Structure-guided unlocking of NaX reveals a non-selective tetrodotoxin-sensitive cation channel

Unlike classical voltage-gated sodium (NaV) channels, NaX has been characterized as a voltage-insensitive, tetrodotoxin-resistant, sodium (Na+)-activated channel involved in regulating Na+ homeostasis. However, NaX remains refractory to functional characterization in traditional heterologous systems. Here, to gain insight into its atypical physiology, we determine structures of the human NaX channel in complex with the auxiliary β3-subunit. NaX reveals structural alterations within the selectivity filter, voltage sensor-like domains, and pore module. We do not identify an extracellular Na+-sensor or any evidence for a Na+-based activation mechanism in NaX. Instead, the S6-gate remains closed, membrane lipids fill the central cavity, and the domain III-IV linker restricts S6-dilation. We use protein engineering to identify three pore-wetting mutations targeting the hydrophobic S6-gate that unlock a robust voltage-insensitive leak conductance. This constitutively active NaX-QTT channel construct is non-selective among monovalent cations, inhibited by extracellular calcium, and sensitive to classical NaV channel blockers, including tetrodotoxin. Our findings highlight a functional diversity across the NaV channel scaffold, reshape our understanding of NaX physiology, and provide a template to demystify recalcitrant ion channels.

M ammalian voltage-gated sodium (Na V ) channel subtypes, Na V 1.1-Na V 1.9, perform essential roles in electrical signaling by initiating and propagating action potentials 1 . A tenth Na V channel-like gene (SCN7A) was cloned three decades ago, but atypical sequence features and the inability to record voltage-activated sodium (Na + )-currents raised speculation that this channel might have distinct physiological roles [2][3][4][5] . Evidence suggests that Na X may contribute to Na + homeostasis [5][6][7][8][9][10] and Na X shows restricted expression in a brain area that specializes in monitoring blood composition [11][12][13] . Na Xknockout mice ingest salt despite dehydration and are resistant to hypertension caused by elevated Na + levels 11,14 . Autoimmunity against Na X is associated with chronic hypernatremia with impaired thirst perception and salt appetite in humans 9 . In the periphery, Na X regulates Na + homeostasis upstream of the epithelial Na + channel (ENaC) and may be a target to treat atopic dermatitis and hypertrophic scarring 10,15 .
Clues to Na X function derive mainly from in vivo studies and cells obtained from knockout mice 5,16 . Murine Na X is reportedly activated by extracellular Na + concentrations above 150 mM 5,17,18 , where Na + -sensitivity can be shifted into the physiological range (135-145 mM) through a mechanism involving the endothelin receptor 17 . Na + influx through Na X may activate the Na + /K + -ATPase 19,20 leading to proton release and the activation of acidsensing channels on nearby neurons 14,16 . However, because all attempts to record currents from traditional heterologous systems have failed 3,4 , the basic biophysical properties of Na X and the molecular determinants involved in Na + -sensing remain unknown. Na X is the most sequence divergent member of the mammalian Na V channel family 1 , which has made structure-function correlations difficult to discern. Because murine Na X contains a reduced number of S4-gating charges and is voltageinsensitive 2-4 , the structure and function of the voltage-sensorlike domains (VSLDs) are unclear. Instead of the characteristic ion-selectivity filter DEKA-locus found in Na V channels, Na X contains a unique DENA-locus [2][3][4] and is thought to be tetrodotoxin (TTX)-resistant 5,12 . Na X is reported to generate noninactivating currents and has a divergent intracellular DIII-DIV linker sequence 5,18 . In this work, we determined the structure of the human Na X channel, which has provided new insights into its unusual biophysical properties and physiology.

Results
Evaluation of human Na X function. Reliable ionic currents were not detected for human Na X expressed in HEK293T cells or Xenopus laevis oocytes in response to changes in membrane voltage ( Fig. 1a-d, Supplementary Fig. 1a-d). Application of Na V channel pharmacological activators like aconitine or veratridine did not invoke any clear Na X -mediated currents in oocytes (Fig. 1b, d, Supplementary Fig. 1g, h). Co-expression with Na V or voltage-gated calcium (Ca V ) channel auxiliary-subunits also failed to produce dependable currents (Fig. 1c, d, Supplementary  Fig. 1i, j). Thus, as reported for the rodent proteins 3,4 , human Na X does not operate as a conventional voltage-gated channel.
In a neuroblastoma (Neuro-2a) cell line expressing murine Na X , increases in the extracellular Na + concentration ([Na + ] e ) reportedly produce a non-inactivating Na + current 17,18 . When human Na X was expressed in Neuro-2a cells, small and flickery inward currents were observed upon changes in [Na + ] e from 140 to 190 mM, but cells expressing the human Na V 1.7 channel showed similar behavior (Fig. 1e, g, Supplementary Fig. 1e). A non-inactivating inward current was only observed when seal resistances were low (<1 GΩ) in both human Na X -and mocktransfected Neuro-2a cells (Supplementary Fig. 1f). Because sequence differences ( Supplementary Fig. 2) or cellular context 21 may account for these seemingly conflicting functional outcomes between murine and human Na X , we next considered if structure determination might provide insight into the function or regulation of Na X .
Structure of the β3-Na X channel complex in lipid nanodiscs. To facilitate cryogenic electron microscopy (cryo-EM) analysis, human Na X was co-expressed with the canonical auxiliary β3subunit, purified in mild detergent (glyco-diosgenin), and reconstituted into phospholipid nanodiscs ( Supplementary  Fig. 3a-c). Prior to sample vitrification, the Na + concentration was raised above the threshold reported to activate Na X (to 200 mM), and the resulting cryo-EM reconstruction of the β3-Na X complex extended to 3.2 Å resolution (Supplementary Fig. 4a-g, Supplementary Table 1).
The Na X channel resembles a four-leaf clover with VSLDs arranged in a domain-swapped organization (Fig. 1h). The β3subunit is wedged between extracellular loops of the pore module and its single transmembrane helix is positioned against VSLD3 (Fig. 1h, Supplementary Fig. 5a). The intracellular DIII-DIV linker is bound alongside the pore (see below), and the carboxylterminal domain is not well-resolved ( Supplementary Fig. 5a), establishing that the overall architecture of Na X is highly similar to available human Na V channel structures determined in detergent ( Supplementary Fig. 5b-e) [22][23][24][25][26] . However, Na X does reveal structural deviations in regions implicated in Na V channel gating ( Supplementary Fig. 5d, e), suggesting that these changes may be a consequence of sequence and functional divergence ( Fig. 1a, b, Supplementary Fig. 2).
The Na X pore module is in a nonconductive state. The Na X pore module contains an outer vestibule, an ion-selectivity filter, a central cavity, and an S6-gate (Fig. 2a). Despite the high Na + concentration in our sample, no distinguishing extracellular Na + binding site was observed ( Supplementary Fig. 5f), and residues lining the narrow S6-gate form a hydrophobic barrier to the passage of hydrated ions, defining a nonconductive state (Fig. 2b, Supplementary Fig. 5c) [22][23][24][25][26] . These structural observations align with our physiological results ( Fig. 1a-f) and overturn expectations that human Na X alone gives rise to a Na + -activated, noninactivating conductance 17,18 .
The Na X central cavity is similar to those in Na V channels, although the DIV-S6 phenylalanine targeted by pore-blocking drugs 26,27 is replaced by DIV-Trp1484 ( Supplementary Fig. 5g). Na X reveals four lateral fenestrations that penetrate the pore, suggesting that lipids or Na V and Ca V channel antagonists might target the central cavity through membrane access pathways (Fig. 2c, d).
The Na X pore is infiltrated by membrane lipids. Four lipids penetrate the lateral pore fenestrations in Na X to enter the central cavity (Fig. 2c, d). These features can unambiguously be assigned as three phospholipids and one cholesterol, where one phospholipid even straddles and seals the intracellular S6-gate (Fig. 2c, d, Supplementary Fig. 5h). To evaluate if nanodisc reconstitution was responsible for lipid visualization, we prepared a β3-Na X sample in the detergent glyco-diosgenin without exogenous phospholipid supplementation and determined its structure in 200 mM Na + to 2.9 Å resolution (Supplementary Fig. 6a-g, Supplementary Table 1). An essentially identical pore structure penetrated by four well-defined lipids was observed ( Supplementary Fig. 6h). The presence of well-resolved, co-purifying lipids in Na X may reflect the hydrophobicity of the central cavity and S6-gate or the intrinsic stability of the nonconductive state.
The intracellular DIII-DIV linker restricts S6-gate dilation. The Na X DIII-DIV linker is bound to a pore module receptor site reminiscent of the fast-inactivated state in Na V channels (Fig. 2b, Supplementary Fig. 7a, b) [22][23][24][25][26] . Mutating the Na X IFI-motif to a Na V channel non-inactivating QQQ-motif 28 was insufficient to generate currents in oocytes ( Supplementary Fig. 7c, d), but disrupting the VSLD3 and DIII S4-S5 linker regions required to form the fast-inactivation receptor site in Na V channels produced robust leak currents in Na V 1.7-Na X chimeric channels (Supplementary Fig. 7e-g) 29 . The DIII-DIV linker therefore appears to restrict S6-dilation in Na X and may stabilize the nonconductive state.   Fig. 1 Characterization of and overall structure of human Na X . a Representative currents from Xenopus laevis oocytes expressing human Na X or Na V 1.7. Na X : steps between +80 to −100 mV, in 20 mV increments from a HP of 0 mV; Na V 1.7: depolarizing steps between −80 and +65 mV, in 5 mV increments, from a HP of −100 mV. b Representative currents from oocytes expressing Na X in response to extracellular application of indicated compounds (BDS-I Blood depressing substance I, ATX-II Neurotoxin 2). Voltage protocols as above. See Methods for concentrations of compounds tested. c Representative currents from oocytes expressing Na X and co-expression of Na V or Ca V channel auxiliary-subunits, Na + /K + -ATPase subunits and synapse-associated protein 97 (SAP97), and in the presence of the indicated extracellular Na + concentration. Voltage protocols as above. d Data summary of independent experiments performed as in parts a-c. Data are shown as mean ± SD; ns not significant; ****p < 0.0001; one-way ANOVA with Dunnett's test (against Na X ). Exact p-values and statistical parameters are provided in Source Data. Numbers of biological replicates (n) are indicated. e Representative currents from murine Neuro-2a cells expressing human Na X or Na V 1.7 in response to changes of the extracellular Na + concentration (HP = −60 mV), as indicated, or voltage: depolarizing steps between −60 to +60 mV, in 10 mV increments from a HP of −100 mV. f Data summary of independent experiments performed as in parts e. Data are shown as mean ± SD; ns not significant; *p < 0.05; **p < 0.01; ****p < 0.0001; one-way ANOVA with Dunnett's test (against mock-transfected cells). Exact p-values and statistical parameters are provided in Source Data. Numbers of biological replicates (n) are indicated. g Western blots of total lysate and surface fraction of proteins extracted from Neuro-2a cells probed for the indicated proteins. Data represent three independent biological replicates. h Side and extracellular view of the β3-Na X channel complex. Approximate membrane boundaries are indicated. DI, DII, DIII, and DIV are colored in green, blue, orange, and pink, respectively, with the β3-subunit in gray surface representation.
Na X -QTT Na X ns ns **** **** (9) (9) (9) (8) (7) (7) **** **** Fig. 2 Na X pore module structure reveals a nonconductive state. a Na X pore volume shown as gray surface with DII and DIV in cartoon rendering (DI and DIII omitted for clarity). b View of the S6-helices with side-chains lining the activation gate shown. Orthogonal view provides a wider perspective with DIII and DIV colored orange and pink, and the IFI-motif (green) from the DIII-DIV linker shown in stick and semi-transparent surface representation. c Orthogonal views sliced through the pore module highlighting lateral fenestrations and bound lipids. The phosphatidylethanolamine that crosses the S6-gate is in purple stick representation. d Similar to middle panel c, but with cryo-EM map shown in blue mesh representation. e Location of S6-gate hydrophobic side-chains targeted by pore-wetting mutations. f Representative currents from Xenopus laevis oocytes expressing the Na X -QTT construct with voltage protocol indicated. g Data summary of independent experiments with indicated constructs (see Supplementary Fig. 8). Data are shown as mean ± SD; ns not significant; ****p < 0.0001; one-way ANOVA with Dunnett's test (against Na X -QTT + H 2 O). Exact p-values and statistical parameters are provided in Source Data. Numbers of biological replicates (n) are indicated.
Na X -QTT currents are outward-rectifying without signs of inactivation ( Fig. 2f, Supplementary Fig. 8b). Current amplitudes or kinetics were not impacted when Na X -QTT was co-expressed with the β1or β3-subunits (Fig. 2g, Supplementary Fig. 8e). Na X -QTT also behaved as a non-inactivating channel in patch-clamp experiments in HEK293T cells (see below), where minor differences in the kinetics of the outward currents are likely due to species differences between the expression systems, and the lack of control over intracellular ionic composition in Xenopus laevis oocytes. Because Na X -QTT contains only three mutations targeted to the S6gate (Fig. 2e), we characterized this construct as a proxy to evaluate ion selectivity and pharmacology of the human Na X channel.
The Na X selectivity filter. The unique DENA-locus of Na X is supported by a local architecture that differs radically from the scaffold which supports the DEKA-locus in Na V channels ( Fig. 3a-d). Consequently, the ion-selectivity filter of Na X is wider and more electronegative (Fig. 3d). At the DENA-locus, Asp353 (DI) bonds to the backbone of DII-Trp689, Glu688 (DII) points into the ion permeation pathway, and Asn1142 (DIII) bonds to the DIV-Phe1433 carbonyl to impose a~1.5 Å radial shift onto Ala1434 (DIV) relative to the DIV alanine in Na V channels (Fig. 3c). Displacement of Ala1434 is accommodated by the distinct DI-Pro355 in Na X , which permits DI-Tyr359 to fill the volume vacated by the traditional DI tryptophan (Fig. 3b, c). This remodeling only slightly repositions side-chains within the Na X outer vestibule, suggesting that ion or toxin binding above the selectivity filter may not be substantially impacted ( Supplementary Fig. 9a, b). Na X -QTT is non-selective among monovalent cations. The DIII-Asn1142 dramatically alters the structure and chemical profile of the Na X selectivity filter relative to the DIII-lysine that  Fig. 3 Structure and characterization of the Na X selectivity filter. a Na X DENA-motif side-chains shown as sticks. b Residues in Na X that form a conserved interaction network around the selectivity filter in Na V channels (Na V 1.7, PDB 6J8J). c Superimposed view of the DI-DIV interface comparing Na X and Na V 1.7 (gray, PDB 6J8J). d View as in b, Na X and Na V 1.7 selectivity filter electrostatic surface rending. Note, central cavity and activation gate excluded for clarity. e Representative currents from HEK293T cells expressing human Na X -QTT with a C-terminal GFP-Flag tag in a physiological (left) or NMDG + -only extracellular solution (middle). See methods for composition of intracellular (IC) and extracellular (EC) solutions. Steps between +80 to −100 mV, in 20 mV increments, from a HP of 0 mV. Right, shows I-V curve data summary from n = 6 cells over two independent experiments. Data are shown as mean ± SD. Numbers of biological replicates (n) are indicated. f Representative currents from HEK293T cells expressing human Na X -QTT with indicated monovalent cations in the extracellular solution. Voltage ramp from −80 to +80 mV was applied. Right, summary of reversal potentials and permeability ratios measured from three independent experiments. Data are shown as mean ± SD. Numbers of biological replicates (n) are indicated. supports Na + -selectivity in Na V channels (Fig. 3d) 32 . Transferring a DEKA-locus into Na X produced no measurable currents ( Supplementary Fig. 9c), establishing that features beyond the selectivity filter contribute to channel activation. When cations in the extracellular solution were replaced with N-methyl-D-glucamine, the reversal potential for Na X -QTT in HEK293T cells shifted from +5 mV to below −100 mV (Fig. 3e), indicating that this construct behaves as a cation channel, and that large cations do not permeate. Replacement of Cl − ions with the large anion methansulfonate did not change the current profile ( Supplementary Fig. 9d), establishing that Na X -QTT behaves as a cation-selective channel. Using a ramp protocol with a simple Na + intracellular solution (150 mM), the reversal potential was close to 0 mV under symmetrical NaCl conditions, and only slight changes in the reversal potentials were measured upon exposure to different extracellular monovalent cations (150 mM; Fig. 3f). These results demonstrate that the Na X -QTT selectivity filter does not effectively discriminate between Na + , K + , Cs + , or Li + ions. Na X -QTT is inhibited by Ca 2+ and sensitive to Na V channel blockers. Ca 2+ does not permeate through Na X -QTT, but 1 mM extracellular Ca 2+ inhibited~70% of the Na + current (Fig. 4a). Zinc (Zn 2+ ), cobalt (Co 2+ ), and gadolinium (Gd 3+ ) cations also inhibited Na X -QTT currents (Fig. 4b). Contrasting long-standing reports 5,12 , tetrodotoxin showed inhibitory effects on Na X -QTT, especially on the inward current (Fig. 4c). The high potency of TTX-block suggests that it occurs at the selectivity filter, where most side-chains implicated in TTX-block in Na V channels are structurally conserved (Supplementary Fig. 2, 9b) 24 . Lidocaine, quinidine, and loperamide blocked Na X -QTT currents in the µM to mM concentration range; however, strong inhibitory effects were not observed by all classical Na V channel pore-blocking drugs tested (Fig. 4c).
Na X reveals atypical voltage-sensor-like domains. Conditions under which human Na X channel might produce voltageactivated currents have not been identified (Fig. 1a-f), although the VSLDs share a high structural similarity with the voltagesensor domains (VSDs) of Na V and Ca V channels (Supplementary Fig. 10a). Numerous features are implicated in noncanonical contributions to Na X channel function, for instance, VSLD4 is found in an unusual deactivated-like state ( Supplementary  Fig. 10a, b).

Discussion
The human Na X channel does not function as a conventional voltage-activated channel, but we find that the transfer of entire homologous domains or subdomains between Na X and Na V 1.7 can produce channels with unusual gating characteristics (Supplementary Fig. 7d, e, 10c). These results suggest that the coupling interfaces or energetics of Na X gating have diverged substantially from Na V channels, underscoring the functional diversity that has evolved across this channel scaffold 33 . This does not exclude the possibility that Na X may function as a voltage-modulated channel, since binding of one or more putative auxiliary proteins may impart voltage sensitivity as described for the NALCN and TMEM16A channels 33,34 .
Murine Na X has been reported to produce non-inactivating inward currents in Neuro-2a cells upon increasing the extracellular Na + concentration >150 mM 5,17,18 . We only observed qualitatively similar recordings with low seal resistance, and failed to reconstitute Na + -activated currents in cells expressing human Na X (Fig. 1d, Supplementary Fig. 1f). Moreover, our β3-Na X cryo-EM samples were prepared with NaCl concentrations above the reported threshold required for Na + -dependent gating 5,16,17 , but no clear Na + -sensing locus or Na + -based activation mechanism was observed ( Fig. 2a, b, Supplementary Fig. 5f). The only map feature that we might assign as a Na + ion is bound near the Na X selectivity filter at an extracellular site that is conserved in Na V channels ( Supplementary Fig. 9a) 23 . We conclude that Na X lacks an intrinsic Na + -sensor, and speculate that the described Na + -sensing mechanism must be contributed by an exogenous factor or pathway, such as the Na + /K + ATPase, the endothelin receptor, a yet to be identified system, auxiliary protein(s), or post-translational modification 17,19,20 .
Inspired by the concept of hydrophobic gating in channels 30,31 , mutation of only three residues around the S6-gate region was sufficient to produce robust ionic currents from Na X -QTT in Xenopus oocytes and HEK293T cells (Figs. 2e, f, 3e, f, 4a-c). Conceptually, the introduced polar side-chains help to hydrate the hydrophobic S6-gate, dislodge bound membrane lipids, and displace the IFI-motif to promote pore dilation and ion conduction ( Fig. 2a-g, Supplementary Fig. 7a) 28 . Na X -QTT currents were non-inactivating and not modulated by voltage, and while Na X has been proposed to function as a leak channel in vivo 13 , future studies are required to define the biophysical properties of wild-type channels.
In the absence of extracellular Ca 2+ ions, Na X -QTT gives rise to a near-Ohmic leak current that does not discriminate between the tested monovalent cations (Fig. 3f). This finding challenges prior suggestions that Na X is Na + -selective 18 and confirms a key role for the DIII-lysine in Na V channels 32 . Extracellular Ca 2+ block of Na X -QTT produces outward-rectifying currents with only a small inward leak component at negative membrane potentials (Fig. 4a). Na X may therefore function as a Na + -leak channel sensitive to modulation by extracellular Ca 2+ . Ca 2+ block of Na X -QTT is~10-fold more potent than a block on Na V channels (Fig. 4a) 35 , but similar to the Ca 2+ sensitivity of the distantly related NALCN Na + -leak channel 33 , perhaps suggesting a shared mechanism to avoid excessive depolarizing Na + influx into cells. Na X -QTT has pharmacological sensitivities expected for a Na V -like channel, including a block by TTX, lidocaine, quinidine, and loperamide (Fig. 4b, c) 1 . However, flecainide, ranolazine, or phenytoin did not produce strong inhibition (Fig. 4c), possibly reflecting unique central cavity-lining residues in Na X (Supplementary Fig. 5g) or the absence of an inactivated state.
Lipids occlude the ion conduction pathway in our Na X channel structures (Fig. 2c, d, Supplementary Fig. 6h). The phospholipid which plugs the S6-gate is reminiscent of cholesterol-like molecules bound in Na V channels 22,23 , where the occupancy of these hydrophobic ligands appears to correlate with the upconformation of a conserved DIV-S6-tyrosine side-chain implicated in drug block ( Supplementary Fig. 6h, 10d) 24,26 . Drug binding can enforce a down-conformation of this S6-tyrosine, concurrent with the displacement of the lipid-like density at the S6-gate (Supplementary Fig. 10d) 26 . Thus, our Na X structures advance an emerging paradigm in channels whereby membrane lipids can directly modulate pore structure, drug binding, and ion conductance.
Under the conditions that we have examined, the human Na X channel does not function independently as a Na + -activated channel, nor as a voltage-activated channel. Using structure-guided engineering, we discovered a Na X -QTT channel construct that displays a non-selective cationic conductance and potent inhibition by extracellular Ca 2+ and TTX. Overall, although the steps, factors, or pathways required for endogenous channel activation still need to be precisely defined, our study reveals that the Na X channel may operate as a Ca 2+ -modulated Na + leak channel and provides new tools to further demystify this once recalcitrant channel.

Methods
Protein expression and purification. Full-length human Na X containing tandem 2xStrepII and 2xFLAG tags at its N-terminus and untagged full-length human β3 were each cloned into a pRK vector behind a CMV promotor. Constructs were co-transfected with PEI into Expi293 cells and cultured for 48 h in SMM 293T-I medium under 5% CO 2 at 37°C. Cells were harvested by centrifugation at 800 × g for 10 min and resuspended and lysed in Buffer A (25 mM ADA pH 6.0, 200 mM NaCl, 1 mM PMSF, 1 μg/mL benzonase, and 1× Roche protease inhibitor cocktail) by dounce homogenization. To solubilize the cell membranes, glyco-diosgenin (GDN) and cholesterol hemisuccinate (CHS) were added to the sample to final concentrations of 2% and 0.3%, respectively, and the sample was gently stirred at 4°C for two hours. Insoluble material was then collected by ultracentrifugation at 125,000 × g for 1 h at 4°C. Supernatants were applied to anti-FLAG M2 agarose resin that had been preequilibrated in Buffer B (25 mM ADA pH 6.0, 200 mM NaCl, 0.01% GDN) and bound in batch at 4°C for 1 h. The sample was then applied to a gravity column and the collected resin was washed with 10 CV Buffer B followed by 10 CV supplemented with 5 mM ATP and 10 mM MgCl 2 . Protein was eluted with 5 CV Buffer B supplemented with 300 μg/mL FLAG peptide. Eluates were applied directly to Strep-Tactin XT Superflow high capacity resin that had been pre-equilibrated in Buffer B and bound in batch for three hours, then washed with 10 CV Buffer B prior to sample elution in 5 CV supplemented with 50 mM biotin. For nanodisc incorporation, the sample was concentrated to 15 μM using an Amicon Ultra centrifugal filter device (100 kDa MWCO). For structural analysis in detergent (GDN), the eluate was instead concentrated to 100 μL and applied to a Superose 6 3.2/300 column that had been preequilibrated in Buffer B.
Mass spectrometry analysis. Following size exclusion chromatography, 10 µg of the Na X -containing peak fraction (co-expression with the β3-subunit) was denatured with 8 M guanidine (1:1 v/v), reduced with 1 M dithiothreitol (DTT, Sigma-Aldrich, St Louis, MO) to a final concentration of 100 mM DTT and incubated at 95°C for 10 min, then centrifuged. 10 µL of 1 M Tris, pH 8 was added to the solution, and water was used to dilute the guanidine to a 2 M final concentration. Deglycosylation was performed with 2 µL of PNGaseF (15000 U, New England Biolabs) followed by overnight incubation at 37°C. Samples were further reduced with 10 mM DTT at 60°C (15 min) followed by alkylation with 20 mM iodoacetamide at room temperature. Proteins were digested with 0.2 µg trypsin (Promega) or chymotrypsin in 50 mM ammonium bicarbonate, pH 8 at 37°C overnight. Digestions were quenched with formic acid and the supernatants were subjected to desalting on C18 PhyTips (PhyNexus), lyophilized, reconstituted in 0.1% formic acid containing 2% acetonitrile and analyzed without further processing by reversed-phase nano-LC/MS/MS on a Waters NanoAcquity HPLC system (Waters Corp.) interfaced to an Elite Orbitrap mass spectrometer (ThermoFisher Scientific). Peptides were loaded onto a Symmetry C18 column (1.7 mm BEH-130, 0.1 × 100 mm, Waters) and separated with a 60 min gradient from 2 to 25% solvent B (0.1% formic acid, 98% acetonitrile) at 1 µL/min flow rate. Peptides were eluted directly into the mass spectrometer with a spray voltage of 1. (3) (4) (3) (4) Fig. 4 Pharmacology of the human Na X -QTT channel. a Representative I-V curves from HEK293T cells expressing human Na X -QTT with indicated extracellular monovalent cations with or without indicated amounts of CaCl 2 in the extracellular solution. Voltage ramp from −80 to +80 mV was applied. Right, percentage of block of outward Na + by indicated concentrations of Ca 2+ at 80 mV. Data are shown as mean ± SD of n = 5 cells over three independent experiments. b Representative currents from Xenopus laevis oocytes expressing human Na X -QTT in standard extracellular solution with or without indicated divalent and trivalent cations (unit in mM), when stepping from 0 to +80 mV. c Representative currents from oocytes expressing human Na X -QTT in standard extracellular solution with or without indicated blockers added, when stepping from 0 to +80 mV (left) or from 0 to −100 mV (right). Middle, summary of currents measured from two independent experiments. Data are shown as mean ± SD. Numbers of biological replicates (n) are indicated. See Methods for concentrations of compounds tested. (+57.0215 Da) of cysteine as a static modification. Data were searched against the specific sequences of in-house constructs overlaid onto the Uniprot mammalian database, including reverse protein sequences. Peptide assignments were first filtered to a 2% false discovery rate (FDR) at the peptide level and subsequently to a 2% FDR at the protein level.
Reconstitution of β3-Na X into lipid nanodiscs. For reconstitution into nanodiscs, multiple 50 μL sample aliquots were each applied to 2 mL tubes with a 200-fold molar excess of a POPC:POPE:POPG lipid mix (3:1:1 ratio solubilized in a sonication bath at 10 mg/mL in a buffer containing 50 mM HEPES pH 7.5, 100 mM NaCl, 5 mM MgCl 2 , and 1% CHAPS) and incubated for 30 min at 4°C. The membrane scaffold protein MSP1E3D1 (Sigma) was then applied to the samples in a 4-fold molar excess and incubated for an additional 30 min at 4°C. Samples were then diluted to 1.5 mL and Bio-Beads were applied to a final concentration of 0.25 mg/mL. Samples were then nutated overnight at 4°C prior to Bio-Bead removal. The samples were then pooled and passed over 1 mL Strep-Tactin XT resin in a column that had been pre-equilibrated in Buffer C (25 mM HEPES pH 7.0, 200 mM NaCl) to remove empty nanodiscs. The column was washed with 5 CV Buffer C prior to sample elution with 5 CV containing 50 mM biotin. The sample was then concentrated to 100 μL and applied to a Superose 6 3.2/300 column that had been pre-equilibrated in Buffer C. Peak fractions were pooled and concentrated to 3 mg/mL.
Cryo-EM sample preparation and data acquisition. Samples were cross-linked with a final concentration of 0.05% EM-grade glutaraldehyde at room temperature for 10 min. The crosslinking reactions were quenched by the addition of 1 M Tris pH 7.0. Ultrafoil R2/2 (200 mesh) cryo-EM grids (Quantifoil GMBH) were plasma cleaned for 5 s using the Solarus plasma cleaner (Gatan, Pleasanton, CA). Three microliters of the final samples at 2.25 mg/mL (nanodiscs) or 1.5 mg/mL (GDN) were applied to the cryo-EM grids and blotted for 2.5 s using a Vitribot Mark IV (ThermoFisher Scientific, Waltham, MA) using a blot force setting of 8 at 100% humidity, and plunged into liquid ethane. Grids were then imaged on a Titan Krios electron microscope (ThermoFisher Scientific, Waltham, MA) operated at 300 keV with a bioquantum energy filter using a K2 (nanodiscs) or K3 (GDN) Summit direct electron detector (Gatan, Pleasanton, CA). Images of the nanodisc were recorded at a magnification of ×165,000, which corresponded to 0.824 Å/pixel using a 20 eV energy slit. Image stacks contained 50 images recorded at 0.2 s intervals over 10 s, giving a total exposure of~50 e − /Å 2 . Images of the GDN sample were recorded in super-resolution mode at ×105,000 magnification, corresponding to 0.419 Å/pixel, using a 20 eV energy slit. Image stacks contained 60 images recorded at 0.05 s intervals over 3 s, giving a total exposure of~64 e − /Å 2 . All data collection was done using serialEM 36 .
For the first dataset, 11,658 movies were corrected for frame movement using MotionCor2 41 in RELION. The resulting images were filtered to retain only those with an accumulated motion total below a value of 250 and contrast-transfer function (CTF) parameters were fit using the 30-4.5 Å band of the spectrum using CTFFIND4.1 42 . Images were filtered to include only those with a detected fit resolution better than 5 Å, giving a total of 9988 good images for further processing. 1,669,107 particles were picked using a deep learning-based algorithm in WARP 37 . Particles were subjected to two rounds of 2D classification in cisTEM, and the best 30 classes were chosen (76,392 particles) and exported to RELION for 3D classification. A 20 Å low-pass filtered (LPF) K V 1.2 reconstruction solved in nanodiscs (EMD-9024) 43 for which all density outside of the nanodisc had been erased was used as the initial 3D reference. The best obtained 3D volume was then used as the reference for the second round of 3D classification using a broader selection of particles from the 2D classification in cisTEM (350,901 particles). The best 3D volume and its corresponding 91,435 particles were then imported back into cisTEM for iterative rounds of auto-refine without a mask and manual refinement with iteratively adjusted masks and 20 Å LPF outside the mask (outside weight of 0.8). The resulting 4.5 Å map was then used as the reference in a final round of unmasked auto-refinements and masked manual refinements in cisTEM using the particle stack from only a single round of 2D classification in cisTEM, giving the final 3.2 Å map.
For the GDN detergent sample, Cryo-EM data were processed using a combination of the RELION, Gautomatch, and cisTEM software packages [38][39][40]44 . 10,850 movies were corrected for frame movement using MotionCor2 41 in RELION and binned to 1 Å/pixel. The resulting images were filtered to retain only those with an accumulated motion total below a value of 250 and contrast-transfer function (CTF) parameters were fit using the 30-4.5 Å band of the spectrum using CTFFIND4.1 42 . Images were filtered to include only those with a detected fit resolution better than 5 Å, giving a total of 10,569 good images for further processing. A total of 2,780,663 particles were automatically picked using Gautomatch 44 using 16 uniform projections of the reconstruction of Na V 1.4 in complex with the β1 auxiliary subunit (EMD-9617) as templates. Particles were subjected to a single round of 2D classification into 250 classes in RELION and a particle stack containing 1,420,422 particles from the best 55 classes were extracted and exported to cisTEM. The structure of β3-Na X solved in nanodiscs was used as an initial reference in an initial round of auto-refine without a mask, followed by several rounds of manual refinement with iterative mask refinement and 20 Å LPF outside the mask (outside weight of 0.8). Manual refinement runs included defocus refinement and used a score threshold of 0.2. Iterative rounds of unmasked autorefinements and masked manual refinements yielded a final reconstruction of 2.85 Å resolution.
Model building and structure analysis. The structure of Na V 1.4 in complex with the β1 auxiliary subunit (PDB 6AGF) 22 was used as a template to generate a Na X homology model using the Phyre2 server 45 . This model along with β1 was rigid body docked into the cryo-EM map for manual rebuilding in Coot 46 to give an initial model of the β3-Na X complex. Multiple rounds of real-space refinement in Phenix 47 and manual rebuilding in Coot were followed by molecular dynamicsassisted manual refinement in UCSF ChimeraX 48 with ISOLDE 49 . After a final refinement in Phenix, the model was validated using the Phenix validation package. Structure figures were generated using PyMol 50 , UCSF Chimera 51 , and UCSF ChimeraX. Caver3.0 was used to analyze the channel pore 52 . Sequence alignments were performed using Clustal Omega 53 and rendered with ESPript 3.0 54 .
Cell culture, cell surface biotinylation, and western blots. HEK293T and Neuro-2a cells (ATCC) were grown and maintained as described previously 33 . Approximately 800,000 HEK293T and 250,000 Neuro-2a cells were seeded in 35 mm cell culture dishes~20 h before transient transfection with 1 µg of cDNA using LipoD293 ver. II (tebu-bio). For biochemical experiments, the cDNAs of Na X -eGFP-2 × FLAG and β3 were mixed in a mass ratio of 1:1. Equal amount of empty vector DNA was added to keep the total cDNA amount constant when β3 was excluded from transfection. Transfected HEK293T cells were used for biochemical experiments 24 h post-transfection. To induce differentiation, Neuro-2a cells were serum-starved 24 h post-transfection (cultured in DMEM for an additional 24-30 h) before they were used for biochemical and electrophysiological experiments. Cell surface biotinylation and western blots were performed as described previously 33 , with only slight modifications: (1) The quenching step was