Introduction

The K_ATP channel expressed in pancreatic islet β-cells is a principal homeostatic regulator of insulin secretion^1–3. Composed of four pore-forming Kir6.2 and four regulatory sulfonylurea receptor 1 (SUR1) subunits^4,5, K_ATP channels influence the resting membrane potential, thus action potential firing, Ca²⁺ entry, and insulin secretion. The activity of K_ATP channels is dynamically regulated by glucose changes via intracellular ATP and MgADP, which bind to inhibitory and stimulatory sites to close and open the channel, respectively^6,7. This enables insulin secretion to follow fluctuations in glucose concentrations. The central role of K_ATP channels in insulin secretion and glucose homeostasis is underscored by dysregulated insulin secretion and blood glucose in humans bearing K_ATP mutations: loss-of-function mutations cause congenital hyperinsulinism characterized by persistent insulin secretion despite life-threatening hypoglycemia^8,9; conversely, gain-of-function channel mutations result in neonatal diabetes due to insufficient insulin secretion¹⁰.

In addition to adenine nucleotides-dependent regulation, membrane phosphoinositides participate in the gating activation of K_ATP channels^11,12, as they do all other eukaryotic Kir channels¹³. PI(4,5)P₂ (denoted as PIP₂ unless specified) is the most abundant phosphoinositide in the plasma membrane¹⁴, with PIP₂ containing arachidonoyl (C20:4) and stearoyl (C18:0) acyl chains being the most dominant species in mammalian cells¹⁵. In inside-out patch-clamp recordings, application of C18:0/C20:4 long-chain PIP₂ to the cytoplasmic side of the membrane patch resulted in stable increases in K_ATP channel open probability (P_o) and a gradual decrease in channel sensitivity to ATP inhibition; whereas scavenging membrane PIP₂ with poly-lysine decreases channel P_o^11,12. Because increasing channel P_o by PIP₂ also decreases channel sensitivity to ATP inhibition^11,12, a joint regulatory mechanism involving allosteric or binding competition between the two ligands has seemed likely¹⁶. In particular, the effect of PIP₂ on K_ATP channels has been assumed to derive from a putative conserved binding site in Kir6.2 having a position homologous to the PIP₂ binding site previously identified in crystal and cryoEM structures of Kir2 and Kir3^17–20. Uniquely however, gating of Kir6.2 by PIP₂ is strongly modulated by its regulatory subunit SUR1. Compared to the fully assembled K_ATP channels, Kir6.2 channels lacking SUR1 partners display more than 10-fold lower P_o^11,16,21, which reflects reduced PIP₂ binding or gating, and are less sensitive to PIP₂ stimulation^11,16. Of particular interest, the SUR1 N-terminal transmembrane domain, TMD0, which interfaces with Kir6.2^{22 -24}, is sufficient to confer the high P_o of wild-type (WT) channels^25,26, implicating a crucial role of SUR1-TMD0 in controlling Kir6.2 response to PIP₂.

To date, cryoEM structures of K_ATP channels determined for a variety of liganded and mutational states have yielded a structural spectrum encompassing open, clo sed, and inactivated conformations (reviewed in^6,7,27). Careful comparisons have suggested mechanisms by which channel activity is regulated by the SUR subunit and nucleotides. However, confusion has grown regarding PIP₂ binding and gating. Two recently reported open- and pre-open mutant K_ATP channel cryoEM structures lack discernible PIP₂ at the putative binding site^28,29, even when excess synthetic short-chain diC8-PI(4,5)P₂ was added to the sample in the case of the pre-open structure³⁰. Furthermore, purified WT K_ATP channels reconstituted in lipid bilayer s lacking PIP₂ showed single channel openings²⁸. These reports suggest that PIP₂ binding may be inessential for open channel transition²⁸. Thus, notwithstanding robust data indicating PIP₂ involvement in physiological channel activation^11,31,32, how PIP₂ binds and modulates K_ATP channels remains enigmatic.

Our approach to understanding K_ATP channel activation and PIP₂ binding involved determining channel structure bound to PIP₂ by (1) incubating membranes expressing K_ATP channels with long-chain PIP₂ known to stably activate the channel^11,12,33, and (2) employing a Kir6.2 variant containing the neonatal diabetes-causing mutation Q52R (denoted Kir6.2^Q52R)^34,35. Kir6.2^Q52R markedly increases channel P_o and causes a 20-fold decrease in channel sensitivity to ATP inhibition^34,35. Higher P_o could result from enhanced PIP₂ binding or its effect on gating. Intriguingly, the effect of this mutation on channel gating is SUR1-dependent. When Kir 6.2 was manipulated to express at the cell surface without co-assembly with SUR1 (by deleting the Arg-Lys-Arg endoplasmic reticulum retention signal at the C-terminal region of Kir6.2), introducing the Q52R mutation had no effect on the P_o and ATP sensitivity³⁶. The SUR1-dependent over-activity, imparted by the Kir6.2-Q52R mutation, indicated the possibility to probe the regulatory interaction between SUR1 and membrane PIP₂ on Kir6.2 gating.

Here, we present the cryoEM structure of the K_ATP channel harboring the neonatal diabetes Q52R mutation (referred to as the SUR1/Kir6.2^Q52R channel hereinafter) bound to C18:0/C20:4 long-chain PI(4,5)P₂ in an open conformation. Surprisingly, two PIP₂ molecules bind in tandem in the open channel at the interface between Kir6.2 and SUR1. A first PIP₂ molecule bound within the site that is conserved among Kir channels. A second PIP₂ molecule was at a novel adjacent position, stacked within the regulatory interface between Kir6.2 and SUR1 TMDs. The open channel conformation is further stabilized by direct interaction between Q52R of Kir6.2 and W51 located in TMD0 of SUR1. Accounting for decades of evidence from functional perturbation studies, the PIP₂-bound SUR1/Kir6.2^Q52R K_ATP structure unveils the agonist binding of PIP₂ in K_ATP channels, revealing that in addition to the ancestral Kir PIP₂ site, the incorporation of the regulatory SUR1 subunit creates a second regulatory PIP₂ binding site. The results implicate a novel mechanism through which SUR1, by cooperating with Kir6.2 for PIP₂ binding, stabilizes the K_ATP channel in an open state to control activity, and illuminate the molecular mechanism by which a human activating mutation in the K_ATP channel causes neonatal diabetes.

Results

Structure of an open SUR1/Kir6.2^Q52R K_ATP channel bound to PIP₂

To purify the neonatal diabetes variant SUR1/Kir6.2^Q52R K_ATP channel, we co-expressed Kir6.2^Q52R and SUR1 as independent polypeptides by transducing adherent mammalian COSm6 cells with recombinant adenovirus packaged with genes for rat Kir6.2 ^Q52R and FLAG-tagged hamster SUR1 (see Methods). Prior functional experiments have shown that the activating effects of long-chain PIP₂ are long-lasting and resistant to subsequent washout, indicating stable partition into the membrane^11,12,33. By contrast, the effects of synthetic short - chain PIP₂ such as diC8-PIP₂ are highly reversible^33,37. Accordingly, we prepared a membrane fraction of COSm6 cells expressing the mutant channel, and enriched the native membranes with brain PIP₂, primarily containing C18:0/C20:4 long-chain PI(4,5)P₂ (Avanti Polar Lipids), before detergent solubilization. We anticipated that addition of long-chain PIP₂ prior to detergent extraction would stably incorporate PIP₂ resistant to washout in subsequent purification steps to facilitate the capture of PIP₂-bound channel structure. K_ATP channels were purified via the FLAG-tag on SUR1 and spotted on graphene oxide (GO)-coated grids for cryoEM (see Methods). To maximize channels in PIP₂-bound open conformations, no ATP or ADP were added to the sample.

CryoEM micrographs showed mostly full SUR1/Kir6.2^Q52R K_ATP channel particles (Fig. S1a). 2D classes indicate an ordered Kir6.2 and SUR1-TMD0, with a relatively more blurred SUR1-ABC core especially the nucleotide binding domains (NBDs) (Fig. S1a). A non-uniform refinement with C4 symmetry using all good particles from 2D classification (70,638) in cryoSPARC resulted in an un-masked reconstruction of the full channel at 3.9 Å using the gold-standard Fourier shell correlation (GSFSC) cutoff of 0.143 (see Methods). The electron potential map shows relatively weak signal for TMD2 and NBD2 of SUR1 (Fig. S1), indicating higher mobility of these domains. Multiple rounds of ab initio reconstruction without symmetry imposed (Fig. S1a) yielded a best 3D class containing 14,115 particles, which following a C4 non-uniform refinement in cryoSPARC with auto-masking that excluded SUR1-NBD2 resulted in a final map reconstruction of 2.9 Å by GSFSC within the autoFSC mask (Figs. 1a, S1), or 3.3 Å resolution by GSFSC within a mask that includes the full K_ATP channel. Since the full channel map derived from the C4 non-uniform refinement has the best overall resolution, showing a highly ordered Kir6.2 tetramer core and SUR1-TMD0 with local resolutions < 2.9 Å, the map was used for model building (Figs. 1, S1, Table S1). The structure revealed bound PIP₂ at the regulatory interface of Kir6.2 and SUR1 and an open Kir6.2 pore and molecular rearrangements at the SUR1-Kir6.2 interface critical to gating, as we describe in detail in the following sections.

Figure 1.

Additional analysis of all particles from the 2D classes employed a strategy of symmetr y expansion and focused 3D classification (using a mask of Kir6.2^Q52R tetramer plus one SUR1 subunit to circumvent the flexibility of the SUR1 subunits), revealing a spectrum of conformational classes with distinct Kir6.2-cytoplasmic domain (CTD) and SUR1 positions (Fig. S2). The two most extreme conformations include a dominant conformation at ∼3.3 Å ( no C4 symmetry imposed) that is nearly identical to the map derived from the C4 non-uniform refinement described above, and a minor conformation at 6.9 Å (Fig. S2a). Superimposition of the two extreme conformations revealed differences in the K⁺ pore path, rotation of the Kir6.2^Q52R-CTD and distance of Kir6.2^Q52R-CTD to the membrane (Fig. S2b-d), as well as the absence or presence of the cryoEM density corresponding to the N-terminal domain of Kir6.2^Q52R (the distal 30 amino acids, referred to as KNtp; Fig. S2e). Relative to the high resolution dominant class, the low resolution minor class shows a more constricted Kir6.2^Q52R pore, Kir6.2^Q52R-CD clockwise rotated (viewed from the cytoplasmic side) and moved away from the membrane, and clear KNtp density in the central cleft of the SUR1-ABC core. Moreover, the SUR1-ABC core is rotated towards the Kir6.2^Q52R tetramer and moves downward from the outer membrane (Fig. S2c,d). These features highly resemble our previously published closed apo (no PIP₂, nucleotides, or pharmacological inhibitors added) WT K_ATP channel structure (PDB ID 7UQR) ³⁸, and indicate the minor conformation isolated from our sample represents an apo closed SUR1/Kir6.2^Q52R structure (and will be referred to as such hereinafter) even with the exogenously added long-chain PIP₂.

Tandem PIP₂ binding at the interface of Kir6.2 and SUR1

The cryoEM map of PIP₂-bound SUR1/Kir6.2^Q52R K_ATP channels showed lipid density corresponding to two adjacent PIP₂ molecules occupying sites in a large pocket at the Kir6.2-SUR1 interface near the inner leaflet of the membrane bilayer ( Fig. 2a). The first PIP₂ occupies the predicted PIP₂ binding site of Kir6.2 based on comparisons with the homologous structures of Kir channels Kir2 and Kir3/GIRK bound to PIP₂^17–20 (colored cyan in all relevant figures). The second PIP₂ binding site, which is occupied by a PIP₂ with stronger map features (colored magenta in relevant figures) than the first PIP₂, is nestled between SUR1 and the first PIP₂ binding site and in contact with both Kir6.2 and SUR1 subunits (Fig. 2). Both PIP₂ molecules were fit into the map by modelling as C18:0/C20:4 PI(4,5)P₂ (Figs. 1b, 2b,c, S3a, b), which is the dominant form present in brain PIP₂ (∼75%)¹⁵, and both PIP₂ sites have map features for the phospho-head groups as well as long acyl chains that can be fit by a C18:0/C20:4 PI(4,5)P₂ structure (monomer ID PT5) (Figs. 2b, c, S3b). Three other well-resolved lipid map features at the inner leaflet of the membrane bilayer were bound between SUR1-TMD0 and SUR1 - TMD1. These were best modeled by the common native phospholipids phosphatidylethanolamine and phosphatidylserine (Fig. S3c,d). The comparative fit of the different lipid species demonstrates that they can be distinguished using the reconstructed cryoEM maps (Fig. S3). Of note, lipid or detergent densities were also observed in the outer leaflet space (Fig. S3a,c); however, they were not sufficiently resolved to allow modeling.

Figure 2.

Structural analysis surrounding the PIP₂ head groups

The head group of PIP₂ at the first binding site forms polar interactions with surrounding residues (Fig. 2b, c). The phosphate groups are coordinated directly by Kir6.2 residu es R176 and K67 while Kir6.2-W68 forms close interaction with the inositol ring, although K170, which is about 4.5 Å away and H175, which is about 5.5 Å away, are also in close proximity. Most of the head group-interacting residues are conserved in Kir2 and Kir3 channels with the exception of Kir6.2-P69 and H175, which in Kir2 and Kir3 are an arginine and a lysine, respectively and shown to coordinate PIP₂ binding in Kir2 and Kir3^17–20. The amino acid difference at these two positions would weaken Kir6.2-PIP₂ interactions, consistent with the lower phosphoinositide head-group specificity compared to other Kir channels³³. Interestingly, Kir6.2-H175 sidechain exhibited two rotamer positions in the PIP₂-bound SUR1/Kir6.2^Q52R channel cryoEM density map: one oriented towards PIP₂ in the conserved binding site and the other towards E179 in the same Kir6.2 subunit (Fig. 2b). Kir6.2-H175 has been previously implicated in acid-induced activation of K_ATP channels between pH 7.4 and 6.8 with a pK of 7.16³⁹. Mutating Kir6.2-H175 to lysine, which mimics protonated state of H175, increased basal channel activity and abolished channel activation by pH, suggesting protonation of H175 favors channel opening ³⁹. Moreover, mutating Kir6.2-E179 to Q greatly attenuated pH-induced channel activation, suggesting Kir6.2 - E179 also has a role in acid activation of K_ATP channels. Our structural observation that H175 sidechain has interaction with PIP₂ head group in the conserved site and with Kir6.2-E179 is consistent with the aforementioned mutation-function correlation studies. Since our protein sample was prepared at pH∼7.5 (see Methods), which predicts H175 to be largely unprotonated based on the pK of acid activation of the channel of 7.16, it remains to be determined how protonation of H175 at lower pH that activates the channel alters interaction with PIP₂ and E179 to stabilize channel opening.

The second PIP₂ molecule sits in a novel site coordinated by both Kir6.2 and SUR1 subunits (Figs. 1b, 2). Here, the PIP₂ head group interacts with surrounding residues including K39 and Q57 of Kir6.2, and N131, F132, P133, and K134 of SUR1. Previously we have noted that in closed K_ATP structures, SUR1-K134, located in intracellular loop (ICL)2 of SUR1 - TMD0, has its side chain directed towards cryoEM densities in the predicted PIP₂ binding pocket tentatively assigned as phosphatidylserines³⁸. Based on the observation, we mutated SUR1-K134 to alanine and found it reduced channel P_o, which was recovered by adding exogenous PIP₂³⁸. Our PIP₂-bound open structure now shows that SUR1-K134 is involved in PIP₂ binding but at the novel second site, directly explaining the mutational study result.

Structural interactions with the acyl chains of PIP_2.

Most phospholipids present a range of molecular species comprising acyl chains of diverse length and saturation. However, mammalian PIPs show the predominance of a single hydrophobic backbone composed of arachidonoyl ( C20:4) and stearoyl (C18:0) chains¹⁵. In the open K_ATP channel structure determined with an enrichment of natural brain PIP₂ in the membrane, we observed cryoEM densities corresponding to branched long acyl chains for both PIP₂ molecules and modeled them to best fit the densities (Figs. 2b, S3a, S3b). The first PIP₂ bound at the Kir-conserved site, mediates interactions between Kir6.2 subunits. While one chain is primarily associated with the outer TM helix of the Kir6.2 subunit that coordinates head group binding (modeled as the arachidonoyl chain), the other chain sits between the TM helicies of the adjacent Kir6.2 subunit (modeled as the stearoyl chain; Fig. 2b). CryoEM density at this conserved PIP₂ binding pocket was previously reported by our group and others, but in those studies the map features lack lipid acyl tails and could not be ascertained as PIP₂^30,38,40,41 (Fig. S4a,b). The lipid tails of the second PIP₂ molecule are sandwiched between SUR1 - TMD0 and Kir6.2-TMs, with one chain running along TM1 and TM4 of SUR1-TMD0 (modeled as the arachidonoyl chain) and the other between TM1 of SUR1-TMD0 and TMs of Kir6.2 (modeled as the stearoyl chain), and have non-polar interactions with SUR1 (V34, F38, I42, P45, I46, L135, I137, A138, I141) and Kir6.2 (L56, V59, I159) respectively (Fig. 2b, c). Long-chain PIP₂ stably activates K_ATP channels^11,12; in contrast, activation of channels by synthetic short-chain PIP₂ such as diC8-PIP₂ is readily reversible³³. The extensive hydrophobic interactions between channel subunits and the long acyl chains of C18:0/C20:4 PI(4,5)P₂ observed in our structure may contribute to the apparent increase in stability of channel activation by long-chain PIP₂.

Functional role of the two PIP₂ binding sites

To probe the functional role of the two PIP₂ binding sites, we compared the PIP₂ response of WT channels to channels containing the following mutations: Kir6.2^R176A, which is predicted to weaken the first PIP₂ binding site; SUR1^K134A, which is expected to weaken the second PIP₂ binding site; Kir6.2^R176A and SUR1^K134A, which weaken both PIP₂ binding sites (Fig. 3a). Exposure to brain-derived C18:0/C20:4 PI(4,5)P₂ increased activity that is resistant to washout and decreased ATP inhibition over time in WT channels, as has been shown previously^11,12. Similar effects of PIP₂ were observed in all three mutant channels (Fig. 3a). However, the initial currents upon membrane excision, the extent of current increase upon PIP₂ stimulation, and the PIP₂ exposure time required for currents to reach maximum showed striking differences among the different channels (Fig. 3b-d). WT channels exhibited robust activity upon membrane excision into ATP-free solution, and currents increased by 1.48 ± 0.23-fold (Fig. 3a) that plateaued within one minute of PIP₂ exposure. Perturbation of the second PIP₂ binding site by SUR1^K134A resulted in channels that showed reduced initial currents (596.91 ± 218.23 pA vs. 1488.53 ± 348.86 pA of WT channels), which increased by 3.96 ± 0.47-fold in response to PIP₂. Perturbation of the first PIP₂ binding site by Kir6.2^R176A markedly reduced initial currents (42.44 ± 11.12 pA), which increased by 48.74 ± 18.25-fold after PIP₂ exposure. Combining SUR1^K134A and Kir6.2^R176A yielded channels that showed barely detectable currents at patch excision (12.62 ± 4.34 pA) and a dramatic 299.65 ± 121.72-fold current increase by PIP₂. All three mutants also required longer PIP₂ exposure to reach maximum currents in the order of double mutant (419.33 ± 57.86 sec) > Kir6.2^R176A (298.57 ± 73.76 sec) > SUR1^K134A (193.83 ± 33.66 sec), compar ed to WT channels (33.17 ± 3.18 sec) (Fig. 3d). These results demonstrate that both PIP₂ binding sites have functional roles in K_ATP channel activity.

Figure 3.

Kir6.2 pore in open conformation

In Kir channels, there are three constriction points in the K⁺ conduction pathway: the selectivity filter on the extracellular side, the inner helix gate at the helix bundle crossing where four inner helices (M2) converge, and the G-loop gate in the cytoplasmic pore just below the membrane¹³. The structure of the PIP₂-bound SUR1/Kir6.2^Q52R channel has an open Kir6.2 inner helix gate (Fig. 4a). Clear cryoEM density shows rotation of the Kir6.2-F168 side chains away from the pore’s center, thus causing a dilated pore size (Figs. 4a, 5a, S4a, movie 1). The radius of the K⁺ pathway at the inner helix gate is ∼3.3 Å, compared to ∼0.5 Å in the WT K_ATP channel bound to inhibitors we reported previously^38,40, exemplified by the ATP and repaglinide (RPG) bound structure (PDB ID 7TYS)³⁸ (Fig. 4b,c). The radius at the inner helix gate in our current structure is comparable to that in the open structure of SUR1/Kir6.2^C166S,G334D (3.3 Å)²⁸, and the pre-open structure of SUR1-Kir6.2^H175K fusion channel (3.0 Å)²⁹ (Fig. S4b). In contrast to the inner helix gate, little difference is observed in the G-loop gate between closed and open structures (Fig. 4), suggesting minimum involvement of the G-loop gate in K_ATP channel gating by PIP₂.

Figure 4.

Figure 5.

The SUR1/Kir6.2^Q52R K_ATP sample was purified in the absence of nucleotides; as expected, no ATP cryoEM density was observed at the inhibitory ATP-binding site on Kir6.2. Nonetheless, the open conformation includes structural rearrangements at the ATP binding site that would disfavor ATP binding (Fig. 5a,c, S4a,c). Specifically, the sidechains of Kir6.2-K39 and R50 flip away from ATP-interacting positions, with the K39 sidechain now coordinating PIP₂ at the second site; moreover, the sidechain of Kir6.2-E51 forms a salt bridge with R54 to partially occlude the ATP binding pocket (Fig. 5c). Our observation agrees with previous hypotheses that the open channel conformation is not compatible with ATP binding at Kir6.2 and that ATP inhibits the channel by stabilizing the channel in the closed conformation¹⁶.

Relative to channels bound to inhibitors, in which the Kir6.2-CTD is predominantly positioned close to the plasma membrane (“CTD-up” conformation)³⁸, the SUR1/Kir6.2^Q52R open channel structure shows the Kir6.2-CTD is also juxtaposing the membrane but further rotated clockwise (viewed from the extracellular side) (compare Fig. 5a and 5b, movie 1). This twisting as well as the other structural changes at the ATP binding site discussed above are similarly observed in the open structure of the SUR1/Kir6.2^C166S,G334D channel²⁸, and the pre-open structure of the SUR1-Kir6.2^H175K fusion channel²⁹ (Fig. S4b,d). The converging observations in three open structures using different mutant constructs suggest WT K_ATP channels undergo similar conformational changes between closed and open states.

Analysis of SUR1 structure

In the PIP₂-bound open SUR1/Kir6.2^Q52R structure, the two NBDs of SUR1 are not well resolved; in particular, NBD2 is highly dynamic with a local resolution > 5 Å (Figs. 1, S1d). However, the two NBDs are clearly separated, as predicted since no MgATP or MgADP was added to the sample. Notably, no cryoEM density corresponding to the distal N-terminal peptide of Kir6.2^{Q5 2 R} was observed in the central cleft lined by TM helices from TMD1 and TMD2 of the SUR1-ABC core even when the map was filtered to 7 Å (Fig. S5a); no continuous density corresponding to KNtp was observed at an alternative site to suggest a stable resting position, precluding structural modeling of this flexible domain. The absence of KNtp cryoEM density in the PIP₂ - bound open SUR1/Kir6.2^Q52R structure is in contrast to the clear KNtp density observed in the central cleft of SUR1-ABC core in the apo closed SUR1/Kir6.2^Q52R structure at comparable resolution and contour level (Fig. S2e, S5). Presence of KNtp in the SUR1-ABC core has been previously reported in inhibitor-bound structures and linked to channel closure^38,41,42. Worth noting, in a publsihed open K_ATP structure where NBDs are bound to MgATP/MgADP and dimerized, KNtp density was also absent²⁸. The lack of KNtp in the SUR1-ABC core in the open SUR1/Kir6.2^Q52R structure further enforce the correlation between the absence of KNtp in the central cleft of the SUR1-ABC core and channel opening, whether SUR1-NBDs are dimerized or not.

The ABC modules of the four SUR1 subunits in the SUR1/Kir6.2^Q52R map from a C4 non-uniform refinement adopt a propeller-like conformation (Fig. 1a, b). Further 3D classification of the symmetry expanded particle set revealed three distinct rotational positions of SUR1-ABC module relative to the Kir6.2-SUR1-TMD0 tetrameric core (Fig. S6a,b). Similar conformational dynamics have been reported for the SUR1 NBDs-dimerized, open structure of the SUR1/Kir6.2^C166S,G334D channel²⁸, thus likely represent intrinsic flexibility of this domain regardless of nucleotide binding and/or NBD dimerization. In addition to the dynamics of the SUR1-ABC module within the PIP₂-bound open SUR1/Kir6.2^Q52R conformation class, more pronounced heterogeneity was also captured in 3D classification of symmetry expanded particles that include all particles from 2D classes (Fig. S2). Here, rotation of the SUR1-ABC module is seen correlated with the corkscrew position of the Kir6.2-CTD, with the SUR1-ABC module rotated in towards the Kir6.2 tetramer in the apo closed CTD-down conformation class, and more flung out in the PIP₂-bound open CTD-up conformation (Figs. S2c,d, 6c,d). Such correlated dynamic movements of SUR1 and Kir6.2 may reflect the conformational transiti on between closed and open channels.

Figure 6.

Compared to inhibitor-bound closed Kir6.2/SUR1 channel structures [reviewed in^6,7], SUR1 subunits in the SUR1/Kir6.2^Q52R open structure are more tilted away from Kir6.2 and elevated towards the plasma membrane (viewed from the side) such that they are nearly in plane with the Kir6.2 tetramer (compare Fig. 1b and 1c). Accompanying this upward SUR1 motion, there is an overall rigid-body rotation of SUR1 driven by a rotation of the TMD0 of SUR1 in the open structure relative to closed structures (Fig. 6). At the outer leaflet, SUR1-F41, which interacts with Kir6.2-L84 in the closed conformation, now interacts with Kir6.2-L85 in the open conformation (Fig. 6c). At the inner leaflet, TMD0 moves away from Kir6.2 (Fig. 6a, 6c), cr eating a space between SUR1-TM1 and Kir6.2-M1. Using PISA analysis⁴³, the contact surface area at this interface (between SUR1-TM1 residues 27-54 and Kir6.2-M1 residues 66-96) is 645.6 Ų (ΔG = −20 kcal/mol) in the closed channel structure (PDB ID 6BAA), which is reduced to 395.9 Ų (ΔG = −12 kcal/mol) in the SUR1/Kir6.2^Q52R open structure. Binding of a second PIP₂ in this space (see Fig. 2b) compensates for the lost surface contact between SUR1-TMD0 and Kir 6.2, thus stabilizing this interface in the open channel.

Similar structural changes were also described in the pre-open SUR1-Kir6.2^H175K fusion structure where SUR1 NBDs are bound to MgATP/MgADP and dimerized³⁰ (Fig. S4). Dimerization of SUR1 NBDs antagonizes ATP inhibition at Kir6.2 and stimulates K_ATP channel activity. It has been proposed that NBD dimerization causes outward bending of the cytoplasmic half of SUR1-TMD0 and the pulling away of K205 in SUR1-L0 from binding ATP at the inhibitory site to stimulate channel activity³⁰. Since our open SUR1/Kir6.2^Q52R structure does not contain MgATP/MgADP and the NBDs are clearly separated, the similar conformational changes we observed at SUR1-TMD0/L0 are not driven by SUR1 NBD dimerization. One possibility is that MgATP/MgADP and PIP₂ converge on the same structural mechanism to open the channel, as has been proposed previously based on functional studies⁴⁴.

Kir6.2-Q52R stabilizes the open conformation by interacting with SUR1-W51

Kir6.2-Q52 is located immediately N-terminal to the interfacial helix (IFH; a.k.a. slide helix) and C-terminal to βA (Fig. 5c,d). In published ATP-bound closed structures, Kir6.2-Q52 lies close to SUR1-E203 in the proximal portion of the SUR1-L0 linker (Fig. 5d). Previous studies have shown that enforcing Kir6.2-Q52 and SUR1-E203 interactions via engineered charged amino acid pair or cysteine crosslinking reduced ATP inhibition IC₅₀ by 100-fold or caused spontaneous channel closure in the absence of ATP, respectively⁴⁵. These results indicate that stabilizing the interface between Kir6.2 βA-IFH and SUR1-L0 stabilizes ATP binding and channel closure. In contrast, in the SUR1/Kir6.2^Q52R open structure, the SUR1-L0 linker moves ∼7 Å towards SUR1 and away from the ATP-binding site on Kir6.2 (compare Fig. 5c and 5d), with SUR1-E203 and K205 flipped away from Kir6.2 such that SUR1-TMD0 ICL1 (aa 52-60) engages with Kir 6.2 - βA to form a main chain β-sheet (Fig. 5c). In particular, the sidechain of Kir6.2 residue 52 (Q in WT, R in mutant) now faces SUR1-TMD0 and interacts with SUR1-W51 (Figs. 5c, S4f), tethering the Kir6.2-CTD to SUR1-TMD0 in a rotated open position.

The Kir6.2-Q52R mutation increases P_o and decreases ATP inhibition of K_ATP channels in a SUR1-dependent manner⁴⁶. The cation-π interaction observed between Kir 6.2-Q52R and SUR1-W51 in the open conformation (Fig. 7a) led us to hypothesize this interaction underlies the gain-of-function gating effect of the Kir6.2-Q52R mutation, by stabilizing Kir6.2-CTD in a rotated open conformation. To test this, we mutated W51 of SUR1 to cysteine and assessed the ATP sensitivity of channels formed by co-expressing Kir6.2^Q52R and SUR1^W51C using inside-out patch-clamp recording. SUR1^W51C reversed the effect of Kir6.2^Q52R such that the ATP sensitivity of the channel resembles WT channels (Fig. 7b). The IC₅₀ values of ATP inhibition for WT, SUR1/Kir6.2^Q52R, SUR1^W51C/Kir6.2, and SUR1^W51C/Kir6.2^Q52R channels are 6.25 ± 0.44 μM, 161.2 ± 16.23 μM, 20.88 ± 1.96 μM, and 8.98 ± 0.49 μM, respectively (Fig. 7c). Corroborating these findings, in Rb⁺ efflux assays cells co-expressing Kir6.2^Q52R and SUR1^W51C showed efflux levels similar to cells expressing WT channels, in contrast to the significantly higher efflux in cells co-expressing Kir6.2^Q52R and WT-SUR1 (Fig. 7d). These results provide strong evidence that Kir6.2-Q52R interacts with SUR1-W51 to enhance channel activity and reduce ATP inhibition, thus explaining the SUR1-dependent pathophysiology of this neonatal diabetes mutation.

Figure 7.

To test whether the interaction between Kir6.2-Q52R and SUR1-W51 that gives r ise to the enhanced activity and reduced ATP sensitivity is dependent on PIP₂ binding at the two sites revealed by our structure, we tested the effect of Kir6.2^Q52R on the background of SUR1^K134A, Kir6.2^R176A, or SUR1^K134A and Kir6.2^R176A, to weaken the novel PIP₂ binding site, the conserved PIP₂ binding site, or both. Inside-out patch-clamp recording experiments showed a leftward shift in the dose response of ATP inhibition of the SUR1^K134A/Kir6.2^Q52R channel (IC₅₀= 54.08 ± 7.14 μM, H = 1.55), the SUR1/Kir6.2^Q52R/R176A channel (IC₅₀=17.97 ± 1.20 μM, H = 1.41), and the SUR1K^134A/Kir6.2^Q52R/R176A channel (IC₅₀= 8.30 ± 1.11 μM, H = 1.18) relative to the SUR1/Kir6.2^Q52R channel (IC₅₀= 161.20 ± 16.23 μM, H = 2.34) (Fig. S7a). SUR1^K134A or Kir6.2^R176A alone had little effect on the IC₅₀ of ATP inhibition (4.12 ± 0.30 μM, H = 1.54, and 11.27 ± 1.95 μM, H = 0.84, respectively, compared to 6.25 ± 0.44 μM, H = 0.93 for WT channels). In agreement, Rb⁺ efflux experiments showed attenuation of the gain-of-function phenotype associated with Kir6.2^Q52R when combined with PIP₂ binding mutations SUR1^K134A, Kir6.2^R176A, or both (Fig. S7b). Taken together, these results support a role of PIP₂ at the two binding sites in manifesting the gain-of-function effect of Kir6.2^Q52R.

Discussion

In this study, we surmounted the technical difficulties in capturing PIP₂ bound to K_ATP channel, firstly by using a neonatal-diabetes Kir6.2^Q52R variant that enhances channel activity in an SUR1-dependent mechanism, and secondly by using natural brain-derived long-chain PIP₂ before membrane solubilization, rather than the previously used approach of adding short-chain synthetic diC8-PIP₂ after channel purification^24,30,47. The cryoEM map of the lipid densities with long acyl chain tails, together with functional studies on designed K_ATP channel mutants, permitted the assignment of PIP₂ at both the expected site and the novel binding site (Figs. 2b,c, S3b, S4a,b). The novel PIP₂ binding site, coordinated by Kir6.2 and its regulatory subunit SUR1, uncovers an unprecedented cooperation between an ion channel and transporter on PIP ₂ binding, and highlights the uniqueness of the K_ATP complex in the evolution of K⁺ channels and ABC proteins. The structure resolves a major problem in the long-standing puzzle of how SUR1, in particular TMD0 of SUR1, modulates the open probability of the Kir6.2 channel^25,26. Moreover, it answers a long-standing question of how a prominent clinical Kir6.2 mutation increase s channel activity to cause neonatal diabetes⁴⁶.

Unique PIP₂ binding pocket in K_ATP channels

As a member of the Kir channel family, Kir6.2 has a homologous PIP2 binding site akin to that identified in Kir2 and Kir3^17–20. Surprisingly, our structure reveals a second, novel PIP ₂ binding site immediately next to the conserved Kir channel PIP₂ binding site. This second site is uniquely coordinated by both SUR1 and Kir6.2 subunits. Mutational studies support functional importance of PIP₂ binding at both sites (Fig. 3). In single channel recordings, Kir6.2 channels lacking SUR1 have low P_o with brief openings, in contrast to Kir6.2 channels assembled with SUR1 or SUR1-TMD0, which exhibit ∼10-fold higher P_o and long bursts of openings^16,25,26,48. We propose the low P_o and brief openings in Kir6.2 channels lacking SUR1 result from PIP₂ binding at the conserved site, which tethers Kir6.2-CTD near the membrane to rotate to the open position. PIP₂ binding at this site may not be as strong as in Kir2 and Kir3 channels due to sequence variations at key residues including Kir6.2 P69 and H175 (both positively charged amino acids in Kir2 and 3)^17–20, explaining the low P_o of Kir6.2. PIP₂ binding at the unique second site formed by SUR1 and Kir6.2 enables SUR1 to stabilize Kir6.2-CTD in the rotated open position, giving rise to bursts of openings and higher P_o.

The PIP₂ binding pocket is smaller in the previously published inhibitor-bound closed structure than in the SUR1/Kir6.2^Q52R open structure. In a solvent accessible surface representation of the model for a closed structure (exemplified by the repaglinide/ATP bound PDB ID 7TYS)³⁸, PIP₂ at the second site would clash with SUR1 (Fig. S8b). Inspection of the closed SUR1/Kir6.2^Q52R CTD-down conformation (see Fig. S2) also finds that PIP₂ binding at the second site would clash with SUR1 (Fig. S8c), indicating this structure represents an apo state, at least with respect to PIP₂ binding at the second novel site. These observations imply that PIP₂ moves in and out of the second site as the channel opens and closes. In this regard, it is interestingly to note that a conformation-dependent change in the PIP₂ binding site in the voltage-dependent KCNQ1 (Kv7.1) channel was recently reported, where movement of the voltage sensing transmembrane helix S4 up or down the membrane is associated with opening or occlusion of the PIP₂ binding site to open or close the channel, respectively⁴⁹. A question that arises is whether smaller lipids may occupy the more constricte d second site as the channel transitions from the open to the closed conformation. Due to resolution differences, it is not possible to resolve lipid density in the apo closed SUR1/Kir6.2 ^Q52R channel map. However, in our published closed structures without exogenous PIP₂, there was also lipid density, which we tentatively modeled as two phosphatidylserines³⁸. Another question is whether the first PIP₂ binding site is always occupied by PIP₂ since it appears available in both the open and closed channel conformations (Fig. S8). More studies are needed to answer these questions.

K_ATP channels exhibit lower specificity towards PIPs compared to other Kir channels, and are activated equally well by PI(4,5)P₂, PI(3,4)P₂, and PI(3,4,5)P₃, and also by PI(4) P and long chain (LC)-CoAs³³. The structural basis underlying the reduced PIPs specificity for K_ATP channel activation requires further investigation. The aforementioned sequence variations at key PIP₂ coordinating residues at the site homologous to other Kir channels may contribute to the lower specificity. Additionally, the large size of the tandem PIP₂ binding pocket may accommodate different PIPs and LC-CoAs. Such degeneracy may account for the observation that purified K_ATP channels reconstituted in lipid bilayers lacking PIP₂ exhibited spontaneous single channel openings²⁸, in contrast to Kir2 and Kir3 channels which require PIP₂ for activity^20,50.

Mechanism of PIP₂ and ATP antagonism

PIP₂ and ATP functionally compete to open or close K_ATP channels (Fig. 3a)^11,12. Kinetic analyses have indicated that PIP₂ and ATP binding are mutually excluded¹⁶. Comparison between PIP₂-bound SUR1/Kir6.2^Q52R open structure and previously published ATP-bound closed structures shows PIP₂ and ATP antagonism occurs via both binding competition and allosteric mechanisms (Fig. 5, movie 1). At the level of binding competition, both ligands compete for a common binding residue, Kir6.2-K39, which interacts with ATP in the ATP-bound closed structures^38,40 but with PIP₂ in the PIP₂-bound open structure (Figs. 2b, 5c). Previous molecular dynamics simulations, using ATP-bound closed structures of Kir6.2 tetramer⁵¹ or tetramer of Kir6.2 plus SUR1-TMD0³⁸ with a single PIP₂ molecule docked in the conserved PIP₂ binding site, found that Kir6.2-K39⁵¹ or both K39 and R54³⁸ switched between ATP binding and PIP₂ binding. In the SUR1/Kir6.2^Q52R open structure, we see Kir6.2-K39, rather than binding PIP₂ in the conserved site, binds the second PIP₂ in the novel site. In contrast, Kir6.2-R54 is not involved in PIP₂ binding; instead, it hydrogen bonds with Kir6.2-E51 in the same subunit and E179 in the adjacent Kir6.2 subunit, which stabilizes the interface between the IFH and the C-linker helix (the linker helix connecting M2 and C-terminal cytoplasmic domain of Kir 6.2) in the open conformation. Mutation of both K39 and R54 to alanine has been shown to reduce ATP as well as PIP₂ sensitivities⁵². Our structure clarifies the structural role of K39 and R54 and suggests while K39A likely reduces PIP₂ sensitivity by weakening PIP₂ binding at the second site, R54A likely reduces PIP₂ sensitivity indirectly by disrupting interactions in Kir6.2 that are needed to stabilize the open channel conformation.

Allosterically, PIP₂ binding diverts ATP binding residues away from the ATP binding pocket to disfavor ATP binding. These include sidechain reorientation of Kir 6.2 K39 and R50, and interaction between Kir6.2 R54 and E51 that partially occludes the ATP binding pocket (Fig. 5c). Moreover, in the ATP-bound closed conformation, SUR1-L0 interfaces with Kir 6.2 - βA, allowing SUR1-K205 to coordinate ATP binding^38,42,45,53 and stabilize channel closure. However, in the open structure, SUR1-L0 is disengaged from Kir6.2-βA, allowing Kir6.2-CTD to rotate such that ICL1 of SUR1-TMD0 engages with Kir6.2-βA, forming a continuous β-sheet to stabilize the open conformation, which also disfavors ATP binding. Similar changes have been described in published SUR1/Kir6.2^C166S,G334D open structure²⁸, and SUR1-Kir6.2^H175K fusion pre-open structure³⁰ (Fig. S4f). The common structural rearrangements from closed to open state seen independent of mutations support the same gating transition in WT channels and highlight a key role of SUR1 in stabilizing the Kir6.2-CTD in two distinct rotational positions to close or open the K_ATP channel.

Insights on disease mutations

The study provides mechanistic insight on how Kir6.2^Q52R causes neonatal diabetes. The strong cation-π interaction engendered by the Kir6.2-Q52R mutation with SUR1-W51 illustrates how changes at this interface have profound effects on channel gating and physiology. In the published SUR1/Kir6.2^G334D,C166S open structure and the SUR1-Kir6.2^H175K fusion channel pre-open structure Q52 of Kir6.2 is similarly in position to interact with W51 of SUR1^28,29 (see Fig. S4f). The polar-π interaction between glutamine and tryptophan is much weaker than the cation-π interaction between arginine and tryptophan. The difference in binding energy for the relevant gas-phase interactions for polar-π is 1.4 kcal/mol (NH₃-Benzene), compared to the cation-π interaction energy of 19.0 kcal/mol (NH₄⁺-Benzene)^54,55. That Q52 and Q52R in Kir 6.2 both interact with W51 of SUR1 suggests an important role of Kir6.2-Q52 in stabilizing Kir6.2-CTD and SUR1-TMD0 interface for channel opening. Functional studies showing that weakening PIP₂ binding at either sites attenuate both WT and SUR1/Kir6.2^Q52R activity indicates endogenous PIP₂ is required to support the activity of both WT and mutant channels. In the absence of normal PIP₂ binding, the gain-of-function effect of Kir6.2^Q52R becomes much weaker. The presence of a minor conformation class corresponding to an apo closed SUR1/Kir 6.2 ^Q52R channel (Fig. S2) even in the presence of excess PIP₂ further strengthens the notion that Kir6.2^Q52R alone is insufficient to cause disease in the absence of PIP₂.

In addition to Kir6.2^Q52R, many disease mutations are located at the SUR1-Kir6.2 interface seen in the SUR1/Kir6.2^Q52R open structure (Fig. 8). These include congenital hyperinsulinism (HI) associated loss-of-function Kir6.2 mutations: R54C, L56G, K67D, R176H and R177W, and SUR1 mutations: I46T, P133R, and L135V, as well as neonatal diabetes (ND) and Developmental delay, Epilepsy, and Neonatal Diabetes (DEND) syndrome associated gain-of-function Kir6.2 mutations: K39R, E51A/G, Q52L/R. G53D/N/R/S/V, V59A/G/M, W68C/G/L/R, K170N/R/T, E179A/K, and SUR1 mutations: P45L, I49F, F132L/V, and L135P (Fig. 8) ⁵⁶. Some of these, such as HI-associated K67D and R176H, and ND/DEND associated K39R, W68C/G/L/R involve residues that coordinate PIP₂ binding; others however, likely affect PIP ₂ or ATP gating allosterically by stabilizing channels in closed or open conformations.

Figure 8.

In summary, the PIP₂-bound K_ATP channel structure elucidates the intimate partner ship between SUR1 and Kir6.2 in coordinating PIP₂ binding and channel gating. The unique PIP₂ binding pocket and the molecular interactions involved in gating provide a framework for designing new K_ATP modulators to control channel activity. The finding that SUR1 binds PIP₂ to regulate Kir6.2 also raises the question whether PIP₂ serves structural and functional roles in other ABC proteins, as has been implicated for CFTR⁵⁷.

Materials and methods

Cell lines

COSm6 cells were cultured in high-glucose DMEM medium (GIBCO) supplemented with 10% Fetal Bovine Serum (Fisher Scientific), 100 U/mL penicillin, and 100 U/mL streptomycin at 37°C with 5% CO₂.

Protein expression and purification

Genes encoding rat Kir6.2^Q52R and N-terminal FLAG-tagged (DYKDDDDK) hamster SUR1 were cloned into pShuttle vectors and then the AdEasy vector (Stratagene), and packaged into recombinant adenoviruses, which were used for protein expression as described previously ⁵⁸. However, instead of inducing K_ATP channel expression in INS-1 cells, which are pancreatic β cell derived and have endogenous WT Kir6.2 and SUR1 expression that would mix with the Q52R variant, a different mammalian adhesion cell line was used (COSm6 cells). COSm6 cells grown to mid-log in 15 cm tissue-culture plates were infected with adenoviruses packaged with Kir 6.2, SUR1, and tTA, using multiplicity of infections (MOIs) optimized empirically. Note the pShuttle vector used for SUR1 contains a tetracycline-regulated response element, necessitating co-infection of a tTA (tetracycline-controlled transactivator) adenovirus for SUR1 expression. At ∼24 hours post-infection, cell medium was changed and included 100 μM tolbutamide post-infection. Tolbutamide enhances K_ATP channel expression at the plasma membrane and washes out easily. At ∼48 hours post-infection, cells were harvested by scraping and cell pellets were frozen and stored at −80°C until purification.

SUR1/Kir6.2^Q52R channels were purified as previously described⁵⁸, except after the total membrane fraction was prepared, membranes were re-suspended in buffer containing 0.2 M NaCl, 0.1 M KCl, 0.05 M HEPES pH 7.5, 4% trehalose, and 1 mg/mL brain PIP₂ (Avanti Polar Lipids), and incubated at 4°C for 30 minutes before increasing the volume 10x in the same buffer without added lipids but including the detergent digitonin at a final concentration of 0.5%, then membranes were solubilized at 4°C for 90 minutes. As previously described⁵⁸, after clarification by ultracentrifugation, the soluble fraction was incubated with anti-FLAG M2 affinity agarose for 10 hours and eluted at 4°C for 60 minutes with a buffer containing 0.2 M Na Cl, 0.1 M KCl, 0.05 M HEPES pH 7.5, 0.05% digitonin and 0.25 mg/mL FLAG peptide. Purified channels were eluted at ∼170 nM (∼150 μg/ml) and used immediately for cryo-EM grid preparation. K_ATP channel particles were also fixed and stained using Uranyl Acetate and sample quality was assessed by negative-stain-EM.

CryoEM sample preparation and data acquisition

To increase protein adsorption to the cryoEM grids, and also to mitigate selective-orientation of K_ATP channel particles that occurs on commercial carbon surfaces, graphene-oxide ( GO) grids were prepared as previously described⁵⁹. Briefly, gold Quantifoil R1.2/1.3 400 mesh grids were cleaned with acetone and glow-discharged for 60 seconds at 15 mA with a Pelco EasyGlow®, and 4 µL of 1 mg/mL Polyethylenimine (PEI, 40,000 MW) in 25 mM HEPES pH 7.9 was applied to each grid and incubated for 2 minutes followed by washing with water. Then, 0.1 mg/ml GO was vortexed vigorously and applied to the grid and incubated for 2 minutes followed by two washes with water. The GO grids were allowed to dry for 15 minutes and used within 2 hours for sample vitrification.

To prepare cryoEM samples, 3 µL of purified K_ATP channel complex was loaded onto fresh GO-coated grids for 30 s at 6 °C with a humidity of 100%. Grids were blotted for 2.5 seconds with a blotting force of −10 and cryo-plunged into liquid ethane cooled by liquid nitrogen using a Vitrobot Mark III (FEI).

Single-particle cryo-EM data was collected on a Titan Krios 300 kV cryo-electron microscope (ThermoFisher Scientific) in the Pacific Northwest CryoEM Center (PNCC), with a multi-shot strategy using beam shift to collect 27 movies per stage shift, assisted by the automated acquisition program SerialEM. Images were recorde d on the Gatan K3 Summit direct-electron detector in super-resolution mode, post-GIF (20eV window), at 105,000x magnification (calibrated image pixel-size of 0.826 Å; super-resolution pixel size 0.413 Å); nominal defocus was varied between −1.0 and −2.5 µm across the dataset. The dose rate was kept around 24 e^-/Ų/sec, with a frame rate of 35 frames/sec, and 78 frames in each movie ( i.e. 2.2 sec exposure time/movie), which gave a total dose of approximately 55 e^-/Ų. Three grids that were prepared in the same session using the same protein preparation were used for data collection, and from these three grids, 2727, 3956 and 1576 movies (8,259 movies total) were recorded.

CryoEM image processing

Super-resolution dose-fractionated movies were gain-normalized by inverting the gain reference in Y and rotating upside down, corrected for beam induced motion, aligned, and dose-compensated using Patch-Motion Correction in cryoSPARC2⁶⁰ without binning. Parameters for the contrast transfer function (CTF) were estimated from the aligned frame sums using Patch-CTF Estimation in cryoSPARC2⁶⁰ and binned by 2 with Fourier cropping. Micrographs were manually curated using sorting by curate exposures. The resulting 5241 dose-weighted motion-corrected summed micrographs were used for subsequent cryo-EM image processing. Particles were picked automatically using template-based picking in cryoSPARC2 based on 2D classes obtained from cryoSPARC live during data collection. For each of the three sets of data, particles were cleaned by three rounds of 2D classification in cryoSPARC2 ⁶⁰. The combined particle stack contained 70,638 particles, which were then used for ab initio reconstruction in C1 requesting two classes. Only classes that had good alignments and contained full channel particles were included in subsequent rounds of classification. 40,896 particles from the best class were then subjected to ab initio reconstruction in C1 requesting three classes, which gave a class with only side, a class with only top/bottom, and a good class with 23,378 particles. These 23,378 particles were used for a final ab initio reconstruction using three classes, and gave 21,663 particles that sorted into a class with a straight transmembrane region (14,992 particles) and a class where the transmembrane region was more bent with the Kir6.2 core puckered up towards the extracellular space (6,657 particles). Further rounds of ab initio reconstruction resulted in equivalent classes without further improvement in particle classification. The particles were re-extracted using a 600 pixel box at 0.8265 Å/pix, duplicate particles within 20 Å from each other on the micrograph were removed, then used as input for non-uniform refinement⁶¹ in C1 with the 6,657 particle map reconstruction at 6.4 Å resolution and the 14,115 particle map reconstruction at 3.4 Å resolution. A non-uniform refinement in C4 symmetry imposed with the 14,115 particles resulted in a 2.9 Å resolution reconstructed cryoEM map using cryoSPARC auto mask tightening (Fig. S1). The auto-tightened FSC mask excluded the disordered NBD2 (Fig. S1c). Masks for FSC calculation that included the full K_ATP channel with or without micelle were generated using molmap in ChimeraX ⁶² (Fig. S1b,c), and resampled on the full map grid (600³ pixels). The FSC calculated using these static masks yielded 3.3 Å resolution for the full channel without micelle (Fig. S1b) and 3.5 Å resolution for the entire particle including the micelle (“Loose” in the FSC plot shown in Fig. S1b).

To assess conformational heterogeneity for the entire dataset, duplicate and corrupt particles were removed from the set of 70,638 particles from 2D classification to give 66,049 particles and refined using cryoSPARC homogeneous refinement with C4 symmetry imposed (4.0 Å resolution), then 4-fold symmetry expanded to give 264,196 particles (Fig. S2a). A mask covering the Kir6.2 tetramer, four TMD0 densities in SUR1, plus a large volume surrounding one of the SUR1 subunits, to have a large region covering all possible SUR1 locations, was generated using Chimera and used as a focused map to conduct 3D classification of symmetr y expanded particles without particle alignment in cryoSPARC2. This yielded two dominant classes: class 1 ( 106,781 particles), which resembles the open structure derived using the workflow shown in Fig. S1a, and class 2 (53,201 particles), which has the Kir6.2^Q52R in the CTD-down position resembling previously published apo closed WT channel³⁸. Particles from each of these classes were further classified into four classes using cryoSPARC 3D classification, and the classes with the most divergent conformations were compared (Fig. S2b-e). The reconstructed map resembling the published apo closed WT structure (PDB ID 7UQR) was well-fit by a rigid-body of that structure, but the resolution of the map was not sufficient to build a detailed atomic model.

To assess conformational heterogeneity within the high-resolution class of 14,115 particles of full-channel complexes that may have individual SUR1 subunits adopting independent conformations within a single K_ATP channel particle, particles were subjected to C4 symmetry expansion and 3D classification without particle alignment in cryoSPARC2 using the same mask described above. An initial focused 3D classification searching for 5 classes sorted particles into three dominant classes (class 0: 925 particles; class 1: 22474 particles; class 2: 10711 particles; class 3: 25664 particles; class 4: 194 particles). Class 1 to 3 showed three distinct SUR1 conformations, and subsequent local refinement using a loose mask covering the Kir6.2 tetramer, four TMD0 domains of SUR1, and one SUR1, resulted in map reconstr uctions of 3.12 Å, 3.99 Å, and 3.07 Å resolutions for class 1, 2, and 3, respectively (Fig. S6a,b). In all three of these particle classes, the NBDs are separated, but with substantial differences between the relative positions of NBD2, and all have better resolved maps corresponding to the NBD2 than for the overall consensus C4 non-uniform refinement.

In the C4 reconstruction of the full K_ATP channel particle, there is good density for nearly every side chain of Kir6.2 except for the N-terminal 31 residues and the C-terminal residues beyond residue 352. TMD0 and the transmembrane helicies of SUR1 also have clear density, with unclear density for the dynamic NBDs and loop regions, especially NBD2 which is poorly resolved due to apparent high flexibility when in the open conformation without NBD dimerization conditions.

To create an initial structural model, PDB ID 6BAA was fit into the reconstructed density for the full K_ATP channel particle using Chimera⁶³, and then refined in Phenix⁶⁴ as separate r igid bodies corresponding to TMD (32-171) and CTD 172-352) of Kir6.2 and TMD0/L0 (1-284), TMD1 (285-614), NBD1 (615-928), NBD1-TMD2-linker (992-999), TMD2 (1000-1319) and NBD2 (1320-1582). All sidechains and missing loops that had clear density were then built manually using Coot⁶⁵. The resulting model was further refined using Coot and Phenix iteratively until the statistics and fitting were satisfactory (Table S1). Two models are presented. The first (PDB ID 8TI2) contains residues 32-352 for Kir6.2, and residues 1-1578 for SUR1 except for an extracellular loop (1043-1060), two loop regions (623-673 and 743-766) in NBD1 and the linker between NBD1 and TMD2 (928-986). The second model (PDB ID 8TI1) has the highly flexible NBD2 removed, and the SUR1 model ends at residue 1317. NBD1/2 and loop regions showed signs of disorder, thus many sidechains of residues in these regions were stubbed at Cβ. The N-terminal of Kir6.2, which in the closed NBD-separated conformation rests between TMD1 and TMD2 of SUR1, is not observed in this conformation.

In addition to protein density, N-acetylglucosamine (NAG), which is the common core of N-linked glycosylation, is modeled in the distinctively large density at the side chain of N10 in each SUR1 (Fig. 1b). Densities corresponding to two PIP₂ molecules were also observed at the interface between Kir6.2 and SUR1 and were modeled after considering fits of other lipid molecules. Three other lipid sites per SUR1 monomer provided a useful contrast to the bulky density of the PIP₂ head group (Fig. S3). The resulting model was further refined using Coot⁶⁵ and Phenix^64,66,67 iteratively until the statistics and fitting were satisfactory (Table S1). All structure figures were produced with UCSF Chimera⁶³, ChimeraX⁶², and PyMol (http://www.pymol.org). Pore radius calculations were performed with HOLE implemented in Coot ⁶⁵.

Electrophysiology

For electrophysiology experiments, COSm6 cells were co-transfected with various combination of WT or mutant SUR1 and Kir6.2 cDNAs along with the cDNA for the Green Fluorescent Protein (to facilitate identification of transfected cells) using FuGENE® 6. Cells were plated onto glass coverslips twenty-four hours after transfection and recordings made in the following two days. All experiments were performed at room temperature as previously described ⁶⁸. Micropipettes were pulled from non-heparinized Kimble glass (Fisher Scientific) on a horizontal puller (Sutter Instrument, Co., Novato, CA, USA). Electrode resistance was typically 1 −2 MΩ when filled with K-INT solution containing 140 mM KCl, 10 mM K-HEPES, 1 mM K-EGTA, 1 mM EDTA, pH 7.3. ATP was added as the potassium salt. Inside-out patches of cells bathed in K-INT were voltage-clamped with an Axopatch 1D amplifier (Axon Inc., Foster City, CA). ATP or porcine brain PIP₂ (Avanti Polar Lipids; prepared in K-INT and bath sonicated in ice water for 30 min before use) were added to K-INT as specified in the figure legend. All currents were measured at a membrane potential of −50 mV (pipette voltage = +50 mV). Initial currents prior to PIP₂ exposure shown in Fig. 3 were calculated by subtracting currents measured in K-INT with 1 mM ATP (which inhibits >99% of K_ATP currents) from currents in K-INT solution to account for leak currents. Data were analyzed using pCLAMP10 software (Axon Instrument). Off-line analysis was performed using Clampfit and GraphPad. Data were presented as mean ± standard error of the mean (S.E.M). ATP inhibition dose-response curves were obtained by fitting data to the Hill equation (I_rel = 1/(1 + ([ATP]/IC₅₀)^H)), where I_rel is the current relative to the maximum currents in K-INT solution (expressed as % current in Figs. 7c, S7a), IC₅₀ is the ATP concentration that causes half-maximal inhibition, and H is the Hill coefficient. Note H was allowed to vary for the curves shown in Figs. 7c, S7a.

Rb⁺ efflux assay

COSm6 cells were transiently transfected with various combination of WT or mutant SUR1 and Kir6.2 cDNAs using FuGENE® 6. Untransfected cells were included as background control. Cells were cultured in medium containing 5 mM RbCl overnight. The next day, cells were washed quickly twice in Ringer’s solution (5.4 mM KCl, 150 mM NaCl, 1 mM MgCl₂, 0.8 mM NaH₂PO₄, 2 mM CaCl₂, 25 mM HEPES, pH 7.2) with no RbCl. Rb efflux was measured by incubating cells in Ringer’s solution (for experiments shown in Fig. 7d) or Ringer’s solution supplemented with 5 mM glucose (for experiments shown in Fig. S7b) over a 40 min period. Inclusion of 5 mM glucose increases intracellular ATP/ADP ratios to suppress channel activity, allowing for better detection of channels showing mild gain-of-function phenotype as shown in some of the mutant channels in Fig. S7b). At the end of the 40 min incubation, Efflux solution was collected and cells lysed in Ringer’s solution plus 1% Triton X-100. Rb concentrations in both the efflux solution and cell lysate were measured using an Atomic Adsorption Instrument Ion Channel Reader (ICR) 8100 from Aurora Biomed. Percent efflux was calculated by dividing Rb in the efflux solution over total Rb in the efflux solution and cell lysate. For each experiment, duplicates were included as technical repeats and the average taken as the experimental value. For all channel combinations, 3-6 separate transfections were carried out in parallel as biological repeats. Data were presented as mean ± standard error of the mean (S.E.M) and statistical analysis performed by one-way ANOVA with Tukey’s post hoc test in GraphPad.

Data availability

The cryo-EM maps have been deposited in the Electron Microscopy Data Bank (EMDB), and the coordinates have been deposited in the PDB under the following accession numbers: PDB ID 8TI2 and EMD-41278 (NBD2 modeled as main chain atoms only); PDB ID 8TI1 and EMD-41277 (NBD2 not modeled).

Conflict statement

The authors declare that they have no competing financial or non-financial interests with the contents of this article.

Author contributions

CMD designed and performed experiments, analyzed data, prepared figures, wrote and edited the manuscript. YYK performed electrophysiology and Rb efflux experiments, analyzed data, prepared figures, and edited the manuscript. PZ analyzed data, prepared figures and edited the manuscript. AE performed Rb efflux experiments, prepared figures and edited the manuscript. SLS conceived the project, performed electrophysiology experiments, prepared figures, wrote and edited the manuscript.

Supporting information

Supplemental Movie

Acknowledgements

A portion of this research was supported by NIH grant U24GM129547 and performed at the Pacific Northwest Cryo-EM Center (PNCC) at Oregon Health and Science University and accessed through EMSL (grid.436923.9), a DOE Office of Science User Facility sponsored by the Office of Biological and Environmental Research, with special thanks to Dr. Nancy Meyer for help with cryoEM data collection. We acknowledge the support by the National Institutes of Health grant R01DK066485 (to SLS) and the OHSU Tartar trust foundation ( to CMD). We are grateful to Zhongying Yang for help preparing adenovirus constructs and John Allen for preparing plasmids. We thank Drs. Min Woo Sung, Katarzyna Ciazynska, and Bruce L. Patton for helpful discussions and Dr. Bruce L. Patton for comments on the manuscript.

Figure S1.

Figure S2.

Figure S3.

Figure S4.

Figure S5.

Figure S6.

Figure S7.

Figure S8.

Table S1.

Movie 1. Morph between PIP₂-bound open SUR1/Kir6.2^Q52R channel structure and ATP/Repaglinide-bound closed WT channel structure.