UC Davis

Current drug discovery efforts for non-addictive, chronic pain management are targeting human voltage-gatedsodium (hNaV) channels: pore-forming transmembrane proteins that evoke the fast action potential in excitable neuronal, cardiac, and skeletal cells. Genetic and preclinical target validation studies have identified hNaV1.7, hNaV1.8, hNaV1.9 channel subtypes as key proteins in pain signaling, however, attempts at selectively targeting these channels fall short due to non-selective binding to other hNaV channel subtypes and other ion channel families; non-selective binding can lead to cardiac arrest, paralysis and seizure. Peptide toxins originating from tarantula, spider, and scorpion have been identified as potent hNaV-inhibiting biologics, while recent cryo-EM structural images of peptide toxins bound to hNaV channels provides a rational, structural context for designing proteins and small molecules with improved selectivity and potency for channel modulation. Further, advances in both classic, physics-based and machine learning-based de novo protein design methods using Rosetta – a protein structure prediction and design software suite – enable a new avenue of drug design and virtual screening prior to experimental validation. This dissertation describes the implementation of Rosetta protein design, small molecule docking, and multiple state-of-the-art deep learning approaches for ion channel research. Chapter 1 introduces the importance of accurately modeling voltage-gated ion channels to generate functional hypotheses, with a comparison of cryo-electron microscopy structures to deep-learning methods. Chapter 2 describes the importance of modulating NaV channels for human health, how current deep learning-based protein design methods can be utilized to accomplish this task, and my assessment of targeting hNaV1.7 voltage sensing domain 2 using classical Rosetta methods and deep learning-based protein design methods. Chapter 3 details the performance of the Rosetta small molecule docking methods and of Chai-1 — a recent deep learning method that can simultaneously predict the channel conformation, small-molecule conformation, and their binding mode — for molecular structure prediction and drug discovery. The appendices provide examples of how Rosetta methods can be applied to other transmembrane protein projects.