O2-tolerant CO dehydrogenase via tunnel redesign for the removal of CO from industrial flue gas

Ni–Fe carbon monoxide dehydrogenases (CODHs) are nearly diffusion-limited biocatalysts that oxidize CO. Their O2 sensitivity, however, is a major drawback for industrial applications. Here we compare the structures of a fast CODH with a high O2 sensitivity (ChCODH-II) and a slower CODH with a lower O2 sensitivity (ChCODH-IV) (Ch, Carboxydothermus hydrogenoformans). Some variants obtained by simple point mutations of the bottleneck residue (A559) in the gas tunnel showed 61–148-fold decreases in O2 sensitivity while maintaining high turnover rates. The variant structure A559W showed obstruction of one gas tunnel, and molecular dynamics supported the locked position of the mutated side chain in the tunnel. The variant was exposed to different gas mixtures, from simple synthetic gas to sophisticated real flue from a steel mill. Its catalytic properties remained unchanged, even at high O2 levels, and the efficiency was maintained for multiple cycles of CO detoxification/regeneration. Ni–Fe carbon monoxide dehydrogenases (CODHs) are able to oxidize CO with a high rate, but their O2 sensitivity is a major drawback for their industrial application. This work shows that CODHs can be tailored for industrial or gas cleaning processes by engineering the selectivity of their gas channels.

A559 was also oriented towards the substrate tunnel, whereas the side chain of M560 faced away from the tunnel.
In addition, we realized that residue A559 forms a triad with its neighbouring residues (V565 and I580) along the junction of the substrate tunnels (Fig. 1d). Interestingly, the substrate tunnels with triad residues appear to be non-selective paths where water molecules and xenon-coordinating residues coexist ( Supplementary Fig. 3). This suggests that various molecules (which include water, CO and O 2 ) can access the active site through putatively non-selective substrate tunnels. Much to our surprise, the triad near the active site is universal in structurally similar Ni-Fe CODHs, even in the different bifunctional acetyl-CoA synthase-MtCODH (Mt, Moorella thermoacetica) subunit (Supplementary Table 1). Moreover, this narrowest position at the conserved triad is likely to act as a bottleneck in gas tunnels that critically affects the transport rate of substrates depending on the tunnel characteristics, such as diameter and dynamic fluctuation 16 (Supplementary Fig. 4a,b). We therefore hypothesized that the selective obstruction of the common bottleneck point in the gas tunnel would decrease the O 2 sensitivity of CODH by blocking O 2 transfer into the active site of the C-cluster. Accordingly, by Through phylogenetic analysis of Ni-Fe CODH proteins, we found that at least two different clades, those that contained ChCODH-IV and DvCODH, are closely localized, distinct from the common lineage and are assumed to be insensitive to and possess unique mechanisms against molecular O 2 . The phylogenetic trees showed orthologous relationships based on the amino acid sequences of the CODH proteins (see Methods for details). b, partial sequence alignment of CODH proteins. Five residues (D94, E101, R187 and Q559 with a yellow shade and Y558 with a pink shade) are conserved in ChCODH-IV and DvCODH but are quite different to those of ChCODH-II. Information on the protein names, including the characterized CODHs with an asterisk and associated non-redundant protein accession numbers, is given in Supplementary Fig. 1. c, Gas tunnels identified in the ChCODH-II and ChCODH-IV structures.
ChCODH-II (pDB 1SU7) shows the highest activity towards CO but is more sensitive to O 2 than ChCODH-IV (pDB 6ELQ). The homodimeric ChCODH-II and ChCODH-IV contain five metal clusters: the [NiFe 3 S 4 ] cluster at the catalytic site (C cluster), the [Fe 4 S 4 ] cubane-type cluster (B cluster) and the [Fe 4 S 4 ] cluster in the dimer interface (D cluster), which is unlike the [Fe 2 S 2 ] D cluster in DvCODH, implying that the catalytic mechanisms of ChCODHs are similar due to the identical metal cluster contents. d, Common bottleneck points of gas tunnels near the active site of Ni-Fe CODHs. A branch of the gas tunnels depicted in the panels ends at cluster C directly above the apical coordination site of the Ni atom. In the structure of ChCODH-II, the gas tunnels and the triad residues (A559, V565 and I580) that constitute the bottleneck are indicated. In less O 2 -sensitive CODHs, the residues that correspond to A559 of ChCODH-II are bulky residues, such as tryptophan and tyrosine.  Table 2 and Supplementary Fig. 5). The variants were screened based on their specific activities of CO oxidation and their residual activities after O 2 exposure (Supplementary Table 3). The effects of the mutations showed a simple and clear correlation between the aromatic side chains and decreased O 2 sensitivity, except for A559S (Fig. 2a). After exposure to dissolved O 2 (10 μM), the residual activities of the variants with aromatic groups, A559H, A559W, A559Y (>90%), A559F and A559P (40-50%), sulfur-containing A559M (58%) and hydroxyl-containing A559S (90%) were detected, whereas ChCODH-II WT exhibited zero residual activity, which shows a complete loss of activity. Additionally, compared with that of WT, the specific activities of A559F, A559H, A559S, A559W and A559Y for the substrate CO were increased by approximately 1.38-2.5-fold (Supplementary Table 3). The amino acid side chains at the 559 position seemed to have a strong impact on the O 2 sensitivity and CO oxidation in ChCODH-II. Figure 2b demonstrates the remarkable effects of mutations to bulkier amino acids on improving the resistance to O 2 . The engineered A559W and A559H ChCODH-II showed an over 50% residual activity due to the notably lower O 2 sensitivity, even in the presence of 38.2-92.9 μM O 2 , whereas the WT showed only a 50% residual activity at 0.63 μM O 2 , which suggested an approximately 60.6-147.5-fold improvement in the half-maximal inhibitory O 2 concentration (Supplementary Table 4). The O 2 sensitivity of the engineered variant of ChCODH-II was much lower than those of the naturally less O 2 -sensitive ChCODH-IV and DvCODH, which would enable its application to industrial flue gas that contains 1.5-6.5 μM O 2 (0.1-0.5%). Table 1 indicates that most ChCODH-II variants have high catalytic rates (k cat ≈ 2,000 s −1 at 30 °C) similar to the value (k cat ≈ 1,500 s −1 at 20 °C) reported by Ragsdale and co-workers 17 , which suggests that the engineering of enzymes did and engineered ChCODHs (closed symbols) was observed for ChCODH-II (squares), ChCODH-IV (circles), DvCODH (triangles), ChCODH-II A559H (diamonds), A559S (hexagons) and A559W (reverse triangles). The grey and pink bars indicate the half-maximal inhibitory O 2 concentrations of ChCODH-II WT and A559H, respectively. The horizontal arrow shows the increase in the half-maximal inhibitory O 2 concentration from ChCODH-II to A559H. The data represent the mean ± s.d., as determined from n = 3 independent experiments (Methods). c, Changes in O 2 sensitivity in ChCODH-IV and its variants. The effect of the O 2 concentration on ChCODH-IV (open symbols) and its variants (closed symbols) was measured for ChCODH-II (squares), ChCODH-IV (circles), ChCODH-IV Y558A (triangles), Y558H (diamonds), Y558S (hexagons) and Y558W (reverse triangles). No activity was observed for ChCODH-IV Y558S or DvCODH W552A. The grey and pink bars indicate the half-maximal O 2 concentrations of ChCODH-IV WT and Y558A, respectively. The horizontal arrow shows the decrease in the half-maximal inhibitory O 2 concentration from ChCODH-IV to Y558A. The data represent the mean ± s.d., as determined from n = 3 independent experiments. d, Time course of O 2 influence on ChCODHs. The residual activities of ChCODH-II (squares), A559W (reverse triangles) and ChCODH-IV (circles) in the presence of 10 μM O 2 were monitored for 3 h. The data represent the mean ± s.d., as determined from n = 3 independent experiments. not disturb the catalytic activity. Moreover, the ChCODH-II variants were as robust as the WT enzyme up to 70 °C (Supplementary Table 5), which presents the additional advantage of higher activities and stabilities even at relatively high temperatures.
Regarding the catalytic properties with CO as a substrate, the apparent Michaelis constant (K M,app ) for CO in ChCODH-II WT was estimated to be 20 μM, which is slightly higher than the value of 8 μM reported by Dobbek and co-workers 9 , but is similar to the value of 18 μM reported by Meyer and co-workers 13 . For the variants, the K M,app value of 24 μM CO for A559Y showed a very similar CO affinity to that of the WT (20 μM), whereas other mutants exhibited a slightly decreased CO affinity. Often, mutations that show a decreased O 2 binding exhibit a corresponding decrease in substrate binding (for example, RuBisCo (ref. 18 ). Owing to the possibility of competition between CO and O 2 for the variants, we determined the ratio K CO M,app /K O 2 I,app (K I,app , inhibition constant), which is crucial to evaluate the efficiency of the variants. Compared with the WT, the variants showed 41-139-fold increases in the relative affinities for CO over O 2 , which indicates that these variants are highly selective for CO even in the presence of O 2 . However, changing the flexibility at the bottleneck of the tunnel will alter catalytic parameters, such as K M,app for CO. In addition, it may be that a small portion of the substrate CO can be transferred by the mutated tunnel, but most of the CO transfer takes place through other tunnels, as shown in Supplementary Fig. 3. A slightly higher K M value for CO implies a lowered affinity of the mutant for CO due to a decreased flexibility at the bottleneck or the possibility of less CO being transferred through this engineered tunnel to bind the active site.
In contrast, mutating the other two residues in the junction triad, V565 and I580, did not decrease O 2 sensitivity, which indicates that the A559 position is a key site for O 2 access in ChCODH-II (Supplementary Table 3). Moreover, among the mutations of M560, another unique position in ChCODH-II and IV, the M560Q variant, showed only a slightly lower O 2 sensitivity ( Supplementary Fig. 5), which suggests that this site is not crucial for decreasing O 2 sensitivity.
As shown in Fig. 2c, in ChCODH-IV, we examined the reverse of the changes that decreased the O 2 sensitivity of ChCODH-II and Y558A to reconfirm the results for the ChCODH-II variants. As expected, Y558A was very sensitive to O 2 , in fact, as sensitive as ChCODH-II, which thus indicates that the role of Y558 (ChCODH-IV) is similar to that in the engineered A559 mutants (ChCODH-II) as a key site for decreased O 2 sensitivity. This supported the hypothesis that the substitution of tyrosine for alanine decreases O 2 sensitivity but does not change the catalytic properties of Y558A relative to those of ChCODH-IV WT (Table 1). With reference to the data on the reduced O 2 sensitivities of the ChCODH-II variants, positive mutations (tryptophan and histidine) were introduced at Y558, and the residual activities of the ChCODH-IV variants were observed (Fig. 2c). These results showed that replacing tyrosine with tryptophan or histidine at 558 also decreased the O 2 sensitivity in ChCODH-IV, as is the case for the ChCODH-II A559 variants. Consequently, we concluded that the 559 or 558 position in either ChCODH-II or IV was the commonly shared key site that affected O 2 sensitivity.
For the stable and efficient conversion of CO in the presence of O 2 , we further estimated the stability of engineered A559W as an oxidizing biocatalyst (Fig. 2d). At 10 μM O 2 , the relative activities of ChCODH-IV and A559W of ChCODH-II were stably maintained over time, but the activity of ChCODH-II WT rapidly diminished. These results revealed that the variant A559W exhibited a stable activity at 10 μM O 2 despite the rapid O 2 inactivation of ChCODH-II WT and the low CO conversion of ChCODH-IV. It is well-known that most Ni-Fe CODHs are inactivated by O 2 within only a few minutes 13,14 , and inactivation is more severe in the absence of O 2 scavengers, such as dithiothreitol (DTT) and dithioerythritol (DTE) 19 . Furthermore, variants exposed to O 2 for several hours can efficiently undergo CO conversion at higher conversion rates than can ChCODH-IV, which has a low specific activity (~90 U mg −1 ). Therefore, it is predicted that low O 2 -sensitive ChCODH-II variants will provide a good performance in removing CO from industrial waste gas mixtures that contain O 2.

Structural analysis of less O 2 -sensitive ChCODH-II variants.
To uncover the reasons for the low O 2 sensitivities of the ChCODH-II  (1)) 30 . ND, not determined.
variants, we solved the crystal structures of the variants A559W (Protein Data Bank (PDB) 7XDM), A559H (PDB 7XDN) and A559S (PDB 7XDP) ( Fig. 3a and Supplementary Table 6). Given that the activity loss of most CODHs under aerobic conditions is closely related to the alteration or destruction of the active site C cluster, which is the most O 2 -sensitive metal site 9,14,20 , to maintain the structural integrity of the C cluster is vital to decrease O 2 sensitivity. Thus, we determined whether the structural integrity of the Fe-S clusters in the three variants was maintained and compared the anaerobic structure of A559W (PDB 7XDM) with the O 2 -exposed structure of A559W (PDB 7ERR) prepared under aerobic conditions ( Fig. 3a and Supplementary Fig. 6). When we examined the positions and the interacting residues around the B, C and D clusters of the variants, they were highly similar root mean squared deviation (r. However, the only local environments around the A559 mutation sites that affect the characteristics of the gas tunnel were significantly changed compared with those of ChCODH-II WT. In the A559W and A559H variants, the side chain of I580 was pushed away through the incorporation of the bulkier residue W559 or H559 instead of alanine, and the position of V582 was also slightly changed by making additional hydrogen bonds with W559 or H559. In the A559S variant, neither hydrogen bonding with V582 residues nor flipping of the I580 side chain was observed. Only the side chain of V582 moved towards S559 to form the additional van der Waals interaction. The incorporation of W559, H559 or S559 made the gas tunnel in the variants narrower than that of ChCODH-II WT ( Fig.  3b and Supplementary Fig. 4c). When we measured the radius of the gas tunnel at the bottleneck point of A559W using the CAVER program 21 , it was reduced by approximately 1.02 Å compared with that of ChCODH-II WT (1SU7). Consequently, we deduced that the obstructed tunnels of the ChCODHs would not allow the transfer of both CO and O 2 to the active site, whereas other gas channels seem to be more selective for CO ( Supplementary Fig. 8).
To estimate the structural rigidity of each tunnel of ChCODH-II WT and its variants, we calculated and compared the distances across the tunnel between two tunnel-forming residues (A/H/S/ W559 and F608) by molecular dynamics (MD) simulation (Fig.  3c,d and Supplementary Figs. 9 and 10). F608 as the opposite tunnel-forming residue with A/H/S/W559 was selected to more accurately reflect the local fluctuation of tunnels as the triad residues that form the bottleneck are not located precisely opposite to position 559. The distance and standard deviation (s.d.) values of A559W or A559S after 3 ns decreased to 10.486 Å and 0.395 Å or 10.467 Å and 0.630 Å, respectively, compared with those of the WT. The simulated distance of A559H and F608 increased to 13.730 Å, but the lower value (0.726 Å) of the local fluctuation was smaller than that in the WT. This analysis showed decreases in local fluctuation in the variant tunnels, which reflects that the constricted tunnels by point mutations become more rigid and more likely to limit O 2 accessibility compared with that of WT. In less O 2 -sensitive Ni-Fe hydrogenases [22][23][24][25] , there is evidence that constriction of the tunnels protects their active sites from O 2 attacks by serving as a barrier to the intrusion of molecular O 2 . Therefore, these results suggest the possibility that the mutation at position 559 is critical for the rigidity of the tunnel and affects O 2 access to the active site of ChCODH-II.
Efficient removal of industrial flue gas from a steel mill. Next, we assessed the activity of engineered A559W for different CO mixtures from synthetic and industrial flue gases (Fig. 4a, Supplementary Table 8 and Supplementary Fig. 11). Compared with the low-content CO (synthetic gas 1, 95% (v/v) CO and 5% (v/v) N 2 ), the relative activities of A559W were similarly observed as a deviation of up to 3.6% towards other CO mixtures: synthetic gases 1 (100%), 2 (95%) and (101%), blast-furnace gas (BFG, 91%), coke-oven gas (COG, 99%) and Linz-Donawitz converter gas (LDG) (96%). This indicates that ChCODH-II A559W has a consistent and sufficient selective performance for gaseous CO mixtures regardless of the low CO content and multicomponent mixtures, such as CO 2 , CH 4 and H 2 . In particular, the lack of severe inactivation by O 2 and unknown trace impurities in the industrial flue gases BFG, COG and LDG suggests that a less O 2 -sensitive ChCODH variant can be used as a suitable and widely applicable biocatalyst for CO-containing flue gases from a variety of industrial environments.
Finally, to evaluate the performance of the low-O 2 -sensitive ChCODH-II variant for the removal of gaseous CO, we tested the variants as an oxidizing biocatalyst in 50% Figure  4b indicates that ChCODH-II WT and A559W readily consumed CO within 2.5 hours under anaerobic conditions, whereas under O 2 , only A559W consumed CO within 3 hours. This experiment confirmed that A559W, unlike ChCODH-II WT, can sufficiently and selectively consume CO without severe O 2 inactivation. Under the aerobic conditions, A559W seemed to traverse a lag phase in half-an-hour, probably until the enzyme reactivation by CO could overcome O 2 -induced arrest. Ni-Fe CODHs can be recovered when CO or dithionite is added 26 . A plausible explanation is that CO can serve as a weak reductant for A559W recovery in the manner of dithionite-treated ChCODH-II 9,14,26 . Furthermore, we observed the reversibility of the inactivation of A559W after a short-term exposure to air ( Supplementary Fig. 12). ChCODH-II WT is almost irreversibly inactivated by O 2 . In contrast, the activity of the variant A559W towards CO reversibly and rapidly recovers from the inactive state caused by air exposure. This reversibility of the O 2 inhibition could be very interesting for industrial applications. However, the exact mechanism of this remarkable reversibility is currently unclear and needs to be further explored. Finally, for practical application purposes, we monitored the CO utilization of A559W using real LDG flue gas from a steel mill in the presence of dissolved O 2 (~13.4%, v/v). The variant A559W displayed an impressive capability to remove the total amount of CO (53.4%) from LDG (Fig. 4c,d), but no CO consumption was observed for ChCODH-II WT. Even under atmospheric conditions of approximately 250 μM O 2 (~20%, v/v), CO uptake of 33% was observed for A559W ( Supplementary  Fig. 13). Moreover, we compared the CO consumptions of variants A559H, A559S and A559W using 50% (v/v) CO-saturated buffer and LDG-saturated buffer (Supplementary Fig. 14). A559W had the fastest CO removal rate in the presence of dissolved O 2 (~13.4%, v/v). This may be because the catalytic efficiency of A559H and A559S is lower than that of A559W, as shown in Table 1. We considered A559W to be the most efficient biocatalyst for the removal of CO. Thus, the less O 2 -sensitive biocatalyst A559W was shown to enable the enhanced CO bioconversion under near-atmospheric conditions and demonstrated outstanding performance in complete CO removal from steel-mill flue gas (LDG).
In addition to an efficient CO consumption, A559W must have the sustainable capacity to be reused and recycled for the treatment of industrial flue LDG. Immobilized biocatalysts can be employed as a simple reuse and recovery method for cost savings in industrial   applications. Figure 5 shows that the conversion rate of CO by immobilized A559W is maintained through repeated cycles of reuse and in the presence of O 2 . The immobilized A559W was incubated at room temperature (r.t.) using flue LDG for the first ten cycles (Fig. 5a,b), and LDG with dissolved 170 μM O 2 (~13.4%, v/v) was then used for the second ten cycles (Fig. 5a,c). In the absence of O 2 , there was no loss of conversion rate during ten cycles, which indicates that the enzyme is still active after repeated uses and is not inhibited by the product CO 2 or other gas impurities from LDG. In the presence of O 2 , ten cycles of the reaction also exhibited complete oxidation of CO to give CO 2 , which indicates that the immobilized A559W is unaffected by O 2 interference. The results suggested that this engineered variant can efficiently remove CO from steel-mill flue gases that contain varying levels of CO and O 2 , even after repeated reuse, and might be adaptable for future industrial applications.

Conclusions
Resistance to atmospheric O 2 exposure is a rare feature among enzymes that express CO dehydrogenase activity. To date, the only known aerobic CODH is the Mo-Cu CODH enzyme. In contrast, Ni-Fe CODHs are highly sensitive to O 2 , with the exception of two slow CODHs, ChCODH-IV and DvCODH, which are less sensitive to O 2 . Thus, the use of low-O 2 -sensitive ChCODH biocatalysts with a marked increase in rate efficiency would provide the advantage of an efficient CO conversion for various waste gases and syngases that contain O 2 (for example, biomass and plastics) in industry or applied green chemistry. Based on our analyses and a rational approach, we selected different strategic points in the enzyme to generate variants able to maintain the high activity rates of ChCODH-II combined with the low O 2 sensitivity of ChCODH-IV. This study thus points towards an effective avoidance of the rapid O 2 inactivation of Ni-Fe CODHs through an increased selectivity of the gas tunnel. Catalytic activity and decreased O 2 sensitivity are often considered to be trade-offs, but the ChCODH-II variants show how to overcome such limits.
In summary, our study presents the key discoveries that one position is enough to decrease the sensitivity to O 2 and that the second, unaffected tunnel seems to be highly specific for CO/CO 2 rather than O 2 ; it provides some reasons as to why all anaerobes have such highly O 2 -sensitive CODHs even though one point mutation would be enough to drastically improve their CODHs by conferring a faster turnover and decreased O 2 sensitivity but probably less efficient in vivo under a lower CO environment and it suggests that the   , v/v)). The CO 2 produced in the reaction of either ChCODH-II (white bars) or A559W (orange bars) was monitored. The data represent the mean ± s.d., as determined from n = 3 independent experiments. d, Gas chromatography (GC) analysis of the reaction products from LDG gas and A559W. The GC chromatograms were obtained after the reaction of A559W using LDG flue gas. The products were identified by comparison with authentic standards (peak 1, H 2 ; peak 2, N 2 ; peak 3, CO; peak 4, CO 2 ).
next step in applying these designed biocatalysts will be coupling on an electrode or direct transfer of the reducing power to another system (for example, CO 2 fixation into formate). Alternatively, these variants could be implanted in anaerobic bacteria to increase their performance as bioconverters. This work provides concrete proof that CODHs can be tailored for industrial or gas-cleaning processes by engineering the selectivity of their gas tunnels.

Methods
Gene synthesis and cloning. The genes encoding ChCODH-II (CHY_0085), ChCODH-IV (CHY_0736) and DvCODH (DVU_2098) with NdeI and BamHI restriction sites were optimized, synthesized and sequence verified by Macrogen based on their GenBank sequences (accession nos. NC_007503 and NC_002937). The resulting fragments were cloned into pET28a (Novagen) that contained a His tag and an additional thrombin-cutting site. Site-directed variants were constructed using the QuikChange site-directed mutagenesis method from Stratagene through Pfu DNA polymerase. PCR products were incubated for 60 min at 37 °C with DpnI Cell lysates were centrifuged and purified by Ni-NTA affinity chromatography (Qiagen) with a standard buffer that contained 2 mM DTE and 2 μM resazurin, but DTE and resazurin were omitted from the washing and elution steps during the O 2 inactivation assay. Protein concentrations were determined using the Bradford method. When the purified proteins were subjected to SDS-PAGE on 12% gels, all the variants were detected as a single soluble band whose size (69 kDa) approximately corresponded to the calculated size of the proteins (67 kDa) with the His 6 -tag (2 kDa). Western blot analyses using anti-His-tag antibodies were also carried out to confirm the presence of the His tag and putative degradation of the His-tagged protein. For the crystallization of ChCODH-II variants (A559W, A559H and A559S) under anaerobic conditions, further purification was anaerobically performed in an anoxic glove box (model B, COY Laboratory Products Inc.) by size-exclusion chromatography (HiLoad 16/600 Superdex 200 prep grade, GE Healthcare Bio-Sciences), which was previously equilibrated with a buffer that contained 20 mM Tris-HCl pH 7.5 and 3 mM DTT. For crystallization of the aerobic A559W variant, the protein purification step using Ni-NTA resin was conducted in a similar manner to that above but under aerobic conditions (outside the anaerobic chamber). Further purification by size-exclusion chromatography was also performed aerobically with a buffer that contained 20 mM Tris-HCl pH 7.5 and 3 mM DTT.
Size-exclusion chromatography with multiangle light scattering. Size-exclusion chromatography with multiangle light scattering experiments on the ChCODH-II A559W variant were performed aerobically using a fast protein liquid chromatography (FPLC) system (GE Healthcare) connected to a Wyatt MiniDAWN TREOS MALS instrument and a Wyatt Optilab rEX differential refractometer. A Superdex 200 10/300 GL (GE Healthcare) gel-filtration column was pre-equilibrated with a buffer that contained 20 mM Tris-HCl pH 7.5 and 3 mM DTT and normalized using ovalbumin protein from hen egg white (Sigma-Aldrich). Proteins were injected (1 mg), and elution was performed at a flow rate of 0.4 ml min −1 . Data were analysed using the Zimm model for static light-scattering data fitting and graphed using an EASI graph with an ultraviolet peak in ASTRA V software (Wyatt), as shown in Supplementary Fig. 15. The first ten cycles of reaction were initiated in the absence of O 2 by the addition of the immobilized A559W enzyme (500 U mg −1 immobilized protein). The immobilized enzyme was recovered through a disposable open column, washed twice with 200 mM HEpES/NaOH buffer pH 8 and reused after each reaction. c, Reusability of immobilized ChCODH-II A559W using LDG with O 2 (closed symbols). After the first ten cycles using LDG, the second ten cycles were then performed in the presence of 170 μM O 2 (~13.4%, v/v).
Activity measurement. CO oxidation activity was measured at 30 °C after saturation with CO using screw-cap cuvettes with a CO head space. The activity was assayed by observing the CO-dependent reduction of oxidized EV (EV ox , ε 578 = 10,000 M −1 cm −1 ) (ref. 28 ). The reactions were started by enzyme injection (0.1-0.2 μg) into a screw-cap cuvette that contained 2 ml of reaction buffer (20 mM EV ox , 0-250 μM O 2 , 50 mM HEPES/NaOH buffer pH 8 with saturated CO from 50 ml of stock solution flushed with CO gas for 1 h) (ref. 29 ). When the activity of the enzyme is measured, trace O 2 during the enzyme preparation may slightly affect the CODH activity due to the lack of a reducing agent (DTE). Thus, small differences can occur in the measurement of the specific activity. For practical applications, the enzymatic oxidation of CO as a carbon and energy source at this mild temperature has the advantages of a spontaneous reaction without the need for additional energy and of easy coupling with other mesophilic enzymes, such as formate dehydrogenase.
To test the effect of O 2 inactivation, we measured the residual activity of the enzyme at appropriate O 2 concentrations. Before assaying the residual activity, the CODHs (2-4 μg) were incubated in 50 mM HEPES (final volume of 200 μl) for 2 min with O 2 addition, and then the reaction mixture was diluted 200-400 times to give a final O 2 concentration below 0.6 μM. One unit of CODH activity was defined as the amount of enzyme required to reduce 1 μmol of EV ox per min at 30 °C and pH 8.

Kinetic analysis.
For calculations of the kinetic parameters for EV ox , a CO-saturated buffer that corresponded to 0.91 mM CO (99.998% according to Henry's law) was used at 30 °C. The headspace of a rubber-septum-stopped cuvette that contained 0.5-8 mM EV in 50 mM HEPES/NaOH buffer (pH 8) was purged with CO for 1 h. To monitor the initial velocity of the CODHs, the reaction was initiated by an injection of recombinant CODHs (0.1-0.2 μg). For measurement of the kinetic parameters towards CO, two CO-saturated buffers that corresponded to 0.91 mM CO (99.998% according to Henry's law) and 45.5 μM CO (5% CO (v/v)/95% N 2 (v/v) according to Henry's law) were used at 30 °C. Three CO concentrations (10 μM, 20 μM and 40 μM) and two CO concentrations (80 μM and 160 μM) that each contained a final concentration of 20 mM EV in 50 mM HEPES pH 8 buffer were immediately prepared before the reaction by diluting two CO-purged buffers of 0.0455 mM and 0.91 mM, respectively. The enzyme reaction was initiated by injecting each reaction buffer with different CO concentrations (0.01-0.16 mM) into the enzyme (0.1-0.2 μg). The absorbance change at 578 nm was spectrophotometrically monitored at 30 °C in an anoxic glove box. The kinetic parameters (k cat,app and K M,app ) for CO and EV ox were calculated by the Hanes-Woolf equation (Supplementary Fig. 16). All of the enzymatic activities were determined in triplicate. For K I determinations, IC 50 values ( Supplementary Fig. 17 Crystals of the ChCODH-II A559W variant were grown under aerobic conditions using the sitting drop method by mixing 0.2 μl of protein (10 mg ml −1 ) with 0.2 μl of reservoir solution that consisted of 0.1 M HEPES/NaOH pH 7.0, 50 mM MgCl 2 and 25% (w/v) polyethylene glycol 3350. The crystals were transferred to a solution that contained the reservoir solution and 5% (w/v) glycerol for cryoprotection. Data were collected at 100 K in 1° oscillation at the 5C beamlines of the Pohang Light Source with a wavelength of 1.0000 Å. The structure calculation methods were identical to those used for the anaerobic proteins. Ramachandran statistics (%) of favoured, allowed and outliers were 96.7, 3.2 and 0.2, respectively. The atomic coordinates and structure factors were deposited in the Protein Data Bank (PDB 7ERR). X-ray absorption spectroscopy and Fe anomalous difference map. Fluorescence scans were performed on crystals of anaerobic ChCODH-II A559W, A559H and A559S variants at the 11C and 5C beamlines of the Pohang Light Source near the X-ray absorption edge of Fe ions. The anomalous diffraction data were collected at the peak position of the Fe absorption edge (7.135 keV). Anomalous datasets were processed and scaled using the HKL2000 software package 31 . Supplementary Table  7 lists the anomalous data-collection statistics. An anomalous difference Fourier map calculation was performed using the fast Fourier transform in the CCP4 refmac5 program 35 .
Molecular dynamics simulation of ChCODHs. The complex structure was optimized by MD simulation using Discovery Studio software (BIOVIA) with the CHARMm force field 36 . The structures of WT (PDB 1SU7), A559W (aerobic, PDB 7ERR; anaerobic, PDB 7XDM), A559H (PDB 7XDN) and A559S (PDB 7XDP) were analysed for simulation. The protein force field parameters were taken from the CHARMm force field. The protonation states of the acidic and basic residues in the protein were analysed using the calculated protein ionization and residue pK protocol. Protons were added to the input CODH structure based on the ionization states of the titratable groups obtained from the Discovery Studio protocol, which calculates the ionization state of a protein at a specified pH (8.0). The transferable intermolecular potential with 3 points (TIP3P) water model 37 was used for the solvent. For the Fe-S clusters, we used the force-field parameters provided by Discovery Studio (Supplementary Table 9). The C clusters of the protein were fixed to prevent an unnatural conformation. Solvation was used to simulate the more natural behaviour of tunnels that consisted of amino acids, which are affected by solvent molecules such as surface residues. The protein was solvated with 19,873 water molecules and 53 sodium and 54 chlorine ions for the WT, 20,924 water molecules and 66 sodium and 55 chlorine ions for the A559W mutant, 19,418 water molecules and 60 sodium and 51 chlorine ions for the A559H mutant and 20,045 water molecules and 60 sodium and 53 chlorine ions for the A559S mutant to create a 0.145 M NaCl solution in a cubic unit cell, thereby neutralizing the protein charge. The periodic boundary condition was set to prevent truncation effects. The particle mesh Ewald method was used to calculate long-range electrostatic forces 38 . All the bonds with hydrogen were constrained with the SHAKE algorithm 39 . A standard dynamics cascade protocol was performed to apply nanoscale MD simulation in Discovery Studio. This protocol consisted of five steps. For minimization, the steepest descent algorithm with a max step of 1,000 and a r.m.s. gradient of 1.0 was used first. For a more detailed minimization, the Adopted Basis Newton-Raphson algorithm with a max step of 2,000 and an r.m.s. gradient of 0.1 were applied. After a brief minimization process, the system was heated for 4 ps in the isothermal-isobaric (NPT) ensemble to a temperature of T = 300 K from T = 50 K and a pressure of P = 1 atm. Next, the system was equilibrated for 10 ps in the NPT ensemble at a temperature of T = 300 K and a pressure of P = 1 atm. A subsequent production run of 10 ps followed by a production run of 10 ps were carried out in the NPT ensemble. The full system of MD was run for 8 ns with the NPT conditions. The trajectories were saved every 2 ps for analysis. Whether the protein reached an equilibrated state was evaluated by the energy and volume under the NPT ensemble 40 .
Reusability of immobilized A559W. The soluble protein from the cell lysate (15-25 ml of lysis buffer per gram of wet cell weight) for immobilization was incubated with Ni-NTA agarose (as an immobilization carrier) with shaking at 80-100 rpm for at least 2 h in an anaerobic chamber. After 2 h of incubation, when the soluble proteins, which were brownish due to the Fe-S clusters, were sufficiently bound to the bluish Ni-NTA agarose, the resin colour visibly changed to dark brown. The resultant Ni-NTA immobilized proteins were washed with 1 column volume (20 ml disposable gravity flow column) of wash buffer (20 mM imidazole, 300 mM NaCl, 50 mM sodium phosphate, pH 8) to eliminate unbound proteins. A small aliquot of these immobilized proteins was eluted to check the specific activity of the protein in the same manner described in the protein purification section above. In addition, we measured the apparent units of protein per immobilization bead volume owing to the difficulty of determining the exact amounts of immobilized proteins.
The reaction buffer for the enzyme reaction was anaerobically prepared in a rubber-capped serum bottle (115 ml) with 95 ml of buffer solution (40 mM EV ox , 200 mM HEPES, pH 8), transferred from the anaerobic chamber, and then purged with LDG for 1 h. The headspace gas of the purged serum bottle was further analysed by the GC-TCD method above. The main enzyme reactions were performed for 20 consecutive cycles: 10 cycles using LDG only and another 10 cycles using LDG with dissolved O 2 . For the 10 cycles without O 2 , the first reaction cycle was performed at r.t. and 150 rpm with shaking by syringe injection of the immobilized protein (50 U mg −1 immobilized protein) into the LDG-purged serum bottle. After 30 min of incubation on the shaker with the bottle placed horizontally, the gas sample from the headspace in the reaction bottle was analysed by a GC instrument. Next, the immobilization beads used after the reaction were collected in a disposable column, and the collected immobilized proteins were washed twice with 20 ml of wash buffer (200 mM HEPES, pH 8). This recycled protein was consecutively reused in the next cycle with a freshly prepared LDG-purged reaction buffer. In the same manner, the immobilized enzyme reactions were repeated for ten cycles.
For the ten cycles using LDG with dissolved O 2 , we also prepared the O 2 -purged buffer, in which the reaction buffer (200 mM HEPES, pH 8) was purged with O 2 (100%, v/v) for 1 h. The rubber-capped serum bottle (115 ml) with 82.3 ml of reaction buffer (40 mM EV ox for a 95 ml final volume, 200 mM HEPES, pH 8) was transferred from the anaerobic chamber and then purged with LDG for 1 h. Then, 12.7 ml of the O 2 -purged buffer was injected into the LDG-purged reaction buffer (final O 2 concentration of 170 μM). The gas sample from the headspace in the reaction buffer was analysed by GC. As in the previous experiments using only LDG-purged buffer, the enzyme reactions using LDG with dissolved O 2 were repeatedly performed by incubation of the same immobilized protein after ten cycles of reuse. All the gas samples from the ten reaction cycles in the presence of LDG and O 2 were analysed by GC.
Bioinformatics analysis. Searches for amino acid sequence homologues and multiple sequence alignments were performed using BLAST and ClustalW, respectively. The hybrid-cluster proteins with no CODH activity were excluded from the sequence analysis. Conserved domains and clusters of orthologous groups of proteins were analysed using the CD-Search tool at the National Center for Biotechnology Information (http://www.ncbi.nlm.nih.gov). Structural alignments for the assessment of the topological similarity of protein structures were analysed using TM score software 41 (https://zhanggroup.org/TM-align). Phylogenetic trees were constructed using MEGA 10 software 42 with the neighbour-joining method (1,000 replicates). Bootstrap values (>90%) are displayed at the branch points. Non-redundant reference sequences (1,710 proteins) of Ni-Fe CODHs belonging to id_1151 (the number in clusters of orthologous groups) from eggNOG 5.0 43 (orthologues database) were manually checked for completeness and length. Sequences shorter than 620 amino acids and longer than 700 amino acids were not used for alignment, and sequences longer than 650 amino acids were shortened. The resultant 208 amino acid sequences were used in the phylogenetic analysis. The optimal tree with the sum of branch length = 33.96697452 is shown. All the ambiguous positions were removed for each sequence pair (pairwise deletion option). There were a total of 854 positions in the final dataset. Moreover, the substrate tunnel analyses were carried out using the tool of CAVER3.0 software 21 (http://www.caver.cz).
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
The datasets generated during and/or analysed during the current study are available from the corresponding authors upon reasonable request. MD trajectories of a total of 5 ns are available on Zenodo at https://doi.org/10.5281/ zenodo.6865415. The atomic coordinates and structure factors for the ChCODH-II A559 variants have been deposited in the Protein Data Bank under accession codes PDB 7ERR, 7XDM, 7XDN and 7XDP.