A Structural Model of the Inactivation Gate of Voltage-Activated Potassium Channels

Abstract

After opening, the Shaker voltage-gated potassium (K_V) channel rapidly inactivates when one of its four N-termini enters and occludes the channel pore. Although it is known that the tip of the N-terminus reaches deep into the central cavity, the conformation adopted by this domain during inactivation and the nature of its interactions with the rest of the channel remain unclear. Here, we use molecular dynamics simulations coupled with electrophysiology experiments to reveal the atomic-scale mechanisms of inactivation. We find that the first six amino acids of the N-terminus spontaneously enter the central cavity in an extended conformation, establishing hydrophobic contacts with residues lining the pore. A second portion of the N-terminus, consisting of a long 24 amino acid α-helix, forms numerous polar contacts with residues in the intracellular entryway of the T1 domain. Double mutant cycle analysis revealed a strong relationship between predicted interatomic distances and empirically observed thermodynamic coupling, establishing a plausible model of the transition of K_V channels to the inactivated state.

Significance

Upon persisted depolarization, voltage-gated potassium channel inactivates rapidly. It is known that the tip of the N-terminus penetrates deep into the permeation pathway during inactivation, impeding ion conduction. Here we use molecular dynamics (MD) simulations coupled with electrophysiology experiments to propose a plausible atomic-scale mechanism of inactivation. In this model, the first six amino acids of the N-terminus establish hydrophobic contacts with residues lining the central cavity of the pore. In addition, a contiguous α-helix forms numerous polar contacts with residues in the intracellular entryway to the pore. Electrophysiology experiments using the double mutant cycle strategy revealed a strong relationship between model predicted interatomic distances and empirically observed thermodynamic coupling, establishing a plausible model of the transition to the inactivated state.

Introduction

Voltage-gated potassium (K_V) channels are integral membrane proteins that function as homotetramer complexes (1, 2). The particular firing properties of neurons depend in part on the dynamics of gates within the K_V channels, which control permeation through a K⁺ selective pore (3). In response to depolarization, all K_V channels open an intracellular gate that allows K⁺ to permeate (4, 5, 6). In some K_V channels, including the Shaker channel from Drosophila melanogaster (2), opening the voltage-dependent gate leads to a rapid inactivation process produced by another gate, termed the inactivation particle or inactivation gate. This inactivation domain is located at the N-terminus of the channel (7, 8), and under certain stimuli enters deep into the central cavity, cutting off ion conduction (9, 10). The inactivation mechanism is coupled to the activation process (7), and once the channel is open the inactivation gate acts like an open-channel blocker (11). Because Shaker K_V channels are homotetramers, each channel has four N-termini; however, only one of these is necessary to occlude the pore and cause inactivation (12, 13). The inactivation particle is highly disordered, which makes it not amenable to x-ray crystallography or cryo-electron microscopy; therefore, the question arises of how such a disordered polypeptide is able to reliably and rapidly induce inactivation. Recent functional studies have revealed several key components of the inactivation mechanism, which requires that the cytoplasmic N-terminus be guided through the intracellular vestibule where it then penetrates the channel pore. During the first stage of inactivation, the N-terminus appears to interact with the T1 domain at the entryway to the intracellular vestibule on its way to the central cavity (14, 15). Once inside the central cavity, the tip of the N-terminus experiences hydrophobic interactions with residues lining the channel pore, blocking the permeation pathway (9, 10).

Despite this body of research, the precise interactions governing the transition of the N-terminus from the cytoplasm to the central cavity of the channel, as well as the conformational state that it adopts upon reaching the pore, have remained unclear. A β-hairpin conformation spanning residues 1–12 has been previously proposed (16, 17), with the hairpin loop (residues 6 and 7) forming its front end. However, mutagenesis experiments have revealed contacts between residue at the position 2 of the N-terminus and residue 470 within the central pore (9, 10), which appears inconsistent with this β-hairpin structure. In this study, we describe molecular modeling, MD simulations, and experimental mutant cycle analyses (18, 19, 20) to characterize the Shaker inactivation mechanism at the molecular level. The computational methods reveal a plausible conformation for the N-terminus during inactivation and identify several contacts between this domain and residues in the central pore and T1 domain. Mutant cycle analyses provide experimental support for the existence of these contacts.

Materials and Methods

Mutagenesis

Wild-type Shaker K⁺ channel DNA was used as a template to introduce single and double mutations, which were produced using standard PCR techniques.

DNA expression

All channel DNAs were cloned into GW1-CMV (British Biotechnology, Oxford, UK) vectors and cotransfected with CD8 and SV40 DNAs into HEK293T cells (21) using electroporation (Nucleofector II; Amaxa Biosystems, Cologne, Germany) system. After transfection, cells were cultured in DMEM with 10% fetal bovine serum. Experiments were performed 1 day after transfection.

Experimental solutions and electrophysiological recordings

The intracellular solution was composed of the following (mM): 160 KCl, 0.5 MgCl₂, 1 ethylene glycol-bis(β-aminoethyl ether)-N,N,N′,N′-tetraacetic acid, and 10 HEPES (pH = 7.4 with KOH). The extracellular solution contained the following (mM): 150 NaCl, 10 KCl, 1 MgCl₂, 3 CaCl₂, and 10 HEPES (pH = 7.4 with NaOH). Current recordings were obtained from inside-out excised patches, sampled at 100 kHz and filtered at 5 kHz using an Axopatch 200B amplifier (Axon Instruments, San Jose, CA). Data acquisition was performed using a Digidata 1440A AD/DA converter and Clampex software (Axon Instruments). Borosilicate glass pipettes (Harvard Apparatus, Holliston, MA) were pulled (Sutter Instrument, Novato, CA) to have 1.9–2.7 MΩ resistance. The experiments were performed at 21–23°C and monitored using a thermocoupler (Thermometrics Corp, Northridge, CA) in the chamber.

Modeling and analysis of ionic currents

Traces were fit in Matlab with a Markov kinetic model to estimate the values of k_on and k_off (15). The equilibrium constant (K_eq = k_on/k_off) was determined for each mutant channel, and thermodynamic cycle analyses were used to quantify the level of interaction (Ω) between two amino acids as follows (18):

$$\mathit{\Omega }=\frac{K_{\mathrm{eq}}^{\left(\mathrm{WT,WT}\right)}\times K_{\mathrm{eq}}^{\left(\mathrm{Mut,Mut}\right)}}{K_{\mathrm{eq}}^{\left(\mathrm{WT,Mut}\right)}\times K_{\mathrm{eq}}^{\left(\mathrm{Mut,WT}\right)}}$$

An Ω value different from unity indicates interaction between two mutations.

Molecular modeling of Shaker K⁺ channel

A Shaker comparative model (residues 96–489) was built using the Kv1.2 channel from Rattus norvegicus (Protein Data Bank [PDB]: 3LUT) (22) as a template. A preliminary pairwise sequence alignment for modeling, initially of only one monomer, was generated using the AlignMe server (23). This alignment was subsequently replicated to build a molecular model of the tetrameric channel structure. Adjustments in the initial alignment were made, and refined models were built to optimize agreement with PSIPRED (24) secondary structure and TOPCONS (25) transmembrane topology predictions. In the final alignment, 74.7% of the residues were identical, and the similarity reached 83.4%. A total of 2000 iterations of model building were performed with MODELER (26) applying symmetry restraints between the C_α atoms of each monomer. The best Shaker model was selected as that with the lowest Molpdf energy value in MODELER and the highest Procheck (27) and global ProQM (28) scores. The final model is of excellent quality according to Procheck, with 98.9% residues in the favored and additional allowed regions of the Ramachandran plot. The global ProQM score is also very high, with an optimal value of 0.614.

Prediction of the N-terminal segment folding

Secondary structure predictions of the full-length Shaker sequence (UniProt: P08510) were performed using the software PSIPRED (24, 29), SOPMA (30), SPIDER2 (31), JPRED4 (32), and PSSPRED (33). The PEP-FOLD server (34) was used to predict a de novo structural model of the inactivation particle, including the sequence of the first 36 amino acids of the protein (the maximum allowed by the software). A total of 200 models were generated, and we selected the one with the lowest sOPEP energy value in PEP-FOLD to use in our subsequent simulations.

Docking simulations

Docking simulations were carried out with a β-hairpin structure (PDB: 2LNY) (17) and the de novo predicted structural model of the inactivation particle. In preparation for docking, a grid box was built using the mass center of the four I470 residues of the homotetrameric model of the channel as a reference point. The grid was large enough (47 × 47 × 47 Å³) to accommodate free motion of the N-terminus structural model into any region of the channel. The docking calculations were performed using AutoDock Vina (35) with the default scoring parameters, and AutoDockTools (36) was used to select the lowest-scoring conformation of the inactivation particle docking with the channel’s inner pore and T1 domain. The best conformation of the complex containing the de novo predicted N-terminus model was corroborated using the Rosetta FlexPepDock protocol (37). Starting from an approximate specification of a peptide structure and its binding site, FlexPepDock performs simultaneous docking and folding taking the receptor surface into account. A total of 11,000 conformations were generated for this complex and compared with those obtained by AutoDock Vina (Fig. S2).

MD simulations

The lowest-scoring conformation of the two complexes yielded by computational docking were used as the starting point for MD simulations. To reduce the computational expense, only the channel pore (residues 378–489) and the T1 domain (residues 142–199) were included in the simulations. Each complex was embedded in a fully hydrated palmitoyl-oleyl-phosphatidyl-choline bilayer solvated with explicit water molecules. Three K⁺ ions were placed in the selectivity filter at K⁺ binding sites S₀, S₂, and S₄, and two water molecules were placed at S₁ and S₃ (38). Potassium and chloride ions (0.15 M KCl) were added to the aqueous phase to mimic physiological ionic strength and to ensure charge neutrality. The protonation states of the residues at neutral pH were checked with the PropKa program (39). The M1 residue was considered protonated for the two N-terminal structures, although an acetylated state of the N-terminus for the α-helical conformation was also evaluated. The initial configuration of each system was optimized by means of 25,000 steps of energy minimization, followed by equilibration and relaxation in a 152−368 ns MD simulation at 298 K in the isobaric-isothermal ensemble. Soft harmonic constraints were applied to the protein and the ions of the selectivity filter during the first 15 ns of simulation, which were decreased gradually from 60 to 0 kcal mol⁻¹ Å⁻² over this period. The distance between the center of mass of each T1 domain and the transmembrane region of the same subunit was restrained to 45 Å using the collective variable module (Colvars) (40). The center-of-mass distances of neighboring subunits of the T1 domain were also restrained to 24 Å. All MD simulations were performed using the program NAMD v2.10 (41) and the CHARMM36 (42) force field. The temperature was maintained using a Langevin thermostat with a damping coefficient of 5 ps⁻¹. The pressure (1 atm) was maintained using the Langevin piston method (43). Long-range electrostatic interactions were computed using the particle-mesh Ewald summation method (44), with a smooth real-space cutoff applied between 8 and 9 Å. All covalent bonds involving hydrogen as well as the intramolecular geometries of water molecules were constrained using the SETTLE algorithm (45). The Verlet I r-RESPA multiple time-step integrator (46) was used with a time step of 2 fs for most interactions and 4 fs for long-range electrostatics. Structural analyses of the complexes were performed using the program VMD (47). The dimensions of the channel’s pore were calculated with the HOLE program (48).

Applied voltage simulations

After equilibration and relaxation of the complexes, MD simulations applying constant electric fields perpendicular to the membrane plane were performed to represent transmembrane potentials (49, 50) of +0.6 and +1.0 V. The simulations were continued with the applied voltages for 0.227–1.45 μs. Distances between the side chains of pairs of key residues were evaluated during the applied voltage simulations (full penetration period, Fig. S6). To evaluate the convergence of the simulation representing the open-inactivated state of the channel, starting from the last frame of the trajectory at 0.6 V, we set the transmembrane potential to 0 V and an additional 270 ns of MD were run. Distances between the pairs of residues were again calculated and compared with those found in the simulation under voltage (Fig. S3).

Results

Difficulties of the β-hairpin model

The conformation adopted by the N-terminus during inactivation of the Shaker K_V channel has been controversial. Early efforts suggested that the inactivation particle acquires a β-hairpin structure (16), a proposal that was subsequently supported by a combination of electrophysiology, nuclear magnetic resonance, Fourier transform infrared spectroscopy, and MD (17, 51). However, these structural studies were performed with peptides embedded in lipid bilayers or interacting with bacterial KcsA channels in the presence of detergent, which may not favor the same structure that is found within the pore of the intact K_V channel. Subsequent mutagenesis experiments demonstrated that residue A2 of the N-terminal tip makes contact with residue I470 inside the central cavity (9). This finding appears to preclude a β-hairpin motif, which predicts that residue L7 enters more deeply than A2, impeding contact of the latter with I470. Based on that observation, our overall goal was to define the specific conformation adopted by the N-terminus during inactivation of the Shaker K_V channel and to characterize the nature of the interactions that allow for channel block.

To begin, we built a molecular model of the Shaker K_V channel on the basis of the K_V 1.2 crystal structure (22) with the pore in the open state. Because both channels exhibit a considerable sequence identity of 75%, our model is expected to have a correct fold and high accuracy in the atomic details. The 20 amino acid–long β-hairpin structure (17) was then docked onto the channel’s intracellular vestibule, resulting in the structural model shown in Fig. 1 A. The complex was relaxed in a 152 ns MD simulation. During this time, the β-hairpin made no progress in penetrating the central cavity. Therefore, to coax the positively charged N-terminal methionine to penetrate the central cavity, a membrane potential of 1.0 V was applied, and additional 227 ns of MD were run. Even though intracellular K⁺ ions could reach the selectivity filter under these simulated conditions, the amino acids forming the β-hairpin still did not penetrate the central cavity. Analysis of pore radius showed values of ∼5 Å in the region surrounding residue I470 (Fig. 1 B). On the other hand, largest cross section of the β-hairpin along its axis exhibited a maximal radius of ∼7 Å, which occurs at the base of the hairpin (residues M1 and E12) and coincides with a distance of 5.3 Å between the C_α atoms of these residues (Fig. 1 C). Structures of potassium-selective ion channels in open-pore conformations exhibit radii values <6 Å (52), which makes the distortions of the inner vestibule necessary to accommodate a β-hairpin structure unlikely. In fact, during the simulation, we observed partial unzipping of the hairpin (reaching 12.9 Å between the C_α atoms at the base) and its migration away from the inner pore. These analyses therefore appear to exclude an inactivation particle with a β-hairpin conformation.

Figure 1.

The central cavity has insufficient space to accommodate a β-hairpin structure. (A) Cartoon representation of Shaker K_V channel (green) in complex with the N-terminus segment (red) in a β-hairpin conformation. Potassium ions and water molecules in the selectivity filter are depicted with space-filling representations. The complex was embedded in a solvated membrane and simulated for 227 ns, applying a transmembrane potential of 1.0 V. (B) Pore-radius profile of the channel calculated as an average over the MD simulation. The initial nonequilibrium portion of the trajectory was discarded for the analysis. The HOLE pore-lining surface (gray) and channel pore (green) in the last frame of the MD trajectory are shown in the inset. The I470 residues are depicted as yellow spheres. (C) Initial (top) and final (bottom) conformation after simulation of the β-hairpin structure in the channel. The distance between the C_α atoms of the M1 and E12 residues indicates that the β-hairpin starts to open during the simulation.

The helix-extension model

We propose a model in which the inactivation particle enters the inner vestibule as an extended structure so that the tip of the N-terminus can reach the central cavity. Secondary structure predictions using different algorithms (Fig. S1), applied to the full-length Shaker sequence, showed mostly an α-helical structure from residues 11 to 54. For the first 10 residues of the N-terminus; however, the algorithms predicted a small α-helical portion or a disordered coil in free solution rather than a β-hairpin. We then generated a de novo model of the first 36 amino acids of the Shaker K_V channel’s N-terminus using the PEP-FOLD3 server (34). This structure was subsequently corroborated in the context of the Rosetta FlexPepDock protocol (37) (Fig. S2). The model contains a short α-helix (residues 1–8) followed by a longer α-helix (residues 12–36). As shown in Fig. 2 A, this structure was placed in the channel’s intracellular vestibule using computational docking and MD simulation. Within the first 50 ns of MD simulation at 0 V, the short α-helix spontaneously began to unfold, and by 368 ns it had completely unfolded and begun to penetrate the central cavity in the absence of any external forces (Fig. 2 B). Penetration of an extended initial model of the N-terminus tip was also evaluated; nevertheless, it exhibited too much flexibility and failed to enter the pore. Recent studies have demonstrated that the tip of the N-terminus establishes direct contact with residues lining the central cavity’s surface (9, 10). We used the distances between the C_α atoms of residue A2 of the N-terminus and the four I470 residues of the central cavity of the channel to quantify the degree of penetration and found that these distances decreased as the N-terminus began to penetrate the central cavity (Fig. 2 D). However, after ∼170 ns, there were no further changes in the position of the inactivation particle. At the end of this 368 ns simulation period, the distances between A2 and the four I470 residues in the central cavity were too long to be compatible with disulfide bond formation (∼7 Å between Cδ of I470 and Cβ of 2A), as is known to occur when these two residues are mutated to cysteines (9).

Figure 2.

The N-terminus enters the inner vestibule as an unfolded structure. (A) Structure of the Shaker K_V channel (green) interacting with the double α-helical model of the inactivation particle (red). K⁺ ions and water molecules within the selectivity filter are shown as spheres. The complex was embedded in a membrane solvated with explicit water molecules. (B) After 368 ns of MD simulation without an applied voltage, the tip of the N-terminus unfolds and partially penetrates the central cavity. Residues A2 and I470 are indicated by blue and yellow spheres, respectively. (C) Final state of the complex after 1.45 μs of simulation under a transmembrane potential of 0.6 V. The close contact between the residues A2 and I470 indicates full penetration of the N-terminus tip and, therefore, complete blocking of the channel. (D) Distance between the C_α atoms of the A2 residue of the N-terminus and the I470 residues of each channel subunit as a function of simulated time without an applied voltage. (E) Distance between the C_α atoms of residues A2 and I470 as a function of simulated time applying a voltage of 0.6 V.

It is possible that further penetration of the N-terminus into the central cavity could happen spontaneously at 0 V on longer timescales, such as the millisecond timescale that characterizes wild-type Shaker inactivation. However, due to the computational cost of significantly longer simulations at the present time, we instead opted to apply an electric potential to promote further penetration of the N-terminal tip. Experimentally, the voltage has minimal intrinsic influence on the process of inactivation (53, 54). Nonetheless, our model predicts that the protonated primary amino group of the N-terminal methionine is affected by the membrane potential as compared with similar simulations having this group acetylated, where penetration of the N-terminus tip was not observed. It should be noted that these effects may become especially apparent at nonphysiological voltages. Beginning from the final state of the 368 ns simulation at 0 V (Fig. 2 B), we applied an electric potential of 0.6 V and simulated the system for additional 1.45 μs (Fig. 2 C). This manipulation was intended to reduce barriers for penetration of the NH₃⁺ terminal group into the central cavity. After ∼570 ns of simulation (Fig. 2 E, dashed line), the tip of the N-terminus entered deep into the central cavity and residue A2 made contact with residue I470 at distances suitable for disulfide bond formation (∼3.5 Å between Cδ of I470 and Cβ of 2A), as previous experiments have demonstrated (Video S1; 9, 10). Despite the large voltages used in the simulation to observe the full penetration of the inactivation particle, the configuration obtained in this way might be physiologically relevant. To further test this notion, we performed an additional simulation to see whether the configuration was stable in the absence of an external voltage. Beginning from the final frame of the 1.45 μs simulation with 0.6 V applied, the transmembrane voltage was removed and additional 270 ns of MD at 0 V were run. As expected, the N-terminus remained in the pore, suggesting that the configuration of the inactivation particle was stable in the absence of an external voltage (Fig. S3).

Video S1. Molecular Dynamic Simulation of the Inactivation Process

This video shows the MD simulations performed that were represented by three snapshots in Fig. 2, A–C.

Interestingly, the N-terminus tip of the inactivation gate was free to rotate inside the central cavity of the channel, and the residue A2 intermingled with various subunits, which is consistent with previous experiments (15). As can be seen in Fig. 2 E, the N-terminus initially made contact with residue I470 of subunit A during the period 570–710 ns. Shortly afterward, the contact switches to residue I470 of subunit D (710–840 ns), and then to subunit C and back to subunit D again (840–950 ns). At this point, the tip of the inactivation gate penetrated even further into the central cavity, so the distances between A2 and all four I470 increased slightly. Additionally, as the N-terminus penetrated the central cavity of the channel, the water molecules residing inside were displaced toward the selectivity filter, promoting the exit of K⁺ ions toward the extracellular medium (Fig. 2 C; Fig. S4). Nevertheless, this may only occur in conditions of particularly high transmembrane voltages and/or due to the specific set of initial conditions of the simulation at the time voltage was changed to 0.6 V.

Specific interactions between the N-terminus structural model and the channel

Once the tip of the N-terminus entered the central cavity, the protonated amino group of the N-terminal methionine made close contact with the polar side chains of the T442 residues at the entrance of the selectivity filter (Fig. 3 A). Furthermore, residues A2, A3, V4, A5, and L7 of the N-terminus established hydrophobic interactions with residues V67, I470, A471, V474, and V478 of the permeation pathway, as has been previously proposed based on experimental data (9, 10). Close proximity between A2 and I470 has been shown by the specific and reversible disulfide bond formation between them (9), whereas A3 was suggested to make contact with residues V474 and V478, based on mutant cycle analysis between the K_V1.4 channel α-subunit and the inactivation gate of β12 subunit (10). Both interactions are supported by our model (Fig. S5). Additionally, Y8 exhibited a hydrogen bond with N482 at the cytoplasmic entrance to the channel pore, potentially stabilizing the inactivation particle in the bound state.

Figure 3.

Association of the N-terminus with the channel is determined by both hydrophobic and polar components. (A) Simulation snapshot of the extended structure of the N-terminus making contact with residues in the central and inner cavity of Shaker K_V channel. The N-terminus (red) makes contact with all four subunits of the channel; however, for clarity only two subunits are shown here in cartoon representation (green). The channel pore (segments S5 and S6) is oriented with the extracellular side on top. (B) A rotation of 90° shows key contacts stabilizing the long α-helical portion of the inactivation particle in the intracellular entryway of the T1 domain. The residues in (A) and (B) are displayed as sticks colored by atom type: C in gray, O in red, N in blue, and S in yellow. (C) Contour map representing the average distance between the center of mass of the residue’s side chains of the N-terminus segment and the channel during the MD simulation (period of full penetration). The x axis corresponds to residues in the inactivation particle, whereas the y axis corresponds to residues in the T1-domain or inner vestibule. The color scale of the distances is ranged from 3 to 15 Å, with larger distances indicated by white.

The long α-helix between residues 12 and 36 made many contacts with residues located at the intracellular entryway of the permeation pathway (Fig. 3, A and B). Specifically, polar interactions were identified between the residues D13, R17, K19, and R164, D193, E194, S479 of subunit D; and E35–K198 of subunit C. A hydrophobic interaction between Q25 and L170 was also found. All these interactions are summarized in Fig. 3 C, which shows a contour-like map of the average distances (up to 15 Å; full penetration period) between residues of the N-terminus structural model and each of the four subunits of the channel. On a general level, our model predicts that the Shaker K_V channel’s inactivation mechanism is mostly regulated by three components: 1) residues 2–7 of the N-terminus making hydrophobic contacts with residues in the central cavity, 2) Y8 establishing a break between the hydrophobic and polar region through a hydrogen bond with N482 and, 3) the long α-helix (residues 12–36) forming polar interactions with residues at the inner pore and the T1 domain.

Model-based selection of amino acids pairs

The interactions between residues of the pore and the first 2–7 residues of the inactivation gate have been previously studied by mutant cycle analysis (10) and cysteine mutagenesis followed by chemical modification and disulfide bond formation (9). Access of the tip of the N-terminus into the inner pore should occur through the side windows, located immediately above the T1 domain (55, 56). Indeed, experimental data point to one specific electrostatic interaction between the positively charged residue R18 and negatively charged residues EDE161–163 in the T1 domain of Aplysia’s inactivating K_V channels (14), as well as their homologs R17 and EDE192-194 in Shaker K_V channels (15). The novel interactions predicted by our model between the long α-helix (residues 12–36) and the T1 domain provide a unique opportunity to extend these empirical studies more broadly.

We selected a total of eight possible pairs from these putatively interacting regions (long α-helix/T1 domain) for which the predicted distances between the amino acids’ side chains ranged from 3.6 to 14.6 Å. In addition, we studied the potential interaction between two pairs: 1) Y8 and N482, at the cytoplasmic end of the inner pore, for which our simulations predicted a hydrogen bond, and 2) K19 and S479, for which the inactivation gate is predicted to establish polar interactions with only subunit D (Fig. 3 C). The distances of these 10 pairs of contacts in the modeled trajectories studied are shown in Fig. S6. Interestingly, these trajectories show substantial dynamics of the inactivation gate outside of the intracellular cavity. For example, Y8 maintained close interactions with N482 throughout most of the full penetration period while transitioning between three different subunits of the channel. Other interactions, such as that between R17 and D193, were relatively more stable and restricted to a single subunit of the channel.

Experimental validation of the proposed structural model of the inactivation gate

At many levels, fast inactivation is a complex kinetic process. It requires two separate regions of the channel that during inactivation interact with a nonuniform stoichiometry. For example, the tip of one bound N-terminus could potentially interact with all four subunits in the intracellular cavity of the channel, whereas in other regions, such as the long α-helix in the N-terminus and the T1 domain, there might exist up to four such interactions with a 1:1 stoichiometry. In addition, the inactivation process is thought to occur in two sequential steps: a conductive “pre-inactivated” state followed by a nonconductive inactivated state (10, 14). Given this complexity, it is difficult to rely on a single variable or argument to validate the proposed structural model of inactivation. However, it is impossible in principle to distinguish between macroscopic ionic currents flowing in an open or a pre-inactivated channel using electrophysiology. For that reason, we fit our current recordings with a Markov kinetic model in which activation occurs in a single step to determine the values for k_on and k_off, which are the only free parameters (15). We found that this model was consistently able to capture our data (Fig. 4), so we proceeded with this simplification of the inactivation kinetics.

Figure 4.

Thermodynamic coupling analysis between D13 and R164. Single current trace from wild-type (WT), single mutants (D13N and R164Q), and double mutant (D13N/R164Q) channels in response to a voltage step from −80 to +60 mV. Red lines overlying the traces represent the best fits of 30 ms current recordings to an inactivation model. The equilibrium constants (K_eq) were estimated from the best-fit parameter values for k_on and k_off. The K_eq values correspond to estimations for each current trace. With these values, the estimated thermodynamic coupling coefficient (Ω) was 0.20 (see Materials and Methods). The center of the cycle shows the location of these two amino acids at the end of the MD simulation.

One at a time, we mutated Shaker K_V channels in 16 positions: eight from the N-terminus, six from the T1 domain and two from the post-S6 region. We report all changes in k_on and k_off with respect of wild-type channels in Fig. S7. Generally, the rate changes introduced by these single mutations were small, affecting the kinetics of inactivation by no more than ∼2-fold. Next, we used double mutant cycle analysis, guided by our model’s predictions as described above (18, 19). Fig. 4 shows the current recordings from one mutant cycle, D13/R164. Each trace is a single current recording in response to a voltage step from −80 to +60 mV. Red lines overlapping the current recordings represent Markov model fits of the ionic current data, giving values for k_on and k_off (15). From these rates, we calculated the equilibrium constant (K_eq) of each construct and used them to estimate the degree of thermodynamic coupling between residues (Ω; see Materials and Methods), where a deviation from unity indicates an interaction. For the current traces shown in Fig. 4, the Ω value for D13/R164 was 0.2, suggesting a close interaction between the two residues.

Table S1 shows the average K_eq values of all single and double mutants used in this study. Using these values of K_eq, we calculated Ω for the 10 selected pairs of amino acids. For those instances in which the value was less than 1, we calculated the reciprocal and then plotted Ω against the average distance between the side chains of each pair of residues predicted in our simulations (Fig. 5). The Ω values for six such pairs drop sharply to unity as the simulation-derived distance between the two amino acids increased beyond ∼5 Å, consistent with such distances being incompatible with direct intermolecular interactions. By contrast, two pairs (D13N/R164Q and Y8A/N482A) with smaller predicted distances exhibited Ω values higher than one. Fig. S8 depicts the values of k_on (top) and k_off (bottom) with respect to wild-type channels of all constructs involved in these two cycles. In both cases, the changes in k_on and k_off of the double mutant channels are not the sum of the effects observed in the channel constructs containing a single mutation. This indicates that the two mutations in each of these cycles are not independent. Fig. S8 also shows an example in which the kinetic consequences on k_on of individual mutations are additive (E35N/K198Q).

Figure 5.

Relationship between thermodynamic coupling coefficients and distances. For those pairs in which the Ω value was <1, the reciprocal was taken and plotted here. Error bars on the y axis represent the propagated mean ± SE of the four K_eq estimations used to calculate each Ω value. Distances (x axis) were measured between the atoms (residue1:atom and residue2:atom) Y8:OH and N482:Nδ2, D13:Cγ and R164:Cζ, R17:Cζ and D193:Cγ, K19: Nζ and S479:OG, R17:Cζ and E194:Cδ, E35:Cδ and K198:Nζ, Q25:Cδ and L170:Cγ, E29:Cδ and R191:Cζ, K18: Nζ and R191:Cζ, and D13:Cγ and R191:Cζ. Distance values represent the mean ± the SD of the entire “full penetration” period of the simulation, except for two cases in which the inactivation gate made contact with more than one subunit (see Fig. S6). In those instances, the average distance and SD were calculated from a period in which contact was with a single subunit. For Y8-N482 pair, this period is between t = 0.71 and 1.125 μs, during which there was contact with subunit B. For the pair D13-R164, contact was with subunit D between t = 0.6 and 1.3 μs.

Interestingly, for those pairs in which the average distances were ∼4 Å, their corresponding Ω values varied from ∼2 to ∼5. This apparent lack of correlation may originate from the fact that inactivation is a complex process governed by multiple interactions at multiple locations on the protein, and the chemical nature of the interactions between each pair is not the same. Thus, although predicted distances less than 5 Å should give rise to Ω values significantly different from unity, the model makes no strong predictions about their relationship within this limited range. In contrast, it makes a very strong prediction that residues above some critical distance threshold should have no measurable interaction, and this prediction is supported by these data, validating the inactivation mechanism suggested by our model.

Discussion

Past experiments have identified the N-terminus as the inactivation gate of Shaker K_V channels (7, 8) and demonstrated several interactions between this domain and residues in the central cavity (9, 10, 57) and T1 domain (14, 15). However, these studies have failed to identify the atomic-scale mechanisms of inactivation as well as the conformation adopted by the N-terminus as it transitions from the intracellular space to the channel pore. Here, we used MD simulations alongside electrophysiological recordings to address these questions. Our study argues against a β-hairpin motif inactivating the channel, mainly because experiments have shown close contact between A2/I470 in the channel pore (9), and analyses of the internal geometry of the permeation pathway in different K_V channels (52) show insufficient space to accommodate such deep penetration of the β-hairpin. Our data instead support a model of the inactivation gate as an extended peptide between residues 1 and 8, followed by a long α-helix from residues 12 to 36. In line with previous findings (9, 10, 57), our model identifies numerous hydrophobic contacts between the extended peptide and the amino acids of the central cavity and inner pore, terminating with a hydrogen bond between Y8 and N482 at the base of the pore. The long α-helix, in turn, serves to stabilize the inactivation particle at the level of the T1 domain through electrostatic interactions. These stabilizing contacts were confirmed by demonstrating a tight correspondence between the interatomic distances predicted by our model and the degree of thermodynamic coupling observed experimentally. This conformational ensemble is consistent with previous reports that the inactivation particle enters the pore through the intracellular entryway above the T1 domain (14, 15), but it goes further than any prior study by showing precisely where and how such a transition may take place.

Once the channel is blocked, our simulations corroborate the conclusion of previous experiments that the inactivation particle interacts with the four I470 residues in the central cavity with equal probability (15). Moreover, MD simulations of the N-terminus with the M1 residue acetylated showed no deep penetration into the pore, suggesting that the N-terminal tip must be protonated to block the channel. This is consistent with the affinity of the central cavity for positively charged ions, like potassium (4) and quaternary ammonium derivatives (10), and rationalizes the functional similarity between the inactivation gate and open-channel blockers (11).

On the other hand, to comply with empirical constraints and allow the tip of the N-terminus to penetrate deeply into the central cavity, we found it necessary to apply a membrane potential of 0.6 V. However, it remains very plausible that such penetration could occur at physiological voltages on longer timescales than those probed by the MD simulations. In fact, permanence of the open-inactivated state of the channel in absence of a transmembrane potential was demonstrated in our model, discarding possible voltage influences in the inactivation mechanism, particularly in the off rate. As computational efficiency and power improves, it will be important to confirm that this is true in the context of realistic physiological conditions. Although 0.6 V is substantially higher than the voltages typically applied in electrophysiological experiments, it is within the range of commonly used electric potentials to simulate membrane proteins (49, 50). Moreover, the fact that the N-terminus was prohibited from entering the channel pore while adopting a β-hairpin conformation, even at higher membrane potentials (1.0 V), casts doubt on the relevance of this structure for inactivation and suggests that a narrower single-stranded conformation of the N-terminus is necessary for full penetration into the channel pore.

Finally, it should be noted that the principal focus of this study was to describe the Shaker K_V channel inactivation mechanism in the open state of the protein; therefore, our analyses do not distinguish the location of the inactivation gate when the channel is closed. However, it is tempting to speculate that many of the same interactions, particularly those between the long α-helix and the T1 domain, also manifest while the channel is in the closed state. This might favor the localization of the N-terminus within the intracellular vestibule even before the channel is open and contribute to rapid inactivation upon opening of the pore. Moreover, taking into account the tetrameric architecture of Shaker K_V channels, a plausible model of the closed state might have all four N-termini residing near the intracellular entryway of the T1 domain until opening occurs.

In conclusion, the present work reveals previously unknown interactions that the N-terminus of Shaker K_V channels establishes with residues lining the ion permeation pathway, with the post-S6 region and with the cytoplasmic T1 domain. The hydrophobic nature of the amino acids threading into the central cavity, as well as the polar region forming the long α-helix in the structural model are homologous to fast inactivation gates of K_V channels from mollusks and mammals (Fig. S9). This provides support that the general mechanistic principles of the process of inactivation are conserved throughout the animal kingdom. Moreover, although the calculations presented here have some limitations (the use of approximate structural models, representation of only the open state of the channel, and the absence of a large portion of the inactivation particle connecting the N-terminus with the T1 domain), they suggest the key interactions stabilizing the bound state of the inactivation particle and dictate the kinetics of Shaker K_V channel inactivation. An improved characterization of the operation of K_V channels should lead to a richer understanding of how neurons sculpt the shape and frequency of action potentials as well as help the development of new pharmaceutical strategies for the treatment of associated channelopathies.

Author Contributions

M.H., A.V.-J., and J.C. conceived the project. A.V.-J. and H.P. carried out computational modeling and simulations. J.C. supervised computational simulations efforts. F.P.-C., A.S., A.d.l.C.L., A.S.L., and M.H. designed, performed, and analyzed the experimental data. A.V.-J., J.C., F.P.-C., A.S.L., and M.H. wrote the manuscript. All authors contributed to the editing of the paper and scientific discussions.

Editor: Philip Biggin.