Structure of the human sodium leak channel NALCN in complex with FAM155A

Functional characterization of NALCN by electrophysiology

We employed patch-clamp recordings in HEK293 cells to characterize the electrophysiological properties of NALCN. The current elicited in response to voltage steps was very small and appeared to be indistinguishable from the mock result when NALCN was expressed alone in HEK293 cells (Supplementary Fig. 1a). In contrast, coexpression of NALCN with UNC79, UNC80, and FAM155A dramatically enhanced the current, which was largely inhibited by 1 mM verapamil, implying that the current measured was mainly mediated by NALCN (Supplementary Fig. 1b). We also recorded the kinetics of the voltage-sensitive NALCN current: hyperpolarization voltages elicited a large inward current with inactivation, while depolarization voltages activated the current with an exponential time course before reaching the steady state (Supplementary Fig. 1b). These observations indicate that the auxiliary subunits are essential for a functional NALCN, which is consistent with previous studies^12,30. The electrophysiological properties of NALCN suggest that it tends to maintain the RMP in a manner reminiscent of Le Chatelier’s principle in chemical equilibria. In comparison, Na_v channels always show rapid current activation and inactivation for fast initiating action potentials³³, while certain K_v channels such as KCNQ channels show slow activation kinetics to avoid fast repolarization³⁴.

Structure determination of the human NALCN-FAM155A complex

Based on our functional characterizations, we focused on the coexpression of the four components in HEK293 cells. Unfortunately, we were unable to obtain a stable complex of the four components. Instead, a stable and homogeneous subcomplex of NALCN and FAM155A was obtained (Supplementary Fig. 2a). The peak fractions from gel filtration purification were pooled and concentrated to approximately 8.5 mg/ml for cryo-EM sample preparation. The EM images were collected on a Titan Krios electron microscope operating at 300 kV and equipped with a K3 direct detector and a GIF Quantum energy filter. After a few rounds of two-dimensional (2D) and 3D classifications, about sixty-five thousand particles were selected, which yielded a final reconstruction map with an overall resolution of 3.1 Å (Fig. 1 and Supplementary Table 1, Supplementary Figs. 2, 3). Local masks of the extracellular and intracellular regions were applied to further improve the local resolutions to 2.9 and 3.1 Å, respectively.

The density of the final reconstruction is of high quality. Most side chains were clearly resolved, allowing for accurate model building (Fig. 1a, b and Supplementary Figs. 4, 5). The model of NALCN was built starting from a homology model based on Ca_v1.1 (PDB: 5GJV), whereas FAM155A was built de novo. In total, 1278 and 182 residues were built for NALCN and FAM155A, respectively. The complex structure was confirmed by cross-linking mass spectrometry analysis (Supplementary Fig. 6). Three glycosylation sites (Asn210, Asn216, and Asn1064) were identified on NALCN and one glycosylation site (Asn217) on FAM155A (Supplementary Fig. 5c). Four disulfide bonds were recognized in the extracellular loops (ECLs) of NALCN and six in FAM155A (Supplementary Fig. 5c). The glycosylation sites and disulfide bonds in turn further verified the accuracy of the structure.

The overall structure of NALCN resembles those of eukaryotic Na_v and Ca_v channels (Figs. 1c and 2). We selected human Na_v1.4 (PDB: 6AGF) and rabbit Ca_v1.1 (PDB: 6JP5) as representative Na_v and Ca_v channels for comparisons with NALCN^35,36. The structures of human Na_v1.7 (PDB: 6J8I), rat Na_v1.5 (6UZ3) and American cockroach Na_vPaS (6A95) were also selected for specific discussions^37–39. The structures of NALCN can be superimposed to the α subunit of Na_v1.4 and α1 subunit Ca_v1.1 with root-mean-square deviation (r.m.s.d.) of 2.99 Å over 760 Cα atoms and 3.894 Å over 905 Cα atoms, respectively. In addition to the four homologous repeats, a C-terminal domain (CTD) after repeat IV, which was observed in the structures of Ca_v1.1 and Na_vPaS but not in human Na_v channels, was also clearly resolved in the structure of NALCN (Fig. 1c). NALCN has no N-terminal domain (NTD) before repeat I, which always exists in Na_v channels^38–40.

NALCN has several unique features that distinguish it from Na_v and Ca_v channels. FAM155A adopts a distinct binding interface on NALCN, which is different from those between the ion-conducting α subunit of Na_v and Ca_v channels and their auxiliary subunits (Fig. 2b). Moreover, the S4-5_I linker unexpectedly forms a flexible loop instead of a juxtamembrane helix as seen in the other three repeats of NALCN and other Na_v/Ca_v structures (Fig. 2c). Meanwhile, S5_I elongates to the cytosol by three-helix turns compared to that of Na_v/Ca_v channels. Notably, the unique structure of S4-5_I and S5_I results in a cavity that accommodates a lipid, which may contribute to the regulation of NALCN (Fig. 2c and Supplementary Fig. 4b). The local structure around the lipid is very different between NALCN and Na_v/Ca_v channels. Intriguingly, the lipid sits in a position that is occupied by S4-5_I linker in Na_v/Ca_v structures (Fig. 2c). In addition, a short helix designated II-III linker, which is connected to S6_II by a short loop, was observed in the structure of NALCN (Fig. 1c).

Specific interactions between NALCN and FAM155A

Human FAM155A contains 456 amino acids, and residues 192–382 were resolved in our structure except for a short loop from residues 250–258 (Fig. 3a). The resolved structure of FAM155A contains two lobes, designated the N-lobe and C-lobe, stabilized by six disulfide bonds. The N-lobe consists of three short β strands and two short α helices (H1–H2), while the C-lobe contains four α helices (H3–H6) (Fig. 3b). The overall structure of FAM155A is different from that of any auxiliary subunit in Na_v or Ca_v channels (Fig. 2b, Supplementary Fig. 7a). A search of the protein data bank using the DALI server⁴¹ revealed that the C-lobe of FAM155A contains a fold (similarity Z-score 4.3) similar to the structure of the frizzled-like cysteine-rich domain (CRD) of receptor tyrosine kinase MuSK (PDB: 3HKL), which had 84 Cα atoms aligned to FAM155A with an r.m.s.d. of 2.8 Å (Supplementary Fig. 7b). Moreover, three disulfide bonds are structurally conserved between the two structures (Supplementary Fig. 7b).

Only the C-lobe of FAM155A is involved in the interaction with NALCN. FAM155A sits above the pore domain of repeat IV of NALCN and forms extensive interactions with the ECL_I, ECL_III, and ECL_IV of NALCN (Fig. 3c and Supplementary Table 2). The loop in FAM155A connecting H4 and H5 and a hairpin loop after H6 are designated L1 and L2, respectively (Fig. 3b). Notably, L2 of FAM155A inserts deeply into a dome formed by ECL_III of NALCN. The H3, H5, H6 helices, and L2 of FAM155A resemble the palm and thumb of a hand, respectively, holding ECL_III of NALCN tightly (Fig. 3c). Detailed analysis of the interaction interfaces revealed extensive salt bridges and hydrogen bonds between NALCN and FAM155A (Fig. 3d and Supplementary Table 2). Unexpectedly, the sugar moiety linked to Asn1064 on ECL_III, seems also contributed to the interaction by potentially forming a hydrogen bond with E350 on FAM155A (Fig. 3d). Electrostatic surface analysis revealed that two surfaces between NALCN and FAM155A are electrically complementary, ensuring a favorable interaction stability (Supplementary Fig. 7c). Sequence alignment among NALCN and Na_v/Ca_v channels revealed that most of the interface residues in NALCN, including Asn1064, are not conserved among other Na_v/Ca_v channels, indicating the high binding specificity of FAM155A towards NALCN, but not other channels (Supplementary Fig. 8). On the other hand, the key residues in the interaction interface are highly conserved among NALCN and FAM155A from different species, indicating that the interaction mode is evolutionarily conserved (Supplementary Fig. 9). Furthermore, the key residues mediating complex formation in FAM155A are also highly conserved in FAM155B, suggesting FAM155B may also be able to form a stable complex with NALCN in vivo (Supplementary Fig. 9). This notion is supported by the finding that FAM155A can be functionally substituted by human FAM155B¹².

As UNC79 and UNC80 were not resolved in the structure, we then asked whether coexpression of NALCN and FAM155A alone was also able to recapitulate measurable NALCN currents. Interestingly, we readily recorded a large current in response to voltage steps when NALCN and FAM155A were cotransfected (Supplementary Fig. 1c). The current–voltage curve of NALCN coexpressed with either all three auxiliary subunits or with FAM155A alone was nonlinear (Supplementary Fig. 1d). The current response of NALCN with FAM155A exhibited similar voltage sensitivity to that of NALCN coexpressed with three auxiliary subunits, as the apparent gating charge (Z_app) was similar (1.22 ± 0.18 e₀ and 1.25 ± 0.55 e₀, respectively) (Supplementary Fig. 1d, e). However, the G–V curve of NALCN coexpressed with three auxiliary subunits was largely shifted towards the hyperpolarization voltage compared to NALCN with FAM155A alone (V_1/2: −65.07 ± 2.11 and −44.66 ± 1.91 mV, respectively) (Supplementary Fig. 1e). Moreover, we observed no inactivation of the steady-state current upon setting the prepulse potential to different levels when NALCN was coexpressed with auxiliary subunits (Supplementary Fig. 1f, g), which is consistent with the physiological role of NALCN as a leaky sodium channel. The observation that FAM155A alone is also able to recapitulate NALCN function is different from several recent studies^31,42.

Structural basis for the voltage dependence of NALCN

Our structural study has provided an opportunity to understand the voltage sensitivity of NALCNs. Detailed analysis of the VSDs of NALCN suggests that they possess several key features shared among functional VSDs in Na_v/Ca_v channels. For instance, each VSD preserves the positively charged residues known as gating charges (GCs) in a pattern of occurring every three amino acids in the S4 segments. As usual, we define the position of the last GC in each S4 as R6⁴³. NALCN has two to four GCs in each repeat distributed at positions R2–R6 (Fig. 4). The last GC of repeat IV is at a position that has one residue shift to R6, a phenomenon also observed in other channels such as Ca_v1.1⁴³. Moreover, the residues of the charge transfer center (CTC) which consists of a negatively charged residue (An2) and an aromatic occluding residue (F) on S2, and a negatively charged residue on S3, are highly conserved in NALCN except for the occluding residue on VSD_III. The An1 sites, which lie seven residues ahead of the occluding residues on S2 segments of NALCN, are mainly negatively charged or polar residues, similar to those of Na_v/Ca_v channels (Fig. 4a, b). In addition, previous structural studies on Na_v channels identified several negatively charged or polar residues on S1 that may also have roles in GC transfer³⁵. These residues were also observed on NALCN (Fig. 4a). When the four VSDs are superimposed relative to An1 and CTC, all R4 residues are above the occluding residues on S2, reminiscent of up or depolarized states (Fig. 4b and Supplementary Fig. 10). Further comparisons with Na_vPaS whose VSD_III and VSD_IV are in a potential resting state, revealed obvious upward movement of the S4 segments from Na_vPaS to NALCN, confirming the depolarized conformations of VSDs in NALCN (Supplementary Fig. 10).

The voltage sensitivity of NALCN, however, seems to be lower than that of Na_v/Ca_v channels (Supplementary Fig. 1). Several unique structural features of NALCN may lead to its relatively weak voltage sensitivity. First, the occluding residue on S2 in repeat III is replaced by a methionine (Met929), instead of a conserved phenylalanine or tyrosine (Fig. 4c). The occluding residue of the CTC is reported to be crucial for the GC transfer during voltage sensing⁴⁴. Second, the S4 segments of VSDs are usually formed as 3₁₀ helices, whereas in repeat IV the S4 largely relaxes to a regular α helix. The only 3₁₀ helix turn in S4_IV, which is at position R5, is a serine instead of a GC. Therefore, VSD_III and VSD_IV may contribute little to the voltage sensitivity, due to lack of occluding residue on S2_III and GCs and 3₁₀ helices on S4_IV, respectively. This structure-based speculation is consistent with a reported functional study, in which neutralization of the GCs in repeat III/IV has little effect on voltage sensitivity¹². In contrast, neutralization of the GCs R146 + R152 in repeat I or R481 + R484 or R481 + K487 in repeat II led to significant changes in voltage sensitivity¹². Third, the upper GCs (R2-R4) in repeat I-III form more extensive hydrogen bond interactions with surrounding negatively charged or polar residues than their counterparts in Na_v/Ca_v channels (Fig. 4c). During the GC transfer process, the GCs break old interactions with surrounding residues and form new interactions with other residues, accompanied by a sliding movement along the S4 segment. The extensive interactions of the GCs and surrounding residues in NALCN may require the VSDs to overcome a higher energy barrier to undergo conformational changes, weakening the voltage sensitivity. Fourth, as S4-5 linkers are supposed to be a key transducer from VSDs to the ion-conducting pore during voltage sensing gating in VGICs⁴⁵, the relaxation to a flexible linker of S4-5_I may hinder the effective transduction of the electromechanical coupling of repeat I in NALCN. Altogether, these structural variations in NALCN may contribute to its low voltage sensitivity.

EEKE selectivity filter in NALCN

Like all reported Na_v/Ca_v structures, the pore domain of NALCN is formed by the S5 and S6 helices of the four repeats, and the SF of NALCN is supported by two-pore helices (P1 and P2) that intervene between S5 and S6 (Fig. 5a). The key residues in the SF of NALCN responsible for the ion selectivity are E280/E554/K1115/E1389 (EEKE), distinct from DEKA in Na_v channels and EEEE/EEDD in Ca_v channels. Sequence alignment of the P1-SF-P2 among NALCN, Ca_v1.1, and Na_v1.4 reveals an invariant tryptophan in the first residue of P2 and a highly conserved residue (Thr/Ser/Cys) in the last residue of P1, while the other residues are not very conserved (Fig. 5b). The overall structure of P1-SF-P2 in NALCN is more closely related to that of Ca_v1.1 than Na_v1.4, with r.m.s.d. values of 0.848 Å over 110 Cα atoms and 1.602 Å over 107 Cα atoms, respectively. The backbone of the pore helices of NALCN aligns well to those of Ca_v1.1, whereas the P2s do not align well to those of Na_v1.4 (Fig. 5c). Despite its structural similarity with Ca_v channels, NALCN is responsible for the background sodium leak conductance in hippocampal neurons, indicating that NALCN is selective for Na⁺ in vivo⁴.

To better understand Na⁺ binding and selectivity in SF, we performed canonical MD simulation to investigate the interaction of the membrane-embedded pore domain in 150 mM NaCl. In three independent trials of 200 ns equilibration, we observed three potential Na⁺ ion-binding sites, designated Site1-3, in proximity to SF (Fig. 5d). In addition to the glutamic acid residues in the SF, the Na⁺ binding sites are formed by several negatively charged residues from the pore helices, including E275, D558, D561, and D1390. Among the three binding sites, Site1 is most stably located in the middle of SF, while the other two are transient and presumably cation attractants (Fig. 5d and Supplementary Movie 1). Probability statistics suggest that there is an average of two Na⁺ ions within the 4 Å range of the SF residues over time (Fig. 5e). Notably, Site 3 formed by E280/E554/D558 (EED site) is spatially close to the Na⁺ binding site formed by a DEE motif in Na_v channels³⁵ (Fig. 5f). However, the sidechains of the EED motif in NALCN are not as closely spaced as those of the DEE site in Na_v channels and may not form a favorable Na⁺ binding site. On the other hand, Site1, the most occupied binding site, is spatially close to a putative Ca²⁺ binding site in Ca_v1.1 and Ca_v3.1 channels^36,43 (Fig. 5f). The replacement of glutamic acid (E) or aspartic acid (D) with lysine (K) in repeat III from Ca_v channels to NALCN shifts the ion selectivity from Ca²⁺ to monovalent ions. Like Na_v channels, the lysine in NALCN favors permeation of Na⁺ but not Ca²⁺ through the SF. However, the EEKE SF may still preserve the ability to bind Ca²⁺, as the structure of SF in repeat I/II/IV is almost identical to that of Ca_v channels (Fig. 5c). In this sense, extracellular Ca²⁺ may compete with Na⁺ in binding to SF, suggesting a potential mechanism of NALCN blockage by extracellular divalent cations¹².

Closed intracellular gate of NALCN

The ion permeation path below the SF in NALCN is enclosed by the S6 tetrahelical bundle, a structural feature that is shared in all reported Na_v/Ca_v structures (Figs. 5a and 6a). Notably, a short seven residue π helix is embedded within each S6 segment (Supplementary Fig. 11a, b). The π helices are stabilized by intra-helical hydrogen bonds. Moreover, W311/L588/M1145/Y1436 of the π helix form close inter-helical contacts with L1439/K314/L591/L1148 from the adjacent π helix, sealing the lateral pore fenestration in NALCN (Supplementary Fig. 11b). We used HOLE⁴⁶ to calculate the radius along the permeation path and compared it with that of Na_v1.4 and Ca_v1.1, whose intracellular gates are in open and closed states, respectively. Due to the existence of FAM155A, ions can enter the permeation path from only one side, which is above the pore domain of repeat II (Fig. 6a, left). Surprisingly, although NALCN is supposed to conduct a “leak” current, the intracellular gate is closed, even more tightly than the closed Ca_v1.1. The narrowest region, sealed by two layers of the hydrophobic residues Val, Ile, and Leu, is ~10 Å in length along the permeation path with a radius <1 Å (Fig. 6a, right). Notably, the lower gate is fully blocked by the III-IV linker. Detailed analysis reveals that the III-IV linker is stabilized by extensive hydrogen bonds or polar interactions with S6_I and the II-III linker (Fig. 6b). Most of the involved residues are not conserved among Na_v/Ca_v channels, indicating that the local interactions among S6_I, the II-III linker and the III-IV linker are highly specific to NALCN, which implies that NALCN has a distinct gating mechanism (Supplementary Fig. 7).

It is intriguing that the position of the III-IV linker in NALCN is distinct from those of both Na_v1.4 and Ca_v1.1 in different ways. NALCN does not have an IFM motif, a highly conserved motif in the III-IV linker of all Na_v channels that is crucial for fast inactivation through an allosteric inhibition mechanism^45,47. In Na_v1.4, the III-IV linker is away from the inner pore, and its helix rotates counterclockwise by ~75° but keeps at a similar vertical height to that of NALCN (Fig. 6c, top). The helix in the III-IV linker, however, shows a similar orientation but distinct vertical height between NALCN and Ca_v1.1. Compared to Ca_v1.1, the helix in the III-IV linker of NALCN is shifted ~6 Å up towards the extracellular side, probably due to a shorter loop connecting S6_III and the helix in the III-IV linker (Fig. 6c, bottom). The II-III linker, which is part of the elongated S6_II in Ca_v1.1, bends towards the center pore. These distinct structural features in NALCN make the S6_I, II-III linker and III-IV linker forms a close contact that tightly seals the inner gate.

Like Ca_v1.1 and Na_vPaS, the III-IV linker of NALCN also interacts with the CTD (Fig. 1c). The CTD of NALCN consists of at least six short helices C1-C6, among which C1-C4 form an EF-hand-like domain that is structurally conserved in the CTD of Ca_v1.1, Na_vPaS, and Na_v1.5. However, C5-C6 helices in NALCN are distinct from other channels (Supplementary Fig. 12). The conserved IQ motif which is a binding site for calmodulin in the C6 helix of Na_v channels is absent in NALCN. Interestingly, when the CTDs are superimposed, the III-IV linker of NALCN displays a shift compared to that of Ca_v1.1 and Na_vPaS, or coincides with the C5 helix of Na_v1.5, suggesting that the CTD and the III-IV linker may undergo relative conformational changes (Supplementary Fig. 12). Speculatively, CTD may provide specialized binding sites for UNC80 or UNC79 in the regulation of the gating of NALCN.

A unique lipid-binding site on NALCN

Similar to S4-5_I and S5_I, the S4-5_II, S5_II, and S6_II segments in NALCN also displays significant structural deviations compared to that of Ca_v1.1/Na_v1.4 (Fig. 6c). The dramatic differences in the pore domain of repeat II may be related to the binding of a lipid molecule (lipid1) in a nearby cavity (Fig. 2c). We have assigned a total of ten lipid molecules in the structure. Most of the lipids in the structure bind at semi-closed clefts around the protein, while only lipid1 sits in a closed hydrophobic cavity surrounded by S4-5_I, S5_I, S6_I, S5_II, and S6_II (Fig. 1c, Supplementary Fig. 13a, b). Although the exact identity of lipid1 is not clear, the two-tail shape density suggests it may be a phospholipid (Supplementary Fig. 4b). Two positively charged residues Lys504 and Lys505 from S5_II can form hydrogen bonds with the phosphate group of lipid1, providing the molecular basis of binding specificity. Notably, such lipid binding is not observed in other channels, implying an important role in regulating the function of NALCN. Indeed, mutations on the two positively charged residues, such as K504A, K504D, K505A, result in a significant reduction in the measured currents conducted by NALCN (Supplementary Fig. 13c). In contrast, mutations on the hydrophobic residues along the lipid-binding cavity, such as F325A, I328W, increase the measured current, while F332A has little effect on the current. These results suggest that except for the auxiliary subunits, the function of NALCN is also fine-tuned by specific lipid–protein interaction.

Protein expression and purification

NALCN, UNC-80, UNC-79, and FAM155A were cloned into pCAG vector. HEK 293F cells (Invitrogen) were transfected with the four plasmids and harvested after 60 h. Cells were resuspended in buffer containing 25 mM Mops (pH 7.4), 250 mM NaCl, 0.5% (w/v) Lauryl Maltose Neopentyl Glycol (LMNG, Anatrace), 0.06% (w/v) cholesteryl hemisuccinate Tris salt (CHS, Anatrace) and protease inhibitor cocktail including 2 mM phenylmethylsulfonyl fluoride (PMSF), 1.3 μg/ml aprotinin, 0.7 μg/ml pepstatin, and 5 μg/ml leupeptin, then incubated at 4 °C for 2 h. The insoluble fraction was precipitated by ultracentrifugation at 255,700×g for 1 h, and the supernatant was applied to anti-Flag G1 affinity resin (GenScript) by gravity. The resin was rinsed several times with the Wash buffer, which contains 25 mM Mops (pH 7.4), 500 mM NaCl, 0.01% (w/v) Lauryl Maltose Neopentyl Glycol, and the protease inhibitor cocktail. The target proteins were eluted by the Elution buffer containing 25 mM Mops (pH 7.4), 500 mM NaCl., 0.02% (w/v) glyco diosgenin (GDN, Anatrace), and the protease inhibitor cocktail plus 200 μg/ml FLAG peptide (GenScript). The eluent was then concentrated using a 50-kDa cut-off Centricon (Millipore) and further purified by Superose 6 Increase 10/300 GL column (GE Healthcare) in 25 mM Mops (pH 7.4) 150 mM NaCl, 0.02% (w/v) GDN, and the protease inhibitor cocktail. Fractions were collected and concentrated to about 8.5 mg/ml for cryo-EM analysis.

Whole-cell electrophysiology

Patch-clamp recordings were performed with a HEKA EPC10 amplifier with PatchMaster software (HEKA) in whole-cell configuration. Patch pipettes were prepared from borosilicate glass and fire-polished to resistance of ~6 MΩ. For whole-cell recording, serial resistance was compensated by at least 65%. Extracellular solution contained 150 mM NaCl, 10 mM Hepes, and 30 mM d-(+)-glucose (pH 7.4) with NaOH, and intracellular solution contained 136 mM NaCl, 5 mM EGTA, 10 mM Hepes, and 2 mM Na₂ATP (adenosine 5′-triphosphate) (pH 7.2) with NaOH. The current amplitude at the steady-state during the last 20 ms of voltage steps was averaged to construct the G–V curve. All recordings were performed under room temperature (~22 °C). Temperature variation was less than 1 °C as monitored by a thermometer. Current signal was sampled at 20 kHz.

HEK-293T cells were grown in Dulbecco’s modified Eagle’s medium (ThermoFisher Scientific) supplemented with 10% fetal bovine serum and 1% penicillin-streptomycin at 37 °C in a 5% CO2 humidified growth incubator. Cells were transiently transfected by Lipofectamine 3000 (Life Technologies) 48–72 h before patch-clamp recordings. The NALCN, UNC-79, UNC-80, and FAM155A cDNAs were mixed in the mass ratio of 1:1:1:1.

To apply solutions containing ligands such as verapamil during patch-clamp recording, a rapid solution changer with a gravity-driven perfusion system was used (RSC-200, Bio-Logic). Each solution was delivered through a separate tube so that there was no mixing of solutions. Pipette tip with a membrane patch was placed directly in front of the perfusion outlet during recording. Each membrane patch was recorded only once.

Data from patch-clamp recordings were analyzed in Igor Pro (WaveMatrix). To characterize the steady-state G–V curves, a single-Boltzmann function was used:

where G/G_max is the normalized conductance, V_1/2 is the half-activation voltage, q_app is the apparent gating charge and F is Faraday’s constant, R is gas constant, and T is the temperature in Kelvin, which is set to 295 K (22 °C).

Cryo-EM sample preparation and data collection

The purified NALCN complex was concentrated to about 8.5 mg/ml for cryo-sample preparation. Aliquots (3.5 μl) of protein solution were loaded onto glow-discharged holey carbon grids (Quantifoil Au R1.2/1.3), which were blotted for 3.5 s and immersed in liquid ethane cooled by liquid nitrogen using Vitrobot (Mark IV, ThermoFisher Scientific). The grids were exposed through Titan Krios operating at 300 kV equipped with Gatan K3 Summit detector and GIF Quantum energy filter in super-resolution mode (81,000× magnification). Movie stacks were automatically acquired using AutoEMation⁵⁰, with a 20 eV slit width and a defocus range from −0.5 to −2.5 μm. Each stack consisting of 32 frames was exposed for 2.56 s with 0.08 s per frame, and for ~50 e⁻/Ų of the total dose.

Cryo-EM data processing

Collected movie stacks were motion-corrected by MotionCor2⁵¹ with 2-fold binning, producing micrographs of 1.087 Å pixel size. Following patch-CTF estimation, around eleven million particles were automatically picked with cryoSPARC v2⁵² from 17,922 micrographs. Three rounds of 2D classification enriched images of good classes, resulting in a total of 905,458 particles being selected. An Ab-initio reconstruction generated a low-resolution initial map for the subsequent homogeneous and non-uniform refinement jobs⁵³, yielding a 3.8-Å reconstruction with identifiable side-chain densities in NALCN pore domain only. To improve map quality, we exported the aligned particles and did a 3D classification without orientation assignment using RELION 3.0 (--skip-align flag)⁵⁴. One out of five 3D classes represented clear secondary structures of the overall channel complex, and the 65,177 particles were fed back to cryoSPARC v2 for further homogeneous and non-uniform refinements that pushed the resolution to 3.14 Å. Adapted masks for protein regions, extracellular subunit, and intracellular domain were applied in particle subtraction and local refinement tasks, which helped in achieving slightly higher resolution at 3.11, 2.94, and 3.04 Å, respectively. Map resolutions were determined by the gold-standard Fourier shell correlation (FSC) 0.143 criterion using Phenix.mtriage⁵⁵.

Model building and refinement

An initial model of NALCN was generated using SWISS-MODEL online server⁵⁶. The template used for the homology modeling was the α1 subunit of Ca_v1.1 (PDB: 5GJV). The model was firstly docked into the final reconstruction map in Chimera⁵⁷ and then manually adjusted in COOT⁵⁸. After building the model of NALCN, an extra density remained unmodelled, which is close to the ECL of NALCN in the extracellular side. The high quality of the map enables us to identify that the extra density belongs to FAM155A. The structure of FAM155A was then de novo built. The sequence assignment was guided by bulky residues such as Phe, Tyr, and Trp. Up to six disulfide bonds and a glycosylation site (Asn217) were identified in FAM155A, further verifying the accuracy of the model. Ten lipid molecules were manually built to fit into the corresponding densities. The density for the C6 helix of CTD is discontinuous and has been tentatively assigned as NALCN (1589-1602) based on the side chain density.

Subsequently, the models were refined against the corresponding maps by PHENIX⁵⁵ in real space (phenix.real_space_refine) with secondary structure and geometry restraints generated by ProSMART⁵⁹. Overfitting of the overall models was monitored by refining the models in one of the two independent half maps from the gold-standard refinement approach and testing the refined model against the other map⁶⁰. Statistics of 3D reconstruction and model refinement can be found in Supplementary Table 2.

Molecular dynamics simulation

The pore domain of NALCN (residue 171–204, 260–328, 490–603, 1001–1042, 1094–1163, 1318–1400, and 1420–1455) was embedded in 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphocholine (POPC) bilayer for molecular dynamics simulations. After the determination of side-chain protonation states by PROPKA3.1⁶¹ and peptide terminal neutralization by acetyl/methylamine capping groups, water molecules and 150 mM NaCl were added using VMD⁶², resulting in a system of size 100 × 100 × 105 Å with ~93,000 atoms. The system was parameterized using tleap in AmberTools18⁶³ by ff14SB⁶⁴ and LIPID17 force fields, and along with TIP3P water model⁶⁵. All the simulations were performed using OpenMM7⁶⁶. We first applied positional constraints (k = 10 kcal/mol/Ų) on protein-heavy atoms for a 5000-step minimization, then released constraints of side chains when heating up the system to 310 K with 2 fs stepsize (H-bonds constraints), 1 ps⁻¹ friction coefficient for Langevin dynamics, Particle Mesh Ewald (PME) method⁶⁷, and 12 Å non-bonded cut-off. During 50-ns pre-equilibration for membrane relaxation in NPT ensemble, the force constant k of backbone atoms was reduced from 10 to 0.1 kcal/mol/Ų. Three independent 200-ns production runs were conducted constraining Cα atoms more than 15 Å away from the selectivity filter residues (EEKE) with k = 0.1 kcal/mol/Ų. Simulation trajectories were analyzed using MDTraj 1.9.3⁶⁸. After superimposition of the pore region to the initial structure, we recorded the positions of sodium ions within 15 Å of SF residues EEKE every 1 ns for each run.

Protein crosslinking and LC–MS/MS analysis

Equal amount (w/w) of BS3 (bis[sulfosuccinimidyl] suberate) was added to the protein mixture, which was incubated at room temperature for 1 h. The reaction was terminated by adding 0.5 M ammonium bicarbonate to a final concentration of 20 mM for 10 min incubation. The SDS-PAGE was used to separate the crosslinked protein and stained with Coomassie Blue G-250. The gel bands of interest were cut into pieces. Sample was digested by trypsin with prior reduction and alkylation in 50 mM ammonium bicarbonate at 37 °C overnight. The digested products were extracted twice with 1% formic acid in 50% acetonitrile aqueous solution and dried to reduce volume by speedvac.

For LC–MS/MS analysis, the peptides were separated by a 65 min gradient elution at a flow rate 0.300 µl/min with the Thermo EASY-nLC1200 integrated nano-HPLC system which is directly interfaced with the Thermo Q Exactive HF-X mass spectrometer. The analytical column was a home-made fused silica capillary column (75 µm ID, 150 mm length; Upchurch, Oak Harbor, WA) packed with C-18 resin (300 A, 3 µm, Varian, Lexington, MA). Mobile phase A consisted of 0.1% formic acid, and mobile phase B consisted of 100% acetonitrile and 0.1% formic acid. The mass spectrometer was operated in the data-dependent acquisition mode using the Xcalibur 4.1 software and there is a single full-scan mass spectrum in the Orbitrap (300–1800 m/z, 60,000 resolution) followed by 20 data-dependent MS/MS scans at 30% normalized collision energy. Each mass spectrum was analyzed using the Thermo Xcalibur Qual Browser and pLink 2⁶⁹ for the database searching and cross-linking analysis.

Reporting summary

Further information on research design is available in the Nature Research Reporting Summary linked to this article.