Structure of voltage-modulated sodium-selective NALCN-FAM155A channel complex

Resting membrane potential determines the excitability of the cell and is essential for the cellular electrical activities. The NALCN channel mediates sodium leak currents, which positively adjust resting membrane potential towards depolarization. The NALCN channel is involved in several neurological processes and has been implicated in a spectrum of neurodevelopmental diseases. Here, we report the cryo-EM structure of rat NALCN and mouse FAM155A complex to 2.7 Å resolution. The structure reveals detailed interactions between NALCN and the extracellular cysteine-rich domain of FAM155A. We find that the non-canonical architecture of NALCN selectivity filter dictates its sodium selectivity and calcium block, and that the asymmetric arrangement of two functional voltage sensors confers the modulation by membrane potential. Moreover, mutations associated with human diseases map to the domain-domain interfaces or the pore domain of NALCN, intuitively suggesting their pathological mechanisms.

N ALCN channel is a voltage-modulated sodium-selective ion channel 1 . It mediates sodium leak currents that are essential for setting the membrane excitability 2 . NALCN channel is involved in several crucial physiological processes including maintaining normal circadian rhythms 3 and respiratory rhythm 4 . NALCN channel functions as a multi-protein complex, which consists of NALCN, FAM155, UNC79, UNC80, and other proteins 5 . Mutations in the genes of human NALCN or UNC80 lead to a spectrum of neurological diseases 6 , including infantile hypotonia, psychomotor retardation, and characteristic facies (IHPRF) 7,8 and congenital contractures of the limbs and face, hypotonia, and developmental delay (CLIFAHDD) 9 . In addition, a gain-of-function mutation of NALCN in the mouse causes the dreamless phenotype, suggesting the function of NALCN in rapid eye movement sleep 10 .
Sequence analysis shows NALCN protein shares a similar domain arrangement to the pore subunit of eukaryotic voltagegated sodium channels Na V or calcium channels Ca V . Four consecutive voltage sensor-pore modules are fused into a single polypeptide chain. However, unique to the NALCN channel, the positive charges on voltage sensors of NALCN are much fewer than that of Na V or Ca V . Moreover, residues in the selectivity filter of NALCN are "EEKE" instead of "DEKA" of Na V or "EEEE" of Ca V . These features render NALCN to be a special clade in the voltage-gated channel superfamily 11 . In addition to the NALCN subunit, FAM155, UNC79, and UNC80 subunits are indispensable for the function of NALCN channel. The coexpression of NALCN, FAM155, UNC79, and UNC80 proteins in a heterologous expression system, such as Xenopus laevis oocytes or HEK293 cells, is necessary and sufficient to generate robust voltage-modulated sodium-selective currents 1 . FAM155 is a transmembrane protein family with a cysteine-rich domain (CRD). There are two FAM155 family members in human, namely, FAM155A and FAM155B, both of which could support the currents of the NALCN complex 1 . FAM155 is the homolog of NLF-1 in C. elegans, which is required for the membrane localization of NALCN 12 . It is proposed that the interaction between NALCN and FAM155 is similar to the interaction between yeast calcium channel Cch1 and its regulator Mid1 13,14 . The other two proteins in the complex, UNC79 and UNC80, are large conserved proteins without any known domains but they are essential for the function of NALCN in vivo [15][16][17] . During the preparation of this manuscript, the structure of human NALCN-FAM155A at 2.8 Å in nanodiscs was published 18 and another group reported a similar structure at 3.1 Å in detergent on preprint sever 19 . Here, we describe the structure of rat NALCN and mouse FAM155A in detergent to the resolution of 2.7 Å, which provides enhanced details and additional structural insights into the mechanism of this important channel complex.

Results
Structure of the NALCN-FAM155A core complex. Despite extensive screening of NALCN homologs and optimization, we failed to purify the stable hetero-tetrameric complex of NALCN-FAM155-UNC79-UNC80. However, we found C-terminal GFP tagged rat NALCN and C-terminal flag tagged mouse FAM155A can generate voltage-modulated NALCN currents upon coexpression of mouse UNC79 and mouse UNC80 ( Supplementary  Fig. 1a). Moreover, rat NALCN and mouse FAM155A can form a stable core complex which can be purified chromatographically ( Supplementary Fig. 1b-d). Because NALCN protein dictates the properties of ion permeation, divalent ion block and voltage modulation 1 and FAM155 are required for correct trafficking of NALCN 12 , we reasoned the structure of NALCN-FAM155A complex would provide clues of how NALCN channel works. Therefore, we started with the cryo-EM studies of the NALCN-FAM155 core complex. We prepared the NALCN-FAM155A complex in a buffer with sodium chloride and GDN detergent, mimicking the symmetric sodium solution used in electrophysiology recording ( Supplementary Fig. 1a). The single-particle cryo-EM analysis yielded a map at a nominal resolution of 2.7 Å (Supplementary Figs. 2-4, and Supplementary Table 1). The map quality was sufficient for modeling the stable portions of NALCN and FAM155A which encompass 73% of NALCN and 37% of FAM155A. The excellent map quality also reveals several putative detergents or co-purified lipid densities ( Supplementary Fig. 4b).
NALCN subunit shows a typical eukaryotic Na V or Ca V structure ( Fig. 1a-d). Pore domains of domain I, II, III, and IV shape the ion permeation pathway (Fig. 1a-d). Voltage sensors (VS, S1-S4 segments) pack around the pore in a domain-swapped fashion ( Fig. 1a-d). Interestingly, the arrangements of VS are highly asymmetric, with the voltage sensor of domain I (DI-VS) being the most off-axis (Fig. 1d). On the intracellular side, the Cterminal domain of NALCN binds onto the intracellular III-IV linker. The majority of the long II-III loop and the C-terminal 167 residues of NALCN were disordered. On the extracellular side, the CRD of FAM155A protrudes from the top of the complex into the extracellular space while the two predicted transmembrane helices of FAM155A were invisible, probably due to their flexibility. The structure of NALCN allowed us to locate all of the reported missense mutations identified in the NALCN subunit that cause diseases in human and model organisms (Fig. 1e, f) 7,9,10,17,[20][21][22][23][24][25][26][27] . The structure presented here is overall similar to the recently published structure of human NALCN-FAM155A complex (EMD-22203 and PDB ID: 6XIW) 18 (Supplementary Fig. 7a), with a root-mean-square deviation of 0.68 Å. However, they differ in ion selectivity filter and cytosolic domain, as described later.
Structure of FAM155A CRD and its interaction with NALCN. The FAM155A has an extracellular cysteine-rich domain (CRD) which is folded by three consecutive α helices and the connecting loops in between. The CRD of FAM155A is highly cross-linked by disulfide bonds owing to the high content of cysteines in this domain. Most of the Cys residue pairs are conserved from NLF-1 of C. elegans to FAM155A/B in human, suggesting their important functions, presumably by stabilizing the structure of CRD (Fig. 2a-c and Supplementary Fig. 6).
The CRD of FAM155A binds at the top of the NALCN channel (Fig. 1b, c). Extracellular loops of domain I, III, and IV (L5 I , L5 III , L5 IV , and L6 IV ) in NALCN form a platform for FAM155A binding (Fig. 2d). The post-α3 loop of FAM155A extends over the extracellular ion entrance of the NALCN pore (Fig. 2e). The L5 III of NALCN covers above the post-α3 loop of FAM155A and locks it in the sandwiched conformation (Fig. 2e). The binding of CRD of FAM155A onto NALCN relies on its extensive interactions with L5 III , L5 IV, and L6 IV of NALCN. In detail, K287 and E290 on α1 of FAM155A make cation-π and electrostatic interactions with W1085 and R1062 on L5 III , respectively (Fig. 2d). K307 on α1-α2 loop interacts with D1408 on L6 IV (Fig. 2d). R1094 on L5 III interacts with the carbonyl group of G364 of FAM155A, D1397 on P2 IV , and E1364 on L5 IV (Fig. 2e). These interacting residues are highly conserved (Fig. 2b, c and Supplementary Figs. 5,6). Notably, it is reported that mutation of R1094Q, which locates on the interface between NALCN and FAM155A (Fig. 2e), leads to the disordered respiratory rhythm with central apnea 27 , suggesting the importance of these inter-subunit interactions.
The pore domain of NALCN. Along the ion permeation pathway, negatively charged residues are broadly distributed. They may function to attract sodium ions for permeation (Fig. 3a).
The calculated pore profile reveals two constrictions within the transmembrane domain (Fig. 3b, c). The constriction close to the extracellular side is the high field strength (HFS) layer of the selectivity filter 28 . At this layer, the side chains of E280 (domain I), E554 (domain II), and K1115 (domain III) point to the center of the pore (Fig. 3c-f), which is similar to Na V or Ca V (Fig. 3g). Unexpectedly, we found the 1387-1390 region of domain IV has two possible alternative conformations, both of which were modeled in the allowed region in the Ramachandran plot validated by Phenix 29 (Supplementary Fig. 7e-h). In one conformation (E1389-I), E1389 makes electrostatic interaction with R1384 on the P1 helix of domain IV and makes hydrogen bonding with Y287. In the other conformation (E1389-II), E1389 interacts with Y1431 and the main chain of G1388 via hydrogen bonds ( Supplementary Fig. 7i). Although the side chain of E1389-II and adjacent D1390 point to the direction of the ion permeation pathway, the side chain of D1390 is above the HFS layer while the E1389-II is below the HFS layer (Fig. 3f, g). This is in agreement with recent findings that the sodium current of NALCN can be blocked by extracellular calcium at physiological concentration and mutations of D1390A has a more profound effect than E1389A on extracellular calcium block 1 . Therefore, there are no side chains on domain IV that is at the corresponding position of the highly negatively charged Glu residue of domain IV in Ca V at the HFS layer, such as the E1323 on Ca V 1.1 (PDB ID: 5GJV, Fig. 3f). Therefore, the arrangement of charged residues at the HFS layer of NALCN selectivity filter (EEK_, − − + 0) is similar to Na V (DEKA, − − + 0) but different from Ca V (EEEE, − − − − ) (Fig. 3g), providing a plausible explanation of the sodium selectivity of NALCN. The conformation and structural arrangement of E1389 and D1390 in NALCN reported here were not previously observed in other Na V or Ca V channels. To further explore the roles of E1389 and D1390 on the ion permeation, we mutated each of them into alanine and measured the ion selectivity of the mutants. We found mutations of E1389A slightly lowered the P Na /P K and increased P Na /P Cs , while D1390A had no effect on P Na /P K but markedly increased P Na /P Cs (Fig. 3h), suggesting although the side chains of both E1389 and D1390 are not on the HFS layer, they may also influence ion permeation through maintaining the local chemical environment. We did not observe strong ion densities inside the pore, indicating sodium ions bind weakly to the pore of NALCN in our sample preparation condition. In comparison with the structure of human NALCN-FAM155A complex (EMD-22203 and PDB ID: 6XIW) 18 ( Supplementary Fig. 7a), we found, the extra density of the putatively modified Y287 in 6XIW 18 is contributed by the guanidine group of R1384 in our model, evidenced by the strong connective side-chain density of R1384 in our high-resolution map ( Supplementary Fig. 7b-d). We noticed their protein sample was supplemented with 3 mM CaCl 2 18 , while our sample was not. Whether these structural differences were due to distinct sample preparation conditions awaits further investigations. At the level of the central vestibule, the S6 helices of all of the four domains show π-bulges in the middle (Fig. 3i). It is suggested that the α-π helix transition is important for the gating of some channels 30    tightly sealed by several hydrophobic residues on S6 helices (Fig. 3b, c, j), indicating the current structure represents a non-conductive state. Below the gate, the intracellular exit of the ion permeation pathway is blocked by III-IV linker at one side but remains laterally open towards the direction of domain III (Fig. 3k). We found many disease mutations of NALCN accumulate in the pore, around the bundle crossing region, suggesting they might exert their pathological functions by modulating NALCN gating directly (Fig. 1e, f).
Conformations of degenerated voltage sensors at depolarizing membrane potential. VS senses changes of membrane potential and drives the gating movement of voltage-gated ion channels. Previous studies showed that the consecutive positive gating charges reside on one face of the S4 helix. On the opposing S2 helix, negative or polar residues are separated by an aromatic residue (Phe or Tyr) and followed by a positively charged residue, forming the E1-X2-F3-E4-R5 structural motif. Together with a negatively charged residue on S3 helix, F3 and E4 on S2 form the so-called charge-transfer center (CTC) at the bottom of VS 31 . The S4 helix of functional VS is in the "down" conformation due to the negative voltage at resting membrane potential, while at depolarizing potential, the S4 helix moves "up" in response to the changes of electrostatic field 32,33 . In contrast to the canonical voltage-gated ion channels, some of the voltage sensors of NALCN are degenerated due to mutations of key residues in either chargetransfer center on S2 or gating charges on S4. However, NALCN is modulated by voltage 1 . Therefore, the structure of NALCN provides a unique opportunity to explore the conformations of degenerated VS at 0 mV, a depolarizing state. The voltage sensor of domain I (DI-VS) has a consensus S2 helix. On the S4 helix, positively charged residues of R2, R3, and R5 are conserved, while the middle R4 is replaced by Ile (I149) (Fig. 4a). We observed R2 (R143) interacts with E128 on S3, R3 (R148) interacts with E1 (D75) on S2, and R5 (R152) interacts with E4 (E85) on S2. The Cα of F3 (Y82) is at a level similar to the Cα of R4 (I149) (Fig. 4b). For reference, the Cα of F3 (Y168) in DI-VS of Na V 1.4 in depolarizing state (PDB ID: 6AGF) is at a similar level to R3 (R225), suggesting S4 of NALCN DI-VS is even "upper" than Na V 1.4 ( Supplementary Fig. 9a). The conserved structural features and depolarized conformation of DI-VS suggest it is a functional voltage sensor. Indeed, recent studies showed wild-type NALCN responses to voltage pulse at holding potential of 0 mV but the R146Q + R152Q (R3 + R5) mutant of DI-VS does not 1 Fig. 4 Voltage sensors of NALCN at the depolarized state. a Sequence alignment of S2 and S4 segments. The sequences of rat NALCN, human NALCN, fruit fly NCA, nematode NCA1, NCA2, human Ca V 3.1, rabbit Ca V 1.1, human Na V 1.2, and human Na V 1.4 were aligned using PROMALS3D 53 . Residues forming the charge-transfer center are shaded in gray. The conserved negative or polar E1 and E4 on S2 segments, F3 on S2 segments, and positive R1-R6 on S4 segments are highlighted in red, brown, and blue, respectively. b Structures of four VSDs in NALCN. Gating charges and charge-transfer center residues are shown as sticks. 3 10 helices are colored in pink. adopts a helical structure which is parallel to the membrane plane. However, the N terminus of S4-S5 linker of NALCN DI shows a loop-like structure while its C terminus is fused with S5 to form an exceptionally long S5 helix (Supplementary Fig. 9b). Moreover, the interaction between DI-VS and the pore of domain II is distinct from other VSs ( Supplementary Fig. 9c), explaining why the DI-VS is off-axis (Fig. 1d). These structural observations indicate DI-VS of NALCN might convey the voltage signal to the pore in a manner that is distinct from other VSs.
DII-VS has an intact CTC and three consecutive gating charges on S4 (R3-R5) and the Cα of F3 (F429) is at a similar level to Cα of R3 (R481), which is similar to DII-VS of Nav1.4 (PDB ID: 6AGF) ( Fig. 4 and Supplementary Fig. 9a). These observations suggest DII-VS is a functional VS. Correlating with this, functional studies showed mutation of R3 (R481) together with one of R4 (R484) or R5 (K487) render NALCN unresponsive to pulse at 0 mV 1 .
Although DIII-VS of NALCN has three conserved consecutive gating charges on S4 (R3-R5), it has a defective CTC on S2 and the essential aromatic residue at F3 is replaced by M929 (Fig. 4a). This is consistent with functional studies showing neutralizing mutation of all the gating charges on S4 (R989Q + R992Q + R995Q) has little effect on the voltage modulation of NALCN channel 1 . We observed S4 residues above R3 (R989) adopt α helical structure while residues below R3 form a 3 10 helix. R3 interacts with E906 on S1-S2 linker and E1 (E922). R4 interacts with E1 (E922) and F925. R5 is stabilized by interactions with E4 and D958 on S3. The Cα of F3 (M929) is at the same level as Cα of R5 (R995), which is similar to Nav1.4 (PDB ID: 6AGF) ( Fig. 4b and Supplementary Fig. 9a), suggesting DIII-VS of NALCN is in an "up" conformation, although its CTC is defective.
In contrast to the DIII-VS, which has a defective CTC but three conserved gating charges, the DIV-VS of NALCN has conserved CTC but only two gating charges R4 (R1310) and R6 (K1317) on S4. Moreover, the majority of S4 adopts an α helical structure on which the spatial arrangement of gating charges is not optimal for movements upon membrane potential changes, indicating DIV-VS is defective. In agreement with this, mutation of R1310Q has no effect on the voltage sensitivity of NALCN 1 . Notably, it is reported the W1287L mutation on S3 of DIV-VS can leads to the loss of currents and IHRRF 7,34 (Fig. 1f), probably by affecting protein folding or stability. Intracellular domains. At the intracellular side of domain III, S6 III helix is connected to the well-ordered III-IV linker via two sharp turns at G1164 and L1168. These structures reroute the intracellular exit for sodium towards the direction of domain III (Fig. 3c). There is an extensive polar interaction network between III-IV linker and the intracellular side of domain I and II, involving E327 and R329 of domain I, E603, K609, and Q613 of domain II, Q1172, and R1173 of III-IV linker. Residues following this turn of III-IV linker fold into the III-IV helix which binds below domain IV. R1174 on this helix interacts with E327 on S6 I , while R1181 interacts with the main-chain carbonyl group of S1451, L1452, and Y1454 (Fig. 5a). The gain-of-function R1181Q mutation on this helix causes intellectual disability, episodic and persistent ataxia, arthrogryposis, and hypotonia 20 . Moreover, T1165P mutation can lead to CLIFAHDD syndrome, further emphasizing the important function of III-IV linker (Fig. 1f). The position of III-IV linker in NALCN shares similarity to Ca V 1.1 (PDB ID: 5GJV) 35 and Na V PaS (PDB ID: 5X0M) 36 channels but distinct from Na V 1.2 (PDB ID: 6J8E) 37 , Na V 1.4 (PDB ID: 6AGF) 38 , Na V 1.5 (PDB ID: 6UZ3) 39 , and Na V 1.7 (PDB ID: 6J8G) 40 (Supplementary Fig. 10a). NALCN does not have the IFM motif on the III-IV linker, which is involved in the fast inactivation of Na V . This is in accordance with the fact that NALCN does not show fast inactivation during depolarization.
The III-IV helix, together with the intracellular surface of NALCN forms a platform for the binding of CTD. The resolved portion of NALCN CTD consists of five helixes (C1-C5) and forms a calmodulin-like domain that packs below the III-IV linker (Fig. 5b). The NALCN CTD shares structural similarity to those of Ca V 1.1 (PDB ID: 5GJV) 35 and Na V PaS (PDB ID: 5X0M) 36 , except the C-terminal C5 helix orients differently ( Supplementary  Fig. 10b, c). We found a strong extra helical density bound in a groove of CTD, which is cradled by C1-C4 helices. The excellent map density of this helix matches the amino acid sequence of a fragment of the II-III linker (751-765) ( Supplementary Fig. 10d and Fig. 5b). Furthermore, GST pull-down results using either protein expressed in HEK293 cells or protein purified from E. coli confirmed the interactions between 751-765 fragment and the CTD of NALCN ( Supplementary Fig. 10e, f). Therefore, we tentatively named this helix (751-765) as CTD Interacting Helix (CIH). Based on this model, the hydrophobic side chains of the Nterminal I753 and L754 of CIH are inserted into a hydrophobic pocket formed by C1-C4 (Fig. 5c). The positively charged R761, R764, and R765 residues on the C terminal of CIH form a complex interaction nexus with the side chains of D1543 on C4-C5 loop, E1557 and E1560 on C5, and several main-chain carbonyl groups on C4-C5 (Fig. 5c). The interactions between II-III linker and CTD are not observed previously in other Na V or Ca V structures. CIH in our model was built as 1589-1602 of  18 , probably due to the poor local map quality.

Discussion
The high-resolution structure of the NALCN-FAM155A complex presented here shows the architecture of the NALCN channel core components, reveals the structure of FAM155 CRD, and depicts the detailed inter-molecular interface between NALCN and FAM155A subunits. More importantly, the non-canonical configuration of selectivity filter suggests the mechanism of sodium permeation and calcium block. Furthermore, the asymmetric spatial arrangement of both functional and degenerate voltage sensors provides mechanistic clues for the voltage sensitivity of NALCN. In addition, our structure also provides a template for drug discovery targeting the NALCN channel in related human diseases.
Electrophysiology. Constructs containing desired point mutations of rNALCN were generated by Quick Change. HEK293T cells were co-transfected with rNALCN-CGFP, mFAM155A-FLAG, mScarlet-mUNC80, and HA-mUNC79 plasmids (in a modified BacMam expression vectors 41 ) at a ratio of 2:1:1:1 using Lipofectamine 3000 (ThermoFisher Scientific) and incubated for 24 h before recording. Patch electrodes were pulled with a horizontal microelectrode puller (P-1000, Sutter Instrument Co, USA) to a resistance of 2-3 MΩ. Whole-cell patch clamps were performed using an Axon-patch 200B amplifier (Axon Instruments, USA), and data were collected with pClamp 10 software (Axon Instruments, USA) and an Axon Digidata 1550B digitizer (Axon Instruments, USA). Pipette solution containing (mM): 10 HEPES (pH 7.2, NaOH), 136 NaCl, 10 NaF, 5 EGTA, 2 Na 2 ATP and bath solution containing (mM): 10 HEPES (pH 7.4, NaOH), 150 NaCl, 30 glucose. In some experiments, 10 µM GdCl 3 (Sigma) was added to the bath solution to block the inward current, as indicated. For ion selectivity experiments, the bath solution was exchanged to 10 mM HEPES (pH 7.4, XOH or HCl), 150 mM XCl (X = Na + , K + , Cs + ), or NMDG and 30 mM glucose by local perfusion equipment (MPS-2, InBio). The membrane potential was increased from holding potential (0 mV) to +80 mV for 0.5 s to activate the channel and then stepped to various testing potentials (−80 to +80 mV). The peak of tail currents was used to generate I-V curves for the determination of reversal potentials (E rev ). When Na + in the bath solution, the measured E rev is close to the theoretical E rev (0 mV). To improve the accuracy of our measurements, patches with loose seals (<−100 pA at −80 mV when exposing to NMDG + bath solution) or small currents (<1.5 nA at +80 mV when exposing to Na + bath solution) were discarded. The liquid junction potentials were calculated using pClamp 10 (−4.2 and −4.7 mV for extracellular K + and Cs + , respectively) and corrected to calculate the reversal potentials. Signals were acquired at 5 kHz and low-pass filtered at 1 kHz. The ion permeability ratios were calculated with the following equation: where E rev , F, R, and T are the reversal potential, Faraday constant, gas constant, and absolute temperature, respectively. The data were processed with Microsoft Excel and GraphPad Prism 6.
Protein expression and purification. Mouse FAM155A was cloned from the cDNA that was reverse transcribed from the total mRNA of the mouse brain. The cDNAs of rat NALCN and mouse FAM155A were cloned into a modified BacMam expression vector with C-terminal GFP-strep and FLAG-tag, respectively 41 . The baculoviruses were produced using the Bac-to-Bac system and amplified in Sf9 cells 42 . For protein expression, HEK293F cells cultured in Freestyle 293 medium at a density of 2.5 × 10 6 ml −1 were infected with 6% volume of NALCN P2 virus and 4% volume of FAM155A P2 virus. 10 mM sodium butyrate was added to the culture 12 h post-infection and transferred to a 30°C incubator for another 48 h before harvesting. Cells were collected by centrifugation at 3990×g (JLA-8.1000, Beckman) for 10 min, and washed with 50 mM Tris (pH 7.5), 150 mM NaCl, 2 μg/ml aprotinin, 2 μg/ml pepstatin, 2 μg/ml leupeptin, flash-frozen, and storage at −80°C.
EM sample preparation. Holey carbon grids (Quantifoil Au 300 mesh, R 0.6/1) were glow-discharged by Solarus advanced plasma system (Gatan) for 120 s using 25% O 2 and 75% Ar. Aliquots of 3 μl concentrated protein sample were applied on glow-discharged grids and the grids were blotted for 2 s before plunged into liquid ethane using Vitrobot Mark IV (ThermoFisher Scientific).
Cryo-EM data acquisition. Cryo-grids were firstly screened on a Talos Arctica electron microscope (ThermoFisher Scientific) operating at 200 kV with a K2 Summit direct electron camera (ThermoFisher Scientific). The screened grids were subsequently transferred to a Titan Krios electron microscope (ThermoFisher Scientific) operating at 300 kV with a K2 Summit direct electron camera and a GIF Quantum energy filter set to a slit width of 20 eV. Images were automatically collected using SerialEM in super-resolution mode at a nominal magnification of ×130,000, corresponding to a calibrated super-resolution pixel size of 0.5225 Å with a preset defocus range from −1.5 to −1.8 μm. Each image was acquired as a 7.12 s movie stack of 32 frames with a dose rate of 7.15 e − /Å 2 /s, resulting in a total dose of about 50.9 e − /Å 2 .
Image processing. The image processing workflow is illustrated in Supplementary Fig. 3. A total of 5,470 super-resolution movie stacks were collected. Motioncorrection, two-fold binning to a pixel size of 1.045 Å, and dose weighting were performed using MotionCor2 43 . Contrast transfer function (CTF) parameters were estimated with Gctf 44 . Micrographs with ice or ethane contamination and empty carbon were removed manually. A total of 2,226,376 particles were autopicked using Gautomatch from 5,428 micrographs. All subsequent classification and reconstruction were performed in Relion 3.1 45 unless otherwise stated. Two rounds of reference-free 2D classification were performed to remove contaminants, yielding 567,397 particles. The particles were subjected to 57 iterations K = 1 global search 3D classification with an angular sampling step of 7.5°to determine the initial alignment parameters using the initial model generated from cryoSPARC 46 . For each of the last seven iterations of the global search, a K = 6 multi-reference local angular search 3D classification was performed with an angular sampling step of 3.75°and a search range of 30°. The multi-reference models were generated using reconstruction at the last iteration from global search 3D classification low-pass filtered to 8 , 15, 25, 35, 45, and 55 Å, respectively. The classes that showed obvious secondary structure features were selected and combined. Duplicated particles were removed, yielding 292,877 particles in total. These particles were subjected to additional three rounds of multi-reference local angular search 3D classification using the same parameters. Each round had 30 iterations, and the particles corresponding to good classes from the last 15 iterations of each round were combined, and duplicated particles were removed, yielding 201,654 particles. These particles were subsequently subjected to local search 3D auto-refinement, which resulted in a map with a resolution of 3.0 Å. In order to further clean up the data set, three rounds of random-phase 3D classification 47 were performed with K = 2, an angular sampling step of 1.875°, and a local search range of 10°, using two reference models, which are the model obtained from previous refinement and the phase-randomized model. Phaserandomized models were generated from the model obtained from the previous refinement using randomize software (from the lab of Nikolaus Grigorieff) for phase-randomization beyond 40, 30, and 20 Å for the first, second, and third round of 3D classification. For random-phase 3D classification, each round had 30 iterations, and the particles corresponding to good class from the last 15 iterations of each round were combined, and duplicated particles were removed, yielding 135,043 particles, resulting in a 2.9 Å map. CTF refinement was then performed with Relion 3.1, the resolution was improved to 2.8 Å. The particles were then reextracted, re-centered, and re-boxed from 240 pixels to 320 pixels. After 3D autorefinement, the resolution reached 2.7 Å.
All of the resolution estimations were based on a Fourier shell correlation (FSC) of 0.143 cutoff after correction of the masking effect. B-factor used for map sharpening was automatically determined by the post-processing procedure in Relion 3.1 45 . The local resolution was estimated with Relion 3.1 45 .
Model building. The initial model of NALCN alone was generated by SWISS-MODEL based on the structure of rabbit Ca v 1.1-α1 subunit (PDB 5GJV) 35 . The model was then fitted into the cryo-EM map using Chimera 48 and rebuilt manually using Coot 49 . Model of FAM155A was built manually using Coot 49 . Model refinement was performed using phenix.real_space_refine in PHENIX 29 . Images were produced using Pymol and Chimera 48 .
For GST pull-down assay using purified proteins from E. coli ( Supplementary  Fig. 10f), CTD was cloned into a pGEX-6P-1 vector, CIH was cloned into pET-21b vector with C-terminal GFP-His tag, GFP was cloned into pET-21b vector with Nterminal His tag and empty pGEX-6P-1 vector was used to express GST. Overexpression of GST-CTD and CIH-GFP was induced in E. coli NiCo21 (DE3) with 0.2 mM isopropyl-β-D-thiogalactoside (IPTG) when the cell density reached OD 600 = 0.6 for 18 h at 16°C, and overexpression of GFP and GST was induced with 1 mM IPTG for 4 h at 37°C. The cells were collected and sonicated in lysis buffer containing 50 mM Tris (pH 7.5), 150 mM NaCl, 2 μg/ml aprotinin, 2 μg/ml pepstatin, 2 μg/ml leupeptin, and 1 mM PMSF. Then the cell debris was removed by centrifugation at 30,966×g (JA-25.50, Beckman) for 30 min. For purification of GST and GST-CTD, the supernatants were loaded onto Glutathione Sepharose 4B and washed with 20 mM Tris (pH 7.5), 150 mM NaCl and eluted with 50 mM Tris (pH 8.0), 50 mM NaCl and 10 mM reduced glutathione. For purification of GFP and CIH-GFP, the supernatants were loaded onto TALON (Clontech) and washed with W1 buffer (20 mM Tris (pH 7.5), 300 mM NaCl) and W2 buffer (20 mM Tris (pH 7.5), 150 mM NaCl, 10 mM imidazole) and eluted with 250 mM imidazole (pH 8.0) and 50 mM NaCl. All proteins were further purified with an anionexchange column (HiTrap Q HP, GE Healthcare) and the peak fractions were pooled for GST pull-down assay. The purified proteins were mixed (10 µM for each) as indicated ( Supplementary Fig. 10f) at room temperature for 30 min and loaded onto Glutathione Sepharose 4B. The column was washed with 20 mM Tris (pH 7.5) and 150 mM NaCl for 5 times and eluted with 100 mM Tris (pH 8.0), 150 mM NaCl, and 20 mM reduced glutathione. Proteins were separated with 12% SDS-PAGE and visualized by GFP fluorescence (2500B, Tanon) and coomassie blue staining.

Data availability
Data supporting the findings of this manuscript are available from the corresponding author upon reasonable request. The cryo-EM map of the NALCN-FAM155A complex has been deposited in the EMDB under code EMD-30470. The atomic coordinate of the NALCN-FAM155A complex has been deposited in the PDB with accession code 7CU3. PDB entries (6AGF, 5GJV, 5X0M, 6J8E, 6UZ3, 6J8G, and 6XIW) used in this study were downloaded from Protein Data Bank. Source data are provided with this paper.