Spin Polarization Reveals the Coordination Geometry of the [FeFe] Hydrogenase Active Site in Its CO-Inhibited State

The active site of [FeFe] hydrogenase features a binuclear iron cofactor Fe2ADT(CO)3(CN)2, where ADT represents the bridging ligand aza-propane-dithiolate. The terminal diatomic ligands all coordinate in a basal configuration, and one CO bridges the two irons leaving an open coordination site at which the hydrogen species and the competitive inhibitor CO bind. Externally supplied CO is expected to coordinate in an apical configuration. However, an alternative configuration has been proposed in which, due to ligand rotation, the CN– bound to the distal Fe becomes apical. Using selective 13C isotope labeling of the CN– and COext ligands in combination with pulsed 13C electron–nuclear–nuclear triple resonance spectroscopy, spin polarization effects are revealed that, according to density functional theory calculations, are consistent with only the “unrotated” apical COext configuration.


Enzyme purification and maturation
Heterologous production of HydA1 from Chlamydomonas reinhardtii in E.coli and artificial maturation of the H-cluster were based on a previously published protocol with minor modifications. 1  2. FTIR spectroscopy FTIR spectra were obtained using an IFS 66v FTIR spectrometer from Bruker Optics with a N2 cooled mercury cadmium telluride (MCT) detector. Sample preparations and measurements were carried out under anaerobic conditions. Samples were placed between CaF2 windows and placed in an air-tight sample holder. Spectra were recorded at 15 o C with 20 kHz velocity in double-sided forward backward mode and are the average of 1000 scans. Data processing was performed using home-written scripts in the Matlab® programming environment.

EPR and Pulse ENDOR spectroscopy
Field swept Q-band EPR spectra were obtained on a Bruker ELEXYS E-580 X-band spectrometer equipped with a home-built Q-band accessory with 10 W available power at 34 GHz using a home built resonator described earlier. 2 The spectra were recorded in pulsed mode using FID detection after a 1 s π/2 excitation pulse. After a pseudo-modulation transformation, the spectra obtained in this way are comparable to those obtained using CW EPR. 3 Cryogenic temperatures (10-20 K) were obtained using a liquid helium cooled Oxford CF935 flow cryostat. ENDOR experiments were performed with the same setup making use of a water cooled 400 W Bonn-Electronic RF amplifier with frequency range 0.1 -400 MHz (BSA 1025-400). A Trilithic TM H4LE35-3-AA-R high-power low pass filter (cut off frequency around 35 MHz) was used to suppress the "harmonics" of the 1 H ENDOR signals.
Absolute signs can be determined using nuclear relaxation effects such as in the "pulsed ENDOR saturation and recovery" (PESTRE) 6 experiment or using a Variable Mixing Time (VMT) 7 experiment.
Making use of ESEEM effects in Mims ENDOR, relative signs can also be obtained without application of an additional pumping pulse. 8 Scheme S1. Top: Pulse sequence of the Mims TRIPLE resonance experiment. Bottom: Resulting signals for a system of two I=1/2 nuclei. The "ENDOR" signal is recorded with RF1 on an off-resonance frequency. The two HFI couplings (1 and 2) have opposite sign. The RF1 pulse inverts the population of the  (2) transition in the -manifold. This reduces the population difference of the  (1) transition in the same () manifold but does not affect the populations in the -manifold. In the difference TRIPLE resonance spectrum, the  (1)  transition appears as a weak positive contribution and the pumped  (2)  transition as a strong positive contribution. A similar scheme is valid for the Davies TRIPLE resonance experiment.

DFT calculations using the truncated model
The H-cluster was modeled using a small structural model as in our previous studies, [9][10] where only the immediate ligands to the two Fep/d iron sites are included ( Figure 2 in main text). In the Hox-CO state these are four CO, two CN -, the ADT (-NH-) ligand, and the cysteine side-chain modeled as CH3-CH2-SH thiol, with the Fep-bound sulfur protonated mimicking effects from the [4Fe-4Fe]H subcluster. The structural optimizations were done in ORCA 4.11 using the BP86 functional and a def2-TZVP basis set [11][12][13][14] while the subsequent spin and EPR analyses was done using the B3LYP 15-18 functional using the same basis set in which relativistic effects were accounted for using the Zero Order Regular Approximation (ZORA) 19 . The calculations include atom pairwise dispersion correction [20][21] . The SOMO properties and the spin population representation in Figure 4 (main text) were generated using ChemCraft. 22

S4
To investigate the effect of Hartree-Fock (HF) exchange on the calculated HFIs, the EPR calculations were performed using functionals with different amounts of HF-exchange: BP86 (0%), TPSSh (10%), B3LYP manually set to 15%, and B3LYP using the default 20% HF-exchange (see Tables S1 and S2). All calculations predict negative HFIs for the basal 13 CNligands (indicated in blue) and a positive HFI for the apical 13 CO (indicated in red). The calculations also consistently predict a negative isotropic HFI for 57 Fed although both irons have a positive spin population (Figure 3 in main text). This can be explained since the calculated spin populations at the iron core are dominated by the d-orbitals which have no density at the iron nucleus. The 57 Fe isotropic HFI is determined by the s-orbital population which is only a fraction of the total and can be negative. The carbons of the basal CO/CNligands have no direct spin density from the SOMO (see Figure 3 main text). However, one can assume that the -bonds to the Fe core are polarized by the spin density in the Fe-dxy type orbitals of the SOMO much like the situation in aromatic organic radicals where the sigma bonds of the protons are polarized by the magnetic orbitals. 24 The g-values as well as the 57 Fe-HFI interactions are best reproduced by the B3LYP calculations. All functionals overestimate the magnitude of the 13 C HFIs but the ratio between 13 CNp, 13 CNd, and 13 COext is best reproduced in the BP86 calculation (with no HF-exchange). It should be noted, however, that the DFT method, in general, has serious and unintelligible deficiencies especially for treatment of open shell TMC systems and in the estimation of Fermi contact interactions. 25 The best and most stable results are obtained using a combination of a hybrid functional (such as B3LYP), a sufficiently large basis set (def2-TZVP) and relativistic corrections. 25 As will be shown below, this level of theory provides a reasonable approximation of the electronic structure of the iron core (i.e. g-parameters and 57 Fe-HFIs) also for the extended models including Broken Symmetry treatment.

DFT calculations using extended models
In order to investigate the effect of the [4Fe-4S]H subcluster on the spin distribution and EPR parameters of the H-cluster in the HoxCO state, we performed calculations on large models L-COa ( Figure S1) and L-COb including the cubane cluster. The coordinating cysteines were modeled as S-CH3 moieties. The EPR parameters were calculated for both the BP86 and the B3LYP functional as listed in Table S3. For all calculations, the broken symmetry configuration BS1 was selected since the L-COa model as well as the models including protein side chains (see below) automatically converge into this configuration. The L-COb model does not converge in this configuration. Therefore, for the EPR property calculations, the BS1 configuration was explicitly imposed during the calculations. For model "LCOb-opt" the H-cluster was separately geometry optimized in the BS1 configuration. Table S3. EPR parameters of the extended models L-COa (apical COext) and L-COb (basal COext) including the [4Fe-4S]H sub-cluster in the BS1 broken symmetry configuration ( Figure S1). The "LCOb-opt" model was geometry optimized in the BS1 configuration. The experimental g-values and 13 C-HFI interactions were obtained in the current study while the 57 Fe HFI for Fep and Fed were obtained from reference. 23 HFIs are given in MHz. For the B3LYP calculations the obtained EPR parameters are not fundamentally different from those obtained using the truncated models (S-COa and S-COb) as presented in Table S1 and S2. Clearly, the BP86 calculation completely fails to predict the correct g-values and 57 Fe-HFI. On the other hand, the HFI values for the 13 CO/ 13 CNligands are comparable to those obtained for the truncated models S-COa and S-COb.

Method
To assess the effect of the protein surrounding, we also ran calculations including the directly coordinating amino acid side chains as well as parts of the protein that have a steric interaction with the CN -/CO ligands on the Fed side (Ser232, Cys299, Lys358, Pro324 in Figure 1). Model FP1-COa contains the selected amino acid side chains. For model PF2-COa the protein surrounding is removed without geometry optimization. For model PF3-COa the positions of the coordinating CH3-S carbons of the cubane were fixed while the binuclear subcluster was geometry optimized. In this process the bond between the bridging thiol and Fep was elongated from 2.4 to 2.75 Å. In Table S4     The proximal CO has its stretch at 1963 cm -1 while the distal CO and the extraneous CO stretches are coupled (2011 and 1969 cm -1 ). Bottom: FTIR of Hox-CO using extraneous 13 CO. It can be seen that the COd and COext stretches are "uncoupled" leaving COext at 1991 cm -1 and COd at 1944 cm -1 as indicated by the arrows.