Previous work by Dr. Sílvia Osuna, supervisor of the current proposal among others, showed that explicit water MD simulations provide a greater discriminative power than analysis of static structures alone, provided by X-ray crystallography or quantum calculations. Nevertheless, a general limitation of MD simulations is the lack of sufficient phase-space sampling due to limited computational resources, which is especially the case for systems with a huge number of degrees of freedom like enzymes. Starting this previously generated simulation data, massive parallel independent trajectories were ran for each LovD system studied to reconstruct long time scale processes. Afterwards, a MSM was generated for each LovD system and check for convergence. A Hidden Markov models (HMM) was used to optimize both, the state decomposition and the transition probabilities between states, for optimal construction of the MSM. The Graphics Processing Unit (GPU)-based cluster at the host organization was used, together with the GPU version of the AMBER16 package, and the PyEMMA (
http://pyemma.org(odnośnik otworzy się w nowym oknie)) software for the generation, analysis, and validation of the Markov models. MSM unravelled the thermodynamics and kinetics of complex conformational rearrangements in wild-type and LovD variants, characterizing the enzyme most populated conformational states, as well as the kinetics for interconversion. Identification of active HMM state was performed by monitoring LovD active site distances, in particular the Lys79-Tyr188 distance. Distance analysis revealed and stabilization of the active HMM macrocluster, as the DE process evolved (WT -> LovD1 -> LovD3 -> LovD6 -> LovD9). Therefore, mutations introduced in LovD along the DE process exerted an stabilisation effect on its active conformations, increasing the population of the conformations preactivated for catalysis.
Markov State Models were also applied to the second proposed enzyme model, TrpS, a more complex α2β2 heterodimer enzyme system catalyzing the condensation of indole and L-serine to form L-tryptophan. In this case, MSM failed to reproduce the conformational dynamics of the open-to-closed COMM domain transition leading to improved catalysis in the evolved variants. Therefore, the contingency plan based on metadynamics was applied. In particular, metadynamics in combination with path-collective variables were used to recover the free-energy profiles associated with the TrpS and TrpS0B2 evolved variant open-to-closed COMM domain transition, at different reaction intermediates. Similar to LovD, energy profiles revealed an stabilisation or population shift towards closed/active COMM domain conformations as the enzymatic reaction progresses. Moreover, mutations introduced in the stand-alone TrpS0B2 variant were able to retain and improve the conformational plasticity observed in the WT TrpS, thus explaining the higher catalytic activity of the stand-alone variant.
The second research goal was to develop new computational tools to identify which residues affect the enzyme active site conformational dynamics, unveiling the most important positions for stand-alone enzyme activity and substrate scope. Computational methods, especially Molecular Dynamics (MD) simulations, are particularly useful to recover the motions of enzymes critical for its function. However, the large number of degrees of freedom present in biomacromolecules, including enzymes, hampers the extraction of essential mo-tions. In this work, statistical or machine-learning techniques were used to learn functional relationships from MD simulation data without requiring a detailed model of the un-derlying physics or biological relations. In particular, a series of machine learning methods (Random Forest, logistic regression, support vector machines, xgboost classifier, etc.) were used to find new patterns from MD simulation data. Feature extraction from the best classification model lead to an understanding of the conformational rearrangements along the experimental directed evolution process. Finally, data gathered from previous work packages lead to the proposal of new point mutations for LovD and TrpB to test experimentally for improved catalytic activity.
