DECIPHERING AND TARGETING TRANSMEMBRANE ALLOSTERIC SITES OF GPCRS: TACKLING CHALLENGES WITH PHYSICS-BASED METHODS AND ARTIFICIAL INTELLIGENCE

Peter Somto Obi

doi:10.7273/000006493

G protein-coupled receptors, a class of membrane-embedded proteins, constitute therapeutic targets for ~ 35 % of FDA-approved drugs. Recent breakthroughs in the structural biology of GPCRs reveal novel transmembrane lipid-facing allosteric binding pockets distinct from the orthosteric site. Interestingly, the ligands bound to these sites are significantly exposed to the surrounding lipid bilayer. For a ligand to bind to these lipid-exposed sites, it must first partition before diffusing laterally to the binding site. However, there is a lack of clear mechanistic understanding of how these sites function, including how these allosteric ligands access these binding pockets and the relative enthalpic and entropic contributions of the membrane lipids in access, binding, and stability at the transmembrane sites. Because these lipid-exposed sites are novel, there haven't been any large-scale studies characterizing them in different membrane-embedded proteins and how they compare to conventional pockets exposed to the aqueous bulk. Additionally, there is no information on how traditional metrics for estimating important drug discovery parameters, such as druggability for soluble proteins' binding pockets, translate to these unique membrane-embedded sites. Furthermore, predicting the permeability of transmembrane ligands, an essential part of in silico drug discovery workflows for developing compounds that bind to these lipid-exposed sites, is computationally expensive and time-consuming. All these hinder the rational, high throughput design of allosteric ligands, making the development process tedious and costly. The background, significance, knowledge gaps, hypothesis, and specific aims of the dissertation work are outlined in Chapter 1. Chapter 2 elucidates the binding process of the CB1R negative allosteric modulator (NAM), ORG27569, to its transmembrane allosteric site, including the role of membrane lipids in this process. Additionally, a retrospective analysis of the effect of ligand-lipid interactions was carried out using a series of the NAM's analogs. This was the first study to characterize the entire binding process of an allosteric ligand to a site embedded deep in the bilayer. We found that the membrane lipids contribute significantly to the binding and eventual stability of allosteric ligands and can also help improve druggability predictions at these lipid-exposed sites. Also, specific ligand-lipid interactions can affect measurable parameters such as binding affinity and cooperativity. Chapter 3 focuses on the differences between transmembrane sites and soluble proteins' binding pockets, the interplay between these properties, and druggability predictions. This is the first large-scale analysis of transmembrane binding sites and how they compare to soluble protein sites. We observed that these pockets have distinct properties, and the current metrics do not work well in predicting druggability for transmembrane sites. Chapter 4 implements a generative AI model, denoising diffusion probabilistic model (DDPM), to reduce the computational time for predicting the membrane permeability of drugs by one-third. By combining just one-third of the original umbrella sampling simulations and the DDPM, we could reproduce the solvation-free energy profile, orientation angle, and other important features identical to those obtained from traditional methods. This dissertation provides unique insights into the access and binding of transmembrane ligands to lipid-exposed pockets and the participation of membrane lipids in facilitating access and subsequent stability after binding, laying a framework for future rational design for transmembrane ligands. The differences elucidated between transmembrane and soluble protein binding sites and the subsequent effect on predicting druggability have huge implications for designing future drug discovery workflows. This work also showcases the potential of generative AI to accelerate drug discovery and development by reducing the required computational time for predicting the membrane permeability of drugs and biologically relevant chemical species.

DECIPHERING AND TARGETING TRANSMEMBRANE ALLOSTERIC SITES OF GPCRS: TACKLING CHALLENGES WITH PHYSICS-BASED METHODS AND ARTIFICIAL INTELLIGENCE

Files and links (1)

Abstract

Metrics

Details