A two state homology model of the hERG K+ channel: application to ligand binding.

Biorg. Med. Chem. Lett. 2005, (15), 1737

Ramkumar Rajamani, Brett A. Tongue, Jian Li, Charles H. Reynolds

The hERG K+ ion channel plays a critical role in repolarization of the cardiac action potential. Interference with the repolarization mechanism can lead to adverse clinical events such as cardiac arrhythmia, long QT syndrome and death. A number of known drug molecules have been shown to have high affinities for the hERG channel. Due to the potential for serious side effects due to hERG binding, prediction and mitigation of binding of potential clinical candidates at the channel has become a priority in the pharmaceutical industry.

The channel is a tetramer and each subunit has six transmembrane domains. Experimental mutational studies have shown that the probable binding site for drug molecules is a cavity under the selectivity filter in the S6 helix. The closest homologous protein structure available is the structure for a bacterial K+ channel KcsA. Earlier work included docking studies into a homology model derived from KcsA and pharmacophore and QSAR models. All of these studies were limited in their predictive value. The earlier studies also failed to account for the open and closed states of the channel.

The present paper is a step forward because it accounts for the flexibility of the hERG K+ channel. This is achieved by examining models derived from two protein structures, one for a K+ channel in an open state (MthK) and one for a K+ channel in a closed state (KcsA). Starting with the closed state (KcsA) as a reference, the S6 helices were translated outwards along the conserved glycine hinge to match the reference open state of MthK. The channel was then closed in 1°increments and subjected to a short protocol of heating (0.4 ps), equilibration (0.6 ps), and dynamics using the CHARMM force field (5 ps). A harmonic constraint of 24 kcal/mol/Å2 was applied on the C atoms during this process. This protocol relieved any unfavorable interactions arising from helix movement. Two states were picked for the study. The partially open state is a 10° translation of the S6 helix from the reference closed state and relieves the strain due to replacement of Thr of KcsA with aPhe656 of hERG. The fully open state is a 19° translation away from the closed state.

Thehomology models for the two states were used for docking studies using GLIDE. After docking, each ligand’s best pose was again minimized within the protein using a conjugate gradient minimizer, the OPLS-AA force field and GB/SA continuum water model. Additionally, residues within an 8 Å distance were also allowed to optimize during the minimization. This procedure modeled the presence of water (implicit) and protein flexibility. Finally the ligand was extracted from the minimized ligand-receptor complex and re-minimized in GB/SA to obtain reference energetics for the free state.

The LIE (Linear Interaction Energy) method was used to predict binding affinities. The authors found that using a single state for modeling and prediction of binding affinities gave a poor correlation with IC50s. Using a combined equation that accounted for binding at either the open or partially open state gave the following equation (Eq 1). An R2 of 0.82 was obtained for 27 compounds out of a test set of 32 compounds. The five outliers may possibly bind in states not accounted for by the 10° or 19° tilt(s)used in the present work. The vdw energy term is the primary contributor to the fit and this correlated with experimental studies wherein the hydrophobicity of the pore residue Phe656 and aromatic residue Tyr652 were found to be crucial for binding. The predicted binding modes of the ligands agreed with experimental findings.

pIC50combined = -0.163(∆vdw) + 0.0009(∆ele) (Eq 1)

In my opinion, this work represents a fairly rigorous treatment of a common problem in industry i.e. there is often very little receptor structure information available and one has to rely on homology models or even partial sequence alignments to predict secondary structure of a protein in order to attempt in silico modeling. This work seems to have some experimental validity and also provides some insights on how to mitigate hERG binding. For instance, one could add a group that would disrupt the hydrophobic pi-stacking with Phe656 or Tyr652.

1