QM/MM calculations reveal a bridging hydroxo group in a vanadium nitrogenase crystal structure

Nitrogen is present in the atmosphere in the form of dinitrogen and the element is essential for life. Only diazotrophs, however, possess the capability of activating and breaking the triple bond of dinitrogen, a reaction performed by the nitrogenase enzymes. Nitrogenases are metalloenzymes that catalyse the reduction of dinitrogen to two molecules of ammonia for each molecule of dinitrogen. Different forms exist that can be distinguished by the nature (or absence) of the heterometal that is present in the active site. Molybdenum nitrogenase contains an iron–molybdenum cofactor (FeMoco), vanadium nitrogenase contains an iron–vanadium cofactor (FeVco) and the iron nitrogenase contains an all-iron cofactor (FeFeco). The Mo nitrogenase is the most active and has been studied most extensively while the V nitrogenase has in recent years gained more attention following the discovery of its unique catalytic properties and recently the first crystal structure became available. Despite intense research efforts, the mechanism of biological nitrogen reduction is far from understood and fundamental information such as the substrate binding site is lacking for all nitrogenase forms. The chemistry of the nitrogenases differ, with V nitrogenase requiring more electrons and protons for N2 reduction and more H2 is produced per N2 reacted, 7 as compared to Mo nitrogenase. This is possibly a consequence of the heterometal but the crystal structure recently revealed that FeVco differs from FeMoco with respect to a bridging ligand where either a carbonate or nitrate ligand was found in place of sulfide S3A. Additionally, the Fe sites are more reduced in FeVco than FeMoco according to Fe XAS experiments and theoretical calculations, a consequence of the different heterometal. Recently, Sippel et al. described the isolation of a turnover state of V nitrogenase that they were able to crystallize. The 1.2 Å crystal structure revealed the same unusual CO3 2 /NO3 ligand but additionally, the S2B bridging sulfide was missing and a lightatom ligand had replaced it (Fig. 1). At this resolution it was not possible to distinguish between C, N and O. A carbon ligand can be plausibly ruled out and only NH and OH were considered as likely candidates for the light-atom ligand, as there is a strong hydrogen bond between the ligand atom and the carbonyl group of nearby residue Gln176. Electron density analysis also found evidence of a protonated ligand fitting better. Interestingly the S2B sulfide atom that the ligand had replaced could be found, likely in the form of an SH ion, 7 Å away. These results led the authors to propose a mechanism for nitrogen reduction, with the light atom ligand interpreted as an NH group. As the crystallographic study was not

Nitrogen is present in the atmosphere in the form of dinitrogen and the element is essential for life. Only diazotrophs, however, possess the capability of activating and breaking the triple bond of dinitrogen, a reaction performed by the nitrogenase enzymes. 1,2 Nitrogenases are metalloenzymes that catalyse the reduction of dinitrogen to two molecules of ammonia for each molecule of dinitrogen. Different forms exist that can be distinguished by the nature (or absence) of the heterometal that is present in the active site. Molybdenum nitrogenase contains an iron-molybdenum cofactor (FeMoco), vanadium nitrogenase contains an iron-vanadium cofactor (FeVco) and the iron nitrogenase contains an all-iron cofactor (FeFeco).
The Mo nitrogenase is the most active and has been studied most extensively while the V nitrogenase has in recent years gained more attention following the discovery of its unique catalytic properties [3][4][5] and recently the first crystal structure became available. 6 Despite intense research efforts, the mechanism of biological nitrogen reduction is far from understood and fundamental information such as the substrate binding site is lacking for all nitrogenase forms.
The chemistry of the nitrogenases differ, with V nitrogenase requiring more electrons and protons for N 2 reduction and more H 2 is produced per N 2 reacted, 7 as compared to Mo nitrogenase. This is possibly a consequence of the heterometal but the crystal structure recently revealed that FeVco differs from FeMoco with respect to a bridging ligand where either a carbonate or nitrate ligand was found in place of sulfide S3A. 6 Additionally, the Fe sites are more reduced in FeVco than FeMoco according to Fe XAS experiments and theoretical calculations, a consequence of the different heterometal. 8 Recently, Sippel et al. 9 described the isolation of a turnover state of V nitrogenase that they were able to crystallize. The 1.2 Å crystal structure revealed the same unusual CO 3 2À /NO 3 À ligand but additionally, the S2B bridging sulfide was missing and a lightatom ligand had replaced it ( Fig. 1). At this resolution it was not possible to distinguish between C, N and O. A carbon ligand can be plausibly ruled out and only NH and OH were considered as likely candidates for the light-atom ligand, as there is a strong hydrogen bond between the ligand atom and the carbonyl group of nearby residue Gln176. Electron density analysis also found evidence of a protonated ligand fitting better. Interestingly the S2B sulfide atom that the ligand had replaced could be found, likely in the form of an SH À ion, 7 Å away. These results led the authors to propose a mechanism for nitrogen reduction, with the light atom ligand interpreted as an NH group. As the crystallographic study was not Fig. 1 The newly discovered bridging ligand, shown in purple, as it appears in the crystal structure of the turnover state of VFe nitrogenase by Sippel et al. 9 The ligand atom has replaced a bridging sulfide that now appears close to the cofactor, probably in the form of SH À .
able to conclusively identify the nature of the ligand, it is of utmost importance to clarify the identity of this ligand conclusively. In this study, we carefully examine the OH and NH possibilities via quantum mechanics/molecular mechanics (QM/MM) calculations. As shown in Fig. 2, the light atom X bridges Fe atoms no. 2 and 6. It makes a strong hydrogen bond to the carbonyl group of the nearby Gln176 residue that appears in a different conformation compared to the VFe resting state structure (or the analogous Gln191 in MoFe protein). This strong hydrogen bond between the carbonyl oxygen and atom X allows one to deduce that the ligand must be protonated, likely by one proton.
The hydrogen bond between X and O Gln176 is strong: 2.51 Å in one instance of the active site pocket and 2.39 Å in the other. In this context it is worth noting that O(H)Á Á ÁO hydrogen bonds found in crystal structures of small molecules are in the range of 2.54-2.82 Å (heavy atom distances) while N(H)Á Á ÁO hydrogen bonds are longer, in the range of 2.75-2.98 Å. 10 The stronger hydrogen bonding of OH donor compared to NH donors would thus suggest OH as a more likely candidate for the XH ligand, but due to the nature of a metal-bridging NH/OH this is not so clear-cut as such a ligand could have unusual electronic properties, particularly seeing as the metal-sulfur clusters of these enzymes are unique. There are few examples of Fe-bridging NH ligands in the literature and the oxidation state of such a ligand is not clear (likely either a nitrene, HN 0 or nitrido, HN 2À ). A structural argument for distinguishing between OH or NH, in our view, necessitates quantum chemical calculations that are capable of both describing the unusual electronic structure accurately as well as accounting for the protein environment.
In a recent study we described a QM/MM model of the resting state of the MoFe protein of nitrogenase. 11 By combining brokensymmetry DFT calculations with the TPSSh functional 12,13 with a classical forcefield (CHARMM36 14 ) we demonstrated that our QM/MM approach allowed direct determination of the charge of the cofactor, that the cofactor is populating a specific electronic state and that the alkoxide group is protonated. This analysis was possible via structural comparison alone and indicates the accuracy of our theoretical approach as well as the high resolution (1.0 Å) of the experimental crystal structure 15 of the resting state.
Here we apply the same QM/MM approach to the VFe protein, using the 1.2 Å crystal structure of the ligand-bound state as a starting point (PDB ID: 6FEA). A 38 917 atom spherical QM/MM model was prepared as detailed in the ESI. † All heavy atoms were kept fixed at crystallographic positions during the preparation phase. QM/MM geometry optimizations (using TPSSh and CHARMM36 for QM and MM descriptions respectively; see ESI † for further information) were performed for both NH and OH ligands. As the cofactor redox state is unknown, but the spin-state was found to be S = 3/2, we considered two possibilities: same redox level as the resting state, labelled E 0 and a state oxidized by two electrons, labelled E ox . We note that resting state FeVco has a more reduced Fe oxidation state than FeMoco according to Fe XAS 8 and the cofactor charge could be defined as either [VFe 7 S 8 C(CO 3 )] 2À or [VFe 7 S 8 C(NO 3 )] À . The S2B-ligand substitution (at the Fe 2 -Fe 6 site) will affect the charge, as will the identity of the ligand at the Fe 4 -Fe 5 site. We considered both CO 3 2À and NO 3 À as reasonable possibilities as discussed in the original crystallographic study of the resting state. 6 CO 3 2À and NO 3 À are isoelectronic and likely to behave similarly but the difference in charge will certainly affect the electrostatic environment in the active site. Due to the already high negative charge of FeVco compared to FeMoco, a more negative charge than 2À or redox state beyond E 0 is highly unlikely. In fact, due to the high negative charge of the E 0 models, the NH group, when interpreted as a nitrido NH 2À group and including a CO 3 2À ligand, resulted in an electronic structure with unbound electrons (positive orbital energies), an obvious sign of something wrong with the model. This did not occur with models with an OH À ligand, likely due to the lower charge. Finally we note that calculating an oxidized state can also be interpreted as assuming a nitrene NH 0 group replacing S 2À instead of nitrido NH 2À . The QM/MM optimized geometries for all models are shown in Fig. 3 and structural parameters are tabulated in Table 1 (details of the QM region are present in table as well as ESI †). We start by discussing the effect of an NH or OH ligand on the hydrogen-bonding in the active site. An NH model with E 0 charge and a CO 3 2À ligand (Fig. 3a)   than the crystal structure) and gives short Fe-N bonds of 1.82 and 1.81 Å (compared to 2.01-2.08 Å in the crystal structure). We note in this context that distances involving light atoms bonded to heavy atoms have higher uncertainties than heavyheavy atom distances due to Fourier ripple effects. 16 This will also affect the hydrogen-bonding distance. The metal-metal distances, however, in contrast will be better resolved. The 0.12-0.13 Å Fe 2 -Fe 6 contraction due to an NH ligand that is predicted by our calculations, is uncomfortably large and should have been resolved in the 1.2 Å crystal structure, were it present. Previous MoFe crystal structures with resolution between 1.0-2.0 Å have shown that these Fe-Fe distances have an average deviation of only 0.03 Å and maximum deviation of 0.05 Å (see ESI †).
An NH model with NO 3 À instead of CO 3 2À as the other bridging ligand (Fig. 3b) gives a shorter N(H)Á Á ÁO distance (2.82 Å), an improvement, but also appears inconsistent with the crystal structure. When oxidized NH models ( Fig. 3c and d) are calculated with CO 3 2À and NO 3 À , the N(H)Á Á ÁO distance is reduced to 2.69 and 2.66 Å, respectively, which is closer to the experimental distance but still far off. This stronger hydrogen bond, however, also results in a weakening of the hydrogen bonding between His180 and Gln176 that does not fit well with the experimental structure (off by 0.5-1.0 Å). Furthermore, the Fe 2 -Fe 6 distance is still always too short and as seen in Table 1 the more oxidized charge results in an increased Fe 2 -Fe 3 distance of 2.81 Å (CO 3 2À ) or 2.79 Å (NO 3 À ) that seems incompatible with the experimental structure. The V-Fe 6 distances are also too long for all NH models. As there is very little observed change in the Fe-Fe and V-Fe distances when going from the resting state V-Fe crystal structure to the turnover V-Fe crystal structure, these changes are hard to accept. All things considered, ligand X in the crystal structure is highly unlikely to be an NH.
In sharp contrast to the NH models, the model with an OH ligand and E 0 charge ( Fig. 3e (Fig. 3f) gives a shorter O(H)Á Á ÁO Gln176 hydrogen bond but this disrupts the hydrogen bonding between His180 and Gln176. Seeing as CO 3 2À and NO 3 À are isoelectronic ligands, this reveals a sensitivity of the hydrogen-bonding network to the electrostatic environment in the active site. The more oxidized OH models ( Fig. 3g and h) result in even stronger O(H)Á Á ÁO Gln176 hydrogen bonding (in better agreement with the experimental distances) but this also disrupts the His180-Gln176 hydrogen bonding.
As with NH, the removal of 2 electrons results in too long metal-metal distances: Fe 2 -Fe 3 , V-Fe 6 and V-Fe 7 , making an oxidized state unlikely. These changes likely occur due to removal of electrons from metal pairs with delocalized electron character. The +Fe 2 -Fe 6 -Q 176 -X dihedral angle in Table 1 also reveals a slightly different orientation of the OH group in the oxidized models as well as the E 0 -NO 3 À model, compared to the E 0 -CO 3 2À model and the crystal structure.
We additionally performed QM cluster calculations where a continuum solvation model replaced the explicit protein MM environment and constraints were used to keep an approximate active site geometry. These results, shown in the ESI, † reveal that the cluster models are not capable of describing the local active site geometry well enough as the hydrogen bond between O(H)Á Á ÁO Gln176 is predicted to be considerably weaker than with the QM/MM models. Seeing as the QM/MM results revealed a sensitivity to the electrostatic environment, this is perhaps not surprising. The OH models still gave Fe 2 -Fe 6 distances closer to experiment than NH models. This demonstrates the utility of our QM/MM models for accurately describing molecular structure of the active site in nitrogenases.
In conclusion, our study clearly resolves the identity of the unknown ligand species in the new VFe crystal structure and reveals it to be a hydroxo group instead of an NH. While there is always uncertainty associated with distances in crystal structures, particularly involving light atoms we feel the data safely rules out all NH models based on deviations of light-light atom distances as large as 0.4-0.5 Å and in particular based on deviations of metal-metal distances of 0.12-0.20 Å. In addition to the OH assignment, a CO 3 2À ligand is preferred over NO 3 À in our calculations and only an E 0 charge state fits well, i.e. the model shown in Fig. 3e. The identity of the ligand species as an

Conflicts of interest
There are no conflicts to declare. Table 1 Structural parameters (distances in Å, dihedral angles in 1) of all NH and OH QM/MM models in Fig. 3 and crystallographic parameters for cofactors in different subunits of the protein. Also shown are total charges for the QM region and the cofactor charge. The QM region in all models contains the cofactor, singly protonated homocitrate (3À charge) and the truncated sidechains of residues Lys83D (1+ charge), Gln176D, His180D, Cys257D (1À charge), Arg339D (1+ charge), Lys361D (1+ charge) and His423D