PDF, English Restricted access: Repository staff only until 1 April 2026. Login+Download (81MB) | Terms of use

Abstract

In recent years, there has been a growing interest in the prediction of drug-protein binding kinetics in the field of drug design due to its importance for drug efficacy in vivo. However, experimental assays are typically expensive to conduct, and time-consuming and they do not directly provide mechanistic insights, highlighting the need to predict binding kinetics in silico. This thesis aims to develop a protocol to accurately calculate bimolecular association rate constants in a feasible computational time. The protocol consists of a computational framework that combines Brownian dynamics and molecular dynamics simulations to study binding pathways and binding mechanisms -conformational selection and induced fit- as well as to calculate the association rate constants under the assumption that a binding process consists of two stages: diffusional and post-diffusional steps. Brownian dynamics simulations are a computational technique to study the diffusion of molecules subject to Brownian motion, and are useful in estimating diffusional association rates and to record early-stage encounter complexes in protein-ligand systems. This simulation technique provides a computationally efficient alternative to molecular dynamics simulations, as they only use implicit solvent models and usually apply rigid body approximation. On the other hand, with molecular dynamics simulations, it is possible to simulate the motion of the atoms individually (i.e. full atomistic detail) and to use explicit water models to include water-mediating interactions. These simulations are more accurate (especially when dealing with short-range interactions) and allow refinement of the diffusional encounter complexes recorded, including conformational changes coming from induced fit mechanisms in the post-diffusional step. I designed the protocol to first record the ligand as close as possible to the active site of the protein with Brownian dynamics simulations. In the next step, molecular dynamics simulations in full atomistic detail and explicit solvent are conducted to accurately simulate short range interactions, water displacement from the binding pocket, conformational changes and the resolution of steric clashes. This approach leverages the strengths of both molecular simulation techniques while minimizing their disadvantages to consider induced fit effects on binding rates. I validated the accuracy of this framework with a variety of protein-ligand complexes of different sizes, chemical nature, flexibility and binding rates, and the results I obtained correlate with experimental data. The protocol also applies the Metropolis-Hastings algorithm to switch between conformers during the diffusion of the ligand in Brownian dynamic simulations. This method uses the information of Markov State Models to estimate the transition between conformers. The approach has been tested with HIV-1 protease and 8 of its inhibitors but did not reproduce differences in the association rate constants of the inhibitors. However, the correlation with experimental data improved after refining the encounter complexes with molecular dynamics simulations, indicating a major contribution of induced fit to the association rate constants. Moreover, I tested the protocol with the FKBP51 protein, which exhibits a strong gating effect against selective compounds. Results indicate an induced fit effect contribution on the binding process. Last, I ran Brownian dynamics simulations of the trypsin-benzamidine system with multiple crowding molecules. For these simulations, I implemented the Northrup-Allison-McCammon algorithm in a periodic box to calculate diffusional association rate constants. Results revealed that the presence of crowders triggers two different opposing mechanisms: enhancing of binding sampling and hampering of the diffusion of the ligand to its protein target. I have implemented the necessary steps for the protocol in a software toolbox named SDAMD and that is available as an auxiliary tool in the SDA (Simulation for Diffusional Association) software package for studying the binding of protein-ligand complexes.