**1. Introduction**

Voltage-gated ion channels play key roles in electrical signalling in cells. They function much like transistors—A change in the membrane potential opens the channel gate and allows passive diffusion of a selected type of ions such as Na+,K+, or Ca2+ across the cell membrane [1]. Dysfunction of ion channels due to mutations in the channel protein or environmental effects are associated with numerous diseases [2]. Thus ion channels are important targets for therapeutic drugs, and there is an ongoing interest in the pharmaceutical industry to find channel blockers with high affinity and specificity. Many toxins from marine animals bind to specific ion channels with high affinity [3,4], and therefore provide natural leads for drug development [5–10].

Once a toxin is identified as a potential drug lead for a target ion channel, more work needs to be done to improve its affinity and selectivity for the target. This is essential to reduce the dosage and avoid side effects that may arise from binding of the drug to unintended proteins. This may be achieved in the lab by creating analogs of the toxin through mutations of selected residues, and testing their affinity for various proteins. Such a trial and error approach could be very time consuming and success is not guaranteed. Provided the structures of the target proteins are available—either from X-ray diffraction or via homology modelling—one can alternatively use computational methods to construct accurate models for the channel-toxin complexes and predict the effect of the mutations in silico. Advances in crystallization of membrane proteins and computer hardware/software in the last fifteen years have made such a computational approach to drug design a distinct possibility.

Availability of a crystal structure of an ion channel is essential for computational studies of toxin binding. Channel models constructed in the absence of a crystal structure are not reliable enough to use in atomistic simulations. The first crystal structure of a bacterial potassium channel (KcsA) was determined in 1998 [11], followed by many others [12]. Of particular importance for toxin binding studies was the solution of the mammalian voltage-gated potassium channel Kv1.2 [13], which has enabled construction of homology models for other Kv1 channels. Sodium channels were relatively harder to crystallize. The first crystal structure for a bacterial voltage-gated sodium channel appeared only recently [14], and a mammalian one is yet to be solved. Unlike potassium channels where the pore domains of bacterial and mammalian channels are very similar, there are substantial differences between the two classes in sodium channels. Thus constructing homology models of the mammalian Nav1 channels from the bacterial crystal structure will be a more challenging task. As yet, there are no crystal structures for the calcium channels and the ligand-gated ion channels, which explains the current focus of the computational studies on the potassium and sodium channels.

The most important progress in computer hardware was the introduction of the cluster architecture and parallel computing, which brought supercomputing power to masses. This was an essential breakthrough because an accurate description of structure and dynamics of a complex system requires an atomic-level treatment via molecular dynamics (MD) simulations and sufficient sampling of the simulation system. Routine simulation of a protein system consisting of ∼ <sup>10</sup><sup>5</sup> atoms in the microsecond range would not have been feasible without the high-performance computing power afforded by the clusters. On the software front, MD programs and their associated force fields such as AMBER [15], CHARMM [16], and GROMACS [17] have been continuously improved since their inception. Used in combination with a docking program, MD simulations have the ability to produce accurate models of protein-ligand complexes [18]. Similarly, one can perform free energy simulations to predict the absolute free energy of binding for a given complex, and predict the change in the binding free energy due to a mutation in the complex near chemical accuracy [19–22].

Here we present a review of the computational methods used in construction of channel-toxin complexes, and calculation of absolute and relative binding free energies in such complexes. Because application of these methods to protein-peptide complexes are relatively new, we provide detailed examples from the potassium channel toxins ShK and *κ*-conotoxin PVIIA. Computational investigation of sodium channel toxins is just starting. Nevertheless, we give an example from *μ*-conotoxin GIIIA to illustrate how the binding modes in sodium channels differ from those in potassium channels.
