First-Principles Calculations on Ni,Fe-Containing Carbon Monoxide Dehydrogenases Reveal Key Stereoelectronic Features for Binding and Release of CO2 to/from the C-Cluster.

In view of the depletion of fossil fuel reserves and climatic effects of greenhouse gas emissions, Ni,Fe-containing carbon monoxide dehydrogenase (Ni-CODH) enzymes have attracted increasing interest in recent years for their capability to selectively catalyze the reversible reduction of CO2 to CO (CO2 + 2H+ + 2e- ⇌ CO + H2O). The possibility of converting the greenhouse gas CO2 into useful materials that can be used as synthetic building blocks or, remarkably, as carbon fuels makes Ni-CODH a very promising target for reverse-engineering studies. In this context, in order to provide insights into the chemical principles underlying the biological catalysis of CO2 activation and reduction, quantum mechanics calculations have been carried out in the framework of density functional theory (DFT) on different-sized models of the Ni-CODH active site. With the aim of uncovering which stereoelectronic properties of the active site (known as the C-cluster) are crucial for the efficient binding and release of CO2, different coordination modes of CO2 to different forms and redox states of the C-cluster have been investigated. The results obtained from this study highlight the key role of the protein environment in tuning the reactivity and the geometry of the C-cluster. In particular, the protonation state of His93 is found to be crucial for promoting the binding or the dissociation of CO2. The oxidation state of the C-cluster is also shown to be critical. CO2 binds to Cred2 according to a dissociative mechanism (i.e., CO2 binds to the C-cluster after the release of possible ligands from Feu) when His93 is doubly protonated. CO2 can also bind noncatalytically to Cred1 according to an associative mechanism (i.e., CO2 binding is preceded by the binding of H2O to Feu). Conversely, CO2 dissociates when His93 is singly protonated and the C-cluster is oxidized at least to the Cint redox state.

Introduction

Use of carbon dioxide as a carbon feedstock for the production of useful chemicals and fuels is considered one of the most promising approaches to overcome the limited supply of fossil fuels and simultaneously reduce the atmospheric concentration of greenhouse gases. Selective CO2 reduction at low activation energy, however, is a critical challenge due to the high thermodynamic stability of the CO2 molecule and the multielectron and multiproduct nature of the reduction process. For the development of large-scale and eco-friendly processes for CO2 conversion, efficient and selective electrocatalysts based on inexpensive metals are therefore required. In this context, biological systems involved in the reductive assimilation of CO2 to organic carbon may be a source of inspiration. In particular, a deep understanding of the chemistry performed by carbon monoxide dehydrogenases (CODHs), evolved over millions of years to efficiently catalyze the otherwise difficult two-electron reduction of CO2, may be extremely useful to design novel and sustainable bioinspired catalysts for high-performance CO2 to CO conversion.

Two chemically distinct types of CODHs are distinguished by their distribution and metal composition. The first of these is the O2-sensitive enzyme from obligate anaerobic bacteria and archaea containing a highly asymmetric [Ni-Fe-S] cluster. This enzyme catalyzes CO oxidation with turnover frequencies (TOFs) of up to 40000 s–1 and CO2 reduction with TOFs of 45 s–1.[1,2] The second class of CODHs is the O2-tolerant enzyme occurring in aerobic carboxidotrophic bacteria. They contain a bimetallic [Mo-(μ2-S)-Cu] system that only catalyzes CO oxidation at a moderate TOF (100 s–1).[3,4] The capability of Ni,Fe-containing CODHs (henceforth simply Ni-CODHs) to catalyze the reversible interconversion between CO2 and CO has led researchers to spend increasing efforts in the study of these enzymes.

The Ni-CODH enzyme is a homodimeric protein of approximately 130 kDa with five metal clusters (see Figure ).[5] Each subunit contains a [Fe4S4] cubane (B-cluster) and an asymmetrical [Ni-Fe-S] cluster (C-cluster), at which the reversible CO2 reduction occurs. An additional [Fe4S4] cubane (D-cluster) is located at the interface between the two monomers. The Ni-CODH can also be a part of the heterotetrameric CO dehydrogenase/acetyl-coenzyme A synthase (CODH/ACS) complex, in which the reduction of CO2 is coupled with the synthesis of acetyl-CoA in autotrophic and acetogenic bacteria,[6,7] or of the multimeric acetyl-CoA decarbonylase/synthase (ACDS) complex, in which the disassembly of acetyl-CoA is catalyzed for producing CH4 in methanogenic archaea.

Figure 1

Cartoon representation of the X-ray crystal structure of the Ni-CODH homodimer from C. hydrogenoformans (PDB code 3B52). The C-cluster and its protein environment are shown enlarged on the right half of the figure. Atoms constituting the small DFT model (SM) are represented by a ball and stick representation, and corresponding residue names are indicated with bold labels; atoms also included in the large model (LM) are depicted as in a stick representation. Aliphatic hydrogen atoms and the CO2 ligand bridging the Ni-Feu site in the 3B52 structure are not shown.

The Ni-CODH active site or C-cluster (see Figure ), in both unifunctional and bifunctional enzymes, is covalently bound to the protein by five cysteine residues and one histidine residue. It is composed by an unusual structure formed by Ni, Fe, and S atoms; three Fe atoms and one Ni atom form a [NiFe3S4] cluster, with a structure very similar to that of a “canonical” [Fe4S4], in which an additional Fe atom extraneous to the cuboidal-like core (unique Fe or Feu) is inserted at an Ni–S edge. Three redox states of the C-cluster have been characterized by spectroscopic data: a fully oxidized inactive state (Cox), an active state obtained from the monoelectronic reduction of Cox (Cred1), and a state obtained from the bielectronic reduction of Cred1 (Cred2). A further undetected diamagnetic state (Cint) is postulated to have an intermediate redox state between Cred1 and Cred2. Cox has the spin state S = 0 and exhibits a Mössbauer spectra typical of [Fe4S4]2+ with no evidence of Feu.[8] Mössbauer parameters of the S = 1/2 Cred1 state (g values at 2.01, 1.81, and 1.65, gav = 1.82)[9] suggest instead high-spin Fe(II), Fe(II), Fe(III) formal oxidation states for the [Fe3S4] subsite and the high-spin Fe(II) state for Feu,[8] whereas L-edge X-ray absorption spectroscopy (XAS) indicates a low-spin diamagnetic Ni(II) ion.[10] The lack of 61Ni hyperfine coupling in the Cred1 EPR signal is consistent with a Ni site electronically isolated from the cluster that does not participate in the spin-coupling mechanism. The electronic structure of the paramagnetic Cred2 state (g values at 1.97, 1.87, and 1.75; gav = 1.86) is even more uncertain.[5,11−14] The similar EPR spectra of Cred1 and Cred2 suggest that the electronic structure of the [Fe3S4] core fragment is unchanged, whereas Ni K- and L-edge XAS studies are consistent with a low-spin diamagnetic Ni(II) for both states.[10,15] However, the accommodation of two electrons at the Feu atom appears unlikely. On the basis of these considerations, two alternative descriptions of Ni in the Cred2 state have been proposed: Ni(0) or the isoelectronic protonated site formulated as the nickel hydride species Ni(II)-H.[5,14]

CODH catalysis should involve the reductive conversion of the inactive Cox state to Cred1 and Cred2. Since Cred1 and Cred2 differ by two electrons and have an operational midpoint potential of −530 mV, which coincides with the values found for the CO2/CO pair (E°′ = −558 mV), they are respectively proposed as the redox states competent for CO oxidation and CO2 reduction.[12,16] However, different mechanisms for the Ni-CODH catalytic cycle have been proposed due to the uncertainty in the oxidation states, the nature of the active ligands, and their coordination mode in Cred1 and Cred2.[5,14,16−18] According to the Ni(II)/Ni(0) assignment for the Ni atom in the Cred1/Cred2 state and Jeoung and Dobbek’s high-resolution X-ray structures,[5,19] the catalytic mechanism reported in Scheme a has been proposed. It involves the formation of a nonbridging hydroxide ligand bound to the Feu atom in both active Cred2 and Cred1 states (see models I and IV, respectively) and a CO2-bound intermediate in which CO2 bridges the Ni-Feu site (model II). In the latter, the C atom is bound to the Ni atom, one O atom of the carboxylate group (O1) is coordinated to Feu and hydrogen-bonded to a conserved Lys residue, and the other O atom (O2) is H-bonded to a conserved His residue. On the basis of these structures, the binding of CO2 to Cred2 is proposed to take place via a dissociative mechanism (i.e. CO2 binds the C-cluster after the dissociation from Feu of the hydroxide ligand as H2O; see the I → II step in Scheme a). Subsequent cleavage of the C–O1 bond and transfer of a proton from the solvent to O1 via a series of His residues results, as very recently observed also by Liao and Siegbahn,[20] in the formation of a CO ligand and a OH– ion, terminally bound to the Ni and the Feu atoms, respectively: an intermediate structurally related to model III in Scheme has been actually proposed by the latter authors on the basis of DFT calculations. CO may then be released, with the C-cluster becoming oxidized by two electrons. Finally, two electrons are transferred, one at a time, from external electron donors through the D-cluster and the B-cluster to the C-cluster. This returns the C-cluster to Cred2.

Scheme 1

Ni-CODH Catalytic Mechanism (in the Direction of CO2 Reduction) Proposed by (a) Jeoung and Dobbek[5] and (b) Appel et al.[18],

Notably, in the first mechanism CO2 binds to the C-cluster after the dissociation of a H2O molecule from Feu (dissociative mechanism), whereas in the second mechanism CO2 binds to Ni when a hydroxide ligand is still coordinated to Feu (associative mechanism).

Notably, in the first mechanism n class="Chemical">CO2 binds to the C-cluster after the dissociation of a H2O molecule from Feu (dissociative mechanism), whereas in the second mechanism CO2 binds to Ni when a hydroxide ligand is still coordinated to Feu (associative mechanism).

A revised version of such a mechanism (see Scheme b) has been reported on the basis of information provided by biochemical experiments and X-ray diffraction studies of the n-butyl isocyanate inhibited enzyme.[21] This form, in which the inhibitor n-butyl isocyanate is terminally coordinated to Ni and an hydroxide ligand is terminally bound to Feu, is proposed to mimic a catalytic intermediate prior to the formation of the Ni–C–O1–Fe bridge (see model II′ in Scheme b). According to this hypothesis, CO2 terminally binds to the Ni atom through an associative mechanism when a hydroxide ligand is still coordinated to Feu (see the I′ → II″ step in Scheme b).

The oxidative addition of CO2 to the Ni atom of the Cred2 state featuring a hydride bound to the Ni(II) ion has also been proposed on the basis of combined structural and theoretical data.[14,17] However, also in this version of the Ni-CODH catalytic mechanism, there are many uncertainties about the binding of the CO2 substrate to the C-cluster.

In order to shed more light on these aspects of the Ni-CODH catalytic cycle, quantum mechanical calculations have been performed on a minimal and a very large model of the active site. In particular, the CO2 binding to the C-cluster has been investigated in the Cred1, Cint, and Cred2 redox states, in the presence and in the absence of a hydroxide ligand bound to Feu. With the final aim of contributing to the provision of significant insights in unveiling the stereoelectronic and catalytic properties of the Ni-CODH enzyme, a detailed analysis of the geometries and electronic structures of relevant intermediates is also provided.

Results obtained from this study allow us to propose possible reaction mechanisms for the binding and release of CO2 to/from the C-cluster. However, it should be pointed out that the latter are discussed only with consideration of minimum energy structures and, therefore, should be interpreted with care. Kinetic aspects, namely the prediction of transition states and the calculation of corresponding energy barriers, which must be considered for a complete and exhaustive mechanistic description of a reactive process, will be the object of a future work. The investigation described in the following indeed represents the initial step of a research line we are currently developing on the catalytic mechanism of Ni-CODHs.

Methods

Models of the Ni-CODH Active Site

The starting structure for the DFT calculations was based on the X-ray geometry of the Carboxydothermus hydrogenoformans Ni-CODH (PDB code: 3B52),[5] in which a CO2 molecule bridges the Ni and the Feu atoms of the active site. In the following, the residues are numbered according to this structure. In the framework of the cluster approach,[22−24] two models of different size (see Figure ) have been considered to investigate the effect of the protein environment on the stereoelectronic properties of the active site of Ni-CODH.

The smallest model (SM), which contains up to 64 atoms, includes the [Fe4NiS4] core of the C-cluster and the side chains of the residues forming its first coordination sphere (see the ball and stick representation in Figure , right). The five cysteine residues coordinated to the nickel and iron atoms (Cys295, Cys333, Cys446, Cys476, Cys526) and the histidine residue coordinated to the Feu atom (His261) are terminated at the Cα atoms and saturated with hydrogens. During the geometry optimization, terminal atoms are constrained to their crystallographic positions, in order to avoid unrealistic distortions of the C-cluster.

The largest model (LM) contains up to 234 atoms and has a size of 24 Å (see the stick representation in Figure , right). This model includes selected atoms of 16 residues belonging to the second coordination sphere (Ala91, Gly92, His93, Ser94, His96, Asp219, Cys223, Asp 231, Glu299, Arg303, Gln332, Ser525, Val527, Lys563, Ile567, and Trp570) and three water molecules, apart from all the atoms contained in the small model. In particular, the entire residue His93 and the side chain of Lys563 have been included in the model because they should be directly involved in the catalytic cycle, interacting with the ligands bounded to the C-cluster[5,19] and participating in acid–base reactions.[25,26] Notably, His93 is positioned at the top of a cationic tunnel composed of histidine residues located on sequential turns of a helix starting near the C-cluster and ending at the protein surface, which is proposed to facilitate transfer of protons during the reaction.[25,27] The importance of His93 and Lys563 in catalysis is confirmed by the loss of enzymatic activity after their mutation.[27] Three protonation states are possible for His93 depending on whether δN, εN, or both atoms are protonated. Since the proton on δN strongly interacts with the carboxylate group of Asp219, it was always included in the model. Conversely, the proton on εN interacts with nonprotein ligands at the active site. Therefore, it is possible to assume that protonation state of εN plays a crucial role in the binding and release of the substrates. On the basis of these considerations, His93 has been modeled as either doubly protonated or singly protonated at the δN atom. Both protonation states of Lys563 (neutral and positively charged) have been also considered. The carbonyl and the Cα atoms of Ala91, the entire residue Gly92, and the N and Cα atoms of Ser94 have been included in the model because they form a small α-helix containing His93, whereas the side chain of Asp219 terminated at the Cα atom has been selected because its carboxylate group is H-bonded to His93. Conversely, the side chain of Trp570 terminated at the Cβ atom, interacting with Asp219, and the side chain of Cys223 terminated at the Cα atom have been included in the model in order to avoid unrealistic conformational changes of the side chain of His93. The side chain of Ile567 terminated at the Cα atom has been added to the model, since it is close to the vacant coordination site on Ni, whereas the side chain, the carbonyl and the Cα atoms of Ser525, and the N and the Cα atoms of Val527 have been selected because they form the peptide chain containing the residue Cys526, belonging to the first coordination sphere. Conversely, the side chain of Glu299 terminated at the Cα atom has been included in the model because it is H-bonded to Ser525. Finally, the side chains of His96, Gln332, and Cys223 terminated at the Cα atoms, the side chain of Arg303 truncated at the Cγ atom, the side chain of Asp231 terminated at the Cβ atom, and three water molecules, selected among those reported in the crystallographic structure, have been included in the model in order to mimic the entire H-bond network in the C-cluster environment. The truncated residues have been saturated with hydrogen atoms. During the geometry optimizations, 31 atoms have been constrained to the crystallographic positions, in order to avoid unrealistic distortions at the boundary of the model.

The Supporting Information contains a detailed list of the atoms composing the SM and the LM models and a list of the atoms that, during geometry optimizations, are fixed at their respective X-ray positions (see Table S1).

A medium model, containing up to 132 atoms, of the active site has also been considered in order to calculate zero-point energy corrections. It contains the [Fe4NiS4] cluster and the side chains and the Cα atoms of Cys295, Cys333, Cys446, Cys476, Cys526, His261, His93, Lys563, Asp219, and Ile567. Constrained-geometry optimizations of this medium model have been performed. Initially, all atoms with the exception of hydrogens were constrained at the positions computed for the large-sized model. Constraints were then progressively removed. Such a procedure allows us to obtain structures that best reproduce the geometries obtained with the large model and to compute consistent reaction energies.

Computational Details

Quantum mechanics (QM) calculations have been carried out in the DFT framework with the Turbomole program suite,[28] using the BP86 exchange-correlation functional[29,30] in conjunction with the resolution of the identity (RI) technique.[31] The BP86 functional is commonly used to study metal-containing molecular systems such as metallo-enzymes, due to the increasing available computational data which indicate that BP86 is one of the most accurate pure functionals to study transition-metal compounds.[32] BP86, coupled with an appropriate basis set, predicts reaction energies with a reasonable accuracy and reproduces experimental geometries within a few hundredths of an angstrom.[33] In the small model, an all-electron valence triple-ζ basis set with polarization functions (def-TZVP)[34] was used for all atoms. In the large model the def-TZVP basis set was used for the [Fe4NiS4] core of the C-cluster, the sulfur, and the Cβ atoms of Cys526 and all atoms of Cys295, Cys333, Cys446, Cys476, and His261. For all other atoms, the double-ζ basis set SVP[35] was used (for details, see Table S1). The BP86/def-TZVP level of theory for the representation of the metal-containing cofactor has been proved to be suitable for the reproduction of both electronic and structural properties of complex antiferromagnetically coupled bioinorganic systems; for example, it was proven to correctly describe the stereoelectronic properties of the Ni-Fe active site of Acetyl-CoA synthase[38] and of the chain of multiple-metal-containing redox-active sites embedded in FeFe-hydrogenases.[39] The effects of the protein environment have been modeled by placing the molecular cluster in a polarizable continuum medium with ε = 4, according to the conductor-like screening model (COSMO).[36,37]

In order to verify the consistency of the results, single-point energy calculations on the BP86-optimized geometries of relevant species along the CO2 binding/dissociation pathway have been also carried out using different functionals: namely, B3LYP,[40−42] PBE0,[43] and M06.[44] The obtained results are consistent with those obtained by using the BP86-COSMO scheme; the reaction energies are similar and show the same trend with respect to the redox state of the C-cluster (see Table S2). BP86, B3LYP, PBE0, and M06 single-point energy calculations have been also performed by including the dispersion interaction correction given by the DFT-D3 Grimme scheme,[45−47] as implemented in Turbomole. Since the binding of CO2 to the C-cluster involves the formation of a short metal–ligand bond, CO2 binding energies could be sensitive to the spatial cutoff function. Therefore, D3 dispersion contributions were also calculated by applying the Becke–Johnson (BJ)-damping function.[48] Since the M06 functional is parametrized for the treatment of short- and medium-range correlation effects, it is incompatible with the BJ-damping function[49] and only the zero-damping function has been applied to it. Structure optimizations including dispersion corrections have not been performed because they have been shown to increase the deviation between DFT and experimental binding energies.[50] As shown in Table S2, for a given reaction, the inclusion of dispersion interactions in the calculations has a constant effect and it is independent of the redox state of the active site; when dispersion interactions are considered, CO2 binding is always predicted to be more favorable. Conversely, the use of the BJ-damping function instead of the standard damping function has only a negligible effect on reaction energies. This result is in line with the work by Grimme and co-workers.[48] Zero-point energy corrections have been also computed for several species using the medium model of the active site. Notably, they are always lower than 3 kcal/mol for CO2 binding and equal to zero for isomerization reactions (see Table S3). Additionally, for a given reaction, they are almost the same for the different redox states of the C-cluster. Therefore, for the sake of clarity, the results obtained by including dispersion and ZPE corrections will be not further commented on in the text.

In the framework of the single-determinant DFT approach, the antiferromagnetic coupling of the Fe atoms in the C-cluster has been treated within the spin-unrestricted broken-symmetry formalism (BS) introduced by Noodleman.[51] The resulting BS state is not a pure spin state but rather a mixed state in which the majority spin and minority spin are arranged either spin-up or spin-down to give a spin coupling pattern with the correct net total spin and an overall either antiferromagnetic or ferromagnetic alignment. To construct a desired BS state, a calculation of the high-spin state is first completed, which is a pure spin state described by a single determinant with all unpaired electrons aligned in the same direction (α spins) to adopt the highest possible total spin state. Subsequently, groups of occupied α and β MOs of the high-spin state are exchanged in order to generate a guess of the spin-flipped state which is obviously submitted to a DFT energy minimization. For instance, in the Cred1 state a mixed-valence Fe(II,HS)Fe(III,HS) pair of spin S = 9/2 and a ferrous Fe(II,HS)Fe(II,HS) pair of spin S = 4 are coupled antiferromagnetically to give an overall low-spin ground state which exhibits an S = 1/2 EPR signal,[9,52] whereas the Ni atom is low-spin diamagnetic Ni(II).[10,15] The S = 1/2 BS state for Cred1 has therefore been obtained by manipulating the density of the highest possible total spin state, S = 17/2. Analogously, the desired BS states for the one- and two-electron-reduced states, Cint (S = 0) and Cred2 (S = 1/2), have been obtained from the S = 18/2 and S = 19/2 high-spin states, respectively. The resulting electronic structures of the six possible nonequivalent spin coupling schemes for the C-cluster (see Scheme S1) which satisfy S = 1/2 for Cred1 and Cred2 and S = 0 for Cint have been checked by computing Mulliken spin densities and NBO atomic charges.

Since the resolution of the X-ray structures of the enzyme is not sufficiently high to establish the broken-symmetry state of the C-cluster, the geometries of all investigated species have been optimized according to all possible spin-coupling schemes using the SM model. The geometries of the different broken-symmetry states are very similar, whereas their relative stabilities are strongly influenced by the electronic structure of the C-cluster. In particular, the spin alignment patterns in which the Fe1Feu pair is coupled antiferromagnetically with the Fe2Fe3 pair (BS-3 and BS-6 in Scheme S1) are always found to be of lower energy (see Table S4). On the basis of these results, geometry optimizations carried out using the large model of the active site have been performed by aligning the Fe site spin vectors according to only these two spin coupling schemes. All relative and reaction energies reported in this work have been calculated by considering the more stable BS state for each species involved. Analogously, geometries and electronic structures described in the Results refer to the more stable BS state.

Due to the uncertainty about the spin state of the EPR-silent Cint redox state, triplet (S = 1) and quintuplet (S = 2) spin states have been also considered for the more relevant species discussed in this work. Notably, as shown in Table S5, the singlet (S = 0) BS state is always more stable than higher-spin states (or at least isoenergetic with the triplet state) and, therefore, for the sake of clarity the latter states will be not further commented on in the text.

Nomenclature

In the following, computational models will be labeled according the general scheme RS-X-KM, where RS is the formal redox state of the C-cluster (C, C, or C), X is the specific chemical nature of the ligand(s) bonded to the active site, and KM is the size of the model (LM or SM). The protonation states of His93 and Lys563 are indicated by the superscript after the LM label; H+ and K+ denote the positively charged states of His93 and Lys563, respectively, whereas H0 and K0 denote their neutral form. Energies of all species investigated in this work are reported in Table S6 in the Supporting Information.

Results

Numerous X-ray structures of Ni-CODHs have been reported so far. Among these, two high-resolution crystal structures (PDB codes: 3B52 and 4UDX) feature a CO2 molecule bound to the C-cluster.[5,19] In both structures, the carboxyl carbon atom of CO2 is coordinated to the Ni ion (η1(C) coordination), completing its distorted-square-planar geometry, whereas one carboxyl oxygen is bound to the Feu metal (η1(O) coordination), resulting in a μ2-η2(C,O) binding mode of CO2 bridging the Ni-Feu site (henceforth referred to as simply μCO2). Although this is the only coordination mode that has been experimentally observed, in order to exhaustively explore the binding and dissociation mechanism of CO2 to and from the C-cluster, other coordination modes have been investigated. Indeed, a new species in which CO2 is terminally bound to the Ni atom through the carbon atom (hereafter referred to as tCO2) has been identified as a genuine minimum on the PES. It may play a key role in the binding/dissociation of the CO2 molecule to the C-cluster. In addition, the possibility that CO2 may bind to the C-cluster according to an associative mechanism, with a binding mode similar to that observed for the n-butyl isocyanate inhibitor in the X-ray structure of C. hydrogenoformans CODH (PDB code: 2YIV),[21] prompted us to investigate the binding of CO2 to the Ni ion in the presence of a hydroxide ligand bound to Feu. Such different binding modes of CO2 have been investigated for the Cred2, Cint, and Cred1 redox states of the C-cluster with different protonation states of the nearby His93 and Lys563 residues to disclose the crucial role of the protein environment in assisting the CO2 binding.

μ2-η2(C,O) binding of CO2 to the Ni-Feu Site of the C-Cluster

The 3B52 CO2-n class="Chemical">containing crystal structure has been solved at 1.5 Å from a sample poised at −600 mV (equivalent in its redox potential to the Cred2 state) in the presence of HCO3– as the CO2 source,[5] whereas the 4UDX structure, which was also determined at −600 mV, has been solved at 1.03 Å in the presence of HCO3–/CO.[19] Since the latter structure was not yet available when we started our study on Ni-CODH, the first structure was used as the starting geometry of the C-cluster for DFT calculations.

Notably, geometry optimization of the large model LMH+,K+ of μCO2 in the Cred2 state does not lead to a significant structural rearrangement of the C-cluster (see Figure a and Table S7). However, the optimized structure differs from the starting structure in the position and in the geometry of the bound CO2 molecule. The predicted Ni–C distance (1.86 Å) is shorter than that found in the 3B52 structure (1.96 Å), whereas the C–O1 (hereafter, O1 refers to the carboxyl oxygen atom of CO2 bound to Feu) distance (1.32 Å) is longer than the corresponding experimental value (1.25 Å). In the optimized structure the bound CO2 molecule is also slightly more bent; the O1–C–O2 angle (hereafter, the O2 label refers to the nonbridging oxygen atom) of 122.2° is about 10° smaller than in the 3B52 structure (132.6°). Nevertheless, it should be noted that the geometry of CO2 predicted by our calculations is very similar to that found in the true atomic resolution 4UDX structure. In the latter, the Ni–C bond (1.81 Å) is substantially shorter than that determined earlier (1.96 Å), whereas the two C–O bonds are considerably elongated (C–O1 and C–O2 distances are 1.32 and 1.30 Å, respectively). The O1–C–O2 angle (117.2°) is instead 15° smaller than that estimated in the 3B52 structure. All of these values are better reproduced by our calculations. In addition, the predicted distances between the CO2 oxygen atoms and the H-bonded nitrogen atoms of His93 and Lys563 are in better agreement with those found in the 4UDX structure. Indeed, the computed O1–N(Lys563) and O2−εN(His93) distances are 2.82 and 2.56 Å, whereas they are equal to 2.72 and 2.70 Å in the 4UDX structure.

Figure 2

Schematic representation of the geometries of the μ-CO2-bound form of the C-cluster in the Cred2 state optimized using the (a) LMH+,K+, (b) LMH0,K+, (c) LMH+,K0, and (d) SM models. Selected interatomic distances are given in Å. Aliphatic hydrogen atoms are not shown.

It is interesting to note that no significant structural differences have been found in the CO2 geometry of the CO2-bound C-cluster in Cint and Cred1, with respect to the Cred2 state. On the other hand, substantial geometry changes have been determined for the [NiFe4S4] cluster and for the side chain of the Cys526 residue (see Table S7). In particular, oxidation of Cred2 to Cint and then to Cred1 leads to the progressive contraction of the C-cluster with the sulfur atom of Cys526 approaching the Feu ion, as indicated by the Ni–S4 and Feu–S(Cys526) distances respectively equal to 3.59 and 3.48 Å in Cred2, 3.53 and 3.38 Å in Cint, and 3.27 and 2.65 Å in Cred1.

In order to evaluate the effect of the protonation state of the His93 and the Lys563 residues on the bonding mode of CO2 to the active site, we have also considered μ-CO2-bound large models in which His93 and Lys563 are modeled as neutral residues (LMH0,K+ and LMH+,K0 large-size models, respectively), as well as the SM small-size model in which such residues are not considered at all (see Figure b–d). We found that the geometry of CO2 as well as of the [NiFe4S4] cluster is only slightly affected by the second coordination sphere. On the other hand, the comparison of the crystallographic O2−εNHis93, O1−εNHis93 and O1−εNLys563 distances with those predicted using the LMH+,K+, LMH0,K+, and LMH+,K0 large models (see Table S7) clearly suggests that in 3B52 and 4UDX His93 and Lys563 are both positively charged. Indeed, these distances are well reproduced in the μCO-LM species, whereas εNHis93–O1 and εNHis93–O2 distances in μCO-LM and the O1–NLys563 distance in μCO-LM are much larger than those found in the two crystal structures. The increase in such distances is the direct consequence of the repulsive interaction between the electron density of the CO2 oxygen atoms and either the lone pair of εNHis93 in μCO-LM or that of εNLys563 in μCO-LM. The movement of neutral His93 and Lys563 away from the CO2 ligand leads to a slight rearrangement of the residues directly H bonded to them: namely, Asp219 and Gln332. No appreciable differences are instead predicted in the geometry of other residues. Superimposition of the geometries of the Cred2-μCO2 species optimized with the LMH+,K+, LMH0,K+, and LMH+,K0 models clearly show such effects (see Figure S1a). A large distortion for the [NiFe4S4] cluster has been instead observed when the small model is used (see Figure d). For the large models, oxidation of the μ-CO2-bound C-cluster leads to a progressive contraction of the [NiFe4S4] cluster and the movement of Cys526 toward Feu, but the Ni–S4 and Feu–S(Cys526) distances predicted with the SM model are significantly shorter than those calculated using the large models and those of the X-ray structures.

The validity of our DFT models for the prediction of the geometry of the Ni-CODH active site has been checked by calculating the root-mean-square deviation (RMSD) values of the [NiFe4S4]-CO2 cluster extracted by theoretical and experimental geometries (see Table S8). The RMSDs calculated using the large models of the active site are significantly lower than those computed using the small model, confirming the key role of the protein environment in tuning the geometry of the active site and justifying the large size of the model used in this work. Furthermore, it should be noted that the RMSD values between the DFT models of the μ-CO2-bound C-cluster and the true atomic resolution 4UDX crystal structure are generally smaller than those computed with respect to the lower resolution 3B52 X-ray structure, despite the fact that the latter has been used as the starting geometry for our DFT calculations.

A comparison of the RMSDs calculated for the [NiFe4S4] core and the CO2 ligand separately confirms that, as discussed above, the protein environment mainly affects the geometry of the metallic cluster rather than the geometry of the bound CO2. The very similar RMSD values for LMH+,K+, LMH0,K+, and LMH+,K0 show instead that the protonation state of His93 and Lys563 does not affect the geometry of the [NiFe4S4]-CO2 cluster (see Table S8). On the other hand, RMSDs calculated also including the side chains and the Cα atoms of His93 and Lys563 clearly show that in the 4UDX X-ray structure His93 and Lys563 are both positively charged; the RMSD for LMH+,K+ is significantly lower than those of LMH0,K+ and LMH+,K0 (see Figure S2). Calculations of RMSD values among the structures of Cred2-μCO2 optimized with the LMH+,K+, LMH0,K+, and LMH+,K0 models including all the (non-hydrogen) atoms of the models confirm instead that the greater difference among such models is the geometry of His93 and Lys563. Indeed, despite the greater number of atoms involved in the calculation, they are smaller than or very similar to those calculated only on the CO2-bound C-cluster and the His93 and Lys563 residues (see Table S9). Finally, the RMSD value computed for the LMH+,K+ CO2-bound C-cluster in the Cred2 state is lower than that calculated for the Cred1 state, supporting the assignment of the 3B52 and 4UDX structures to the Cred2 state (see Table S8).

The bent geometry of the bound CO2 molecule and the elongation of both C–O bonds is a clear manifestation of the reductive activation of CO2. An electron transfer from the C-cluster to the CO2 ligand is confirmed by the partial NBO charge of the bound CO2 (see Table ). In the μCO-LM species, the charges of CO2 in the Cred2, Cint, and Cred1 states are equal to −1.03, −0.98, and −0.91, respectively. The decrease in the CO2 negative charge in species in which His93 and Lys563 are deprotonated (LMH0,K+ and LMH+,K0) or absent (SM model) highlights the effect of the protonation state of these residues in the electron transfer from the C-cluster to CO2.

Table 1

Computed NBO Charges for the CO2 Molecule Bound to the C-Cluster in μCO2, tCO2, and CO2-OH, Calculated Using the LMH+,K+, LMH0,K+, LMH+,K0, and SM Models of the Active Site

	μCO2			tCO2			CO2-OH
	Cred2	Cint	Cred1	Cred2	Cint	Cred1	Cred2	Cint	Cred1
LMH+,K+	–1.03	–0.98	–0.91	–1.00	–0.92		–0.94	–0.80	–0.67
LMH0,K+	–0.95	–0.88	–0.79	–0.73	–0.56	–0.37	–0.71	–0.61	–0.49
LMH+,K0	–0.96	–0.91	–0.84				–0.87	–0.81	–0.69
SM	–0.98	–0.91	–0.82	–0.89	–0.78	–0.63	–0.97	–0.86	–0.71

An analysis of the CO2 Mulliken spin population, which is always equal to zero (see Table S10), allows us to rule out the reduction of CO2 to the CO2•– radical and strongly suggests a bielectronic reduction to the formal diamagnetic CO22– species, as also indicated by previous calculations.[20] Such a bielectronic reduction is also indicated by the geometry parameters of the bound CO2. Indeed, the C–O bond lengths and the O1–C–O2 angle of the CO2 ligand coordinated to the large models of the C-cluster best match those of CO22– (1.32 Å and 118°, respectively). The CO2 charge of about −1, lower than the expected value of −2, can be attributed to the large charge delocalization provided by the covalent linkage between CO2 and the metal sites.

The reductive activation of CO2 corresponds to the oxidation of the C-cluster. A detailed analysis of the electronic structure of CO2-bound and unbound species allowed us to identify the metal sites that are mainly involved in the oxidation. In this respect, due to the well-known electron spin and charge delocalization in FeS clusters, an analysis of the electronic structure of the C-cluster can be still more insightful, considering the net charges of the two subunits {Fe1NiFeuS1S2} (hereafter labeled layer L1) and {Fe3Fe4S3S4} (hereafter labeled layer L2). L1 and L2 correspond respectively to the blue and red layers of the BS coupling schemes shown in Scheme S1. A comparison of the electronic structure of the C-μCO-LM model, corresponding to X-ray structures of the CO2-bound C-cluster,[5,19] and the unbound form of the C-cluster in the same redox state, C-LM, indicates that the oxidation of the C-cluster occurs predominantly on the L1 layer. Indeed, the charge of the L1 layer in C-μCO-LM is 0.69 more positive than that in C-LM. In particular, the atomic charge of Feu is significantly affected by the CO2 binding; in C-μCO-LM it is 0.18 more positive than that in C-LM (see Table S11). Conversely, atomic charges on Ni and other Fe atoms do not differ by more than 0.05. These results suggest that the CO2 binding to the Cred2 state of the C-cluster promotes an electron transfer from the Feu-containing layer to the CO2 ligand. Analogously, also in C-μCO-LM and C-μCO-LM the CO2 ligand is reduced mainly at the expense of the L1 layer of the C-cluster, and in particular of the Feu atom. A similar electronic structure is observed with the LMH0,K+, LMH+,K0, and SM models (see Table S11).

The effect of the CO2 binding on the electronic structure of the C-cluster is also evidenced by an inspection of the frontier orbitals of unbound and CO2-bound adducts of the C-cluster. In all redox states, indeed, binding of CO2 leads to a significant stabilization of both the HOMO and LUMO levels (see Table S12). The stabilization of the HOMO and LUMO indicates that, after the binding of CO2, the C-cluster is simultaneously more difficult to oxidize and easier to reduce. To confirm such a finding, energy differences between reduced and oxidized species (Ered– Eox) have been calculated for μCO2 adducts and unbound forms of the C-cluster. Such values are expected to provide an estimate of the propensity of an oxidized species to be reduced and, therefore, to follow the same trend of the experimental reduction potentials; the more negative the Ered– Eox value (the more positive the potential), the greater its tendency to be reduced. Notably, Ered– Eox values for μCO2 adducts are at least 0.5 eV lower than those calculated for the unbound forms of the C-cluster (see Table S13). For instance, in the reduction from Cred1 to Cint the Ered– Eox values for unbound and CO2-bound LMH+,K+ models are respectively −1.43 and −1.92 eV, whereas in the reduction from Cint to Cred2 they are +0.09 and −0.60 eV. These results clearly indicate that the binding of CO2 to the C-cluster promotes the reduction of the active site. This is consistent with electron transfer from the cluster to the CO2 ligand.

Finally, to understand how the protein environment and the redox state of the C-cluster tune the reactivity of the active site, the binding energies of CO2 in the Cred2, Cint, and Cred1 states have been evaluated using the different models of the active site (see Table ). Binding of CO2 in the μ2-η2(C,O) mode to the SM model is exoergic only in the case of the Cred2 state (−8.4 kcal/mol), whereas it turns out to be endoergic by 1.4 and 6.4 kcal/mol for Cint and Cred1, respectively. On the other hand, binding of CO2 to the LMH+,K+ model is always calculated to be energetically favored, with CO2 binding energies becoming less negative with the oxidation of the C-cluster from Cred2 (−35.4 kcal/mol) to Cint (−19.5 kcal/mol) and to Cred1 (−8.3 kcal/mol). The large difference between the binding energies calculated for the SM model and those calculated for the LMH+,K+ model highlights the fundamental role of the protein environment in tuning the stability of the μCO2-bound forms of the enzyme. The interaction of CO2 with the C-cluster is critically assisted by the network of H bonds formed near the active site. In particular, His93 and Lys563 may be crucial for the CO2 coordination. The elongation of H-bond distances between these residues and the oxygen atoms of CO2 with the oxidation of the active site from Cred2 to Cred1 (see Table S7) reflects the smaller stabilization of the CO2-bound adduct by the protein environment.

Table 2

Binding Energies (in kcal/mol) of CO2 to the C-Cluster and the OH-Bound C-Cluster (ΔE(μCO2), ΔE(tCO2), and ΔE(CO2-OH)) and Relative Stabilities of μCO2 and tCO2 (ΔE(μCO2) → ΔE(tCO2)), Computed Using the LMH+,K+, LMH0,K+, LMH+,K0, and SM Models of the Active Site

	ΔE(μCO2)			ΔE(tCO2)			ΔE(μCO2 → tCO2)			ΔE(CO2-OH)
	Cred2	Cint	Cred1	Cred2	Cint	Cred1	Cred2	Cint	Cred1	Cred2	Cint	Cred1
LMH+,K+	–35.4	–19.5	–8.3	–24.5	–8.8		+10.9	+10.7		+4.7a	+8.9a	+5.3
LMH0,K+	–19.6	–4.7	+11.4	–3.3	+3.5	+13.1	+16.3	+8.2	+1.7	+5.4a	+17.1a	+21.1
LMH+,K0	–34.8	–17.5	–2.6							+19.4a	+24.5a	+25.8
SM	–8.4	+1.4	+6.4	–13.3	–2.9	+4.9	–4.9	–4.3	–1.5	–19.6	–14.4	–5.0

Energy values also involve the reaction energy associated with the deprotonation of H2O by His93 or Lys563 and migration of the resulting hydroxide to the Feu site.

Notably, deprotonation at the εN atom of His93 (LMH0,K+ model) induces an unfavorable interaction between that atom and the O2 atom of CO2. This makes the binding of CO2 a less favorable process; in Cred2 and Cint it is still favored by 19.6 and 4.7 kcal/mol, respectively, whereas in the Cred1 redox state it turns out to be endoergic by 11.4 kcal/mol (even more endoergic than that computed with the SM model). On the other hand, the protonation state of Lys563 only slightly affects the CO2 binding, as indicated by the similar binding energies calculated for the LMH+,K+ and the LMH+,K0 models. The main stabilizing interaction in μCO2 adducts is therefore the H bond between the hydrogen atom bound to the εN atom of His93 and the O2 oxygen atom of CO2. Accordingly, the μCO-LM species have been always calculated to be more stable than the μCO-LM isomers by more than 8 kcal/mol. On the basis of these results it is also possible to suggest that the His93 residue, depending on its protonation state, may favor the binding or the release of the CO2 molecule from the C-cluster. CO2 can bind to the active site when His93 is doubly protonated, where the formation of two strong H bonds promotes the CO2 coordination. CO2 may be released instead when the histidine residue is deprotonated on εN. This conclusion is based on the plausibility of different protonation states for His93 during catalysis; this residue was in fact proposed to be involved in proton transfer from/to the C-cluster.[25,26]

Terminal Binding of CO2 to the Ni Atom of the C-Cluster

All species in which CO2 is terminally bound to Ni or Feu by one oxygen atom are unstable, resulting in CO2 release from the active site or isomerization to the more stable structure in which CO2 bridges the Ni-Feu site (μCO2 species) during geometry optimization. Conversely, species in which the CO2 ligand is terminally bound to the Ni ion through the carbon atom (tCO2 species) have been generally identified as stable isomers (see Figure ). When calculations are carried out using the SM model, the tCO2 species are even more stable than the corresponding μCO2 isomers by about 5, 4, and 2 kcal/mol in the Cred2, Cint, and Cred1 states, respectively. Notably, the relative stability is reversed in the case of the LM models, pointing out the effect of the protein environment in modulating the stability of the two isomers. The lower stability of tCO2 with respect to μCO2, due to the lower interaction between the oxygen atoms of CO2 and the nearby residues (see Figures a,b and 3a,b), is in agreement with the fact that only the bridging CO2 species has been characterized by X-ray studies.[5,19]

Figure 3

Schematic representation of the geometries of the tCO2-bound form of the C-cluster in the Cred2 state optimized using the (a) LMH+,K+, (b) LMH0,K+ and (c) SM models. Selected interatomic distances are given in Å. For the sake of clarity, aliphatic hydrogen atoms are not shown.

When both His93 and Lys563 are positively charged (LMH+,K+ model), tCO2 is a stable species only in the Cred2 and Cint redox states. Even though both Cred2-tCO2 and Cint-tCO2 are less stable than the corresponding μCO2 isomers by about 11 kcal/mol, binding of CO2 is still energetically favored by 24.5 and 8.8 kcal/mol, respectively. This result supports the hypothesis that the tCO2 species corresponds to a catalytic intermediate in which a first covalent interaction is established between CO2 and the metallic cluster. Notably, deprotonation of His93 (LMH0,K+ model) strongly affects the binding energies of tCO2 (as already observed for μCO2), as well as the relative stability of the tCO2 vs μCO2 isomers. tCO2 binding energies are equal to −3.3, 3.5, and 13.1 kcal/mol for the Cred2, Cint, and Cred1 states, respectively. This results in the tCO2 isomer being less stable than the μCO2 isomer by as much as 16.3 kcal/mol in the Cred2 state, whereas the energy difference decreases by about 2 kcal/mol in the Cred1 form (see Table ). Conversely, deprotonation of Lys563 (LMH+,K0 model) did not allow for the identification of the tCO2 adduct as a genuine energy minimum on the PES, for all of the redox states considered, since during geometry optimizations they isomerized to the μCO2-LMH+,K0 or the tCO2-LMH0,K+ species.

In addition, the geometries of the tCO2 adducts are significantly affected by the protonation state of His93. When both His93 and Lys563 are protonated, the Ni atom retains a square-planar arrangement of ligands, even if it is slightly more distorted than that observed in μCO2, whereas when His93 is neutral, the Ni atom adopts a distorted-tetrahedral geometry (see Figure and Table S14). The difference in the Ni geometries can be attributed to the different interactions established by CO2 with the protein surroundings. In the tCO species, the square-planar coordination of Ni is favored by the strong H-bond interactions between the O1 oxygen atom of CO2 and the positively charged His93 and Lys563. On the other hand, deprotonation of His93 generates a repulsive interaction that moves the lone-pair electrons of the O1 oxygen atom far away from the neutral N atom of His93 and therefore pushes the CO2 ligand to a more apical position. The rearrangement of CO2 and His93, due to the deprotonation of the histidine residue (see Figure S1b) results in large RMSD values between the tCO and tCO adducts (see Table S9).

In the tCO2 species, as observed for the μCO2 isomers, the bound CO2 molecule features a large negative charge, and the charge is transferred from the Feu-containing layer (see Table and Table S11). Interestingly, when His93 is doubly protonated, such a negative charge is very similar to that computed for the μCO2 isomers, indicating a formal reduction of CO2 to a carboxylate anion, whereas when His93 is singly protonated, the negative charge of CO2 is significantly smaller. In fact, the charges are −0.95 and −0.88 in C-μCO-LM and C-μCO-LM, respectively, whereas they are −0.73 and −0.56 in C-tCO-LM and C-tCO-LM, respectively. Deprotonation of His93 therefore promotes an electron transfer from CO2 to the C-cluster, possibly decreasing its tendency to be reduced and simultaneously increasing its tendency to be oxidized. A comparison of HOMO and LUMO energies in μCO2 and tCO2 species and calculation of the Ered– Eox values (see Tables S12 and S13) confirm this picture. When His93 is doubly protonated, HOMO and LUMO energies in tCO2 and μCO2 do not differ by more than 0.05 eV. Analogously, Ered– Eox values are almost identical (−0.60 eV for μCO2 and −0.59 eV for tCO2 in the reduction from Cint to Cred2). On the other hand, when His93 is singly protonated, the HOMO and LUMO in tCO2 are higher in energy than those in μCO2 by respectively 0.24 and 0.28 eV in Cred2, 0.36 and 0.33 eV in Cint, and 0.58 and 0.27 eV in Cred1. Accordingly, the Ered– Eox value of the tCO2 adduct is less negative than that of μCO2 by 0.35 eV in the reduction from Cred1 to Cint and by 0.28 eV in the reduction from Cint to Cred2. This different trend highlights the effect of the protonation state of His93 on the electronic structure of the C-cluster. When such a residue is doubly protonated, the tCO2 and μCO2 adducts have similar reduction and oxidation potentials, whereas when it is singly protonated, tCO2 adducts are more difficult to reduce and easier to oxidize in comparison to the μCO2 isomers.

It is also worth noting that, as observed for the μCO2 adducts, the energies of frontier orbitals and Ered– Eox values of tCO2 adducts are always more negative than those of the unbound forms of the C-cluster (see Tables S12 and S13). CO2 binding therefore promotes the reduction of the active site independently from the coordination mode of CO2 and the protonation state of His93. However, only when His93 is singly protonated, the isomerization from the tCO2 to the μCO2 adduct further promotes the reduction of the cluster. In this case, Ered– Eox energy differences calculated for μCO2 adducts are more negative (i.e., the reduction potentials are more positive) than those computed for tCO2 species. Consequently, the reduction of μCO2 requires potentials less negative than those needed for the reduction of tCO2.

Binding of CO2 to the Ni Atom of the OH-Bound Form of the C-Cluster

The binding of CO2 to the Ni atom of the C-cluster in which a hydroxide is terminally bound to the Feu atom is finally investigated. The resulting adducts, hereafter called CO2-OH, are stable complexes in all redox states, using both large and small models (see Figure and Figure S1 and Table S15). The binding mode of CO2 is very similar to that observed for the n-butyl isocyanate inhibitor in the 2YIV X-ray structure,[21] with the Ni atom featuring a distorted-tetrahedral coordination, which is in contrast with the square-planar geometry observed in the μCO2-bound C-cluster. In CO2-OH species optimized using the LMH+,K+ model, one oxygen atom of CO2 forms H bonds with His93 and the OH– ligand coordinated to Feu, which in turn is strongly H bonded to Lys563 (see Figure a). In species optimized using the LMH0,K+ and SM models in which the His93 residue is deprotonated at the εN atom (see Figure b) or absent (see Figure d), the same oxygen atom of CO2 only interacts with the hydroxide ligand bound to Feu. On the other hand, the H bond with Lys563 is lost in the CO2-OH adducts optimized with the SM and LMH+,K0 models (see Figure c,d). It is interesting to note that during the geometry optimization of C-CO-OH-LM a proton of Lys563 is transferred to the hydroxide ligand, leading to the formation of a water molecule still bound to Feu (Feu–OOH2 and HOH2–NLys563 distances equal to 2.17 and 1.56 Å, respectively).

Figure 4

Schematic representation of the geometries of the CO2-OH-bound form of the C-cluster in the Cred2 state optimized using the (a) LMH+,K+, (b) LMH0,K+, (c) LMH+,K0, and (d) SM models. Selected interatomic distances are given in Å. Aliphatic hydrogen atoms are not shown.

In order to calculate the binding energies of CO2 in the CO2-OH adducts, the OH-bound forms of the C-cluster have been characterized. In the case of the SM model, for which no second-sphere residues are included, a stable species with the OH– ligand terminally coordinated to Feu has been identified for all of the redox states investigated (Cred1, Cint, Cred2). Binding of CO2 to such species is always an exoergic process, being equal to about −20 kcal/mol in the Cred2 state and decreasing to about −5 kcal/mol in the Cred1 state (see Table ). Nevertheless, the results are very different when second-sphere residues are included in the LM models. In the Cred1 state the hydroxide ligand is still terminally coordinated to the Feu site, completing its distorted-tetrahedral geometry in accord with the crystal structures of C. hydrogenoformans CODH[5] and M. thermoacetica CODH/ACS[53] (structures 3B53/3B51 and 3I01, respectively). On the other hand, in Cint and Cred2 the OH– ligand is protonated to H2O that dissociates from the C-cluster. In particular, in the LMH+,K+ model, the proton is transferred from His93 and the dissociation of water is favored by the formation of a strong H-bond network with Lys563, His93, and a conserved water molecule, whereas in the case of the LMH0,K+ model the proton is transferred from Lys563, and the formed water molecule remains weakly bound to Feu (the Feu–OH2 distance is equal to 2.24 and 2.42 Å in Cint and Cred2, respectively). Interestingly, the OH-bound forms of the C-cluster optimized using the LMH+,K0 model converge to those optimized with the LMH0,K+ model through the transfer of the εN proton from His93 to the OH– ligand.

According to the results presented above for the LM models, only the Cred1 state is compatible with the OH-bound form of the C-cluster, since in the Cint and Cred2 states the hydroxide ligand dissociates as a water molecule. The absence of the hydroxide ligand in the Cred2 state is supported by spectroscopic experiments; the loss in Cred2 of the strong ENDOR signal observed for the Cred1 state was indeed attributed to the release of OH–.[54] Furthermore, Fontecilla-Camps et al.[14] noticed that the X-ray structures, featuring a nonbridging hydroxide ligand bound to the Feu atom, solved at −320 mV (3B53) and −600 mV (3B51) that have been previously assigned to the Cred1 and Cred2 states, respectively, are almost identical, suggesting that the latter is a mixture of the Cred1 state (3B53) and the CO2 adduct of the C-cluster in the Cred2 state (3B52).

Since the OH-bound form of the enzyme in the Cred2 and Cint states is unstable, binding of CO2 to the OH-bound C-cluster can be considered only for the Cred1 state for which, in contrast to the SM model, is always an endoergic process (+5.3, +21.1, and +25.8 kcal/mol for the LMH+,K+, LMH0,K+, and LMH+,K0 models, respectively). Instead, for Cint and Cred2, we can calculate the ΔE value associated with the formation of the CO2-OH adduct by the simultaneous binding of CO2 and OH– (see values indicated by footnote a in Table and Scheme S2), according to the reactions represented in Scheme A. In this case as well, the formation of the CO2-OH adducts is predicted to be an endoergic process for all of the protonation states investigated; even if the CO2-OH adducts are genuine minima on the PES, they are unstable with respect to the release of the CO2 and H2O substrates. However, the CO2-OH species may still be intermediates along the CO2 binding/dissociation pathways (see below). In this respect, it is also interesting to evaluate the ΔE value associated with the deprotonation of a water molecule and coordination of the resulting hydroxyl ligand to Feu in the tCO2 forms of the enzyme to give the CO2-OH adducts, according to the reactions in Scheme B. Such a process calculated for the LMH+,K+ model is endoergic by 7.1 and 2.2 kcal/mol in the Cred2 and Cint states, respectively, whereas it turns out to be exoergic by about 5 kcal/mol in the case of Cred1.

Scheme 2

Possible Mechanisms for the Formation of (a) C-CO-OH-LM, (b) C-CO-OH-LM, and (c) C-CO-OH-LM

As in μCO2 and tCO2 species, in CO2-OH adducts the bound CO2 molecule features a large negative charge (see Table and Table S16), indicating an electron transfer from the C-cluster to CO2. Notably, as observed for the binding of CO2 to the unbound C-cluster, the binding of CO2 to the OH-bound form of the enzyme in the Cred1 state promotes the reduction of the C-cluster and simultaneously makes its oxidation more difficult. The HOMO and LUMO in C-CO-OH are indeed lower in energy than those in C-OH by 0.42 and 0.26 eV, respectively, when His93 is doubly protonated, and by 0.39 and 0.29 eV when His93 is singly protonated (see Table S12).

Discussion

In light of the results presented above it is possible to propose a mechanism for the binding and the dissociation of CO2 to and from the C-cluster. However, such processes are differently affected by the protonation state of the surrounding residues and the redox state of the C-cluster. The two mechanisms are therefore discussed in different schemes.

Our calculations strongly suggest that the binding of CO2 to the C-cluster is energetically more favored when both His93 and Lys563 are in their positively charged form. Protonated His93 and Lys563, apart from stabilizing the resulting CO2 adduct through their interaction with the oxygen atoms of bound CO2, can also act as proton donors, as in the direction of the CO2 reduction (i.e., in the direction of CO2 binding) two protons have to be supplied to the active site. Our calculations, according to experimental data,[55] also indicate that CO2 binds preferentially to the Cred2 state of the C-cluster. Cred2 is indeed the redox state of the active site, which always features the highest affinity for CO2. In this state, binding of CO2 occurs according to a dissociative mechanism (i.e., CO2 binds to the unbound form of the C-cluster, after the release of possible other ligands). The associative mechanism (i.e., binding of CO2 to the OH-bound form of the C-cluster) is ruled out by the incompatibility of the Cred2 state with the OH-bound form of the C-cluster.

CO2 approaches therefore the active site in the Cred2 state via a hydrophobic tunnel apical to the Ni atom and initially binds to the terminal position of the Ni ion (−24.5 kcal/mol), where initial favorable interactions with protonated His93 and Lys563 are possible (see Scheme ). Once the C-tCO(H,K) intermediate is formed, it easily isomerizes to the more stable μCO2 adduct, which is strongly stabilized by the H-bond network with His93, Lys563, and a conserved water molecule (−10.9 kcal/mol). These two chemical steps are significantly less favored if the active site is oxidized to Cint or Cred1. However, binding of CO2 to the Cint redox state and subsequent isomerization to the μCO2 adduct are still energetically favored by 8.8 and 10.7 kcal/mol, respectively. Nevertheless, the formation of C-μCO(H,K) can be preceded by the reduction of C-tCO(H,K) to C-tCO(H,K). The CO2 binding step indeed induces a transfer of electron density from the cluster to CO2, that in turn promotes the reduction of the C-cluster, as clearly indicated by the significant decrease in the Ered– Eox value (from 0.09 eV in C(H,K) to −0.59 eV in C-tCO(H,K); see Scheme ).

Scheme 3

Schematic Representation of CO2 Binding to the CODH Active Site

In Cred2 and Cint, CO2 initially binds to the terminal position of the Ni atom of the unbound C-cluster. Subsequent isomerization yields the more stable μCO2 species (see red and yellow arrows). In Cred1, CO2 binds to the OH-bound C-cluster. Reduction and H2O release are required for the formation of μCO2 (see blue arrows). Red, yellow, and cyan backgrounds denote species in Cred2, Cint, and Cred1, respectively.

Differently from the binding of CO2 to the Cred2 and the Cint redox states, binding of CO2 to Cred1 should proceed by an associative mechanism. Our calculations indeed support the experimental assignment[5,13,53,54] of Cred1 to the OH-bound form of the C-cluster. Therefore, in this state, CO2 should bind to the Ni atom of the C-cluster when a hydroxide ligand is still coordinated to Feu. Such a step is endoergic by 5.3 kcal/mol. This value, however, is not sufficiently high to exclude the possibility that CO2 binds to Cred1. Accordingly, several experimental studies support the binding of CO2 also to Cred1, even if CO2 has been proposed to bind to the Cred2 redox state. Indeed, exposure to CO2 slightly affects the Cred1 EPR g values,[56] whereas exposure to the CO2 analogue and competitive inhibitor CS2 under reducing conditions leads to the disappearance of the Cred1 EPR signal and the slow formation of a novel signal.[1,55] Furthermore, exposing particular batches of CODH, which are unable to convert Cred1 to Cred2, to CO2/dithionite (but not dithionite itself) can “cure” such batches, allowing them to attain the Cred2 state.[57] It is difficult to envisage that CO2 has these effects unless it could bind to the enzyme when the C-cluster is in a state more oxidized than Cred2. Therefore, it has been proposed that CO2 binds Cred1 noncatalytically, perturbating its EPR signal but without accepting an electron pair from the enzyme.[57] The resulting species could correspond to the C-CO-OH (H,K) adduct. As shown in Scheme , conversion of such an intermediate in the μCO2 adduct requires the transfer of a proton from His93 to the hydroxide ligand that is released from the active site as a H2O molecule. Notably, such a step in the Cred1 state is energetically disfavored by 4.6 kcal/mol, whereas it is slightly exoergic in the Cint state (−2.2 kcal/mol). This result suggests that the formation of the μCO2 adduct (whose formation is essential for CO2 reduction) is preceded by the reduction of the C-cluster from Cred1 to Cint through the transfer of one electron from the auxiliary clusters. Studies of the dependence of CO2 reduction by R. rubrum CODH on the redox state of the C-cluster support this hypothesis; Cred1 is indeed not competent to reduce CO2, whereas its one-electron-reduced state is active for CO2 reduction.[58] Actually, the point that CO2 binds more strongly to the reduced form of the C-cluster can be equivalently investigated by looking at reduction propensities: the decrease in the LUMO energy of about 0.3 eV after the binding of CO2 (see Table S12) suggests that the CO2-OH adduct is more easily reduced than the OH-bound form, in accordance with the experimental observation in which the reduction of the C-cluster from Cred1 to Cred2 is strongly speeded up by the presence of the CO2 substrate.[57] The formation of C-tCO+HO (H,K) therefore follows a CEC mechanism. Finally, the tCO2 to μCO2 isomerization can take place before or after further reduction of the C-cluster (see Scheme ). Due to the rather complex picture described above for CO2 binding, we consider it useful to summarize in a bullet point list the most likely and most intriguing routes for CO2 binding. Such a list—included in the following—will contain most of the information coming from the colored arrows in Scheme , as they indicate steps that are relatively favored in the energy landscape composed by the intermediates investigated in the present contribution:

The most likely route for CO2 binding involves the Cred2 state featuring both His93 and Lys563 in their protonated (charged) states, after H2O has left the active site following completion of the previous catalytic cycle; as far as the regiochemistry of binding is concerned, CO2 may first bind as tCO2 and then evolve toward μCO2.

Intriguingly, CO2 may also bind to Cint, which would be more likely when oxygenic ligands are absent from the active site and, at the same time, both His93 and Lys563 attain their charged (protonated) state.

CO2 could even bind to Cred1; however, such a possibility depends on the occurrence of specific conditions, which are somehow likely to be associated with formation of states not necessarily functional for catalysis to occur.

As discussed above, the release of CO2 from the C-cluster does not occur at the same redox and protonation state of the CO2 binding. Indeed, dissociation of CO2 in the Cred2 state when both His93 and Lys563 are protonated (i.e., conditions under which CO2 binding is more favored) is a strongly endoergic process; the initial displacement of CO2 from the bridging position to the terminal position on the Ni atom and the subsequent CO2 dissociation are energetically disfavored by 10.9 and 24.5 kcal/mol, respectively.

Our calculations indicate that CO2 release can take place by following a reaction pathway that implies deprotonation of a nearby residue and oxidation of the C-cluster. Accordingly, in the direction of CO oxidation (i.e., in the direction of CO2 release) electrons and protons must be released from the active site. As shown in Scheme , after deprotonation of His93, the first step of CO2 dissociation involving the μCO2 to tCO2 isomerization is predicted to be energetically disfavored by 16.3, 8.2, and only 1.7 kcal/mol in Cred2, Cint, and Cred1, respectively. Conversely, deprotonation of His93 makes the release of CO2 from the tCO2 adduct slightly endoergic in Cred2 (+3.3 kcal/mol) and exoergic in Cint and Cred1 (−3.5 and −13.1 kcal/mol, respectively). On the basis of these considerations, we assume that the CO2 release occurs when His93 is singly protonated and the C-cluster is oxidized at least to the Cint redox state.

Scheme 4

Schematic Representation of CO2 Dissociation from the μCO2 Adduct of the Active Site in the Cred2 Redox State

The most plausible mechanisms are indicated by red arrows. Deprotonation of His93 promotes oxidation of Cred2-μCO2 to Cint-μCO2. The latter isomerizes to Cint-tCO2, from which CO2 can dissociate. Cint-tCO2 can also be oxidized to Cred1-tCO2, from which CO2 dissociation is more favorable. Species in the Cred2, Cint, and Cred1 redox states are denoted by red, yellow, and cyan backgrounds, respectively.

Notably, deprotonation of His93 can immediately induce the oxidation of the C-cluster from Cred2 to Cint. Proton-coupled electron transfers (PCETs) are indeed very common reactions in chemistry and biology to balance the charge of the system. The energetics of the C-μCO(H,K) → C-μCO(H,K) PCET oxidation is therefore calculated according to the procedure described in ref (20). In such a calculation, in which an (e–, H+) couple is removed from the active site, the experimental oxidation potential of −0.3 V[12] has been used and translated, by using the energy of a proton in water at pH 7, to 371.6 kcal/mol. Such a step is predicted to be energetically disfavored by 8.4 kcal/mol. This value, however, is not sufficiently high to exclude the feasibility of the PCET process. Accordingly, the μCO2 to tCO2 isomerization in the Cred2 state is predicted to be strongly endoergic (+16.3 kcal/mol) and, therefore, unlikely to take place. Such a step in the Cint redox state is still endoergic (+8.2 kcal/mol), but much less than in Cred2. Hence, C-μCO(H,K) is oxidized to C-μCO(H,K), which then isomerizes to C-tCO(H,K), from which CO2 can dissociate (−3.5 kcal/mol, see Scheme ). On the other hand, the possibility that a PCET oxidation occurs at the level of the Cint redox state is ruled out by the high endoergicity of the process (+23.2 kcal/mol). The μCO2 to tCO2 isomerization can, however, promote a further oxidation of the cluster to the Cred1 redox state. Indeed, as discussed above, when His93 is singly protonated, such a step results in the transfer of electron density from CO2 to the C-cluster that makes its oxidation more favorable. Calculations of Ered– Eox energy differences for μCO2 and tCO2 adducts confirm this picture. Since the oxidation and reduction potentials for a given species are identical in value but opposite in sign, –(Ered– Eox) values are expected to provide an estimate of the oxidation potentials. The –(Ered– Eox) value computed for C-tCO(H,K) is less positive (i.e., the oxidation potential is more positive) than that calculated for C-μCO(H,K) by about 0.28 eV (see Table S13). Oxidation of C-tCO(H,K) therefore requires potentials less positive than those needed for oxidation of C-μCO(H,K).

The release of CO2 from C-tCO(H,K) to give the unbound form of the C-cluster in a strongly exoergic process (−13.1 kcal/mol) is questionable. In the Cred1 redox state a H2O/OH– ligand is indeed expected to occupy the vacant site at the Feu atom that is created upon μCO2 to tCO2 isomerization. Interestingly, in the Cred1 state, the binding of a H2O molecule, initially H-bonded to His93 and Lys563, to Feu and the transfer of one of its hydrogen atoms to His93 to give a CO2-OH adduct is exoergic by 4.6 kcal/mol. In Cint and Cred2, such a process is instead endoergic by 2.2 and 7.1 kcal/mol, respectively (see Scheme ). Subsequent CO2 release from Cred1-CO2-OH, yielding the OH-bound form of the C-cluster in Cred1, is energetically favored by 5.3 kcal/mol.

As shown in Schemes and 4, the His93 residue is proposed to be involved in proton transfers from/to the C-cluster both in the binding and in the release of CO2 from the active site. In this respect, it is worth noting that the involvement of His93 in such processes is more likely than that of Lys563. Indeed, in all redox states investigated, the unbound form of the C-cluster in which His93 is deprotonated (LMH0,K+) is slightly more stable than the corresponding species in which deprotonation occurs at Lys563 (LMH+,K0) (5.7, 1.7, and 3.2 kcal/mol in the Cred2, Cint, and Cred1 states, respectively). Conversely, during geometry optimization of the OH-bound forms of the C-cluster in the Cint and Cred2 redox states with the LMH+,K+ model a proton is spontaneously transferred from His93 to the OH– ligand. On the other hand, optimization of Cred1-OH with the LMH+,K0 model results in the spontaneous transfer of a proton from His93 to OH– and then to Lys563. This process can be explained by an analysis of the geometry of the system. The removal from Lys563 of the proton pointing toward the NiFeu site indeed produces a strong electrostatic repulsion between the N atom of Lys563 and the oxygen atom of OH–. Analogously, if one of the other two protons of Lys563 is removed, the NLys563 atom is repulsed by lone pairs of nearby residues (in particular, either the O atom of the carboxamide group of Gln332 or the S atom of Cys333). To avoid the formation of such unfavorable interactions, the system prefers to deprotonate His93. Furthermore, the reciprocal orientation of OH– and His93 allows the formation of a strong H bond between the deprotonated εN atom of His93 and the hydrogen atom of the hydroxide ligand. All of these results strongly suggest that the deprotonation of His93 is more feasible than that of Lys563.

Conclusions

The disclosure of the stereoelectronic and catalytic properties of the active site of Ni,Fe-containing carbon monoxide dehydrogenases is important not only in the context of the efforts aimed at elucidating structure–function relationships but also for the development of bioinspired catalysts for CO2/CO interconversion that may be used for the removal of such gases from the environment. To contribute to such an effort, the mechanism of binding and dissociation of CO2 to/from the C-cluster has been investigated.

A comparison of results obtained using a minimal DFT model (metal ions and first coordination sphere) and a very large model of the active site (metal cluster, first and second coordination spheres; 270 atoms) highlights the crucial role of the His93 and Lys563 residues in tuning the coordination geometry of the C-cluster and the stability of CO2 adducts. His93 and Lys563 residues are also both shown to be involved in proton transfers from/to the C-cluster. Mutational studies on MtCODH[27] confirm the fundamental role of such residues. Indeed, the enzyme activity is significantly attenuated in mutants in which His93 and Lys563 are individually changed to alanine, whereas it is abolished in the double mutant. The latter result indicates that His93 and Lys563 are involved in catalysis of CODH serving the same function. Our calculations suggest, however, that the catalytic cycle more likely involves proton transfers from/to the C-cluster and His93. Still, in consideration mainly of the high flexibility and spatial extension of the Lys side chain, elucidation of the aforementioned reciprocal compensation of the latter residue for the former will require further theoretical investigations based on an explicit treatment of the dynamic properties of the active site’s second coordination sphere.

The protonation state of His93 is also predicted to highly influence the direction in which the CO2/CO interconversion occurs; the charged protonated form of His93 indeed favors the binding of CO2, whereas the neutral form of this residue, in which only the δN atom is protonated, promotes its release. The redox state of the C-cluster is also shown to affect the energetics of the chemical binding and dissociation of CO2. In particular, binding and release of CO2 are respectively favored by the reduction and oxidation of the active site.

On the basis of these results, a mechanism for CO2 binding and CO2 release to/from the C-cluster has been proposed. CO2 initially binds, according to a dissociative mechanism, to the terminal position of the Ni atom of the C-cluster in the Cred2 state when His93 is doubly protonated. Subsequent displacement of CO2 to the bridging position of the Ni-Feu site leads to the formation of the well-characterized CO2 adduct of the C-cluster.[5,19] Our calculations, however, also support the noncatalytic binding of CO2 to the OH-bound form of the C-cluster in the Cred1 state according to an associative mechanism.[57]

While binding of CO2 is a strongly favored process, the dissociation of CO2 from the active site according to both dissociative and associative mechanisms is predicted to be more complex. Oxidation of the C-cluster at least to the Cint redox state and the endoergic μCO2 to tCO2 isomerization are required. Accordingly, NMR and steady-state kinetic studies showed that the release of CO2 is partially rate limiting.[59] In such experiments, after the binding of CO to the C-cluster and its oxidation to CO2, bound CO2 is reduced back to CO that then dissociates from the active site. The rate of CO2 release has been shown therefore to be slower than the rate of cluster reduction.