In search of a consensus model of the resting state of a voltage-sensing domain

Abstract

Voltage-sensing domains (VSDs) undergo conformational changes in response to the membrane potential and are the critical structural module responsible for the activation of voltage-gated channels. Structural information about the key conformational states underlying voltage-activation is currently incomplete. Using experimentally determined residue-residue interactions as structural constraints, we determine and refine a model of the Kv channel VSD in the resting conformation. The resulting structural model is in broad agreement with the results originating from various labs using different techniques, indicating the emergence of a consensus for the structural basis of voltage sensing.

Introduction

Voltage-gated ion channels are transmembrane proteins that control and regulate the flow of small ions across cell membranes. They undergo conformational changes in response to changes in the membrane potential, thereby allowing or blocking the passage of selected ions. Structurally, these channels are formed by four subunits surrounding a central aqueous pore for ion permeation. Each subunit comprises six transmembrane α-helical segments called S1 to S6. The first four α-helices, S1–S4, constitute the voltage sensor domain (VSD). VSD's respond to changes in the membrane potential by moving charged residues across the membrane field. Although there has been much progress over the last decade, atomic details of the voltage-sensing process are not known.

Three idealized mechanistic models have been proposed to describe the voltage-sensing motion in Kv channels (Tombola et al., 2005). In the helical-screw/sliding-helix model, the S4 segment is assumed to retain its helical conformation as its moves along its long axis (Ahern & Horn, 2004, 2005; Yarov-Yarovoy et al., 2006). In the transporter-like model, it is assumed that the translational movement of S4 is modest because the membrane field is focused over a small spatial region (Chanda et al., 2005). In the paddle model, the S3-S4 helix-turn-helix is assumed to undergo a fairly large displacement through the lipids (Jiang et al., 2003; Ruta et al., 2005).

Ultimately, to fully understand the mechanism of voltage activation, knowledge of the three-dimensional structure of a channel in its various functional states is needed. At a minimum, structures of the two main endpoints in the conformational transitions, the active and resting states, are required to begin to understand voltage sensing. Yet, even for voltage-gated K⁺ (Kv) channels, knowledge of those two conformations is currently incomplete.

Atomic resolution X-ray structures of the Kv1.2 and the Kv1.2/Kv2.1 chimera channels provide information on the active state conformation (Long et al., 2007). The available crystal structures show that the VSD is formed by four antiparallel helices (S1-S4), packed in a counterclockwise fashion as seen from the extracellular side. The first two arginine residues (R1 and R2) along the S4 helix are close to the membrane-solution interface, while the following two arginines (R3 and R4) are involved in electrostatic interactions with acidic residues in S2 and S3. Given the experimental conditions, it seems that the VSD in the X-ray structures is likely to correspond to a relaxed inactivated conformation that appears after prolonged activation (Villalba-Galea et al., 2008). While this process might involve additional structural changes, it is nonetheless reasonable to assume that the available X-ray structures are broadly representative of the active state without further information.

There is currently no atomic-resolution structure of a Kv channel in the resting state. This has motivated efforts aimed at translating the results from various experiments into structural information using modeling (Jiang et al., 2003; Lainé et al., 2003; Ruta et al., 2005; Posson et al., 2005; Chanda et al., 2005; Yarov-Yarovoy et al., 2006; Campos et al., 2007; Grabe et al., 2007; Lewis et al., 2008; Pathak et al., 2007). Despite their inherent approximate nature, such experimentally-constrained structural models can serve to provide a context for the rational interpretation and design of future experiments. The can also be used to interpret and validate future experimental structures targeting the resting state of the VSD.

Information about the conformation of the VSD in the resting state has come from a wide range of experiments, including mutagenesis studies (Starace et al., 1997; Starace & Bezanilla, 2001, 2004; Ahern & Horn, 2004, 2005; Grabe et al., 2007; Lin et al., 2010; Tao et al., 2010), cross-linking (Jiang et al., 2003; Lainé et al., 2003; Ruta et al., 2005; Campos et al., 2007), fluorescence (Pathak et al., 2007), resonance energy transfer (Cha et al., 1999; Chanda et al., 2005; Posson et al., 2005), and inhibitory toxins (Phillips et al., 2005b). Despite the wealth of experimental information, not all measurements can be easily translated into simple structural constraints. In that regard, experimental observations involving residue-residue interactions are of interest because they provide highly specific spatial constraints for the resting conformation of Kv channels (Campos et al., 2007; Lin et al., 2010; Tao et al., 2010). Engineered metal bridges are particularly informative because they involve strong chemically-specific interactions occurring between residues that are within atomic proximity from one another. Furthermore, the presence of a high affinity metal bridge indirectly implies that the interaction reliably reports the protein conformation, since large distortions would be expected to cause unfavorable strain energy that would result in a low affinity site.

Here, we review the available information on the resting state of the VSD and assess how its conformation is constrained by the available experimental data. We set out to explicitly simulate several of the key interactions associated with the resting state using molecular dynamics (MD). While MD simulations are limited by approximations, the approach enables an objective evaluation of how these interactions can contribute to restricting the conformation of the VSD. It is particularly important to consider models that treat each interaction individually with all atomic details to fully grasp its implications. This analysis converges towards an average computationally-derived consensus model that is consistent with a wide range of available experimental data. The overall conformation that satisfies all the available constraints appears to be defined within about 3 Å RMS (backbone), which indirectly reflects the semi-quantitative “resolution” of the available knowledge of the resting state as interpreted via MD simulations of atomic models. These results help better circumscribe the current views of voltage sensing and highlight the emerging consensus regarding the resting state conformation of a VSD in K channels.

Results

Our aim is to examine the structural implications from four experimental residue-residue interactions known to occur in the resting state conformation. Our starting point is the model of the Kv1.2 channel proposed by Pathak et al (Pathak et al., 2007), which was subsequently refined by Khalili-Araghi et al using all-atom molecular dynamics (MD) simulations of a membrane environment in explicit solvent (Khalili-Araghi et al., 2010). Previous calculations of the sensing charge using this model resulted in a value of 12-13 elementary charges per tetramer (Khalili-Araghi et al., 2010), consistent with experimental estimates observed in Shaker channels (Aggarwal & MacKinnon, 1996; Seoh et al., 1996; Schoppa et al., 1992). For the sake of simplicity, only a single VSD embedded in a solvated bilayer is considered. Four all-atom models of the VSD were constructed and simulated. In each model, specific site-directed mutations were introduced, and harmonic restraints were applied to steer the model toward a configuration in which the interaction is realized. The results are shown in Figure 1, and those interactions are discussed below. The overall deviations of the models are shown in Supplementary Figure S1.

Figure 1.

Instantaneous configuration of the VSD of the Kv1.2 channel before and after restrained MD simulations. In each panel, the transparent helices show the initial coordinates while the final coordinates are represented in solid colors. (A) Bridge I: residues I177C and R294C form a Cd²⁺-cysteine bridge between the S1 and S4 helices. (B) Bridge II: residues I230C and R294C form a Cd²⁺-cysteine bridge between the S2 and S4 helices. (C) Bridge III: residues I230D and F267D form a Mg²⁺-aspartate bridge between the S3 and S4 helices. (D) Bridge IV: residues E236 and R294K in S2 and S4 interact to stabilize the resting state.

Bridges I and II; Two Cd²⁺ bridges

Functional recordings of the gating current in Shaker have identified pairs of cysteine residues that are amenable to metal bridge formation via a cadmium (Cd²⁺) ion (Campos et al. 2007). The cysteine-cysteine Cd²⁺ bridge involves residues R362C (S4) and I241C (S1), corresponding to Kv1.2 residues R294C in S4 and I177C in S1. To examine the Cd²⁺ bridge between S1 and S4, a model was constructed by introducing the mutations R294C and I177C in the resting state model of the Kv1.2 VSD using the PsfGen module of the program VMD. The cysteine residues were introduced in the deprotonated form (carrying a charge of -1) and a Cd²⁺ ion was inserted. In the initial model, the Cβ atoms of these residues are 12.2 Å apart. However, the Metalloprotein Database and Browser (MDB) (Castagnetto et al., 2002) shows that cysteine pairs bridged by Cd²⁺ ions exhibit a Cβ-Cβ distance of roughly 5 to 7 Å. Therefore, the Cβ-Cβ distance was initially too large for a metal bridge to be formed. To form the metal bridge, a time-dependent harmonic restraint was applied over five nanoseconds between the Cd²⁺ ion and the sulfur atoms on cysteine residues R294C and I177C. The simulation was allowed to continue for an additional nanosecond to allow the protein to equilibrate. For the final nanosecond, the average Cβ-Cβ distance had been reduced to 7.8 Å, in accordance with the expected distance of a cysteine metal bridge as found in the MDB. The backbone RMSD (root-mean-square deviations) between the initial model and the equilibrated model is 1.7 Å as calculated in the transmembrane region, indicating that this constraint was satisfied with minimal rearrangement of the backbone atoms (see Figure 1A).

A second model was constructed according to the same method to reflect the observation of a separate Cd²⁺ bridge between residues R362C (S4) and I287C (S2) (Campos et al., 2007), corresponding to Kv1.2 residues R294C and I230C. The Cβ atoms of these residues were initially separated by a distance of 13.2 Å. Again, the metal bridge was formed by applying a harmonic restraint between the metal ion and the sulfur atom on the cysteine residues, and therefore bringing the Cβ atoms to within 6.2 Å. The backbone RMSD between the initial and final conformation in the transmembrane region is 2.7 Å, indicating that the interaction was formed with little movement of the protein backbone (see Figure 1B).

Bridge III; One Mg²⁺ bridge

Functional recordings of ionic currents have shown that magnesium (Mg²⁺) slows the kinetics of activation of the Shaker double mutant I287D in S2 and F324D in S3 (I230D and F267D in Kv1.2), while having little or no effect on deactivation (Lin et al., 2010). This behavior is reminiscent of the voltage-sensing K⁺ channel ether-a-go-go, which is known to contain a functional divalent cation binding site at those respective locations (Tang et al., 2000). Therefore, this site provides a separate set of constraints between S2 and S3 that must be satisfied in the resting state of the VSD. To examine the Mg²⁺ bridge between S2 and S3, a model was constructed based on the VSD of Kv1.2 where the I230D and F267D mutations were introduced, and a Mg²⁺ ion was inserted. In the initial model, the Cβ-Cβ distance between I230D-F267D is 5.2 Å, indicating that the Mg²⁺ bridge is readily satisfied without the need to alter the initial conformation of the VSD. Nevertheless, for methodological consistency, a six nanosecond MD simulation was performed. The average Cβ-Cβ distance of the final nanosecond of simulation is 6.5 Å and the backbone RMSD between the initial and final conformation is 1.2 Å (see Figure 1C).

Bridge IV; S2-S4 electrostatic interaction

Functional recordings of ionic currents show that the resting state of the Shaker double mutant F290W-R362K (F233W-R294K in Kv1.2) is more energetically stable than the resting state of the single mutant F290W (Tao et al., 2010). Although the nature of the interaction was not identified, one possibility is that in the resting state, the double mutant enables an electrostatic interaction to occur between R1 and the acidic residue E2, which is one turn below the mutated phenylalanine. To examine the possibility of this electrostatic interaction, a model was constructed by introducing the R294K and F233W mutations in the VSD of the Kv1.2 channel. Then, a harmonic restraint was imposed on residues R294K and E236 to pull them together. The restraint was applied between the NZ atom of the lysine side chain and atom OE2 of the carboxylate anion in E2. The Cα-Cα distance separating R1 and D2 is 19.8 Å in the initial model. Surprisingly, despite the seemingly long distance attraction between residues, the backbone RMSD before and after the harmonic restraints were applied is 2.5 Å (see Figure 1D).

Average Consensus Model

The results from the four restrained simulations were used to generate a consensus structural model for the resting conformation of the Kv1.2 VSD. The four structures were first aligned by minimizing the RMSD of the Cα atoms in the transmembrane region. A spatial average of the coordinates of all four simulations was performed to generate a target structure for the consensus model. Finally, to alleviate any artifacts caused by performing the geometric average on the coordinates, a targeted molecular dynamics (TMD) simulation was performed over 250 ps. The resulting structure from the TMD provides a realistic consensus model that closely approximates the average of the four restrained simulations. The superposition of all the constrained models is shown in Figure 2, and the consensus structural model is shown in Figure 3. An animation displaying the superimposed models is provided in Supplementary Information. Importantly, despite their conformational differences, the activated and resting states present the same overall topology, with the S1-S4 helical segments packed in counter-clockwise fashion as seen from the extracellular side. The PDB coordinates are also provided as supplementary information.

Figure 2.

Overlay of simulated VSD structures of the Kv1.2 channel after imposing harmonic restraints. Color scheme is as follows: TM helix S1 is gray, TM helix S2 is yellow, TM helix S3 is red, and TM helix S4 is blue. The Cα shown in the figure correspond to the residues that have been subjected to harmonic restraints. RMSD values given in parenthesis indicate the deviation of the Cα atoms among the four restrained simulations. The overall RMSD of each structure relative to the average consensus model does not exceed 2.2 Å.

Figure 3.

Side-by-side comparison of the active (left) and consensus model resting (right) states for a single VSD of the Kv1.2 channel. The Cα of R294 is displaced 7-10 Å vertically. RMS fluctuation values were obtained from Khalili-Araghi's extended simulations and reflect the spread of vertical z coordinates for the backbone Cα atom. Coordinates for the two models are provided as supplementary information.

Discussion

We have sought to identify a structural model for the resting state conformation of the VSD of Kv1.2. Our approach has been to mimic the experimental conditions in four separate simulations and generate a consensus structural model. The restraints were applied individually by carrying out the proper mutations in the model in order to realistically mimic the actual experimental conditions and because there is no indication that the interactions can be satisfied simultaneously. In all cases, the restrained MD simulations of VSD mutants resulted mainly in rearrangement of the side chains involved, indicating that the starting model is able to accommodate the experimental residue-residue interactions without large-scale conformational changes of the backbone (Figures 1 and S1). The rigid body motion of the transmembrane helices justifies our use of such few restraints as compared to NMR methods, which require an average of ~15 restraints per residue to determine a precise structure (Clore et al., 1993).

The final coordinates of the backbone Cα from the four restrained models do not differ markedly, and the relative overall root-mean-square (rms) deviations do not exceed 2.2 Å (Figures 2 and S1). This suggests that it is reasonable to determine an average moderate-resolution consensus model of the VSD in the resting state consistent with available experimental data. Despite having identified a structure whose attributes are in consensus with experimental data however, it is prudent to note that other models could be found that also satisfy the constraints used.

The consensus model displays features that are consistent with all three idealized mechanistic models that have been proposed previously. On the one hand this may appear to be somewhat surprising in view of the sharp divergences among the idealized models. However, it is not entirely unexpected given the fact that the resulting consensus conformation must ultimately be consistent with all the available experimental results at the origin of these idealized models. For example, one of the most stringent constraints from the biotin-avidin trapping data used in support of the paddle model corresponds to position L121C in KvAP, which is accessible to a 10 Å biotinylated linker from the intracellular side of the membrane (Ruta et al., 2005). However, a model of the VSD with a cysteine-attached biotin inserted at position L298 in the Kv1.2 channel, and complexed with avidin (pdb 1AVD) indicates that this constraint can be satisfied while remaining near the average consensus model (Figure 4). As in the sliding helix model, the predominant motion appears to involve a translation of S4 along its main axis, together with some rotation and tilting. However, S4 clearly does not move within a proteinaceous pore shielding it completely from the surrounding lipids, as was traditionally imagined. Consistent with the paddle model, many of the residues of the VSD are extensively exposed to the membrane lipids. However, the charged residues along S3 or S4 do not point directly into the low dielectric lipid hydrocarbon; they are either involved with electrostatic interactions with other charged residues in S1, S2 and S3, or with the polar headgroup of the lipids. Finally, there appear to be extensive rearrangement of the internal aqueous crevices contributing to a focusing of the membrane field, as depicted in the transporter model. This feature is consistent with the general idea that the internal and external solutions are electrostatically separated by a relatively thin isolating region (Starace and Bezanilla, 2004; Ahern & Horn, 2005; Freites et al., 2006; Sands & Sansom, 2007; Jogini & Roux, 2007; Asamoah et al., 2003). Previous MD computations showed that the membrane field is indeed focused over a distance of about 10 Å between E1 and E2 (see Fig. 4 of Khalili-Araghi et al. (2010)), which is about 2 to 3 times more intense than the membrane field across a bilayer, in accord with experiments (Asamoah et al., 2003).

Figure 4.

Model of Kv1.2 VSD bound to avidin via biotin with 10 Å IPEO linker. The coordinates correspond to the consensus model of the resting state. The biotin linker was attached via a disulfide bond to L298C (in Kv1.2). The corresponding residue in KvAP was experimentally shown to be accessible to intracellular avidin (Ruta et al., 2005). The docking was performed while keeping the VSD fixed to ensure that the biotin-avidin did not perturb the protein.

The current consensus model suggests the voltage-sensing motions are of intermediate magnitude. Figure 3 depicts the average consensus model of the resting state and the active state (taken from the MD of Khalili-Araghi et al. (2010)) in a side-by-side comparison. The comparison indicates a vertical motion of the S4 that is about 7-10 Å, as measured at the Cα atom of R1, though it should be noted that the estimate of vertical motion of the S4 helix depends on how it is defined and aligned with respect to the rest of the structure (see Method). Without additional information, it is prudent to envision a range of values for the vertical motion in order to reflect the uncertainty among the different models depicted in Figure 2 (see also Figure S1) and the thermal fluctuations estimated by MD simulations (see Figure S2). The displacement of S4 is considerably larger than the 1-2 Å initially proposed on the basis of lanthanide resonance energy transfer (LRET) measurements (Cha et al., 1999). Nonetheless, the displacement of S4 in the model is consistent with the LRET results, once the molecular structures of the donor and acceptors with their linker are taken into account. Similarly, the motion is considerably smaller than the vertical motion typically associated with the paddle model, originally proposed to be 20-25 Å on the basis of biotin-avidin trapping data obtained with the KvAP bacterial channel (Jiang et al., 2003). However, further analysis shows that the movement displayed by the S4 helix in the consensus model is actually consistent with the biotin-avidin trapping data (Figure 4). It is also important to keep in mind that those conformational states are dynamic and undergo significant fluctuations (Figure S2).

While the three idealized models proposed to explain voltage sensing are often contrasted by the magnitude of the S4 movements, this is clearly an oversimplification. For example, the helix-screw/sliding-helix model pictures predominantly a rigid body motion of S4. However, available X-ray structures and several independent MD simulations provide support for the intriguing possibility that voltage-sensing might be accompanied by a transformation of S4 from an α helix to a 3-10 helix in the resting state (Long et al., 2007; Clayton et al., 2008; Villalba-Galea et al., 2008; Bjelkmar et al., 2009; Khalili-Araghi et al., 2010; Vieira-Pires & Morais-Cabral, 2010). Indeed, in our own consensus model the S4 segment retains a portion of the 3-10 helix that was exhibited in Khalili-Araghi's model. Similarly, one implication of the paddle model is that the helix-turn-helix S3-S4 moves together as a rigid body. Such concerted motion is neither observed in the consensus model nor in experiments. Based on disulfide bond pattern of cysteine pairs substituted between S3 and S4 in Shaker, Broomand and Elinder concluded that the two helices can move relative to each other (Broomand & Elinder, 2008). Therefore, while the consensus model recapitulates many of the suggestions embodied by the three idealized models, some specific details from those models are not supported.

A comparison of the consensus model of the VSD in the resting state with other available models in the literature shows that they all share a common overall topology (Figure S3), although some differences are worth noting. The most similar is the model of the KvAP channel obtained via the Rosetta protein folding algorithm (Yarov-Yarovoy et al., 2006), which was also utilized to generate an early version of the model analyzed here (Pathak et al., 2007). The differences are somewhat larger with a recent model of the KvAP channel (Schow et al., 2010) and another model of the Kv1.2 channel (Delemotte et al., 2010). To generate the KvAP model, the biotin-avidin trapping data of Ruta et al (Ruta et al., 2005) was converted into a set of specific z-position constraints, which were all applied simultaneously to residues in S3 and S4 during all-atom MD simulations (Schow et al., 2010). The resulting VSD is broadly similar to the consensus model, with the exception of a local unfolding of the S3 helix. The model of Kv1.2 channel was generated in a similar way, by imposing several residue-residue distances from experiments (Delemotte et al., 2010). Again, the overall structure is similar to the consensus model, although the model exhibits a kink at the center of the S3 and S4 helices and the R294 side chain is in close proximity to E2.

Although the overall picture is consistent, one disagreement concerns the position of the side chain of R1. The consensus model predicts that R1 is stabilized by interactions with E1 in the resting state (Figures 3 and S4). Some other models place R1 near the acidic side chain E2, closer to the intracellular membrane surface (Tao et al., 2010). Even assuming that the backbone remains roughly at the same position, it is possible that R1 might actually interact with E1 or with E2, or that it is located somewhat in between these two residues. This aspect of the resting state conformation is not strongly constrained with the currently available information. Arginine and glutamic acid side chains are about 5-6 Å long and the backbone Cα-Cα distance in the consensus model (Figure 3) is ~12 Å between R1 and E1 and ~17 Å between R1 and E2, suggesting that either interactions could be possible. However, several experimental observations are broadly indicative that R1 remains above the center of the bilayer in the resting state in functional K channels, corresponding roughly to the position of F233 in S2 (Figure 3). Substituting a histidine at the position of R1 is known to produce a proton pore for the resting state of Shaker (Starace et al., 1997; Starace & Bezanilla, 2001, 2004). Other mutations at the position of R1 allow the passage of the so-called Omega currents through the VSD (Tombola et al., 2005, 2007). The latter were interpreted in terms of a model in which the displacement of S4 undergoes an inward movement of 13 Å at the extracellular end of S4, and 10 Å at the Cα of R1 (Tombola et al., 2007) The proton pore and the Omega current are consistent with the notion that R1 in the resting state must be positioned in such a way to form a narrow barrier between the internal and external solutions for the water molecules and protons. Consistent with this suggestion, Ahern and Horn (Ahern & Horn, 2004, 2005), found that the membrane field acting on R1 falls over a distance of about 4 Å using functional measurements with tethered charges attached to the S4 segment. Similarly, Swartz and co-workers concluded using a tarantula toxin that the voltage-sensing S3-S4 helix-turn-helix traverses no more than the outer leaflet of the membrane bilayer during activation (Phillips et al., 2005b). Fluorescence measurements with the lipophilic ion dipicrylamine distributes on either side of the lipid bilayer used as a resonance-energy-transfer acceptor from donor molecules attached to several positions in the Shaker K channel indicated that the S4 segment does not translocate across the whole lipid bilayer (Chanda et al., 2005). Luminescence resonance energy transfer used to measure distances between the voltage sensors and a pore-bound scorpion toxin indicated that the VSD segments do not undergo significant transmembrane translation in functional Shaker K channels (Posson et al., 2005). A high degree of structural complementarity within the S3-S4 helix-turn-helix paddle motif is not required for the voltage sensing, since it is possible to delete much of S3b while retaining functional channels (Xu et al., 2010). Moreover, the S3b segment is not seen to change accessibility during gating in Shaker channels (Gonzalez et al., 2005). Therefore, while the exact position of the R1 side chain in the resting state conformation is uncertain and could be either close to E1 or E2, most experimental measurements appear to support the notion that the S4 backbone at the level of R1 is bounded by the mid-point of the membrane bilayer.

It is unclear whether the remaining differences can be attributed to the experimental methods used to study them, or to evolutionary variations within the channels themselves. Many conformational states are accessible to the VSD, and their relative populations could be highly sensitive to small perturbations. For instance, extreme hyperpolarization of the squid giant axon results in delayed activation and onset of K currents, consistent with the existence of deeper resting conformational states (Cole & Moore, 1960). The possibility of multiple resting states is further supported by the temperature dependent conformational transition in Shaker, which exhibits a decrease in entropy upon activation (Rodríguez & Bezanilla, 1996; Rodríguez et al., 1998). Therefore, while we may speak of a single open state conformation, it is conceivable that the “resting state” of the VSD actually comprises a population of conformations.

The conformational equilibrium of the VSD–a small structure displaying a fair amount of internal dynamics—may be sensitive to several factors. For instance, it is know to be affected by lipids (Xu et al., 2008; Schmidt et al., 2009). It seems plausible that introduction of chemical probes or amino acid substitutions along S4 affect the sensing motions, thus increasing the difficulties in interpreting experimental observations. In this regard, some ambiguity cannot be avoided with the rough experimental approaches that are used. After all, the distance between E1 and E2 is about 10 Å, and the length of an arginine side chain is about 5-6 Å. It is conceivable that some perturbation may result in R1 being either closer to E1, while others resulting in R1 being closer to E2. As an illustration of the high sensitivity of functional measurements, a recent kinetic study shows that even in the case of the active state, different engineered metal bridges are correlated with subtle movements of S4 (Phillips & Swartz, 2010). This suggests that some experimentally engineered residue-residue cross-links may either slightly distort the conformation of the VSD or stabilize alternate states, and therefore one should exert caution when using such restraints to build structural models.

Additional experimental interactions will hopefully help better define the resting state of the VSD. But the concept of an average consensus model will make sense only if a consistent picture continue to emerge from the growing body of data. In this regard, a very recent result by Lin et al (2011) offers an opportunity to test the present approach. The authors describe a double mutant in the Shaker channel, I287H (along S2) and A359H (3 residues preceding R1 in the S3-S4 loop), which allows the formation of a new Zn²⁺ metal bridge site trapping the resting state of the VSD of (Lin et al, 2011). To clarify the structural implication of this result, an atomic model of the VSD with the double mutation and the Zn²⁺ bridge was generated with MD simulations following the same protocol used for the other interactions (Supplementary Information). The resulting model shows that it is possible to satisfy the requirement of this interaction without a considerable displacement of the backbone of S1-S4 relative to the consensus model (Figure S5). Specifically, the Cα of R1 remains within ~1.5 A from the consensus model (see also animation). Interestingly, the model suggests that the side chain of R1 might point toward Phe233 when this bridge is present, a feature that is not observed in the models based on the other metal bridges. From this perspective it is worth re-emphasizing once more that while the backbone of the average consensus model may be well defined with a high level of confidence, the configuration of the side chains remains somewhat uncertain.

In summary, we have used restrained MD simulations based on the Khalili-Araghi et al model of the Kv1.2 VSD in the resting state to reproduce experimental interactions. It was possible to satisfy the experimental constraints with limited alterations to the protein backbone, as judged by RMSD values. The resulting consensus structural model also places the S4 in a position that is accessible for binding avidin, as shown in experimental electrophysiology recordings. The agreement of these results originating from various labs using different techniques is broadly indicative of an emerging consensus for the resting conformational state of a VSD. The consensus experimentally-constrained structural model provides a clearer picture of the voltage sensing pathway in Kv-like proteins.

Methods

It is useful to adopt a naming convention for the most important charged residues of the VSD. Accordingly, the basic residues along S4 are: R1, R2, R3, R4, K5, R6, corresponding to R362, R365, R368, R371, K374, R377 in Shaker, or R294, R297, R300, R303, K306, R309 in Kv1.2. The acidic residues are: E0 along S1, corresponding to E247 in Shaker or E183 in Kv1.2, E1 and E2 along S2, corresponding to E283 and E293 in Shaker or E226 and E236 in Kv1.2, and D3 along S3, corresponding to D316 in Shaker or D259 in Kv1.2. In the active state conformation, salt-bridges are formed between R4 and E1, and between K5 with E2 and D3 (Tiwari-Woodruff et al., 1997, 2000). In the model of the resting state conformation, R1 interacts with E1, R3 and R4 interact with E2 and D3 (Khalili-Araghi et al., 2010).

Molecular dynamics (MD) simulations on a single VSD subunit in an explicit water/lipid/ions system were performed in NAMD 2.6 (Phillips et al., 2005a) using the CHARMM27 force field and the TIP3P water model. The FreeEnergy module was used to impose harmonic restraints on groups of atoms in the spirit of steered molecular dynamics. A spring constant of 20 kcal/mol/Ų was used for the restraints, and the reference distance was decreased over a period of five nanoseconds to pull the atom groups together. After reaching the target distance, the simulation was continued for another nanosecond to allow the system to equilibrate. The coordinates of each simulation were averaged over the last 1 ns to obtain an average structure. All simulations were subjected to a constant electric field equivalent to a -500mV membrane potential. Mutagenesis of residues was performed with the PsfGen module of VMD. The residue-residue interactions are shown in Figure 1 and the overlay of all the resulting models is shown in Figure 2.

The coordinates obtained by averaging all of the simulations were used in targeted MD (TMD) simulations of the initial reference structure of Khalili-Araghi's Kv1.2 resting state to obtain the final consensus model (shown in Figure 3). The four restrained simulations resulted in structures that are very similar to Khalili-Araghi's resting state model (RMSD ≤ 3.0 Å). To obtain an average configuration of the restrained VSD's, the resulting structures from the four restrained simulations were each aligned to Khalili-Araghi's VSD model and a geometric average was performed on the atom coordinates. Finally, these averaged coordinates were used as target coordinates for Khalili-Araghi's initial resting state model and a TMD simulation was performed. This final step was done to ensure that the non-physical averaging of the coordinates did not produce an energetically unfavorable conformation. The relative rms deviations among the four models do not exceed 2.2 Å, suggesting that the magnitude of the charge movement within the VSD is not expected to differ markedly from the previous MD calculations by Khalili-Araghi et al. (2010). An animation rotating the superimposed models along the Z-axis is given in Supplementary Information.

To accurately compare the active and resting states, the tetramer structures provided by Khalili-Araghi where aligned with their principal axis along the membrane normal axis. To lend weight to each of the subunits, four-fold symmetry was imposed on the full-length tetrameric channels by performing a spatial average on the carbon alpha atoms. The Oriented Proteins in Membranes (OPM) database was consulted for placing the tetramers vertically in the membrane by making sure the S1 and S2 helices exhibited no vertical movement (Lomize et al., 2006). Finally, the isolated VSD was aligned to the VSD of the resting state conformation of the symmetric tetramer for the comparison shown in Figure 3. It is noted that the isolated VSD in the membrane is tilted with respect to the orientation observed in the full-length tetrameric channel.

The biotin-avidin trapping model system was generated using a multi-stage protocol. The complex was assembled using restrained MD and then relaxed for 2 ns. First, residue 298 was mutated to cysteine to conjugate the biotinylated linker as was done in Ruta et al. (2005). Next, the avidin (1AVD) was oriented along z and kept rigid while being steered towards the VSD in vacuum. A restraint was used at z = -12 A to prevent the avidin from penetrating the membrane region. The rigid body constraints on the avidin were removed when the structure was in close proximity to the VSD. The Antechamber package was used to generate force field parameters required for simulating the biotinylated linker (Wang et al., 2006). Harmonic restraints were then used to slowly steer the linker towards residue 298 and to keep the biotin end bound to avidin. Finally, CHARMM-gui provided scripts to generate the all-atom explicit water/lipid membrane system (Jo et al., 2008). The resulting model is shown in Figure 4.

All figures were generated by the molecular visualization package VMD (Humphrey et al., 1996).

Highlights.

Resting state conformation of voltage-sensing domain, residue-residue interactions as structural constraints from experiments