Structural basis for voltage-sensor trapping of the cardiac sodium channel by a deathstalker scorpion toxin

Voltage-gated sodium (NaV) channels initiate action potentials in excitable cells, and their function is altered by potent gating-modifier toxins. The α-toxin LqhIII from the deathstalker scorpion inhibits fast inactivation of cardiac NaV1.5 channels with IC50=11.4 nM. Here we reveal the structure of LqhIII bound to NaV1.5 at 3.3 Å resolution by cryo-EM. LqhIII anchors on top of voltage-sensing domain IV, wedged between the S1-S2 and S3-S4 linkers, which traps the gating charges of the S4 segment in a unique intermediate-activated state stabilized by four ion-pairs. This conformational change is propagated inward to weaken binding of the fast inactivation gate and favor opening the activation gate. However, these changes do not permit Na+ permeation, revealing why LqhIII slows inactivation of NaV channels but does not open them. Our results provide important insights into the structural basis for gating-modifier toxin binding, voltage-sensor trapping, and fast inactivation of NaV channels.


Introduction
Eukaryotic voltage-gated sodium (NaV) channels generate the inward sodium current that is responsible for initiating and propagating action potentials in nerve and muscle 1,2 . The sodium current is terminated within 1-2 milliseconds by fast inactivation 1,2 . A wide variety of neurotoxins bind to six distinct receptor sites on NaV channels and modify their function 3,4 . α-Scorpion toxins and sea anemone toxins bind to Neurotoxin Receptor Site 3, dramatically inhibit fast inactivation of NaV channels, and cause prolonged and/or repetitive action potentials [3][4][5] . Scorpions utilize these toxins in their venoms to immobilize prey by inducing paralysis and causing cardiac arrhythmia 4,6-8 . Because of their high affinity and specificity, scorpion toxins are used widely to study the structure and function of NaV channels. α-Scorpion toxins bind to the voltage sensor (VS) in domain IV (DIV), which is important for triggering fast inactivation [9][10][11][12][13] . Therefore, structures of the high-affinity complexes of α-scorpion toxins and NaV channels will provide critical information for understanding the structural basis for toxin binding, voltage-sensor trapping, and fast inactivation.
Eukaryotic NaV channels contain four homologous, nonidentical domains composed of six transmembrane segments (S1-S6), organized into a voltage-sensing module (VS, S1-S4) and a pore module (PM, S5-S6) with two intervening pore helices (P1 and P2) 14,15 . The S4 segments contain four to eight repeats of a positively charged residue (usually Arg) flanked by two hydrophobic residues. These positively charged residues serve as gating charges, moving outward upon depolarization to initiate the process of activation 14,15 . Chemical labeling and voltage clamp fluorometry suggest that DI-VS and DII-VS are primarily responsible for activation of the channel, whereas DIV-VS induces fast inactivation 14,15 . A triple hydrophobic motif, Ile-Phe-Met (IFM), located in the DIII-DIV linker, serves as the fast inactivation gate 14,15 . Mutation of the IFM motif can completely abolish fast inactivation 14,15 .
Determination of the structures of prokaryotic [16][17][18] and eukaryotic [19][20][21] NaV channels has remarkably enriched our understanding of their structure and function. Those structures revealed that NaV channels share similar key structural features 22 . The central pore is formed by the four PMs with the four VSs arranged in a pseudosymmetric square array on their periphery. The four homologous domains are organized in a domain-swapped manner, in which each VS interacts most closely with the PM of the neighboring domain. The four S6 segments come together at their intracellular ends to form the activation gate [16][17][18] . Intriguingly, in the structures of mammalian NaVs, the IFM motif binds in a receptor site formed by the DIII S4-S5 linker and the intracellular ends of the S5 and S6 segments in DIV, which suggests a local allosteric mechanism for fast inactivation of the pore by closing the intracellular activation gate [19][20][21] .
The α-scorpion toxins bind to DIV-VS in its resting state, trap it in an intermediate activated conformation, and inhibit fast inactivation, providing an attractive target for studying the coupling of DIV-VS to pore opening and fast inactivation [9][10][11][12][13]23,24 . Strong depolarization can reverse voltage-sensor trapping and drive the a-scorpion toxin off its receptor site, providing direct evidence for a toxin-induced conformation of the VS 9,24,25 . The cryo-EM structure of the αscorpion toxin AaHII was resolved bound to two different sites on a nonfunctional chimera of the cockroach sodium channel NaVPas, which contained 132 amino acid residues of the DIV-VS of the human neuronal sodium channel NaV1.7 embedded within 1449 residues of NaVPas 26 . These results revealed structures of AaHII bound to the voltage sensors in both DI and DIV but did not resolve whether AaHII bound to either of these sites was functionally active in the chimera 26 .
Therefore, the precise structural mechanism by which α-scorpion toxin binds to the DIV-VS in a native sodium channel and blocks fast inactivation still remains elusive.
LqhIII from the 'deathstalker scorpion' Leiurus quinquestriatus hebraeus (also known as the Israeli yellow scorpion and the North African striped scorpion) is classified as an α-scorpion toxin and shares the common βαββ scaffold containing four pairs of Cys residues that form disulfide bonds 7 . Most scorpion toxins paralyze prey by targeting the sodium channels in nerve and skeletal muscle specifically 7 . In contrast, LqhIII binds with highest affinity to the human cardiac sodium channel, with an estimated EC50 of 2.5 nM 27,28 . It prevents fast inactivation efficiently, and it dissociates at an extremely slow rate 27,28 , making it exceptionally potent.
In this work, we elucidate the molecular mechanisms of voltage-sensor trapping and block of fast inactivation by α-scorpion toxins in the context of a functional native toxin-receptor complex by determining the cryo-EM structure of rat cardiac sodium channel NaV1.5 in complex with the α-scorpion toxin LqhIII at 3.3 Å resolution. Our experiments provide important insights into the structural basis for gating-modifier toxin interaction, voltage-sensor trapping, electromechanical coupling in the VS, and fast inactivation of the pore.

Results
Voltage-sensor trapping of NaV1.5 by LqhIII. For our structural studies, we took advantage of the fully functional core construct of the rat cardiac sodium channel NaV1.5 (rNaV1.5C), which can be isolated with high yield and high stability 21 . Expression of rNaV1.5C in the human embryonic kidney cell line HEK293S GnTIand recording from single cells in whole-cell patch clamp mode (see Methods) yields inward sodium currents that activate rapidly and inactivate within 6 ms (Fig.   1a, black trace; inward current is plotted as a negative quantity by convention). Perfusion of increasing concentrations of LqhIII progressively slows the fast inactivation process and makes it incomplete (Fig. 1a, colored traces). We measured the sodium current remaining 6 ms after the depolarizing pulse as a metric of LqhIII toxin action (Fig. 1a, dotted line), because the unmodified sodium current has declined to nearly zero by this time, whereas substantial toxin-modified sodium current remains. The EC50 value for the increase in sodium current remaining at 6 ms following the stimulus is 11.4 nM (Fig. 1a). This effect of LqhIII and other a-scorpion toxins is achieved by trapping the voltage sensor in DIV of sodium channels in a conformation that allows sodium channel activation but prevents coupling to fast inactivation 4,9,10 . Voltage-sensor trapping develops slowly and progressively over more than 20 min, with a half-time of 11.3 min at 100 nM ( Fig. 1b). As expected from previous work 4,9,10 , strong depolarizing pulses to +100 mV cause dissociation of the toxin and loss of its blocking effect on fast inactivation (Fig. 1c). The molecular mechanism for this long-lasting voltage-dependent block of fast inactivation of NaV1.5 sodium currents by LqhIII is unknown.
Structure determination of rNaV1.5C/LqhIII complex by cryo-EM. We analyzed the structural basis for the potent voltage-sensor trapping effects of LqhIII by cryogenic electron microscopy (cryo-EM). LqhIII was incubated with purified rNaV1.5C for 30 min. The regulatory proteins FGF12b and calmodulin were added to stabilize the isolated protein, and the toxin/channel complex was further purified by size-exclusion chromatography (SEC). A symmetric peak of the toxin/channel complex was collected from the second SEC run ( Supplementary Fig. 1a, b). Detailed descriptions of protein expression, purification, cryo-EM imaging, and data processing are presented in Methods.
Cryo-EM data were collected on a Titan-Krios electron microscope and processed using RELION ( Supplementary Fig. 1c, d; Supplementary Fig. 2a-c). A 3D reconstruction map of the rNaV1.5C/LqhIII complex was obtained at an overall resolution of 3.3 Å, based on the Fourier Shell Correlation (FSC) between independently refined half-maps (Fig. 2a). Strong density specifically localized near the extracellular side of DIV-VS shows that there is only one LqhIII molecule bound to rNaV1.5C ( Fig. 2b; purple), as expected from previous biochemical studies of scorpion toxin binding to sodium channels 9 . The local resolution for the PM core region is ~3.0-3.5 Å, whereas the four peripheral VSs have local resolutions of ~3.5-4.0 Å (Fig. 2c). The resolution for the toxin is lower than the channel protein (~4.0-5.0 Å, Fig. 2c). However, the interacting surface of the toxin that binds to DIV-VS has a resolution of ~3.5-4.0 Å for the amino acid side chains that form the complex, as they are tightly bound ( Supplementary Fig. 2d and e). The 3D structure of the tightly disulfide-crosslinked toxin is well-known from previous studies ( Supplementary Fig. 3a) 29 , allowing it to be accurately fit into the observed density. No significant density was observed at high resolution for the C-terminal domain (CTD), FGF12b, or calmodulin (Fig. 2b), indicating that these components of the purified protein complex are mobile.
Overall structure and LqhIII binding site. The high-resolution cryo-EM density map allowed us to build an atomic model for the rNaV1.5C/LqhIII complex ( Fig. 3a and b). The overall structure of the rNaV1.5C/LqhIII complex is very similar to our previous apo-rNaV1.5C structure 21 , with a minimum RMSD of 0.78 Å over 1164 residues. However, local conformational differences give many important insights. The structure of LqhIII is rigidly locked by disulfide bonds, except for the β2β3 loop and C-terminal region, which are highly flexible in solution as revealed by NMR analyses (Fig. 3c). Remarkably, LqhIII uses these two flexible regions to bind to the extracellular side of DIV-VS by wedging its β2β3 loop and C-terminus into the aqueous cleft formed by the S1-S2 and S3-S4 helical hairpins (Fig. 3b). These features are in close agreement with previous molecular-mapping studies of neurotoxin receptor site 3 11 and with the structure of the AaHII/NaVPas-NaV1.7 chimera 26 (see Discussion). The toxin may attack Neurotoxin Receptor Site 3 in the DIV-VS using its most flexible regions to allow it to dock in a stepwise manner that results in a tight induced-fit complex.
The close interactions of the C-terminus and the β2β3 loop of LqhIII with rNaV1.5C are illustrated in Fig. 3b and d. At the C-terminus of LqhIII, Glu63 interacts with the Asn329-linked glycan from DI-PM, and Lys64 dips into the aqueous cleft and interacts with Gln1615 ( Fig. 3b and   d). The end of the β2β3 loop inserts into the DIV-VS cleft and partially unwinds the last helical turn of the S3 segment. Mutagenesis studies mapping Neurotoxin Receptor Site 3 revealed a negatively charged residue in the extracellular S3-S4 linker that is conserved among NaV channels and is critical for α-scorpion toxin binding 4,9 . In agreement with those studies, the conserved negatively charged residue Asp1612 at an equivalent position in NaV1.5 mediates this crucial interaction with the bound toxin. His43 and His15 wrap around Asp1612 like pincers forming a hydrogen bond (~2.5 Å) and a potential salt bridge (~4.0 Å), respectively (Fig. 3d).
Moreover, we note that the backbone carbonyl of His43 engages the backbone carbonyl of Thr1608 at a distance of 2.8-3.5 Å, which may contribute to the affinity or specificity of interactions with the β2β3 loop ( Fig. 3d) 30 .
The complementary interacting surfaces of LqhIII and Neurotoxin Receptor Site 3 are depicted in a space-filling model in Fig. 3e (left), and the functionally important interacting residues are highlighted in yellow with embedded sticks and displayed in an 'open-book' format in Fig. 3e (right). The interacting surface area of neurotoxin receptor site 3 covers ~ 836 Å 2 located on an arc stretching from the S3-S4 linker to the S1-S2 linker (Fig. 3e, right). The LqhIII toxin latches onto that arc, gripping it between the β2β3 loop and the C-terminus (Fig. 3e). It is likely that the flexibility of these regions of the toxin in solution is important for its initial approach and final tight grip on its target site.

An intermediate activated state of DIV-VS trapped by LqhIII. Fast inactivation of NaV channels
requires activation of DIV-VS 9,10,12,13 . Because there is no membrane potential during solubilization and purification, the VSs of published NaV structures are usually in partially or fully activated states. In our apo-rNaV1.5C structure, four of the six gating charges of DIV-VS pointed outward on the extracellular side of the hydrophobic constriction site (HCS), as expected for an activated state 21 . As a result, the fast inactivation gate in the apo-rNaV1.5C structure binds tightly in a hydrophobic pocket next to the activation gate 21 . α-Scorpion toxins bind to NaV channels in the resting state with higher affinity and trap the channel in a partially activated state, in which both the rate and extent of transition to the inactivated state are impaired ( Fig.1) 9,10 . Because of its high affinity and specificity, LqhIII is able bind to the purified rNaV1.5C protein in its activated state and chemically induce voltage-dependent structural changes to partially deactivate the VS.
Remarkably, LqhIII binding drives DIV-S4 approximately two helical turns inward to form an intermediate, partially activated structure ( Fig. 4a and b). Each gating charge Arg in the intermediate activated DIV-VS is positioned ~10-12 Å further inward than in the fully activated DIV-VS ( Fig. 4a and b). Importantly, in the toxin-bound intermediate activated state reported here, R1 to R4 adopt a 310-helix conformation, with the last helical turn of the S4 segment relaxing R5 into alpha-helical form. In contrast, in the fully activated state, the region between R2 to R6 is in 310-helical form, but R1 is alpha-helical. As a consequence of the 310-helix conformation from R1-R4 in the toxin/channel complex, the residues between R1-R2 and R3-R4 bridge the HCS such that R1-R2 and R3-R4 share the same vertical plane in their interactions with the negative side chains of the extracellular negative cluster (ENC) and intracellular negative cluster (INC), respectively. This unique linear voltage-sensor-trapped conformation would be strongly stabilized by these simultaneous gating charge interactions outside and inside the HCS, which may provide the chemical energy required for potent voltage-sensor trapping against the force of the transmembrane electrical field and therefore for effective modification of sodium channel gating.
The potential gating charges R5 and R6 translocate to the intracellular side of the VS completely.
These charged residues were proposed to interact with the CTD in the structure of the NaVPas/NaV1.7 chimera 26 . However, the CTD was not resolved in our structure, preventing visualization of the potential binding positions of R5 and R6.
Superposition of the fully activated state (grey) and toxin-induced intermediate activated state (blue) of the DIV-VS revealed a remarkable conformational difference (Fig. 4c). From S1 through most of S3 there is little or no structural change, whereas the final two helical turns of S3 and the entire S4 segment undergo dramatic conformational shifts. Notably, Gly1607 serves as a pivot point for S3 rotation, and the rotation of upper S3 in turn moves S4 downward ~11 Å, such that R1 and R2 in the intermediate activated state are approximately in the positions of R3 and R4 in the fully activated state (Fig. 4c). This toxin-induced conformational change in the S3-S4 linker is further documented by our fit to the cryo-EM density, which is illustrated in Extended Fig. 3c-e. At the intracellular end of S4, an elbow-like bend is formed between S4 and the S4-S5 linker, which pushes the S4-S5 linker ~4.6 Å inward at its N-terminal end (Fig. 4c). Intriguingly, our previous resting-state structure of NaVAb showed that a similar elbow pushes the S4-S5 linker and its connection to S4 strikingly inward and twists this segment in order to close the intracellular activation gate 31 . This conformational change in the S4-S5 linker is further supported by the close fit of our structural model to the cryo-EM density ( Supplementary Fig. 4a). Superposition of the intermediate activated DIV-VS structure (blue) upon the resting state NaVAb-VS structure (orange) further illuminates these conformational differences (Fig. 4d). The connecting S3-S4 loop of the intermediate activated state of the LqhIII/rNaV1.5C complex is not located as deeply inward as that of resting state of NaVAb and is not as tightly twisted (Fig. 4d). Moreover, the R1 and R2 gating charges are both located fully outward from the HCS in the partially activated S4 segment in the LqhIII/rNaV1.5C complex, whereas R1 is positioned only partially outward from the HCS in the resting state of NaVAb (Fig. 4d). In addition, the S4-S5 linker in the intermediate activated state has not moved as deeply into the cytosol as in the resting state (Fig. 4d). These differences suggest that the toxin-induced intermediate activated state of NaV1.5 VS is indeed an intermediate state between the resting state and the fully activated state.
A hallmark feature of the action of a-scorpion toxins is strongly voltage-dependent dissociation from their receptor site, which correlates with the voltage dependence of activation of sodium channels ( Fig. 1c; 9,24,25 ). The structure of the rNaV1.5C /LqhIII toxin complex reveals the molecular basis for this important aspect of scorpion toxin action. In the complex of the toxin with the partially activated state of the DIV VS, the positive charge of the e-amino group of K64 on LqhIII interacts with the same negatively charged side chain in the ENC that interacts with R1 and R2 in the activated conformation of the VS ( Fig. 4a and b). In light of these structures, it seems likely that outward movement of the S4 segment during activation of the DIV VS creates a clash with K64  Fig. 4b and c). Considering that the two proteins were purified following the same procedures, it is likely that the same set of lipid and detergent molecules would be available for binding. Therefore, we believe the larger lipid or detergent molecule is able to bind in the lumen of the activation gate of the intermediate activated structure because of its larger diameter rather than because of a change in lipid or detergent concentrations between the two protein preparations. does not appear to be open enough to conduct hydrated Na + 32,33 . To test this hypothesis, we used molecular dynamics methods similar to those we previously applied to the NaVAb structure 32,33 in order to investigate the effect of LqhIII on pore hydration and dilation of the intracellular activation gate (Fig. 6). The inner pore of rNaV1.5C is depicted lying from right (extracellular) to left (intracellular) with the surrounding S5 and S6 helices illustrated in orange (Fig. 6a). Water molecules (red) fill the inner part of the central cavity on the right and the intracellular exit from the pore on the left. However, in this snapshot, there is a gap in hydration in the intracellular activation gate itself (white), where the S6 segments come together in a bundle (orange helices, Fig. 6a). In fact, statistical analysis of the conformational ensemble shows that the average probability density of water molecules in the intracellular activation gate (purple band) is near zero in simulations of both rNaV1.5C (black) and rNaV1.5C/LqhIII (red; Fig. 6b). Accordingly, Na + did not permeate through the dehydrated activation gate in any of the simulations, suggesting that the pore is functionally closed. Not only is the activation gate the least hydrated region of the pore on average, but it is also nearly always dehydrated ( Fig. 6a and b). Even when a pathway connecting the central cavity to the intracellular space is transiently present, water molecules are usually excluded from entering this region due to the hydrophobic effect (Fig. 6a). As such, the activation gate is predominantly dehydrated (dewetted) and occupied by 4 water molecules on average,  Fig. 6c and d). However, the activation gate is not significantly more likely to be wetted in simulations of rNaV1.5C with LqhIII than in simulations without LqhIII ( Supplementary Fig. 5b), consistent with the fact that toxin binding does not open the gate sufficiently for passage of Na + . Nevertheless, the analysis of fluctuations in diameter and hydration of the intracellular activation gate provide an initial suggestion that LqhIII binding may facilitate the transition to the open state of NaV1.5, which requires activation of the voltage sensors in domains I, II, and III for completion.

Discussion
We determined the structure of rat NaV1.5C in complex with the α-scorpion toxin LqhIII by single particle cryo-EM. Biochemical and biophysical studies support only a single neurotoxin receptor site 3 per sodium channel located in the VS in DIV, at which α-scorpion toxins, sea anemone toxins, and related gating-modifier toxins bind 9,10 . Consistent with this expectation from functional studies, we found a single molecule of LqhIII bound to the VS in DIV. The toxin binds at the extracellular end of the aqueous cleft formed by the S1-S2 and S3-S4 helical hairpins in the VS through its b2b3 loop and its C-terminal. Many conserved amino acid residues that are important for α-scorpion toxin binding and its functional effects on sodium channels are located in key positions in the toxin-receptor binding interface (Fig. 3e). These results provide convincing evidence that we have correctly identified the pharmacologically important Neurotoxin Receptor Our complex structure provides an excellent model for investigating the coupling between gating charge transfer and fast inactivation. The cryo-EM structure of AaHII/NaVPas-NaV1.7-DIV-VS chimera suggested that R5 and K6 of NaV1.7-DIV-VS were stabilized by interaction with the NaVPas CTD. By contrast, in our fully functional LqhIII/rNaV1.5C structure, the CTD was not observed. In fact, the CTD's also were not observed in the high-resolution structures of human NaV1.2, 1.4, and 1.7 channels either [19][20][21]40 . These results suggest that the CTD's of native mammalian NaV channels are disordered and/or mobile and differ substantially from the cockroach CTD in the NaVPas-NaV1.7 chimera, whose amino acid sequence is not similar to the CTD's of mammalian NaV channels. Based on this comparison, it seems likely that the CTD plays a secondary role or a regulatory role in fast inactivation in mammalian NaV channels, which may In previous work, the structure of a chimera of the cockroach sodium channel NaVPas with the AaHII toxin bound was determined by cryo-EM 26 . The functional significance of this sodium channel in the cockroach is unknown, and this chimera containing a segment of the DIV-VS of human NaV1.7 was nonfunctional; therefore, it is difficult to precisely compare this prior work to the structures we present here. Unexpectedly, AaHII bound to the NaVPas chimera in two positions, one on the VS in DI of NaVPas and one on the DIV-VS contributed in part by NaV1.7, and it was not shown whether either of these sites was functional in the chimera 26 . In contrast, we found only a single toxin binding site, as expected from previous structure-function studies 9,10 .
Neurotoxin Receptor Site 3 identified in our study is generally similar to the AaHII binding site found in DIV of the AaHII/NaVPas-NaV1.7-DIV-VS chimera structure 26 , but we found an important difference in the binding poses of the two toxins. Compared with AaHII bound to the NaVPas-NaV1.7-DIV-VS chimera, LqhIII bound to rNaV1.5C is rotated ~26 o downward, further away from the glycan and DI of the channel (Supplementary Fig. 7). This difference may reflect alteration in the position of the receptor site within the NaVPas-NaV1.7-DIV-VS chimera tertiary structure caused by artifactual constraints from formation of the chimera 26 . Alternatively, the structure of the functionally active LqhIII/rNaV1.5C complex described here may be characteristic of the cardiac sodium channel, which has numerous distinct features compared to neuronal sodium channels like NaV1. 7 Thus, the toxin-bound state we have characterized here may have broad significance for voltage sensor trapping by a wide range of gating-modifier toxins from hundreds of species of spiders, scorpions, mollusks, and coelenterates, which all use this universal mechanism to immobilize their prey.

Methods
Electrophysiology. All experiments were performed at room temperature (21-24 °C) as described previously 21 . Human HEK293S GnTIcells were maintained and infected on cell culture plates in Dulbecco's Modified Eagle Medium (DMEM) supplemented with 10% FBS and glutamine/penicillin/streptomycin at 37°C and 5% CO2 for electrophysiology. Unless otherwise mentioned, HEK293S GnTIcells were held at -120 mV and 100-ms pulses were applied in 10 mV increments from -120 mV to +60 mV. A P/-4 holding leak potential was set to -120 mV. CryoEM grid preparation and data collection. Three microliters of purified sample were applied to glow-discharged holey gold grids (UltraAuFoil, 300 mesh, R1.2/1.3), and blotted for 2.0 -3.5 s at 100% humidity and 4 °C before being plunged frozen in liquid ethane cooled by liquid nitrogen using a FEI Mark IV Vitrobot. All data were acquired using a Titan Krios transmission electron microscope operated at 300 kV, a Gatan K2 Summit direct detector and Gatan Quantum GIF energy filter with a slit width of 20 eV. A total of 4,222 movie stacks were automatically collected using Leginon 49 at a nominal magnification of 130,000x with a pixel size of 0.528 Å (superresolution mode). Defocus range was set between -1.2 and -2.8 μm. The dose rate was adjusted to 8 counts/pixel/s, and each stack was exposed for 8.4 s with 42 frames with a total dose of 60 e -/ Å 2 .
Cryo-EM data processing. The movie stacks were motion-corrected with MotionCorr2 50 , binned 2-fold, and dose-weighted, yielding a pixel size of 1.056 Å. Defocus values of each aligned sum were estimated with Gctf 51 . A total of 3,805 micrographs with CTF fitted better than 6 Å were used for particle picking, and a total of 1,817,940 particles were automatically picked in RELION3.0 52 .
After several rounds of 2D classification, 882,608 good particles were selected and subjected to one class global angular search 3D classification with an angular search step at 7.5°, at which low-pass filtered cryo-EM map of rNaV1.5C was used as an initial model. Each of the last five iterations was further subjected to four classes of local angular search and 3D classification with an angular search step at 3.75°. After combining particles from the best 3D classes and removing duplicate particles, 570,843 particles were subjected to per-particle CTF estimation by GCTF followed by Bayesian polishing. The polished particles were subjected to last round three-class multi-reference 3D classification. The best class containing 267,595 particles was auto-refined and sharpened in Relion3.0. Local resolution was estimated by ResMap in Relion3.0. A diagram illustrating our data processing is presented in Supplementary Fig. 2.
Model building and refinement. The structures of rat rNaV1.5C (PDB code: 6UZ0) and LqhIII (PDB code: 1FH3) were fitted into the cryo-EM density map in Chimera 53 . The model was manually rebuilt in COOT 54 and subsequently refined in Phenix 55 . The model vs map FSC curve was calculated by Phenix.mtrage. Statistics for cryo-EM data collection and model refinement are summarized in Supplementary Table 1.

Molecular dynamics model. The cryo-EM structure of rNaV1.5C/LqhIII lacking DI-DII and DII-DIII
linkers is composed of three chains which correspond to DI, DII, and DIII-DIV. The MODELLER software (ver. 9.22) was used to insert missing residues and sidechains within the polypeptide chains, followed by quick refinement using MD with simulated annealing 56 . Neutral N-and Ctermini were used for the three polypeptide chains in our refined model of rNaV1.5C. N-termini from chains DII and DIII-IV were acetylated, and a neutral amino terminus (-NH2) was used for DI. Neutral carboxyl groups (-COOH) were used for all C-termini. Disulfide bonds linking residues 327-342, 909-918, and 1730-1744 were included in our model of the channel as they were present in the cryo-EM structure; however, no glycans were added to the protein. Charged N-and Ctermini were used for LqhIII and disulfide bonds linking residues 12-65, 16-37, 23-47, and 27-49 were included. Molecular dynamics simulations. Molecular models of rNaV1.5C/LqhIII with and without the toxin were prepared using the input generator, Membrane Builder 57-61 , from CHARMM-GUI 59 . The rNaV1.5C/LqhIII model was embedded in a hydrated DMPC bilayer, with approximately 150 mM NaCl. The protein was translated and rotated for membrane embedding using the PPM server 62 .
The lipid bilayer was assembled using the replacement method, and solvent ions were added at random positions using a distance-based algorithm. A periodic rectangular cell with approximate dimensions of 14x14x13 nm was used, which comprised ~240,000 atoms.
The CHARMM36 all-atom force field [63][64][65] was used in conjunction with the TIP3P water model 66 . Non-bonded fixes for backbone carbonyl oxygen atoms with Na + 67 , and lipid head groups with Na + 68 were imposed. Electrostatic interactions were calculated using the particle-mesh Ewald algorithm 69,70 and chemical bonds were constrained using the LINCS algorithm 71 .
The energy of the system was minimized with protein position restraints on the backbone (4000 kJ/mol/nm 2 ) and side chains (2000 kJ/mol/nm 2 ), as well as lipid position and dihedral restraints (1000 kJ/mol/nm 2 ) using 5000 steps of steepest descent. The simulation systems were pre-equilibrated using multi-step isothermal-isovolumetric (NVT) and isothermal-isobaric (NPT) conditions for a total of 10.35 ns (see Table MDS1 for parameters). Unrestrained "production" simulations of approximately 300 ns were then generated with a 2 fs time integration step. The first 100 ns of all production simulations were considered part of equilibration based on RMSD analyses of Cα atoms (Supplementary Fig. 6) and were excluded from subsequent data analysis.
Thirty independent replicas (ten of them 400 ns-long and twenty of them 300 ns-long) were generated for each system using random starting velocities, yielding a total simulation time of After performing the spatial transformations, the z-axis of the simulation box was used as the pore axis of NaV1.5 and the transformed positions were used for subsequent analyses.
The axial distribution of water was computed by counting the number of water O-atoms within a cylindrical of radius 8.5 Å centered on the pore axis. The probability distribution of water was calculated for each replica by counting the number of water molecules in uniform cylindrical slices along the pore-axis and normalizing the counts by the slice with the highest number of water molecules (solvent slice). The average and SEM of the probability distribution was computed across replicas.
Pore hydration analysis indicated a dehydrated region located at the ICAG (-2.8 nm < z < -1.5 nm). The number of water molecules in the gate was counted for each frame and normalized by the total number of frames to obtain the probability distribution. The average and s.e.m. were computed across replicas.
To measure the size of the intracellular activation gate, residues at the ends of the S6 helices        MD parameters used for each pre-equilibration and production step. The ensemble (ENS), time integration step, and total time/replica are indicated. Berendsen (B) (1) or Nosé-Hoover (NH) (2, 3) thermostats and B or Parrinello-Rahman (PR) (4, 5) barostats with their respective coupling times constants (CTC) are shown. Protein backbone and sidechain position restraints as well as lipid position and dihedral restraints are also specified.