Structural modeling of hERG channel: Drug interactions using Rosetta

Human Ether-a-go-go-Related Gene (hERG) encodes a potassium-selective voltage-gated ion channel essential for normal electrical activity in the heart but is also a major drug anti-target. Genetic hERG mutations and blockage of the channel pore by drugs can cause long QT syndrome (LQTS), which predisposes individuals to potentially deadly arrhythmias.  However, not all hERG blocking drugs are pro-arrhythmic, and their differential affinities to discrete channel conformational states have been suggested to contribute to arrhythmogenicity. We used Rosetta electron density refinement and homology modeling to build structural models of open-state hERG channel wild-type (WT) and mutant variants (Y652A, F656A, and Y652A/F656A), and a closed state WT channel based on cryo-electron microscopy structures of hERG and EAG1 channels. These models were used as protein targets for molecular docking of charged and neutral forms of amiodarone, nifekalant, dofetilide, d/l-sotalol, flecainide, and moxifloxacin. We selected these drugs based on their different arrhythmogenic potentials and abilities to facilitate hERG current. Our docking studies and clustering provided atomistic structural insights into state-dependent drug–channel interactions l that play a key role in differentiating safe and harmful hERG blockers and can explain hERG channel facilitation through drug interactions with its open-state hydrophobic pockets.

Methods
Rosetta modeling of hERG in open and closed states: We used the Rosetta structural modeling software (Bender et al., 2016;Leman et al., 2020) and the CryoEM structures of a putatively open-state hERG (PDB ID: 5VA2) (Wang and MacKinnon, 2017) and a closed-state EAG1 (PDB ID: 5K7L) (Whicher and MacKinnon, 2016) as templates (Figure 1). Each structure was passed through the cryo-EM density refinement protocol in Rosetta (Wang et al., 2016) (Supplement Script 1). The lowest scoring density-refitted models were then used in RosettaCM (Song et al., 2013) to model the channel’s unresolved residues and atoms in the extracellular region (Supplement Script 2). We generated 10,000 structural models of both open and closed state and selected the top 1,000 from each for RosettaLigand modeling of hERG interaction with drugs (described below). The lowest energy structures were visually inspected before being selected for the docking study. UCSF Chimera’s rotamer tool was used to make the F656A, Y652A, and Y652A/F656A mutations based on the final wild-type open-state model.

RosettaLigand modeling of hERG interaction with drugs: We obtained the molecular structures of each drug from the ZINC (Irwin and Shoichet, 2005) and PubChem (Kim et al., 2019) databases. OpenEye OMEGA (OpenEye Scientific Software) was used to generate conformers for the drugs (Hawkins et al., 2010). At a physiological pH 7.4, each drug exists in a dominant ionized (cationic or zwitterionic) form. However, because the drug receptor site in the pore lumen region is hydrophobic, this may shift the ionization equilibrium. This indicates that we need to study both ionized and neutral forms of each drug when analyzing its interactions with the hERG channel (Table 1) (Chatelain et al., 1986;Cross et al., 1990;Kodama et al., 1997;Hille, 2001;Lemaire et al., 2011;Kazusa et al., 2014). Previous computational studies suggest that the cationic form of sotalol predominantly remains in an aqueous solution while the neutral form embeds into the membrane and interacts with the hERG channel (Yang et al., 2020;DeMarco et al., 2021), hence our inclusion of neutral drug docking results in the main text.

To uniformly and efficiently sample the pore region, drugs were placed at 10 different initial locations spanning the top and bottom of the pore lumen region and the four fenestration regions. As part of the standard Rosetta docking protocol, we set the initial random perturbation to a translation distance less than 5Å and the sampling radius to 5Å (Supplement Script 3). The details of the RosettaLigand docking algorithm have been described previously (Meiler and Baker, 2006;Davis and Baker, 2009;Lemmon and Meiler, 2012;Combs et al., 2013;Bender et al., 2016;Leman et al., 2020). A total of 100,000 docking models were generated for each drug and each protein. The top 10,000 were selected based on the total_score of the protein-ligand complex and then ranked by ligand binding energy represented by the Rosetta interface_delta_X score term. The top 50 most favorable interface score models were visually analyzed using UCSF Chimera (Pettersen et al., 2004). The representative poses were further analyzed using the Protein-Ligand Interaction Profiler (PLIP) (Salentin et al., 2015) web service.

Clustering analysis of docking results was done in R Studio by calculating a similarity matrix between all top 50 poses clustered based on a cutoff parameter and minimum cluster size parameter using equation (1) where z is drug center of mass (COM) position with respect to hERG SF Ca COM along the z-axis, l is length of the vector between end-point atoms of a drug molecule, and  is the polar angle of the drug end-point atoms’ vector away from the z-axis (Supplement Table 1 and Supplement Scripts 4 and 5).

This ensures invariance to the rotation around the z-axis and, therefore, can account for the 4-fold symmetry of the hERG channel. Unique clusters were then identified using K-means optimization initialized using the lowest interface score structures from each cluster as the cluster centers.

Percentage within hydrophobic pocket was calculated by proportion of poses of the top 50 models of each docking simulation with at least one atom positioned at or in the hydrophobic pocket of the hERG channel as visualized for each pose in Chimera (Figure 2). Percentage within closed pore of hERG channel calculated by counting number of poses of top 50 models for each docking simulation that are fully encapsulated within hERG channel pore or fenestration region (Figure 1D).

IUPAC drug nomenclature: Amiodarone, (2-butyl-1-benzofuran-3-yl)-[4-[2-(diethylamino)ethoxy]-3,5-diiodophenyl]methanone Nifekalant, 6-[(2-((2-hydroxyethyl)[3-(4-nitrophenyl)propyl]amino)ethyl)amino]-1,3-dimethylpyrimidine-2,4(1H,3H)-dione Flecainide, (RS)-N-(piperidin-2-ylmethyl)-2,5-bis(2,2,2-trifluoroethoxy)benzamide Moxifloxacin, 1-Cyclopropyl-7-[(1S,6S)-2,8-diazabicyclo[4.3.0]nonan-8-yl]-6-fluoro-8-methoxy-4-oxoquinoline-3-carboxylic acid Sotalol, (RS)-N-[4-[1-hydroxy-2-(propan-2  ylamino)ethyl]phenyl]methanesulfonamide Dofetilide, N-[4-(2-([2-(4-methane sulfonamidophenoxy)ethyl] (methyl)amino)ethyl)phenyl]methanesulfonamide