Cryo-EM structure of the open human ether-à-go-go related K⁺ channel hERG

Summary

The human ether-à-go-go related potassium channel (hERG, Kv11.1) is a voltage-dependent channel known for its role in repolarizing the cardiac action potential. hERG alteration by mutation or pharmacological inhibition produces Long QT syndrome and the lethal cardiac arrhythmia torsade de pointes. We have determined the molecular structure of hERG to 3.8 Å using cryo-electron microscopy. In this structure the voltage sensors adopt a depolarized conformation and the pore is open. The central cavity has an atypically small central volume surrounded by four deep hydrophobic pockets, which may explain hERG’s unusual sensitivity to many drugs. A subtle structural feature of the hERG selectivity filter might correlate with its fast inactivation rate, which is key to hERG’s role in cardiac action potential repolarization.

In brief

Structural analysis of the hERG channel may explain known human channelopathy mutations and why the channel is extremely sensitive to a wide range of drugs.

Introduction

hERG is a voltage-dependent K⁺ (Kv) channel found in neurons and cardiac cells (Sanguinetti and Tristani-Firouzi, 2006; Trudeau et al., 1995; Vandenberg et al., 2012; Warmke and Ganetzky, 1994). Its function is best understood in the heart where it terminates the action potential (AP) (Curran et al., 1995; Sanguinetti and Tristani-Firouzi, 2006). Although hERG contains a K⁺-selective pore and four voltage sensors like other Kv channels, its gating properties are atypical (Sanguinetti et al., 1995; Smith et al., 1996; Trudeau et al., 1995; Warmke and Ganetzky, 1994). Inactivation occurs so rapidly that relatively little K⁺ current is observed immediately following membrane depolarization, during the early phase of the AP. As the membrane repolarizes, however, hERG reopens as it recovers from inactivation, facilitating rapid AP termination. When hERG is defective the cardiac AP is prolonged, resulting in a hallmark lengthening of the QT interval on the electrocardiogram (ECG) (Curran et al., 1995; Sanguinetti et al., 1996). This clinical condition, known as Long QT syndrome (LQT), is a harbinger of the lethal arrhythmia torsade de pointes (Viskin, 1999). LQT from hERG malfunction is observed in two distinct contexts, heritable missense mutations in the hERG gene (LQT2) and drug-induced (LQTdi) (Curran et al., 1995; Roden, 2004; Roden et al., 1996; Sanguinetti et al., 1995).

LQTdi results from off-target inhibition of hERG, most commonly because the drug blocks the ion pathway (Mitcheson et al., 2000; Sanguinetti and Mitcheson, 2005; Sanguinetti and Tristani-Firouzi, 2006). The list of drugs that inhibit hERG includes (but is not limited to) antibiotics (grepafloxacin (Bischoff et al., 2000)), antimalarials (halofantrine (Nosten et al., 1993) and quinine (Sanchez-Chapula et al., 2003)), gastroprokinetic agents (cisapride (Vitola et al., 1998)), antihistamines (astemizole (Zhou et al., 1999)), antiarrhythmics (quinidine (Roden et al., 1986) and dofetilide (Jurkiewicz and Sanguinetti, 1993)) and anti-psychotics (sertindole (Rampe et al., 1998)). These compounds are chemically diverse and yet they all inhibit hERG. The question naturally arises: why is hERG so susceptible to such a wide array of compounds?

In this study we determined the structure of hERG using cryo-electron microscopy (cryo-EM). The voltage sensors are depolarized and the inner helical gate is open. An unusual geometry of the ‘central cavity’ within hERG’s ion conduction pathway offers a likely explanation for this channel’s extreme susceptibility to a wide range of drugs. Furthermore, hERG contains a subtle difference in the structure of its selectivity filter compared to all other K⁺ channels so far analyzed. This difference appears to be correlated with its unusually rapid inactivation rate.

Results

Biochemical and functional analysis

The wild-type hERG (hERG_WT) channel aggregates during purification. To overcome this limitation we deleted segments of the amino acid sequence that are predicted to be disordered (Buchan et al., 2013) while monitoring biochemical and electrophysiological behavior. We identified a construct, hERG_T, with functional properties very similar to the wild-type channel, in which two segments (141–350 and 871–1005) were deleted (Fig. 1A). The truncated construct contains 814 amino acid residues while the full length hERG contains 1159. Under voltage clamp hERG_WT and hERG_T both produce relatively small outward currents during membrane depolarization and large inward ‘tail currents’ during subsequent repolarization (Fig. 1B, C). These well-known properties of hERG reflect rapid inactivation during depolarization and quick recovery from inactivation but slow deactivation (closing) during repolarization (Smith et al., 1996; Trudeau et al., 1995). The voltage-dependent activation curve (Fig. 1D) and pharmacological properties (Fig. 1E) of hERG_T are also similar to hERG_WT (Snyders and Chaudhary, 1996; Zhou et al., 1999). Therefore the deleted segments in hERG_T do not appear to alter the major functional properties of the hERG channel that we wish to understand.

Figure 1. A hERG construct for cryo-EM structure determination.

(A) The primary structure of hERG with domains indicated in different colors. Amino acid residues between 140–351 and 870–1006 were removed in the hERG_T construct. (B) and (C) Representative whole-cell voltage family recordings in transfected HEK-293T cells of (B) hERG_WT and (C) the hERG_T construct. Bath solution contained 20 mM K⁺ and pipette solution contained 120 mM K⁺. Voltage protocol is illustrated at the bottom of (C). Cell membrane was held at −80 mV, depolarized to −80 mV to 60 mV for 2 s in 10 mV steps, and then repolarized back to −80 mV. The open arrows denote the steady state currents during depolarization and the closed arrows denote the tail currents upon repolarization. Dashed lines indicate the 0 current level. (D) Normalized tail currents (mean ± SEM) of hERG_WT (blue triangles, n = 6) and hERG_T (black circles, n = 11) are plotted as a function of the depolarization step voltage and fitted with Boltzmann equation (solid lines. blue, hERG_WT; black, hERG_T. See STAR Methods). Fitted half activation voltage V_1/2 and apparent valence z are shown beside the plots. (E) Dose response characteristics of hERG_T to astemizole (black circles, n=3, mean ± SEM) and dofetilide (blue triangles, n=3, mean ± SEM). IC₅₀ values from rectangular hyperbolar fit (solid lines. blue, dofetilide; black, astemizole, see STAR Methods) are shown in the figure.

Milligram quantities of hERG_T were expressed and purified. Single-particle cryo-EM method was used to obtain a density map with a resolution of 3.8 Å (Fig. S1) (~144k particles in the final reconstruction). We also determined the structures of two additional mutations, hERG_Ts and hERG_Ts S631A at 3.7 Å and 4.0 Å, respectively. The details of these structure determinations are given in STAR Methods. Local resolution was estimated with program “blocres” (Heymann and Belnap, 2007). A Fourier Shell Correlation (FSC) cutoff of 0.5 was selected for local resolution estimation so that the maximum and minimum local resolutions are above and below the global resolution of the map. Estimated local resolution range from ~3.2 Å in the central core of the transmembrane domain to ~5.6 Å in peripheral (especially cytoplasmic) domains (Fig. 2A, Fig. S2 and S6). For hERG_Ts a new model with side chains (using the structure of EAG1 as a starting point (Whicher and MacKinnon, 2016)) was built for the transmembrane region and the central portion of the cyclic nucleotide binding homology domain (CNBHD). The Per Arnt Sim (PAS) domain was modeled by docking the crystal structure (PDB ID 1BYW, Morais Cabral et al., 1998) into the density with minor adjustment where side chain differences were apparent. These differences were small and mostly occurred in regions where the PAS domain contacted the CNBHD, which was not present in the crystal structure. hERG_T and hERG_Ts S631A models were built starting from the hERG_Ts model. The final model excludes the first two N-terminal amino acids, residues 132–397 linking the PAS domain to the first transmembrane helix (predicted to be unstructured (Buchan et al., 2013)), extracellular loops of the transmembrane domain (434–451, 512–519, 578–582, 598–602) and C-terminal residues 864–1159, also predicted to be unstructured. We note that all mechanistic points addressed in this study concern regions of the channel (transmembrane domain, cytoplasmic cavity and selectivity filter) in which side chain density was well defined.

Figure 2. Overall architecture of the hERG channel.

(A) Local resolution of reconstructed density map as estimated using software “blocres” (Heymann and Belnap, 2007) with 20 voxel box size and 0.5 Fourier Shell Correlation (FSC) cutoff. Density map is colored according to the local resolution using UCSF Chimera (Pettersen et al., 2004). (B) hERG overall architecture viewed down the four-fold axis from the extracellular side. hERG channel is represented as semi-transparent molecular surface with ribbon representation for one subunit. Domains are colored as in figure 1A. (C) Stereo view of the hERG channel from the side. See also figures S1, S2 and S5.

Domain architecture of hERG

The three-dimensional domain organization of the hERG channel is shown in figure 2B and C. Each subunit of the four-fold symmetric tetramer contains a voltage sensor (helices S1–S4) and pore-forming region (S5–S6) inside the membrane bilayer and a PAS domain, C-linker helices and a CNBHD projecting into the cytoplasm (Fig. 1A and Fig. 2B, C). hERG and its molecular cousin the EAG1 K⁺ channel share these same structural domains (Whicher and MacKinnon, 2016). EAG1, however, is regulated by Ca²⁺ and calmodulin in addition to membrane voltage. This distinction has resulted in a substantial conformational difference in the cryo-EM structures of these two channels, which lends insight into the activation gating mechanism in this class of K⁺ channels.

A conceptual model for voltage sensor activation

The pore of the hERG channel is open (Fig. 3A, B). The closely related EAG1 channel is closed owing to the presence of Ca²⁺ and calmodulin, which lock its pore shut (Whicher and MacKinnon, 2016). The hERG-EAG1 comparison provides an informative picture of the conformational changes that seem to occur in this class of K⁺ channels. At its narrowest point the gated region of the pore in EAG1 (Q476 on the S6 helix) is closed to a radius less than 1 Å (Fig. 3B). At the corresponding location in the hERG channel (Q664) the pore’s radius is almost 6 Å (for reference, the smallest radius of the ion conduction pore in the open structure of Kv1.2–2.1 paddle chimera (KvChim) is around 5 Å) (Fig. 3B). Superposition of the two structures shows that the large conformational difference in the S6 helices is enabled by a glycine gating hinge (G648 in hERG, G460 in EAG1) (Jiang et al., 2002) (Fig. 3C).

Figure 3. hERG channel is in the open conformation.

(A) The central pore generated with program “hollow” (Ho and Gruswitz, 2008) is shown as yellow surface. Only two of the opposing hERG subunits are shown as blue ribbons for clarity. (B) The radius of the transmembrane pore (calculated with “HOLE” software (Smart et al., 1993)) plotted against the displacement from the top of the selectivity filter (blue: hERG, cyan: KvChim (PDB ID 2R9R), gray: EAG1 (PDB ID 5K7L)). Positions of constriction site residues Q476 in EAG1 and its counterpart Q664 in hERG are indicated with arrows. (C) Stereo view of two opposing subunits of hERG (blue) and EAG1 (gray) overlaid. hERG and EAG1 tetramers were aligned using the selectivity filter and the pore helix. Secondary structure elements are labeled in bold. Constriction site residues (hERG Q664 and EAG1 Q476) are shown as sticks. “Gating hinge” glycines (hERG G648 and EAG1 G460) are shown as spheres.

The voltage sensor in the hERG structure, as it was observed in EAG1, is not domain-swapped, meaning it is packed against its own (i.e. same polypeptide chain) pore subunit (Fig. 2C and Fig. 3C). The hERG S4 helix contains five positively charged amino acids. The first three – K1, R2 and R3 – are located above (i.e. on the extracellular side of) the gating charge transfer center, with R4 and R5 below (Fig. 4) (Tao et al., 2010). Electrophysiological studies of wild-type and mutant hERG channels showed that K1, R2 and R3 all influence voltage-dependent gating while R4 and R5 do not (Zhang et al., 2004). These functional measurements therefore suggest that hERG’s voltage sensor is in a depolarized conformation in the structure because K1, R2 and R3 are closer to the extracellular side of the membrane. An electric field within the membrane, displacing these residues toward the inside during pore closure, would account for their contribution to gating charge.

Figure 4. Conformation of the hERG voltage sensor.

Stereo view of the hERG voltage sensor. Gray lines indicate the membrane-solution interfaces. Basic residues (labeled in blue) at positions K1 to R5 and the gating charge transfer center residues (red labels for acidic residues and green for the aromatic) are shown as sticks. Secondary structure elements are labeled in bold.

If the closed pore in EAG1 (produced by the presence of Ca²⁺ and calmodulin) is representative of hERG’s closed pore, then we can ask what kind of voltage sensor movement could bring about the same conformational change in the pore (Fig. 3C)? To be clear, the voltage sensors in both hERG and EAG1 are depolarized: in hERG this conformation is compatible with the open state of the pore; in EAG1 it is not (calmodulin has an overriding effect). It would appear that an inward (toward the cytoplasm) and centric (toward the pore axis) displacement of S4 (driven by a membrane electric field) could produce similar pore closure.

The large body of work on voltage sensing has focused mainly on Shaker-like (i.e. Kv1) channels, which have domain-swapped voltage sensors. Domain-swapping occurs because an α-helical linker connects each voltage sensor to a neighboring subunit (i.e. the polypeptide chain of an adjacent subunit). In such channels α-helical linkers (one for each of four voltage sensors) form a cuff around the S6 helices, which form the pore’s gate. Conformational changes within the voltage sensors therefore can be transmitted to the gate through the linkers (Long et al., 2005a, b; Lu et al., 2001). The recent discovery of the non-domain-swapped architecture in EAG1 and observed again in hERG implies a very different mechanism of voltage sensor coupling because α-helical S4–S5 linkers do not exist. How might the voltage sensors exert force on the pore in hERG and EAG1? Figure 3C suggests a possible answer. In the superposition of the closed EAG1 channel (mediated by Ca²⁺ and calmodulin) with the open hERG channel there is no translation of S4 across the membrane. However, S5 maintains an extensive anti-parallel contact with S6 in both structures. This observation leads us to propose that voltage sensors in hERG and EAG1 channels transmit force through the S5–S6 interface. An inward and centric displacement of S4 would thus close the S6 helical gate by compressing the S5 helices.

This transmission of force would be somewhat different than the lever mechanism proposed for Shaker-like Kv channels (Long et al., 2005a, b; Lu et al., 2001). Functional measurements also point to a substantial difference in the conformational changes within hERG and Shaker-like Kv voltage sensors. In Shaker the total gating charge is between 12 and 16 elementary charge units (Aggarwal and MacKinnon, 1996; Ledwell and Aldrich, 1999; Seoh et al., 1996) whereas in hERG the value is about 6 (Zhang et al., 2004). These functional measurements imply that voltage sensor conformational changes are smaller in the hERG K⁺ channel.

Unique central cavity and hERG drug sensitivity

Mutagenesis studies have identified amino acids that appear to form the high-affinity drug-binding site in hERG (Chen et al., 2002; Mitcheson et al., 2000; Perry et al., 2004). The positions of these amino acids in sequence and in the molecular structure are highlighted (Fig. 5A–D). In the structure they are located near the central cavity and in the selectivity filter adjacent to the central cavity (Fig. 5B–D). Some drug-important amino acids are unique to the EAG family of K⁺ channels and some are not (Fig. 5A). The structure offers an explanation as to why these particular amino acids are important: they line the surfaces of elongated, relatively hydrophobic pockets that extend from the central cavity (Fig. 5D). The pockets are not present in other K⁺ channels, including Kv1 (Fig. 5E), Slo1 and Kir (Fig. S3). They exist in hERG because the S6 inner helix is displaced to create a separation between the pore helix (PH) and S6 helix (Fig. 5C, D). The hydrophobic pockets are roughly cylindrical in shape with a diameter of about 8 Å and a depth about 11 Å (measurements between atomic centers). The longest dimensions of astemizole and dofetilide are more than 20 Å and the narrowest dimension around 3~5 Å. Thus, we propose that one part of a blocking drug fits into a pocket. We note that the substituted aromatic rings of astemizole and dofetilide fit snugly into the pockets. If this occurs, these drugs occupy the center of the cavity and insert a functional group into one of the hydrophic pockets.

Figure 5. Local chemistry of a putative drug-binding site.

(A) Sequence alignment of selected K⁺ channels near the putative drug-binding site. Conserved residues are shown in bold. Residues related to drug binding are colored in yellow with arrows indicating the amino acid residue numbers in hERG. (B) EM density map of the putative drug-binding site is shown as blue mesh and molecular model shown as sticks. Residues related to drug binding are highlighted in yellow. Only one subunit is shown for clarity. (C) Overlay of hERG (blue) and KvChim (cyan) near the central cavity with drug binding related residues shown as sticks. (D) Internal molecular surface around the central cavity of hERG is represented as translucent surface colored by eletrostatic potential according to the scale shown. Residues related to drug binding are shown as sticks on the otherwise ribbon representation of the channel. (E) Internal molecular surface around the central cavity of KvChim represented similarly as in (D). See also figure S3.

We also think that the small volume of hERG’s central cavity may contribute to its susceptibility to drugs. The central cavity of K⁺ channels in general is electronegative owing to the orientation of four pore helices, which direct their C-terminal partial negative ‘end charge’ toward the cavity (Roux and MacKinnon, 1999, Zhou and MacKinnon, 2004). A smaller volume of high dielectric medium (i.e. water) surrounded by a relatively lower dielectric medium (i.e. protein) will exhibit a more negative electrostatic potential. For example, the calculated potential (see STAR Methods) at the center of the larger cavity in KvChim is approximately −125 mV compared to −625 mV in hERG (Fig. 5D, E, 1 kT/e ≈ 25 mV at room temperature). This substantial difference is due to cavity size. Since many drugs that block hERG contain a positive charge, or can form a cationic conjugate acid, a more negative electrostatic potential will increase drug affinity.

A possible basis for rapid inactivation

Having analyzed the molecular structures of many different K⁺ channels we noticed a subtle but clear difference in hERG’s selectivity filter (Fig. 6 A, D). The side chain of F627 in the GFG (or GYG in some K⁺ channels) sequence, well defined in the map (Fig. S1G-I and S6), is uniquely positioned in hERG (Fig. 6D, E). This point is illustrated in figure 6D, which shows a superposition of hERG with EAG1, a closely related K⁺ channel in a closed conformation, with KvChim, a more distantly related K⁺ channel in an open conformation, and with KcsA, a very distantly related bacterial K⁺ channel in a closed conformation. These other K⁺ channels, different in type and conformational state with respect to the S6 inner helical gate, are nearly identical with respect to the position of F (or Y) in the selectivity filter. Because hERG is functionally unique with respect to its nearly instantaneous inactivation, we wondered whether the structure could represent an inactivated conformation, manifest as a uniquely appearing selectivity filter. To test this possibility we examined a hERG mutant, S631A, that does not inactivate very much (Fig. 6G, H) (Schonherr and Heinemann, 1996). The structure of the S631A mutant shows that its selectivity filter is now like other K⁺ channels with respect to the position of F627, supporting the idea that instantaneous inactivation is correlated with the unusual position of F in the GFG sequence (Fig. 6F and Fig. S4). We note that structural studies on the S631A mutant were carried out in a background truncation mutant (hERG_Ts) that is slightly different than hERG_T and shares essentially the same overall structure as hERG_T (Fig. S4A, see STAR Methods). In transfected HEK-293T cells, hERG_Ts starts to open at more negative voltages, exhibiting a 20 mV leftward shift in its voltage dependent activation curve (at half maximum activation) compared to hERG_WT (Fig. S4E). Voltage-dependent rapid inactivation and slow deactivation (Fig. S4B) as well as pharmaceutical properties of hERG_Ts (Su et al., 2016) are similar to hERG_WT. Control experiments show that the full correlation between inactivation and the selectivity filter conformation exists in the hERG_Ts construct. Namely, hERG_Ts inactivates (Fig. S4B) and its F position is like that in hERG_T (Fig. S4C) and hERG_Ts S631A does not inactivate (Fig. S4D) and its F position is like other K⁺ channels (Fig. 6D, F and Fig. S4C). At the resolution of these structures the position of the F627 side chain is easily discernable. Detection of more subtle conformational differences within the selectivity filter will require higher resolution data.

Figure 6. Selectivity filter rearrangements accompanying hERG rapid inactivation.

Side view of the cryo-EM density maps (translucent surfaces) of the selectivity filters of (A) hERG_T (blue), (B) EAG1 (gray) and (C) a non-inactivating mutant hERG_Ts S631A (green). “*” indicate the positions of the selectivity filter aromatic residues (F627 in hERG, F439 in EAG1). Only two opposing subunits are shown for clarity. (D) Overlay of the equivalent aromatic residues in the selectivity filter shown as sticks, hERG_T F627 (blue), hERG_Ts S631A F627 (green), EAG1 F439 (grey), KcsA Y78 (yellow) and KvChim Y373 (cyan). (E) Density maps of F627 in hERG_T (blue mesh) and F439 in EAG1 (grey mesh) viewed from the extracellular side along the four-fold axis. Stick models are shown for one subunit of each channel. Arrows highlight the counter-clockwise rotation of this residue. (F) Overlay of the density maps and model of residue F627 in hERG_T (blue) and hERG_Ts S631A (green) in a similar way as in (E). (G) and (H) Representative whole-cell voltage family recordings in HEK-293T cells of (G) hERG_T and (H) a non-inactivating mutant hERG_T S631A. See also figures S1, S4–S6.

Discussion

A structure of the hERG K⁺ channel advances our understanding in three respects. First, hERG is a Kv channel that is structurally distinct from the more thoroughly studied Shaker-like Kv channels (Long et al., 2005a). It was discovered only in the past year that Kv10 (EAG), Kv11 (ERG) and Kv12 (Elk) channels have, compared to Kv1–Kv9 channels, a fundamentally different manner of adapting their voltage sensors to the pore (Whicher and MacKinnon, 2016). Instead of encircling the S6 helical gate with a cuff of α-helices, Kv10–Kv12 voltage sensors make more tenuous contacts with the pore through a short non-helical linker. The comparison of hERG and EAG1 channels presented here lead us to propose that voltage sensor movements in Kv10–Kv12 channels are transmitted at least in part through S4-mediated displacements of S5, which is packed against the S6 helical gate.

Second, the small central cavity in hERG contains extended pockets that can account for its sensitivity to a wide range of drugs. The pockets are non-existent in most other K⁺ channels and are constricted shut in the closed state of the closely related EAG1 channel. The unusually small volume of the main central cavity probably also favors the binding of cationic (or potentially cationic) drugs by amplifying the electrostatic potential. Two aspects of the cavity have confounded our attempts to determine the structure of hERG with dofetilide and astemizole. The first confounding factor is the presence of strong density at the center of the cavity even in the absence of drug. This density undoubtedly reflects the strong negative electrostatic potential inside the cavity, which results in the presence of a cation (whether a metal ion from solution or a drug) to form a more stable structure. The second is the asymmetric manner in which drugs will bind, resulting in partial occupancy (probably quarter) of a functional group inside the hydrophobic pockets.

Third, a subtle difference in the conformation of the hERG selectivity filter compared to other K⁺ channels – a difference that is rectified by a mutation that interferes with inactivation – suggests that the subtle difference might be related to hERG inactivation. More data are required to substantiate this suggestion. We raise the point to stimulate further study. We also note that the filter conformation is unlike that attributed to C-type inactivation in the KcsA K⁺ channel (Cuello et al., 2010; Hoshi et al., 1990). In the KcsA filter the conformational change is large and similar to that observed when K⁺ ions are depleted from the filter (Zhou et al., 2001). In hERG the conformational difference from the most commonly observed filter structure is slight.

STAR Methods

Contact for Reagent and Resource Sharing

Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact Roderick MacKinnon (mackinn@rockefeller.edu).

Experimental Model and Subject Details

Cell Lines

HEK293S GnTI⁻ cells were used for protein expression and maintained in Freestyle 293 media supplemented with 2% Fetal Bovine Serum (FBS) at 37°C in 5% CO₂.

HEK293T cells were maintained in high glucose DMEM media supplemented with 10% FBS and and 2mM L-Glutamine in 5% CO2 at 37°C. They were used in electrophysiological recordings.

Method Details

Cloning of hERG constructs

DNA coding human ERG1 (KCNH2, hERG) was synthesized by Bio Basic Inc. Regions encoding residues 141–350 and 871–1005 were removed for hERG_T. Regions encoding residues 141–380 and 871–1005 were removed for hERG_Ts. Point mutation of serine 631 to alanine was achieved using quick-change PCR. Wild-type and mutant hERG DNAs were cloned into a modified BacMam expression vector (Goehring et al., 2014) with a C-terminal green fluorescent protein (GFP) followed by a 1D4 peptide tag (Molday and Molday, 2014). A PreScission protease site (LEVLFQ/GP) (Walker et al., 1994) is present between hERG and C-terminal GFP.

Expression and purification of hERG

The expression and purification procedure of hERG_T, hERG_Ts and hERG_Ts S631A is essentially the same as previously described (Su et al., 2016) with lower detergent and lipid concentration in the last size exclusion chromatography step. Briefly, viruses encoding hERG constructs were produced and GnTI-cells were infected at a density of 3 ×10⁶ cells/ml. Expression was induced 24 hours post infection with 10 mM sodium butyrate for 36 hours. Cells were collected and extracted with buffer containing 1% n-Dodecyl β-D-maltoside (DDM) and 0.2% Cholesteryl hemisuccinate (CHS) following a quick membrane prep. GFP-binding nanobody (Fridy et al., 2014) resin was used to isolate GFP fused hERG from the extracted supernatant and then washed in buffer containing 0.1% DDM, 0.02% CHS and 0.1 mg/ml phospholipids (1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (POPE): 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC): 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphate (POPA) = 5:5:1, ECA551). PreScission protease was used to elute hERG from GFP-binding nanobody resin. Eluent was concentrated and loaded onto Superose 6 size exclusion column in buffer containing 0.025% DDM, 0.005% CHS and 0.025 mg/ml ECA551. Peak fractions were collected, concentrated and used immediately.

Electrophysiology

BacMam expression vector bearing hERG constructs were transfected into HEK-293T cells cultured in DMEM (Gibco) media supplemented with 10% fetal bovine serum (FBS) and 2 mM L-Glutamine (ThermoFisher) using Lipofectamine 2000 (Invitrogen) following manufacturer’s protocol. Specifically, for each transfection of a 6-well plate well, 2 μg of DNA was diluted into 100 μl of Opti-Mem media. 6 μl of Lipofectamine 2000 was also diluted into 100 ul of Opti-Mem media. These two diluted solutions were mixed for around 5 minutes and added to the cells. Room-temperature whole-cell recordings were performed ~24 hours after transfection. The bath solution contained (in mM): 10 HEPES pH 7.4, 120 NaCl, 20 KCl, 2 MgCl₂, 1 CaCl₂ and 10 glucose. The pipette solution contained (in mM): 10 HEPES pH 7.4, 50 potassium phosphate pH 7.4, 5 EGTA-K, 2 MgCl₂, 60 NMDG chloride. Borosilicate glass pipettes with resistance between 2~5 MΩ were used. A digidata 1440 digitizer (Molecular Devices) interfaced to pClamp10.5 software was connected to an Axopatch 200B amplifier (Molecular Devices) for data acquisition. Analog signals were filtered at 1 kHz and subsequently sampled at 20 kHz and stored on a computer. Unless otherwise mentioned, a voltage protocol of holding at −80 mV, stepping to −80 mV to 60 mV for 2 seconds in 10 mV steps, then stepping back to −80 mV was used for measurements of hERG tail current. For characterization of voltage dependent activation, tail currents were normalized using I_norm = (I – I_min)/(I_max – I_min) and fit to Boltzmann equation I_norm = 1/(1 + exp(−z × (V − V_1/2)/(RT/F))) where the constant RT/F ≈ 25.7 mV at 25 °C, z is the apparent valence and V_1/2 is the half activation voltage. For astemizole and dofetilide titrations, the bath solution contained (in mM): 10 HEPES pH 7.4, 135 NaCl, 5 KCl, 2 MgCl₂, 1 CaCl₂ and 10 glucose. A voltage protocol of holding at −80 mV, stepping to 60 mV for 2 seconds then stepping back to −40 mV was used for tail current measurements. Compounds were continuously perfused in at different concentrations. Data was fitted to simple 1:1 rectangular hyperbolar inhibition isotherm taking the form of I = 1 – x/(IC₅₀ + x), where I is the normalized tail current (normalized to the current level without compound in the same cell), x is the compound concentration and IC₅₀ is the apparent half inhibition concentration.

Cryo-EM sample preparation and data collection

Surface property of Quantifoil 400 mesh gold R1.2/1.3 holey carbon grids were modified by 12 s of glow discharge in low pressure air. 3.5 μl of protein sample at approximately 6 mg/ml were applied to the carbon coated side of the grid, blotted and vitrified in liquid ethane using a Vitrobot Mark IV (FEI). The blotting parameters used were: 98% humidity, 3 s blot time, 0 blot force and 1 total blot. The blotted grids were stored in liquid nitrogen until imaged.

Titan Krios operating at 300 keV equipped with a Gatan K2 Summit camera controlled by SerialEM (Mastronarde, 2005) was used for automated data collection of frozen single particles. Movies were recorded in super-resolution mode (0.65 Å/pixel) with a defocus range of −0.8 μm to −3.5 μm. 50 of 0.3 s frames were recorded per movie at a dose rate of 1.8 electrons per Ų per frame.

Cryo-EM data processing

For all three structures gain reference was applied to the super-resolution frames and compressed 2 × 2 by Fourier cropping in “mag_distortion_correct” (Grant and Grigorieff, 2015a). Whole frame motion correction was performed and summed using Unblur (Grant and Grigorieff, 2015b) for particle picking. The contrast transfer function parameters were estimated for each summed micrograph using CTFFIND4 (Rohou and Grigorieff, 2015). Motion of individual particles were subsequently corrected using an algorithm, alignparts, developed by Rubinstein and Brubaker (Rubinstein and Brubaker, 2015). All refinement and classification steps are performed in C4 symmetry.

For hERG_Ts around 4500 particles were manually picked and subjected to RELION (Scheres, 2012) 2D classification requesting 30 classes. Autopick was performed using 4 good 2D class averages representing typical particle orientations as template on 4100 aligned micrographs, resulting in a total of ~830k particles. Particles were subjected to 2D classification from which nicely averaged classes containing ~630k particles were manually selected. Good class averages were used to generate an initial model with C4 symmetry using EMAN2 (Tang et al., 2007). A RELION 3D auto-refine was then performed with the 630k particles using the initial model as the reference resulting in a 4.3 Å resolution map according a 0.143 cutoff criterion on the FSC. Post processing in RELION using a soft mask to suppress contribution by the detergent micelle followed by repeated FSC yielded a resolution estimate of 3.9 Å (0.143 cutoff). Refined particles were 3D-classified requesting 10 classes without particle alignment using a soft mask encompassing the protein and excluding the detergent micelle. The best 3 classes containing ~213k particles were selected and subjected to another 3D refinement run, resulting in a final 3.7 Å resolution map after postprocessing.

For hERG_T, the same template from hERG_Ts was used for autopicking 2443 aligned micrographs, resulting in a total of ~555k particles. After individual particle alignment all particles were subjected to 3D auto-refine using the hERG_Ts map as a reference (low-pass filtered to 60 Å), yielding a 4.9 Å resolution map (4.3 Å after post-processing). Refined particles were 3D-classified into 8 classes without alignment using a soft mask excluding the detergent micelle. Particles from the two best classes (~144k) were pooled and subjected to 3D refinement resulting in a 4.9 Å resolution map (4.1 Å after post-processing). This set of particles were further refined using Frealign (Grigorieff, 2007) using an 8 Å resolution cutoff and a soft mask excluding detergent micelle, resulting in the final map at 3.8 Å resolution.

For hERG_Ts S631A, the same template was used for autopicking 1505 aligned micrographs, resulting in a total of ~333k particles. After individual particle alignment, all particles were subjected to 3D auto-refine using the hERG_Ts map as a reference (low-pass filtered to 60 Å), yielding a 6.2 Å resolution map (5.9 Å after post-processing). Refined particles were 3D-classified into 8 classes without alignment using a soft mask to exclude the detergent micelle. Particles from 4 best classes (~206k) were pooled and a 3D refinement was carried out resulting in a 5.7 Å resolution map (4.4 Å after post-processing). Frealign was used to further refine these particles using a 7 Å resolution cutoff and a soft mask excluding detergent micelle, resulting in the final map at 4.0 Å resolution.

Model building

Atomic models were built in software Coot (Emsley et al., 2010). For the transmembrane domain and CNBHD, the EAG1 structure (PDB ID 5K7L) (Whicher and MacKinnon, 2016) was fitted into the density map of hERG_Ts, mutated into hERG sequence and rebuilt. The crystal structure of the PAS domain of hERG (PDB ID 1BYW) (Morais Cabral et al., 1998) was fitted into the density map and adjusted locally where the map quality was sufficiently good. The hERG_Ts model was subsequently docked into the density maps of hERG_Ts S631A and hERG_T. Adjustments were made to yield the corresponding atomic models. The models are mostly complete with the following exclusions: two N-terminal amino acids, the unstructured linker between the PAS domain and transmembrane domain (132–397), extracellular loops of the transmembrane domain (434–451, 512–519, 578–582, 598–602) and the unstructured C-terminal residues 864–1159.

Model refinement and validation

One of the two half maps resulting from RELION or Frealign refinement was used to refine the atomic models. For each structure, the half map used for refinement is named the “work half map” while the other one is named the “free half map”. Models and maps were both translated to a box 5 Å offset from the edge in all 3 directions to reduce computational intensity. Real-space refinement was performed with PHENIX real-space refinement (Adams et al., 2010) with secondary structure restraints. A mask was created by extending 3 Å from the real-space refined model and applied to the work half map for solvent flattening. Solvent flattened structure factors were then calculated and used for reciprocal-space refinement with Refmac (Brown et al., 2015; Winn et al., 2011) using secondary structure restraints generated by ProSMART (Nicholls et al., 2014). MolProbity was used to validate the geometries of the refined models (Chen et al., 2010). Corrected Fourier shell correlation curves were calculated (Oldham et al., 2016; Sindelar and Grigorieff, 2012) between each refined atomic model and the work/free half maps as well as the full map to assess the correlation between the model and density map. Statistics of cryo-EM data processing and model refinement are listed in figure S5.

Electrostatics calculation

Atomic coordinates were prepared using PDB2PQR (Dolinsky et al., 2007) using the Assisted Model Building with Energy Refinement (AMBER) force field (Cornell et al., 1995). Electrostatics were then calculated using continuum solvation methods with the Adaptive Poisson-Boltzmann Solver (APBS) tools incorporated in PyMOL (The PyMOL Molecular Graphics System, Version 1.8 Schrödinger, LLC.). The calculation was performed assuming a protein relative dielectric of 2 and solvent dielectric of 78. 0.15 M monovalent cation/anion were included in the solution. The nonlinear Poisson-Boltzmann equation was solved assuming a temperature of 37°C. Solutions were visualized by coloring the van der waals surfaces of channels in a blue to red gradient to depict electrostatic values.

Quantification and Statistical Analysis

Cryo-EM

Resolution estimations of cryo-EM density maps are based on the 0.143 Fourier Shell Correlation criterion (Rosenthal and Henderson, 2003).

Electrophysiology

Error bars in Figure 1 and Figure S4 represent standard error of the mean for 5 to 11 independent experiments. Data regression was performed using software OriginPro (OriginLab) and regression quality assessed by R² values.

Data and Software Availability

Data Resources

Atomic coordinates and maps of hERG_T (PDB ID: 5VA2, EMDB ID: EMD-8651), hERG_Ts (PDB ID: 5VA1, EMDB ID: EMD-8650) and hERG_Ts S631A (PDB ID: 5VA3, EMDB ID: EMD-8652) have been deposited with the Protein Data Bank at http://www.rcsb.org.

Supplementary Material

1. Figure S1. Structure determination of hERG_T, hERG_Ts and hERG_Ts S631A using single particle cryo-EM method, related to figures 2 and 6.

(A) A representative micrograph with scale bar shown in white. Micrographs for all three constructs have similar appearance. (B) Eight good 2D class averages from the hERG_Ts dataset (scale bar: 10 nm). (C) Orientation distribution (as reported by RELION) of the particles included in the final reconstruction of hERG_Ts. Taller red columns represent more particles than shorter blue ones. Distribution profiles of hERG_T and hERG_Ts S631A are similar (not shown). (D, E, F) Fourier shell correlation curves of (D) hERG_T, (E) hERG_Ts and (F) hERG_Ts S631A. Horizontal dashed lines represent the 0.143 FSC cutoff value. Vertical dashed lines in (D) and (F) indicate the high resolution limit during Frealign refinement. (G, H, I) Local resolution of reconstructed density map of (G) hERG_T, (H) hERG_Ts, and (I) hERG_Ts S631A estimated using software “blocres” with 20 voxel box size and 0.5 FSC cutoff. Density map is cut open along the 4-fold symmetry axis and colored according to local resolution.

2. Figure S2. Cryo-EM density map of selected regions of hERG_T, related to figure 2.

Density map was sharpened with different b-factors (−120 for N-terminus, −150 for PAS domain and S2–S3 linker, and −180 for other regions, using the “bfactor.exe” in the Frealign package) and low-pass filtered at 3.8 Å. Atomic model is represented as sticks with carbon atoms colored according to figure 1A (orange: PAS domain, magenta: voltage sensor domain, green: pore domain, red: C-linker domain, cyan: CNBHD).

3. Figure S3. Molecular surface around the central cavities, related to figure 5.

(A) Molecular surface of Slo1 (PDB ID 5TJ6) channel. (B) Molecular surface of Kir3.2 (PDB ID 3SYO) channel. Surfaces are colored according to electrostatic potential calculated using APBS tools (see STAR Methods).

4. Figure S4. Effects of S631A mutation on hERG_Ts, related to figure 6.

(A) Overlay of hERG_T (blue) and hERG_Ts (red) structures. For clarity, only PAS and transmembrane domains of two opposing subunits and CNBHD domains of the other two subunits are shown. (B) A representative whole-cell voltage family recording of hERG_Ts in HEK-293T cells. Voltage protocol is shown below panel D (see STAR Methods for details). (C) F627 side-chain positions in hERG_T (blue), hERG_Ts (red) and hERG_Ts S631A (green). (D) A representative whole-cell voltage family recording of hERG_Ts S631A in HEK-293T cells. Voltage protocol is shown below the current traces (see STAR Methods for details). (E) Normalized tail currents ((I – I_min)/(I_max – I_min), mean ± SEM, n = 5) of hERG_Ts (red circles) are plotted as a function of the depolarization step voltage and fitted with Boltzmann equation (red solid line, see STAR Methods). Fitted half activation voltage V_1/2 = −42 ± 2 mV and apparent valence z = 2.2 ± 0.2. For reference, Boltzmann fitted curves of hERG_WT and hERG_T from figure 1D are plotted as blue and black dashed lines, respectively.

5. Figure S5. Statistics of single particle cryo-EM data and atomic model refinement, related to figures 2 and 6.

(A) Statistics values. (B) – (D) FSC curves between atomic models and free half map (black lines), working half map (blue lines) and full map (red lines) for (B) hERG_T, (C)hERG_Ts and (D) hERG_Ts S631A. Dashed lines indicate 0.5 FSC value. Resolution values for the full map at 0.5 FSC are indicated behind arrows.

6. Figure S6. Cryo-EM density maps of selected regions of hERG_Ts and hERG_Ts S631A, related to figure 2 and 6.

Atomic models are shown as sticks with carbon atoms colored according to figure 1A (orange: PAS domain, magenta: voltage sensor domain, green: pore domain, red: C-linker domain, cyan: CNBHD). Density maps are shown as blue wire mesh. (A) hERG_Ts density map was sharpened with RELION post process using a b-factor of −140 and low-pass filtered at 3.7 Å. (B) hERG_Ts S631A density map was sharpened with different b-factors (−180 for N-terminus, S2 and S6; −150 for PAS domain, S2–S3 linker; and −200 for other regions, using the “bfactor.exe” in the Frealign package) and low-pass filtered at 4.0 Å.

Highlights.

• hERG channel structure is determined at 3.8 Å using single particle cryo-EM.

• hERG channel is open with the voltage sensors in a depolarized conformation.

• Unique properties of the central cavity likely contribute to hERG block by many drugs.

• A subtle structural rearrangement in hEGR’s selectivity filter correlates with rapid inactivation.