Unraveling the strokes of ion channel molecular machines in computers

Voltage-gated ion channels are fascinating micromachines responsible for all electrical signaling in biology. In PNAS, Amaral et al. (1) use molecular simulations to show how the new NavAb sodium channel from Arcobacter Butzleri (2) moves between intermediate states similar to K⁺ channels, which means it is likely that these principles are generally valid for activation of voltage-gated ion channels.

The story of voltage-gated ion channels during the past decade is a beautiful example of experiments and computational studies working hand in hand to advance the state of science. In contrast to most other membrane proteins that are largely hydrophobic, the voltage-sensing domains of these proteins all have an α-helix (S4) with several charges that are responsible for the gating current that can be measured during activation of the channel (3). There has been vivid discussion about the molecular explanation for this process (4); does the segment rotate to alternate between conformations exposed to the intra- vs. extracellular side, does it rotate and tilt, or does it simply translate vertically when the electrical field changes?

MacKinnon’s determination of an X-ray structure of the open state of the Kv1.2 potassium channel in 2005 (5) made it possible to use Kv1.2 to bootstrap discovery of closed ion channel structures by combining molecular modeling with constraints based on cross-linking and other experiments. In only a couple of years, the first early attempts have converged into a remarkably similar view of the relaxed or closed “down-state” of the Kv1.2 voltage sensor (6–10) without any X-ray structures available yet. Other combined studies (11) have further been able to identify and model intermediate conformations that provide direct information about how this particular channel activates (Fig. 1), and, only a few months ago, Jensen et al. managed to simulate the entire actual gating process (12). All this has been a tremendous success of computational modeling, and the sliding helix is now the dominant model for gating.

Fig. 1.

Deactivation process for a voltage sensor from a voltage-gated ion channel (illustrated with data from ref. 11). From top to bottom, the sensor starts in a fully activated state corresponding to a depolarized membrane. As hyperpolarization is applied, the voltage sensor domain moves through at least two intermediate states in which the charged arginine side chains in the S4 helix (blue) move one position down for each (leaving the next arginine in the charge transfer center) before it reaches a fully relaxed down state. In the down state, a linker causes the voltage sensor domain to push inward on the pore domain (not shown), which in turn will close the pore.

However, none of it would have been possible without exceptional efforts in electrophysiology to derive the molecular restraints (13–15).

Yet, in the middle of this success, it is easy to forget that most of the attention has been focused on a fairly narrow class of potassium channels, in particular Kv1.2 and Shaker. Although the voltage sensors of many channels have similar sequences, there is huge functional divergence that gives them completely different roles in our cells.

Amaral et al. (1) explore a structure of a sodium channel, NavAb from the bacterium Arcobacter butzleri, recently determined by Payandeh et al. (2). This structure is particularly interesting because the ion channel pore itself is in a closed conformation whereas the four surrounding voltage sensors are almost in the open state—essentially, it appears to be an intermediate conformation corresponding to preopening. Amaral et al. (1) use this intermediate state as a starting point to drive the entire ion channel toward the open and closed states, based on previous X-ray structures and models for the Kv1.2 potassium channel.

It is still difficult to simulate the natural deactivation process, and excess hyperpolarization has to be used (12). However, the pattern of the hydrogen bond (or salt-bridge) network between the S4 helix and the rest of the voltage sensor has been shown to fully characterize the different states in previous studies (16), and, by using steered molecular simulation techniques, it is possible to systematically drive the channel between states without directly using specific target coordinates. Amaral et al. (1) use this to parameterize a reaction path with the fully relaxed and fully open conformations as end states, which results in a reaction coordinate R along which the channel transitions can be steered. The authors identify as many as six intermediate conformations along this pathway, they show that many of these are identified and transiently stable for several of the channels, and they illustrate how NavAb moves between them in sequence during activation. Even for a system as large as an ion channel, techniques like this make it quite straightforward to sample complex conformational transitions—although the scientific implications are far-reaching, the underlying simulations are within in reach for almost any modeling group in the world today. Related studies have been reported for voltage sensors in potassium channels (16), but, in the work of Amaral et al. (1), the initially closed ion channel pore follows the voltage sensor to achieve full opening, which is a remarkable illustration of the close interplay between these two domains.

NavAb rearranges and assumes intermediate conformations very similar to Kv1.2, with arginine side chains in the S4 segment moving across a hydrophobic region in a step-wise fashion. In itself, this provides strong support for the sliding helix with charges shielded from the membrane as the general mechanism for gating in cationic voltage-gated channels, but it also goes further to show that simulations today can be used quite reliably to study conformational transitions, in some cases even when the end states of those transitions might not be fully known (1, 16). In fact, the predicted open state conformations of NavAb from the work of Amaral et al. (1) are very similar to those of the homologous channel NavRh (itself a homologue to NaChBac) for the voltage sensor (17), and the pore domain to NavMs, without any information from these channels used to guide the simulations.

Biomolecular simulations are moving beyond qualitative conclusions, and the modeling of the transient conformations confirms that the activation occurs by charged residues in the S4 segment moving through a hydrophobic region called the catalytic or charge transfer center (18) in the middle of each voltage sensor, where glutamic acid side chains provide stabilization for basic residues in S4. The intermediate states correspond to the catalytic center successively being occupied by three arginines also found in S4 for potassium channels (R2, R3, R4) and finally a threonine (T111), where Kv1.2 would have a lysine. This makes it possible to characterize specific metastable conformations as well as extreme relaxed and open states. Further, by directly measuring the displacement of charges in the simulation, it is possible to calculate and predict the gating charge expected to occur in experiments. As this gating charge is straightforward to measure (19), the authors are able to show that only a complete transition between their most closed and the fully open state will generate a gating charge large enough to agree with experiments. The value of 12.08 e is exceptionally close to those in previous models for potassium channels, and means both types of voltage sensor effectively transport three unit charges from the inside to the outside of the cell membrane during activation, and the additional states observed must be transient intermediates on this pathway. This agrees very well with other recent studies of experimentally trapped intermediates, and is a historic nod to Zagotta et al. (20), who argued for the absolute necessity of intermediate states during voltage gating-based Markov models to explain kinetics of electrophysiology experiments in 1994.

With the ion channel pore moving from the closed NavAb X-ray structure to an open conformation in the fully activated voltage-sensor state, simulations like the study of Amaral et al. (1) directly make predictions about the opening mechanism. Amaral et al. (1) confirm early ideas of a linker between S4 and the S5 segment in the pore, which pushes on the pore domain in the relaxed state. When S4 translates vertically by as much as 15 Å, the linker in turn moves up almost 4 Å. A series of simulations with the voltage sensor constrained to successively more extracellular states suggests that the linker itself does not pull the pore open, but, by relaxing the inward pressure on S5, the pore domain is free to explore increasingly open conformations. This is closely coupled with motions of the S6 helix in the pore, but, in contrast to the potassium channels, where glycine and a Pro-Val-Pro motif lead to kinks in the S6 helix, the NavAb channel appears to maintain S6 mostly straight, but tilted. This prediction should hopefully be straightforward to test with cross-linking or metal-ion bridges in future experiments.

Although there are still some remaining differences in the exact amount of vertical translation for the S4 segment (approximately in the 8–15-Å range) for voltage gating in general, there is now remarkable consensus about the mechanism of gating and the relaxed down state of these channels, and that each arginine side chain has to move over a narrow hydrophobic zone before reaching the stable catalytic site position. This causes the central part of the S4 segment facing the catalytic site to adopt 3₁₀-helix conformation (21), and this structure moves in the helix sequence as S4 slides during activation (1, 21). Models based on simulations and experimental data are able to simultaneously explain both gating currents, contacts between residues reported in experiments, kinetic effects of mutations around the catalytic site, and why some of the gating residues transfer more charge than others. This reliability of models gives us great hope for the use of computational modeling of activation to better understand disease-related mutation and predict differences for a wide range of cationic voltage-gated ion channels. Although there is a large amount of hard work remaining in this exciting field, the paper by Amaral et al. (1) brings us one step closer to the end of the beginning.