Structural insights into the mechanism of pancreatic KATP channel regulation by nucleotides

ATP-sensitive potassium channels (KATP) are metabolic sensors that convert the intracellular ATP/ADP ratio to the excitability of cells. They are involved in many physiological processes and implicated in several human diseases. Here we present the cryo-EM structures of the pancreatic KATP channel in both the closed state and the pre-open state, resolved in the same sample. We observe the binding of nucleotides at the inhibitory sites of the Kir6.2 channel in the closed but not in the pre-open state. Structural comparisons reveal the mechanism for ATP inhibition and Mg-ADP activation, two fundamental properties of KATP channels. Moreover, the structures also uncover the activation mechanism of diazoxide-type KATP openers.

T he activity of K ATP channels is inhibited by cytosolic ATP and activated by Mg-ADP 1 . The opening of K ATP channels leads to the hyperpolarization of the cell, while the inhibition of K ATP results in depolarization 2 . Therefore, K ATP channels translate the cellular metabolic status into the excitability of the plasma membrane to control the electrical activity of the cell 1 . Because of its unique properties, K ATP channels play essential roles in many key physiological processes, such as hormone secretion 2 , cardiac preconditioning 3 , and vasodilation 4 . The genetic mutations of genes encoding K ATP channels lead to a spectrum of diseases, ranging from metabolic syndrome to cardiovascular diseases and CNS disorders, including neonatal diabetes or even the Developmental delay, Epilepsy, and Neonatal Diabetes" (DEND) syndrome 5 , hyperinsulinaemic hypoglycemia of infancy 5 , dilated cardiomyopathy 6 , familial atrial fibrillation 7 , Cantú syndrome 8,9 and intellectual disability myopathy syndrome 10 . K ATP channels are also important drug targets. K ATP inhibitors promote insulin release to treat diabetes. These drugs include glibenclamide (GBM) and repaglinide (RPG), the socalled insulin secretagogues 11 . K ATP activators (K ATP openers) are used to pharmacologically activate K ATP channels in clinic 12 . Diazoxide is an oral K ATP opener and has been used in the treatment of hypoglycemia and hypertension for nearly half a century 12 .
Functional K ATP channels are hetero-octamers composed of four Kir6 and four SUR subunits 13 . Kir6 are inward rectifier potassium channels that require PI(4,5)P 2 for maximum activity [14][15][16][17][18] . It harbors the nucleotide-binding pocket which can bind the inhibitory ATP, and also ADP to a lesser extent 14,19 . SUR subunits are ABC transporter-like proteins that undergo Mg-nucleotide-dependent conformational changes 20 . The SUR subunits bind to activating Mg-ADP and drugs, including insulin secretagogues and K ATP openers 11,21 . Recent advances in cryo-EM structure determination of K ATP channels in the presence of different ligand combinations by three groups have provided instrumental information about how K ATP channels are assembled from individual subunits, how inhibitory ATP binds the channel, and how chemically distinct insulin secretagogues bind at SUR subunits [22][23][24][25][26][27][28] . Moreover, the conformational changes of the SUR1 subunit upon Mg-nucleotide binding have been visualized 22,26,29 . Despite the progress, the fundamental questions about how K ATP channels work, including the mechanism of ATP inhibition and Mg-ADP activation, remain elusive. In this work, we obtain the structures of the K ATP channel in both the closed state and the pre-open state, revealing the mechanism of K ATP channel regulation by nucleotides.
Based on these observations, we made H175K mutation on the SUR1-Kir6.2 fusion constructs, in which the C terminus of SUR1 is covalently linked to the N terminus of Kir6.2 by a long linker to ensure the correct 4:4 stoichiometry between Kir6.2 and SUR1 26 , yielding the H175K cryo-EM construct. The H175K cryo-EM construct can be inhibited by ATP and activated by Mg-ADP and NN414 ( Supplementary Fig. 1d). These results suggest that H175K cryo-EM recapitulates the basic electrophysiological properties of the wildtype K ATP channel and could be used for structural studies. We purified H175K cryo-EM protein in detergent and supplemented Mg-ADP, PI(4,5)P 2 diC 8 , and NN414 into the protein for cryo-EM sample preparation ( Supplementary Fig. 1e, f).
Single particle cryo-EM analysis showed the H175K cryo-EM protein shows the "propeller" shape ( Fig. 1c-e, and Supplementary Figs. 2, 3), similar to our previous wild-type K ATP protein in a similar condition 26 . The consensus refinement revealed that the peripheral ABC transporter modules of the SUR1 subunit (TMD1-NBD1-TMD2-NBD2) show motions relative to the central K ATP channel core, consisting of Kir6.2 and SUR1-TMD0 domains. We further exploited symmetry expansion, signal subtraction, and local refinement to improve the resolution of the ABC transporter module to 3.1 Å 26 ( Supplementary Fig. 2). The focused 3D classification revealed that the K ATP channel core has obvious conformational heterogeneity at the bundle crossings of the Kir6.2 channel, showing a close to open transition at the gate. Subsequent refinement resolved two 3D classes: one class has a closed and the other has a widened inner helix gate. The resolution of them reached 3.16 Å and 2.87 Å for the Kir6.2 channel after focused refinement, respectively ( Supplementary  Fig. 2). The maps obtained from local refinement were aligned to consensus maps and combined to yield two composite maps for model building and interpretation (Supplementary Figs. 2, 3, and  Supplementary Table 1).
Conformational changes of Kir6.2 TMD during channel opening. The structure of Kir6.2 in the closed state of H175K cryo-EM is similar to our previous ATP + RPG state structure (PDB ID: 6JB1) 27 , with a root-mean-square deviation (RMSD) of 0.7198 Å ( Fig. 2 and Supplementary Fig. 4). Residues on the M2 helix tightly seal the pore at the bundle crossing (Fig. 2a, b). The side chains of L164 and F168 form the gate, where the radius of the narrowest restriction is below 1 Å (Fig. 2a, b, e). In contrast, in the structure with widened gate, the inner part of the M2 helix moves outward (Fig. 2c, d). Particularly, the side chains of L164 and F168 move away from the center, resulting in the dilation of the pore (Fig. 2c, d, f, g). The radius of the ion permeation pathway at the bundle crossing of TMD now increases to 3 Å (Fig. 2e). However, the constriction at the cytosolic G-loop gate still shows a radius of 2.6 Å (Fig. 2e). Although the Kir6.2 channel in this structure is not fully open compared to the CNG channel (PDB ID: 6WEK) 34 and does not allow the passage of fully hydrated potassium ions with estimated radii of 3.3 Å 35 , the channel is clearly in transition to the open state. Therefore, we tentatively assign the current structure with the widened gate as the "pre-open" state.
Associated with the expansion of the pore at the center, there are concomitant movements of the inner part of the Interfacial Helix (IH) and M1 helices (Fig. 2f, g). In the closed state, the side chains of F60 on IH pack against T171 on M2 stabilize the closed pore ( Fig. 2f, g). While in the pre-open state, the side chains of F60 swing away, which allows the expansion of M2 (Fig. 2f, g). In contrast to the obvious structural rearrangements of the inner portion of the pore, the structure of the outer region of the pore, especially at the selectivity filter, has little change (Fig. 2g).
In the structure of Kir3.2 in complex with PI(4,5)P 2 diC 8 31 , the PI(4,5)P 2 diC 8 molecules bind at the subunit interface at the inner leaflet of the membrane. We found lipid-like densities in both closed state and pre-open state of H175K cryo-EM at similar positions ( Supplementary Fig. 3), but the lack of head group feature hindered confident identification of their identities. Therefore, whether PI(4,5)P 2 diC 8 molecules were bound in the H175K cryo-EM structures awaits further investigation.
Conformational changes of Kir6.2 CTD during channel opening. In the closed state, there are ADP densities inside each nucleotide-binding pocket of Kir6.2 CTD (Supplementary Fig. 3). The binding mode of the adenosine group of ADP is similar to that observed previously [25][26][27] . In contrast, there is no ADP density in the pre-open state, and we observed an obvious structural reorganization around the nucleotide-binding pocket of Kir6.2 (Fig. 3). The conformation of R50-R54 on the βA-IH loop has large changes ( Fig. 3a, b, and Supplementary Movie 1). The side chains of Q52 flip from a solvent-exposed conformation to a buried conformation, while the side chains of E51 move in the opposite direction, occupying the nucleotide-binding pocket (Fig. 3b, Supplementary Fig. 4g-h, and Supplementary Movie 1).
During channel opening, there is a 6°anti-clockwise rotation of CTD viewing from the intracellular side, resulting in the clash between E51 and the nucleotide-binding pocket in the pre-open state ( Fig. 3c and Supplementary Fig. 4i). Therefore, the structure of Kir6.2 in the pre-open state is not compatible with the binding of nucleotides anymore, in agreement with the fact that no ADP was observed in the nucleotide-binding pocket of the pre-open state.
Conformational changes of SUR1 TMD0 domain during channel opening. TMD0 domain of SUR1 has a five-helix-bundle structure 23,25 . The N terminal region and TM1 of SUR1 TMD0 interact with the M1 helix of Kir6.2. The extracellular side of TMD0 harbors the docking groove for ECL3 of SUR1 ABC transporter module (335-347) 27 . We observed the outward movements of the TMD0 domain in the inner leaflet and cytosolic region (ICL1, ICL2, and ICL3) during channel opening, while the structure of TMD0 in the outer leaflet largely stays the same (Fig. 4a). These observations suggest the outer half of TMD0 is a structural scaffold that is responsible for tethering Kir6.2 with the ABC transporter module, while the inner half of TMD0 has conformational plasticity for regulatory function. In detail, part of the ICL1 (51-60) of TMD0 is disordered in the closed state, but it is ordered in the pre-open state and forms main-chain hydrogen bonding with βA of Kir6.2 CTD (Fig. 4a, b). In the RPG + ATP bound state structure (PDB ID: 6JB1), K205 on the ATP-binding loop of SUR1 (ABLOS) motif of ICL3 from TMD0 interacts with β and γ phosphates of ATP 27    H175K cryo-EM , we found K205 is away from ADP and does not form interactions with β phosphate of ADP (Fig. 4a, c, d and Supplementary Fig. 4 and Supplementary Movie 1). Moreover, the side-chain densities of residues on the ABLOS motif are not well resolved, indicating their high mobility. By measuring the distances between marker atoms of adjacent subunits, we found coordinated outward movements of Kir6.2 M1 helices (using Cα of Kir6.2 K67 as marker atoms), the ABLOS motif (using Cα of SUR1 K205 as marker atoms), and SUR1 ABC transporter module (using Cα of SUR1 K394 as marker atoms) (Fig. 4c, d).
KCO binds inside the SUR1 subunit in the NBD-dimerized conformation. The ABC transporter module of SUR1 shows an asymmetric NBD-dimerized structure as observed previously 22,26 ( Fig. 5a). We observed that Mg-ADP is bound in the partially closed consensus site, while Mg-ATP is bound in the fully closed degenerate site (Fig. 5b, c). Since we did not supplement additional ATP into the cryo-EM sample, the ATP molecules might be carried through purification. The excellent local map quality allowed us to unambiguously identify the NN414 molecule and to determine its binding pose (Fig. 5d). NN414 binds inside the transmembrane domain of the SUR1 subunit. The dioxide group and its adjacent nitrogen atom of NN414 form polar interactions with H584 on TM11 ( Fig. 5e-g). One NH group on the benzothiadiazine ring and the other NH group between benzothiadiazine ring and methylcyclopropyl group of NN414 form hydrogen bonds with D1031 on TM12 ( Fig. 5e-g). Several hydrophobic interactions further stabilize the binding of NN414. The methylcyclopropyl group of NN414 forms hydrophobic interactions with M1290, Y1287, and T1286 on TM17 (Fig. 5e-g). The central benzothiadiazine ring of NN414 is sandwiched by C1072 of TM13, L1030 of TM12, and I552 of TM10 on one side and V555 of TM10 and L580 of TM11 on the other side ( Fig. 5e-g). Rb + efflux assay showed that H584A or D1031A mutation does not impair K ATP channel activation induced by metabolic inhibitors, indicating the correct folding and trafficking of these mutants to the plasma membrane, but these two mutations abolished the activation by NN414 in Rb + efflux assay (Fig. 5h, i)  activation of K ATP channel by NN414 in the presence of Mg-ATP is also abolished by H584A or D1031A as measured in the insideout patch clamp mode, suggesting their essential role in the activation process of NN414 ( Fig. 5j and Supplementary  Fig. 6a-c).

Discussion
During the preparation of this manuscript, another group reported the K ATP structure composed of SUR1-Kir6. Close-up views of electron densities at the degenerate site. NBD1, NBD2, nucleotides, and Mg 2+ are colored in pink, blue, gray, and green, respectively. c Electron densities at the consensus sites. d NN414 density (orange) in the SUR1 subunit (gray). The map is shown as mesh and the protein is shown as sticks. e, f Close-up views of the NN414-binding site. TMD1 and TMD2 are colored in pink and blue, respectively. NN414 (orange) and residues that interact with NN414 are shown as sticks. g Cartoon representation of the interaction between NN414 and SUR1. The key residues on TMD1 and TMD2 are shown as pink and blue ovals, respectively. h The dose-response activation curves of SUR1-Kir6.2 K ATP channel by NN414 measured by Rb + efflux assay. Curves were fitted to the Hill equation. Data are shown as mean ± SD and WT and D1031A n = 4, H584A n = 3 independent Rb + efflux assays, respectively. i Effects of metabolism inhibitors (MI) on K ATP channel containing various SUR1 mutants. Data are shown as mean ± SD and WT and D1031A n = 4, H584A n = 3 independent Rb + efflux assays corresponding to (h). j K ATP channel activation by 1 µM NN414 in the presence of 0.1 mM Mg-ATP. Data are shown as mean ± SD and n = 3 independent patches. Source data are provided as a Source Data file.  Fig. 7a, b). Moreover, both the C166S + G334D cryo-EM and H175K cryo-EM structures show a "propeller" architecture, which is similar to our previous structures of pancreatic K ATP in the presence of ATP + RPG 27 , ATP + GBM, or Mg-ADP + NN414 26 . The "propeller" architecture is dramatically different from the "quatrefoil" structure of pancreatic K ATP observed in amphipol 22 or the vascular K ATP observed in detergent 36 . The "propeller" architecture allows the communications between the ABC transporter module of SUR and Kir6 through the ICL3 (L0), which might be crucial for the nucleotide regulation of the K ATP channel. In great contrast, the ICL3 is completely disordered in the "quatrefoil" structures. The physiological relevance of the "quatrefoil" structure awaits further investigation. In many K ATP structures 26,27 , we observed two conformations of Kir6.2 CTD, designated as "R" state and "T" state. There is a 12-13°anti-clockwise rotation from the "R" state to the "T" state viewed from the cytosolic side. These structures are similar to the Kir3.2 structure which shows "undocked state" (similar to "R" state) in the absence of PI(4,5)P 2 or docked state (similar to "T" state) in the presence of PI(4,5)P 2 31 . In the current work, we present the structures of H175K cryo-EM at both closed state and pre-open state. The CTD of H175K cryo-EM in the closed state is in a similar position to the "T" state of ATP + RPG structure (PDB ID: 6JB1) ( Supplementary Fig. 4c). While the CTD of the pre-open structure has an additional 6°anti-clockwise rotation compared to the "T" state (Fig. 3c). These structural observations support the idea that K ATP opening is associated with the rotation of Kir6.2 CTD, similar to that proposed for Kir2.2 30 or Kir3.2 channel 31 , indicating a conserved "rotate to open" mechanism for Kir channel gating. The cryo-EM density maps showed that in the same sample, inhibitory nucleotides exclusively bind Kir6.2 in the closed state but not in the open state. Further structural analysis revealed that the conformational changes of Kir6.2 during channel opening, including rotation of CTD and structural rearrangement of the βA-IH loop, disrupt the inhibitory nucleotide-binding pocket of Kir6.2 (Fig. 3a, b). Conversely, the wedging of nucleotide inside the nucleotide-binding pocket of the Kir6.2 channel would block conformational changes that are required for channel opening, providing a plausible mechanism for Kir6.2 channel inhibition by nucleotides. The signals of nucleotide-binding in Kir6.2 CTD are further transmitted to the central ion channel pore via several structural elements, including CTD, βA-IH loop, IH, and M2 gating helix (Fig. 3a). Corroborating with our observations, these structural elements are hotspots for genetic mutation of neonatal diabetes outside the ATP-binding pocket, such as E51, Q52, and G53 on βA-IH linker; V59, F60, and V64 on IH helix; A161, L164, C166, I167 and K170 on M2 ( Supplementary Fig. 5) 5 , suggesting that mutations in these structural elements might allosterically affect ATP inhibition and channel gating.
The binding of Mg-ADP to SUR1 induces the asymmetric dimerization of NBD1 and NBD2, which further drives the closure between TMD1 and TMD2 of the SUR1 ABC transporter module. Our structure shows that diazoxide-type K ATP openers, exemplified by NN414, interact with both TMD1 and TMD2 (Fig. 5) to promote the closure of TMD (Fig. 6). The converged structural changes induced by Mg-ADP and K ATP openers suggest their synergistic effect on K ATP activation. Moreover, by aligning the structures of SUR1 in the presence and absence of NN414 29 , we found NN414 binding induced the enlargement of its binding site surrounded by TM10, TM11, TM12, TM14, and TM17 ( Supplementary Fig. 7c), suggesting an induced-fit mechanism for NN414 binding on SUR1.
Comparing the structure of K ATP in the presence of ATP + RPG (PDB ID: 6JB1) 27 with the structure of H175K cryo-EM in the closed state ( Supplementary Fig. 4a, b and Supplementary Movie 2), we found that their Kir6.2 channels are all in the same nucleotide-bound inhibited conformation ( Supplementary Fig. 4c and Supplementary Movie 2). In contrast, there is a large structural rearrangement of SUR1 due to Mg-ADP and NN414 binding ( Supplementary Fig. 4d-f and Supplementary Movie 2). The conformational changes of the ABC transporter module are transmitted to the lasso motif and finally arrive at TMD0, resulting in outward tilting of the inner half of TMD0 (Supplementary Fig. 4d-f, Supplementary Movie 1, and Supplementary  Movie 2). Notably, the important ATP-coordinating residue K205 on the ABLOS motif of SUR1 27 moves outward and is away from the inhibitory nucleotide bound on the Kir6.2 ( Supplementary  Fig. 4d-f, and Supplementary Movie 1). This would certainly weaken the binding of inhibitory nucleotides and in turn, promote channel activation (Fig. 6a, b and Supplementary Movie 1). During the opening of Kir6.2, there is a further outward tilting of TMD0 and its associated intracellular loops, leaving more space for the expansion of Kir6.2 TMD (Figs. 4 and 6b, c and Supplementary Movie 1). Together with the outward movement of the ABLOS motif on SUR1 ICL3 where E203 locates (Fig. 4), there is a dramatic flipping movement of Q52 on βA-IH of Kir6.2 (Fig. 3). It is reported that Q52E mutation of Kir6.2 paired with E203K of SUR1 mutation greatly enhances the ATP sensitivity of K ATP , and oxidative crosslinking of Q52C (Kir6.2) and E203C (SUR1) mutant could lock the channel in a closed state 37 . Our structural observation suggests that double mutations of Q52C and E203C, on one hand, block the opening conformational change of Kir6.2 directly and, on the other hand, fix the relative distance between the SUR1 ABLOS motif and Kir6.2 to inhibit channel opening.
Kir6 N-terminal peptide (KNtp) plays important role in regulating K ATP function 38 . KNtp not only enhances the ATP sensitivity of Kir6.2 but also mediates the inhibition of insulin secretagogue in the absence of nucleotides, possibly by binding to the central vestibule of SUR and restraining the mobility of Kir6 CTD (Fig. 6a) 36,38 . In the H175K cryo-EM construct, the KNtp is covalently fused to the C-terminus of SUR1 and therefore could not bind inside the SUR1 central vestibule anymore 26,27 . However, the H175K cryo-EM construct could be inhibited by ATP and activated by Mg-ADP, as wild-type K ATP channel ( Supplementary  Fig. 1d). Therefore, our current work uncovers the KNtpindependent nucleotide regulation mechanism of K ATP channels (Fig. 6). Notably, the protein of C166S + G334D cryo-EM has free KNtp, but no density of KNtp is observed in the cryo-EM maps 29 , supporting the hypothesis that KNtp is released from its binding site inside the central cavity of SUR1 and is flexible when the NBDs of SUR1 are dimerized (Fig. 6). Although the ion permeation pathway of the pre-open state does not allow the permeation of fully hydrated potassium ions (Fig. 2e), the hallmarks of K ATP channel activation, such as the enlargement of inner helix gate of Kir6.2, the dissociation of inhibitory nucleotide from Kir6.2, the rotation of Kir6.2 CTD, and the binding of activator NN414 and activatory Mg-nucleotides on NBD-dimerized SUR1, all suggest the current pre-open structure represents the "activated" state of K ATP . It is reported that the open probability of K ATP is less than 100% in the activated condition 39 , suggesting that K ATP can sample the non-conductive activated state, which might be represented by the current pre-open structure.
Construct of H175K cyro-EM . We used cDNA of SUR1 from Mesocricetus auratus (maSUR1) and cDNA of Kir6.2 from Mus musculus (mmKir6.2) for our studies. We made a maSUR1-mmKir6.2 H175K fusion construct (H175K cryo-EM ) which is similar to our previous K ATP fusion construct 26 . There were a 41-residue linker (VDGSGSGSGSAAGSGSGSGSGSGAAGSGSGSGSGSGAAALE) and an 8-residue Prescission Protease cleavage site (LEVLFQGP) between SUR1 and Kir6.2 H175K mutant. The first Met of Kir6.2 was removed to minimize internal translation initiation. This construct contains C-terminal GFP tag and strep tag which were used for protein purification. For electrophysiological experiments, Kir6.2 was cloned into a modified C-terminal GFP-tagged BacMam expression vector and SUR1 into non-tag BacMam expression vector as described previously 26 .
Electrophysiology. K ATP constructs were transfected into FreeStyle 293 F cells using polyethylenimine at the cell density of 0.8-1.1 × 10 6 cells/ml. Cells were cultured in 293TI medium supplemented with FBS for 24-36 h before recording. Macroscopic currents were recorded in inside-out mode at +60 mV via Axonpatch 200B amplifier (Axon Instruments, USA). Patch electrodes were pulled by a horizontal microelectrode puller (P-1000, Sutter Instrument Co, USA) to 2.0-5.0 MΩ resistance. In inside-out mode, both pipette and bath solution were KINT buffer, containing (mM): 140 KCl, 1 EGTA and 10 HEPES (pH 7.4, KOH). For neomycin inhibition, the 50 mM stock neomycin (Sigma) was made in DMSO, stored at −20°C and diluted into KINT buffer to the desired concentration. For NN414 activation, the 50 mM NN414 stock (Sigma) was dissolved in DMSO, stored at −20°C and diluted into KINT buffer to final working concentration. ATP and ADP stocks were prepared on ice, aliquoted and stored at −20°C. ATP and ADP were dissolved in water and adjusted to pH 7 by KOH. The nucleotide concentration was determined by its extinction coefficient and the UV absorption at 259 nm. Recordings were acquired at 5 kHz and low-pass filtered at 1 kHz. Data were further analyzed by pClampfit 10.0 software. The fraction of NN414 activation was calculated as the ratio of the NN414-activated currents over the total currents of K ATP channel.
Rb + efflux assay. AD293 cells were cultured in six-well plate till 90-95% confluence. Wild type mmKir6.2 with C terminal GFP tag was co-transfected with wild type maSUR1 or maSUR1 with H584A mutation into AD293 cells. Cells were continually cultured for 24 h for protein expression and assembly, and GFP signal was detected by microscope in vivo. Then cells were digested by 0.25% trypsin supplemented with EDTA and were equally separated into 96-well plate, finally incubated with 100 μl medium per well. For Rb + efflux determination, 6 mM Rb + was supplemented into the medium for Rb + pre-incubation into the cells. The 96well plate was pre-treated by polylysine for about 24 h in 37°C, and washed out by DMEM medium without FBS before incubation cells. After incubation in 96-well plate for 12-14 h, cells were washed by Ringer's solution (mM): 118 NaCl, 10 HEPES (pH 7.4), 25 NaHCO 3 , 4.7 KCl, 1.2 KH 2 PO 4 , 2.5 CaCl 2 , and 1.2 MgSO 4 and incubated with Ringer's solution supplemented with NN414 in different concentration for 10 min. As a baseline control, plasmids expressing GFP were transfected into cells at the same time. To detect whether K ATP channels were functionally expressed, metabolic inhibitors including 3 mM de-O-glucose and 1 μM oligomycin were used to activate K ATP channels. After drug treatment, the supernatants were transferred for Rb + efflux determination. The cells plated in the well were dissolved by 1% Triton X-100 for 30 min and also transferred for total Rb + quantification. The quantification of Rb + was carried out on Ion Channel Reader 8000 (Aurora Group Company).
Protein expression and purification. K ATP channels were expressed as described previously and the purification process was carried out with minor modification 26 . For protein purification, membrane pellets were homogenized in TBS (20 mM Tris-HCl pH 7.5, 150 mM NaCl) and solubilized in TBS with 1% digitonin (biosynth), supplemented with protease inhibitors (1 mg/ml Leupeptin, 1 mg/ml Pepstatin, 1 mg/ml Aprotinin, and 1 mM PMSF), 1 mM EDTA and 1 mM EGTA for 30 min at 4°C. Unsolubilized materials were removed after centrifugation at 100,000 g for 30 min and the supernatant was loaded onto two 5 mL columns packed with Streptactin 4FF resin (Smart Lifesciences). Strep column was first washed by TBS buffer supplemented with 0.1% digitonin, protease inhibitors (1 mg/ml Leupeptin, 1 mg/ml Pepstatin, 1 mg/ml Aprotinin), 1 mM EDTA and 1 mM EGTA. Then the columns were washed with TBS supplemented with 0.1% digitonin, 3 mM MgCl 2 and 1 mM ATP. The last washing step buffer was TBS supplemented with 0.1% digitonin. The K ATP octamers were eluted by TBS supplemented with 0.1% digitonin and 8 mM desthiobiotin. The eluate was concentrated and loaded onto Superose 6 increase column (GE Healthcare) running with TBS supplemented with 0.1% digitonin. Peak fractions were collected and concentrated to A 280 = 15 (estimated as 15 µM K ATP octamers). The protein purification was completed within 15 h and the purified protein was instantly used for cryo-EM sample preparation.
Cryo-EM sample preparation. K ATP octamers were supplemented with 5 mM MgCl 2 , 0.5 mM ADP, 0.5 mM NN414 and 10 μM PI(4,5)P 2 diC 8 . Acidic PI(4,5) P 2 diC 8 (Echelon Biosciences) was dissolved in water and buffered by Tris-HCl (pH 7.5) before usage. The final protein concentration was estimated to be 13.5 µM octamer. Cryo-EM grids were prepared with Vitrobot Mark IV (FEI) and GIG R1/1 holey carbon grids, which were glow-discharged for 120 s using air before making Cryo-EM sample grids. 2.5 µl K ATP octamers sample was applied to the glowdischarged grid and then the grid was blotted at blotting force in level 2 for 2 s at 100% humidity and 20°C, before plunge-frozen into the liquid ethane.
Cryo-EM data acquisition. Cryo-grids were screened on a Talos Arctica microscope (Thermo Fisher Scientific) operated at 200 kV for small-scale data collection. For grids of high quality, a large data set for K ATP channel structure determination was collected in Titan Krios microscope (Thermo Fisher Scientific) operated at 300 kV.
Images were collected using K2 camera (Gatan) which was mounted post a Quantum energy filter with 20 eV slit, operated under super-resolution mode with a pixel size of 1.324 Å at the object plane, and controlled by Serial EM. Defocus values were set to range from −1.3 μm to −1.8 μm for data collection. The dose rate on detector was 8 e − s −1 A −2 . And the total exposure was 50 e − /A 2 . Each 12 s movie was dose-fractioned into 50 frames.
Image processing. Movies collected were exposure-filtered, gain-corrected, motion-corrected, mag-distortion-corrected and binned with MotionCor2-1.3.2 40 , producing dose-weighted and summed micrographs with pixel size 1.324 Å. CTF models of dose-weighted micrographs were determined using Gctf-1.18 41 . Gautomatch-0.56 (developed by Kai Zhang, MRC-LMB) was used for particles auto-picking and Gautomatch-0.56 templates were produced by projecting K ATP density map generated from small-scale data collected from 200 kV microscope. Data processing was initially executed in Relion_3.1 42 . Particles were extracted from dose-weighted micrographs. After 2 rounds of 2D classification and 3D classification with C4 symmetry, particles of good quality were re-extracted and recentered. The remaining particles were used for 3D refinement and CTF refinement. Upon convergence, the particles were expanded using C4 symmetry and signals for the SUR1 ABC transporter module were subtracted. The subtracted particles were refined using local search within 5°range. The refined SUR1 ABC transporter particles were subjected to no alignment 3D classification, with K = 4 and T = 20. The 3D classes with good map quality were selected and refined in cryoSPARC-3.1.0 by non-uniform refinement 43 , CTF refinement and local nonuniform refinement to reach a resolution of 3.11 Å (map-A). Focused 3D classification on Kir6.2 CTD was carried out with K = 4 and T = 20 without alignment. Two 3D classes with good features but with relative rotations were selected and refined using non-uniform refinement, CTF refinement and local non-uniform refinement in cryoSPARC-3. Model building. The structure of K ATP in complex with ATP and RPG (PDB ID: 6JB1) 27 or Mg-ADP and NN414 (PDB ID: 5YWC) 26 was divided into individual domains and fitted into the cryo-EM maps using UCSF chimera-1.14 44 . The model was further manually rebuilt in Coot-0.8.6 and refined against the maps using Phenix version 1.18-3777 45 . Permeation pathways were calculated using HOLE2 46 .
Quantification and statistical analysis. Global resolution estimations of cryo-EM density maps are based on the 0.143 Fourier Shell Correlation criterion 47 . The local resolution map was calculated using cryoSPARC 48 . Rb + efflux assay curves were fitted to the Hill equation using GraphPad Prism 5. Electrophysiological data reported were analyzed with pclampfit 10.0 software. The number of biological replicates (N) and the relevant statistical parameters for each experiment (such as mean or standard error) are described in figure legends. No statistical methods were used to pre-determine sample sizes.

Data availability
Atomic coordinates and cryo-EM maps are deposited in EMDB and PDB as follows: preopen state: EMD-32310 and PDB 7W4O; closed state: EMD-32311 and PDB 7W4P. Previously published structures: 5YWC, 6JB1, and 6WEK are available from PDB. Source data are provided with this paper. Reagents generated in this study will be made available on request, but we may require payment and/or a completed Materials Transfer Agreement if there is potential for commercial application. Source data are provided with this paper.