AI-Based Prediction of Ligand Binding to Kv7 Channels for the Design of Selective Therapeutics
Abstract
The Kv7 family of potassium ion channels, comprising five isoforms (Kv7.1 to Kv7.5), perform the crucial physiological role of conducting potassium ions and regulating cellular electrical excitability. Kv7.1, paired with the β-subunit KCNE1 (E1), is found in heart muscle cells and is vital for proper cardiac function. Kv7.2 and Kv7.3 form heteromeric channels in neurons, where they regulate neuronal excitability. Kv7.4 is located in cochlear hair cells, contributing to sound amplification, while Kv7.5 helps regulate excitability in smooth muscle tissue. Because Kv7 channels are widely distributed and involved in essential physiological processes, their dysfunction is linked to conditions such as long-QT syndrome (LQTS), epilepsy, hearing loss, and urinary incontinence. Despite their clinical relevance, no approved drugs directly target Kv7 channels. Effective treatments must be highly specific to each Kv7 subtype to be therapeutically beneficial while minimising side effects.
Towards the goal of developing of Kv7 subtype-selective compounds, we secured a large KAW grant (no. 2022.0105) worth 27.1 MSEK over five years, led by Prof. Lucie Delemotte at Scilifelab, Stockholm. This grant unites a wide range of academic expertise, including molecular modelling, electrophysiology, machine learning, and structural biology. Published work from the electrophysiology lab of A/Prof. Sara Liin at Linköping University, a co-investigator on the grant, has characterised compounds with selective effects on Kv7 channels. This includes studies on cannabidiol, estradiol, endocannabinoids, and polyunsaturated fatty acids. Our current work focuses on characterising the selectivity of 30+ chemical derivatives of these compounds and other known Kv7 activators: Zinc Pyrithione, QO-58, and maxipost. A key strategy driving chemical modifications is rational drug design, which aims to determine drug binding sites and optimise drug-protein complementarity. However, this has been difficult to do at scale due to our extensive compound inventory.
A revolutionary development in this field has been the creation of AI models that predict drug-protein binding, such as AlphaFold3 and Chai-1. Based on published literature and our own testing, these are substantially more accurate than previous docking-based approaches. Using these models to characterise binding poses a substantial computational cost, especially since we aim to predict how our compounds bind to several Kv7 channels to assess selectivity. The Berzelius supercomputer, equipped with state-of-the-art Nvidia GPUs, is highly optimised for these calculations. Predicted compound binding stability will be tested via molecular dynamics simulations using an existing allocation (NAISS 2024/3-36) in collaboration with the Delemotte lab, which specialises in this technique. Further, site-directed mutagenesis experiments in the Liin lab will be used to validate that residues predicted to interact with the proteins are indeed important for their binding. The Berzelius supercomputer will enable drug binding site characterisation for Kv7 channels, a key step toward developing targeted therapeutics.
