Addressing Long-Standing Challenges in Computational Enzymology With Large QM-Cluster Models of the...

Hydrogenase enzymes play a crucial role in generating energy for microorganisms by catalyzing the reversible oxidation of molecular hydrogen to protons. This catalytic mechanism has been well studied using computational models of varying complexity, ranging from smaller QM-cluster models of the enzyme active site to QM/MM models that capture the full enzyme structure. However, differences among studies have produced conflicting predictions for the energetics of certain reaction steps. This work focuses on characterizing one step-a cysteine-histidine proton transfer of Desulfovibrio fructosovorans [Ni, Fe]-hydrogenase-using a series of QM-cluster models to explore how model design influences predicted reaction thermodynamics. The Residue Interaction Network-based ResidUe Selector (RINRUS) toolkit was used to systematically create QM-cluster models based on either inter-residue distances or contact metrics from the active site [Ni, Fe] cluster. It is shown that QM-cluster models can achieve reaction energy predictions comparable to QM/MM and "big-QM" models when active site models are designed based on inter-residue contact interactions and with careful consideration of charged residues. Distance-based residue selection, a common strategy for QM-cluster model design, is not as effective compared to the RINRUS rules-based residue ranking approach from inter-residue contact counts. Large differences between previously reported QM and QM/MM reaction energies are resolved with RINRUS-based models, even at a modest level of electronic structure theory (B3LYP with modified LANL2DZ(d) basis sets/effective core potentials on metal atoms and 6-31G(d')/6-31G on nonmetal atoms). Overall, this [Ni, Fe]-hydrogenase case study underscores the need for careful model design when studying complex biological systems and demonstrates how RINRUS can provide a framework towards addressing this challenge.