Conformational landscape and ligand-dependent clustering of the human type 2 IP₃ receptor

Abstract

Inositol 1,4,5-trisphosphate (IP₃) receptors (IP₃Rs) are tetrameric ER Ca²⁺ channels that shape intracellular Ca²⁺ signaling in response to IP₃, regulating diverse physiological processes. The structural basis for subtype-specific regulation among the three subtypes (IP₃R-1–3) remains incompletely understood due to the lack of IP₃R-2 structures. Here, we report cryo-electron microscopy (cryo-EM) structures of human IP₃R-2 in distinct conformations in the presence and absence of IP₃, Ca²⁺, and ATP. These structures define the conformational landscape of IP₃R-2, delineate ligand-binding interactions, and reveal shared architectural features alongside isoform-specific differences. We also resolve ligand-dependent IP₃R-2 assemblies, identifying a conformation-dependent inter-channel interface. Live-cell imaging demonstrates that IP₃R-2 undergoes clustering following ligand-induced Ca²⁺ release, and disruption of this interface selectively abolishes clustering without impairing channel activity. Together, these findings provide a structural framework for human IP₃R-2 and establish a mechanism linking ligand-dependent conformational changes to inter-channel interactions and post-activation cellular clustering.

Here the authors define the conformational landscape of human IP₃R-2 and identify a ligand-dependent inter-channel interface that mediates receptor clustering during calcium signalling.

Introduction

Calcium (Ca²⁺) is a ubiquitous second messenger that regulates diverse physiological processes, including apoptosis, gene expression, and muscle contraction^1,2. Ca²⁺ homeostasis is essential for cell viability, whereas its dysregulation can disrupt cellular functions and lead to disease^3–8. Inositol 1,4,5-trisphosphate (IP₃) receptors (IP₃Rs) integrate diverse inputs to shape Ca²⁺ signals by tightly controlling their spatial and temporal characteristics within cells, thereby directing downstream responses^3–6,9. Primarily located on the endoplasmic reticulum (ER) membrane, IP₃Rs are activated upon IP₃ binding, generated via phospholipase C-mediated hydrolysis of phosphatidylinositol 4,5-bisphosphate (PIP₂) downstream of G protein-coupled receptors (GPCRs) or receptor tyrosine kinases (RTKs)^3–6,9. The resulting channel opening releases Ca²⁺ from ER stores into the cytosol. IP₃Rs also require Ca²⁺ binding to cytosolic activating sites at nanomolar Ca²⁺ concentrations, whereas higher cytosolic Ca²⁺ concentrations inhibit gating, producing a biphasic dependence of channel activity on Ca²⁺ (refs. ^9–14). IP₃R activity is further regulated by small molecules such as ATP^15,16, posttranslational modifications such as phosphorylation¹⁷, and interactions with regulatory proteins¹⁸.

In mammals, the IP₃R family has three subtypes (IP₃R-1, IP₃R-2, and IP₃R-3), encoded by three different genes (ITPR1, ITPR2, and ITPR3, respectively)^19–21. All three subtypes share 60–70% amino acid sequence identity and assemble as homo- or heterotetramers, forming one of the largest intracellular ion channel complexes (~1.2 MDa). Despite this high sequence identity, the subtypes exhibit distinct spatiotemporal expression patterns and physiological properties, resulting in isoform-specific channel behavior. For instance, IP₃R-2 exhibits the highest affinity for IP₃, whereas IP₃R-3 has the lowest^3,20,22–24; IP₃R-1 and IP₃R-2 are more sensitive to potentiation by ATP than IP₃R-3^24–26. Under comparable conditions, channels composed of IP₃R-2 tend to support cell-wide Ca²⁺ oscillations, whereas IP₃R-3-containing channels more often produce large, sustained Ca²⁺ transients with a reduced propensity to oscillate^27–29.

IP₃Rs can assemble into clusters on the ER membrane, shaping the spatial and temporal properties of intracellular Ca²⁺ signaling. Super-resolution imaging of endogenously tagged IP₃Rs has revealed that a subpopulation of immobile IP₃R clusters, residing adjacent to the plasma membrane, serves as the preferential sites for initial Ca²⁺ puff generation³⁰. Furthermore, electrophysiological studies suggest that clustering modulates channel function by altering sensitivity to IP₃ and Ca²⁺ and promoting inter-channel coupling^31–33. In parallel, IP₃Rs undergo dynamic, ligand-dependent clustering upon cellular stimulation, transitioning from a diffuse distribution to larger aggregates^34–37. The molecular and structural basis of these inter-channel interactions and how ligand binding drives clustering remain unclear. In recent years, structural studies, particularly cryo-electron microscopy (cryo-EM), have yielded high-resolution structures of IP₃R-1^9,38–42 and IP₃R-3^43–46 in multiple ligand-bound states, illuminating their gating mechanisms. However, structural information for IP₃R-2 has been notably absent, a gap often attributed to difficulties in expression and purification²⁴.

Here, we report cryo-EM structures of human IP₃R-2 (hIP₃R-2) in distinct conformations in the presence and absence of IP₃, Ca²⁺, and ATP. These structures define the conformational ensembles adopted by hIP₃R-2, delineate ligand-binding interactions, and reveal both conserved and subtype-specific features. Notably, we resolve ligand-dependent assemblies of hIP₃R-2, identifying a conformation-dependent inter-channel interface that mediates channel clustering. Complementary live-cell imaging and mutational analyses demonstrate that disruption of this interface selectively impairs ligand-induced clustering without affecting channel activity. Together, these findings provide a structural basis for inter-channel interactions and establish a mechanism linking ligand-dependent conformational changes to IP₃R clustering and Ca²⁺ signaling.

Results

Expression, purification, and structure determination of hIP₃R-2 in lipid nanodiscs

To enable structural characterization of hIP₃R-2, we synthesized the full-length gene, expressed the protein using Sf9 insect cells, and reconstituted it into MSP1E3D1/DOPC nanodiscs (Fig. 1a). Channel activity was confirmed by planar DOPC/DOPE bilayer recordings. In the presence of IP₃ and ATP, hIP₃R-2 exhibited robust single-channel openings with a biphasic dependence on free Ca²⁺ and maximal open probability (Po) at ~300 nM (Fig. 1b, c).

Fig. 1. Functional and structural characterization of hIP₃R-2.

a Domain boundaries of hIP₃R-2 construct. b Representative single-channel traces of hIP₃R-2 at different Ca²⁺ concentrations. c Plots of open probability as a function of free Ca²⁺ concentration. Error bars correspond to s.e.m. n = 6, 9, 19, 8, 7, and 10 independent recordings for 90 nM, 140 nM, 290 nM, 500 nM, 1.8 µM, and 6.0 µM free Ca²⁺, respectively. Each independent recording corresponds to a newly incorporated channel. Source Data are provided. d Unsharpened consensus map of apo hIP₃R-2. e, f Composite maps of hIP₃R-2. Each domain in one of the subunits is colored as in (a). g Ribbon diagram of the apo hIP₃R-2 showing only two opposing subunits. h The close-up view of the zinc-finger motif. i Close-up view of the TMD (only one subunit is shown). j Close-up view of the TMD as viewed through the pore.

We prepared three cryo-EM samples to capture distinct conformational states under different ligand conditions: (1) ligand-free (apo), (2) with 50 µM IP₃ and 1 mM ATP (IP₃/ATP), and (3) with 50 µM IP₃, 1 mM ATP, and 4 mM CaCl₂ (IP₃/ATP/Ca²⁺). Because the protein buffer contained 5 mM EDTA, the calculated free Ca²⁺ concentration for the IP₃/ATP/Ca²⁺ condition was ~10 nM (MaxChelator), and the apo and IP₃/ATP conditions contained no CaCl₂ intentionally added and were predicted to have negligible free Ca²⁺. We observed decreased image quality in samples containing higher Ca²⁺ concentrations, due to protein aggregation/clustering, which precluded structural characterization of hIP₃R-2 under these conditions. Moreover, the free Ca²⁺ experienced by particles during vitrification likely deviates from calculated values due to trace Ca²⁺ introduced during grid preparation (e.g., from blotting materials) and the limited buffering capacity in the thin aqueous film after blotting^39,44,45. Accordingly, we cannot exclude partial Ca²⁺ occupancy in the apo and IP₃/ATP conditions or higher effective Ca²⁺ levels in the IP₃/ATP/Ca²⁺ sample, which could influence the conformational states observed in the final reconstructions. We therefore refer to the apo and IP₃/ATP conditions as “no added Ca²⁺” (nominally Ca²⁺-free).

Structure of apo hIP₃R-2

Under apo conditions, we identified two major conformational classes, hereafter termed compact and loose, based on the arrangement of the cytosolic domains (Supplementary Fig. 1). The compact class shows well-resolved cytosolic domains that form extensive intersubunit contacts and closely resembles rIP₃R-1 and hIP₃R-3 conformations observed at low Ca²⁺ (refs. ^{38–41,43–46}). 3D refinement of the particles in the compact class with C4 symmetry yielded a high-quality map at 2.95 Å overall resolution (Fig. 1d, Supplementary Fig. 1, and Supplementary Table 1). Symmetry expansion and focused local refinements further improved map quality in select regions (Fig. 1e, f and Supplementary Fig. 2). Given the close similarity to published rat IP₃R-1 (rIP₃R-1) and human IP₃R-3 (hIP₃R-3) apo structures^{40,41,43,45,46} and the absence of density for IP₃, ATP, and Ca²⁺ at expected sites, we designate this reconstruction as the hIP₃R-2 apo state (Fig. 2a). By contrast, the loose class displays weakened or lost cytosolic intersubunit contacts and markedly increased flexibility; notably, the cytosolic domains of two protomers are largely disordered relative to the rest of the protein in this reconstruction (Supplementary Fig. 1). This global “loosened” architecture resembles conformations previously observed for hIP₃R-3 at elevated (including inhibitory) Ca²⁺ concentrations^44–46. The loose class refined without symmetry enforcement (C1) yielded a 3.99 Å map (Supplementary Fig. 1). Refinement strategies that improved the compact class did not markedly improve this class, consistent with increased conformational heterogeneity. Because trace Ca²⁺ during vitrification cannot be excluded, yet the map does not permit confident Ca²⁺ assignment, we refer to this ensemble as inactive-like based on its structural features and similarity to prior hIP₃R-3 conformations, without inferring Ca²⁺-dependent inhibition.

Fig. 2. Comparison of the hIP₃R-2 and hIP₃R-3 structures.

a, b Structures of apo hIP₃R-2 and hIP₃R-3 (PDB ID: 6UQK) overlayed on their TMDs. Domains in one hIP₃R-2 subunit are colored as in Fig. 1, and hIP₃R-3 structure is colored in grey. b The same overlay as in (a), but focusing on the ARM2 domains. c The close-up view of the cytosolic domains of two apo hIP₃R-2 subunits; one shown in ribbon and another in surface representation. The grey mesh shows the cryo-EM density for the subunit shown in ribbon. The region resolved in hIP₃R-2 relative to hIP₃R-3 is colored red. The dashed lines indicate the disordered SBP region.

The overall architecture of apo hIP₃R-2 closely matches that of rIP₃R-1 and hIP₃R-3, forming a characteristic mushroom-shaped assembly with a cytosolic cap and membrane-embedded stem that constitutes the ion conduction pathway^{40,41,43,45,46} (Figs. 1d–g and 2a). A small portion of the stem is exposed to the ER lumen and includes glycosylation sites located within flexible, unresolved regions. The channel is a homotetramer with fourfold symmetry around the pore axis (Fig. 1f). Each subunit adopts the same domain organization as in rIP₃R-1 and hIP₃R-3, comprising two β-trefoil domains (βTF1 and βTF2), three Armadillo repeat domains (ARM1, 2, and 3), a central linker domain (CLD), a juxtamembrane domain (JD), a transmembrane domain (TMD), and a C-terminal domain (CTD) (Fig. 1e–g). The map quality enabled the modeling of most residues; however, regions with no interpretable density were excluded from the final model. Although the presence of a left-handed coiled-coil helical bundle within the CTD is evident in the cryo-EM map, unambiguous assignment of the residues forming this bundle was not possible. Therefore, we modeled this segment as a poly-alanine.

There are many parallels between hIP₃R-2 and rIP₃R-1/hIP₃R-3. Individual domains superpose on their counterparts with Cα RMSD < 1 Å, indicating a conserved fold. As in rIP₃R-1 and hIP₃R-3, a zinc-finger motif within the JD coordinates Zn²⁺ via Cys2562, Cys2565, His2582, and His2587 (Fig. 1h). The TMD adopts a fold similar to voltage-gated ion channels, with S1-S4 forming a pseudo-voltage sensor-like domain (pVSD) and S5-S6 forming the pore, connected by the S4-S5 linker helix (Fig. 1i, j). Unlike canonical voltage-gated channels, the cryo-EM map reveals two additional TM helices between S1 and S2. One of these helices (S1″) is well ordered with clear side chain density, whereas the other (S1′) appears flexible and could not be modeled with confidence (Fig. 1i, j). Analogous helices were observed in rIP₃R-1 and hIP₃R-3 structures, suggesting that they are a conserved feature of IP₃Rs^{38–40,43–46}. The narrowest region of the pore is formed by Phe2537 and Ile2541 on the S6 (Fig. 1j). In the apo state, the solvent-accessible pore radius at this constriction is about 0.7 Å (calculated using HOLE⁴⁷), consistent with a closed gate. As in rIP₃R-1 and hIP₃R-3, the S6 helix contains a π-helical segment on the luminal side of these pore-blocking residues (Fig. 1i).

The apo cryo-EM map reveals two continuous belts of density encircling the TMD on the luminal and cytosolic sides, consistent with the lipid headgroup layers of a bilayer within the nanodisc (Fig. 1d). In addition, within the cavity formed by the pVSDs and pore domain, we observe several strong non-protein densities (Supplementary Fig. 3). We tentatively assign these features as phospholipids based on their size, shape, and location; however, we did not model them because their identity and headgroup orientation cannot be unambiguously determined, as is typical for flexible lipids. Overall, the pattern of annular/bound lipid-like densities around the TMD is highly similar to that reported for rIP₃R-1 reconstituted in nanodiscs⁴⁰, and to ordered lipid/detergent densities observed in prior rIP₃R-1 and hIP₃R-3 reconstructions^{38,39,43–46}. Notably, several of these densities lie adjacent to the S1′/S1″ region, consistent with a potential role in protein-lipid interactions (Supplementary Fig. 3).

Despite the high overall similarity, apparent differences emerge in tetrameric assemblies when apo hIP₃R-2 and apo hIP₃R-3 (PDB 6UQK⁴³) are aligned on their TMDs (Fig. 2a). Most differences are subtle, reflecting rigid-body misalignments among cytosolic domains and likely reflecting intrinsic conformational variability of these regions. The most noticeable difference is in the arrangement of ARM2, which is more tilted toward the adjacent subunit in hIP₃R-2 (Fig. 2b).

We previously showed that a flexible loop (self-binding peptide; SBP) within the ARM2 domain of hIP₃R-3 can occupy the IP₃-binding site of the same subunit⁴³. In the apo hIP₃R-2 consensus map, we do not observe comparable density attributable to an SBP at or near the IP₃-binding site, and the segment linking ARM2-α1 and ARM2-α2 is also unresolved, suggesting high flexibility (Fig. 2c). Notably, additional portions flanking this loop are better ordered in hIP₃R-2 than in hIP₃R-3: three residues from ARM2-α1 project toward βTF2 of the adjacent subunit, consistent with a potential intersubunit contact, and five residues at the N-terminal end of ARM2-α2 bend by ~180° and run roughly parallel to ARM2-α2 while forming intradomain contacts, whereas the intervening loop remains disordered (Fig. 2c and Supplementary Fig. 4). To test whether the potential SBP-like configurations are obscured by averaging, we performed focused 3D classification using a mask encompassing the IP₃ pocket (Supplementary Fig. 4). This analysis revealed heterogeneous, low-resolution density in a subset of classes that approaches or partially occupies the pocket, and in other classes extends toward the neighboring βTF region (Supplementary Fig. 4); however, the local maps were insufficient to assign or model SBP residues unambiguously.

Structure of hIP₃R-2 in complex with IP₃ and ATP

As in the apo condition, the IP₃/ATP dataset yielded compact and loose classes (Supplementary Fig. 5). Sequential 3D classification of compact particles resolved six states: two high-resolution reconstructions (3.31 and 3.33 Å) with apparent fourfold symmetry corresponding to ligand-bound resting and preactivated, and four intermediates (intermediate states 1–4) at 6.22, 6.91, 4.47, and 4.17 Å (Supplementary Figs. 5 and 6 and Supplementary Table 1). Intermediate states 1 and 2 are the lowest-resolution maps and were not analyzed further; intermediate states 3 and 4 supported rigid-body fitting of individual domains and capture asymmetric domain arrangements between the resting and preactivated ensembles. The loose class adopts an inactive-like conformation. In all reconstructions, IP₃ and ATP show well-defined density at their expected binding sites.

In the ligand-bound resting state, hIP₃R-2 adopts an apo-like architecture while exhibiting well-defined densities for IP₃ and ATP, mirroring the resting-state conformation reported for hIP₃R-3^45,46 (Fig. 3a, b). IP₃ occupies the cleft between the βTF2 and ARM1 domains and induces a modest clamshell-like motion, with ARM1 rotating ~4° toward βTF2 (Fig. 3b, c). The most noticeable conformational change is on the loop that harbors Arg269, which moves closer to the IP₃ molecule and forms a salt bridge with the P5 phosphate of IP₃ (Fig. 3c). Additional interactions are predominantly electrostatic contacts between basic residues and the phosphate groups of IP₃: Arg265 (βTF2) with P4; Lys507, Arg510, and Lys569 (ARM1) with P5; and Arg503 and Arg568 (ARM1) with P1. Thr267 (βTF2) and Tyr567 (ARM1) are within hydrogen-bonding distance to P4 and P5, respectively (Fig. 3b).

Fig. 3. Structures of hIP₃R-2 in the presence of IP₃ and ATP.

a Composite cryo-EM maps of hIP₃R-2 in ligand-bound resting (left) and preactivated (right) states. Domains in one subunit are colored as in Fig. 1. Densities for IP₃ and ATP are shown in red and labeled. b Close-up views of the IP₃ binding sites of hIP₃R-2 in resting (left) and preactivated (right) states along with the cryo-EM densities (grey mesh). Residues interacting with IP₃ are shown as sticks. c Comparison of the IP₃ binding sites between apo (orange) and resting state (blue and yellow for ARM1 and βTF2, respectively) on the left and between the resting and preactivated (magenta) states on the right. Structures are aligned on their βTF2, and the rotation of the ARM1 is indicated by a curved arrow. d, e Ribbon representation of the hIP₃R-2 structures superposed on their TMD. One subunit is shown in full; the others are semi-transparent. Domain coloring for the ligand-bound resting state is the same as in Fig. 1. The apo and preactivated structures are colored in orange and magenta, respectively. The black rods denote the rotation and translation axes. Red curved and straight arrows indicate the direction of the rotations and translations, respectively, for the labeled domains. In the preactivated ensemble, ARM2 retracts toward the CLD while the neighboring ARM1/βTF2 cleft closes.

Relative to the resting state, the IP₃-binding network in the preactivated state remains largely unchanged, but ARM1 rotates by an additional ~16° toward βTF2, accompanying coordinated rearrangements across the cytosolic assembly when aligned on the TMD (Fig. 3b–e). The most prominent change involves ARM2. In the apo and resting states, ARM2 adopts an extended conformation in which it forms extensive contacts with ARM1 of the adjacent subunit (Fig. 3d, e). In contrast, the preactivated state ARM2 undergoes a rigid-body rotation of ~60° and 11 Å translation that brings it closer to the ARM1/CLD of the same subunit, yielding a retracted conformation (Fig. 3d, e). This rearrangement reduces intersubunit contacts between ARM2 and ARM1 and strengthens intrasubunit interactions near the ARM1/CLD interface. These changes likely prime the channel for opening, consistent with the preactivated state assignments previously reported for rIP₃R-1^38,39 and hIP₃R-3^44–46.

3D classifications revealed deviations from fourfold symmetry, indicating non-uniform switching across subunits (Supplementary Figs. 5–9). In the resting state, all four subunits adopt an identical conformation: ARM2 is in the extended position, forming intersubunit contacts, and the ARM1/βTF2 cleft surrounding bound IP₃ is not fully closed. In the preactivated state, all four subunits again converge to a single conformation: ARM2 is retracted, strengthening intrasubunit contacts near ARM1/CLD, and the ARM1/βTF2 cleft is fully closed. By contrast, the intermediate states are asymmetric, with individual subunits sampling distinct combinations of ARM2 positioning and IP₃-site closure. For example, in the intermediate state 3 structure, subunit A resembles the resting state (ARM2 extended; ARM1/βTF2 cleft not fully closed) (Fig. 4). Subunit B differs from both resting and preactivated conformations, combining a retracted ARM2 with a cleft that remains partially open. Subunit C matches the preactivated state (ARM2 retracted; cleft fully closed). Subunit D adopts a distinct configuration relative to the other three subunits, maintaining an extended ARM2 while the ARM1/βTF2 cleft is fully closed. In the intermediate state 4 structure, the pattern is likewise asymmetric but distinct from intermediate state 3: subunit C adopts a conformation similar to the preactivated state (ARM2 retracted; cleft fully closed), whereas subunit D resembles subunit C from intermediate state 3 (ARM2 retracted; cleft not fully closed). Thus, the intermediate-state structures illustrate that ARM2 retraction and cleft closure can proceed independently within the same subunit. Notably, the position of ARM2 in one subunit correlates with the degree of cleft closure in the adjacent subunit. Retraction of ARM2 in a given subunit is associated with closure of the neighboring ARM1/βTF2 cleft, suggesting intersubunit coupling. Similar asymmetric subunit rearrangements have been reported for hIP₃R-3^45,46, suggesting a conserved symmetry breaking during the resting-to-preactivated transition.

Fig. 4. Subunit-asymmetric intermediates in hIP₃R-2.

a–d Consensus cryo-EM maps of hIP₃R-2 at ligand-bound resting (a), intermediate state 3 (b), intermediate state 4 (c), and preactivated state (d). Maps are low-pass filtered at 5 Å and colored by domains as in Fig. 1. SubA-D denote individual subunits. Orange arrows indicate ARM1 rotation toward βTF2; red arrows indicate ARM2 retraction toward the CLD. Intermediates display subunit-specific combinations of ARM2 positioning and IP₃-pocket cleft closure, whereas the preactivated ensemble shows a uniform retracted ARM2/closed cleft across all four subunits.

ATP binding site

All reconstructions from the IP₃/ATP dataset show strong, well-defined density for ATP bound within the JD, in the ATP-binding pocket previously identified for rIP₃R-1³⁸ and hIP₃R-3^44,46 (Fig. 5a, b). The adenosine base stacks within a hydrophobic pocket formed by Phe2181, Phe2563, Ile2583, Met2589, and Trp2590, located near the zinc-finger motif, and forms hydrogen bonds with the backbone amide of Phe2563 and the carbonyl group of His2587. The phosphate moieties are coordinated by three basic residues, Arg2174, Lys2177, and Lys2584. Asn2588 is positioned to form hydrogen bonds with the α-phosphate. The ribose ring does not directly interact with protein residues; however, we consistently observe density extending from the ring oxygen across multiple conditions and datasets, plausibly corresponding to a bound water molecule (Fig. 5b).

Fig. 5. ATP binding to the JD.

a Overall structure of hIP₃R-2 in the ligand-bound resting state, highlighting the ATP binding site in the JD. ATP density is shown in red and labeled. b Close-up view of the ATP binding site with the cryo-EM map (grey mesh). Side chains contacting ATP are shown as sticks, and putative hydrogen bonds are indicated by dashed lines. Arg2174 is well resolved and engages the β-phosphate of ATP; in other states, its density is weaker but remains oriented toward the phosphate moieties. c Comparison of the ATP site in hIP₃R-2 (JD, cyan; TMD, magenta) and hIP₃R-3 (PDB 7T3P, grey), aligned on the JD. Residues are labeled for both subtypes (hIP₃R-2 above, hIP₃R-3 below).

IP₃R-2 has the highest affinity for ATP among IP₃R subtypes, whereas IP₃R-3 has the lowest^24–26. Based on hIP₃R-3 structures and sequence alignment, we previously proposed that a charge difference in a residue near the phosphate moieties contributes to this disparity⁴⁴. Structural alignment of the JD from hIP₃R-2 and hIP₃R-3 shows that the ATP-binding pocket is highly similar and that ATP adopts a comparable pose in both (Fig. 5c). As expected, the most notable difference is Arg2174 in hIP₃R-2, which corresponds to Glu2149 in hIP₃R-3. In the ligand-bound resting state, the side chain of Arg2174 is clearly resolved and engages the β-phosphate of ATP. In other structures, the side chain is less well defined but remains oriented toward the phosphate moieties. It is plausible that the additional electrostatic stabilization/coordination of the phosphate moiety of ATP by Arg2174 in hIP₃R-2 versus Glu2149 in hIP₃R-3 yields tighter binding. These observations support our prior proposal that this residue difference partly explains the higher ATP affinity of IP₃R-2 relative to IP₃R-3.

The structure of hIP₃R-2 in an inactive-like state

In both the apo and IP₃/ATP datasets, a substantial fraction of particles fall into a loose class in which cytosolic intersubunit contacts are weakened or lost (Supplementary Figs. 1 and 5). This architecture parallels the hIP₃R-3 conformation that increases with Ca²⁺ and predominates at inhibitory concentrations (~2 mM)^44–46, and we therefore refer to it as inactive-like. No Ca²⁺ was intentionally added prior to grid preparation in either condition; however, residual free Ca²⁺ in thin films cannot be excluded (see “Methods”).

In the IP₃/ATP/Ca²⁺ dataset, the vast majority of particles adopted the loose conformation, with only a small fraction populating the compact class, yielding a low-resolution map insufficient for high-resolution refinement (Supplementary Figs. 10). The loose class particles segregated into two subclasses: class 1 closely resembles the loose class observed without added Ca²⁺, whereas class 2 exhibits additional features that will be discussed in the next section (Supplementary Figs. 10–13). We did not observe pore dilation consistent with an open state in either subclass.

We first compared the loose classes from the IP₃/ATP and IP₃/ATP/Ca²⁺ (class 1) datasets (Fig. 6). Consensus reconstructions from the global refinement are highly similar (Fig. 6a). The JDs and TMDs are well defined, whereas resolution decreases markedly for the remainder of the protein. Cytosolic domains are resolved in two subunits but largely invisible in the other two, indicating greater conformational heterogeneity. Local refinement improved the map quality for visible cytoplasmic domains, but not for the invisible ones (Supplementary Figs. 9 and 11). Close inspection of the maps from the IP₃/ATP/Ca²⁺ dataset reveals substantial spherical density at the ARM3-JD interface, corresponding to the activatory Ca²⁺ binding site reported for rIP₃R-1^38,48 and hIP₃R-3^44,46 (Fig. 6b). The site and coordination are consistent with previous structures. Ca²⁺ is coordinated by Glu1930 and Glu1994 in ARM3, and the backbone carbonyl of Thr2605 in JD, with His1932 and Gln1997 in proximity and potentially involved via water-mediated contacts. Occupancy of this site is associated with a ~10° clamshell-like closure of JD relative to ARM3 (Fig. 6c). In the IP₃/ATP dataset (no added Ca²⁺), density is also present at the same location but is substantially weaker (Fig. 6b), and the ARM3-JD arrangement closely matches the Ca²⁺-bound reconstruction (Fig. 6c), suggesting potential partial occupancy. However, focused 3D refinement yielded an even weaker peak (Supplementary Fig. 12), precluding confident ion assignment in this condition. Because trace Ca²⁺ during grid preparation cannot be excluded, the weak density in the IP₃/ATP dataset could reflect partial Ca²⁺ occupancy or local disorder at the site. Accordingly, we avoid mechanistic conclusions that require this sample to be strictly Ca²⁺-free.

Fig. 6. Inactive-like structures of hIP₃R-2.

a Sharpened consensus maps of the inactive-like structures from IP₃/ATP (left) and IP₃/ATP/Ca²⁺ conditions. b Close-up view of the activatory Ca²⁺ binding site at the ARM3-JD interface along with the cryo-EM maps (grey mesh) from consensus refinements from IP₃/ATP (left) and IP₃/ATP/Ca²⁺ (right) datasets. In the IP₃/ATP/Ca²⁺ dataset, a spherical density at the site is consistent with Ca²⁺ occupancy, while the comparable ion density (labeled with red *) is substantially weaker in the IP₃/ATP dataset, and ion identity is not assigned. c Comparison of ARM3-JD across ensembles: preactivated (magenta), inactive-like IP₃/ATP (orange), and inactive-like IP₃/ATP/Ca²⁺ (ARM3, yellow; JD, cyan), aligned on ARM3. The black rod denotes the JD rotation axis relative to ARM3.

A second Ca²⁺ binding site at the CLD-ARM2 interface has been reported in Ca²⁺-inhibited structures of hIP₃R-3^45,46, and mutational analysis suggests that this site contributes, at least in part, to Ca²⁺-dependent inhibition of human IP₃R-1 channels⁴⁹. In our hIP₃R-2 maps, we do not observe density that can be attributed with high confidence to a bound Ca²⁺ ion at the analogous CLD-ARM2 site or elsewhere.

Ligand-dependent clustering of hIP₃R-2

About half of the particles from the IP₃/ATP/Ca²⁺ dataset (class 2) exhibited substantial density adjacent to a complete hIP₃R-2 tetramer (Supplementary Fig. 10). Inspection of this additional density indicated that it corresponds to a second hIP₃R-2 channel interacting with the first. To define the nature of this interaction, we re-extracted the particles with a larger box size to accommodate both channels (Fig. 7a and Supplementary Fig. 13). Despite the limited global resolution, the resulting map clearly reveals two hIP₃R-2 tetramers interacting through their cytosolic domains with a twofold symmetry (Fig. 7a). Rigid body fitting of the structural models indicates that the channels forming the complex adopt a loose (inactive-like) conformation. In this arrangement, the TMDs of both channels align along a nearly planar surface, consistent with a geometry compatible with co-residence in the ER membrane.

Fig. 7. Clustering interface of hIP₃R-2.

a Consensus cryo-EM map of hIP₃R-2 dimer viewed from the cytosol (left) and along the membrane plane (right). The TMDs are nearly coplanar with a small tilt relative to a common membrane plane, consistent with ER membrane geometry. Red boxes indicate the viewed regions in (b–e). b, c Locally refined cryo-EM map (b) and ribbon representation (c) of the domains forming the dimerization interface, viewed as in (a). Domains are colored as in Fig. 1. Asterisk (*) denotes the approximate twofold symmetry axis. d, e Close up view of the boxed regions in (c). Select residues are shown as sticks and labeled. Labels with a prime (′) denote residues and domains from the partner channel. Red sphere indicates the Cα atom of G706. f Representative 2D class averages revealing higher-order assemblies involving three or more channels. These classes support a shared interface but did not yield 3D reconstructions in the current dataset.

Although the dimer map was not sufficiently resolved for de novo model building across the entire assembly, the initial reconstruction generated with the smaller box size provided a substantially better map for one subunit and for the domains of the second channel that form the inter-channel interface (Supplementary Figs. 10 and 13). To enhance local map quality at the putative clustering interface, we performed focused refinements using masks spanning ARM1, βTF1, and βTF2. This approach yielded a local resolution of ~3.5 Å at the interface, permitting side-chain assignment for most residues (Fig. 7b and Supplementary Fig. 13).

The two channels interact through their ARM1 and βTF1 domains, forming a defined and predominantly hydrophilic interface with approximate twofold symmetry (Fig. 7b, c). Near the symmetry axis, loops comprising residues Asp540, Asp543, Gln544, and Arg545 from both subunits are in close proximity, but their contribution to the stability of the complex appears limited, as they bury about 10% of the total interfacial area (Fig. 7c, d). The dominant contacts flank this central region and are formed by ARM1 on one channel engaging βTF1 on the other, accounting for the remainder of the interface (Fig. 7c, e). In total, the interface buries ~2800 Ų (~1400  Ų per channel), as computed with NACCESS⁵⁰. The three-stranded β-sheet within ARM1 (residues 667–713) interacts with the loop comprising residues 135–144 on βTF1, forming the core of the interface. Within the β-sheet, the loop comprising residues 676–689 is not resolved in the cryo-EM maps, suggesting greater flexibility, but it may also contribute to the complex formation. Additional contacts within this region are observed through the two α-helices extending from the βTF1 toward the ARM1 (Fig. 7c, e). This region is less well-defined, and residues 79–85 are completely invisible on the maps.

While 3D analysis revealed a dimeric arrangement, 2D classifications indicate higher-order assemblies involving three or more hIP₃R-2 channels (Fig. 7f). Although 3D reconstructions of these larger assemblies were not attainable with the current dataset, the 2D class averages indicate that higher-order clustering is mediated by the same interaction mode observed in the dimeric structure. Notably, the TMDs of these channels in the assemblies are aligned on a nearly planar surface, supporting the plausibility of such arrangements within the ER membrane.

Fluorescence microscopy studies across diverse cell types have shown that stimulation with IP₃-generating agonists induces redistribution of IP₃Rs from a diffuse, reticular pattern into large punctate aggregates^34–37. These clusters persist for the duration of the stimulus and are reversible upon its removal. Previous studies have suggested that clustering requires channel activation and is mediated by protein-protein interactions, but does not depend on large-scale ER restructuring^35–37.

Our cryo-EM analysis indicates that the clustered assemblies observed here can only form in the presence of IP₃ and Ca²⁺, as alternative conformations would result in steric clashes. The planar arrangement of the TMDs further supports compatibility with the ER membrane. These observations led us to hypothesize that the ARM1-βTF1 interface identified in the cryo-EM structures underlies ligand-induced clustering of IP₃R-2 in cells.

To test this hypothesis, we generated a construct in which monomeric EGFP (mGFP) was fused to the C-terminus of hIP₃R-2 (hIP₃R-2-mGFP) and performed live-cell fluorescence imaging in IP₃R-null HEK293 (HEK293-3KO) cells transfected with this construct. These cells lack endogenous IP₃Rs due to CRISPR-mediated gene knockout⁵¹. Cells were also loaded with Cal-590 to monitor intracellular Ca²⁺ release as a readout of channel activity. Before stimulation, hIP₃R-2-mGFP fluorescence showed a fine reticular distribution throughout the cytoplasm, consistent with ER localization, with occasional puncta. Stimulation with carbachol, an acetylcholine analog that activates muscarinic receptors, resulted in rapid Ca²⁺ release into the cytosol followed by sustained clustering of IP₃R-2-mGFP into discrete puncta (Fig. 8a). Similar ligand-induced clustering was observed in wild-type HEK293 cells expressing endogenous IP₃Rs and transfected with hIP₃R-2-mGFP, confirming that clustering is not dependent on the IP₃R-null background (Supplementary Fig. 14a).

Fig. 8. Ligand-induced clustering of hIP₃R-2 in cells depends on the ARM1-βTF1 interface.

a Live-cell fluorescence imaging of HEK293-3KO cells expressing hIP₃R-2-mGFP (green) and loaded with Ca²⁺ indicator Cal-590 (red). Images are shown before and after stimulation with carbachol at the indicated time points. Prior to stimulation, hIP₃R-2-mGFP displayed a predominantly reticular distribution throughout the cytoplasm, with occasional puncta. Upon stimulation, the fluorescence redistributed into prominent clusters, coincident with Ca²⁺ release. Insets (5X) show the magnified views of the indicated regions. Similar results were observed in nine independent transfections. b Same as in (a), but for G706K mutant. The mutant displayed a similar reticular distribution prior to stimulation and did not show a comparable increase in clustering following stimulation, despite robust Ca²⁺ release. Similar results were observed in nine independent transfections. c Same as in (a), but for P704K mutant. The mutant displayed a similar reticular distribution prior to stimulation and showed stimulation-induced clustering comparable to wild-type hIP₃R-2. Similar results were observed in four independent transfections. Scale bars, 2 µm.

To identify mutations predicted to disrupt the inter-channel interface without perturbing overall channel architecture or function, we performed Rosetta-based computational analysis of the dimer interface. This analysis identified G706, located within the ARM1 β-sheet at the core of the interface (Fig. 7e), as a key hotspot residue. Substitutions at this position produced varying degrees of interface destabilization (positive ΔΔG, reported in Rosetta energy units [REU]) while minimally affecting monomer stability. Among these, G706K was selected for experimental validation based on its strong predicted disruption of dimerization (ΔΔG = 808 REU). We also selected P704K (ΔΔG = 98 REU), which is located two residues away but positioned at the periphery of the interface rather than the buried core occupied by G706 (Fig. 7e).

In cells, the G706K mutant displayed a reticular distribution prior to stimulation, indistinguishable from wild-type IP₃R-2-mGFP (Fig. 8b and Supplementary Fig. 14b). However, stimulation with carbachol did not result in a comparable increase in clustering in any of the cells visualized, despite robust Ca²⁺ release confirming preserved channel activity (Fig. 8b and Supplementary Fig. 14b). In contrast, the P704K mutant exhibited ligand-induced clustering comparable to wild-type following carbachol stimulation, while also supporting normal Ca²⁺ release (Fig. 8c). The contrasting phenotypes of these two mutations demonstrate that the loss of clustering is specifically due to steric disruption at the buried core of the interface, where the small glycine residue is required for tight packing. The functional dissociation between clustering and Ca²⁺ release demonstrates that G706K specifically disrupts inter-channel interactions without impairing channel gating. Consistent with this, FSEC analysis showed no substantial change in the elution profile of the G706K mutant relative to the wild-type receptor, supporting proper tetrameric assembly (Supplementary Fig. 15). Together, these results establish that the ARM1-βTF1 interface identified by cryo-EM is required for ligand-induced hIP₃R-2 clustering in cells. While these experiments do not exclude contributions from additional cellular factors, they demonstrate that disruption of the identified interface is sufficient to prevent clustering under these conditions.

Discussion

Here, we determined the cryo-EM structures of hIP₃R-2 across multiple ligand conditions, defining compact (apo, resting, and preactivated), loose (inactive-like), and asymmetric intermediates. These structures delineate IP₃ and ATP recognition and support a mechanism in which ARM2 retraction is coupled to closure of the adjacent ARM1/βTF2 cleft. We also observe ligand-dependent dimerization and higher-order clustering mediated by ARM1/βTF1 contacts with nearly planar alignment of TMDs, consistent with the endoplasmic reticulum geometry.

hIP₃R-2 shares the core architecture of rIP₃R-1 and hIP₃R-3, with a modest isoform-specific tilt of ARM2 in the apo state compared to hIP₃R-3 (Fig. 2). The IP₃-binding sites in all three subtypes are highly similar and undergo similar rearrangements upon IP₃ binding^{38,39,42,44–46,52–54}. Likewise, global conformational changes upon ligand binding are consistent with a conserved mechanism for activation and gating. At the same time, physiological differences among the subtypes likely arise from subtle distinctions that shift ligand sensitivity and the kinetics of conformational transitions. One such difference is observed at the ATP binding site, where the presence of Arg2174 in hIP₃R-2 versus Glu2149 in hIP₃R-3 provides additional electrostatic stabilization/coordination of the phosphate moiety of ATP, potentially yielding tighter binding (Fig. 5c). hIP₃R-2 also has the highest affinity for IP₃; however, our analysis did not identify a specific structural feature that explains this difference.

Inspired by our previous work on hIP₃R-3 reporting SBP engagement at the IP₃-binding site⁴³, we performed focused 3D classification of the apo hIP₃R-2 IP₃-binding pocket to assess whether a similar interaction occurs. Our analysis reveals heterogeneous density consistent with the SBP adopting distinct conformations, including one that approaches the IP₃-binding site and another that extends toward the βTF region of the adjacent subunit. However, the limited resolution of the local maps precludes unambiguous assignment of SBP residues, and the functional significance of these conformations remains to be determined.

Cryo-EM reconstructions of hIP₃R-2 in the presence of IP₃ and ATP reveal conserved IP₃ recognition across distinct protein conformational states (Fig. 3). In a ligand-bound resting state, IP₃ induces local changes, including a modest clamshell motion (ARM1 rotates ~4° toward βTF2), and the overall architecture resembles apo. In the preactivated state, ligand contacts are preserved, but the cytosolic assembly rearranges substantially: ARM2 retracts toward the CLD and the adjacent ARM1/βTF2 cleft fully closes with an additional ~16° ARM1 rotation. Asymmetric intermediates show that ARM2 retraction and cleft closure can occur independently within subunits, while ARM2 retraction in one subunit correlates with closure of the neighboring cleft, indicating intersubunit coupling. The coexistence of resting, preactivated, and asymmetric intermediates under saturating IP₃ suggests that additional determinants modulate the conformational equilibrium.

IP₃-induced rearrangements in hIP₃R-2 closely parallel those reported for rIP₃R-1^38,39 and hIP₃R-3^44–46, suggesting a conserved trajectory from resting to preactivated toward pore opening. In hIP₃R-3, Ca²⁺ binding at the activating ARM3-JD interface correlates with pore opening, with the JD playing an essential role in this coupling^44,46. Recent mutagenesis studies have further identified a conserved leaflet domain within the JD as essential for gating in mouse IP₃R-1⁵⁵. Although we did not capture a clearly dilated pore in hIP₃R-2, the similarity of IP₃-driven rearrangements supports a comparable pathway. In the IP₃/ATP/Ca²⁺ dataset, we observe density at the activating ARM3-JD site consistent with Ca²⁺ occupancy; nevertheless, these particles adopt a loose, inactive-like conformation. Unlike hIP₃R-3^45,46, we did not detect density for additional Ca²⁺-binding sites in these inactive-like structures. Absence of detectable density does not necessarily imply absence of binding, as Ca²⁺ density can be obscured by imperfect coordination or local flexibility. Accordingly, it remains uncertain whether the loose structures represent desensitized or Ca²⁺-inhibited states, and we conservatively refer to them as inactive-like, pending further structural or functional evidence.

In addition to defining the conformational landscape of hIP₃R-2, our structures provide insight into ligand-dependent inter-channel interactions that may contribute to IP₃R clustering. Under IP₃/ATP/Ca²⁺ conditions, we observe dimeric and higher-order assemblies in which neighboring channels interact through a defined cytosolic interface, primarily mediated by ARM1 and βTF1 domains. This interface buries a substantial area and involves specific side chain contacts, arguing against nonspecific aggregation. Notably, the interface appears to become accessible only upon ligand-dependent conformational changes and is not a dominant feature in the apo or IP₃/ATP datasets, despite identical protein preparation and reconstitution conditions. In all observed assemblies, the TMDs of adjacent channels align on a nearly planar surface, consistent with a geometry compatible with co-residence in the ER membrane. Dimeric assemblies have also been observed previously for hIP₃R-3 in the presence of IP₃/ATP/Ca²⁺; however, the reported arrangement was structurally distinct, with TMD orientations that were not readily compatible with a planar ER membrane organization⁴⁶.

IP₃Rs have been shown to assemble into discrete groups and exhibit distinct physiological properties compared to individual receptors^31–34,37. Such clustering is central to the generation of Ca²⁺ puffs, which are localized and transient ER Ca²⁺ release events arising from coordinated channel opening within a cluster^29,33. While these observations establish a clear link between clustering and signal amplification, the molecular mechanism underlying cluster formation and inter-channel communication has remained poorly defined. Previous studies have suggested that IP₃R clustering may involve auxiliary proteins, cytoskeletal interactions, and spatial organization within specialized ER domains, particularly near ER-plasma membrane junctions^30,56,57.

In addition to these assemblies, IP₃Rs also undergo a distinct, dynamic clustering event upon cellular stimulation, characterized by rapid redistribution from a diffuse reticular pattern into larger punctate aggregates^34–37. The relationship between these stimulus-induced clusters and the smaller functional assemblies inferred from electrophysiological and imaging studies remains unclear, including whether they represent different organizational states or stages within a common pathway. Our structural and cellular data provide a potential mechanistic framework for this process by identifying a conformation-dependent inter-channel interface that is required for ligand-induced clustering. In this model, ligand binding promotes structural rearrangements that expose or stabilize the ARM1-βTF1 interface, enabling direct channel-channel interactions and higher-order assembly.

Although the clustering observed in our reconstituted system likely represents a minimal structural unit and does not capture the full complexity of the cellular environment, it suggests that intrinsic protein-protein interactions within IP₃Rs are sufficient to drive cluster formation. In cells, these interactions may be further modulated by accessory factors and membrane organization, ultimately shaping the size, stability, and functional properties of IP₃R clusters. Notably, identification of the G706K mutation, which disrupts the inter-channel interface without affecting channel structure or activity, provides a means to selectively uncouple ligand-induced clustering from channel function and offers a valuable tool for probing the role of clustering in IP₃R signaling.

In conclusion, the cryo-EM structures presented here establish a framework for understanding hIP₃R-2 regulation across multiple ligand conditions. By defining its conformational landscape, ligand-binding interactions, asymmetric intermediates, and ligand-dependent inter-channel assemblies, our results reveal the structural underpinnings of IP₃R-2 gating and spatial organization. The experimental validation of the inter-channel interface through side-directed mutagenesis and live-cell imaging demonstrates that these structurally identified contacts are required for ligand-induced clustering in cells. Future work integrating these structural findings with electrophysiological characterization of gating transitions, identification of additional clustering factors, and comparative analysis across IP₃R isoforms will be important for establishing a complete mechanistic understanding of IP₃R signaling and its role in intracellular Ca²⁺ dynamics.

Methods

Construct design and cloning

The full-length hIP₃R-2 coding sequence (ITPR2; UniProt Q14571) was codon-optimized for expression in Spodoptera frugiperda (Sf9) cells and chemically synthesized (Gene Universal Inc.) as five overlapping double-stranded DNA fragments (1,966, 1,958, 1,957, 1,958, and 425 bp). Fragments were assembled by overlap-extension PCR using primers annealing to the 5′ end of fragment 1 and the 3′ end of fragment 5 to yield the complete open reading frame (ORF). The assembled ORF was fused at its C terminus to a thrombin cleavage site and a Twin-Strep tag, and subcloned into the pACEBac1 vector using BamHI and XbaI (Supplementary Data 1). A Kozak consensus sequence was incorporated immediately upstream of the start codon to enhance expression in Sf9 cells. The accuracy of the final constructs was verified by Sanger sequencing before being incorporated into baculovirus using the MultiBac expression system⁵⁸.

For fluorescence microscopy and FSEC analysis, an hIP₃R-2-mGFP fusion construct was generated by fusing mGFP (EGFP A206K mutant lacking the initiating methionine) to the C-terminus of hIP₃R-2 via a 17-amino-acid linker (ASNGLVPRGSAAAQNNA). The construct was cloned into a modified pACEBac1 vector in which the polyhedrin promoter was replaced with a CMV promoter and the 3′ untranslated region was replaced with WPRE sequence. The sequence of the entire plasmid (pAW-CMV-hIP₃R-2-mGFP) was verified by next-generation sequencing. The P704K and G706K mutations were introduced by site-directed mutagenesis using the following primers: P704K, 5′-CAAAGAGAAGCACGGCAAGGCTATC-3′ and 5′-TGCCGTGCTTCTCTTTGTTGGAGTC-3′; G706K, 5′-GCCTCACAAGAAGGCTATCCGTCAC-3′ and 5′-AGCCTTCTTGTGAGGCTCTTTGTTG-3′. The complete plasmid sequences were verified using next-generation sequencing.

Protein expression and purification

Sf9 cells at 3 × 10⁶ cells/mL were infected and harvested 48 h post-infection by centrifugation (2000×g, 10 min, 4 °C). Pellets were washed once in PBS, collected again (3000×g, 10 min, 4 °C), flash-frozen in liquid nitrogen, and stored at −80 °C.

For purification, frozen pellets were thawed on ice and resuspended in lysis buffer (Buffer A: 200 mM NaCl, 40 mM Tris-HCl pH 8.0, 5 mM EDTA pH 8.0, 2 mM DTT) supplemented with 1 mM PMSF. Cells were disrupted using an Avestin EmulsiFlex-C3. The lysate was clarified at 7000×g for 10 min, and membranes were collected by ultracentrifugation at 185,000×g for 1 h (Beckman Coulter Type 45 Ti rotor). Membrane pellets were homogenized in ice-cold buffer A and solubilized with 0.5% lauryl maltose neopentyl glycol (LMNG; Anatrace, cat. NG310) and 0.1% glycodiosgenin (GDN; Anatrace, cat. GDN101) at ~100 mg/mL membrane concentration for 4 h with gentle mixing. Insoluble material was removed by ultracentrifugation (185,000×g for 1 h), and the supernatant was applied by gravity to Strep-Tactin XT 4Flow resin (IBA Lifesciences). The resin was washed with 20 column volumes of buffer A containing 0.5% CHAPS (Anatrace, cat. C316S), and bound protein was eluted in the same buffer supplemented with 100 mM D-biotin (pH 8.0). Peak fractions containing hIP₃R-2 were pooled and polished by size-exclusion chromatography (SEC) on a Superose 6 Increase 10/300 GL column (Cytiva) equilibrated in SEC buffer (200 mM NaCl, 20 mM Tris-HCl pH 8.0, 5 mM EDTA pH 8.0, 2 mM DTT, 0.5% CHAPS). The sample was used immediately for nanodisc preparation.

The membrane scaffold protein MSP1E3D1 expression vector (pMSP1E3D1; Addgene plasmid # 20066; http://n2t.net/addgene:20066; RRID: Addgene_20066) was a gift from Stephen Sligar⁵⁹. MSP1E3D1 was produced in E. coli BL21(DE3) cells and purified by Ni²⁺-chelate affinity chromatography as described previously⁶⁰. The histidine tag was removed by TEV protease digestion, and the protein was further purified by SEC using a HiLoad 16/600 Superdex 200 pg column (Cytiva) equilibrated with 300 mM NaCl and 40 mM Tris-HCl, pH 8.0. Peak fractions were pooled, aliquoted, flash-frozen in liquid nitrogen, and stored at −80 °C.

Nanodisc incorporation of hIP₃R-2

1,2-dioleoyl-sn-glycero-3-phosphocholine (DOPC), dissolved in 1% CHAPS, was mixed with hIP₃R-2 detergent micelles at a lipid-to-protein molar ratio of 200:1 and incubated for 30 min at 4 °C with gentle agitation. MSP1E3D1 was then added at a molar ratio of 4:1 (MSP1E3D1:hIP₃R-2), and the mixture was incubated for an additional hour at 4 °C. The detergent (CHAPS) was removed by dialysis overnight at 4 °C against 2 L of SEC buffer lacking detergent (200 mM NaCl, 20 mM Tris-HCl pH 8.0, 5 mM EDTA pH 8.0, 2 mM DTT), allowing nanodisc self-assembly. hIP₃R-2 nanodiscs were isolated by SEC on a Superose 6 Increase 10/300 GL column (Cytiva) equilibrated in the same detergent-free buffer. Fractions corresponding to hIP₃R-2 nanodiscs were pooled and concentrated to ~4 mg/mL using a 100 kDa centrifugal filter (Millipore), clarified by ultracentrifugation (260,000×g, 10 min; ThermoFisher S110AT rotor). The concentration typically dropped to ~2.5 mg/mL.

Planar lipid bilayer electrophysiology

Single-channel recordings of hIP₃R-2 were performed using MECA-4 chips with 100 µm cavities (Ionera Technologies) on an Orbit Mini system equipped with a temperature control unit (Nanion Technologies)⁶¹. Planar lipid bilayers were formed using the air-bubble technique from a DOPC/DOPE mixture (5:3 molar ratio, Avanti Polar Lipids) dissolved in n-decane (Sigma-Aldrich) to a final concentration of 20 mg/mL. Chips were filled with 150 µL of symmetrical recording buffer containing 20 mM HEPES (pH 7.4), 150 mM KCl, 2 mM TCEP, 1 mM ATP, 0.5 mM MgCl₂, 10 µM IP₃, 5 mM EGTA, and varying CaCl₂ to reach desired free Ca²⁺ concentrations. Cavities were wetted prior to bilayer formation. A lipid-coated pipette tip was used to form an air bubble beneath the buffer surface near a cavity. The bubble was slowly expanded and retracted to paint the lipid monolayer. Bilayers were ruptured and reformed several times until a single, stable bilayer was achieved, which was confirmed by monitoring its capacitance and conductance. Channel incorporation was performed using hIP₃R-2 nanodiscs following an approach described in refs. ^62,63. In total, 0.5 µL of hIP₃R-2 reconstituted in nanodiscs (20 µg/mL in storage buffer containing 20 mM HEPES pH 7.4, 150 mM KCl, 2 mM TCEP, 5 mM EGTA, and 2.5% glycerol) was added near the cavities. The solution was gently mixed, and a holding potential of +40 mV was applied across the membrane to promote fusion. Free Ca²⁺ concentrations were calculated using MaxChelator and measured using a calcium-selective electrode (World Precision Instruments) calibrated with a commercial calcium buffer kit (11 standards ranging from 10 nM to 10 mM; World Precision Instruments). Experiments were repeated using protein from at least three different purifications for each condition tested.

All recordings were performed at 22 °C under a holding potential of +60 mV. Signals were acquired at a sampling rate of 10 kHz and a final bandwidth of 1 kHz. Single-channel analysis, including the calculation of open probability (Po), was conducted using Clampfit (v10.7). For analysis and presentation, currents were digitally filtered at 500 Hz and 200 Hz, respectively⁶⁴.

Cryo-EM sample preparation and data collection

Apo hIP₃R-2 nanodiscs (~2.5 mg/mL) were supplemented immediately prior to vitrification with 0.025% (w/v) fluorinated Fos-Choline-8 (FFC-8; Anatrace, cat. F300F). For IP₃/ATP and IP₃/ATP/Ca²⁺ samples, hIP₃R-2 nanodiscs were supplemented with 50 μM IP₃ (from a 10 mM stock in water), 1 mM ATP (from a 100 mM stock, pH 7.2), either without CaCl₂ (IP₃/ATP) or with 4 mM CaCl₂ (IP₃/ATP/Ca²⁺), followed by the addition of 0.025% (w/v) FFC-8 immediately prior to plunging. In the presence of 5 mM EDTA, the calculated free Ca²⁺ for the IP₃/ATP/Ca²⁺ mixture is ~10 nM (MaxChelator). We note that the free Ca²⁺ during grid preparation may deviate from calculated values due to trace Ca²⁺ leaching and the limited buffering capacity in the small volumes applied to grids.

For grid preparation, 2.5 μL of sample was applied to Quantifoil R1.2/1.3 Cu 300 mesh grids (Electron Microscopy Sciences) glow-discharged for 30 s at 25 mA. Grids were blotted for 4 s at blot force 10 using two layers of PELCO 595 filter paper (Ted Pella, cat. 47000-100) and plunge-frozen into liquid ethane with a Vitrobot Mark IV (Thermo Fisher) set to 8 °C and 100% humidity. Filter papers were not pretreated with Ca²⁺ chelators or other chemicals. Optimization of the plunging conditions was performed by imaging the grids using Glacios microscope (Thermo Fisher).

All data collection was performed using Titan Krios G4i (Thermo Fisher) equipped with a BioQuantum K3 detector at Vanderbilt cryo-EM facility. Movies were recorded as 50-frame exposures at a nominal magnification of 105,000×, yielding a calibrated pixel size of 0.822 Å/pixel using the automated imaging software EPU (Thermo Fisher). The total dose was 50.2-55.5 e⁻/Ų. Four to five shots per hole were acquired, and defocus values ranged from −0.8 to −2.2 μm. Apo and IP₃/ATP/Ca²⁺ datasets were collected from single grids, whereas the IP₃/ATP dataset was collected from three grids.

Cryo-EM data processing

All image processing was performed using CryoSparc (v4.6.2 to v4.7.1)⁶⁵. Motion correction and CTF estimations were performed locally using Patch Motion Correction and Patch CTF Estimation. Micrographs with poor image properties (thin or thick ice, CTF fit resolution worse than 6 Å, large drift) were excluded. Initial particle picking was performed by blob search and particles were then binned 4x and extracted (Supplementary Figs. 1, 5, and 10). After 2D classification, classes with clear structural features were re-extracted at full size (512px). Ab initio reconstructions with 4 classes (apo) or 6 classes (IP₃/ATP and IP₃/ATP/Ca²⁺) were performed, followed by heterogeneous refinement. High-quality particles from these reconstructions were used to train a Topaz picker⁶⁶ (200 micrographs), and a new particle set was selected using Topaz⁶⁶ (v0.2.5a) and processed as above. This process was repeated three times to maximize particle yield. All particles from each iteration were merged into a single set after removing the duplicate particles, re-extracted at full size (512px), and subjected to a final heterogeneous refinement. Classes with unique features are grouped and refined using non-uniform (NU) refinement⁶⁷. After refinement, reference-based motion correction⁶⁸ and local CTF refinement⁶⁹ were performed, followed by another round of NU refinement (Supplementary Figs. 1, 5, and 10).

Under apo conditions, particles segregated into two major classes: compact and loose (Supplementary Fig. 1). The compact class, refined with C4 symmetry, yielded a reconstruction at 2.95 Å average resolution that we designate as the apo state. The map quality for the loose class was considerably lower, precluding model building. Nevertheless, its overall architecture is consistent with an inactive-like conformation.

To assess the SBP occupancy and conformations, we applied C4 symmetry expansion to the apo particle stack, subtracted density for all regions except the N-terminal cytosolic assembly of a single subunit (βTF1, βTF2, ARM1, ARM2, and CLD), and carried out focused 3D classification without angular or translational alignment using a soft mask encompassing the IP₃-binding site. Class volumes were low-pass filtered to 5 Å for visualization (Supplementary Fig. 4).

In the IP₃/ATP dataset, particles in the compact class exhibit structural heterogeneity (Supplementary Fig. 5). To separate the particles, we first performed 3D classification without angular or translational alignment using a mask encompassing the entire cytosolic domains (Supplementary Fig. 5). The resulting classes fell into three major groups: resting (with all four ARM2s extended), preactivated (all four ARM2s retracted), and intermediate (mixed ARM2 configurations). To further resolve the intermediate particles, we prepared four masks encompassing the ARM2s of each subunit and performed 3D classification for each mask separately. We then regrouped the particles into all possible ARM2 arrangements. Some classes lacked interpretable ARM2 density, suggesting higher flexibility, and were excluded from further analysis. Particles with four extended or four retracted ARM2s were merged with the resting or preactivated state classes, respectively. NU-refinement of the intermediate classes revealed 3D reconstructions with varying resolutions. Intermediate states 1 and 2 were worse than 6 Å and were not used for further analysis, although their quality was sufficient to confirm these conformations (Supplementary Fig. 6). Intermediate states 3 and 4 produced better maps that allowed assessment of the relative orientation of individual domains but did not permit unambiguous model building (Supplementary Fig. 6). We also observed heterogeneity in the TMD of the preactivated class. To improve the map quality, we performed focused 3D classification using a mask encompassing TMD and retained particle subset that yielded reconstructions with well-defined TMD density (Supplementary Figs. 5 and 6). Particles in the loose class from the IP₃/ATP dataset were subjected to another round of 3D classification to remove the particles with low quality. NU-refinement of the remaining particles, without enforcing any symmetry, resulted in a reconstruction with an average resolution of 3.14 Å (Supplementary Figs. 5).

In the IP₃/ATP/Ca²⁺ dataset, two major classes, both in loose conformation, were observed (Supplementary Fig. 10). The class 1 particles, refined without symmetry enforcement (C1), yielded a map that is highly similar to the loose-class maps from the other datasets. The second class was also refined in C1. While the main portion of the map resembled class 1, substantial amount of extra density was observed. To determine the nature of this extra density, particles were re-extracted with a larger box size (960 px) and twofold binned to 480 px (Supplementary Fig. 13). We performed ab initio reconstruction searching for 3 classes and without enforcing symmetry. Following heterogeneous refinement, particles that produced a hIP₃R-2 dimer map were then refined with C2 symmetry, resulting in a reconstruction at 4.40 Å resolution.

To improve map quality, we performed local refinements using masks spanning subregions of the consensus reconstructions (Supplementary Figs. 2, 7–9, 11, and 13). For reconstructions processed with C4 symmetry (apo, resting, and preactivated), we prepared four masks that cover distinct cytosolic domains of one subunit and a mask for the TMD tetramer (Supplementary Figs. 2, 7 and 8). After symmetry expansion (C4), we performed local refinement for the cytosolic domains. The local refinements for the TMDs were performed without symmetry expansion while enforcing C4 symmetry. For reconstructions without enforced symmetry (inactive-like (IP₃/ATP), inactive-like (IP₃/ATP/Ca²⁺), and dimer), we prepared masks that cover different domains of the entire assembly and performed local refinement using C1 symmetry (Supplementary Figs. 9, 11, and 13). Local refinement maps were aligned onto the consensus maps in ChimeraX⁷⁰ and merged using the “VOP maximum” command to generate composite maps.

Model building

Models were built using Coot⁷¹. We first placed the hIP₃R-2 model generated using the Alphafold server⁷² into the composite apo map, followed by rigid-body placement of individual domains for a single protomer. Manual rebuilding was then performed, and the protomer was symmetry-expanded (C4) to generate the tetramer. Real-space refinement⁷³ using Phenix⁷⁴ was performed with iterative build-refine cycles till a satisfactory model was obtained. Manual building was performed using local refinement maps, while real-space refinements used composite maps. The resulting apo model served as the starting reference for the other datasets, which were processed with the same workflow. Regions lacking interpretable density were omitted. Residues with poorly defined side-chain density were retained with side chains truncated to alanine while preserving residue identity. Coiled-coil segments at the CTD were modeled as poly-alanine based on features visible in unsharpened maps. Validation of the structural models was performed using MolProbity⁷⁵ implemented in Phenix⁷⁴.

Rosetta mutational analysis

Computational mutational scanning of the dimer interface was performed using Rosetta^76,77 with the ref2015 all-atom energy function to identify mutations that selectively disrupt inter-channel interactions. Interface residues were defined based on heavy-atom proximity (8 Å cutoff) between protomers and systematically substituted with a disruptive panel (A, D, E, K, R, P, W) designed to probe steric, electrostatic, and conformational perturbations. For each mutation, side-chain repacking was performed on a fixed backbone using RosettaScripts (PackRotamersMover), and changes in interface binding energy (ΔΔG) were calculated using InterfaceAnalyzerMover by comparing the energy of the bound complex to that of the separated monomers. To confirm that top-ranked mutations did not destabilize the monomer fold, isolated protomers were separately subjected to side-chain repacking (fixbb) followed by constrained coordinate-restrained relaxation (relax), and changes in total energy, steric repulsion (fa_rep), and backbone Cα RMSD were evaluated. Mutations were prioritized based on large interface ΔΔG combined with minimal perturbation to monomer stability.

FSEC analysis

HEK293-3KO cells, in which all three IP₃R subtypes were knocked out using CRISPR⁵¹, were purchased from Kerafast and used for FSEC analysis. Cells were maintained in DMEM supplemented with 10% FBS at 37 °C with 5% CO₂. For FSEC analysis, cells were seeded at 0.5 × 10⁶ cells/well into six-well plates and transfected after 24 h with 2 µg plasmid DNA using FuGENE HD transfection reagent. Cells were harvested 24 h post transfection and solubilized in 200 µl buffer (200 mM NaCl, 40 mM Tris-HCl pH 8.0, 2 mM DTT, 5 mM EDTA pH 8.0, 0.5% LMNG, 0.1% GDN, 1 mM PMSF, and complete Protease Inhibitor Cocktail (Roche)) for 2 h with gentle mixing at 4 °C. Insoluble material was removed by centrifugation at 150,000×g for 20 min at 4 °C. Aliquots (100 µl) of the supernatant were loaded onto a Superose 6 Increase 10/300 GL column equilibrated in buffer composed of 200 mM NaCl, 40 mM Tris-HCl pH 8.0, 2 mM DTT, 5 mM EDTA pH 8.0, 0.005% LMNG and 0.005% GDN. Fluorescence was monitored with excitation at 475 nm and emission at 515 nm.

Fluorescence microscopy

HEK293-3KO and HEK293 cells were transfected using a reverse transfection protocol on µ-Slide 18-well #1.5 glass bottom plates (Ibidi) coated with 0.1 mg/ml poly-D-lysine. For each well, 0.1 µg plasmid DNA and 0.3 µl FuGENE HD were mixed with 10 µl OptiMEM and incubated for 20 min prior to the addition of 100 µl of cells (1.5 ×10⁵ cells/ml in DMEM supplemented with 10% FBS). Cells were incubated for 24 h at 37 °C with 5% CO₂.

Prior to imaging, cells were loaded with Cal-590 AM. A 5 mM Cal-590 AM stock solution in anhydrous DMSO was diluted to 5 µM in Hanks’ Balanced Salt Solution (HBSS) containing Ca²⁺/Mg²⁺ and 20 mM HEPES supplemented with 0.04% Pluronic F-127, and 2.5 mM probenecid. Cells were incubated with the dye for 1 h at 37 °C, washed, and maintained in HBSS containing Ca²⁺/Mg²⁺ and 20 mM HEPES supplemented with 2.5 mM probenecid.

Two-color imaging was performed on a Nikon Eclipse Ti2 microscope equipped with a Yokogawa CSU-W1 SoRa spinning disk confocal, a Hamamatsu Orca-Fusion BT, and a Tokai Hit Stage Top incubator held at 37 °C. hIP₃R-2-mGFP was excited using a 488 nm laser with emission light collected using a 525/36 nm emission filter. Cal-590 was excited using a 561 nm laser with emission light collected using a 605/52 nm emission filter. The Nikon Perfect Focus system was used to maintain focus during imaging. All images were taken with a Nikon Plan Apo 100×/1.45 NA oil immersion objective with Immersion oil Type F. Ten seconds after an initial image acquisition, 500 µM carbachol was added to the sample chamber. Images were then acquired every 30 s for 6 min. Experiments were performed in at least three independent transfections.

Figure preparation

Figures were prepared using ChimeraX⁷⁰ and The PyMOL Molecular Graphics System (Version 2.0, Schrödinger, LLC).

Reporting summary

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

Author contributions

E.K. conceived the project, performed cryo-EM data analysis, and wrote the manuscript with input from all authors; C.L. optimized and performed protein expression and purification, performed grid preparation and screening, model building and refinement, construct preparation and cell culture for imaging; Y.-J.L. performed protein purification and electrophysiology; M.G.K. performed live cell imaging; Q.T. helped with protein purification, grid preparation, screening, and model building and refinement.

Peer review

Peer review information

Nature Communications thanks Amedee des Georges, Katsuhiko Mikoshiba, and Tobias Raisch for their contribution to the peer review of this work. A peer review file is available.

Funding

This work was supported by the NIH (R01 GM141251 and R35 GM161399 to E.K.) and Vanderbilt University School of Medicine Basic Sciences.

Data availability

The cryo-EM density maps and atomic coordinates have been deposited in the Electron Microscopy Data Bank (EMDB) and the Protein Data Bank (PDB), respectively, under the following accession numbers. Apo hIP₃R-2: PDB 9YKK; EMDB EMD-73048 (consensus), EMD-73054 (composite), and EMD-73049, EMD-73050, EMD-73051, EMD-73052, EMD-73053 (locally refined subregions). Inactive-like hIP₃R-2 (ligand-free): EMDB EMD-73225(consensus). hIP₃R-2 in the ligand-bound resting state: PDB 9YKY; EMDB EMD-73061 (consensus), EMD-73067 (composite), and EMD-73062, EMD-73063, EMD-73064, EMD-73065, EMD-73066 (locally refined subregions). hIP₃R-2 in the preactivated state: PDB 9YLI; EMDB EMD-73082 (consensus), EMD-73088 (composite), and EMD-73083, EMD-73084, EMD-73085, EMD-73086, EMD-73087 (locally refined subregions). hIP₃R-2 in the intermediate state 3: EMDB EMD-73223 (consensus) hIP₃R-2 in the intermediate state 4: EMDB EMD-73224 (consensus) Inactive-like hIP₃R-2 (IP₃/ATP): PDB 9YMZ; EMDB EMD-73111 (consensus), EMD-73129 (composite), and EMD-73112, EMD-73113, EMD-73114, EMD-73115, EMD-73116, EMD-73117, EMD-73118, EMD-73119, EMD-73120, EMD-73121 (locally refined subregions). Inactive-like hIP₃R-2 (IP₃/ATP/Ca²⁺): PDB 9YNK; EMDB EMD-73198 (consensus), EMD-73210 (composite), and EMD-73199, EMD-73201, EMD-73202, EMD-73203, EMD-73204, EMD-73205, EMD-73206, EMD-73207, EMD-73208, EMD-73209 (locally refined subregions). hIP₃R-2 dimer (IP₃/ATP/Ca²⁺): PDB 9YNO; EMDB EMD-73212 (consensus), EMD-73219 (composite), and EMD-73213, EMD-73214, EMD-73215, EMD-73216, EMD-73217, EMD-73218 (locally refined subregions). hIP₃R-2 dimer (complete; IP₃/ATP/Ca²⁺): EMDB EMD-73222 (consensus). The following previously published datasets were used: 6UQK, Cryo-EM structure of type 3 IP₃ receptor revealing presence of a self-binding peptide⁴³. 7T3P [10.2210/pdb7T3P/pdb], IP₃ and ATP bound type 3 IP₃ receptor in the pre-active A state⁴⁴. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74494-y.