Structural basis for the regulation of inositol trisphosphate receptors by Ca²⁺ and IP₃

Abstract

Inositol trisphosphate receptors (IP₃R) are ubiquitous Ca²⁺-permeable channels that mediate release of Ca²⁺ from the endoplasmic reticulum to regulate numerous processes including cell division, cell death, differentiation and fertilization. IP₃R is activated by both IP₃ and its permeant ion Ca²⁺. At high concentrations, however, Ca²⁺ inhibits activity ensuring precise spatiotemporal control over intracellular Ca²⁺. Despite extensive characterization of IP₃R, the mechanisms by which these molecules control channel gating have remained elusive. Here, we present structures of full-length human type 3 IP₃R in ligand-bound and ligand-free states. Multiple IP₃-bound structures demonstrate that the large cytoplasmic domain provides a platform for propagation of long-range conformational changes to the ion conduction gate. Structures in the presence of Ca²⁺ reveal two Ca²⁺ binding sites that induce the disruption of numerous interactions between subunits, thereby inhibiting IP₃R. These structures thus begin to provide a mechanistic basis for understanding the regulation of IP₃R.

Introduction

Inositol trisphosphate receptors (IP₃R) are a family of large tetrameric Ca²⁺-permeable cation channels predominantly expressed in the endoplasmic reticulum (ER)^1–5. Three IP₃R subtypes exist in humans that share between 64 and 70% sequence identity but differ in their tissue expression and subcellular localization, resulting in homo- and heterotetrameric channels that perform a variety of physiological roles^6,7. Activation of IP₃R by cytoplasmic Ca²⁺ and the second messenger inositol 1,4,5-trisphosphate (IP₃) allows Ca²⁺ to flow from the ER into the cytoplasm^2,8–11. The dual regulation of IP₃R by Ca²⁺ and IP₃ integrates extracellular signaling pathways that lead to the generation of cytoplasmic IP₃ with intracellular Ca²⁺ signaling to jointly regulate the activity of numerous cellular processes including fertilization, differentiation, cell division and cell death^12–14. Underscoring the essential role of these channels in diverse cell signaling pathways, the dysregulation of IP₃R is linked to cardiac disease, cancer, neurological disorders and other pathologies^6,15–19.

Electrophysiological and optical recordings of IP₃R activity in a variety of contexts have demonstrated that IP₃R displays a biphasic response to cytoplasmic Ca²⁺ (refs^9,18,20–22). While low concentrations of cytoplasmic Ca²⁺ activate IP₃R, high concentrations of Ca²⁺ inhibit channel activity. The biphasic regulation of IP₃R by its permeant ion ensures precise spatiotemporal control of Ca²⁺ signaling. Depending on the input stimuli, IP₃R can generate cell-wide Ca²⁺ spikes or oscillatory waves as well as more localized Ca²⁺ “blips” or “puffs”^23–25. Proper signal encoding requires Ca²⁺-dependent inhibition as well as Ca²⁺- and IP₃-dependent activation.

Despite intensive investigation into the regulation of IP₃R by Ca²⁺, the mechanisms of Ca²⁺-driven modulation remain open questions^26–29. FRET experiments, partial proteolysis experiments, and negative-stain electron microscopic images suggest that Ca²⁺ stimulates large conformational changes, though the molecular details of these changes are unknown^30,31. Analysis of single channel recordings suggest a high affinity activating Ca²⁺ site and a low affinity inhibiting Ca²⁺ site, though the residues that comprise these sites have yet to be identified³². In contrast, the molecular details of IP₃ binding to IP₃R have been better characterized. Mutational analysis identified the first ~600 residues of IP₃R as the IP₃-binding domain^3,33. Crystal structures of isolated IP₃-binding domains in the presence and absence of IP₃ reveal that IP₃ binding is accompanied by a closing of the clamshell-like IP₃-binding domain around IP₃ (refs^34–37). A cryo-electron microscopy (cryo-EM) structure of rat type 1 IP₃R (rIP₃R1) in a ligand-free state at a resolution of 4.7 Å revealed that this IP₃-binding domain is located more than 70 Å from the pore³⁸. However, side-chain detail is lacking for most residues in the rIP₃R1 density map and the majority of the structural model was therefore built as a poly-alanine backbone. Thus, the architecture of IP₃R at atomic resolution remains unknown and it is unclear how IP₃ binding induces conformational changes that are propagated over such a long distance to influence the gating state of the pore. Therefore, high-resolution structural studies are required to understand the mechanisms through which Ca²⁺ and IP₃ regulate IP₃Rs. Here, we determined structures of full-length human type 3 IP₃R (hIP₃R3) in different states, allowing us to start to delineate the mechanistic basis for IP₃R regulation by IP₃ and Ca²⁺.

Results

Architecture of ligand-free human type 3 IP₃ receptor

To resolve the atomic structure of a full-length human IP₃R, we first determined the structure of hIP₃R3 in a ligand-free state by single-particle cryo-EM at an overall resolution of 3.5 Å with C4 symmetry imposed (Supplementary Fig. 1 and Table 1). The map was of excellent quality, enabling us to de novo model most parts of the channel. However, some of the peripheral regions in the cytoplasmic domain and the carboxy-terminal domain were less well resolved. The density for the cytoplasmic domain was improved by employing symmetry expansion, signal subtraction and focused refinement³⁹ (Supplementary Fig. 1 and Supplementary Table 1). Using the cytoplasmic domain focused refinement map, we were able to build and register most of the cytoplasmic domain. In contrast, focused refinement approaches were unable to sufficiently improve the density for the small carboxy-terminal domain due to its extreme conformational flexibility. Thus, while the carboxy-terminal domain is present in the density map as a four-helix bundle as has been observed for rIP₃R1³⁸, it is not modeled in the final structure (Supplementary Fig. 2).

Table 1.

Cryo-EM data collection, refinement and validation statistics.

	hIP3R3 apo (EMD-7978) (PDB 6DQJ)	hIP3R3 IP3 class 1 (EMB-7981) (PDB 6DQN)	hIP3R3 IP3 class 2 (EB-7984) (PDB 6DQV)	hIP3R3 IP3 class 3 (EMB-7983) (PDB 6DQS)	hIP3R3 IP3 class 4 (EMB-7986) (PDB 6DQZ)	hIP3R3 IP3 class 5 (EMB-7987) (PDB 6DR0)	hIP3R3 Ca2+-bound (EMB-7988) (PDB 6DR2)	hIP3R3 Low IP3-Ca2+ (EMB-7991) (PDB 6DRA)	hIP3R3 High IP3-Ca2+ (EMB-7994) (PDB 6DRC)
Data collection and processing
Magnification	105,000x	22,500x	22,500x	22,500x	22,500x	22,500x	22,500x	22,500x	22,500x
Voltage (kV)	300kV	300kV	300kV	300kV	300kV	300kV	300kV	300kV	300kV
Electron exposure (e–/Å2)	60	61	61	61	61	61	61	61	61
Defocus range (μm)	−1.0 – 2.5	−1.0 – 2.5	−1.0 – 2.5	−1.0 – 2.5	−1.0 – 2.5	−1.0 – 2.5	−1.0 – 2.5	−1.0 – 2.5	−1.0 – 2.5
Pixel size (Å)	1.096	1.088	1.088	1.088	1.088	1.088	1.088	1.088	1.088
Symmetry imposed	C4	C4	C4	C1	C1	C1	C4	C4	C4
Initial particle images (no.)	52,767	302,966	302,966	302,966	302,966	302,966	49,060	74,277	170,308
Final particle images (no.)	26,325	38,777	40,531	37,910	9,535	27,334	33,807	49,087	131,437
Map resolution (Å)	3.49	3.33	3.82	4.12	6.01	4.47	4.33	3.96	3.92
FSC threshold	0.143	0.143	0.143	0.143	0.143	0.143	0.143	0.143	0.143
Map resolution range (Å)	300-3.49	300-3.33	300-3.82	300-4.12	300-6.01	300-4.47	300-4.33	300-3.96	300-3.92
Refinement
Initial model used (PDB code)	3UJ4
Model resolution (Å)	3.80/3.35	3.88/3.36	4.53/3.93	4.83/3.90	7.63/6.33	7.06/4.23	7.44/4.34	4.49/3.91	4.43/3.92
FSC threshold	0.50/0.143	0.50/0.143	0.50/0.143	0.50/0.143	0.50/0.143	0.50/0.143	0.50/0.143	0.50/0.143	0.50/0.143
Model resolution range (Å)	300-3.5	300-3.3	300-3.8	300-4.1	300-6.0	300-4.5	300-4.3	300-4.0	300-4.0
Map sharpening B factor (Å2)	−90	−90	−90	−90	−90	−90	−90	−90	−90
Model composition
Non-hydrogen	69,220	69,412	70,144	69,600	69,828	69,986	69,3572	69,572	69,508
Protein residues	8,744	8,744	8,752	8,750	8,748	8,750	8,764	8,764	8,744
Ligands	4	8	8	8	8	8	12	12	16
Mean B factors (Å2)
Protein	95.1	116.1	169.0	193.5	376.7	217.0	239.9	240.0	217.9
Ligand	53.5	96.95	133.0	177.4	362.0	182.9	174.7	163.6	263.3
R.m.s. deviations
Bond lengths (Å)	0.007	0.003	0.005	0.003	0.003	0.003	0.004	0.003	0.004
Bond angles (°)	0.688	0.490	0.648	0.520	0.536	0.538	0.660	0.517	0.623
Validation
MolProbity score	1.53	1.21	1.36	1.28	1.22	1.26	1.26	1.24	1.33
Clashscore	2.16	2.09	2.22	2.41	2.43	2.37	2.07	2.96	2.14
Poor rotamers (%)	1.97	1.06	0.00	0.00	0.00	0.00	0.00	0.42	0.00
Ramachandran plot
Favored (%)	95.52	96.70	94.74	96.14	96.8	96.33	95.91	97.12	95.04
Allowed (%)	4.48	3.30	5.26	3.86	3.20	3.67	4.09	2.88	4.96
Disallowed (%)	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00	0.00

hIP₃R3 is a tetrameric channel comprised of 2671 residues per subunit with a large amino-terminal cytoplasmic domain (CD) followed by a juxtamembrane domain (JD) and a transmembrane domain (TMD), an architecture that resembles that of rIP₃R1³⁸ (Fig. 1a). Starting at the amino-terminus, the large CD can be divided into two β-trefoil-containing domains (BTF1, BTF2), an armadillo repeat domain (ARM1), the first segment of the central linker domain (CLD), a second armadillo repeat domain (ARM2), the second segment of the CLD and the third armadillo repeat domain (ARM3) (Supplementary Note 1). The CD is stabilized by a number of inter-subunit interactions including a central BTF ring that is composed of the BTF1 and BTF2 domains from all four subunits (Fig. 1b). The BTF ring, which contains several residues involved in IP₃ binding, is connected to the JD through the largely helical ARM1, CLD and ARM3 domains while the helical ARM2 domain is positioned at the periphery of the channel. The JD is comprised of the two segments that flank the TMD and is stabilized by a Zn²⁺ ion that is coordinated by Cys2538, Cys2541, His2558 and His2563. The Zn²⁺ binding site is also present in the structure of rIP₃R1 and in ryanodine receptor structures, indicating that the Zn²⁺-binding site is broadly conserved^38,40.

Figure 1. Structure of human type 3 IP₃R in a ligand-free state.

a, Structure of hIP₃R3 viewed in the plane of the membrane with the cytoplasmic domain at the top. A single subunit is colored according to domain with BTF1 in purple, BTF2 in blue, ARM1 in violet, the CLD in cyan, ARM2 in green, ARM3 in yellow, the JD in orange and the TMD in red. Gray lines represent the approximate position of the membrane. b, Structure of the cytoplasmic domain viewed from the cytoplasm. A single subunit is colored according to domain. ARM3 is removed for clarity. c, Structure of the transmembrane domain viewed from the cytoplasm. S1–S4 helices are colored red, S1′ and S1″ blue and S5, S6 and pore helix green. d–e, Structure of the ion conduction pathway viewed in the plane of the membrane and plot of pore radius. Residues comprising the luminal vestibule, the selectivity filter and S6 gate are highlighted. Front and rear subunits are removed for clarity.

While modeling the TMD, we noticed the presence of multiple hydrophobic lipid or detergent molecules forming a bilayer-like structure at the periphery of the TMD as well as two additional densities that could not be attributed to lipid molecules (Supplementary Fig. 3 and Supplementary Table 1). Inspection of these densities revealed that they correspond to two additional transmembrane helices located between the canonical S1 and S2 helices of the 6TM ion channel fold, which we will refer to as S1′ and S1″ (Fig. 1c). S1″ is positioned adjacent to S1 and S2 and is relatively well ordered in the density map, which allowed us to register its sequence. S1′ is positioned further from the S1–S4 domain on the outside of S1″ and is less well ordered in the density map. Focused classification and refinement improved the density for S1′, but the register could not be confidently determined and we thus modeled S1′ as an unregistered poly-alanine helix (Supplementary Fig. 3). Sequence conservation of the region between S1 and S2 is high among the three human IP₃R types, especially within S1′. Indeed, close inspection of the rIP₃R1 density map revealed densities in similar positions as S1′ and S1″, suggesting that S1′ and S1″ are conserved features of IP₃Rs (Supplementary Fig. 3). S1′ and S1″ extend away from the channel in the membrane where they may potentially facilitate regulatory protein-protein interactions with other membrane-embedded proteins⁴¹.

The ion conduction pathway of hIP₃R3 is formed by S5, S6 and the re-entrant pore helices of each subunit (Fig. 1d). Starting from the luminal side of the membrane, ions first pass through an electronegative vestibule formed by the extended loops between S5 and the pore helix and between the pore helix and S6. A total of 12 negatively charged residues (Glu2398, Asp2400 and Asp2478 from each subunit) are resolved in the luminal vestibule. The electronegative environment of the vestibule may serve to concentrate cations before entering the pore, potentially facilitating Ca²⁺ conductance. Two constrictions are present in the pore between the luminal vestibule and cytoplasmic opening. The first constriction is created by the backbone carbonyl oxygen of Asn2472, which resides on the pore helix (Fig. 1d and Fig. 1e). The constriction has a radius of about 3 Å at its narrowest point and along with neighboring residues may contribute to ion selectivity, although IP₃R3 is only weakly selective for Ca²⁺ over other cations⁴². The second constriction is a hydrophobic gate formed by the side chains of Phe2513 and Ile2517 of S6, which constricts the pore to a radius of less than 1 Å, too narrow to conduct Ca²⁺. The ligand-free state of hIP₃R3 is thus a non-conductive state, consistent with electrophysiological recordings showing that Ca²⁺ and IP₃ are required for IP₃R activation^2,18,32.

IP₃-bound conformations

To resolve the conformational changes associated with IP₃ binding in the context of the full channel, we collected images of hIP₃R3 in the presence of saturating (50 μM) IP₃. Successive rounds of three-dimensional classification revealed an ensemble of distinct IP₃-bound conformational states (Supplementary Fig. 4). Five of the conformational states were sufficiently well populated to generate 3D reconstructions at resolutions up to 3.3 Å, from which we could model their structures.

A reconstruction of the particles assigned to the first class, IP₃ class 1, at 3.3 Å revealed an overall conformation that is similar to the apo state (Fig. 2a and Table 1). The structure of IP₃ class 1 has an all-atom RSMD of approximately 1 Å compared to the apo structure. Despite this overall structural similarity, local conformational changes are present in the IP₃-binding domain. The ARM1-proximal region of BTF2 moves 4 Å towards ARM1 in IP₃ class 1, narrowing the interface between the two domains without disrupting the BTF ring. In the interface between BTF2 and ARM1, we identified a three-lobed density that we modeled as an IP₃ molecule (Supplementary Fig. 5). The IP₃ molecule is coordinated by the positively charged side chains of Arg266 and Arg270 from BTF2 and Arg503, Lys507, Arg510, Arg568 and Lys569 from ARM1 (Fig. 2b and Supplementary Fig. 5), many of which have been previously identified as residues that directly bind IP₃ in crystal structures of isolated IP₃-binding domains^34,36,37.

Figure 2. IP₃ binding site in two IP₃-bound conformations.

a, d, Superposition of (a) IP₃ class 1 (colored by domain) and apo (grey) or (d) IP₃ class 2 (colored by domain) and apo (grey) viewed from in the plane of the membrane (left) and the cytoplasm (right). b, e, Superposition of IP₃-binding domain of (b) IP₃ class 1 (colored by domain) and apo (grey) or (e) IP₃ class 2 (colored by domain) and apo (grey) aligned by BTF1 and BTF2. IP₃ and the side chains of IP₃-coordinating residues are shown as sticks. c, f, Superposition of ARM3, the JD and S6 of (c) IP₃ class 1 (colored by domain) and apo (grey) or (f) IP₃ class 2 (colored by domain) and apo (grey) aligned by the TMD and viewed from the lumen.

Whereas IP₃ class 1 closely resembles apo hIP₃R3, a reconstruction of the particles assigned to IP₃ class 2 at 3.8 Å reveals significant conformational changes (Supplementary Fig. 4 and Table 1). Compared to the apo structure, the length of IP₃ class 2 is 5 Å shorter along the four-fold axis and the conformation of the CD is substantially different (Fig. 2d). To better understand the origin of these conformational changes, we examined the IP₃-binding site of class 2. Similar to class 1, an IP₃ molecule is coordinated by the side chains of Arg266, Arg270, Arg503, Lys507, Arg510, Arg568 and Lys569 from BTF2 and ARM1 (Fig. 2e and Supplementary Fig. 5). However, the configuration of the IP₃-binding domain differs in class 2. In class 2, IP₃ binding is accommodated by a rigid-body rotation of ARM1 towards BTF2, which adopts an apo-like conformation (Fig. 2e). The only changes in BTF2 compared to the apo state are slight movements of the loops that directly coordinate IP₃.

Because ARM1 moves relative to the fixed BTF ring structure in IP₃ class 2, IP₃ binding is accompanied by conformational changes throughout the channel (Fig. 2d). These conformational changes can be separated into two major movements. The first is a coupled movement of the CLD and ARM3 that results in a 5˚ rotation of the JD (Fig. 2f, Supplementary Video 1 and Supplementary Video 2). While the TMD is nearly unchanged relative to the apo state, these concerted movements provide a model for how ligand binding can be transmitted through multiple helical domains to the pore. The second conformational change is a 17 Å movement of ARM2 to position itself near the CLD of the same subunit (Fig. 2d, Supplementary Video 1 and Supplementary Video 2). Besides coming nearly into contact with the CLD, the movement of ARM2 also alters its interactions with the neighboring subunit. ARM2 no longer makes contacts with ARM1 of the neighboring subunit and its interaction with BTF2 of the neighboring subunit is shifted by 2 armadillo repeats from α63 to α68.

When examining the density maps of IP₃ classes 3, 4 and 5, we realized that they are not four-fold symmetric like the maps of classes 1 and 2 (Supplementary Fig. 4). The asymmetry of these classes is most apparent in the positions of the four ARM2 domains, which can adopt either a class 1-like extended conformation or a class 2-like compact conformation (Fig. 3a). We therefore calculated reconstructions of classes 3, 4 and 5 without imposing symmetry at resolutions of 4.1, 6.0 and 4.5 Å, respectively (Table 1). Structures of classes 3, 4 and 5 were initially modeled by rigid-body docking domains of classes 1 or 2 into the density map and then comparing which of the two structures fit best into the density map. The docking demonstrated that these classes were composed of both class 1-like and class 2-like domains (Fig. 3b).

Figure 3. Ensemble of IP₃-bound conformations.

a, Structures of IP₃ class 1, IP₃ class 32, IP₃ class 4, IP₃ class 5 and IP₃ class 2 colored by domain. The ARM2 domain of subunits marked with an * adopt a class 2-like conformation. b, Structures of IP₃ class 1, IP₃ class 3, IP₃ class 4, IP₃ class 5 and IP₃ class 2. Domains adopting class 1-like domains are colored orange and domains adopting class 2-like conformations are colored cyan.

Even at the moderate resolutions of these reconstructions, inspection of the IP₃-binding domains reveals densities consistent with IP₃ molecules at the BTF2-ARM1 interfaces in all three density maps (Supplementary Fig. 5). Thus, the asymmetry in these classes is not caused by variations in IP₃-binding site occupancies. Rather, the asymmetry arises from variations in the conformation of the individual subunits. Notably, a single subunit can be composed of domains that adopt class 1-like and domains that adopt class 2-like conformations. For example, one of the subunits in class 5 has a class 1-like IP₃-binding domain, while the rest of the subunit is class 2-like (Fig. 3b).

Thus, despite IP₃R possessing a single IP₃ binding site, the configuration of the IP₃-binding domain can vary when IP₃ is bound and these variations are coupled to large changes in global channel conformation (Fig. 2). Notably, both configurations can be present in the same channel indicating that the binding modes are in dynamic exchange and that stable intermediates exist between them (Fig. 3). Multiple crystal structures have been determined of isolated IP₃-binding domains in complex with IP₃ (refs^34,36,37). The position of ARM1 relative to BTF2 varies among these structures, adopting a range of conformations between those present in IP₃ class 1 and IP₃ class 2 suggesting that the rest of the channel may influence the configuration of the IP₃-binding domain (Supplementary Fig. 5).

Ca²⁺-bound conformations

To identify the Ca²⁺ binding sites and visualize the effects of Ca²⁺ binding on IP₃R conformation, we next determined a structure of hIP₃R3 at 4.3 Å resolution in the presence of approximately 2 mM Ca²⁺ with C4 symmetry imposed, which we will refer to as Ca²⁺-bound (Supplementary Fig. 6 and Table 1). Examination of the map revealed that the peripheral portions of the CD, including BTF1, BTF2 and ARM1 were poorly resolved and not strictly four-fold symmetrical. We therefore employed two different focused refinement strategies to better resolve the channel. We first calculated a four-fold symmetric reconstruction of ARM3, the JD and the TMD to a resolution of 4.2 Å (Supplementary Fig. 6 and Supplementary Table 1). Next, to improve the density of the flexibly attached CD, we employed signal subtraction, symmetry expansion and focused refinement, resulting in a 4.7 Å reconstruction in which the CD is clearly resolved (Supplementary Fig. 6 and Supplementary Table 1). The structure of Ca²⁺-bound hIP₃R3 was modeled using both the full channel and focused refinement maps. In addition to densities corresponding to hIP₃R3, both the full channel and focused refinement maps contained densities corresponding to two bound Ca²⁺ ions per subunit (Fig. 4a and Supplementary Fig. 7). The first Ca²⁺, which we will call the CD Ca²⁺, is located at the interface between CLD and ARM2 where it is coordinated by main-chain and side-chain oxygen atoms from both domains (Fig. 4a,b). The CLD contributes the main-chain carbonyl oxygen of Arg743, while ARM3 contributes the side chain of Glu1125 and the main-chain carbonyl oxygen of Glu1122. From the apo state, ARM2 moves 28 Å to complete the binding site (Fig. 4c). Notably, ARM2 in the Ca²⁺-bound state is in a similar conformation to its location in IP₃ class 2. From its position in IP₃ class 2, Glu1122 and Glu1125 on ARM2 must move only an additional 3 Å to complete the coordination of the CD Ca²⁺. The second Ca²⁺, which we will refer to as the JD Ca²⁺, occupies a cavity between the JD and ARM3 (Fig. 4a and Supplementary Fig. 7). The JD Ca²⁺ is coordinated by the side chains of two glutamate residues on ARM3 (Glu1882 and Glu1946) and by main-chain and side-chain carbonyl oxygens of one residue on the JD (Thr2581) (Fig. 4d). To bring the side chains of Glu1882 and Glu1946 into close proximity of Thr2581 and complete the Ca²⁺ coordination, ARM3 must rotate 11˚ compared to the apo state (Fig. 4e). In the Ca²⁺-bound conformation, the structure of the JD binding site almost perfectly matches that of a Ca²⁺ binding site identified in the rabbit ryanodine receptor structure suggesting that some aspects of channel regulation are conserved between IP₃Rs and ryanodine receptors⁴⁰ (Supplementary Fig. 7d).

Figure 4. Ca²⁺ binding sites in Ca²⁺-bound hIP₃R3.

a, Monomeric structure of Ca²⁺-bound hIP₃R3 colored by domain viewed in the plane of the membrane. Ca²⁺ and Zn²⁺ ions are shown as spheres. b, CD Ca²⁺-binding site at the CLD-ARM2 interface. Residues coordinating the Ca²⁺ are shown as sticks and the CD Ca²⁺ ion is shown as a magenta sphere. c, Superposition of the CLD and ARM2 of Ca²⁺-bound (colored by domain) and apo (grey, left) or Ca²⁺-bound (colored by domain) and IP₃ class 2 (grey, right) aligned by the CLD. d, JD Ca²⁺-binding site at the ARM3-JD interface. Residues coordinating the Ca²⁺ are shown as sticks and the Ca²⁺ ion is shown as a green sphere. e, Superposition of ARM3 and the JD of Ca²⁺-bound (colored by domain) and apo (grey) aligned by the JD.

Figure 5 shows a comparison of the tetrameric Ca²⁺-bound and apo structures revealing the conformational changes induced by Ca²⁺ binding at the CD and JD sites. The most obvious differences between the structures are a nearly 30 Å dilation of the CD and a complete disruption of the BTF ring that anchors inter-subunit interactions (Fig. 5 and Supplementary Video 3). Indeed, close inspection reveals that Ca²⁺ binding eliminates all of the inter-subunit interactions present in the apo and the IP₃-bound structures. Dissociation of the CDs yields a channel in which the four CDs are physically uncoupled from one another, explaining the increased mobility of the CDs compared to the apo state. The dissociation also explains the reduced retention time of Ca²⁺-bound hIP₃R in gel filtration experiments compared to unliganded channels and is consistent with FRET measurements demonstrating global conformational changes in the presence of Ca²⁺ (ref³¹) (Supplementary Fig. 6a).

Figure 5. Ca²⁺ binding disrupts CD intra- and inter-subunit interactions.

a–b, Cytoplasmic domain of (a) apo and (b) Ca²⁺-bound colored by domain, viewed from the cytoplasm. Spheres represent the Cα position of Trp168 of BTF1 and Lys426 of BTF2 (pink) and Pro140 of BTF1 and Ala1291 of ARM2 (brown). c–d, Structure of (c) apo and (d) Ca²⁺-bound colored by domain, viewed in the plane of the membrane. Front and rear subunits are removed for clarity. Ca²⁺ and Zn²⁺ ions are shown as spheres. Grey lines represent the approximate position of the membrane.

Ca²⁺- and IP₃-bound conformations

To understand how IP₃ and Ca²⁺ coordinate to regulate channel gating, we determined structures of hIP₃R3 in the presence of approximately 2 mM Ca²⁺ and either 10 μM (low) or 50 μM (high) IP₃ at 4.0 Å and 3.9 Å, respectively (Supplementary Fig. 8, Supplementary Fig. 9 and Table 1). Similar to the Ca²⁺-bound state, the CD was poorly ordered in both the low IP₃-Ca²⁺ and high IP₃-Ca²⁺ reconstructions. We therefore calculated two sets of focused refinement maps for each condition. The TMD focused refinement maps were resolved to 3.8 Å for the low IP₃-Ca²⁺ state and 3.7 Å for the high IP₃-Ca²⁺ state while the CD focused refinement maps were resolved to 4.2 Å for the low IP₃-Ca²⁺ state and 3.8 Å for the high IP₃-Ca²⁺ state (Supplementary Figure 8, Supplementary Figure 9, Supplementary Table 1). Inspection of the ligand-binding sites in the high IP₃-Ca²⁺ condition revealed density consistent with the IP₃-binding site and both Ca²⁺ sites being occupied (Supplementary Fig. 10). To bind IP₃, ARM1 in the high IP₃-Ca²⁺ condition is rotated 15˚ towards BTF2 to adopt a conformation that is similar to IP₃ class 2. In the low IP₃-Ca²⁺ condition densities are clearly present at both of the Ca²⁺ sites, but the IP₃-binding site is unoccupied and the IP₃-binding domain adopts an apo-like conformation.

Despite IP₃ being bound, the overall structure of the high IP₃-Ca²⁺ state closely resembles the Ca²⁺-bound structure (Fig. 6a,c and Supplementary Video 3). In both structures the BTF rings are dissociated and the four CDs are detached from one another. Indeed, the only difference in the high IP₃-Ca²⁺ structure compared to the Ca²⁺-bound structure is a clamping down of the IP₃-binding domain around IP₃. In contrast to the similarity between the Ca²⁺-bound and high IP₃-Ca²⁺ structures, the structure of IP₃ class 2 demonstrates that global conformational changes can occur following IP₃ binding (Fig. 6b,d). Whereas the network of inter-subunit interactions in the cytoplasmic domain enable conformational changes to be propagated throughout the channel in IP₃ class 2, their disruption in the high IP₃-Ca²⁺ structure results in four CDs that are physically uncoupled from one another and from the pore. Thus, the Ca²⁺ binding sites stabilize the high IP₃-Ca²⁺ structure in an inhibited conformation. Furthermore, Ca²⁺ binding also stabilizes the separation of the CD domains in the low IP₃-Ca²⁺ and Ca²⁺-bound structures indicating that Ca²⁺-dependent IP₃R inhibition may be independent of IP₃-binding state (Fig. 5).

Figure 6. IP₃-induced conformational changes.

a, High IP₃-Ca²⁺ structure colored by distance of Cα deviation from the Ca²⁺-bound structure viewed in the plane of the membrane. b, IP₃ class 2 colored by distance of Cα deviation from the apo structure viewed in the plane of the membrane. c, Cytoplasmic domain of high IP₃-Ca²⁺ structure colored by distance of Cα deviation from the Ca²⁺-bound structure viewed from the cytoplasm. d, Cytoplasmic domain of IP₃ class 2 colored by distance of Cα deviation from the apo structure viewed from the cytoplasm. IP₃ and Ca²⁺ ions are shown as spheres.

Discussion

The ensemble of structures presented here outlines a model for how IP₃ and Ca²⁺ control the gating of IP₃R. The structures support a model in which a network of inter-subunit interactions enables the propagation of conformational changes from peripheral ligand binding domains to the JD and TMD. It has long been recognized that domains such as BTF1, which do not directly contribute to ligand binding, are essential for IP₃R gating^42–45. The IP₃-bound structures demonstrate that these domains are essential because of their incorporation into the inter-subunit interactions network. BTF1, for example, is a key component of a central ring structure that serves to anchor the inter-subunit network and forms a fulcrum against which movements of the IP₃-binding domain are propagated throughout the channel. Thus, this network ensures that conformational changes in the IP₃-binding domain are coupled to the gating state of the pore (Fig. 2). Ca²⁺ binding disrupts this network leading to an inhibited state where movements due to IP₃ binding are insulated from global channel conformation (Fig. 6).

While high concentrations of Ca²⁺ inhibit IP₃R, low concentrations of Ca²⁺ are required for activation. In the Ca²⁺-bound inhibited structures two Ca²⁺ ions are present: one in the periphery of the CD (CD Ca²⁺) and one at the CD-JD interface (JD Ca²⁺). The presence of two Ca²⁺ binding sites at distinct locations within IP₃R is consistent with different apparent Ca²⁺ binding affinities for activation and inhibition³². While the functional role of the two sites cannot be distinguished from the structural data alone, it is possible that the two sites play distinct roles in gating, with one regulating IP₃R activation and the second inhibition.

Aligning the structures into a trajectory allows us to posit a model for how IP₃ and Ca²⁺ binding leads to activation of hIP₃R3 (Fig. 7). Due to the similarity between the apo state and IP₃ class 1, we hypothesize that upon IP₃ binding the channel first adopts a structure similar to IP₃ class 1 in which conformational changes are limited to the local environment of the IP₃-binding domain. Spontaneous rearrangements of individual IP₃-binding domains into class 2-like configurations may then allow the channel to adopt asymmetric intermediate conformations analogous to IP₃ classes 3, 4 and 5 in which subunits within the channel can adopt different conformations. The presence of multiple stable intermediate states suggests that the conformational exchange between the two IP₃-binding domain configurations occur on a per-subunit basis rather than as a concerted mechanism in which all four subunits act as one (Fig. 3). Compared to the apo state, ARM3 and the JD are rotated as more of the IP₃-binding domains adopt the IP₃ class 2-like configuration suggesting that IP₃-bound structures are intermediates on the pathway to activation (Fig. 2 and Fig. 3). We speculate that Ca²⁺ binding at one of the sites may induce further rotation of the JD leading to an opening of the S6 gate and a conductive state. Between the selectivity filter and the S6 gate, a portion of S6 adopts a short five-turn or π-helical region in the apo structure. Relaxation from this high-energy π-helix into a lower energy α-helix upon JD rotation may help to stabilize the pore in a conductive state. Transitions between π-helical and α-helical conformations within the S6 helices during gating have been recently described in TRPV6⁴⁶. It is also possible that all of the IP₃-bound states can directly access an activated state. Single-channel recordings have identified three reversible modes with distinct gating kinetics whose open channel probability vary by more than 100-fold without changing Ca²⁺ or IP₃ concentrations⁴⁷. These gating modes may result from the existence of different IP₃-bound states that open with stably different probabilities. IP₃ class 1, which is nearly identical to the apo state, would require large conformational changes and despite the presence of bound IP₃ may be the least likely to open. In contrast, IP₃ class 2, in which IP₃ binding is accompanied by conformational changes that propagate towards the pore, may be more likely to open. Thus, the ensemble of structures provide a foundation for beginning to understand the complex gating of IP₃Rs.

Figure 7. Model for IP₃R regulation by IP₃ and Ca²⁺.

In the absence of Ca²⁺ and IP₃, IP₃R adopts a closed state. In the presence of IP₃, IP₃R adopts one of an ensemble of pre-activated states that are in equilibrium. In the presence of high Ca²⁺, the BTF ring is dissociated and the channel adopts an inhibited state. In the presence of high Ca²⁺ and IP₃, the BTF ring is dissociated and the channel adopts an inhibited state. In the presence of low Ca²⁺ and IP₃, the channel adopts a hypothetical activated state in which the S6 gate is opened.

Online Methods

Cell Lines

HEK293S GnTI- (human embryonic kidney, N-acetylglucosaminyltransferase I knockout) cells were used for protein expression and maintained in Freestyle 293 Expression Media (Gibco) supplemented with 2% fetal bovine serum (FBS) at 37˚ C.

Constructs

A gene for human type 3 IP₃R (HsITPR3) in pENTR223 vector was obtained from HMS PlasmID (HsCD00399229). Internal restriction sites for XhoI and EcoRI were removed by site directed mutagenesis. Polymerase chain reaction (PCR) was used to amplify the gene and add restriction sites for XhoI and EcoRI for in-frame subcloning into a bacmam expression vector with an N-terminal GFP and PreScission protease cut site^48,49. The final product (N-GFP-HsITPR3) was sequenced in its entirety.

hIP₃R3 Expression & Purification

N-GFP-HsITPR3 was used to transform DH10Bac E. coli cells to generate bacmid DNA. Recombinant baculovirus was produced by three rounds of viral amplification in Sf9 (Spodoptera frugiperda) cells maintained in Grace’s media supplemented with 10% FBS, 1% penicillin/streptomycin, and 0.1% Pluronic F-68. Viral particles were separated from cell debris by centrifugation and filtration, and used to infect HEK293S GnTI- cells for protein production. For the first 24 hours of infection, cultures were grown at 37˚ C, after which sodium butyrate was added to a final concentration of 10 mM and cells moved to 30˚ C to grow for an additional 48–72 hours prior to harvesting. Cell pellets were washed in phosphate-buffered saline solution and flash frozen in liquid nitrogen.

Thawed cell pellets were solubilized for 1 hour in 2% lauryl maltose neopentyl glycol, 4 mM cholesterol hemisuccinate (CHS), 20 mM HEPES pH 7.5, 150 mM NaCl with protease inhibitor cocktail (1 mM PMSF, 2.5 ug/mL aprotinin, 2.5 ug/mL leupeptin, 1 ug/mL pepstatin A, 0.5 mM AEBSF, 1 mM benzamidine), and DNAse. Whole cell lysates were centrifuged at 75,000xg for 40 minutes, and the soluble fraction was bound to GFP-nanobody beads for 1–3 hours on a rotator⁵⁰. Beads were collected on a column by gravity and then washed with 50 mL of wash buffer (150 mM NaCl, 50 mM Tris pH 8.0, 2 mM DTT, 0.06% digitonin). The protein was digested with PreScission protease diluted in wash buffer on the column for 3 hours to separate the channel from the GFP and then eluted with 12 mL of wash buffer.

Concentrated protein was further purified by size-exclusion chromatography on a Superose 6 Increase (GE Life Sciences) column in gel filtration buffer (150 mM NaCl, 50 mM Tris pH 8.0, 2 mM DTT, 0.06% digitonin) with either addition of 100 μM Ca²⁺, 1 mM Ca²⁺, 5 mM EGTA or 10 mM HEDTA, based on the conditions required for vitrification. Peak fractions were pooled and concentrated to ~8 mg/mL for cryo-EM analysis. For cryo-EM samples containing IP₃ (d-myo-Inositol 1,4,5-Trisphosphate), a stock solution of 1 mM IP₃ (Calbiochem, SR5440-1102-669) in ddH2O was added to bring the solution to the appropriate concentration approximately 30 minutes before vitrification.

Electron Microscopy Sample Preparation and Imaging

3.5–5 μl of purified channel at a concentration of ~8 mg/mL was pipetted onto glow-discharged 400 mesh gold Quantifoil R1.2/1.3 holey carbon grids (Quantifoil, Q25869) approximately 5–20 seconds before blotting (blot force = 1) for two seconds and plunging into liquid nitrogen-cooled liquid ethane using an FEI Vitrobot Mark IV (FEI ThermoFisher). Precise control over free Ca²⁺ concentrations in the vitrified samples is complicated by the presence of high concentrations of Ca²⁺ in the filter paper used for blotting excess solution away from the grid. Prior to our data collection, we performed initial screening experiments using various Ca²⁺ and EGTA concentrations, which suggested that the final concentration of free Ca²⁺ may be as much as 2 mM higher than those used during sample preparation. This estimate was based upon the concentrations of Ca²⁺ and EGTA needed to completely eliminate the inhibited conformation. Efforts to more precisely estimate the free Ca²⁺ following blotting were complicated by the extremely small quantity of solution remaining on the grid following blotting (less than 0.01 μl). The Ca²⁺ concentrations presented here are thus best estimates and may not reflect the true concentration of free Ca²⁺ to which the samples are exposed. Additional conditions were tested, but the presence of channels in Ca²⁺-inhibited states indicated that the free Ca²⁺ was not well controlled. For the ligand-free condition, we added no free Ca²⁺ and either 5 mM EGTA or 10 mM HEDTA and predict a free Ca²⁺ below 100 nM. For the IP₃-bound condition, we added no free Ca²⁺ and 5 mM EGTA and predict a free Ca²⁺ below 100 nM. For the Ca²⁺-bound condition, we added 1 mM Ca²⁺ and no EGTA and predict a free Ca²⁺ above 1 mM. For the low IP₃-Ca²⁺ condition, we added 100 μM Ca²⁺ and no EGTA and predict a free Ca²⁺ above 1 mM. For the high IP₃-Ca²⁺ condition, we added 100 μM Ca²⁺ and no EGTA and predict a free Ca²⁺ above 1 mM.

Grids were transferred to an FEI Titan Krios cryo-EM operating at 300 kV. Images were recorded in an automated fashion using SerialEM or Leginon with a defocus range of −1.0 to −2.5 μM over 8 seconds as 40 sub-frames using a Gatan K2 Summit direct electron detector in super-resolution mode^51,52. For the apo data collection, images were recorded at a calibrated super-resolution pixel size of 0.548 Å/pixel with energy filter slit width of 20 eV. For all other data sets, images were recorded at a calibrated super-resolution pixel size of 0.544 Å/pixel. Electron dose was 9 e⁻/pix/s at the detector for a total accumulated dose of 60 e⁻/Ų for the apo data set and 61 e⁻/Ų for the other data sets. Images were two-times Fourier cropped and aligned using whole frame and local correction algorithms by Motioncor2 resulting in a final calibrated pixel size of 1.096 Å/pixel for the apo data set and 1.088 Å for the other data sets⁵³. The effects of the contrast transfer function were estimated using CTFfind 4.1.8 (ref⁵⁴). A total of 1801, 4795, 944, 1409, and 4076 images were recorded for the apo, IP₃-bound, Ca²⁺-bound, low IP₃-Ca²⁺ and high IP₃-Ca²⁺ data sets, respectively. After removing images with excessive drift, bad ice or poor contrast, 1767, 4519, 926, 1409, and 3974 images were kept for processing for the apo, IP₃-bound, Ca²⁺-bound, low IP₃-Ca²⁺ and high IP₃-Ca²⁺ data sets, respectively.

Cryo-EM analysis

Approximately 1000 particles from each data set were manually selected to generate initial templates for autopicking using Relion³⁹. Several rounds of 2D classification and autopicking were used to generate improved templates for autopicking. An initial model of the apo condition was generated from the autopicked particles using CryoSPARC⁵⁵ and used as the starting model for all data sets. Two-dimensional and three-dimensional classification in Relion were used to eliminate false-positives or damaged particles resulting in 53k, 303k, 49k, 74k, and 170k images for the apo, IP₃-bound, Ca²⁺-bound, low IP₃-Ca²⁺ and high IP₃-Ca²⁺ data sets, respectively. The particle images were polished in Relion to realign particles on a per-particle basis and correct for dose-dependent beam-induced specimen damage.

For the apo condition, the polished particle stack was subjected to one round of three-dimensional classification resulting in a particle stack of 26k homogenous particles. For refinement, the particles were analyzed by cisTEM with C4 symmetry imposed⁵⁶. For the initial global search in cisTEM, the reference map generated in Relion was filtered to 20 Å. For the final iterations, the references were filtered to 5 Å. Using the 0.143 FSC criterion, the resolution was estimated to be 3.49 Å⁵⁷.

To improve the features of the CLD and ARM2 in the apo reconstruction, the stack of 53k particles was subjected to focused three-dimensional classification, resulting in a stack of 30k particles that was symmetry expanded to 120k particles in C1 and the signal was subtracted for the CD and JD of three of the subunits, and the entire TMD. Focused refinement in Relion using a soft mask that included only the CD and JD of one subunit resulted in a reconstruction whose resolution was estimated to be 3.76 Å by the 0.143 FSC criterion.

To improve the features of S1′ and S1″ in the apo reconstruction, the stack of 30k particles was symmetry expanded to 120k particles in C1 and the signal was subtracted for the CD of three of the subunits. Focused refinement in Relion using a soft mask that included the TMD and JD of all four subunits and the CD of one subunit followed by focused classification using a mask that just included the S1–S4 domain of one subunit yielded one major class (81k particles) in which S1′ and S1″ were improved. Focused refinement in Relion of the 81k particles using a soft mask that just included the TMD and JD and all four subunits and the CD of one subunit resulted in a reconstruction whose resolution was estimated to be 3.69 Å by the 0.143 FSC criterion.

For the IP₃ condition, the stack of 303k polished particles was subjected to one round of three-dimensional classification specifying 7 classes without imposing symmetry using a reference in which the four ARM2 domains had been removed. After the first round, one class was poorly resolved (40k particles) and excluded from later processing steps. The remaining particles were merged into 3 groups based upon the positions of the ARM2 domain and reclassified using a reference lacking the ARM2 domains into 7 classes without imposing symmetry. After the second round, the classes were merged into 5 groups based upon the positions of the ARM2 domain. The particles were then reclassified and merged two additional times. The merged particle stacks were then refined without symmetry and classified without angular sampling to remove any misclassified particles. The final cleaned particle stacks were then refined in cisTEM. Classes 1 (39k particles) and 2 (41k particles) were refined in cisTEM with C4 symmetry. For the initial global search in the refinement of classes 1 and 2 in cisTEM, the reference maps generated in Relion were filtered to 20 Å. For the final iterations, the references were filtered to 4.8 Å for class 1 and 6 Å for class 2. Using the 0.143 FSC criterion, the resolutions were estimated to be 3.33 Å for class 1 and 3.82 for class 2. Classes 3, 4 and 5 were refined in cisTEM without symmetry. For the initial global search of classes 3 (38k particles), 4 (10k particles) and 5 (27k particles) in cisTEM, the reference map generated in Relion was filtered to 20 Å. For the final iterations, the references were filtered to 6, 10 and 7 Å, for classes 3, 4 and 5 respectively. Using 0.143 FSC criterion, the resolutions for classes 3, 4 and 5 were estimated to 4.12, 6.01 and 4.47 Å, respectively.

For the Ca²⁺-bound condition, the polished particle stack was subjected to one round of focused three-dimensional classification using a soft mask including ARM3, the JD and the TMD with C4 symmetry imposed resulting in a final particle stack of 34k particles. The particle stack was analyzed in cisTEM with C4 symmetry imposed. For the initial global search in cisTEM, the reference map generated in Relion was filtered to 20 Å. For the final iterations, the references were filtered to 7 Å. Using the 0.143 FSC criterion, the resolution was estimated to be 4.33 Å. To improve the alignment of the more rigid ARM3, JD and TMD, which includes the JD Ca²⁺ binding site, a soft mask was applied to those domains during refinement in Relion, yielding a reconstruction at 4.2 Å. To improve the features of the CD, the stack of 34k particles was symmetry expanded to 135k particles in C1 and the signal was subtracted for the CD of three of the subunits. Focused refinement in Relion using a soft mask including the TMD and JD of all four subunits and the CD of one subunit resulted in a reconstruction whose resolution was estimated to be 4.69 Å by the 0.143 FSC criterion.

For the low IP₃-Ca²⁺ condition, the polished particle stack of 74k particles was classified into 6 classes without imposing symmetry. Four of the six classes, totaling 49k particles, were selected for refinement in cisTEM with C4 symmetry imposed. For the initial global search in cisTEM, the reference map generated in Relion was filtered to 20 Å. For the final iterations, the references were filtered to 7 Å. Using the 0.143 FSC criterion, the resolution was estimated to be 3.96 Å. To improve the alignment of the more rigid ARM3, JD and TMD, which includes the JD Ca²⁺ binding site, a soft mask was applied to those domains during refinement in Relion, yielding a reconstruction at 3.80 Å. To improve the features of the CD, the stack of 196k particles was symmetry expanded to C1 and the signal was subtracted for the JD and CD of three of the subunits, and the entire TMD. Focused classification in Relion using a soft mask that included the JD and CD of one subunit yielded a single major class of 158k particles. Refinement of the 158k particles in Relion using a soft mask that included the JD and CD of one subunit resulted in a reconstruction whose resolution was estimated to be 4.22 Å by the 0.143 FSC criterion.

For the high IP₃-Ca²⁺ condition, the polished particle stack of 170k particles was analyzed in cisTEM with C4 symmetry imposed. One round of three-dimensional classification into seven classes identified six similar classes containing a total of 131k particles. Refinement of these 131k particles in cisTEM resulting in a reconstruction with an estimated resolution of 3.92 Å using reference filtered to 7 Å. To improve the alignment of the more rigid ARM3, JD and TMD, which includes the JD Ca²⁺ binding site, a soft mask was applied to those domains during classification and refinement in Relion, yielding a reconstruction at 3.66 Å. To improve the features of the CD, the stack of 170k particles was symmetry expanded to C1 and the signal was subtracted for the CD of three of the subunits. Focused classification identified a population of 350k particles that when refined in cisTEM using a soft mask that included the JD and CD of one subunit resulted in a reconstruction whose resolution was estimated to be 3.78 Å by the 0.143 FSC criterion.

Model building and coordinate refinement

The structure of ligand-free IP₃-binding domain of rIP₃R1 (PDB: 3UJ4)³⁷ was docked into the full channel apo state density map using UCSF Chimera⁵⁸ and then manually rebuilt according to the density in Coot⁵⁹. The remainder of the protein was built de novo into the apo state full channel, apo state CD focused refinement and S1–S4 focused refinement density maps. The refined apo state model contains residues 5-80, 85-321, 350-527, 534-674, 690-893, 961-1003, 1024-1036, 1043-1129, 1167-1432, 1587-1687, 1717-1804, 1863-2038, 2043-2074, 2111-2226, 2260-2403 and 2449-2611. Unregistered poly-alanine helices were modeled for S1′ and α71-α75 at the C-terminal end of ARM2 using Jpred4⁶⁰. The coordinates were refined against one of the full-channel half-maps (work) using Rosetta 3.7⁶¹ and phenix.real_space_refine⁶² with secondary structure and Ramachandran restraints maintained throughout. To monitor the effects of over fitting during coordinate refinement, Fourier shell correlations were compared against the half-map excluded from refinement (free). Map to model correlations were determined using phenix.mtriage (https://www.phenix-online.org/newsletter/CCN_2017_07.pdf).

For IP₃ classes 1 and 2 and the Ca²⁺-bound structures, the final refined model of apo hIP₃R was docked into the density map using UCSF Chimera and then manually rebuilt according to the density in Coot. The coordinates were refined against one of the full-channel half-maps (work) using phenix.real_space_refine with secondary structure and Ramachandran restraints maintained throughout. To monitor the effects of over fitting during coordinate refinement, Fourier shell correlations were compared against the half-map excluded from coordinate refinement (free). Map to model correlations were determined using phenix.mtriage.

For IP₃ classes 3, 4 and 5, the refined models of IP₃ classes 1 and 2 were docked into the density on a subunit-by-subunit basis by rigid-body fitting in coot. Once the overall fit of the models were completed, the models of IP₃ classes 1 and 2 were split into domain fragments and docked into the density map by rigid-body fitting. The rigid-body fit model was refined by real-space refinement for classes 3 and 5 in phenix.real_space_refine against one of the half-maps and rigid-body refinement for class 4 in phenix.real_space_refine against the full reconstruction.

Software was compiled by SBGrid⁶³. Local resolution estimates were performed using ResMap⁶⁴ and structure figures were prepared with UCSF Chimera and PyMol (PyMol version 1.8.0 Schrodinger LLC).

Data availability

Cryo-EM maps have been deposited in the EMDB under accession codes EMD-7978 for the full channel apo map, EMD-7979 for the apo CD focus refinement map, EMD-7980 for the apo S1–S4 focus refinement map, EMD-7981 for the IP₃ class 1 map, EMD-7982 for the IP₃ class 1 CD focus refinement map, EMD-7984 for the IP₃ class 2 map, EMD-7982 for the IP₃ class 2 CD focus refinement map, EMD-7985 for the IP₃ class 3 map, EMD-7986 for the IP₃ class 4 map, EMD-7987 for the IP₃ class 5 map, EMD-7888 for the full channel Ca²⁺-bound map, EMD-7990 for the Ca²⁺-bound TMD focus refinement map, EMD-7989 for the Ca²⁺-bound CD focus refinement map, EMD-7991 for the full channel low IP₃-Ca²⁺ map, EMD-7993 for the low IP₃-Ca²⁺ TMD focus refinement map, EMD-7992 for the low IP₃-Ca²⁺ CD focus refinement map, EMD-7994 for the high IP₃-Ca²⁺ map, EMD-7995 for the high IP₃-Ca²⁺ TMD focus refinement map and EMD-7996 for the high IP₃-Ca²⁺ CD focus refinement map. Atomic coordinates are available from the RCSB Protein Data Bank under accession codes 6DQJ for the apo structure, 6DQN for IP₃ class 1, 6DQV for IP₃ class 2, 6DQS for IP₃ class 3, 6DQZ for IP₃ class 4, 6DR0 for IP₃ class 5, 6DR2 for the Ca²⁺-bound structure, 6DRA for the low IP₃-Ca²⁺ structure and 6DBC for the high IP₃-Ca²⁺ structure. All other source data are available from the corresponding author upon request.