Please refer to Supplementary Table S2 for Key Resources.Ethical statementThe Garvan/St Vincent’s Animal Ethics Committee approved all experiments (23/11).Electrophysiology studiesXenopus oocyte recordingsOocytes were isolated from Xenopus laevis frogs (2–6 years of age; purchased from Nasco, Fort Atkinson, WI, USA). Frogs were anaesthetized in 0.17% w/v tricaine and segments of ovarian lobes were removed through a small abdominal incision. The ovarian lobes were digested with collagenase A (2 mg ml−1, Boehringer Mannheim USA) in a buffered salt solution containing (mm): NaCl 82.5, KCl 2.0, MgCl2 1.0 and Hepes 5.0 (pH adjusted to 7.5 with NaOH). Oocytes were manually teased apart and rinsed with ND96 containing (mm): KCl 2.0, NaCl 96.0, CaCl2 1.8, MgCl2 1.0 and Hepes 5.0 (pH adjusted to 7.5 with NaOH). Batches of 100–200 Stage V/VI oocytes were isolated and stored at 16 °C in ND96 contaqining freshly prepared Na pyruvate (2.5 mM), theophylline (0.5 mM), and gentamicin (10 μg ml−1)45.WT HERG cDNA was a gift from Gail Robertson (University of Wisconsin). cRNA was synthesized using the Invitrogen™ mMESSAGE mMACHINE T7 Transcription Kit (Thermo Fisher Scientific, Riverstone NSW) according to the manufacturers’ protocols. Oocytes were injected with 23 nL cRNA and incubated for ~24 h prior to electrophysiological recordings, using a Geneclamp-500B amplifier (Molecular Devices, San Jose, CA). All experiments were undertaken at room temperature (21–22 °C). Perfusion solutions contained 3 mM KCl, 97 mM NaCl, 1.8 mM CaCl2, 1 mM MgCl2, 5 mM HEPES, pH adjusted to 7.5 with NaOH. Glass microelectrodes had tip resistances of 0.2–0.7 MΩ when filled with 3 M KCl. Data analysis was performed using pClamp software (Version 10.6, Molecular Devices) and Excel software (Microsoft Corporation, Seattle, WA). All data are shown as mean ± S.E.M.Steady-state inactivation was measured from a two-step voltage protocol, as previously described45. The voltage range for measurement of inactivation in Xenopus oocytes extended to −200 mV for mutants with enhanced inactivation and up to +100 mV for mutants with reduced inactivation. For HEK293 cells inactivation was measured over a voltage range from −150 mV to +20 mV. Peak tail currents were obtained by fitting an exponential function to the decay phase of the tail currents and extrapolating back to the start of the voltage step, i.e., to account for channel deactivation2,4. Normalized conductance values were then fitted with a Boltzmann function:$$g/{g}_{\max }=[1+\exp ((\Delta {G}_{0}{{\rm{\hbox{-}}}}{z}_{g}E\,F)/{RT})]^{-1}$$
(1)
where ΔG0 is the free energy difference between the open and inactivated states at 0 mV. zg is the charge transferred during inactivation, E is the electric field strength, F is Faraday’s constant, R is the Universal Gas constant and T is temperature.Protein expression and purificationThe cryo-EM construct used for the study of Human KCNH2 (HERG) was based on the Ts construct developed by Wang and Mackinnon22. In brief cytoplasmic loops 141–380 and 871–1005 were removed using quikchange PCR methods. A TEV protease site (ENLYFQG) was inserted between the gene and a GFP epitope followed by a double strep tag (AWSHPQFEK) with a linker (GGGS)2 (GGSA) between the first and second repeat. The construct was cloned using EcoR1 and Xba1 into PEG vector47.The PEG-HERG construct was used to generate baculovirus as described in ref. 47. Cells were cultured and purifications were performed with changes as described22. Briefly, HEK 293 GnTI- cells were induced with baculovirus when reaching a density of 2.5–3.5 × 106 cells/ml. After 24 h from induction the temperature was lowered to 30 °C and a final concentration of 10 mM sodium butyrate was added for a further 24 h where cells were harvested.Cells were lysed in 30 mM KCl and 20 mM HEPES pH 7.4 and spun down at 40,000 rpm for 90 min. The insoluble fraction was collected and resuspended in 300 mM KCl, 20 mM HEPES pH 7.4, 1% DDM n-Dodecyl-b-D-maltoside (DDM), 0.2% Cholesteryl hemisuccinate (CHS) and 5 mM DTT, samples were incubated at 4 °C for one hour under gently agitation. The soluble fraction was collected and loaded onto strep-tactin superflow high-capacity resin (IBA) and allowed to bind at a flow rate of 1 ml/min at 4 degrees. Resin was washed (300 mM KCl, 20 mM HEPES pH 7.4, 0.1% DDM, 0.02% Cholesteryl hemisuccinate (CHS), 5 mM DTT and 0.1 mg/ml phospholipids POPC, POPE and POPA in a ratio of 5:5:1) with 5 column volumes or until UV absorption returned to baseline. The sample was eluted with an addition of 5 mM Desthiobiotin to the wash buffer. For WT HERG, GPF was cleaved with TEV protease overnight at 4 °C under gentle agitation. The Tetramer peak was collecting using a Superose 6 using either 300 mM KCl or 3 mM KCl and 297 mM NaCl, 20 mM HEPES pH 7.4, 10 mM DTT, 0.025% DDM, 0.005% CHS and 0.025 mg/ml of phospholipids POPC, POPE and POPA in a ratio of 5:5:1. Protein was concentrated to 7.5 mg/ml.Cryo-EM grid preparation and data collectionQuantifoil® 200 Cu mesh R1.2/1.3 holey Cu-carbon grids were plasma cleaned for 60 s in low pressure gas (0.5 mbar, 80% argon 20% oxygen mixture) using a Diener plasma cleaner. Protein sample (3.5 µl; 7.5–8.5 mg/ml) was applied to the carbon-coated side of the grid. The grid was blotted for 10 s with blot force 10 and vitrified in liquid ethane using a Vitrobot Mark IV (FEI) equilibrated to 4 °C and 100% humidity.For all datasets, grids were imaged on a Titan Krios operating at 300 keV equipped with a Gatan K2 detector. Images were collected in electron counting mode at a nominal microscope magnification of 130kx (1.05 Å/pixel) with 1e-/A2 per frame with a total dose of 50 e-/A2 or 60 e-/A2 and nominal defocus range from −0.5 to −2.5 µm.Cryo-EM data processingData was processed using RELION-3.1.1. Motion correction was done using RELION48 and defocus values were estimated using CTFFIND449. Auto-picking was first performed on a subset of micrographs for each datasets using a Laplacian of Gaussian filter to generate templates for templated-based auto-picking of the whole dataset. Particles were extracted from micrographs with a box size of 300 pixels and binned to 64 pixels for all datasets The binned particles were subjected to five rounds of 2D classification. Good classes were manually selected and reextracted without binning. An initial model was generated without imposing symmetry. The extracted particles were initially subjected to one round of 3D classification with 6 classes and no symmetry imposed. No asymmetry was observed in the initial model or any of the classes (see Supplementary Fig. S1). Further 3D classification was therefore performed with C4 symmetry imposed and classes which clearly resemble HERG were selected. Final 3D auto-refinement followed by Bayesian polishing and CTF refinement were performed in RELION. Focused refinement using a soft mask covering only the transmembrane region (398–667) was done using the final polished particles. Focused refinement did not improve resolution or provide better quality map and, therefore, the focused refined map was not used.Post-processing and resolution estimates were performed with a soft mask including only the transmembrane domains. Local resolution was calculated using the local resolution function in Post-processing job of Relion. Pixel size was calibrated against the published structure (EMD-8650) by maximizing the cross-correlation between the two maps. Calibrated pixel size of the final map was 1.05 Å/pixel for all datasets.Model building and refinementModel was built and refined using ISOLDE50 and PHENIX51. The HERG structure (PDB: 5VA1) was used as starting model. Only the transmembrane domains (residues 398–709) were modeled due to the poorer local resolution for the cytoplasmic domains. MolProbity52 was used to validate the geometries of the refined models. Difference maps were calculated using TEMPy:Diffmap53. The WT high-K and WT low-K maps were first low pass filtered to 3.3 Å, then matched by amplitude scaling in resolution shells and the difference map generated as fractional differences with respect to the globally scaled map values. All images were rendered using ChimeraX-1.454. Internal cavity was calculated using HOLLOW55 with 1.4 Å radius probe and grid spacing of 0.2 Å.Molecular dynamicsAll systems were built with CHARMM and simulated with NAMD2.13 or NAMD2.1456. The CHARMM36 lipid57 and CHARMM22 protein force fields58 with CMAP corrections59 were used with modified K+-backbone carbonyl interaction parameters (depth 0.102 kcal/mol and position 3.64 Å of minimum) to achieve a small experimental preferential solvation of K+ in N-methyl-acetamide over water60. The NPT ensemble was maintained using the Langevin piston Nose-Hoover method61,62 for pressure and Langevin dynamics to maintain a temperature of 303 K. Bonds to hydrogen atoms were maintained with the RATTLE algorithm63 and electrostatic interactions calculated with Particle Mesh Ewald64 with a grid spacing of 1.0 Å and 6th order B-spline mesh interpolation with a neighbor list distance of 15 Å and a real space cut-off of 12 Å with energy switch distance of 10 Å.Molecular dynamics flexible fittingTo best mimic the experimental system for Molecular dynamics flexible fitting (MDFF)65 simulations, detergent micelles were built around proteins using PDB:5VA1. The optimal size of the micelle was determined by simulating 6 different sizes of a simple pure detergent micelle (n-Dodecyl-B-Maltoside Detergent (DDM)). The number detergent molecules in contact with the protein plateaus at 400–450 molecules, Supp Fig. S5, suggesting that this is a sufficient number of detergent molecules for embedding HERG66. WT HERG, N588K_HERG and N588E_HERG were embedded in micelles containing 404 molecules, consisting of detergent (DDM (300)), lipid (POPE (20), POPC (20), POPA (4)) and cholesterol derivative (cholesteryl hemisuccinate (60)), and surrounded with explicit TIP3P water molecules67. Micelles were surrounded with 110 853 water molecules and either 660 K+ ions and 600 Cl− ions (300 mM KCl) or 70 K+ ions and 6 Cl− ions (3 mM KCl), respectively.MDFF65 was used to refine the cryo-EM structures in the presence of a potential energy function based on the cryo-EM density map \(\Phi \left({{\boldsymbol{r}}}\right)\), given by \({U}_{{EM}}\left({{\boldsymbol{R}}}\right)=\sum {\omega }_{j}{V}_{{EM}}({{{\boldsymbol{r}}}}_{j})\), where \(\omega\) is the weight for each atom \(j\) at \({{{\boldsymbol{r}}}}_{j}\), and$${V}_{{EM}}\left({{\boldsymbol{r}}}\right){{\boldsymbol{=}}}\left\{\begin{array}{ll}\xi \left(1-\frac{{{\rm{\phi }}}\left({{\rm{r}}}\right)-{{{\rm{\phi }}}}_{{{\rm{thr}}}}}{{{{\rm{\phi }}}}_{\max }-{{{\rm{\phi }}}}_{{{\rm{thr}}}}}\right);&{\Phi }\left({{\boldsymbol{r}}}\right){{\ge }}{{{\rm{\phi }}}}_{{{\rm{thr}}}}\hfill\\ {{\mathrm{\xi}}};\hfill & \Phi \left({{\boldsymbol{r}}}\right)\, {{ < }}\, {{{\rm{\phi }}}}_{{{\rm{thr}}}}\quad\end{array}\right.$$
(2)
with threshold, \({\Phi }_{{thr}}\), to remove noise and scaling factor \(\xi\). Five independent simulations were performed for each map, where \(\xi\) was progressively increased from 0 to 5 kcal/mol over the first 50 ns followed by 50 ns of constant \(\,\xi\) = 5 kcal/mol. For the WT simulations, the RMSD for the backbone atoms of the pore domain helices and the selectivity filter plateaued after 10 ns and 40 ns, respectively. 50 final structures were generated by energy minimizing with 1000 steps of steepest descent every ns with \(\xi\) = 10 kcal/mol starting after 50 ns simulation. For N588K and N588E HERG, a final structure was saved from an MDFF simulation to use as the starting point for subsequent K+ ion permeation US simulations (see below).Molecular dynamics simulations of HERG in membranesSimulations of HERG with maintained ion configurationsMD simulations were performed starting with the WT HERG high-K structure (pdb: 9CHP). The pore domain (residues 545 to 667) was embedded in a lipid bilayer consisting of palmitoyloleoyl-phosphatidylcholine (POPC) lipids (238 molecules). It was surrounded with 18,771 explicit TIP3P water molecules67 and 150 mM KCl (49 K+ ions and 53 Cl− ions). During initial equilibration, all heavy atoms in the protein, as well as ions in the selectivity filter, were constrained with harmonic constraints with force constants of 10 kcal/mol/Å2. Constraints on the protein were slowly released during 5 ns simulations before production runs.Libraries of 10 independent simulations each of 50 ns were performed for the WT channel with each of 9 different ion configurations in the selectivity filter. Multi-ion configurations in sites S0/S2/S4, S2/S4, and S1/S3 as well as single ion configurations in sites S0–S4 and an empty filter were constrained with flat-bottomed potentials (k = 10 kcal/mol/Å2 and thickness 3.5 Å)) holding the ions in their sites. Means and distributions were calculated after excluding the first 10 ns of equilibration. Errors bars were calculated as the standard error of means for the 10 different simulations for each ion configuration.The angle of the selectivity filter backbone carbonyls, θ, were calculated as the angle between the projection on the xy-plane of the vector between the backbone carbonyl C and the backbone carbonyl O and the projection on the xy-plane of the vector between the backbone carbonyl C and the center of mass of the selectivity filter. Distributions of backbone carbonyl angles (0° indicates pointing towards center of pore) have been plotted as violin plots, and the proportion of time 0, 1, 2, 3, or 4 backbone carbonyls were pointing inwards (defined as <70° deviation from 0°) reported.Cluster analysis was performed with k-means algorithm68 in MATLAB using k = 10 clusters and clustering on the backbone Φ and Ψ angles of the selectivity filter (residue 624 to 628). Initial cluster centroids were generated using the k-means++ algorithm. Each point in space was then allocated to its nearest cluster and the cluster centroid recalculated. This was done iteratively until self-consistent. Each subunit was clustered separately to allow for asymmetries. Analysis of backbone carbonyl angles and interaction distances (d) was performed based on which cluster the system was in (blue conducting, orange non-conducting). Free energy maps, as function of S620 side chain interactions with V625 backbone carbonyl, G626 amide, F627 amide or Y616 backbone carbonyl plotted against rotation of the V625 backbone carbonyl have been obtained from analysis of all frames of the constrained ion MD simulations of HERG in the membrane. Maps were created from calculation of \(\Delta G\left(d,\theta \right)=-{k}_{B}T{\mathrm{ln}}\rho \left(d,\theta \right)+C\), where ρ is the probability distribution as a function of the interaction distance d and V625 backbone carbonyl orientation \(\theta\).Conduction simulationsSimulations starting with either the high-K structure (pdb: 9CHP) or low-K structure (pdb: 9CHQ) were similar to those described above, but included 500 mM KCl solution (165 K+, 169 Cl− ions and 18,653 water molecules), including 5 K+ ions that were initially placed at the centers of sites S0–S4 in the equilibrated structure. The CHARMM36 force field without modifications was used for these simulations. This system was run for 0.5 μs (two independent simulations) with a constant electric field equivalent to an applied membrane potential of −500 mV (negative inside) to attempt to observe conduction. During equilibration, heavy atom harmonic position restraints on the selectivity filter and resident ions were applied with a force constant of 5 kcal/mol/Å2, relaxed to zero slowly over 10 ns. Conduction events were counted after excluding the first 10 ns of equilibration.To maintain this high-K cryo-EM-like structure, flat-bottom restraints were applied to loosely maintain H bonds identified in the cryo-EM structure. This required a weak flat-bottom restraint (force constant 5 kcal/mol/Å2 applied when 3.2 Å was exceeded) acting on the distance between the S620 side chain hydroxyl O, and both the G626 and F627 backbone amide H atoms for each subunit, for the conducting structure. This restraint is designed to not be felt unless the H-bond is attempting to break, as would occur during a structural isomerization of the backbone linkage. In addition, a flat-bottom harmonic restraint was applied to the intracellular gate. Specifically, the 6 Cα-Cα distances connecting adjacent carbon atoms of the same residue were weakly restrained with flat-bottom distance restraints, for the four Cα atoms (each) in residues F656, G657, and N658. The 18 flat-bottom restraints had a force constant of 5 kcal/mol/Å2, with no force applied until the Cα distance deviated more than ±2 Å from the distances observed in the experimental structure. These distance restraints were applied to maintain an open gate at the narrowest point.Two identical simulations were also run starting with the low-K structure for 0.5 μs each. In this case, an equivalent constraint was applied to maintain the identified H bond between the S620 hydroxyl H atom and the flipped backbone carbonyl O atom of residue V625.Umbrella sampling simulations of the V625 backbone carbonylUmbrella sampling simulations69 of the rotation of the V625 backbone carbonyl was performed by constraining the V625 backbone Ψ angle (N625-C625-Ca625-N626). The backbone Ψ angle controls the V625-Gly626 linkage and thus the orientation of the V625 backbone carbonyl oxygen (with Ψ∼−50° leading to the backbone carbonyl oxygen pointing in and Ψ∼+70° leading to the backbone carbonyl oxygen pointing out). This backbone dynamics has previously been shown in both KcsA and MthK15,29. Initial windows were created using steered MD with a harmonic force constant of 0.03 kcal/mol/°2 moving at a rate of 0.2 ns/°. The complete 360° Ψ rotation was divided into 72 windows separated by 5°. The backbone dihedral was constrained with a force constant of 0.03 kcal/mol/°2 in the center of each window. During the production run, multi-ions configurations in sites S0/S2/S4, S2/S4, S2/S3, and S1/S3 were constrained in their sites with flat-bottom potentials (k = 10 kcal/mol/Å2 and width 3.5 Å). Simulations were performed for the WT channel with ions in S0/S2/S4, S1/S3. A convergence criterion of a free energy change of less than 1 kcal/mol was used and this was achieved after 9 ns and 11 ns for WT with ions in S0/S2/S4 or S1/S3. All data prior to equilibration was discarded for final calculations. WHAM70 with periodic boundary conditions was used to calculate the free energy profile. Error bars were calculated as standard error of mean by dividing the data into 1 ns long blocks.Replica exchange with solute temperingTo study the behavior of the filter at 300 K temperature, replica exchange with solute tempering (REST2) simulations31 starting with the conductive structure were performed. 3 K+ ions were trapped in the vicinity of the selectivity filter/cavity region using a tall cylinder of 30 Å height and 15 Å width with flat-bottom half-harmonic constraints of 10 kcal/mol/Å2. Sixteen replicas with effective temperatures for ion interactions from 300 K to 900 K were used, where only the interactions involving the 3 trapped ions were scaled. Two independent systems, with ions starting in S0/S2/S4, or with S1/S3/cavity, were simulated for 500 ns each. The free energy profile, \(\Delta G\left(z\right)=-\!{k}_{B}T{\mathrm{ln}}\rho \left(z\right)+C,\) using only trajectory from the individual replicas of both simulations, is shown in Fig. S9 to demonstrate reproducibility across replicas and convergence over time. Here, ρ is the probability distribution as a function of reaction coordinate \(z\), where \(z\) is the position of an ion along the \(z\) coordinate with respect to the center of mass of the backbone atoms of the selectivity filter. The constant, \(C\), was chosen to set the free energy to zero in the extracellular solution. This effective free energy profile based on ion density due to the 3 ions trapped in the channel, reveals the apparent locations and depths of the binding sites, as well as the apparent barriers for ion movement between sites, but is distinct from the barriers that would be seen in a multi-ion permeation mechanism. It is thus used as a guide to understand the propensities for ion binding and movement.MBAR free energy analysis of REST2 simulationsWe have used MBAR33 to make full use of the sampling obtained in all 16 replicas of the two independent REST2 simulations. MBAR has been used for capturing the barriers in the 1D ion free energy profile in cases with 0–4 flipped V625 backbone carbonyls (Fig. 3D), and to improve sampling of barriers in 2D maps for S620 interactions (Fig. 4C–F). Our approach is based on the publicly available pyMBAR code for replica exchange33, modified to use ion interaction energies from REST2 with our own in-house codes. We sampled from trajectory infrequently (every 20 ps), to ensure uncorrelated samples.Analysis for the free energy of ions inside the selectivity filter (Fig. 3D) was performed by analyzing z position of each trapped ion (grid size 0.4 Å), together with the number of flipped V625 backbone carbonyls, to form 2-dimensional histograms. Individual PMFs for different number of flipped carbonyls were obtained by extracting cross-sections from the 2D map for each individual number of flipped carbonyls to produce separate 1D PMFs. The number of carbonyls that were pointing towards the channel axis was defined by testing whether the Ψ dihedral angle exceeded a cutoff of +30°, based on the flipping barriers seen in Fig. 3C; with the value chosen to ensure the barrier is overcome regardless of the ion configuration. Error bars in 1D free energy profiles represent ± one standard deviation obtained from MBAR33.Free energy maps for S620 side chain hydrogen-bonding interactions with surrounding residues (Fig. 4C–F) were also analyzed using MBAR. Two-dimensional histograms were formed as a function of the carbonyl angle away from the channel axis (θ; defined above; grid size 10°) and a distance involving the S620 side chain. The distances analyzed included that between S620 side chain hydroxyl H and V625 backbone carbonyl O or Y616 backbone carbonyl O, as well as S620 side chain hydroxyl O with V626 backbone amide H or F627 backbone amide H (bin size 0.2 Å). Errors in barrier estimates were obtained from MBAR errors at the location of highest free energy along the minimal free energy pathway between two states.Reporting summaryFurther information on research design is available in the Nature Portfolio Reporting Summary linked to this article.
Potassium dependent structural changes in the selectivity filter of HERG potassium channels
