Cryo-EM structure of type 1 IP3R channel in a lipid bilayer

Type 1 inositol 1,4,5-trisphosphate receptor (IP3R1) is the predominant Ca2+-release channel in neurons. IP3R1 mediates Ca2+ release from the endoplasmic reticulum into the cytosol and thereby is involved in many physiological processes. Here, we present the cryo-EM structures of full-length rat IP3R1 reconstituted in lipid nanodisc and detergent solubilized in the presence of phosphatidylcholine determined in ligand-free, closed states by single-particle electron cryo-microscopy. Notably, both structures exhibit the well-established IP3R1 protein fold and reveal a nearly complete representation of lipids with similar locations of ordered lipids bound to the transmembrane domains. The lipid-bound structures show improved features that enabled us to unambiguously build atomic models of IP3R1 including two membrane associated helices that were not previously resolved in the TM region. Our findings suggest conserved locations of protein-bound lipids among homotetrameric ion channels that are critical for their structural and functional integrity despite the diversity of structural mechanisms for their gating. 3D structure of full-length rat type 1 inositol 1,4,5-trisphosphate receptor reconstituted in lipid nanodisc is determined using single-particle cryo-electron microscopy. The study suggests conserved locations of protein-bound lipids among structurally diverse, homo-tetrameric ion channels.

I nositol 1,4,5-trisphosphate receptors (IP 3 Rs) are tetrameric intracellular cation channels ubiquitously expressed in mammalian cells and located predominantly in the endoplasmic reticulum (ER) membranes. IP 3 Rs are activated by inositol 1,4,5trisphosphate (IP 3 ), a second messenger produced through hydrolysis of phosphatidylinositol 4,5-bisphosphate (PIP 2 ) by phospholipase C, which is activated in response to diverse cellular stimuli such as hormones, growth factors, neurotransmitters, neurotrophins, odorants, light, etc. [1][2][3] . The resulting Ca 2+ release from the ER via IP 3 R channels elevates the cytoplasmic free Ca 2+ concentration, a signal that triggers markedly diverse cellular actions, ranging from contraction to secretion, from proliferation to cell death. The most notable feature of IP 3 R channels is biphasic regulation 1,2 by Ca 2+ , such that at concentrations in the nanomolar range (resting levels in cellular cytosol) Ca 2+ stimulates IP 3 R-mediated Ca 2+ release from intracellular stores. In contrast, at higher micromolar concentrations of Ca 2+ IP 3 Rmediated Ca 2+ release is inhibited by Ca 2+ . To understand how IP 3 R channels convey their gating through the interplay of its two primary agonists, IP 3 and Ca 2+ , structures of IP 3 Rs in both the apo-and ligand-bound states have been determined at nearatomic resolutions by single-particle electron cryo-microscopy (cryo-EM). These structures were determined in a detergentbased aqueous environment, allowing for channel solubility [4][5][6][7] . However, ion channels reside in biological membranes, and lipids are proven to play important physiological roles in maintaining their structural-functional integrity. Although the latter structures described the IP 3 R architecture and the structural determinants for its function, the structural basis for how IP 3 Rs function in a physiological lipid membrane environment and how lipids affect intrinsic channel protein tasks, such as ligand-binding and gating, remain unknown.
In this study, we present an atomic model of the full-length rat IP 3 R1 channel that was built based on the 3.30 Å resolution cryo-EM map of the channel embedded in a lipid nanodisc. In addition, the structure of IP 3 R1 solubilized with detergent in the presence of phospholipids was solved to 2.96 Å resolution also using the single-particle cryo-EM approach. Structural comparisons of the channels visualized in these two different milieus in a ligand-free state provide insights into IP 3 R1-lipid interactions enabling a deeper understanding of IP 3 R channel function in a physiologically relevant membrane environment.

Results
Overall structure of IP 3 R1 in a lipid bilayer. To study the structure in a lipid bilayer environment, we first optimized the purification of full-length IP 3 R1 from rat cerebellum utilizing our previously described procedure that produces functional channels, as demonstrated by IP 3 -induced Ca 2+ fluxes after reconstitution into lipid vesicles 8 . In this study, the cerebellar microsomal membranes were solubilized with Lauryl Maltose Neopentyl Glycol (LMNG) in the presence of phospholipids (see "Methods" section). We next reconstituted LMNG-solubilized IP 3 R1 into lipid nanodiscs composed of MSP1E3D1/POPC, which form nanodiscs of~12-14 nm diameter 9 and are sufficient to accommodate the IP 3 R1 transmembrane (TM) domains without imposing spatial constraints. Importantly, embedding IP 3 R1 into nanodiscs and protein purification was essentially a one-step procedure using immunoaffinity chromatography (Supplementary Fig. 1a; see "Methods" section), minimizing exposure of the solubilized channel protein to detergent and avoiding further displacement of lipid molecules while easily reconstituting the purified channel particles in a near-native lipid environment.
Both IP 3 R1 reconstituted in lipid nanodiscs (IP 3 R1-ND) and IP 3 R1 solubilized in LMNG (IP 3 R1-LMNG) preparations were analyzed by cryo-EM under similar conditions in the presence of EGTA and without the addition of any channel-specific ligands. Corresponding cryo-EM images showed monodisperse samples with particles evenly distributed across the hole (Supplementary Figs. 1b and 2a). Side views of the channel exhibit a characteristic "screw-shape" displaying the TM and cytoplasmic (CY) regions and are easily recognizable. 2D class-averages generated from both samples illustrated different particle orientations with many features in both the CY and TM domains (TMDs) (Supplementary Figs. 1c and 2b). 3D reconstructions were calculated for both data sets at an average nominal resolution of 3.30 Å for IP 3 R1-ND and 2.96 Å for IP 3 R1-LMNG using the "gold standard" Fourier shell correlation (FSC) = 0.143 criterion (Supplementary Figs. 1f, 2e; Table 1; see "Methods" section). However, the local resolution in most of the TM domains and some of the CY domains reached to better than 2.8 Å resolution that enabled reliable assignment of the side chains in these regions ( Supplementary Figs. 1g and 2f).
The overall architecture of both IP 3 R1-ND and IP 3 R1-LMNG are nearly identical to each other and to the previously determined cryo-EM structures of the IP 3 R1 channel in a  elements of the TM region enabling their unambiguous determination. The MA1-MA2 structure is positioned~14 Å from the center of the TM1-TM4 helical bundle and tilted~24°a way from it ( Supplementary Fig. 6d, e). Notably, in the IP 3 R1-LMNG structure, the MA1 helix proximate to the lumenal side of the membrane has a 14 Å displacement from MA1 in IP 3 R1-ND. The cryo-EM map of IP 3 R1-ND clearly shows densities consistent with the presence of a lipid nanodisc that forms a round-shaped matrix surrounding the entire TM region and delineates lipid bilayer boundaries with well-defined cytosolic and lumenal leaflets (Fig. 1a-c). The lipid nanodisc measures about 30 Å high × 130 Å wide, which is in agreement with previously reported nanodisc dimensions 9 . We observed several "hairpinshaped" and elongated densities, tightly adhering to the TM domains and most often appearing perpendicular to the membrane plane (Fig. 1b-e and Supplementary Movie 1). Notably, these well-ordered densities were observed in both the IP 3 R1-ND and IP 3 R1-LMNG maps ( Supplementary Fig. 5b, c). The shape of these densities is consistent with phospholipid molecules. These lipid-like densities were putatively modeled as phosphatidylcholine (PC), the major component (>50%) of ER membranes 10 , which was added during protein solubilization and nanodisc formation (see "Methods" section).
Ion-permeation pathway in IP 3 R1-lipid complexes. Consistent with the previous studies, the ion-permeation pathway in IP 3 R1 is formed around the central axis of the tetrameric channel assembly and defined by the pore-lining TM5 and TM6 helices, the short reentrant pore helix (P-helix), and pore loop containing the selectivity filter (SF) (Fig. 2a, c) 4,5 . The calculated solventaccessible ion-conduction profile shows that the residues F2586 and I2590 form two constrictions with the distances of 1.4 and 1.4 Å in IP 3 R-ND and 1.4 and 3.6 Å in IP 3 R-LMNG, respectively ( Fig. 2a, b). This pore conformation is remarkably similar to the pore architecture in the previous structures of apo-IP 3 R1 and is designated as a closed pore conformation 4,5 . Notably, seven lipid molecules per subunit were revealed in the TM region of IP 3 R1, regardless of whether the channel was reconstituted in lipid nanodiscs or solubilized in LMNG (Fig. 3a, c and Supplementary Figs. 5b, c, 6d, e). Moreover, the structural comparisons of the density maps and models of IP 3 R1-ND and IP 3 R1-LMNG revealed that lipid molecules are tightly associated with the TM domains at nearly identical locations in both structures. The locations of ordered lipid densities mimic the lipid bilayer and form extensive interfaces with the TMDs appearing in inter-and intra-subunit crevices. The lipid buried surface area accounts for 10,307 Å 2 and 9,866 Å 2 in the TMDs of IP 3 R1-ND and IP 3 R1-LMNG, respectively. The lipid-protein interfaces exhibit hydrophobic packing that appears to maintain a seal across the membrane. Due to the domain-swapped architecture of the TMDs, some lipid-binding sites involve multiple IP 3 R1 subunits.
Lipid densities (L1-L3) were observed in the hydrophobic crevices formed close to the lumenal vestibule of the ion- Fig. 2 The ion-permeation pathway of IP 3 R1 in nanodisc. a Solvent-accessible pathway along the IP 3 R1-ND pore mapped using the program HOLE 54 . A series of residues within the ion-conduction pathway are labeled. Dashed line box indicates the zoomed-in region in c. b Comparison of the pore dimensions for IP 3 R1-ND (pink), IP 3 R1-LMNG (purple), and IP 3 R1-CHAPS (PDB ID: 6MU2; dashed gray line). c A wire representation of the SF in IP 3 R1-ND; two opposing subunits are viewed along the membrane plane; narrowest distances between Cα atoms (G2546) or side-chain atoms along the SF are indicated. d The surface electrostatic potential along the ion-permeation pathway. The left panel shows a slice through the channel pore along the four-fold axis with the location of slices perpendicular to the four-fold axis (right panels) indicated by dotted lines. Panels on the right show slices through the pore at E2470, D2551, G2546, and R2597 viewed from the cytosolic side along the symmetry axis. ARTICLE COMMUNICATIONS BIOLOGY | https://doi.org/10.1038/s42003-021-02156-4 permeation pathway of IP 3 R1 (Fig. 3a, c). One lipid molecule (L1) is found at a junction between three IP 3 R1 subunits: the central pore-lining TM6 and P-helix of two neighboring subunits and the MA helices from another subunit (Fig. 3c). The intra-subunit binding site for L2 is formed by the hydrophobic and aromatic residues of TM5 and TM6 helices with one acyl chain contacting TM6 and the other acyl chain intercalating between the lumenal ends of TM5 and TM6 (Fig. 3c). L3 fills the lumenal crevice between the TM4 and TM5 helices of two neighboring subunits. H2409 is in a position to potentially form a salt bridge with the L3 headgroup (Fig. 3c).
It appears that the lipid environment stabilizes the lumenal vestibule of the channel, so that the P-helix and SF, defined as 2546 GGGVGD 2551 , are well resolved in both the IP 3 R1-ND and In the apo-state, Ca 2+ ions from the ER lumen can access the channel's lumenal vestibule and encounter negatively charged residues (E2525, E2470) lining the entrance to the ion-conduction pathway. These residues contribute to an electronegative environment that may hold or stabilize cations in the aqueous vestibule before they traverse across the lipid bilayer (Fig. 2d). In the nonconducting (apo-) state, the SF is defined by the side chains of D2551 and S2545 along with the Cα of G2546 constituting the narrowest point of the SF (Fig. 2c). The distance across the SF is 13 Å at D2551 and~9.5 Å at G2546. This implies that the selectivity filter in both the IP 3 R1-ND and IP 3 R1-LMNG does not constitute a barrier that would preclude the passage of fully hydrated Ca 2+ ions 11 . In previous work, by removing the carboxylic acid group of the aspartic acid residue, the D2551A mutation creates a pore dead channel unable to translocate Ca 2+ ions, yet the conservative mutation, D2551E, alters the channel selectivity for divalent cations to favor monovalent ions 12 . This suggests that the D2551 side chains from each subunit could form a single Ca 2+ binding site that might be responsible for the weak selectivity of IP 3 R1 for Ca 2+ over Na + or K + (Fig. 2c). Moreover, a mutation in the SF of the corresponding glycine residue to arginine of human IP 3 R1 results in Spinocerebellar Ataxia Type 29 and Gillespie Syndrome (Fig. 5) 13 and in vitro G2546A mutations eliminate Ca 2+ flux 14 . Interestingly, the highly conserved, positively charged R2544 is positioned at the C-terminal end of the P-helix, which would not seem conducive to the rapid movement of cations through the channel. However, the sidechain of R2544 points towards the TM6 helix of the neighboring subunit and is positioned to form a salt bridge with D2570 ( Supplementary Fig. 7d). It was found that mutating this charged residue to alanine alters Ca 2+ regulation of the receptor 14 . Specifically, R2544A channels appeared to be activated normally by low concentrations of Ca 2+ , but exhibit enhanced inhibition by high Ca 2+ when compared with wild-type channels. However, the R2544A mutation has little effect in 45 Ca 2+ flux assays 12 , suggesting that the R2544 side-chain projects away from the permeation path, which is in good agreement with our IP 3 R1 structures. Altogether, these studies suggest that mutations within the P-helix resulting in structural changes in the SF might regulate ion flow through IP 3 R channels. A similar regulatory mechanism by conformational changes in SF has been observed in K + channels 15,16 .
Three lipid molecules (L4-L6) are found lining the amphipathic TM4-5 helix that runs parallel to the membrane plane at the cytosolic leaflet of the lipid bilayer and connects the peripheral bundle of the TM1-TM4 helices with the central pore-lining TM bundle. The phospho-head groups of the lipids are located at the boundary of lipid bilayers, in a position to potentially interact with polar residues from the channel. L4 occupies an intra-subunit hydrophobic cleft formed at the interface between the TM3, TM4, and TM4-5 helices (Fig. 3c). Both hydrocarbon chains of L4 appear to nestle among the hydrophobic side chains of TM3 and TM4 while the bulkier density of the polar head group is located next to the N-terminus of the TM4-5 horizontal helix at the lipid/water interface. Mutations of charged residues in the TM4-5 helix (D2419A, R2423A, E2424A, E2425A) have been found to inhibit channel function, pointing to the importance of this domain for the channel activation 17 . Proper lipid integration into the TM4-5 helix may play a critical mechanistic role in the channel gating. Structures of several other tetrameric cation channels (such as TRPV1, TRPV3, NOMPC, and K V 1.2), have also shown lipid binding at this junction and this lipid plays a regulatory role in K v 1.2 channel function ( Supplementary Fig. 8) 18 . The polar head group of another lipid (L5) is positioned near R2436, while its two hydrocarbon tails curve under the TM4-5 helix and continue toward the pore-lining TM5 and TM6 helices of the neighboring subunit (Fig. 3c). Noteworthy, the TM6 helices in the lipid-bound structures exhibit a conformation satisfying a canonical α-helix hydrogen bonding, however, the hydrocarbon tail of L5 is observed near the site where a π-helix bulge in TM6 (residues M2576, I2580) was observed under activating conditions 5 . L6 is located between TM4-5 and TM5 helices with the phospho-head group residing between two arginine residues, R2436 and R2439, which could provide favorable electrostatic interactions between the negatively charged lipid headgroup and the channel protein (Fig. 3c). The two branches of the hydrocarbon tail traverse~2 turns down the TM5 helix to I2442, filling in a hydrophobic pocket formed by TM5 with TM6 of the neighboring subunit.
A long, ovoid density attributed to L7 is found at the intersubunit interface between the TM5 helix and TM1 and TM4 helices from the neighboring subunit (Fig. 3c). Both hydrocarbon tails of L7 are well resolved with one branch of the acyl chain occupying the majority of the hydrophobic pocket formed by TM5, TM1, and TM4. The second hydrocarbon chain fills a hydrophobic pocket between the TM5 and TM6 helices of the neighboring subunits. It is noticeable that the well-defined bulky phospho-head group of L7 runs~6 Å deeper into the cytosolic leaflet of the lipid bilayer than L4-L6. This lipid placement may better execute a stabilizing role on TMDs supporting optimal channel gating.
Allosteric nexus at cytosolic-lipid bilayer interface. Improved resolvability in the density maps was not limited to TM regions, but also extended to the CY regulatory domains. The intervening lateral (ILD) and linker (LNK) domains form a metastable nexus above the channel's cytosolic vestibule that represents the only link between the CY and TM domains and is essential for the channel gating (Fig. 4a) 4,5 . Our previous studies demonstrated that the ILD/LNK assembly undergoes substantial structural rearrangements upon IP 3 R1 activation and is responsible for the transmission of ligand-evoked signals to the pore 5 . Although the ILD/LNK structure of the apo-IP 3 R1 in lipid bilayer is similar to that in detergent [4][5][6][7] , it provides an immense amount of structural details, including unambiguous side-chain placement, revealing putative hydrogen bonding within the domains and coordination of a Zn 2+ ion in its binding pocket 4

.
A unique feature in the superfamily of intracellular Ca 2+ release channels, which includes the closely related IP 3 Rs and RyRs, is the intra-subunit layering of two non-contiguous domains with the C-terminal LNK domain sandwiched between two β-strands and the helix-turn-helix motif of the ILD domain, which connects the ARM3 and TM1 domains ( Fig. 4a and Supplementary Fig. 6a). The ILD/LNK domains are positioned to couple multiple cytosolic regulatory signals to gating motions of the channel 4,5 . Based on our structural analysis of IP 3 R1 in a lipid environment, potential non-covalent interactions can be proposed within the ILD/LNK nexus that might be essential for stabilization of ILD/LNK assembly providing structure-functional integrity of the channel (Fig. 4b) 19 . Specifically, R2618 and E2629 from the LNK domain are in a position to form hydrogen bonds with E2235 and Q2236, respectively, which reside on the ILD of the neighboring subunit. Additionally, K2608 from the cytosolic extension of TM6 is in a position to form a salt bridge with E2235, further linking the ILD/LNK domains with the TMD. The deletion of the nearby residue K2603 (K2563 in human IP 3 R1) at the cytosolic extension of TM6 has been linked to Gillespie Syndrome and causes dysfunctional IP 3 R1 channels that are unable to release Ca 2+ (Fig. 5) 20,21 . This deletion would effectively shorten the length of TM6 likely altering the intermolecular contacts of the ILD/LNK nexus with the TMDs.
Four Zn 2+ -binding sites in the tetrameric channel are defined by a C2H2-like Zn 2+ finger motif (C2611, C2614, H2631, and H2636) in each IP 3 R1 subunit. The cryo-EM maps determined in a lipid environment show strong density peaks (>15 σ) within the binding pocket, which we have assigned as Zn 2+ ions ( Fig. 4c and Supplementary Fig. 9). The Zn 2+ -binding site shows a classic tetrahedral coordination revealing the mechanism of stabilization due to Zn 2+ binding. Zn 2+ binding at the C2H2-like motif appears to be integral to the native IP 3 R1 structure, as purification of the channel was conducted in the presence of cation chelators and no additional Zn 2+ was added. Of note, the stability of the tetrameric structure of IP 3 R is directly tied to sequences within the channel's C terminus, which includes the C2H2-like Zn 2+ finger in the LNK domain 22 .

Discussion
In the present study, we have solved the first structure of the fulllength neuronal IP 3 R1 channel reconstituted in lipid nanodiscs, which provides a native-like lipid bilayer environment suitable for cryo-EM structural characterization of the membrane protein in a soluble, detergent-free form. We have also determined the structure of IP 3 R1 solubilized in LMNG in the presence of phospholipid in order to assess to what extent the detergentsolubilized channel adopts a native structure when compared with the channel embedded in a membrane environment. Our work was motivated by the release of numerous cryo-EM structures of many tetrameric ion channels obtained both in the aqueous detergent-based and lipid-based environments 23 . Based on these studies, it has become evident that the environmental conditions, specifically the lipid bilayer, are responsible for many of the structural differences identified, e.g., the TM pore-lining S6 helix of the TRPV3 channel [24][25][26][27] and the VSLD and peripheral TM helices of the RyR1 channel 28 .
For IP 3 R1, we found that our two structures, with the channel embedded in lipid nanodiscs or solubilized in LMNG in the presence of phospholipid, are virtually identical with only subtle structural differences identified in the CY and TM domains, as well as a displacement of the peripheral membrane-associated helices (Fig. 3a, b and Supplementary Figs. 6b-e, 7a-c). These differences can be accounted for by the embedding medium in cryospecimens (detergent-free vs. detergent-solubilized) and may conceivably be due to the lack of the full lipid shell in IP 3 R1-LMNG structure. Strikingly, both cryo-EM structures of IP 3 R1 presented here revealed well-ordered lipid densities that are immersed at near-identical locations in cavities and clefts formed by the TM domains. However, the IP 3 R1-ND structure likely represents a more complete picture of the lipid topology in IP 3 R1 that includes the nanodisc lipids forming an annular shell around the TM region, which resembles the bilayer structure and mediates communication between the protein and lipid bilayer (Supplementary Movie 1). Notably, several lipid-binding sites in IP 3 R1 are located in intermolecular hydrophobic crevices, which are formed due to the domain-swapped architecture of the TMDs. As such, this allows one lipid to impart structural and functional influence to more than one subunit. The precise molecular identity and origin of protein-bound lipids is uncertain given that they could arise from either the native ER membrane or exogenous phosphatidylcholine added during protein solubilization in detergent and nanodisc formation.
In the IP 3 R1-LMNG map, lipid densities tightly associated with the TM domains were found at similar positions as in IP 3 R1-ND ( Supplementary Fig. 5) suggesting strong lipid-IP 3 R1 interactions at these sites that may represent a specific lipid-protein architectural arrangement in the native lipid bilayer. Similar binding of lipid molecules in the hydrophobic crevices between adjacent subunits has also been shown in other homotetrameric ion channels. We found that the location of the L4-binding site between TM4-5, TM4, and TM3 helices is highly conserved among these ion channels (Supplementary Fig. 8). The interface between ILD and the LNK/TM6 of neighboring subunits overlapped with corresponding EM densities (displayed at 4σ). c Close-up view of Zn 2+ binding site in LNK domain; residues in the C2H2-like Zn 2+ finger are labeled; the density corresponding to Zn 2+ is depicted as magenta mesh, contoured at 15σ.
Our observations led to the conclusions that the presence of protein-bound lipids stabilize the channel architecture, in particular the TM region, which shows higher local resolutions (2.4-3 Å) as compared to the CY domains (3-4.5 Å) ( Supplementary  Figs. 1e-g and 2d-f). In both the IP 3 R1-ND and IP 3 R1-LMNG structures, the conformation of the SF is wide enough to accommodate a fully hydrated Ca 2+ ion, which is consistent with our previous cryo-EM structures of apo-IP 3 R1 determined in detergent 4,5 . These studies suggest that ligand-dependent activation on the channel involves conformational changes in the SF that could provide a plausible mechanism of Ca 2+ recognition by the SF. Furthermore, we found that the pore-lining TM6 forms a continuous α-helical structure in both the lipid-based and detergent environments when the channel is visualized in the absence of activating ligands 4,5 . However, we have shown previously that a π-helix is observed in TM6 in the presence of Fig. 5 Model of disease-associated mutations in TM domains of IP 3 R1. a TM4-TM6 helices (gray, wire representation) from two opposing subunits of IP 3 R1-ND are shown in side view; lipid molecules (orange) are depicted as ball-and-stick models overlapped with corresponding surface representations (mesh). Residues in the rat IP 3 R1 sequence that correspond to the human IP 3 R1 mutations associated with Gillespie Syndrome, microcephaly with pontine and cerebellar hypoplasia, spinocerebellar ataxias, and pontocerebellar hypoplasia are depicted as spheres and colored magenta, pink, blue, and yellow, respectively. b TM4-TM6 helices of four subunits with lipids viewed from the cytosol (left panel) and lumen (right panel) perpendicular to the membrane plane.
activating ligands 5 . This observation raised the possibility that the α-to-π transition in TM6 underlies the gating in the IP 3 R1 channel. Notably, a π-helix has been also identified in pore-lining S6 of the superfamily of TRP channels 26,27,29 . It appears that TRP channels adopt the π-helical conformation of S6 quite differently upon their gating, with some channels utilizing an α-to-π transition, whereas some exhibit a π-to-α transition, and still others have the S6 π-helix in both closed and open states. Taken together, our results and aforementioned studies suggest that the structural mechanism for channel gating involving the α-to-π transition may not be generalized.
Furthermore, the two identified membrane-associated helices, MA1 and MA2, expand the membrane topology of IP 3 R1, and this finding reconciles a long-standing question on the correlation between TM domain predictions and cryo-EM structural observations. While there is no clear evidence for how these helices affect channel function, this region encompasses the determinants previously described to target IP 3 R to the ER 30,31 . Similar TMD features have begun to be described for RyR1 32,33 and IP 3 R3 6 , albeit at lower resolutions. This suggests that the peripheral MA helices may be a conserved feature of the Ca 2+ release channel family.
In summary, it is now largely appreciated that membrane lipids are integral components of ion channels, and by employing specific binding sites in the TM domains, lipids are likely to confer enhanced stability for proper channel gating. However, the exact molecular mechanism by which lipids exert their effects on IP 3 R channel function remains to be elucidated. It is conceivable that lipid binding to the protein may modulate the activation of the channel gate by providing a structural basis for integration and transmission of activating signals from primary stimuli, such as channel-specific ligands. Several Ca 2+ signaling neurodegenerative diseases (e.g., Alzheimer and Huntington diseases), cardiac or muscular diseases (e.g., Duchenne muscular dystrophy) have been linked to abnormal Ca 2+ signaling, as well as to altered membrane lipid composition, organization in lipid domains, and/ or biophysical properties of membranes (Fig. 5). Given that IP 3 R channels play a key role in the control of cellular Ca 2+ signaling, lipids might represent a missing link in human diseases, where initial perturbation of IP 3 R-lipid interactions resulting in altered regulation and gating of IP 3 R channels could lead to perturbations in cytosolic Ca 2+ levels.
Methods IP 3 R1 purification and reconstitution into nanodiscs. Purification steps were performed as previously described 4,8 with the following modifications. Solubilization of native IP 3 R1 from rat cerebellum was carried out in 2 mM Lauryl Maltose Neopentyl Glycol (LMNG, Anatrace) and 0.1% (w/v) L-α-phosphatidylcholine (PC, Sigma) for 2 h at 4°C. CnBr-sepharose beads (GE Healthcare) were coupled to purified monoclonal antibodies raised against the T433 epitope 34 . The hybridoma producing high-affinity monoclonal antibody against IP 3 R1 was raised in mice using synthetic peptide T433. The hybridoma was grown in serum-free media supplemented with 5% ultra-low IgG from fetal bovine serum (ThermoFisher Scientific). Antibodies were purified from culture supernatants by affinity chromatography using a fast flow protein G column (GE Healthcare) according to the manufacturer's instructions. Immunoaffinity purification of IP 3 R1-LMNG was performed in 50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1 mM DTT, 1 mM EDTA, 0.02 mM LMNG and eluted with 500 μM of T433 peptide.
Cryo-EM data acquisition. Cryo-EM data collection was performed on a Titan Krios G3 microscope (ThermoFisher Scientific) operated at 300 kV and aligned for parallel illumination. BioQuantum energy filter (Gatan) was operated with an energy width of 20 eV. Images for both data sets were acquired using the EPU software (ThermoFisher Scientific) at a nominal TEM magnification of ×130,000 and recorded on a K2 Summit direct electron detector (Gatan) operated in superresolution counting mode with a calibrated physical pixel size of 1.07 Å. The dose rate used on the detector was 8 electrons/Å 2 /s (for IP 3 R1-ND) and 6 electrons/Å 2 /s (for IP 3 R1-LMNG), respectively. The total exposure time of 7 s was fractionated into 35 subframes, each with 0.2 s exposure time, resulting in a total accumulated dose of 56 and 42 electrons/Å 2 at the specimen plane, respectively. The images were acquired at a defocus range of 0.8-3.5 μm (Table 1). A total of 22,000 (IP 3 R1-ND) and 19,105 (IP 3 R1-LMNG) dose-fractionated movie stacks were collected as described above.
For IP 3 R1-ND data, the motion-corrected micrographs were imported to cryoSPARC. The contrast transfer function (CTF) was determined using CTFFIND4 39 .~1000 particles were selected manually for 2D classification and the selected 2D class-averages were used for template-based autopicking. A total of 4,801,610 initial particle picks were subjected to 2D classification, identifying 1,495,402 particles that were subjected to further processing in cryoSPARC. Our previously published map (EMDB-9246) was low pass filtered to 60 Å resolution and used as an initial model for Heterogeneous 3D Refinement with C1 symmetry. 573,723 particles from the best 3D classes were subjected to Homogeneous 3D refinement with a soft mask using C4 symmetry. The final round of refinement and post-processing were carried out in RELION3.
For IP 3 R1-LMNG data, 1,407,714 particles were extracted using the NeuralNet autopicking procedure implemented in EMAN2 40 and were imported to RELION3 for the following image processing. The CTF parameters were determined using GCTF 41 . We then followed up with 2 rounds of 2D classification that yielded 1,011,190 particles ( Supplementary Fig. 2b). These particles were subjected to 3D classification performed without imposing any symmetry. Our published map (EMDB-9246), low-pass filtered to 60 Å resolution, was used as an initial model. Based on 3D classification results, the best 303,481 particles were selected and subjected to further 3D refinement by applying C4 symmetry and a soft mask.
For both maps, the post-processing step was performed in RELION3, a soft mask was calculated and applied to the two half-maps before the Fourier shell correlation (FSC) was calculated. B-factors were estimated and applied to the map sharpening step (Supplementary Figs. 1d and 2c). The resolutions for the final 3D reconstructions using the standard 3D refinement approach were 3.30 Å for IP 3 R1-ND and 2.96 Å for IP 3 R1-LMNG based on the gold standard criteria 42,43 . Local resolution variations were estimated using ResMap 44 ( Supplementary  Figs. 1g and 2f).
Model building and refinement. The same strategy was utilized for building 3D models for both the IP 3 R1-ND and IP 3 R1-LMNG cryo-EM density maps. Initial models were generated from our previously published cryo-EM structure of apo-IP 3 R1 (PDB ID: 6MU2) using rigid body fitting tool in UCSF Chimera 45 . "phenix. resolve_cryo_em" density modification 46 was performed on the half-maps resulting in improved resolvability in the density maps allowing for further model optimization. After rigid-body fitting, initial flexible fitting was performed in COOT 47 by manually adjusting the entire protein-peptide chain of a single IP 3 R1 subunit. After the coarse manual fit, real-space refinement was performed with "phenix.real_space_refine" 48 and subsequently adjusted again manually in Coot. This process was repeated in order to maximize fit to density, minimize Ramachandran angle outliers and eliminate steric clashes. Further model validation was carried out using EMRinger 49 and MolProbity 50 in PHENIX. The full tetrameric model was completed by calculating map symmetry in UCSF Chimera and applying it to the model. A final round of real-space refinement in Phenix and manual optimization in Coot was performed on the entire tetramer before modeling the lipids.
Non-IP 3 R1 protein densities visualized in the cryo-EM maps were isolated by subtracting the protein model from the corresponding map using the "zone" tool in UCSF Chimera 45 . Nanodisc lipids and MSP appear to form a toroidal layer of densities in the IP 3 R1-ND reconstruction. Lipid-like densities with a characteristic head-and-two-tails shape were found to be associated with the TM domains of IP 3 R1-ND and IP 3 R1-LMNG structures. These densities were putatively modeled as phosphatidylcholine (PC) lipids (https://pubchem.ncbi.nlm.nih.gov), which are a major component of ER membranes (>50%) 10 and were also added during the protein solubilization. The PC models were obtained from PubChem (https:// pubchem.ncbi.nlm.nih.gov) and fit into the corresponding cryo-EM densities using Coot. The final full refinement of the entire IP 3 R1-lipid model was performed in Phenix using "phenix.real_space_refine". The overall PDB model validation statistics are presented in Table 1.
The  Table 1). The residue numbering includes the first methionine according to the primary sequence with the GenInfo Identifier (GI) code 17380349. The full model vs. map FSC plots was calculated using MTRIAGE 51 Table 1). This resulted in a shift of the fall-off to a lower resolution in the FSC between the final refined model and cryo-EM map ( Supplementary Figs. 1h and 2g). While map-vs-model FSC for the extracted TM domains, calculated using EMAN2 40 , indicated that the resolution based on the FSC = 0.5 criteria is consistent with the gold-standard resolution at the FSC 0.143 cut-off ( Supplementary Figs. 1i and 2h).
Map-model visualization was performed in Coot and UCSF Chimera. Interfaces described in the paper were identified with PDBSum 52 , PDBePisa 53 , and HOLE 54 . The figures and movie were produced using UCSF Chimera and VMD 1.9.4. 55 .
Statistics and reproducibility. No statistical method was used to predetermine the sample size, and the experiments were not randomized. Each cryo-EM data set was collected from one grid. Individual images with bad ice were excluded from the data set by visual inspection. Data collection, processing, and refinement statistics were summarized in Table 1.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.