THERMODYNAMIC INSIGHTS INTO PROTEIN, LIPID AND SMALL MOLECULE INTERACTIONS AT THE CELL SURFACE USING MOLECULAR DYNAMICS

Abstract

Molecular Dynamics is a powerful computational tool to study molecular interactions at the atomistic scale. When done correctly, MD simulations yield accurate kinetic and thermodynamics of molecular interactions with applications in protein-drug interactions, protein-cell membrane interactions, drug discovery, structure prediction and others. However, the simulations are limited by inaccurate or incomplete forcefields, and by computer resources. The research presented here spans across three different aspects of MD simulations: forcefield development, protein-drug affinity calculations, enhanced sampling using generative AI.In my thesis I present my work where I employed enhanced sampling methods to explore the free energy of ligand binding to the bacterial multi-drug exporter, EmrE which is a member of the multidrug resistance family of proteins. I show that EmrE has a cleft region between the monomers which is accessible to the membrane lipids. Using enhanced sampling methods I propose that bulky, lipophilic ligands should preferentially bind to EmrE via this portal connecting protein cavity with the lipid bilayer. I have also used experimental thermodynamic data to improve divalent cation (Be2+)-lipid interaction parameter for the CHARMM36 forcefield. The Be2+ ion causes Chronic Beryllium Disease (CBD) potentially by replacing Ca2+ in key proteinlipid interactions. The current CHARMM36 forcefield is not fully parameterized for Be2+ and overestimates the strength of interactions between ion, water and lipids. I used experimental heat data to parameterize the forcefield, thus reducing the overbinding and achieving the correct hydration number for both Be2+ and Ca2+. In another project I collaborated with experimental scientists on profiling the membrane interactions of two anti-psychotic drugs commonly used for schizophrenia, chlorpromazine and clozapine. Our results from biophysical methods - ITC, AFM, DSC - combined with all-atom MD simulations highlight the similarities and differences in their aggregation propensity, lipophilicity, electrostatic interaction, effect on membrane lipid order and thickness. We conclude that the two drugs interact differently and is dependent on the membrane lipid composition among others (neuronal membrane vs. simple model membrane). Finally, I propose that applying generative AI with pre-existing MD simulation data can improve efficiency of enhanced sampling and free energy calculations, allowing one to sample better from the distribution and potentially develop an iterative protocol that can be used to build a large free energy database of complex systems.