Molecular modelling of GABA-AT reactivity: From small representative models to the full protein, from molecular mechanics to quantum chemistry, from static to dynamics

Understanding enzymes and their catalytic mechanisms is very important in order to develop more effective drugs having little to no side effects. The active site residues of an enzyme may only be responsible for the facilitation of the enzymatic reaction or may be responsible for holding the substrates in place or both. In order to decipher the catalytic behaviour of enzymes, different approaches such as QM, MM-MD, and QM/MM can be used and their results can be correlated. The quantum chemical approach is often the only reliable way to model chemical transformations. Unfortunately, the numerical solution of the SchrÖdinger equation becomes increasingly expensive in terms of computational power when size increases. Numerical procedures in quantum chemistry require the calculation of numerous integrals and the resolution of (non)-linear equations, thus limiting the size of the systems that can be modelled. As of today, only linear scaling approximate quantum chemical methods are capable of combining both the necessary speed and accuracy to model large biomolecular systems with reasonable CPU cost. SEBOMD (SemiEmpirical Born-Oppenheimer Molecular Dynamics) is a method which is based on the Born-Oppenheimer approximation and uses a QM semiempirical NDDO Hamiltonian to perform molecular dynamics. It uses the linear scaling Divide & Conquer (D&C) algorithm to perform MD simulations of molecular systems. Today, the current D&C implementation is not the most efficient to run on very large supercomputers due to the diagonalization process of the Fock matrix. Thus, the diagonalization procedure should be optimized to be able to run MD simulations of molecular systems containing several hundreds of atoms. The main aim of this thesis is to get a deeper understanding of the mechanistic insights of the reactivity and of the dynamics of the pyridoxal 5-phosphate (PLP) dependent enzyme gamma-aminobutyric acid aminotransferase (GABA-AT). GABA-AT catalyzes the reaction in which GABA is degraded to succinic semialdehyde. Reduction of GABA concentration in the brain is at the origin of many neurological diseases, and since GABA cannot cross the blood brain barrier, it cannot be directly administered. As a consequence, the inactivation of GABA-AT is considered as a valid target for the development of antiepileptic drugs. Because GABA-AT resembles many other PLP-dependent enzymes, understanding it could be of importance for the broad community of biochemists and computational chemists who study such class of proteins. Our work has consisted of five stages to pursuit the comprehension of GABA-AT, and by extension that of PLP dependent enzymes. First, the reaction and the preferred binding mode of the natural substrate GABA has been elucidated with different isomers by means of model systems with Density Functional Theory (DFT). The stereoelectronic differences between the studied structures enabled us to discuss the possible particularities of the transimination and HF elimination reactions in GABA-AT. Second, the dynamics and the behavior of the enzyme has been studied with MM-MD through the use of apoenzyme, holoenzyme and holoenzyme with an inactivator. Our results demonstrate that several active site residues play a key role, and their protonation states as well as that of PLP require special attention. Third, the effect of the active site residues in the inactivation mechanism has been investigated with the modeling of clusters at the QM level involving key residues. The biggest cluster has been modelled with 165 atoms and the effect of the solvent has been taken into account using implicit solvent. The inclusion of the active site residues within the models enabled us to emphasize the effects of these residues on the reactivity of GABA-AT. The presence of similar residues in many PLP dependent enzymes enabled us to provide some general aspects for modelling these enzymes. Fourth, new diagonalization routines for the SEBOMD approach implemented in the Amber suite of programs, have been incorporated using LAPACK and SCALAPACK libraries, tested and evaluated. Our results showed that the main restriction on the improvement of the D\&C algorithm is the diagonalization of the Fock matrix which is not efficient on parallel computers. However, when large number of cores were used in SEBOMD simulations of large systems, using LAPACK solvers with threads that are provided by the Intel MKL did improve the efficiency of the SEBOMD calculations. Fifth, reaction free energies of PLP containing systems have been investigated with SEBOMD simulations. Each molecular system contains 747 atoms and their reactivity has been simulated using umbrella samplings and the newly implemented diagonalization routines. This work is the first example of the SEBOMD simulations on chemical transformations, and provides some general insights on the capabilities of the SEBOMD methodology in modelling reactions and in computing reaction free energies.