Priming of hIP₃R-3 for activation

To compare the structures presented here, we aligned their selectivity filters and pore helices (residues 2460-2481), which reside at the luminal side of the TMD and are virtually identical in all classes. The pre-active A structure is almost identical to the previously published IP₃-bound hIP₃R-3 structure⁶ (Supplementary Fig. ^9a) and reveals that IP₃ binding causes the ARM1 to rotate about 23° relative to the βTF-2, causing global conformational changes within the cytoplasmic domains, as observed in previous cryo-EM and X-ray crystallography experiments^6,8,21–24 (Supplementary Fig. ^{9b, c}). The pre-active B and C structures adopt distinct conformations that are intermediates between the pre-active A and open state structures. Based on these conformational changes, we propose a sequential transition from pre-active A to B, then C, although the alternative transitions cannot be ruled out entirely. During the transition to the pre-active B state, the N-terminal domain (NTD) of each protomer comprising βTF1, βTF2, ARM1, ARM2, and CLD rotates about 4° counter-clockwise relative to the TMD and moves about 2Å closer to the membrane plane (Fig. 2a; Supplementary Movie ¹). In the pre-active C state, the NTDs remain primarily unchanged compared to the pre-active B state, while the ARM3 and JD are rotated by 7°, causing mild distortions at the cytoplasmic side of the TMD without opening the channel (Fig. 2b; Supplementary Movie ¹). Compared to the ligand-free conformation, the βTFs move about 7Å closer to the membrane plane, and ARM3-JD rotates about 11° in the pre-active C conformation.

Open in a separate window

Fig. 2

Conformational changes in the pre-active states.

a, b Ribbon representations of hIP₃R structures superposed on the residues forming the selectivity filter and P-helix of the TMDs, emphasizing the conformational changes between the states indicated above. Domains with substantial conformational changes are shown in full colors only on one subunit, while the rest of the protein is transparent. Curved and straight red arrows indicate the rotation and translation of the domains with red labels relative to the rotation axis (black bars), respectively.

Ca²⁺-mediated conformational changes leading to pore opening

In the absence of Ca²⁺, the ARM3 and JD act as a rigid body, where there are no significant conformational changes relative to each other (Fig. 2a, b). When Ca²⁺ is bound, the JD rotates about 11° relative to the ARM3 (Fig. 3a), resembling a clamshell closure, which leads to global conformational changes in the whole receptor, including the movement of the NTD closer to the membrane plane by 2Å (Fig. 3b; Supplementary Movie ¹). In contrast to the limited rotation of the ARM3 (about 5°), the JD rotates about 14° around an axis roughly perpendicular to the membrane plane, leading to conformational changes at the TMD and resulting in pore opening in the active state (Fig. 3b).

Open in a separate window

Fig. 3

Conformational changes coupling Ca²⁺ binding to pore opening.

a Comparison of the JD (shown in full colors) orientation relative to the ARM3 (shown transparent) in the pre-active-C (orange) and active (blue) structures. The black bar indicates the axis for the rotation of the JD. Ca²⁺ and ATP are shown as spheres. b Global conformational changes induced by Ca²⁺ binding are depicted similar to Fig. 2. c Close-up view of the Ca²⁺ binding site in the active conformation. Domains are colored as in Fig. 1.

Ca²⁺ is coordinated by E1882 and E1946 on the ARM3 and the main-chain carboxyl group of T2581 on the JD (Fig. 3c). H1884 and Q1949 are also close and may interact with Ca²⁺ through water molecules (Fig. 3c). These residues are highly conserved in the homologous ion channel family, ryanodine receptors (RyRs), suggesting a common activation mechanism in IP₃Rs and RyRs²⁵ (Supplementary Fig. ^{10a, b}). Mutation of the corresponding residues in RyRs markedly reduced the sensitivity to Ca²⁺, further supporting this site’s involvement in the Ca²⁺ induced activation^26–28.

ATP binding site

Within the JD, we observed a well-resolved cryo-EM density for ATP in all the structures (Supplementary Fig. ⁷). The quality of the maps obtained through local refinement allowed unambiguous modeling of ATP, revealing its key interactions with the protein residues (Fig. 4a, b). The adenosine base intercalates into a cavity surrounded by F2156, F2539, I2559, M2565, and W2566 near the zinc finger motif and forms hydrogen bonds with the sulfur of C2538, the backbone amide group of F2539, and the carbonyl groups of H2563 and I2559 (Fig. 4b). The phosphate moieties interact with K2152, K2560, and N2564 (Fig. 4b). There are no apparent structural changes around the binding site upon ATP binding, suggesting that ATP’s potentiating effect is likely due to the increased rigidity of the JD (Supplementary Fig. ^9d). ATP binding site is highly conserved among the subtypes except for E2149 which corresponds to lysine and arginine in IP₃R-1 and IP₃R-2 (Fig. 4c). A positively charged residue instead of E2149 in the proximity of the phosphate moieties may cause tighter interaction of ATP with IP₃R-1 and IP₃R-2, explaining the low binding affinity of ATP to IP₃R-3 compared to IP₃R-1 and IP₃R-2¹⁷.

Open in a separate window

Fig. 4

ATP binds to the JD.

a The cryo-EM density of ATP (red mesh) from the composite map of the pre-active A state and the modeled ATP molecule. b Close-up view of the ATP binding site in the pre-active A state. Dashed lines indicate hydrogen bonding. c Sequence alignment of hIP₃R subtypes around the residues forming the ATP binding site. Residues shown in (b) are highlighted. Select residues are indicated by arrows, and E2149 is labeled in red.

ATP binds to a similar location near the zinc finger motif in RyRs^25,29,30. However, its binding mode differs, potentially due to the differences in the residues that form the binding pocket, most notably the basic residues interacting with the phosphate moieties (Supplementary Fig. ^{10a, c, d}). In RyR-1s, the phosphate moieties interact with K4211, K4214, and R4215, all located on a single helix (Supplementary Fig. ^{10c, d}). In hIP₃R-3, there is only a single lysine residue (K2152) on the corresponding helix, and the phosphate moieties interact with K2560, located on the opposite side of the binding pocket. A leucine residue (L4980) occupies this position in RyR-1s. The differences in the number and location of the basic residues likely force the phosphate moieties of ATP to adopt different conformations. Furthermore, F2156 in hIP₃R-3 points toward the adenosine binding pocket, prohibiting ATP from adopting the conformations observed in RyR-1s due to steric clash in hIP₃R-3s (Supplementary Fig. ^{10c, d}).

Structure of the TMD in the open conformation

The TMD of IP₃Rs has the same overall architecture of voltage-gated ions channels with a central pore domain, consisting of S5, S6, and pore (P) helix, surrounded by pseudo-voltage-sensor domains (pVSDs), consisting of S1, S2, S3, and S4 helices along with two IP₃R/RyR specific TM helices (S1’ and S1”) (Fig. 5). In the closed channel, F2513 and I2517 of the S6 helix form two layers of hydrophobic constriction at the pore, blocking the path for the permeation of hydrated ions (Fig. 5). JD’s rotation upon Ca²⁺ binding pushes the pVSD’s cytoplasmic side away from the pore domain by about 7Å and tilts the cytoplasmic side of the S6 (S6_cyt) by 12° (Fig. 5a; Supplementary Movie ¹). Concurrently, the S4-5 linker and S5 helix move away from the S6 helix, thereby inducing a distortion of S6 around the constriction site and moving F2513 and I2517 away from the pore. As a result, the diameter of the water-accessible pore increases to 8Å, large enough to permeate hydrated cations (Fig. 5b). The flexibility introduced by the neighboring glycine residue (G2514), mutation of which to alanine in IP₃R-1 is associated with spinocerebellar ataxia 29 (SCA29)³¹, is likely critical to the movement of F2513. The tilting of the S6_cyt breaks the salt bridge between D2518 and R2524 of the neighboring subunits, moving D2518 towards the pore while pulling R2524 away, which creates an electronegative path on the cytoplasmic side of the pore (Supplementary Fig. ¹¹). In contrast to the prediction of a π- to α-helix transition at the S6_lum during channel opening^6,10, the π-helix remains intact, and its tip acts as a pivot for the S6_cyt tilting and bulging (Fig. 5).

Open in a separate window

Fig. 5

Structure of the IP₃R-3 in the open conformation.

a Comparison of the hIP₃R-3 structures in the pre-active C and active conformations aligned as in Fig. 2. b Ion permeation pathways of hIP₃R-3 in pre-active C and active conformations (radii coloring: red, <0.8Å; green, 0.8-4.0Å; gray, >4.0Å) along with the 1D graph of the pore radius. The JD and TMD of the active state are shown in cyan and violet, respectively. The pre-active C structure is colored orange.

Although the TMDs of IP₃Rs and RyRs are highly similar, there are noticeable differences in their pore structures (Supplementary Fig. ^{10e, f}). In RyRs, the constriction site is formed by glutamine and isoleucine, corresponding to F2513 and I2517 in hIP₃R-3, respectively³². In the open state of RyRs, the isoleucine is positioned similarly to I2517 of hIP₃R-3^25,33,34. On the other hand, the glutamine residue faces the pore in the open state, forming part of the hydrophilic permeation pathway, unlike F2513. Interestingly, N2510 in hIP₃R-3, which corresponds to alanine in RyRs, faces the permeation pathway similar to the glutamine of RyRs, suggesting that the amide group plays an important role in the ion permeation. However, since the side chain of N2510 extends from a different position on the S6 helix than the side chain of glutamine in RyRs, the binding pocket for ryanodine²⁵, a RyR-specific inhibitor, is not present in IP₃R, potentially causing IP₃Rs to be unresponsive to ryanodine³².

Several missense mutations identified in the IP₃R subtypes are associated with diseases, including spinocereblar ataxia, Gillespie syndrome, anhidrosis, and neck squamous cell carcinoma (Supplementary Fig. ¹²; reviewed in^32,35,36). Perhaps not surprisingly, most of these mutations are localized around the IP₃ binding site and alter IP₃ binding affinity^32,35–37. Another hot spot for these mutations is the constriction site of the pore, which undergoes conformational changes during channel opening (Supplementary Fig. ¹²). It is plausible that these mutations either affect the Ca²⁺ permeability (e.g., mutation of N2510³⁸ or I2517³⁹) or restrict conformational changes required for dilation of the pore (e.g., mutation of G2514³¹). Two of the mutated residues (T2519⁴⁰ and F2520⁴¹) interact with the residues on the S4-S5 linker, which couples the tilting of the pVSD to the bulging of the constriction site (Supplementary Fig. ^12b). Mutations of these residues are likely to impair this coupling and thus hinder gating.

The flexibility of the CTD

The CTD, extending from the JD along the symmetry axis, forms a left-handed coiled-coil motif before interacting with the βTF2 of the neighboring subunit. The density for the CTD was poorly resolved in all of the states (Supplementary Fig. ^{13a, b}). However, the coiled-coil motifs were visible in the unsharpened maps in the pre-active and active states, enabling us to model poly-alanine peptides without assigned registries (Supplementary Fig. ^{13a, b}). The densities for the extensions from the coiled-coil motif towards the βTF2 become visible when viewed at lower thresholds, whereas the linkers between the JD and the coiled-coil motif remain invisible, indicating higher flexibility for this region (Supplementary Fig. ^{13a, b}). We did not observe any interpretable density for the CTD in the inactive state (Supplementary Fig. ^{13a, b}).

For IP₃R-1, the CTD was proposed to transmit the conformational changes induced by IP₃ at the NTD to the JD⁸. In IP₃R-3, there are no apparent changes on the coiled-coil motif in the pre-active states, but the coiled-coil motif rotates about 20° around the symmetry axis and moves closer to the TMD by 6Å in the active state (Supplementary Fig. ^{13c, d}). However, the linker between the coiled-coil motif and JD remains flexible, suggesting that the structural rearrangements of this domain are not directly enforcing the channel opening (Supplementary Fig. ¹³). In line with these observations, removing CTD residues interacting with the βTF2 or swapping the C-terminal region of IP₃R-1 with the RyRs, which lack the extended CTD, did not diminish receptor activation^21,24,42.

Cryo-EM sample preparation and data collection

Purified hIP₃R-3 in the SEC buffer containing 1mM EDTA was supplemented with 500µM IP₃ (from 10mM stock in water), 0.1mM CaCl₂, and 5mM ATP (from 100mM stock, pH 7.2). 2.0μL of the protein sample was applied to 300 mesh Cu Quantifoil 1.2/1.3 grids (Quantifoil Microtools) that were glow discharged for 20s at 25mA. The grids were blotted for 7s at force 10 using single-layer Whatman ashless filter papers (Cat. #: 1442-055, GE Healthcare) and were plunged into liquid ethane using an FEI MarkIV Vitrobot at 8°C and 100% humidity. The filter papers were not pre-treated with Ca²⁺ chelators or any other chemicals. Four grids prepared using the same sample were imaged using a 300kV FEI Krios G3i microscope equipped with a Gatan K3 direct electron camera in four different data collection sessions at Case Western Reserve University. Movies containing 40–50 frames were collected at a magnification of ×105,000 in super-resolution mode with a physical pixel size of 0.828Å/pixel and defocus values at a range of −0.8 to −1.6µm using the automated imaging software SerialEM⁴⁷ and EPU (ThermoFisher Scientific).

Cryo-EM data processing

Datasets from four sessions were initially processed separately using Relion 3.0⁴⁸. We used MotionCor2⁴⁹ and Gtcf⁵⁰ to perform beam-induced motion correction and CTF estimations, respectively. We performed auto picking using the Laplacian-of-Gaussian option of Relion, extracted particles binned 4×4, and performed 2D class classification. Using the class averages with apparent features, we performed another round of particle picking. We cleaned the particles, extracted as 4×4 binned, through 2D classification and performed 3D classification using the hIP₃R-3 map (EMD-20849¹⁰), which was converted to the appropriate box and pixel size. We observed two predominant conformations. One had a compact NTD and tight interactions between subunits as in previously published IP₃R structures in the absence of Ca²⁺, hereafter called “compact” conformation^6–8,10 (Supplementary Fig. ¹). The other one had the NTD of each subunit tilted away from the central symmetry axis resembling hIP₃R-3 structures obtained in the presence of high Ca²⁺ concentrations, hereafter called “loose” conformation⁶ (Supplementary Fig. ¹). These particles were separately selected and reextracted using a box size of 480×480 pixels at the physical pixel size. After 3D refinements, we performed CTF refinement and Bayesian polishing⁵¹.

We combined all the polished particles and performed another round of 3D classification, using one of the compact structures as a reference map. We grouped particles into “compact” and “loose” classes (Supplementary Fig. ¹). Refinement of the particles in the “compact” conformation yielded a 3D reconstruction with an average resolution of 3.9Å. Although there were slight changes at the TMD compared to the structure in the closed state, these changes were not significant enough to suggest that the channel was open. To more clearly resolve the density around the TMD, we performed another round of 3D classification using a mask that only covers the ARM3, JD, and TMD and without performing an angular or translational alignment in Relion3 (Supplementary Fig. ¹)⁵². 3D refinements of the particles in each class were performed using non-uniform refinement in CryoSPARC, enforcing C4 symmetry and local CTF refinements (Supplementary Fig. ¹)⁵³. Five classes led to four high-resolution (better than 4Å) 3D reconstructions, whereas the 3D refinement of the other three classes resulted in poorly resolved maps. The particles in the “loose” conformation were processed using non-uniform refinement in CryoSPARC, but without enforcing any symmetry. Local resolution estimates were calculated using CryoSPARC⁵³ (Supplementary Figs. ²–⁶). Some of the data processing and refinement software was supported by SBGrid⁵⁴.

To improve the quality of the maps, we performed local refinements using masks covering parts of the original cryo-EM maps (Supplementary Figs. ²–⁶). We prepared five masks that cover distinct domains of one of the subunits for the pre-active A, B, C, and active conformations. After symmetry expansion using C4 symmetry, we performed local refinement using CryoSparc (Supplementary Figs. ²–⁵). For the inactive state, we prepared four masks that cover the cytoplasmic domains of each subunit and another mask that covers the tetrameric ARM3, JD, and TMD (Supplementary Fig. ⁶). The local refinements were performed using C1 symmetry for the cytoplasmic domains and C4 symmetry for the tetrameric ARM3, JD, and TMD. The resulting local refinement maps were aligned onto the original maps using Chimera⁵⁵ and merged using the “VOP maximum” command of Chimera⁵⁵ to prepare the composite maps (Supplementary Figs. ²–⁶).

Model building

Model building was performed using Coot⁵⁶. We first placed the hIP₃R-3 structure in ligand-free conformations (PDB ID: 6UQK¹⁰) into the composite map of Pre-active A, and performed rigid-body fitting of individual domains of one of the protomers. We then manually fit the residues into the density and expanded the protomer structure into a tetramer using the C4 symmetry. We performed real-space refinement using Phenix⁵⁷. We repeated build-refine iterations till a satisfactory model was obtained. This model was used as a starting model for the other structures following the same workflow. Regions without interpretable densities were not built into the model. Residues without apparent density for their side chains were built without their side chains (i.e., as alanines) while maintaining their correct labeling for the amino acid type. The coiled-coil regions were modeled as poly-alanines without residue assignment using the unsharpened maps. Validations of the structural models were performed using MolProbity⁵⁸ implemented in Phenix⁵⁷.

Figure preparation

Figures were prepared using Chimera⁵⁵, ChimeraX⁵⁹, and The PyMOL Molecular Graphics System (Version 2.0, Schrödinger, LLC). Calculation of the pore radii was performed using the software HOLE⁶⁰.

We thank Dr. Kunpeng Lee for cryo-EM data collection at Case Western Reserve University. We thank Theo Humphries and other support staff at the Pacific Northwest Center for Cryo-EM (PNCC), Drs. Elad Binshtein, Melissa Chambers, and Scott Collier at the Cryo-EM facility at Vanderbilt University for their assistance with cryo-EM sample screening. We thank Drs. Hassane Mchaourab, Terunaga Nakagawa, and Silvia Ravera for discussions and review of the manuscript. This work was conducted in part using the CPU and GPU resources of the Advanced Computing Center for Research and Education (ACCRE) at Vanderbilt University. We used the DORS storage system supported by the U.S. National Institute of Health (NIH) (S10RR031634 to Jarrod Smith). This work was supported by the NIH (R01GM141251 to E.K.), Vanderbilt University, Vanderbilt Diabetes and Research Training Center (NIH P30DK020593 to E.K.), and the Molecular Biophysics Training Program (NIH T32GM008320 to E.A.S.).