Investigation of kinetics and pathways of ligand binding to NiFe hydrogenases using molecular dynamics simulation and machine learning models

NiFe hydrogenases are ancient enzymes with the potential to be used as an efficient catalyst to produce H2 biofuel. This makes this family of enzymes very interesting since it can be used for industrial applications. However, some NiFe hydrogenases can be inhibited by common gas molecules in the atmosphere, such as O2 and CO, which can deactivate and damage the enzyme. Over the last years, there have been multiple studies carried out to engineer inhibitor-tolerant enzymes, and one of the strategies is to modify the access of the gas molecules to the catalytic site by tunnel engineering. Point mutations of the residues inside the tunnels can block or facilitate the movements of the gas molecules inside the tunnels heading towards the catalytic site. In this thesis, by utilizing unbiased molecular dynamics and an enhanced sampling method called τ-Random Accelerated Molecular Dynamics (τRAMD), we investigated the binding and unbinding pathways and kinetics of H2, CO and O2 in complex with 10 different mutants of NiFe hydrogenase from Desulfovibrio fructosovorans at positions 74 and 122. The resulting relative residence time values could successfully reproduce the ranking of the experimental residence time values which validated the simulations. Regarding the bottlenecks, it was found that the bottleneck between residues 74 and 122 can control the residence time values and the two residues can restrict the width of the bottleneck. It was also found that there is a total of 9 different tunnels inside the enzyme and the gas molecules can have different preferences in various mutants for using the tunnels during the unbinding events. We also realized that three of the tunnels (T1, T2 and T7) are significantly used more than the other tunnels, so we designated these two as “primary” and the rest as “secondary” tunnels. Across different mutations, we found that although the population of the primary tunnels is the same, the population of the secondary tunnels is different. We proposed that introducing mutations to block the primary pathways, in combination with mutations to open the secondary pathways used by H2, can be an efficient strategy to engineer inhibitor-tolerant enzymes. In the second study of this thesis, unbiased molecular dynamics revealed that the main bottleneck can have two main states, open and closed conformations which gave us an important insight into the dynamics of this bottleneck. In the last study of this thesis, a set of machine learning models was developed in order to automatically identify binding and unbinding pathways of gas molecules through the 9 different pathways available in NiFe hydrogenases. The τRAMD simulations from our first study was used as training data and also a second NiFe hydrogenase, the one from Megalodesulfovibrio gigas, was added to the test sets to test the generalizability of the ML models to other NiFe hydrogenases. The main purpose of these ML models was to provide a fast and accurate workflow to skip the manual visual inspection of trajectories, which is one of the few ways to identify pathways accurately and is very time-consuming. Overall, we believe the studies carried out in this thesis deepened our insight into the ligand transport in NiFe hydrogenases and provide a rational framework for protein engineering to design inhibitor-tolerant hydrogenases for sustainable energy applications.