Closed state structure of the pore revealed by uncoupled Shaker K⁺ channel

Abstract

Voltage gated potassium (Kv) channels regulate processes from cellular excitability to immune response and are major pharmaceutical targets. Despite recent structural advances, the closed state structure of the strictly coupled Kv1 family remains elusive. Here, we capture the structure of the Shaker potassium channel with a closed pore by uncoupling its voltage sensor domains from the pore domain. Structural determination of the uncoupled I384R mutant by single particle Cryo-EM reveals a fully closed pore coexisting with activated, non-relaxed voltage sensors. Comparison with the open pore structure suggests a roll-and-turn movement along the length of the pore-forming S6 helices, contrasting with canonical gating models based on limited movements of S6. These rotational-translational motions place two hydrophobic residues, one in the inner cavity and the other at the bundle crossing region, directly at the permeation pathway, limiting the pore radius to less than 1 Å. The selectivity filter is captured in a noncanonical state, partially expanded at G446, unlike previously described dilated or pinched filter conformations. Together, these findings suggest a reinterpretation of the mechanism of activation gating for strictly coupled Kv1 channels, highlighting the strictly sensor-pore coupling that underlies different functional states.

Introduction

Structural studies on a variety of Kv channels have provided a great deal of information on the conformational landscape of the channels as a whole. However, the vast majority of the structures correspond to activated or inactivated conformations, since they were obtained in the nominal absence of an electric field. In strictly coupled Kv channels, the electromechanical coupling (EMC) between the voltage-sensing domain (VSD) and the pore domain (PD) is obligatory: channel opening or closure requires activation or deactivation of the voltage sensors, respectively^1,2,3. This is in contrast to a typical allosteric coupling, as in BK⁴, HCN⁵, or hERG^6,7 channels, for instance. While this tight coupling ensures a remarkably low leak potassium conductance at rest, it has also hindered the structural understanding of the gating mechanisms, since at 0 mV the voltage sensors typically populate the activated (Up) conformation, with the pore domain displaying an open inner bundle gate. As a result, the current gating model of strictly gated Kv channels, such as the Kv1 and Kv2 families, relies primarily on bacterial potassium channel structures⁸ and electrophysiological data^9,10.

A conserved isoleucine I384 in the S4-S5 linker region was identified as pivotal for the electromechanical coupling in Kv1 family¹¹. Here, we report the closed structure of the pore in a strictly coupled Kv1 channel, Shaker K⁺ channel, by disrupting the coupling between the pore and the voltage sensors from mutations at I384. Single-channel, ionic current, gating current, and fluorescence measurements demonstrated that mutations to the side chain of I384 directly affect the strength of electromechanical coupling. In the most extreme case of I384R, the VSDs become completely uncoupled from the PD: the pore remained closed within a large range of voltages (−120mV to 180 mV) while the VSDs activate/deactivate independently. Single-particle cryo-EM structure of the uncoupled channel revealed a collapsed permeation pathway, consistent with a fully closed conformation. Site-directed fluorimetry measurements utilizing a fluorescent unnatural amino acid (UAA) strongly support our structural observations. Structural rearrangements were also observed in the selectivity filter (SF) and the voltage sensors likely underlying the structural basis for activation-inactivation coupling and hysteresis of the VSDs, respectively. Modifying the canonical hinge model, we propose a roll and turn gating model for strictly coupled Kv channels and a molecular mechanism of interactions among different conformational states.

Results

Conserved isoleucine controls electromechanical coupling in the Kv1 family

A systematic mutagenesis survey of the S4-S5 linker, a region that has been demonstrated to be important for EMC^12,13,14, led to the identification of a single residue that is able to tune the coupling strength between VSDs and PD. The conserved isoleucine 384¹¹ is located at the N-terminus end of the S4-S5 linker and forms elaborate interactions with the intracellular end of the pore-forming S6 helix¹⁵ (Fig. 1A, B). Replacing I384 with smaller residues such as alanine or cysteine, strengthened the already efficient electromechanical coupling (Fig. 1C and Supplementary Fig. 1). I384C, for instance, activated in more negative voltages with much steeper voltage dependency, as illustrated by its conductance-voltage (GV) curve. More importantly, channel activation closely followed the voltage sensor movement, measured in the gating-charge-voltage (QV) curve (Fig. 1D). While in wild type (WT) channels, the difference between the V_1/2 for QV (measured with the nonconductive W434F mutant¹⁶ and GV curve was > 25 mV (Fig. 1E, F, Supplementary Fig. 1, and Supplementary Tables 1, 2), that difference was less than 7.5 mV in I384C, a hallmark for strengthened electromechanical coupling. On the other hand, mutating I384 to glutamate, asparagine or leucine led to severe uncoupling between the VSD and PD (Supplementary Fig. 1). This is characterized, in the case of I384L (Fig. 1G), by large GV curve shifts (>70 mV) to more depolarized potentials, when compared to WT Shaker, together with a shallower slope in the GV curve (Fig. 1H). Strikingly, while the GV was shifted to the right, the QV curve of I384L was shifted in the opposite direction (Fig. 1H). Indeed, similar changes in the current kinetics and GV curves were also seen in hKv1.2 and hKv1.3 with mutations at equivalent positions (I316 position in hKv1.2 and I386 position in hKv1.3), suggesting that I384’s critical role in the electromechanical coupling is conserved among strictly coupled Kv channels (Supplementary Fig. 2).

Fig. 1: I384 residue controls electromechanical coupling in the Shaker potassium channel.

A Structure of the Shaker channel (PDB:7SIP). Voltage-sensing domains (VSD) and pore domains (PD) and S4–S5 helixes are labeled. B Location of I384 at the VSD–PD interface and its contacting residues. C, D The I384C mutant shows markedly slower deactivation compared to WT, with GV curves (N = 4) sharper and more closely aligned with QV (N = 5). Inset shows the tail current of I384C in black and WT in red returning to −120 mV. E, F WT displays typical ionic and gating currents, with GV and QV (n = 3 for both) separated. G–H I384L requires stronger depolarization for activation, with a shallower GV slope (N = 4) and altered gating current (N = 5) The inset highlights the coexistence of both gating and ionic current at the beginning of the pulse. I, J Fluorescence recordings from a TMR labeled I384L (A359C) suggest normal VSD movements (N = 3 for GV/QV/FV curves). Note there is no slow component in the fluorescence traces. K Single-channel recordings from I384L reveal unchanged conductance but increased flickering. L, M I384R abolishes ionic conduction in 12 external K⁺, leaving only gating currents, with the QV shifted strongly to negative voltages (N = 4 for I384R). N Gating current kinetics of I384R are faster than WT in the W434F background (N = 3 for both). All the data are shown as Mean ± SEM and N is the number biological replicates. For C, E, G, prepulse and returning pulse are both -120 mV and ∆V = 10 mV. For D, F, H, the data were fitted with a two-state model (details in method section) for easy visualization and calculation of change in ∆V_1/2. Fitting results could be found in Supplementary Tables 1-2.

Given the large V_1/2 gap between GV and QV in I384L, gating currents were easily resolved in the presence of ionic currents (Fig.1 G inset). To confirm that no additional slow component in the VSD movements was present in I384L that is responsible for the opening of the pore, tetramethyl rhodamine (TMR) was introduced at A359C in the extracellular loops of the VSD so that the movement of the voltage sensors would be evaluated via its fluorescence signal^17,18,19. The fluorescence records showed no additional component in the VSD movement, and the voltage-dependent fluorescence (FV) curve fully overlaps with the QV curve (Fig. 1I, J). This result demonstrates that our gating current measurement reflects the true movement of VSDs. To confirm that the observed effects were indeed a consequence of impaired coupling rather than due to changes in single channel properties, we performed noise analysis and single channel recordings^20,21 in I384L mutant. Noise analysis demonstrated that at 195 mV, the maximum opening probability was less than 70% (Supplementary Fig. 3) and single channel recordings showed that the unitary conductance level in I384L (Fig. 1K, ~10 pS, Supplementary Tables 3-4) was similar to the WT (~12 pS)²¹, but channel flickering was increased (Fig. 1K and Supplementary Fig. 3).

While in I384L, the impairment in the coupling was severe, we discovered another mutant that completely uncouples the voltage sensors from the pore, I384R. Despite the presence of K⁺ ions, I384R showed no ionic conduction, and only gating current could be recorded (Fig. 1L). The gating current itself activated at more hyperpolarized voltages compared to WT (Fig. 1M) and was kinetically faster compared to the WT with W434F background (Fig. 1N), as if an energetic load was taken off from the voltage sensors. This is also consistent with left-shifted QV curves seen in other uncoupled mutants such as I384N, I384L and I384E (Supplementary Fig. 1, and Supplementary Tables 1, 2). In I384R, our recordings showed no discernible K⁺ current from −120mV to +180 mV, demonstrating the exceptional structural stability of the closed pore. Additionally, a left-shifted QV curve allows the VSDs to transition more easily to the up conformation. As a result, at 0 mV, I384R possesses a stably closed pore and four activated VSDs with little structural heterogeneity. This represents an appealing target for structural investigation.

Closed state structure revealed by a completely uncoupled channel

We expressed, purified, and solved the structure of Shaker-IR-I384R, by single-particle cryo-EM. The structure was globally resolved to 3.5 Å, with clear densities for all the transmembrane helices (Fig. 2A, B, Supplementary Figs. 4 and 5). Similar to previously determined structures, the uncoupled channel assembled as a domain-swapped homotetramer (Fig. 2A, B)¹⁵. However, unlike the strictly coupled WT channel structure, captured with an open pore, the uncoupled channel clearly displayed a collapsed permeation pathway. (Fig. 2C, D). Compared to the open conformation of Shaker, the pore-forming S6 helices in the closed pore underwent a roll and turn movement, where a translational movement brings the backbone of the S6 helices closer together (Fig. 2C), and a rotational movement places hydrophobic residues I470 and V474 directly into the pore. These form the two narrowest points of the closed pore (Fig. 2D). Pore radius calculations showed a closed permeation pathway with radius less than ~1 Ų², leaving ion conduction an impossibility (Fig. 2E, F). Unexpectedly, the rotating-in of the I470 residue led to a total collapse of the water-filled inner cavity underneath the selectivity filter (Fig. 2E, F). Previous gating models predict a more limited hinge-like movement for the channel activation where a kink is created in the middle of the S6 helices around the conserved PVP motif and the intracellular half of the helices crosses or separates to close or open the channel^{9,10,23,24,25,26}.

Fig. 2: Cryo-EM structure of uncoupled I384R channel, captured in a closed state.

A Side view of I384R. The model is shown overlapping with the electron density, with the intracellular T1 domain omitted. Global resolution is 3.5 Å. B Intracellular view of I384R. Like previous solved structures, I384R adopts a domain-swapped arrangement. C Structure of the pore in I384R (green) compared to the open state (orange, PDB:7SIP) from the side view. A rotational movement in upper S6 helices (highlighted by arrows) positions two hydrophobic residues, I470 and V474, from the side directly into the permeation pathway. D Intracellular view of the pore in I384R (green) and WT (orange). In addition to the rotational movement, a translational movement of S6 helices towards the pore is observed in the closed state structure, further constraining the pore. E Radius profiles of the open pore from the WT channel. The inner cavity and bundle crossing are both open. The dashed vertical line indicates the approximate size of hydrated K⁺ ion. F Radius profile of pore in the uncoupled channel. The pore shows two constriction sites, one at I470 in the inner cavity and the other at V474 in the PVP motif. The Narrowest point of the pore is less than ~1 Å, indicated by the dashed red line. Clearly, the pore is captured in the closed state. G ANAP is a fluorescent unnatural amino acid sensitive to the hydrophobicity of its local environment. With the filter set used in this work, a more hydrophobic environment leads to an increase in fluorescence signal. H Representative ionic traces from I470ANAP. ANAP is incorporated at position 470 in a site-specific manner through amber stop codon suppression. I Fluorescence signal from I470ANAP. A transient signal could be seen among voltages where the channel opens (above −40 mV). A slightly slower signal is seen at the repolarizing pulse as well (highlighted by a red dashed line). At more negative voltages, however, no such signal could be resolved. It seems the observed fluorescence signal is associated with the opening and closing of the channel.

To validate our structural observations, we set out to measure the local conformational changes around position 470 in the WT channel utilizing a fluorescent unnatural amino acid probe. ANAP is comparable in size to a tryptophan, and its fluorescence changes according to the hydrophobicity of its local environment (Fig. 2G)^27,28. Utilizing the amber stop codon suppression method²⁹, we incorporated ANAP in a site-specific manner at the 470 position and recorded the ionic current and fluorescence signal simultaneously from I470ANAP (Fig. 2H, I). As channels opened, a fast transient fluorescence change was observed at the start of the depolarizing pulse and a slower one at the beginning of the hyperpolarizing pulse (Fig. 2I, highlighted with a dashed line). The transient nature of the fluorescence signal seems to suggest a rotational movement where the different environments are sampled before reaching its final state, consistent with our structural observations. In the negative voltage range (<-40 mV), where the channel does not open, no fast transient fluorescent signal was observed, demonstrating that the fluorescence signal at I470 is only observed when the channel opens. These results are fully consistent with the idea that the structure of the uncoupled channel most likely represents a true closed state of the pore and suggests interactions between the bundle crossing region and the selectivity filter.

A tripartite interaction pocket essential for electromechanical coupling

Globally, I384R does not cause a kink in the S4-S5 linker or a local movement at the elbow region as was seen before in the resting bacteria Nav channels structures³⁰. Instead, we see a lateral, translational movement along the whole length of the S4-S5 linker towards the pore (Fig. 3A). Closer inspection revealed a tripartite interaction pocket among the N-terminus end of the S4-S5 linker, S6 helix in the same subunit and the C-terminus end the S4-S5 linker helix from the adjacent subunit (Fig. 3B). In the strictly coupled WT structure, I384 was securely lodged in a hydrophobic pocket formed by F484 and Y485 within the same subunit in the S6 helix (Fig. 3B – orange color). The hydroxyl group of Y485, on the other hand, interacted intimately with R394 and E395 in the S4-S5 linker from the adjacent subunit, establishing a structural coupling between the pore and the S4-S5 linker (Fig. 3C – orange color). In the uncoupled I384R structure, however, these interactions are all abolished. The full positive charge of the introduced arginine at position 384 forces itself out of the hydrophobic pocket where the side chain swings towards the adjacent S4-S5 linker. This conformational rearrangement pushes the adjacent R394, facing towards the pore previously, away from the S6 helix and allows for the formation of a salt bridge between R384 and E395 (Fig. 3B, C – green color). These newly formed interactions allow for a closer interaction among the S4-S5 linkers, creating a tight collar around the S5 and S6 segments, stabilizing the closed state, and abolishing the previous interactions with Y485 in the S6 helices. The observed disruption of the tripartite interactions likely underlies the structural basis for the uncoupling mechanism of I384R, similar to what was shown physiologically elsewhere^31,32.

Fig. 3: A tripartite interaction pocket for electromechanical coupling and the structure of activated but not relaxed voltage sensors.

A Intracellular view of the uncoupled structure (green) and coupled structure (orange) with their S4-S5 linkers highlighted. In the uncoupled structure, S4-S5 linkers undergo a translational movement towards the pore, forming a much tighter collar around the S6 helices. B, C In the coupled structure (orange), Y485 at the bottom of the S6 helix interacts with E395 and R394 in the C-terminus end of the S4-S5 linker from the adjacent subunit. I384 from the same subunit is lodged in a hydrophobic pocket formed by F484 and Y485. In the uncoupled structure (green), however, R384 jumps out of the hydrophobic pocket and forms a salt bridge with E395 in the adjacent unit, which repulses R394 away from facing the S6 helix, abolishing completely the interactions with Y485. D Side view of the VSD of the uncoupled channel and the WT channel. E Coulombic density for the VSD in I384R. All the side chains of the gating charges can be clearly resolved. F Comparison of the WT and I384R voltage sensors. In both cases, all gating charges (R362, R365, R368, and R371) have moved passed the hydrophobic plug around the I287 and F290 region. Both VSDs are in a fully activated state. G, H Hysteresis of the VSDs in WT and I384R (N = 4 for both). Holding at 0 mV for a prolonged time (>15 s) leads to a ~20 mV shift of the QV curve to the left in the WT channels. However, in I384R, no such shift was observed. It seems that at 0 mV, the VSDs in I384R do not enter the relaxed state. I Helical movements in S4 due to the S4-S5 linker. The tight conformation of the S4-S5 linker shifted the C terminus of the S4-S5 linker by 4.1 Å. This shift is transduced to the N-terminus end of the linker and the S4 helix as well, shifting them 1.7 Å and 1.4 Å, respectively. Data are shown as Mean ± SEM. N is the number of biological replicates.

Voltage-sensing domain captured in its activated but not relaxed state

Gating current measurements demonstrated that at 0 mV, the QV curve of I384R had reached its maximum, suggesting that all the voltage sensors had activated. However, since there are multiple intermediate states for the VSDs, the electrophysiological data cannot unequivocally define whether the voltage sensors reach the fully activated state in the uncoupled channel, or even if they move in a similar way as WT channels. To address this, we compared the VSD structures in I384R and the WT (Fig. 3D). Clear densities were resolved for all gating charges in S4 (R362, R365, R368, R371)^33,34 as well as the key residues that form the hydrophobic plug in S2 (I287, F290)^35,36,37 and the countercharge (E283) (Fig. 3E). The I384R structure shows that all four gating charges have moved pass the hydrophobic plug in the uncoupled conformation and are accessible to the extracellular solution (Fig. 3F), as seen in the WT structure. This suggests that in the uncoupled channel, voltage sensors move similarly to the WT channels to reach the fully activated state.

However, unlike the WT, voltage sensors in I384R do not appear to enter the relaxed state, a conformation that has been observed in Kv, Nav, Cav channels, and voltage-sensitive phosphatases (VSP) and is driven by prolonged depolarization^{38,39,40,41,42,43}. In WT channels with the W434F background, holding the channels at 0 mV for extended periods of time shifts the QV curve almost 20 mV to more negative potentials when compared to holding at -90 mV (Fig. 3G). This relaxed state was not observed in the uncoupled channel. Holding I384R at 0 mV for > 15 s did not cause significant shifts in the QV curve, suggesting that the voltage sensors in I384R did not enter the relaxed state, at least at this voltage (Fig. 3H). Since the cryo-EM structures were captured at 0 mV, the physiological evidence would then argue that the structure of VSDs in the open channel represented the relaxed state, while in the uncoupled channel, the VSDs likely resided in the non-relaxed state. Structurally speaking, the major difference between VSDs in the WT and I384R lies mostly in the lateral displacement of the S4 and S4-S5 linker alpha helixes. In I384R, the S4-S5 linkers displayed a considerable lateral shift, particularly at the C-terminus end of the helix, 4.1 Å away from the WT structure (Fig. 3I). This movement was transduced to the N-terminus end of S4-S5 linker and the S4 helixes, dragging them 1.7 Å and 1.4 Å away from the open state structure, respectively. Since entry into the relaxed state has been associated with opening the pore¹¹, it is possible that the observed helical displacement in I384 formed the structural basis for the relaxed state of the VSDs.

Noncanonical conformation of the selectivity filter and the decreased volume of the closed state channel

One surprising observation of the I384R closed state structure comes from the selectivity filter conformation. Instead of the now classical linear coordination of 4 K⁺ ion densities seen in the conductive filter (Fig. 4)^15,44, only 2 putative bound K⁺ ions, at the S2 and S4 sites of the SF were resolved (Fig. 4A). It is intriguing to see that the S3 K⁺ was absent in the structure, since generally, the S3 position typically displays the strongest K⁺ occupancy in K channel structural determinations⁴⁵. In our case, the coulombic density suggests that occupancy was similar for these two positions with a slightly higher occupancy at S4 position (Supplementary Fig. 6). Structurally, two small twists were observed at the S1 and S4 binding site when compared to the WT structure (Fig. 4C). The carbonyl group of the last glycine in the TVGYG selectivity filter, G446, flipped away from the pore, directly altering the binding site at S1 position (Fig. 4D). A similar twist was also observed at the bottom of the selectivity filter at the T442 position, which might account for the slight shift of the K⁺ ion at the S4 position compared to the WT.

Fig. 4: Noncanonical conformation of the selectivity filter and global decrease in protein volume in the closed state.

A The SF captured in a closed state. The atomic model is overlaid with the density map. Clearly, there exist only two K⁺ in the SF. B Conductive SF captured in the open state. (PDB:7SIP). C Overlay of the SF in the closed (green) and open state (orange). In the closed state, K⁺ ions are seen bound at S2 and S4 positions. D Difference in SF between open and closed state structure. The carbonyl on the backbone of G446 flipped away from the permeation pathway, abolishing the S1 binding site on the top of the selectivity filter. A similar twist is seen at the bottom of the SF at the T442 position, resulting in a small displacement of the threonine side chain. E The channel is more expanded in the open state (orange) compared to the closed state (green). The expansion can be seen in the VSDs as well as the pore. F Cross-section area calculation utilizing CHARMM_GUI.

Another intriguing observation in the closed state structure was the decrease in the protein volume in the transmembrane region. When comparing the WT with I384R structures, we noticed that the protein expanded laterally in the open state (Fig. 4E). This expansion was due to the translational movement of the S4-S5 linker and the S6 helices. Area calculations with CHARMM_GUI show an asymmetric increase of the cross-section area of the open channel compared to the closed one (Fig. 4F)⁴⁶. The most significant expansion happened around the I470 and V474 region, where the cross-section of the open channel increased by almost 10% (Supplementary Fig. 7). This expansion in volume might be the underlying mechanism of the reported mechanosensitivity of the Kv1 channels^47,48.

Discussion

Electromechanical coupling in voltage-gated ion channels and its energetics

Most voltage-gated channels share two basic functional modules: the voltage-sensing domain and the pore domain. The process of electromechanical coupling (EMC) describes the communication between these two modules. In the present study, we identified a tripartite pocket that, we argue, is essential for EMC in the Shaker potassium channel, a strictly coupled channel. Structurally, a set of interactions among Y485, F484, I384, E395, and R394 establishes the functional connectivity between the pore and the S4-S5 linker. These intersubunit interactions likely contribute to the cooperativity of the voltage sensors and the pore opening as well⁴⁹. It has been demonstrated previously that reducing the side chain volume at positions Y485 and F484 in the S6 inner bundle gate leads to shallower and right-shifted GV curves^11,50,51. This is similar to what we show here in I384E, I384L, and I384N. Mutagenesis experiments and thermodynamic cycle analysis among E395, R394, and Y485 have confirmed their energetic coupling and demonstrated their importance for electromechanical coupling^31,32,52. This tripartite pocket is thus likely responsible for the efficient coupling between voltage sensor movement and pore opening seen in Kv channels.

It has long been debated whether the EMC is energetically favorable for the pore to stay in the open or the closed state. In other words, are the voltage sensors doing work to pull the channel open or to push to keep the pore closed? While some computational work suggests the pore prefers to stay open in the absence of an external energy bias⁵³, our results argue otherwise for the Shaker K⁺ channel. In the I384 mutants, all the uncoupled mutants (I384L, I384E, and I384N) show a right-shifted GV curve and a left-shifted Q-V curve. This behavior is expected if an energetic load to the sensor is decreased. For partially uncoupled mutants, our results indicate that the pore is now less firmly coupled to the VSD, making it more difficult to open for a given charge movement, preferring to stay in the closed state.

In a fully uncoupled mutant, it is expected that the movement of the voltage sensor is independent of pore opening; therefore, the left-shift of the QV curve observed in such a mutant would reflect the energy required to open the pore. However, in the case of the I384R mutant, the presence of four salt bridges between R384 and E395 introduces an additional stabilization of the preopen state, a bias that is not expected to be present in the normal operation of the channel. While the newly formed interactions stabilize the closed state of the pore, they are likely to influence the voltage sensor movements as well, given the intimate connection between the voltage sensors and the S4-S5 linkers. The QV curve of I384R is shifted leftwards compared to the WT, indicating a lessened energy load, which was estimated to be at 3.41 ± 2.25 kcal/mol using the V_median approach⁵⁴. While it is tempting to conclude that this would be the energy for the pore opening, the presence of the aforementioned salt bridges invalidates this assumption. Most likely, the energy estimated has not only contributions of salt bridges, but also all the other energetic components present in the gating process.

Kv channel structure captured with a closed pore

As the most extreme case among the uncoupling mutants studied here, I384R only displays gating currents under voltage clamp. This is quite different from the classical W434F mutant¹⁹, even though they both have minimal ionic conductance. The W434F mutant speeds up the C-type inactivation and stabilizes the channel in the inactivated state⁵⁵. Thus, I384R represented an ideal candidate to probe the conformation of the inner bundle gate, and we were able to capture the closed pore structure of the Shaker potassium channel by disrupting its EMC (Figs. 1, 2). It is worth pointing out that this uncoupled pore most likely represented the pore domain in the closed state, instead of an inactivated state, another nonconductive state in Kv channels. Indeed, the relative shifts in QV and GV curves (Fig. 1 & Supplementary Fig. 1) strongly argue that substituting isoleucine at position 384 exerts a direct effect on EMC. There are informative differences in the behavior of the gating currents measured in mutants W434F and I384R. In I384R, the gating current was significantly faster, consistent with a sharp reduction in EMC between PD and VSD, as the sensor is free to move in the absence of a mechanical load. This suggests that we are not tampering with the PD itself and therefore, the structure of the closed pore likely represents the true closed state of the pore in WT channels. In contrast, gating currents in W434F are a reflection of the VSD movement under a physiological load. The suppression of ionic currents is derived from effects downstream to the inactivation gating (C-type inactivation).

Another issue that needs to be addressed is whether the I384R mutant represents the Shaker’s native pore domain closed conformation. We believe this is the case. While new salt bridges were observed in the I384R structure, these newly formed interactions are not necessary to uncouple the VSD from the PD, as significant uncoupling was observed with I384E, I384N, and even I384L (Supplementary Fig. 1). These salt bridges likely stabilized the closed pore, but the major driving force for the uncoupling came from the disrupted tripartite pocket. A surprising finding from the closed pore structure is the large degree of conformational changes happening along the entire length of the S6 helices, leading to the collapse of the inner cavity. In contrast to the simple hinge-like movement seen in prokaryotes^9,56. Inner gate opening in strictly coupled eukaryotic Kv channels appears to be more complex, with both a lateral movement of the PVP motif and an additional rotational movement at the inner cavity of the pore. This type of movement leads to the collapse of the inner cavity in the closed state, where hydrophobic I470 rotates and points directly into the permeation pathway. Such a mechanism might provide additional insights into the pharmacology of channels, since many pharmacological agents specifically bind to this region. It is worth pointing out that our current closed state structure presents a very different pore compared to the recently resolved Kv2.1 under electric field ⁸⁹ (Supplementary Fig. 9). When compared with the present structure, the I384R closed conformation shows a more significant S6 movement. This leads to a completely closed permeation pathway compared to the polarized conformation of Kv2.1. It is thus possible that the Kv2.1 structure represents an intermediate pore conformation instead of a fully closed one.

Our fluorescence experiments with ANAP directly demonstrated that as the pore opens and closes, a conformational change takes place near position 470, providing additional evidence that I384R represents a closed pore seen in WT channels. Our results are also consistent with early pharmacological studies with internally applied tetraethyl ammonium (TEA) ion⁵⁷. Mutating I470 to a smaller residue like cysteine allows TEA to stay in the pore domain inner cavity in the closed state⁵⁸. The transient nature of the ANAP fluorescence signal indicates that at least two different processes happened at position 470. A naïve interpretation would be that rolling and turning movements created two fluorescence signals with different polarities. While the exact conformational changes and potential existence of intermediate states require further investigation, particularly taking recent studies into considerations^59,60, it is certain that during activation, significant conformational changes occur at the inner cavity of the channel.

Conformation of the selectivity filter in the closed and open states

The selectivity filter (SF) in the I384R closed state structure was captured in what appears to be a noncanonical conformation. It deviates from both, the dilated inactivated filter where that abolishes S1 and S2 binding sites (Supplementary Fig. 8A)⁶¹ and the pinched inactivated filtered captured in prokaryotic Kv channels^6,60 (Supplementary Fig. 8B). In I384R, two bound K⁺ ions were found at S2 and S4 position in the absence of any expansion at Y445 position or pinching at G444⁶² (Supplementary Fig. 8). While it is possible the observed conformational state is a consequence of the forced uncoupling between VSD and pore domain, we would like to argue otherwise. Changes in SF filter conformation have been reported previously associated with different functional states of the channel^5,60,63. Furthermore, previous fluorescence measurements seem to be in line with this hypothesis; TMR-labeled channels at the T449 position, around the selectivity filter region, gave rise to a fluorescence change when the channels opened (appearance of ionic currents) and the FV curve followed very closely the GV curve¹⁷. It is possible that the S6 movements during activation are allosterically communicated to the selectivity filter, triggering the transition from the noncanonical conformation to the typical conductive conformation⁶⁴. Additionally, given that ion occupancy was only observed in S2 and S4 positions, the current SF conformation appeared to be more in favor of a soft knock-on model of permeation. However, further investigation is certainly required before a definitive conclusion can be reached.

Isoleucine 470 is a shared structural element for both activation and inactivation

Pore radius calculation identified I470 in the inner cavity as the upper gate in the closed state structure, a residue that has been mostly associated with the C-type inactivation of Kv channels. Physiological experiments mutating I470, together with T449, demonstrated that the inactivated state was rendered conductive by the double mutations⁶⁵, and computationally, long-time scale simulation also observed a coincidence of I470 movement and closure of the permeation pathway in the inactivated channels⁶⁶. Moreover, structural results from Kv4.2 identified the role of the equivalent isoleucine in inactivation mechanisms⁶⁷. In the case of the Shaker K⁺ channel, the inactivated state is entered after the activation of the channel, with the open inactivated state as the most dominant inactivated state in the channel^68,69,70. It appears that I470 is a shared structural element for both activation and inactivation and could serve as the link for the interplay of these two functional states.

Mechanosensitivity of Kv channel gating

In the closed state structure, we saw a collapsed inner cavity in the center of the protein, which led to a global shrinkage of the protein volume. Compared to the open state structure, the protein occupied less area on the membrane in the closed state. It is worth noting that the open state structure was solved in lipid nanodisc, and the current closed state structure was solved in detergents. The difference in structural determination method may result in an underestimation of the actual change in protein volume between closed and open states, since it has been reported that reconstitution of proteins into nanodisc could introduce additional lateral forces from the lipid bilayer to the proteins⁷¹. Therefore, it is possible that the actual cross-sectional area of the channel in the closed state would be smaller, and the actual expansion from closed to open would be more significant in a physiological condition.

Interestingly, the observed decrease in the protein volume (and membrane footprint) between open and closed conformations predicts that to open the channel, additional energy is required to push against the lateral pressure from the membrane. Consequently, membrane lateral pressure should modulate the gating of the channel, giving rise to some mechanosensitivity to the channel. Some physiological experiments support this prediction. Excised patch clamp experiments have demonstrated that the GV curve of Shaker-IR-WT channels displays a leftward shift upon suction application in the pipette^47,72,73. This mechanical modulation of channel gating seems to be mostly limited to changes in V_1/2 (reflecting changes in steady-state energy levels), without altering the effective number of gating charges. This putative expansion of the protein footprint could underlie the molecular basis of the mechanosensitivity in strictly coupled Kv channels and possible other members of voltage-gated channels, such as Nav channels⁷⁴.

Activation mechanism for strictly gated Kv channels

The present experimental results argue for a revised gating model in strictly gated Kv channels (Fig. 5). At very hyperpolarizing voltage, the channels occupy the deepest closed state, where all four VSDs are in the down state. In this deep, closed state, the permeation pathway is gated by I470 in the inner cavity as well as V474 at the bundle crossing region. Upon depolarization, asynchronous activation of the VSDs moves the channels into an ensemble of intermediate closed states, where some but not all VSDs are activated. The movement of the S4 in the VSDs creates a direct pull on the S4-S5 linker. However, this pull does not provide enough energy to open the channel, and the VSDs are stabilized in the non-relaxed state. When all four voltage sensors move up, the channels enter a transient pre-open state, which is represented by the structure of the uncoupled channel. In this pre-open state, all the VSDs populate the activated (up) conformation, which provides enough energy into the linkers to drive the pore into its open conformation. As part of a last concerted movement, S4-S5 linkers expand, allowing the opening of the pore and the relaxation of the VSDs. During opening, the S6 undergoes a roll and turn movement that rotates away the upper and lower activation gates, creating a water-filled permeation pathway. This movement is allosterically communicated to the selectivity filter, leading to a small local conformation change, allowing for the selective permeation of the K⁺ ions and predisposing itself for C-type inactivation. This allosteric communication forms the structural and molecular basis for interaction among different functional states.

Fig. 5: Proposed activation mechanism and interactions with different functional states in Kv1 channels.

The progression to activation is from left to right. In the deepest closed state of the channel, all the VSDs are down, and the SF likely resides in the noncanonical state (1). The closed pore is sealed closed by hydrophobic I470 and V474 (1, 2). Upon depolarization, VSDs transit to an active but not-relaxed state (3). The upward movement of the VSDs creates a pull on the S4-S5 linker, yet since not all VSDs are up, the energetic input is not enough to open the pore (4). In the last closed state before opening, or the pre-open state, all four VSDs move up (5), creating enough pull on the S4-S5 linkers (6). In the case of I384R, the channel is most likely stabilized in this pre-open state with newly introduced salt bridges. From the pre-open state, the last concerted movement happens, and the S6 helices undergo a roll and turn movement, opening up the permeation path and expanding the channel laterally (8). At the same time as the pore opens, the lateral movement of the S4-S5 linker drives the VSDs into a different state, the relaxed state (7).

Methods

Site-directed mutagenesis and cRNA in vitro synthesis

Shaker zH4 K+ channel with fast inactivation removed (∆6–46 - Shaker-IR⁷⁵)in pBSTA vector, flanked by β-globin sequences, was used in this study for all the physiological experiments. Point mutations were generated using mismatched mutagenic primers. The PCR product was digested with DpnI to remove the template and was used to transform the XL-gold ultra-competent cells. After ampicillin resistance screening, plasmids were purified from the colonies using standard miniprep protocols. Purified plasmids were sent for either Sanger (Genomics Facility at University of Chicago) or nanopore (Plasmidsaurus) sequencing to confirm the introduction of the point mutation and the absence of off-target mutations. cRNA was then transcribed in vitro from the linearized plasmids (T7 RNA expression kit; Ambion Invitrogen, Thermo Fisher Scientific, Waltham, MA).

Xenopus laevis oocyte preparation and channel expression

Ovaries of Xenopus laevis were purchased from XENOPUS1 (Dexter, Michigan). The follicular membrane was removed using collagenase type II (Worthington Biochemical Corporation) 2 mg/mL with bovine serum albumin at 1 mg/mL (BSA). After defolliculation, stage V–VI oocytes were then selected and microinjected with 5-100 ng cRNA. Injected oocytes were incubated at 18 °C for 1-5 days in SOS solution (in mM: 100 NaCl, 5 KCl, 2 CaCl₂, 0.1 EDTA, and 10 HEPES at pH 7.4) supplemented with 50 µg/mL. Unless otherwise stated, all chemicals were purchased from Sigma-Aldrich.

Cut-open voltage clamp on Xenopus laevis oocytes

Macroscopic ionic and gating currents were recorded using the cut-open voltage clamp technique⁷⁶. Micropipettes filled with 3 M CsCl or NaCl, with resistance between 0.4 and 0.8 MΩ, were used to measure the internal voltage of the oocytes. Current data were online filtered at 20 kHz with a low-pass 4-pole Bessel filter and sampled by a 16-bit A/D (USB-1604; Measurement Computing, Norton, MA) converter at 1 MHz. All experiments were conducted at room temperature (~17 °C). For ionic current experiments, unless otherwise stated, were conducted in an external solution consisting of in mM: 12 K methylsulfonate (MES), 108 N-methyl-D-glucamine (NMG) MES, 2 Ca MES, 10 HEPES, 0.1 EDTA, pH = 7.4, and internal solution consisted of in mM: 120 K MES, 10 HEPES, 2 EGTA, pH = 7.4. The capacitive transient was manually compensated with a dedicated circuit and, in some cases, further removed by an online P/-4 protocol with a holding voltage of either −80 or −90 mV⁷⁷. For gating current experiments, all experiments were conducted in external solutions consisting of in mM: 120 NMG MES, 2 Ca MES, 10 HEPES, 0.1 EDTA, pH = 7.4, and in internal solution consisted of in mM: 120 NMG MES, 2 Ca MES, 10 HEPES, 2 EGTA, pH = 7.4. The gating current was recorded with a W434F background with the exception of I384R, which by itself produced no discernible ionic current.

Single-channel recordings and noise analysis

Single-channel recordings and noise analysis were performed on excised inside-out Xenopus Laevis oocyte patches. Briefly, the cells were injected with 5 ng of RNA and maintained at 12 °C the day before the experiment. To remove the vitelline membrane, oocytes were incubated for 5 minutes in a hypertonic solution (SOS solution supplemented with 300 mM sucrose). This procedure shrank the oocyte, separating the plasma membrane from the vitelline membrane, making the mechanical removal of the vitelline membrane easier without compromising the integrity of the cell. After removal of the vitelline membrane, oocytes were immediately and gently washed three times with intracellular solution containing in mM: 120 KMES, 2 EGTA, 10 HEPES, pH = 7.40. They were placed in a recording chamber under an inverted microscope. Current was recorded with an Axopatch 200B patch-clamp amplifier. The pipette solution consisted of, in mM, 12 KMES, 108 NMGMES, 2 KCl, 2 CaMES, 10 HEPES, 0.1 EDTA, pH = 7.40. The resistances of the tips were between 13~17 MΩ for single-channel experiments and 6~10 MΩ for noise analysis experiments. To reduce stray capacitance, the tips of the pipette were covered by Sylgard 184 (Dow Corning Corporation). Current was filtered with a digital 8-pole Bessel filter set at 10 kHz (3384 Krohn-Hlite).

For noise analysis, we applied hundreds of depolarizing voltage pulses. The ionic currents elicited were averaged to obtain the mean. The variance and the mean were obtained using our Analysis software. For single channels, we recorded several current traces from voltage pulses to +140 mV. The histograms were obtained using Analysis.

Unnatural amino acid incorporation and voltage-clamp fluorimetry

To incorporate ANAP, we utilized the amber suppression technique. Briefly, the oocytes were injected with a mixture of ANAP-synthetase, ANAP methylester, ANAP-tRNA, Xenopus release factor 1 with D55E mutation, and messenger RNA encoding the channel with the amber stop codon introduced at I470 position^27,28,78. The oocytes were incubated in the dark for 3 to 5 days prior to recording. The voltage clamp fluorometry setup and the filter set used in the study were similar to what was previously described^18,29. However, instead of using a photo diode, a photomultiplier was used to maximize the sensitivity of the signal, while allowing a decrease in the excitation light, decreasing photobleaching. The fluorophore was excited with an LED at 365 nm (Thor lab M365L3). TMR labeling was done similarly to previously described¹⁷. Briefly, oocytes were incubated in 1 mM DTT for 15 minutes. Three washes were administered before transferring the cells to the labeling solution with 20 µM Tetramethylrhodamine-5-Maleimide (Thermo Fisher T6027) in a depolarizing solution (in Mm: 120 KMES, 2 CaMes, 10 HEPES, 0.1 EDTA, pH = 7.40) for 30 minutes.

Data analysis

Ionic current was taken by the steady-state current level and converted to conductance using the following relationship:

$$G\left(V\right)=\frac{I}{V-{V}_{{rev}}}$$

(1)

where, I is the ionic current in steady state, V is the membrane voltage, and V_rev is the reversal potential for the conducting ion. The GV curves were then fitted using a two-state model given by the equation:

$$G\left(V\right)=\frac{1}{1+\exp \left(-\frac{{zF}}{{RT}}\left(V-{V}_{1/2}\right)\right)}$$

(2)

where z is the apparent charge expressed in units of elementary charge (${e}_{0}$), V is the voltage and V_1/2 is the voltage of half-maximal conductance. R is the ideal gas constant; T is the temperature in Kelvin and F is the Faraday constant.

The gating charge was obtained by integrating the on and off-gating currents. They were plotted against the voltage to obtain the QV curves.

For the analysis of the normalized QV curves, we used a two-state model fitting equivalent to the one in Eq. 2 to fit the gating from the I384E, I384L, I384N, and I384R mutants. The equation was the following:

$$Q\left(V\right)=\frac{1}{1+\exp \left(-\frac{{zF}}{{RT}}\left(V-{V}_{1/2}\right)\right)}$$

(3)

For WT, I384A, I384C, we used a three-state model fitting given by the following equation:

$$Q\left(V\right)=N\frac{{z}_{2}+{z}_{1}\left(1+\left(\exp \frac{{z}_{2}F}{{RT}}\left({V}_{2}-V\right)\right)\right)}{1+\exp \left(\frac{{z}_{2}F}{{RT}}\left({V}_{2}-V\right)\right)\left(1+\exp \left(\frac{{z}_{1}F}{{RT}}\left({V}_{1}-V\right)\right)\right)}$$

(4)

where N, z₁, z₂, V₁ and V₂ are the number of channels, the charges associated and equilibrium voltages for the first and second transition, respectively.

For nonstationary noise analysis, the mean variance data were fitted using a parabolic equation as follows:

$${\sigma }^{2}=i < I > -\frac{ < {I > }^{2}}{N}$$

(5)

where ${\sigma }^{2}$ is the variance, $i$ is the single-channel current, $< I >$ is the mean current, and $N$ is the number of channels in the patch. We can estimate the maximal open probability (${{Po}}_{\max }$) by knowing the maximum mean current (${I}_{\max }$) using the following equation:

$${{Po}}_{\max }=\frac{{ < I > }_{\max }}{{Ni}}$$

(6)

For single channel analysis data, we obtained all-points histogram binned at 0.05 pA using our Analysis software. The data was then fitted using a mixture of k Gaussian Distributions as follows⁷⁹:

$$f\left(x\right)={\sum}_{i=1}^{k}{a}_{i}{f}_{i}\left(y\right)$$

(7)

Where ${a}_{i}$ and ${f}_{i}\left(y\right)$, are, respectively, the relative areas of the components and the Gaussian equation described by:

$${f}_{i}\left(y\right)=\frac{1}{{\sigma }_{i}\sqrt{2\pi }}{e}^{-\frac{{\left({n}_{i}\right)}^{2}}{2}}$$

(8)

And

$${n}_{i}=\frac{y-{\mu }_{i}}{{\sigma }_{i}}$$

(9)

Where ${\mu }_{i}$ and ${\sigma }_{i}$ are the mean and the standard deviation of the individual component i.

Data is presented as Mean ± SEM. For noise analysis, ionic and gating currents experiments we used at least 3 oocytes from different batches. For single channels, we recorded several oocytes and only for presentation purposes show the representative traces from one cell.

Expression and purification of mutant shaker potassium channel

Shaker-IR-I384R was subcloned into a modified pEG BacMam vector containing a C-terminal HRV 3 C protease site, an eGFP tag, and an 8×-His tag utilizing the 5′ NotI and 3′ XbaI restriction sites. The Bacmid plasmid was generated using the Bac-to-Bac system, which was then used to transfect the Sf9 insect cells with Cellfectin (Thermo Fisher). After 4 to 7 days of incubation at 27 °C, P₀ virus was then collected after removing the remaining cells and debris. P₀ was then amplified to produce the P₁ and P₂ viruses. 2 L of HEK293S GnTI^- cells were infected with 200 mL of P₂ virus. After 24 h of shaking incubation at 37 °C, sodium butyrate was added to the cells at a final concentration of 10 mM and the culture was transferred to 30 °C. Cells were collected 48 h after infection. The cell pellets were then washed with phosphate-buffered saline (PBS) at pH 7.4, collected by centrifugation, flash-frozen, and stored at −80 °C for later purification.

All purification steps were conducted at 4 °C. Frozen cell pellets were thawed in a water bath and homogenized with a Dounce homogenizer in suspension buffer (150 mM KCl, 2 mM TCEP, 1 mM EDTA, 50 mM TRIS, Pierce protease inhibitor tablet (Thermo Fisher Scientific), pH 7.5). Resuspended cell culture was supplemented with DDM: CHS (10:1) at 1% final concentration (m./v.) and was extracted for 2 hours with gentle agitation. The cell debris was pelleted by ultracentrifugation for 1 hour at 100,000 x g in the Ti45 rotor (Beckman Coulter). The supernatant was collected and incubated with 2 ml CNBR-activated Sepharose beads (GE Healthcare) coupled with 4 mg high-affinity GFP nanobodies purified in-house for 3 hours. The Sepharose beads were then rinsed three times, 10 column volumes with the suspension buffer supplemented with 0.1% DDM:CHS (10:1), 0.05% DDM:CHS + 0.02% GDN, 0.02% GDN, respectively. Then beads were incubated with HRV 3 C protease overnight to release the purified channels. The next day, 3column volumes of suspension buffer was used to elute the beads. The eluted solution was concentrated with an Amicon Ultra Centrifugal Filter unit (Millipore) with 100 kDa cutoff to approximately 1 mL and loaded onto a Superose6 (10 × 300 mm) gel filtration column (GE Healthcare) and separated with suspension buffer supplemented with 0.02% GDN.

Cryo-EM sample preparation and data acquisition

Samples purified from size-exclusion chromatography were pooled and concentrated to around 0.7 mg/mL measured with a nanodrop machine. 3.5 µL concentrated protein was applied to glow-discharged (30 s, 20 W Solarus Plasma Cleaner) Quantifoil grids (R 1.2/1.3 Au 300 mesh). The grids were blotted for 3 s with blot-force 3 in a FEI Vitrobot Mark IV (Thermo Fisher) chamber with 100% humidity at room temperature before plunge freezing in liquid ethane. Clipped grids were subsequently loaded onto a Titan Krios microscope. Single-particle movies were acquired using a K3 direct electron detector in super-resolution mode, coupled with a 20 eV GIF energy filter. Data collection was performed at a nominal magnification of ×81,000 binned by 2 during acquisition, corresponding to a final pixel size of 1.06 Å. The total electron dose was calibrated to 60 e − /Ų, distributed across 50 frames per movie.

Single-particle cryo-EM analysis

All steps for structure determination were performed using CryoSPARC⁸⁰, including motion correction and contrast transfer function (CTF) estimation. A subset of 2000 particles was initially picked and classified in 2D to generate templates for template-based particle picking. Approximately 4,500,000 initial particles were picked and subjected to multiple rounds of 2D classification. From these, 180,000 particles were selected to generate three ab initio models with C1 symmetry. Particles from the best class (~116,000) were then processed for 3D refinement with C4 symmetry, yielding a nominal resolution of 5.7 Å. We performed 3D refinement using non-uniform refinement algorithm⁸¹, which consistently yields the best result in our data. A subsequent 3D classification was performed using the best-refined model from the previous round and a junk model generated ab initio. The best class from this step was further refined using C4 symmetry, resulting in a 3.55 Å nominal resolution. After identifying a subset of ~80,000 particles for final refinement, Reference-Based Motion Correction was applied to estimate per-particle movement trajectories and empirical dose weights. A final 3D refinement step was performed, followed by postprocessing using a tighter mask and C4 symmetry enforcement. Local resolution was calculated using CryoSPARC. Validation was done using MOLprobity⁸².

Model building and structural refinement

The Shaker-IR model (PDB 7SIP) was used as a template to build atomic models for the Shaker-IR-I384R mutants into our density map. All structural models were constructed using unsharpened maps. Initial models were generated through iterative rounds of manual model building in COOT⁸³ and real-space refinement in Phenix⁸⁴. Final refined atomic models were obtained using interactive flexible fitting in ISOLDE⁸⁵. All structural analyses and figure generation were performed using UCSF ChimeraX^86,87.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.