Refinement of a cryo-EM structure of hERG: Bridging structure and function

Abstract

The human-ether-a-go-go-related gene (hERG) encodes the voltage-gated potassium channel (KCNH2 or Kv11.1, commonly known as hERG). This channel plays a pivotal role in the stability of phase 3 repolarization of the cardiac action potential. Although a high-resolution cryo-EM structure is available for its depolarized (open) state, the structure surprisingly did not feature many functionally important interactions established by previous biochemical and electrophysiology experiments. Using molecular dynamics flexible fitting (MDFF), we refined the structure and recovered the missing functionally relevant salt bridges in hERG in its depolarized state. We also performed electrophysiology experiments to confirm the functional relevance of a novel salt bridge predicted by our refinement protocol. Our work shows how refinement of a high-resolution cryo-EM structure helps to bridge the existing gap between the structure and function in the voltage-sensing domain (VSD) of hERG.

Significance

Cryo-EM has emerged as a major breakthrough technique in structural biology of membrane proteins. However, even high-resolution cryo-EM structures contain poor side-chain conformations and interatomic clashes. A high-resolution cryo-EM structure of hERG has been solved in the depolarized (open) state. The state captured by cryo-EM surprisingly did not feature many functionally important interactions established by previous experiments. Molecular dynamics flexible fitting (MDFF) used for refinement of the hERG channel structure in a complex membrane environment re-establishes key functional interactions in the voltage-sensing domain in the open state.

Introduction

Human-ether-a-go-go-related gene (hERG) is responsible for encoding the voltage-gated potassium channel (KCNH2 or Kv11.1, commonly known as hERG) (1). Inherited mutations or nonregulatory blockade of hERG channels lead to long QT syndrome and, hence, can lead to a potentially lethal form of cardiac arrhythmia. Several drugs designed for other cellular targets have been found to nonspecifically block the hERG channel in ventricular myocytes. Therefore, the establishment of the high-resolution structure of hERG in its active (drug-binding) conformation is essential for the rational design of drugs that do not bind to hERG and was the focus points for many research labs over the last decade (2). The specific salt bridges involved in the stabilization of open (depolarized) and closed states of the channel have been studied by several groups using an arsenal of experimental methods ranging from electrophysiology studies combined with targeted mutagenesis (3, 4, 5, 6, 7, 8, 9, 10) combined with tryptophan scanning (11) or use of fluorescent tags (12) and Förster-resonance-energy-transfer-based techniques (13,14). The lack of experimentally determined structures led to the development of a plethora of structural models based on various templates as well as kinetic models aimed at providing a structural basis to a multitude of functional studies (1,3,4,8,15, 16, 17, 18, 19, 20, 21, 22). A pivotal event in understanding of structure-function relationships in the hERG channel was the publication of a high-resolution cryogenic electron microscopy (cryo-EM) structure by Wang and MacKinnon in 2017 (23). The hERG-channel, three-dimensional structure was refined to 3.8-Å resolution in presumably the depolarized or open state, which is the most stable functional state in the absence of a transmembrane voltage. This structure contains the entire transmembrane domain (TMD) as well as a large portion of the intracellular domains (Per-Arnt-Sim (PAS) and CNBD).

Although this structure is obtained at a high resolution, it also raised several questions about how the structure relates to the extensive body of experimental data on the state-specific salt bridges and other interactions proposed to exist in the depolarized state of the voltage-sensing domain (VSD). For example, several salt bridges involved in the direct stabilization of the VSD in its depolarized state (1,3,15,24,25) are missing in the structure deposited to the Protein Data Bank (PDB) (PDB: 5VA2). This indicates that the solved cryo-EM structure may not represent a fully functional state and needs to be refined further to obtain a structure that directly corresponds to the functional open conformation of the channel. Another striking feature of the resolved structure was its different, “nonswapped” VSD packing topology against the pore domain (PD) in stark contrast to the “domain-swapped” topology reported from structural studies of voltage-gated K⁺ channels from the Shaker family (26,27). The structural differences in packing of the VSD against the PD are likely reflected in differences in the gating process in the hERG channel and other related channels with a nonswapped topology compared with Shaker channels (12,28, 29, 30).

One of the critical features of VSD in a nonswapped topology channels such as hERG or HCN could be apparent plasticity of the voltage sensor (30). Using tryptophan mutagenesis scanning, Subbiah et al. proposed a loosely packed configuration of the VSD for the open state of the channel (11), with significant accessible space between the S1–S4 helices and the PD. More recently, a break in the helical structure of S4 and its apparent plasticity were proposed to be important gating determinants for the HCN channel (31). The gating charge transfer associated with hERG activation process is also surprisingly small compared with the gating charge of the Kv1.2 channel, suggesting a different mechanism (5). MD simulations performed with the cryo-EM structure embedded in a lipid bilayer also highlighted a loosely packed VSD configuration (32) in line with the observation of Subbiah et al. (11) However, the VSD in the cryo-EM structure is tightly packed against the PD. Shi et al. also emphasized the importance of further refinement of this cryo-EM structure because of the uncertain positioning of D509, which acts as a proton sensor and stabilizes the open state (25). All in all, these works highlight the importance of further investigating the structure of hERG, starting from the existing hERG cryo-EM structure, to obtain a fully functional state.

Our main goal is to refine the existing structure to overcome these inconsistencies between the structural state captured in cryo-EM and the plethora of functional and modeling studies that have established key interactions in the VSD of hERG. We would like to address two fundamental questions:

• 1)Can the apparent plasticity of the VSD be reconciled with the state deposited into PDB captured by the original cryo-EM density map and yet capture the missing functionally relevant interactions in the deposited PDB structure?
• 2)Is the VSD in hERG channel loosely packed as suggested by current computational (28) and previously published experiment studies and especially data from tryptophan/mutagenesis scanning (11,25)? If not, will it still allow significant VSD hydration and changes in helicity as proposed by other studies performed on the channels with nonswapped topology?

To address these questions, we refined the existing hERG structure using molecular dynamics flexible fitting (MDFF) (33). We performed structural refinement in different model environments (vacuum, implicit solvent, and bilayer) to investigate the possible effect of more native-like conditions on the fitting procedure and the importance of explicit lipids in establishing state-specific interactions in the VSD. We find that our bilayer-fitted MDFF structure has the highest structural quality (based on comparison of the global model fit with the reference map and the reference structure) and side-chain quality and re-establishes the missing salt bridges in the original cryo-EM structure, which is supported by the literature. In addition, this procedure predicted a novel salt bridge. We have tested this novel salt-bridge prediction with mutagenesis and electrophysiological experiments to show that this interaction is functionally important for activation of the hERG channel. Our work offers a computational approach followed by electrophysiology experiments to bridge the gap between the structure and function of hERG, which can be used for other ion channels and transporters.

Methods

Systems setup

The cryo-EM structure of hERG (PDB: 5VA2) was our starting point. Because the cryo-EM structure was missing some loops (for example, loops involving the following residues: 434–451, 511–519, 578–582, and 598–602), we used the cryo-EM structure with modeled loops of the open state from Perissinotti et al. as the target structure for the MDFF (28). This structure contained the missing loops built using ROSETTA loop modeling and was minimized and briefly equilibrated to avoid steric clashes and did not contain the PAS domain (28). This structure (before the simulation) was used to generate a map at 5-Å resolution, excluding atoms forming the nanodisk. We chose a resolution equal to 5 Å because it represents average resolution of the VSD portion of hERG channel (23).

Next, we used the MDFF (33, 34, 35) technique for structure refinement. We extracted the starting structure for fitting from our multi-microsecond MD simulation in which the protein was embedded in a bilayer mimicking the native environment (32). We extracted a representative structure from the fully equilibrated system (production phase of the MD trajectory) in which the VSD was in its fully expanded (relative to the starting structure) configuration. The reason behind this approach is twofold: 1) we wanted to make sure that the density of the excluded PAS domain does not contribute to the unfolding of the VSD during fitting, and 2) to investigate whether the expanded configuration of VSD can be fitted in the density of the compact VSD and simultaneously still retain or re-establish functionally relevant interactions in the open state of this protein. Different environmental conditions, vacuum, generalized born implicit solvent (gbis), and bilayer, were tested during fitting to check the importance of the presence of lipids during the fitting process. Vacuum and gbis systems were prepared using the MDFF plugin (36) in VMD (37). For the bilayer system, the protein was embedded in a POPC bilayer (380 lipids in total and 190 in each leaflet). For the bilayer system, 0.15 M KCl solution was used. CHARMM-GUI was used to prepare the bilayer system (38).

As a control, we also performed fitting with the original cryo-EM density and the published refined structure. For this case, we built the missing loops using the ROSETTA loop modeling protocol (39) and excluded the PAS domain because the large missing connecting part cannot be reliably built using any existing method. We generated 500 models for missing loops, and the structure with the lowest energy loop conformation was used to start the fitting. Finally, we performed an unconstrained, all-atom MD simulation with a refined structure (Data S1; bilayer MDFF, simulated map) embedded in the explicit bilayer and solvated. The protein was embedded in a 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) lipid bilayer using CHARMM-GUI (40) (380 lipids in total and 190 in each leaflet). The CHARMM36m (41) force field was used for the protein and the latest CHARMM36 lipid parameters used to describe lipids in the system (42). The TIP3P model and standard CHARMM36 parameters for K⁺ and Cl⁻ were used to simulate 150 mM aqueous KCl solution (43,44). The protein-membrane system was equilibrated for 50 ns and then subjected to a 1.2-μs production run on Anton 2 platform. The production simulations were performed with a semi-isotropic (NPaT) ensemble at a temperature of 305.15 K. The multigrator scheme was used for temperature and semi-isotropic pressure coupling (45). The time step for the production runs was set to 2 fs, and trajectories were saved every 120 ps. Nonbonded and long-range electrostatic interactions were evaluated every 2 and 6 fs, respectively. Long-range electrostatics were calculated using the k-Gaussian Ewald method (45,46).

Simulation protocol

Simulations were carried out using the CHARMM36m force field for proteins (41) and CHARMM36 force field for lipids (42). NAMD (47) (v. 2.13) was used to carry out all the simulations. The gbis model was used for implicit solvent simulations. For the bilayer system, POPC lipids and explicit TIP3P (44) waters were used. Cutoff and switching distances for nonbonded calculations were as follows: 9 and 10 Å (vacuum), 15 and 16 Å (implicit solvent), and 10 and 12 Å (bilayer). Integration of equations of motion was done using a 1-fs timestep for vacuum and implicit solvent and 2 fs for the explicit bilayer system. Temperature coupling was applied to all the systems. Semi-isotropic pressure coupling was applied to the bilayer system to preserve the bilayer shape using the Langevin piston method (48) with an oscillation period of 50 fs and a damping timescale of 25 fs. Vacuum and implicit solvent simulations were performed at 300 K, whereas bilayer systems were performed at 310 K and 1 atm pressure. For the gbis system, an ion concentration of 0.1 M was used, and a surface tension value of 0.005 kcal/mol/Ų was used for the nonpolar solvation. The solvent dielectric was set to 80. In all cases, secondary structure restraints along with cis-restraints on peptide bonds and chirality restraints were applied to the protein structure during fitting. We used a gscale value of 0.3 for all the cases after testing several other gscale values because those led to overfitting. For the bilayer system, the coupling of the structure to the map was done gradually and at 10-ns steps for Cα atoms, the backbone atoms, and finally the heavy atoms.

Analyses

Root mean-square deviation (RMSD) and cross-correlation coefficient (CCC) were calculated with MDFF plugins in VMD. Together, they provide information on the overall structural quality compared with the target structure and the target cryo-EM map, respectively. The CCC is calculated between the generated maps from the fitted structure over the course of fitting simulation with the target map (33) and excluding hydrogen atoms. Final fitted structures were validated using MolProbity (49,50) (Phenix (51) implementation) and using all four chains. MolProbity provides information on the overall side-chain quality of the structure. We also tested the backbone quality using CaBLAM (52). For the fitting with the original cryo-EM map, we also calculated several map-to-model fits using Phenix. Salt bridges were analyzed using the salt-bridge plugin in VMD using an oxygen-nitrogen distance cutoff value of 3.2 Å. For the calculation of salt-bridge occupancies, a distance cutoff of 5 Å was used for the distance between the center of mass of the oxygens in the acidic side chain and center of mass of the nitrogens in the basic side chain. Occupancies are defined as the number of frames in which the interactions were present divided by the total number of frames.

Molecular biology

Methods for site-directed mutagenesis have been previously reported (53,54). Single- and double-mutant constructs of hERG were produced using conventional overlap PCR with primers synthesized by Sigma-Genosys (Oakville, Ontario, Canada) and sequenced using Eurofins MWG Operon (Huntsville, AL). The hERG constructs were transfected into human embryonic kidney (HEK) 293 cells by the calcium phosphate method and cultured in Dulbecco’s modified Eagle’s medium supplemented with 10% fetal bovine serum and 1% penicillin and streptomycin (GIBCO, Carlsbad, CA).

Electrophysiology

The extracellular solution contained 140 mM NaCl, 5.4 mM KCl, 1 mM CaCl₂, 1 mM MgCl₂, 5 mM HEPES, and 5.5 mM glucose and was kept at pH 7.4 with NaOH. Micropipettes were pulled from borosilicate glass capillary tubes on a programmable horizontal puller (Sutter Instruments, Novato, CA). The pipette solution contained the following: 10 mM KCl, 110 mM K-aspartate, 5 mM MgCl₂, 5 mM Na₂ATP, 10 mM EGTA, 5 mM HEPES, and 1 mM CaCl₂. The solution was adjusted to pH 7.2 with KOH. Standard patch-clamp methods were used to measure the whole cell currents of hERG mutants expressed in HEK 293 cells using the AXOPATCH 200B amplifier (Axon Instruments, Union City, CA). The holding potential was −80 mV. The amplitudes of tail currents were measured when the voltage was returned to −100 mV after +50-mV, 1-s depolarization.

Voltage-dependence of activation

From a holding potential of −80 mV, cells were depolarized for 1 s to a range of voltages from −100 to +50 mV followed by a step to −100 mV (or −50 mV, 1 s; see Results) to record the tail currents. The isochronal tail current-voltage plots were fitted to a single Boltzmann function (1):

$$\frac{I}{I_{max}}=\frac{1}{1+\exp \left(\frac{\left(V_{1/2}-V_m\right)}{k}\right)},$$ (1)

where I/I_max is the normalized current, V_1/2 is the voltage of the half-maximal activation, k is the slope factor, and V_m is the membrane potential.

Statistical analysis of electrophysiology data

The data are presented as the mean ± standard deviation. One-way analysis of variance test was used to analyze the data. A value of p <0.05 was designated as being significant.

Results

Environmental effects are important for the structure quality

The backbone RMSD with respect to the target structure is reported in Fig. 1 A. The variation is minimal in the final segment of the trajectory for all three environments. This means that the final structures obtained using different environments during fitting converged to very similar configurations in terms of backbone RMSD. The obtained final structures retain very similar backbone configurations in the helical membrane parts of the protein (Fig. 1 B). There are small variations in the loop regions connecting the helices. The global CCC (33) values calculated over the course of fitting also converged to similar values (Fig. 2). As can be seen, the final structures are very similar to the target structure in terms of their backbone configurations in the VSD and PD of the hERG. The resulting global CCCs from these structures are identical, indicating a high degree of global similarity to the target density map. These global comparisons highlight that we cannot distinguish which protocol performs better, either compared with the target structure or compared with the target map because they converge to a very similar state.

Figure 1.

MDFF simulations performed in different environments. (A) Backbone RMSD of the protein during the course of the MDFF simulation and (B) the final fitted structures obtained from the MDFF simulation: vacuum (blue), gbis (red), and bilayer (gray). To see this figure in color, go online.

Figure 2.

Cross-correlation coefficient (CCC) during the MDFF simulations. To see this figure in color, go online.

The global measurements of RMSD and CCCs showed little difference in the helical backbone configurations of the hERG-fitted structures, particularly in the VSD regions. As a more local measurement, we used MolProbity (49,50) to check the side-chain quality of the models. The MolProbity statistics are reported in Table 1 along with the MolProbity score, which is a single metric to compare overall quality of the models. We also tested the cryo-EM structure. Ramachandran outliers, measuring the backbone ϕ- and ψ-angles, are most frequent in the vacuum model (3.46%) and least frequent in the gbis model. However, side-chain rotamer outliers are most frequent in the gbis model (4.56%) and least common in the cryo-EM model (1.08%). Cβ deviations are most dominant in the gbis model (198), are similar for vacuum and the bilayer model (169 and 167, respectively), and are not present in the cryo-EM model. Clash scores show an opposite trend: 11.12 for cryo-EM model and 0 for all the others. RMSDs of bonds and angles are lowest for the cryo-EM structure and remain low for all other fitted structures. Finally, the MolProbity score for the overall side-chain quality is worst for the cryo-EM model (2.12) and is best for the bilayer-fitted MDFF model (1.32). Vacuum and the gbis-fitted models perform almost similarly (1.52 vs. 1.50). In light of these side-chain quality analyses, it seems that the bilayer-fitted MDFF structure quality is superior to other counterparts, including the published cryo-EM structure. This is not surprising because even high-resolution cryo-EM structures often contain poor side-chain conformations and interatomic clashes (55). Our backbone quality analysis based on the CaBLAM approach shows only modest to no improvement (Table 1). The control tests with the original map highlight modest improvements in the side-chain quality and little to no improvement in the backbone fitting quality (Table S1).

Table 1.

Model quality based on MolProbity and CaBLAM statistics for cryo-EM and MDFF-fitted structures of hERG using different environmental conditions and using the simulated map

	cryo-EM	vacuum	gbis	bilayer
MolProbity statistics
Ramachandran outliers (%)	1.65	3.46	1.08	2.27
Ramachandran favored (%)	90.66	89.85	91.58	91.63
Rotamer outliers (%)	1.08	4.14	4.56	2.62
Cβ deviations	0	169	198	167
Clash score	11.12	0.00	0.00	0.00
RMS (bonds)	0.01	0.04	0.03	0.03
RMS (angles)	0.99	3.52	3.53	3.57
MolProbity score	2.12	1.52	1.50	1.32
CaBLAM statistics
Disfavored conformations (%)	10.0	12.6	11.7	11.3
Outlier conformations (%)	5.1	4.6	4.9	4.2
Cα geometry outliers (%)	0.49	1.36	1.74	1.36

Analyses are carried out on the entire tetramer. The cryo-EM structure refers to the structure (PDB: 5VA2) by Wang et al. (23).

Refinement of state-dependent interactions in the VSD establishes additional salt bridges

After assessing the overall backbone and side-chain properties in previous sections, we now turn to the refinement of state-specific interactions in the MDFF-fitted hERG models, in particular the presence of functionally important salt bridges in the VSD in the open state. A summary of all intradomain salt bridges in the VSD is presented in Table S3. We analyzed all four chains separately to check the symmetry of the interactions, which may shed light on the activation dynamics of the channel. The cryo-EM structure contains three salt bridges (D411-R541, D460-R528, and D501-R537), and these salt bridges are present in all four chains in the symmetric tetramer. Several other physiologically relevant salt bridges in the depolarized VSD are absent in the cryo-EM structure (3,6,8,11,15,25). We now want to explore whether the refinement protocol preserves these three salt bridges in the cryo-EM structure and further establishes other salt bridges in our fitting procedure. The vacuum MDFF procedure leads to the formation of several additional salt bridges in the VSD of hERG. However, the three pairs present in the cryo-EM are lost in some chains. For example, D411-K538 is lost in the B and D chains and D501-R537 in the A and D chains. The D460-R528 pair is lost in chains B, C, and D. Among newly formed salt bridges, only D466-K538 is present in more than one chain. Hence, the vacuum fitting protocol may not be sufficient to further refine the hERG structure and establish the missing interactions.

The gbis and bilayer MDFF procedures lead to the formation of more functionally relevant salt bridges, as compared with the experimental data from mutagenesis and electrophysiology studies (Fig. 3; Table S3). All salt-bridge pairs formed by D411 are present in simulations performed with different protocols (gbis and membrane). However, both procedures failed to produce a stable D411-K538 pairing in chain D in comparison with the cryo-EM structure. D456-R528 pairs are present in all the chains in the gbis structure and are missing in chains A and D in the bilayer structure. However, R528 forms a salt bridge with D509 in the bilayer structure. D509 is known to be important in hERG function (25). The salt bridge D460-R528 is present in all the chains in the bilayer structure. The gbis structure contains this salt bridge in three chains, but it is missing in chain D. The salt bridge D460-R531 is also thought to be important for the stabilization of the open state (15). This salt bridge (Fig. 3) was restored in the bilayer structure. D466 establishes multiple bifurcating salt bridges with R537, K538, R534, and K407 in the gbis model and, to a lesser extent, in the bilayer model (only with K538 and R534). The D501-R537 salt bridges present in the cryo-EM structure are retained fully in the gbis model; however, this salt bridge is missing in chain D for the bilayer model. In both the gbis and bilayer model, D501-R534 is formed in all chains except chain D. The D509-R528 salt bridge is crucially important for the activation and stabilization of the open state of the hERG (10,25). This salt bridge is established in chains A and D in the presence of the bilayer (Fig. 3). We also see the formation of several other glutamate pairs in both the gbis and the bilayer-fitted structures, which do not occur in the vacuum structure or in the cryo-EM structure. In total, the number of salt bridges formed in the gbis or bilayer fitting protocol (44 and 43, respectively) is almost four times higher than the vacuum protocol or the actual cryo-EM (13 and 12, respectively) structure (Table S3). Our control tests with the original cryo-EM map highlight similar salt-bridge patterns after the refinement (Table S4).

Figure 3.

Formation of functionally relevant salt bridges after fitting. (A) Cryo-EM structure for comparison and (B) the structure obtained from MDFF-bilayer fitting. Only newly formed salt bridges involving aspartates in chain A are shown for clarity. Dashed lines in (B) represent formation of the salt bridges. (C) Distance between nitrogen of arginine/lysine and oxygen of aspartate in the corresponding pairs. To see this figure in color, go online.

A novel salt bridge predicted in the VSD of hERG is important functionally

The MDFF fitting with cryo-EM maps suggested the presence of a novel salt-bridge interaction in hERG VSD formed by D501-R534. To investigate a possible functional role of this salt bridge in activation, we performed electrophysiology experiments (Fig. 4) of single and double mutants of the channel. Interestingly, the single R534E mutation elicits a functional hERG-like channel but with altered activation kinetics (Fig. 4 B). These altered functions include a shift of the V_1/2 of activation to more depolarized potentials (Fig. 4 E) and the development of a slow component of deactivation. The slow τ of deactivation increases from 299 ms in wild-type (WT) channels to 500 ms in the R534E currents. These features could be interpreted in terms of changes in the barrier associated with the voltage-sensor movement during the activation process. The shift to longer τ of deactivation in R534E suggests an increase in energy penalty associated with the activation gate opening, but once the gate is open, it allows WT-like currents. Interestingly, the R534E mutation has only a modest impact on the channel recovery from inactivation. The reversal potential for R534E is −78 mV, and the inward rectification and tail currents features associated with recovery from inactivation are comparable with that observed in the WT hERG. Importantly, the D501K channel does not conduct any functional current (Fig. 4 C).

Figure 4.

Representative current traces of WT (A), R534E (B), D501K (C), and D501K-R534E (D). D501K did not show a visible current. I-V relationships of the tail currents are shown in (E). The experimental protocol is shown in the inset of (A). (F) The current densities of maximal tail current of each mutation. N = 8, 4, 4, and 19 in WT, D501K, R534E, and D501K/R534E, respectively. NS, no significance. ^∗p < 0.5, ^∗∗p < 0.001, comparing with WT in one-way analysis of variance test. (G) In D501K-R534E, reversal potential of tail currents (H) and the reversal potential in response to the test pulses (D). p = 0.11, paired Student t-test. N = 7. (H) and (D) were from a same cell.

To further test the presence of the salt bridge proposed from MDFF modeling, we created the salt-bridge reversal mutation D501K-R534E. The double-mutant rescues a functional current expression (Fig. 4 D) but with an altered function providing direct evidence for a potential importance of the salt-bridge spatial organization (3,15,56). Its alterations include the following: 1) a near-instantaneous leak conductance that has a normal reversal potential, and 2) tail currents that show a markedly abnormal slope factor for activation. The leak conductance with WT-like reversal potential may reflect presence of the partially opened activation gate in the PD in all gating states of the voltage sensor. We hypothesize that the ion flux occurs through the PD, retaining the WT-like selectivity filter.

Although we are not measuring selectivity or single-channel permeation directly, both the R534E and the D501-R534E currents manifest significant inward rectification, suggesting that WT-like C-type inactivation is intact. These mutations are also quite distant from the known selectivity filter region. Accordingly, a major change in the selectivity filter seems unlikely. During repolarization, a very slow deactivation component is observed with a persistently open leak component in keeping with an activation gate in the partially open state. We hypothesize that the change in the slope factor for gating and the leak current may reflect abnormalities in the cooperative transitions on the path to channel closing.

Discussion

The presence of negative charges in the S1–S3 helices and positive charges in the S1 and S4 helices naturally raises the question of their role in gating processes. Although the direct modeling of the gating process is outside the scope of this study, we can use extensive functional studies published to date to first assess the state of the sensor captured in the original cryo-EM structure and then connect to fitting protocols as well as VSD plasticity features emerging from simulation studies. Here, we focus our discussion on the interactions present in the bilayer MDFF structure because this structure seems to be the best in our structure quality checks, as outlined in the Results.

Structure-function relationship is established in the VSD of the refined structure for open state of hERG

In the cryo-EM structure, only three pairs of salt bridges are present (Table S3). They are the following: D411-K538, D460-R528, and D501-R537. Hence, there is a clear gap between the structure of this “open state” resolved by cryo-EM and the other previous functional studies. In the following paragraphs, we discuss the additional salt bridges that were observed in our refined structure and their functional relevance in open, closed, or transition state stabilizations based on previous works.

In a series of studies, Tseng and co-workers highlighted the importance of these salt bridges in stabilizing hERG functional states and its gating process (5), reviewed by Vandenberg et al. (1). Here, we focus our discussion on the interactions present in the bilayer MDFF structure (Fig. 3). The negatively charged residues form essential interactions as found by Liu et al. (3). In charge neutralization experiments, D460 and D509 form salt bridges characteristic of the open state of hERG. However, the authors could not determine the positively charged counterparts for these salt bridges in those experiments (3). We also see bifurcating salt bridges involving D411 (D411-K538 and D411-K407), which are thought to stabilize the closed state of hERG by Liu et al. (3). This is surprising because we start from the putative open state of hERG from the cryo-EM structure, yet we see salt-bridge formation in our fitted model by a negatively charged residue that is supposed to stabilize the closed state (3, 4, 5). In a later charge reversal mutagenesis analysis by Zhang et al., it was postulated that D411 might be involved in stabilizing “multiple gating states” (4), which supports the presence of salt bridges involving D411 in the putatively open (depolarized) structure emerging from MDFF studies.

We also see the formation of the D411-K407 and D411-K538 pairs. The MDFF simulations show the simultaneous presence of D411-K407, D411-K538, and D456-R528 pairs but not the D456-K525 pair (Table S3). However, we cannot investigate the functional couplings of these pairs from our simulation alone because MDFF runs in this study focus on only one conformational state. The modeling data have to be contrasted to the results of functional mutagenesis studies, providing structural basis for the tentative state-specific interactions. For example, R531, R534, and R528 form salt bridges in the open state by interacting with D411 and D456 in good accord with the experimentally proposed state-specific interaction patterns (4). The importance of R531 has been tested by Piper et al. using double-mutant cycle experiments (15). Based on their work and work on the homologous Kv1.2 structure, the authors suggested a functionally important interaction between R531 and D456 to stabilize the open conformation of the channel supplemented by cooperative interactions between R531 and D456, D460, and D509 in the inactivation process (15). Although we do not see the R531-D456 pair, we do find a persistent salt bridge between R531 and D460. Using tryptophan scanning mutagenesis, Subbiah et al. suggested the salt-bridge pairing between R528 and R531 in different conformational states: R528 being involved in “at least one closed state and a transition state” and R531 being involved in “at least one closed and one open state” (11). Our salt-bridge analysis is in accord with their observation. D509 has been identified to be very important in stabilizing both the activated and the relaxed state of the VSD and proposed to be directly interacting with a basic residue (R528) or water-mediated electrostatic interaction with a basic residue (K525) in S4 (25). We also observe the direct interaction of D509 with R528 present in the two chains. Several other salt bridges involving D509 were also present in our fitted structure. Overall, the upper portion of the VSD in hERG appear to be very flexible, allowing for dynamical formation of direct or water-mediated salt bridges. We have not found clear evidence yet in the literature about possible salt-bridge formation by these glutamates in VSD (E435, E444, E481, E518, E519, and E544) and their role in the gating process. Hence, they might be simulation artifacts and should be interpreted with caution. We plan to carry out further work to establish their functional significance.

What are the functional consequences of the salt bridges between aspartates located in S1–S3 and S4 gating charges? Like salt bridges present in Shaker channels (57), these interactions seem to be state specific; interrupting them leads to slow activation or faster deactivation or affects voltage sensitivity during the activation gating process. For example, substitution of these aspartates by cysteine leads to an accelerated deactivation process, leading to the suggestion by the authors that these acidic residues and their state-specific interactions contribute to the deactivation process in an additive or cooperative manner (3). Nevertheless, this highlights their role in stabilizing the open state of hERG. Shi et al. also observed a similar role by the protonating of these residues and found the destabilization of the activated state (58). Particularly, D509 plays an important role in stabilizing the relaxed state of the VSD (25). In short, the salt-bridge pairing observed in our refined structure appears to represent the characteristic interactions previously identified as open-state “stabilizers.” Therefore, the refined structure seems to represent a functional open state as supported by literature work and our electrophysiology experiments and the MDFF-enabled relaxation of the structure restores many of the key salt bridges missing in the deposited cryo-EM coordinates of hERG channel. Our microsecond long MD simulation of the refined structure also supports these observations. Salt-bridge stability analysis highlights that the pairs involving aspartates are mostly long-lived and highly stable, whereas the pairs involving glutamates are mostly transient (Table S5).

VSD in open state of hERG can be less plastic and tightly packed and is still hydrated by water

One of the important features of the VSD in nonswapped topology channels is its relatively loose packing, which might be important functionally (3,11,25). That is, representing important ensemble of states for protein domain with significant plasticity and linking it to its functional manifestation may be particularly challenging for these channels. The initial structure we started with for fitting was derived from long, 1-ms MD simulations, and the VSD was loosely packed and showed average conformation with RMSD to cryo-EM state of ∼6 Å (Fig. 1). During the course of fitting, this loosely packed VSD fits into the cryo-EM density map. Hence, the S4 containing VSD is tightly packed in our final structure, even in the presence of lipid during fitting (Fig. 5).

Figure 5.

MDFF simulation with POPC bilayer. (A) Structures before fitting and (B) the structures after fitting. For clarity, only chain A is presented. Both the initial (pale cyan) and the target (raspberry) structures are presented in a cartoon representation. (C) Final structure after MDFF simulation in the bilayer and the corresponding target cryo-EM map. To see this figure in color, go online.

We observed a compact configuration of VSD regardless of the environment during the fitting. This is different from the observation on the voltage-sensor flexibility found in a microseconds-long MD simulation with the reported RMSDs reaching a plateau at around 5 Å (32). A study by Miranda et al. was centered at the permeation across the PD and employed a strong biasing electrical field (750 mV) to enable ion movement that may impact VSD conformation (32). Applications of electrical field are known to induce significant conformational plasticity of VSD in the previously studied Kv channels (59). The authors observed the rapid relaxation of the VSD in the presence of lipids during MD simulations with an increase in VSD hydration and subsequent deviations from the cryo-EM structure reported by Wang et al. (23). Interestingly, we see formation of water wire even in this compact VSD in our bilayer fitting (Fig. 6). However, the number of water molecules is significantly less than what has been observed in our unbiased, 1-ms-long simulation of the refined structure, in which the VSD adopts an expanded configuration and allows for more water molecules to hydrate the VSD. Because the VSD is constrained by the density map in our fitting procedure, it is natural that the VSD will have a tightly packed configuration (resembling the resolved cryo-EM state and deviating less from it) and has smaller hydration numbers. Hydration of the VSD observed in the fitting process highlights the importance of the VSD plasticity for hERG function, which albeit established experimentally, has been largely overlooked so far in the channel modeling/structural data interpretation.

Figure 6.

Formations of water wire in the VSD during bilayer fitting highlighting VSD plasticity. To see this figure in color, go online.

Conclusions

By using MDFF, we have further refined the hERG open-state cryo-EM structure. Our bilayer-fitted MDFF structure seems to be the most relevant open-state representation of the hERG considering the global structural quality, side-chain model quality, and functionally relevant salt bridges. The state-specific interactions present in refined bilayer-fitted MDFF structure are in excellent agreement with a plethora of functional studies establishing interaction patterns in hERG VSD. We also identified the presence of a novel salt bridge (D501K/R534E) in the bilayer-fitted MDFF state of VSD. The functional significance of this interaction has been further confirmed with electrophysiology experiments performed on single and double mutants of hERG. Our study also highlights apparent functional plasticity of VSD in a nonswapped topology channel. We observed significant hydration of the voltage-sensor domain in our bilayer-fitted MDFF states, a potentially important functional feature linked in other channels to formation of ω-currents. It can be concluded that our integrated approach led to a refined salt-bridging pattern present in the VSD in its open (depolarized) state. The bilayer-fitted MDFF protocol can be recommended for other ion channels and transporters for refinement of key state-specific interactions.

Author contributions

H.M.K. designed and performed the MDFF and MD simulations, analysis of cryo-EM maps, and resulting data analysis. J.G. designed and performed all molecular biology and electrophysiology experiments to test the novel salt bridges predicted by the MDFF refinements. S.Y.N., D.P.T., and H.J.D. conceptualized the study, designed the experiments, and interpreted the results. H.M.K. wrote the first draft of the manuscript. All authors wrote, edited, and contributed intellectually to writing the manuscript and interpreting the data.

Editor: Philip Biggin.