Ligand-mediated structural dynamics of a mammalian pancreatic K_ATP channel

Abstract

Regulation of pancreatic K_ATP channels involves orchestrated interactions of their subunits, Kir6.2 and SUR1, and ligands. Previously we reported K_ATP channel cryo-EM structures in the presence and absence of pharmacological inhibitors and ATP, focusing on the mechanisms by which inhibitors act as pharmacological chaperones of K_ATP channels (Martin et al., 2019). Here we analyzed the same cryo-EM datasets with a focus on channel conformational dynamics to elucidate structural correlates pertinent to ligand interactions and channel gating. We found pharmacological inhibitors and ATP enrich a channel conformation in which the Kir6.2 cytoplasmic domain is closely associated with the transmembrane domain, while depleting one where the Kir6.2 cytoplasmic domain is extended away into the cytoplasm. This conformational change remodels a network of intra- and inter-subunit interactions as well as the ATP and PIP₂ binding pockets. The structures resolved key contacts between the distal N-terminus of Kir6.2 and SUR1’s ABC module involving residues implicated in channel function and showed a SUR1 residue, K134, participates in PIP₂ binding. Molecular dynamics simulations revealed two Kir6.2 residues, K39 and R54, that mediate both ATP and PIP₂ binding, suggesting a mechanism for competitive gating by ATP and PIP₂.

Graphical Abstract

Introduction

Pancreatic ATP-sensitive potassium (K_ATP) channels functionally couple glucose metabolism to insulin release and are crucial for glucose homeostasis (Ashcroft, 2005; Nichols, 2006). Structurally, the pancreatic K_ATP channel is an octameric complex composed of two distinct integral membrane proteins (Clement et al., 1997; Inagaki et al., 1997; Lee et al., 2017; Li et al., 2017; Martin et al., 2017b; Shyng and Nichols, 1997). A tetrameric core of Kir6.2 subunits form the central transmembrane pore of the channel. A coronal array of four sulfonylurea receptor 1 (SUR1) subunits surrounds the channel core, each SUR1 companioned with one Kir6.2 subunit. Genetic mutations of these subunits that dysregulate K_ATP channel activity are causes of neonatal diabetes (gain of function) and congenital hyperinsulinism (loss of function) (Ashcroft, 2005). K_ATP channels harbor multiple distinct and antagonistic binding sites for their primary physiological regulators, intracellular ATP and ADP, which close the ion channel through a binding site in Kir6.2, but open the channel through Mg-dependent binding on SUR1 (Nichols et al., 1996; Puljung, 2018). In addition, channel activity is operationally governed by binding sites for specific membrane phospholipids, particularly PIP_2, which directly promote opening as well as antagonize the ATP inhibition at the Kir6.2 binding sites (Nichols et al., 1996). Finally, the pancreatic K_ATP channel is the drug binding target for sulfonylurea and glinide anti-diabetic medications, which inhibit channel activity and thus stimulate insulin secretion (Gribble and Reimann, 2003). The long held principal objective of K_ATP channel research has been to understand the protein dynamics by which these several ligand interactions, separately and in concert, ultimately determine levels of K_ATP channel activity, and hence control insulin release.

CryoEM structures of K_ATP channels have provided direct insights into the structural mechanisms of ligand recognition and gating regulation. In a previous study, we reported comparative cryoEM structures for pancreatic K_ATP channels in the absence of ligands (apo); in the presence of ATP; and in the combined presence of ATP with alternative pharmacological inhibitors: glibenclamide, repaglinide, or carbamazepine (Martin et al., 2019). The study found all pharmacological inhibitors occupy a common binding pocket located within SUR1 and that this binding pocket lies adjacent to the deep binding site for the Kir6.2 N-terminal tail, which courses through the prominent cleft between the two halves of the ABC (ATP Binding Cassette) module of SUR1. The findings offered mechanistic insight into how distinct pharmacological inhibitors inhibit channel activity and also facilitate channel assembly by stabilizing the interaction between Kir6.2 N-terminus and SUR1. However, it was noted during image analysis that each dataset in the study possessed considerable conformational heterogeneity, suggesting classification analyses within datasets may further illuminate channel structural dynamics relevant to ligand binding and gating.

Here, we show results from reprocessing of cyroEM datasets previously reported, focusing on conformational analysis and augmented by molecular dynamics (MD) simulations. Most notably, we found that the cytoplasmic domain (CTD) of Kir6.2 adopted two distinct conformations. In one, the CTD is tethered close to the membrane (Kir6.2-CTD-up). In the other, the CTD is counterclockwise corkscrewed away from the membrane, towards the cytoplasm (Kir6.2-CTD-down). Across structure datasets, the ratio of CTD-up versus CTD-down conformations strongly correlated with the occupation of inhibitory ligand binding sites. Importantly, drug binding and CTD conformation were associated with significant structural reorganization at the ATP and PIP₂ binding sites, and at domain interfaces within and between subunits, suggesting ligands act as molecular glues to stabilize the channel in the Kir6.2-CTD-up conformation. Of further importance, improved cryoEM maps and functional analysis revealed that binding of the activating ligand PIP₂ involves a direct interaction with SUR1 lysine-134 in TMD0, implicating a mechanism by which SUR1 enhances Kir6.2 PIP₂ sensitivity. Moreover, MD simulations uncovered Kir6.2 residues that participate in both ATP and PIP₂ binding, providing an explanation for how ATP and PIP₂ compete to control K_ATP channel gating. Together, our findings provide a framework for understanding how ligands shift channel conformation to regulate channel activity and how mutations, now observed to affect key protein-protein and protein-ligand interfaces, disrupt channel function and cause disease.

Results

K_ATP channel conformation analysis

Focused 3D classification of the Kir6.2 tetramer core plus one SUR1 subunit (denoted K₄S hereinafter) following symmetry expansion and signal subtraction (Scheres, 2016) was performed on our previously published five datasets: apo, ATP only, carbamazepine and ATP (CBZ/ATP), glibenclamide and ATP (GBC/ATP), repaglinide and ATP (RPG/ATP) (Martin et al., 2019) (see Methods). This strategy was employed to circumvent alignment difficulty due to flexible SUR1 (Fig.S1, S2). The analysis revealed two major K₄S conformations: Kir6.2-CTD-up and Kir6.2-CTD-down, wherein the Kir6.2-CTD was alternatively located closer to, or further from, the Kir6.2 membrane spanning channel domains, respectively. Particularly, translation of the CTD between up and down conformations further involved a rotation, together comprising a corkscrew movement wherein the CTD (from an extracellular point of view) was rotated clockwise in Kir6.2-CTD-up conformation relative to Kir6.2-CTD-down (Fig.1). The CTD-up and CTD-down conformations appear qualitatively similar to the T(tense)-state and R(relaxed)-state previously reported by others using a fusion SUR1-Kir6.2 protein under three different ligand conditions, GBC+ATPγS, ATPγS, or MgADP, wherein the T-state exhibits 10.6–12.5° CW rotation viewed from the extracellular side and 3–4.2 Å translation towards the membrane relative to the R-state (Wu et al., 2018). Within both the CTD-up and CTD-down conformations, rocking and rotation of the Kir6.2-CTD was discernable using Relion 3 multibody refinement principal component analysis (Nakane et al., 2018). The heterogeneity was greater in the CTD-down than the CTD-up population of particles (Fig.S3), consistent with an increase in CTD mobility when detached from the membrane. Similar Kir6.2-CTD dynamics were observed using cryoSPARC 3D variability analysis (Punjani and Fleet, 2021).

Figure 1. Two distinct conformations of the Kir6.2-CTD in RPG/ATP bound K_ATP channels.

(A) CryoEM maps for the Kir6.2 tetramer core plus one SUR1 subunit are shown (semi-transparent grey, 1.0 σ contour), with the Kir6.2 subunit including KNtp in the Kir6.2-CTD-up (blue, 0.7 σ contour) and the Kir6.2-CTD-down (magenta, 0.7 σ contour) conformations. Compared to the CTD-up conformation, the Kir6.2-CTD in the CTD-down conformation is translocated from near the lipid bilayer towards the cytoplasm by ~4 Å (distance measured from the center of mass of G295- Cα in the G-loop gate of all four Kir6.2 subunits to the center of mass of F168-Cα in the helix bundle crossing gate of all four Kir6.2 subunits), and rotated counterclockwise by 12° (viewed from the extracellular side), measured by aligning structures onto the TM domain (residues 55–175) of the RPG CTD-up reference model, and calculating dihedral angles between K338-Cα of Chain A of Kir6.2 and the centers of mass of G245-Cα. (B) Fraction of particles in Kir6.2-CTD-up (blue) and Kir6.2-CTD-down (magenta) in K_ATP channels bound to different ligands or in apo state. (C) Variations in the extent of CTD translation and rotation observed for Kir6.2-CTD-up (blue dots) and Kir6.2-CTD-down (magenta dots) are shown for all datasets using RPG/ATP-Kir6.2-CTD-up as reference (circled blue dot). CTD translation away from the membrane is shown in negative value. There is a linear correlation between Kir6.2-CTD translation and Kir6.2-CTD-rotation (R² = 0.9682, y = −0.3031x − 0.1755). Translation and rotation of the CTD from a human open K_ATP structure (Zhao and MacKinnon 2021, PDB: 7S5T) relative to the RPG/ATP Kir6.2-CTD-up reference structure is included for comparison (green dot).

Both the Kir6.2-CTD-up and Kir6.2-CTD-down conformations were observed in the GBC/ATP, RPG/ATP, CBZ/ATP and ATP-only datasets; however, relative abundance of the two conformations varied in the different liganded states (Fig. 1B, Table S1). The GBC/ATP and RPG/ATP datasets had the highest percentages of particles in the CTD-up conformation, with 92.5% and 71.2%, respectively. The CBZ/ATP dataset, which only had CBZ density in SUR1 but no clear ATP density in Kir6.2 likely due to lower concentrations of ATP used during sample preparation (see Methods) had a significantly lower percentage (22%) of particles in the CTD-up conformation, comparable to the ATP only dataset of 17.8%. In the absence of added ligands, i.e. the apo state, only Kir6.2-CTD-down conformation was observed. These findings show Kir6.2-CTD exists largely in two discrete conformations. That the distribution of the two states correlated with the binding of inhibitory ligands implicates the switching between these conformations as a crucial mechanistic event controlling channel opening and closing. The RPG/ATP dataset gave the highest resolution maps for both the Kir6.2-CTD-up and CTD-down conformations (3.4 Å and 3.6 Å, respectively; Fig.S1, S2, Table S1). The improved map quality compared to our previously published structure (Martin et al., 2019) allowed us to reevaluate ligand and protein densities that were previously ambiguous. We have therefore focused on the RPG/ATP dataset for structural analyses hereinafter.

In the RPG/ATP state, the predominant SUR1 conformation is arranged like a propeller when symmetrized, as described in our previous study (Martin et al., 2019). In addition, we identified a minor conformation (~27% of all particles; Fig.S1, Table S1) showing a large clockwise rotation of SUR1 towards the Kir6.2 tetramer (viewed from the extracellular side). This conformation is qualitatively similar to the quatrefoil conformation previously reported in the MgATP/MgADP-bound, NBDs-dimerized Kir6.2-SUR1 fusion channel structure (Lee et al., 2017), and likewise our recently reported quatrefoil-like Kir6.1-SUR2B vascular K_ATP channel structure bound to GBC and ATP with separate NBDs (Sung et al., 2021). The overall map resolution of this minor class is ~7 Å (Fig.S1), which precluded detailed structural analysis. Nonetheless, it reveals that even in the presence of RPG and ATP, a large rotation of SUR1 resembling that seen in NBDs dimerized SUR1 quatrefoil conformation occurs, albeit much less frequently. Heterogeneity of SUR1 within the dominant propeller conformation with more subtle rotations of SUR1’s ABC core around the Kir6.2 tetramer was also observed regardless of whether the Kir6.2-CTD is up or down. We explored the details of SUR1 dynamics using the Kir6.2-CTD-up class of particles, classifying the dynamic motion as discrete eignevectors using Relion 3 multibody refinement (see Methods). Particles with amplitudes between 5 and 20, and between −5 and −20, along eigenvector 1 were refined separately to generate two maps, referred to as SUR1-in and SUR1-out conformations at 3.9 Å and 3.8 Å (Fig.S4), respectively, for model building (Table S2) and structural analysis.

Comparison of different K_ATP conformations

The changes in conformation of Kir6.2-CTD and SUR1 were accompanied by remodeling of subunit and domain interfaces as well as protein-ligand interactions pertinent to gating. Comparing CTD-down to the CTD-up conformation, the Kir6.2-CTD is translated down into the cytoplasm by ~4 Å along an axis perpendicular to the embedding membrane, and simultaneously counterclockwise (CCW) rotated by 12° about that axis viewed from the extracellular side (Fig.1, movie 1). The descended location of Kir6.2-CTD reconfigured the interfacial (IF) helix (also called the slide helix, herein taken to include G53-D65) in the Kir6.2 N-terminus, and also the C-linker (herein taken to include H175-L181) by which inner helix M2 interacts with the CTD (Fig.2). Specifically, in the CTD-up conformation, the IF helix adopted a 3₁₀ helix characteristic (Vieira-Pires and Morais-Cabral, 2010) wherein a directional kink at D58 demarcated the helix into N-terminal and C-terminal halves, with the N-terminal half pivoted towards SUR1 instead of along the adjacent Kir6.2 subunit. In contrast, the IF helix in the CTD-down conformation formed a continuous helix that extended towards the neighboring Kir6.2 subunit. Respecting the C-linker, in the CTD-up conformation, the C-linker formed a helical structure that participated in membrane PIP₂ binding; while in the CTD-down conformation, the C-linker was fully unraveled into a loop, in which a key PIP₂ interacting residue R176 (Baukrowitz et al., 1998; Shyng and Nichols, 1998) was distant from the membrane and incapable of direct PIP₂ interaction. In comparisons between the SUR1 propeller in and out structures (Fig.S1, S4), the ABC module in the SUR1-in structure is rotated clockwise closer to a neighboring Kir6.2 (viewed from the extracellular side). In this rotated position, the SUR1-L0 loop was pulled away from its interaction with SUR1’s direct Kir6.2 subunit partner. Specific molecular changes at the subunit and domain interfaces and ligand binding sites in different conformations and their relevance to gating are described below.

Figure 2. Structural comparison between RPG/ATP CTD-up and CTD-down conformations.

(A) Superposition of the Kir6.2-CTD-up (blue) and CTD-down (magenta) structures. The red boxed region shows significant secondary structural difference at the IF helix and the C-linker. (B) Close-up view with structural differences highlighted (green boxes) at the IF helix and the C-linker in the two conformations. (C) Close-up view of the inter-subunit H-bond interactions between D65 and T293, and an inter-subunit salt-bridge between D58 and R206, as well as an intra-subunit salt-bridge between R177 and D204, and an inter-subunit ATP-binding interaction from K39 in the Kir6.2-CTD-up conformation. (D) Same close-up view as in (C) but of the Kir6.2-CTD-down conformation showing disruption of those interactions seen in (C). Structural elements and residues from chain a or chain b are labeled with a or b at the end.

Intra-Kir6.2 and inter Kir6.2-Kir6.2 interactions--

Inspection of the Kir6.2 CTD-up and CTD-down structures revealed changes in intra- and inter-Kir6.2 interactions involving structural elements which occupy the transitional space between the membrane spanning and cytoplasmic domains of Kir6.2. These include the IF helix, the C-linker, the DE loop (the loop that connects βD and βE), the G-loop (the lower cytoplasmic gate) as well as the N-terminus and ATP. Specifically, in the CTD-up conformation, D204 at the start of the DE loop forms an intrasubunit salt bridge with R177 in the C-linker, which connects to Kir6.2 TM helix 2; and R206, also in the DE loop forms a salt bridge with D58 in the IF helix of the adjacent Kir6.2 subunit (Fig.2C). Two hydrogen bonds: one between D65 in the IF helix and T293 in the G-loop from adjacent subunits, and one between K39 at the N-terminus and ATP at a neighboring Kir6.2 subunit were also observed (Fig.2C). The salt bridges and hydrogen bonds bind the different structural elements together and hold the CTD in the up conformation. In contrast, the aforementioned salt bridges and hydrogen bonds were eliminated in the CTD-down structures (Fig.2D). In particular, separation of D204 from R177 in CTD-down was accompanied by uncoiling and extension at the end of the C-linker helix including R177, while separation of R206 from D58 was accompanied by a significant straightening of the kink in the IF helix and reorientation of the continuing chain, which connects to the base of the KNtp (Fig.2).

Consistent with the notion that these labile salt bridges have critical roles in channel gating, previously published functional studies have implicated the participating Kir6.2 residues in channel regulation. D58 was previously suggested to be involved in anchoring Kir6.2 CTD to the TM domain through results of targeted mutagenesis (Li et al., 2013). Our findings resolve the salt bridge partnership with R206 in the neighboring Kir6.2 polypeptide, and further reveal such tethering is dynamically incorporated into the conformational changes of ion channel activation. Consolidating this view, R206 has been separately implicated in channel activation by PIP₂, through scanning mutagenesis investigations of positive residues involved in effecting bound PIP₂, wherein mutation R206A was found to abolish PIP₂ response and thus diminish channel activity (Li et al., 2013; Lin et al., 2005; Shyng et al., 2000). Similar to R206A, mutation R177A also abolishes or greatly attenuates channel activity by diminishing PIP₂ response (Li et al., 2013; Shyng et al., 2000). Thus, in corroborating earlier functional studies, our structural findings here elucidate key molecular interactions that place the Kir6.2-CTD close to the membrane in position to interact with membrane-bound phospholipids for channel opening. These insights are directly relevant to human health. Mutations of each of the four residues in the above salt bridges have been identified in congenital hyperinsulinism, a disease caused by loss of function of K_ATP channels. These include D58V (De Franco et al., 2020), R177W (Arya et al., 2014), D204E (Pinney et al., 2008), and R206H (Boodhansingh et al., 2019). Our structures here provide a mechanistic illustration of how perturbation of residues involved in conformational dynamics cause loss of channel function and hyperinsulinism.

Kir6.2 and SUR1 interactions--

Our previous study of pancreatic K_ATP channel structure suggested that a key regulatory interface through which SUR1 controls Kir6.2 channel activity is formed by the extended N-terminus of Kir6.2 (residues 1–30; referred to as KNtp) that is inserted within its SUR1 subunit partner, wherein the KNtp is located within the SUR1 ABC transporter module (Martin et al., 2019). We showed in particular that the KNtp is located between the two transmembrane helix bundles (TMBs) of SUR1 ABC module, and adjacent to the drug binding pocket of GBC, RPG, and CBZ. More detailed structural analysis was hindered by insufficient resolution for the density of KNtp in those published cryoEM maps. The additional analysis methods applied in our current study yielded clearer, contiguous densities and produced significantly improved maps revealing specific interactions between residues in KNtp and SUR1 (Fig.3). In the distal part of KNtp, which lies deep in the SUR1 ABC core cavity, Kir6.2-R4 is in bonding position with SUR1-T1139 and N1301. In the middle section of KNtp, which is near the entrance to the SUR1 ABC core cavity, Kir6.2-L17 interacts with R826 of NBD1 and G1119 and N1123 of TMB2. In the proximal end of KNtp, cryoEM density corresponding to residues P24, Y26 and R27 comes into close contact with cryoEM density correponding to SUR1’s NBD1-TMD2 linker around residue S988. In the Kir6.2-CTD down structure, interactions at the distal and middle segments of KNtp with SUR1 remain largely unchanged; however, the proximal section of KNtp is significantly further away from the NBD1-TMD2 of SUR1 (Fig.3D). In a recent Kir6.1-SUR2B cryoEM structure we showed that the NBD1-TMD2 linker has a role in regulating MgADP-dependent interactions between SUR2B-NBD2 and Kir6.1-CTD (Sung et al., 2021). Our structures presented here reveal an additional contact between NBD1-TMD2 linker and KNtp. Whether this contact and changes at this interface in different conformations have functional significance warrants future investigation. As reported previously (Martin et al., 2019), the cryoEM density of KNtp is the strongest in the GBC/ATP, RPG/ATP, and CBZ/ATP datasets, followed by ATP only, and is the weakest in the apo state, indicating inhibitory ligands stabilize the KNtp-SUR1 interface. Deletion of KNtp is known to increase channel open probability (Babenko et al., 1999; Koster et al., 1999; Reimann et al., 1999), while immobilizing KNtp in the SUR1 ABC core cavity via engineered crosslinking between Kir6.2-L2C and SUR1-C1142 inhibits channel activity (Martin et al., 2019). Worth noting, mutations L2P (Alkorta-Aranburu et al., 2014), R4C/H, L17P, R24C, R27C/H in Kir6.2 (De Franco et al., 2020) as well as R826W (de Wet et al., 2008) and N1123D (Suzuki et al., 2007) of SUR1 have all been reported in neonatal diabetes or congenital hyperinsulinism, which underscores the importance of the KNtp-SUR1 interface in channel gating.

Figure 3. KNtp and SUR1 interface.

(A) KNtp (Kir6.2 aa 1–30) cryoEM density (purple mesh, 1.0 σ contour) in RPG-Kir6.2-CTD-up structure, with key residues that interact with SUR1 shown as orange spheres within red circles. The distal portion of KNtp is located deep in the cavity of the SUR1-ABC core module. The middle section of KNtp lies near the entrance of the cavity. The proximal part (i.e. C-terminal part) of KNtp including P24, Y26 and R27 contacts the SUR1 N1-T2 linker density (shown as green mesh; 1.0 σ contour) near S988. (B) Left: Close view of distal KNtp cryoEM density (purple mesh, 1.0 σ contour) showing interaction of R4 with SUR1 T1139 and N1301 in TMD2 and proximity to bound RPG. Right: Close-up view of middle KNtp in cryoEM density (purple mesh, 1.0 σ contour) showing interaction of L17 with SUR1 R826, G1119 and N1123. (C) KNtp viewed without SUR1, showing its interconnectivity with two ATP binding sites and the LM-loop in the CTD of a neighboring Kir6.2. (D) Superposition of Kir6.2-CTD-up (blue) and CTD-down (pink) structures viewed from the extracellular side showing divergent positions of proximal KNtp.

In addition to KNtp forming interactions with the SUR1-ABC module, regions C-terminal to KNtp in the Kir6.2-N terminal domain were also observed to be intimately involved in protein-protein and protein-ligand interactions. First, in the CTD-up structure, Kir6.2 R31-R34 is close to the short loop that connects βL and βM (LM loop) (Martin et al., 2017b) of the neighboring subunit (Fig.3C). A mutation D323K in the LM loop has been shown to disrupt ATP inhibition (Brennan et al., 2020). Second, further downstream K39 has its sidechain oriented towards ATP bound to the neighboring Kir6.2 on the other side. Thus, we found the Kir6.2 N-terminus is connected simultaneously to two ATP binding pockets. Dynamic movement of the KNtp between the CTD-up and CTD-down conformations may therefore impact the interactions of downstream Kir6.2-N terminal domain with neighboring subunits on both sides and with ATP. Finally, the refined structures showed that a loop (K47-Q52) N-terminal to the IF helix of Kir6.2 has close interaction with SUR1’s TMD0-intracellular loop 1 (ICL1), ICL2 and ICL3 (i.e. L0). Here, a compact network of interactions stabilizes the Kir6.2-CTD close to the membrane and also reinforces ATP binding. In the CTD-down conformation, the Kir6.2 pre-IF helix loop becomes more distant from the SUR1-ICLs such that the Kir6.2-CTD is no longer tethered close to the membrane, which also impacts the ATP binding pocket (see below).

ATP binding pocket--

Rapid and reversible closure upon non-hydrolytic binding of ATP at Kir6.2 is a cardinal feature of K_ATP channels (Nichols et al., 1996). The improved map quality in the current study allowed us to refine interpretation of the ATP cryoEM density and the interaction network that coordinates ATP binding and follow how the ATP binding pocket becomes reconfigured in different conformations.

In our improved maps, the ATP density could be modeled with ATP in two alternative poses. In the first, the γ-phosphate is oriented upward towards R50 of Kir6.2, which is consistent with functional studies indicating that R50 interacts with the γ-phosphate of ATP (Trapp et al., 2003). This pose was used to model ATP density in our previously published structure bound to GBC and ATP (PDB: 6BAA) (Martin et al., 2017a) and also to model ATPγS bound to a rodent SUR1–39aa-Kir6.2 fusion K_ATP channel (Wu et al., 2018). In the second pose, the ATP’s γ-phosphate is oriented downward facing N335. This alternative orientation is also supported by functional data showing that N335Q decreases ATP sensitivity (Drain et al., 1998) and used to model ATP density bound to Kir6.2 in cryoEM structures of a human SUR1–6aa-Kir6.2 fusion K_ATP channel (PDB: 6C3O and 6C3P) (Lee et al., 2017). Of note, we also observed an unassigned protruding density in ATP in our initial GBC/ATP map (EMD-7073), which we speculated may be contaminating Mg²⁺ (Martin et al., 2017a) but which can be well modeled by the alternative pose of the γ-phosphate. The simplest interpretation is that the cryoEM density of ATP we observed is likely an ensemble of the two possible γ-phosphate poses.

The improved map also showed clear cryoEM density for the side chain of K205 in the L0 of SUR1. We have previously proposed that K205 participates in ATP binding (Martin et al., 2017b) based on an early finding that K205E reduces ATP inhibition (Pratt et al., 2012). However, our previously published cryoEM map (EMD-7073) does not resolve the side chain density of K205 sufficiently to permit definitive conclusion. In our current map, K205 side chain was clearly oriented to the bound ATP (Fig.4B), stabilizing interactions with the β- and γ-phosphates of ATP. Similar observations have been reported by Ding et al. (Ding et al., 2019). The role of K205 in ATP binding is further substantiated by Usher et al. (Usher et al., 2020) in which binding affinity between a fluorescent ATP analogue and the channel was assessed by FRET measurements between the ATP analogue and a fluorescent unnatural amino acid ANAP (3-(6-acetylnaphthalen-2-ylamino)–2-aminopropanoic acid) engineered at Kir6.2 amino acid position 311. The study found that SUR1-K205A and K205E mutations reduce ATP binding affinity by ~5 and 10-fold.

Figure 4. Comparison of ATP binding pocket and SUR1-TMD0/Kir6.2-CTD interface between Kir6.2-CTD-up and CTD-down conformations.

(A) Surface representations of the SUR1-TMD0/Kir6.2-CTD interface and the ATP binding pocket in Kir6.2-CTD-up (left) and the CTD-down (right) conformations. SUR1 is shown in deep blue or pink hues, Kir6.2-CTD in pale blue or pink hues, and ATP as stick model. The arrows point to contacts between SUR1 and Kir6.2 in CTD-up panel which are lost (dashed arrows) in the CTD-down panel. Loss of the tight interaction between the SUR1-TMD0 and Kir6.2-CTD domains renders the ATP binding pocket less compact. (B) Detailed views of residues surrounding ATP in the Kir6.2 CTD-up conformation (blue, left) which become more distant from ATP and/or interacting partners in the CTD-down conformation (pink, right), including K205 of SUR1, Q52, R54 and K39 of Kir6.2. ATP cryoEM density (3.3 σ contour) is represented by a mesh.

The conformational dynamics we observed in Kir6.2-CTD and SUR1 had significant impact on the structure of the ATP binding site. In the Kir6.2-CTD-up conformation, the Kir6.2-CTD was packed tightly against SUR1’s ICL1, ICL2, and the initial segment of L0 (Fig.4A; contact surface area ~144 Ų, calculated using PDBePISA). In this conformation, ATP was fitted snugly into the pocket formed by the N- and C-terminal domains of adjacent Kir6.2 subunits and L0 of SUR1, and had an ATP interface area ~430 Ų. In the CTD-down conformation, the Kir6.2-CTD became disengaged from the SUR1-ICLs (Fig.4A; contact surface area ~9 Ų), which disrupted several interactions that had stabilized ATP binding, and reduced the ATP interface area to ~380 Ų. Specifically, R54, which was oriented towards the ATP in the CTD-up state became distant from the ATP binding pocket in the CTD-down structure. K39 in the N-terminus of neighboring Kir6.2, which also coordinated ATP binding in the CTD-up structure, was reoriented away from the ATP in the CTD-down structure (Fig.2C, D). Moreover, the distance between K205 in the L0 of SUR1 and ATP increased in the CTD-down conformation. These changes together weakened the ATP binding pocket and exposed ATP to solvent. In addition to impacting Kir6.2-CTD dynamics, SUR1 rotation also affected ATP binding. When SUR1-ABC module rotated toward the Kir6.2 tetramer core (SUR1-in), L0 pulled away from the ATP binding pocket. As a result, SUR1-K205 lost interaction with ATP, destabilizing binding (Fig.4B).

PIP₂ binding site--

At the binding pocket where PIP₂ is predicted to bind based on homology with Kir2 and Kir3 channels for which PIP₂ bound structures are available (Hansen et al., 2011; Lee et al., 2016b; Niu et al., 2021), a lipid cryoEM density is seen in both Kir6.2-CTD-up and CTD-down conformations. Interestingly, the lipid density in the CTD-up conformation is significantly larger than that in the CTD-down conformation (Fig.5A). This was consistent for all datasets that include both conformations. We were able to fit, and tentatively model, the lipid density in the CTD-up structure with two phosphatidylserine (PS) molecules and that in the CTD-down structure with one PS molecule (Fig.S5). Since no PIP₂ was added to our samples prior to imaging, we reasoned that the more abundant PS may have entered the binding pocket. In a recent study by Zhao and MacKinnon (Zhao and MacKinnon, 2021), it was shown that PIP₂ is not required for K_ATP channel activity, suggesting other phospholipids that occupy the PIP₂ binding site could potentially support channel activity. Whether the density in our structure represents PS, co-purified endogenous PIP_2, or other phospholipids requires further investigation.

Figure 5. Comparison of the PIP₂ binding pocket in Kir6.2-CTD-up and CTD-down conformations.

(A, B) PIP₂ binding pocket of Kir6.2-CTD up (A, blue) and Kir6.2-CTD down (B, pink) conformations viewed from the side. Lipid densities are shown as a mesh (1.0 σ contour). In (A), in addition to Kir6.2 residues previously implicated in phospholipid binding, SUR1-K134 side chain is pointed directly at the lipid density. SUR1 is shown in deep blue or pink, Kir6.2 in pale hues. (C, D) Same as (A, B) viewed from the extracellular side with the helical bundle crossing shown (F168). Note the PIP₂ binding pocket is more compressed in Kir6.2-CTD-down than CTD-up conformation due to secondary structural change at the IF helix that brings Q57 to interact closely with F60 and W68 (red dashed circle). (E, F) Inside-out patch-clamp recording (examples in E) show greater fold current increase in response to PIP₂ of the SUR1-K134A mutant channel than WT channel (left), with statistically significant difference (*p<0.05, student’s t-test).

In the Kir6.2-CTD-up structure, in addition to the set of Kir6.2 residues K67 and W68 in the outer helix, and R176 in the C-linker previously implicated in PIP₂ binding (Brundl et al., 2021; Cukras et al., 2002; Shyng and Nichols, 1998), we found K134 in TMD0 of SUR1 was in close contact with the density corresponding to lipid headgroups (Fig.5A). To test whether SUR1-K134 has a role in the PIP₂ sensitivity of K_ATP channel opening, we functionally characterized a mutant K_ATP channel in which this residue is mutated to alanine (SUR1-K134A), using inside-out patch-clamp recording (see Methods). Compared to WT channels, the SUR1-K134A mutants exhibited substantially smaller initial currents in ATP-free solution. Upon PIP₂ addition, however, the mutant channel currents increased by 5.93±2.17-fold, which is significantly higher than the 1.27±0.23-fold current increase seen in WT channels (Fig.5E, F), indicating the SUR1-K134A mutation reduced intrinsic P_o and PIP₂ interactions. It is well documented that channel P_o, determined by channel interaction with PIP₂, is primarily conferred by SUR1 association with Kir6.2. The Kir6.2 channel itself has low P_o, but co-expression with SUR1 or just the TMD0 domain of SUR1 increases channel P_o by more than 10-fold (Babenko and Bryan, 2003; Chan et al., 2003; Enkvetchakul et al., 2000; Pratt et al., 2011). Our results show SUR1-TMD0 participates in PIP₂ interaction, at least in part through K134, which strengthens PIP₂ interactions with Kir6.2.

In the CTD-down structure, the IF helix was closer to W68 near the cytoplasmic end of the outer helix of the neighboring Kir6.2 than in the CTD-up structure, causing compression of the PIP₂ binding pocket (Fig.5A–D). This provides an explanation for why the lipid cryoEM density in the CTD-down structure was significantly weaker than that in the CTD-up structure and could be tentatively fit by only one PS molecule (Fig.S5). Moreover, the simultaneous unwinding of the C-linker in the CTD-down structure withdrew the key PIP₂-interacting residue R176 to a position too distant for interaction. In this conformation, Kir6.2 is expected to be inactive.

Elucidating the relationship between ATP and PIP₂ binding by MD simulations

ATP and PIP₂ compete with each other to close and open the channel, respectively (Baukrowitz et al., 1998; Shyng and Nichols, 1998). However, the structural mechanism underlying this functional competition has remained unresolved. Previous mutation-function correlation studies led to a proposal that ATP and PIP₂ have overlapping but non-identical binding residues (Cukras et al., 2002; Shyng et al., 2000; Tucker et al., 1998). To test this hypothesis, we conducted MD simulation studies using as a starting point the Kir6.2 (32–352) plus SUR1-TMD0 (1–193) tetramer part the RPG/ATP Kir6.2-CTD-up structure. Previous studies have shown that Kir6.2 and TMD0 of SUR1 form “mini K_ATP channels” (Babenko and Bryan, 2003; Chan et al., 2003), which like WT channels exhibit functional antagonism between ATP and PIP₂ (Babenko and Bryan, 2003; Pratt et al., 2011). The mini K_ATP channel system is therefore suitable for simulating residues which may participate in binding of both ligands. Three independent 1 μs simulations were carried out for each of two ligand conditions, either without ATP or PIP₂ (apo), or with both ATP and PIP₂ in their respective binding pockets (ATP+PIP₂) (Fig.S6A, B; for details see Methods).

Comparing the two different conditions, there was an overall increase in dynamics of the Kir6.2-CTD in the apo simulations versus the ATP+PIP₂ simulations (Fig.6, Fig.S6, movie 2–5). First, significant secondary structural changes at the IF helix and the C-linker were observed in the apo simulations, resembling the changes from the CTD-up to the CTD-down conformation we observed in cryoEM structures. Second, the entire CTD relaxed towards the cytoplasm in the apo simulations. This was quantified by measuring the distance between the helix bundle crossing (HBC) gate at F168 and the G-loop gate at G295 (Fig.6A). In apo simulations, this distance increased over time in all three runs, whereas it remained relatively unchanged for the ATP+PIP₂ simulations (Fig. 6B), except in run 2 during which the distance increased when ATP became partially dissociated at around 500 ns (Fig.6B). These findings show that in the absence of ligands, the Kir6.2-CTD has a tendency to relax toward the CTD-down conformation.

Figure 6. MD simulations of Apo versus ATP+PIP₂ state.

(A) Positioning of the helix bundle crossing (HBC) and G-loop gate in a representative apo-simulation and ATP-PIP₂-simulation. The time-varying position of the geometric center (plotted in red) of G295 (G-loop gate) Cα carbons of all four chains is overlaid on beginning and end snapshots of G295 Cα carbons, after aligning trajectories to the Kir6.2 TM domain. (B) Plots of the distance between the geometric centers of all four Cα atoms of F168 and of G295 for the entire simulations. Greater distances between the two gates in the apo state compared to the ATP+PIP₂ state indicate relaxation of the CTD in the absence of ligands. In one apo simulation (light blue), ATP became partially dissociated at around 500 ns (red arrow). (C) Snapshots of simulations in the presence of ATP and PIP₂ showing interactions of R54 and K39 with either ATP or PIP₂. Kir6.2-K185 and SUR1-K134 which only interact with ATP or PIP₂ respectively are also shown. (D) Heatmaps showing fraction of time residues spend at increasing distances from ATP (horizontal axis) or PIP₂ (vertical axis). A representative trajectory corresponding to chain 1 in run 1 and colored to show time evolution is shown for each residue. K39 and R54 (top row) exhibit switching behavior and occasional simultaneous contact, while essentially exclusive ATP binding residues are shown in the middle row, and PIP₂ binding residues in the bottom row.

Analysis of the minimum distance between each of the ligands and their surrounding residues within 4 Å over the entire simulation revealed that K39 and R54 of Kir6.2 engaged in both ATP and PIP₂ binding. Fig.6D shows the fraction of time over the entire simulation each residue in each subunit and each run came into contact with ATP or PIP₂. R54 and K39 each showed partial ATP and PIP₂ occupancy (Fig.6C,D, Fig.S6C,D, movie 4, 5), which was in contrast to well established ATP binding residues, including R50 and K185, and PIP₂ binding residues K67 and R176, which showed nearly 100% ATP or PIP₂ occupancy. Of the two dual occupancy residues, R54 showed greater interactions compared to K39 with both ATP and especially with PIP₂. K39, while showing interaction with ATP in all three runs, only showed significant interaction with PIP₂ in one of the three runs (Fig.S6D). The analysis also identified residues that had specific, although less stable, interactions with either ATP or PIP₂ as defined by distance between residue and ligand < 4 Å. In particular, Kir6.2-Q52 specifically interacted with ATP, and SUR1-K134 with PIP₂, contrasting with the dual ligand binding mode of R54 and K39. A previous mutagenesis study has shown that mutation of either K39 or R54 to alanine reduces channel open probability as well as sensitivity to ATP inhibition (Cukras et al., 2002), which early implicated a role for these residues in channel gating by PIP₂ and ATP. Confirming a role in physiological regulation, mutations R54C and R54H are linked to congenital hyperinsulinism (De Franco et al., 2020) and mutation K39R to transient neonatal diabetes (Zhang et al., 2015). Our MD simulation results suggest both residues participate directly in ATP and PIP₂ binding, providing mechanistic insight into how mutation of these residues affect PIP₂ and ATP sensitivities and cause disease.

Discussion

Cryo-preserved purified protein samples may contain multiple protein structures that represent distinct functional or transitional states and can provide mechanistic insight (Nogales and Scheres, 2015). In this study, analysis of five K_ATP channel cyroEM datasets collected in different ligand conditions revealed conformational heterogeneity of the Kir6.2-CTD and SUR1 ABC module. We observed the Kir6.2-CTD in either an “up” position tethered close to the plasma membrane, or a “down” position corkscrewed away from the membrane towards the cytoplasm. The ratio of the two conformations correlated with occupancy of inhibitory ligands at the SUR1 and Kir6.2 binding sites (Fig.1), suggesting inhibitory ligands help stabilize the Kir6.2-CTD close to the membrane. Furthermore, in both Kir6.2-CTD conformations the SUR1 ABC module was observed oriented with a range of rotation around the Kir6.2 tetramer central axis (Fig.S4). We observed a restructuring of protein-protein and protein-ligand interfaces in different conformations that sheds light on how ligands shift channel conformational dynamics to regulate gating.

Correlation between Kir6.2-CTD conformation and channel function

The structures analyzed in this study all represent closed channels. Recently, an open human K_ATP channel structure containing Kir6.2 C166S and G334D mutations was reported (Zhao and MacKinnon, 2021), which showed a Kir6.2-CTD that is further CW rotated (extracellular view) and slightly upward translated compared to our Kir6.2-CTD-up conformation (Fig.1C). Rearrangement of the molecular interactions between the IF helix, the C-linker, and TM residues as well as increased distance between the pre-IF helix loop and SUR1 L0 compared to the ATP-bound closed WT channel structure (equivalent to our Kir6.2-CTD-up structure) were observed (Fig.7). The restructuring widens the ATP-binding pocket, explaining the absence of ATP cryoEM density despite high concentrations of ATP in the sample, and stabilizes HBC gate opening via side chain interactions between F60 in the IF helix and the HBC gate residue F168. Another pre-open K_ATP channel structure using a rat SUR1–39aa-Kir6.2 H175K fusion construct published while this manuscript was under review (Wang et al., 2022) showed similar structural characteristics as the open channel structure. Taken together, a picture emerges wherein rotational and translational position of the Kir6.2-CTD determines the functional state of the channel. When the CTD is in the down position, the phospholipid binding pocket is compressed due to a change in secondary structures of the IF helix (Fig.5) and the C-linker is unwound by the increased separation of the CTD from the membrane domain. We propose this conformation corresponds to an “inactivated” state in which the CTD is unable to engage with membrane phospholipids and thus open the channel. When the CTD is in the clockwise up-screwed position with ATP and/or drug bound, the channel is primed for opening but is arrested in an inhibited state due to an interaction network between SUR1-L0 and the Kir6.2-N terminal domain, an interaction stabilized by ATP that prevents further rotation of the CTD needed to open the HBC gate. Upon ATP dissociation, the CTD is released for further CW rotation, enabling the gate to widen and open the channel.

Figure 7. Comparison of an inhibitor-bound CTD-up closed structure and a mutant open structure.

(A) Overlay of RPG+ATP CTD-up structure (blue) with a Kir6.2 double mutant (C166S, G334D) structure (PDB:7S61; yellow). Only two Kir6.2 subunits are colored. A small upward translation of the mutant open structure relative to the inhibitor-bound closed structure is indicated by the small red arrow next to the red box, which is shown in a 90° rotated enlarged view in B. (B) Overlay of the Kir6.2 tetramer with part of the SUR1-L0 viewed from the extracellular side. A CW rotation (5°) of the CTD from the inhibitor-bound closed structure to the mutant open structure is noted. Structural elements and residues from chain b are labeled with b at the end to distinguish from those from chain a. Residues which show significant differences in the two structures are labeled in red. (C) Same as B but with the two structures shown separately and residues in the C-linker.b visible. The red dashed circles illustrate the enlargement of the potassium ion path at the helix bundle crossing (F168) in the open structure.

To explain our structural data and a wealth of electrophysiological data we propose the Kir6.2-CTD undergoes transitions between four principal conformation states in dynamic equilibrium: CTD-down inactivated, CTD-up unliganded and closed, CTD-up bound to inhibitory ligands, and CTD-up open, with the probability to occupy a given conformation driven by ligands (Fig.8), similar to the model previously proposed by Borschel et al. (Borschel et al., 2017). In the absence of ligands the CTD-down conformation dominates, as seen in our apo state dataset, and the channel is inactivated. While not observed in our structural studies here, inactivated channels can transition into a short-lived unliganded CTD-up conformation at low probability. Binding of physiological inhibitor ATP at Kir6.2, and/or a pharmacological inhibitor at SUR1, shifts Kir6.2 towards a stable CTD-up but closed conformation. Binding of phospholipids such as PIP₂, when coupled with unbinding of inhibitory ligands, shifts the equilibrium towards the Kir6.2-CTD-up open position. Under physiological conditions with high intracellular ATP concentrations and ambient phospholipids, we expect the Kir6.2-CTD to be mostly in an ATP inhibited CTD-up conformation, with a small fraction in a CTD-up state having the phospholipids bound and a further-rotated-open conformation, and rarely in an unliganded CTD-up conformation; the CTD-down inactivated conformation would also be rare. However, in pathological conditions the CTD-down inactivated state could be prevalent. We and others have previously reported several mutations including congenital hyperinsulinism-causing mutations at the Kir6.2 subunit-subunit interface (such as R192A, E229A, R314A, R301A/C/H) that promote channel inactivation (Borschel et al., 2017; Lin et al., 2008; Lin et al., 2003; Shyng et al., 2000). Channels containing such mutations briefly open upon patch excision into ATP-free solution but then quickly inactivate. Interestingly, inactivation can be overcome by transiently exposing channels to high concentrations of ATP, followed by washout of ATP. We propose that these mutations increase an energy barrier for Kir6.2-CTD to transition from the CTD-down conformation to CTD-up conformation, thus trapping the CTD in the down inactivated state. ATP thus effectively acts as a molecular glue at its Kir6.2 domain interfaces. Exposure to ATP shifts the Kir6.2-CTD back to the CTD-up position such that channels can open again when ATP is washed out. The model similarly explains the ability of PIP₂ to prevent and reverse inactivation mutants from inactivation (Borschel et al., 2017; Lin et al., 2008; Lin et al., 2003; Shyng et al., 2000) by stabilizing Kir6.2-CTD in the up and further rotated open position.

Figure 8. Correlating Kir6.2-CTD structures with functional states of pancreatic K_ATP channels.

Cartoon representation of the structural and functional states of K_ATP channels. In the presence of high concentrations of intracellular ATP and ambient PIP₂, channels are mostly in ATP-bound closed state in which the Kir6.2-CTD is in the up-conformation and docked near the membrane and rotated CW from an extracellular perspective. Removal of ATP results in a transient unliganded closed state with the CTD in the up position conducive to binding PIP₂. Binding of PIP₂ opens the channel in which the Kir6.2-CTD is further CW rotated and moved up towards the membrane. Inactivation occurs when the CTD in the unliganded state transitions into the down conformation, a process that is enhanced by inactivation mutations including those known to cause hyperinsulinism. ATP facilitates channel recovery from inactivation by shifting the equilibrium towards the ATP-bound closed state in which the Kir6.2-CTD is in the up-conformation to allow the channel to transition to the unliganded CTD-up closed state upon subsequent removal of ATP, primed for PIP₂ binding and channel opening. Addition of PIP₂ also counters inactivation by shifting the equilibrium via the unliganded CTD-up closed state towards PIP₂ bound open state. Note the unliganded, CTD-up closed state shown to account for functional and kinetic modeling data in the literature is likely short-lived and its structure is yet to be captured by cryoEM.

The Kir6.2-CTD-up and CTD-down conformations observed in our structures are similar to the T and R states observed in a SUR1–39aa-Kir6.2 fusion channel alternatively bound to GBC+ ATPγS, ATPγS alone, or MgADP by Wu et al. (Wu et al., 2018). However, the percentage of particles in the T-state (corresponding to our Kir6.2-CTD-up state) in their ATPγS+GBC or ATPγS datasets is ~40% and 43% respectively, which differ significantly from the ~93% and 22% in our GBC+ATP and ATP alone datasets. Compared to our CTD-up state in the ATP condition, the higher percentage of T-state particles in the fusion channel’s ATPγS condition could potentially derive from the 10-fold higher concentrations of ATPγS used to generate the Wu et al. structure. An explanation for the markedly higher percentage of Kir6.2-CTD-up state particles in our GBC+ATP condition, over T-state particles in their GBC+ATPγS condition is not obvious. In principle, however, the extra 39aa linker between SUR1 C-terminus and Kir6.2 N-terminus in the fusion construct could uncouple drug binding from Kir6.2-CTD conformation. Consistent with this, GBC was shown to be ineffective in inhibiting the fusion channel in contrast to the WT channel formed by separate SUR1 and Kir6.2 proteins (Wu et al., 2018).

Previous K_ATP channel cryoEM studies have provided evidence that KNtp interacts with the central cavity of the SUR ABC core (Ding et al., 2019; Martin et al., 2019; Sung et al., 2021; Wu et al., 2018). The structures presented here refines our view of the molecular interactions between the KNtp and different parts of SUR1. We have previously shown that engineered crosslinking between Kir6.2-L2C and SUR1-C1142 reduces channel activity (Martin et al, 2019). A likely scenario is that stapling KNtp along SUR1 via the contact sites we observe stabilizes the inhibited Kir6.2-CTD-up conformation and prevents further rotation of the CTD needed to open the channel. This interpretation can explain why deletion of KNtp increases channel open probability (Babenko et al., 1999; Koster et al., 1999; Reimann et al., 1999), while drugs such as GBC, RPG and CBZ, which stabilize KNtp in the transmembrane cavity of the SUR1 ABC module, mimic the physiological inhibitor ATP and block channel activity (Devaraneni et al., 2015).

Comparison to other Kir channels

Differential proximity of the CTD to the membrane has also been reported in Kir2 (Hansen et al., 2011; Lee et al., 2016b; Zangerl-Plessl et al., 2020) and Kir3 (Niu et al., 2020), suggesting there may be a common theme in Kir channel conformation and gating transitions. Supporting this, mutations which disrupt cytoplasmic domain subunit interface in Kir2.1 also reduce channel activity, akin to the inactivation mutations reported in Kir6.2 (Borschel et al., 2017). Distinctively however, unlike Kir2 and Kir3 channels, K_ATP channels have an additional ATP-bound Kir6.2-CTD-up conformation stop between the CTD-down inactivated and CTD-up open conformations, which allows for a rapid and reversible inhibition of the channel in response to metabolic signals.

Another unique feature of Kir6.2 channels is the requirement of SUR1 co-assembly to achieve the high ATP sensitivity and open probability of native K_ATP channels (Inagaki et al., 1995; Tucker et al., 1997). Several studies have now provided functional, biochemical and structural evidence that SUR1 directly participates in ATP binding via K205 in the L0 linker (Ding et al., 2019; Pratt et al., 2012; Usher et al., 2020). SUR1 increases the open probability of Kir6.2 by more than 10-fold, an effect that is largely mediated by TMD0 (Babenko and Bryan, 2003; Chan et al., 2003). We show in our structure that K134 in SUR1-ICL2 is oriented towards the lipid headgroup density in the PIP₂ binding pocket, suggesting SUR1-TMD0 increases channel open probability by directly contributing to binding of PIP₂ or other phospholipids. Supporting this, MD simulations show K134-PIP₂ interactions (Fig.6D, Fig.S6D) and functional experiments show that mutation of SUR1 K134 to alanine reduces channel P_o (Fig.5E, F).

Structural basis of ATP and PIP₂ antagonism

ATP and PIP₂ functionally compete to inhibit and activate K_ATP channels, respectively (Baukrowitz et al., 1998; Shyng and Nichols, 1998). Molecular dynamics (MD) simulations reveal that two Kir6.2 residues, K39 and R54, interact with both ATP and PIP₂, providing evidence that competition for binding residues between the two ligands underlies, at least in part, functional competition between ATP and PIP₂. Interestingly, in one of the ATP+PIP₂ simulation runs, ATP dissociates from its binding pocket (Fig.6B). This dissociation event is likely captured because SUR1-L0, which contains the ATP stabilizing residue K205 is not included in the simulation structure. It offers a view of how ATP may dissociate such that binding residues shared between ATP and PIP₂ have greater freedom to interact with PIP₂, favoring channel opening. The rotational movement of SUR1 towards the Kir6.2-tetramer observed in our multibody refinement analysis (Fig.S4) increases the distance between SUR1-K205 and bound ATP, which may initiate ATP dissociation by weakening ATP binding and thus provide a pathway for channel transition from ATP bound inhibited state to PIP₂ bound open state.

In summary, the structural analysis, MD simulations, and functional studies presented here together with the recent open K_ATP channel structure reported by others (Zhao and MacKinnon, 2021) offer insight into several longstanding questions on K_ATP channel gating mechanisms. A principal unresolved question regards the full extent of the conformational dynamics of the SUR1 subunit and how it may relate to channel function. The large rotation of the SUR ABC module that leads to a quatrefoil channel conformation has previously been reported in a human SUR1–6aa-Kir6.2 fusion protein channel in which the NBDs are bound to MgATP/MgADP and dimerized (Lee et al., 2017). Recently, a similar large rotation is reported in a SUR2B/Kir6.1 vascular K_ATP channel bound to GBC and ATP (Sung et al., 2021). However, the quatrefoil conformation is not observed in the most recent human SUR1 NBDs dimerized-Kir6.2 mutant open channel (Zhao and MacKinnon, 2021). Whether the variable findings stem from protein preparation methods or involve differences in data processing will be important to resolve in order to fully understand K_ATP channel structure and function relationship.

Methods

Image processing and particle classification

CryoEM images of pancreatic K_ATP channels (co-assembled from hamster SUR1 and rat Kir6.2) collected in different liganded conditions in our previous publication (Martin et al., 2019) were reprocessed from the initial 2D classification step that we described previously using RELION-3.1 (Zivanov et al., 2018). Classes displaying fully and partially assembled complexes with high signal/noise were selected. The particles were re-extracted at 1.045 Å/pix for RPG/ATP, 1.399 Å/pix for GBC/ATP, 1.72 Å/pix for CBZ/ATP and ATP only, and 1.826 Å/pix for apo state datasets, and then used as input for 3D classification in RELION-3.1. Fig.S1 shows the data processing workflow for the RPG/ATP dataset. Channel particles refined in the final C4 reconstruction (150,707 total particles) were subjected to C4 symmetry expansion and yielded 4 fold more copies. Further refinement was performed without symmetry restraints or masking such that possible heterogeneous particles can be aligned without any restraints. 2D class averages from all data sets showed significant heterogeneity of the SUR1-ABC module, indicating dynamic SUR1 motions were captured during vitrification of the cryoEM samples. To probe potential novel conformations due to dynamics of SUR1-ABC module relative to the Kir6.2 tetramer, or for novel conformations that arise due to dynamic motions of other domains of the K_ATP channel, a soft mask that includes the Kir6.2 tetramer and one SUR1 in a propeller form was created in Chimera using our previously published model (PDB:6BAA) (Martin et al., 2017a), and extensive focused 3D classification was performed without particle alignment. This revealed two major classes with different Kir6.2-CTD conformations that are either anchored up towards the plasma membrane (CTD-up) or extended down further towards the cytoplasm (CTD-down).

Focused refinement of SUR1 was carried out after partial signal subtraction that removed signals outside the masked region, followed by further 3D classification without alignment at higher regularization T values (ranged from 6 to 20) and local refinement of signal subtracted particles (Scheres, 2016). Extensive 3D classification sorted out remaining minor groups of particles that did not align well with the propeller conformation but no other conformations emerged. The dominant class then underwent three iterations of CTF refinement and 3D refinement. To test whether some of the SUR1 particles adopt quatrefoil-like conformation reported previously, a mask that includes the Kir6.2 tetramer and one SUR1 was also created using a quatrefoil-like model of our previously published Kir6.1/SUR2B structure (PDB:7MJO) (Sung et al., 2021), which was used for classification following the same scheme described for the propeller form mask. A minor quatrefoil form at 7.1 Å overall resolution was identified from the RPG/ATP dataset (Fig.S1). Final maps were subjected to Map-modification implemented in Phenix with two independent half maps and corresponding mask and model as input. They were then sharpened with model-based auto sharpening with the corresponding model using Phenix, a step that was iterated during model building.

The same workflow was used to process the other four datasets, GBC/ATP, CBZ/ATP, ATP only and apo states. With the exception of the apo dataset which yielded only the CTD-down conformation, all other datasets showed both Kir6.2-CTD classes similar to those identified in the RPG/ATP dataset but with varying ratios of the two conformations. Particle distributions and final map resolutions for all datasets are summarized in Table S1. Upon carrying out this further analysis, we noted the CBZ/ATP dataset previously reported to be collected in the presence of 10μM CBZ and 1mM ATP did not yield a map with clear ATP density at the Kir6.2 ATP binding site. Upon inspection of the ATP used it was discovered that the concentration had been mistakenly reported as 1mM rather than 0.1mM, which likely explained the lack of clear ATP cryoEM density.

Multibody refinement

For Kir6.2 tetramer multibody refinement, particles pooled from both the RPG/ATP CTD-up and CTD-down classes and also each class separately were used (Nakane et al., 2018). To interrogate the Kir6.2 CTD movements relative to its TM domain, we masked out the SUR1 density and assigned the Kir6.2 TM domain (58–173) and CTD (174–352) as two separate rigid bodies (Fig.S3A). Principal component analysis showed that a dominant eigenvector accounted for 25.3% of the overall variance (Fig.S3B). Histograms of the amplitudes along this eigenvector shows a bimodal distribution with two peaks, indicating two conformationally distinct populations differing in the distance between the CTD and the TM domain and rotation of the CTD as expected. Rotation and rocking motions of the CTD were also observed within the Kir6.2-CTD-up and Kir6.2-CTD-down class particles. Although similar in degrees of motion, greater heterogeneity was seen in the CTD-down population of particles than in the CTD-up (Fig.S3C, D), consistent with increased mobility of the CTD when extended away from the membrane.

For (Kir6.2)₄-SUR1 multibody refinement, the map was divided into three bodies: body 1, Kir6.2-tetramer (E30-D352); body 2, ABC-core (Q211-V1578) of SUR1 plus KNtp (M1-E19) of Kir6.2; and body 3, TMD0 (M1-L210) of SUR1 (Fig.S4A). Multibody refinement was repeated with varying standard deviations for the degrees of rotation, and pixels of translation, to rule out artifacts. Principal component analysis in the relion_flex_analyse program revealed that approximately 17.5% of the variance is explained by the dominant eigenvector 1 (Fig.S4B,C) corresponding to horizontal swinging motion of SUR1 (Nakane et al., 2018). We then used the program to generate two separate STAR files, each containing ~35,000 particles with eigenvalues less than −5 or greater than +5 along eigenvector 1 (Fig.S4D). These two sets of particles were further refined using a soft mask, yielding two maps which we refer to as SUR1-out and SUR1-in with an overall resolution of 3.8 and 3.9 Å, respectively.

Model building and refinement

The RPG/ATP dataset yielded the highest resolution maps and were used for modeling (Table S2). Initial models for the Kir6.2₄-SUR1 channel were obtained by docking Kir6.2-TMD (32–171) and Kir6.2-CTD (172–352) from our previously published model (PDB:6BAA) (Martin et al., 2017b), and TMD0/L0 (1–284), TMD1 (285–614), NBD1 (615–928), NBD1-TMD2-linker (992–999), TMD2 (1000–1319) and NBD2 (1320–1582) of SUR1 (PDB:6PZA) (Martin et al., 2019), into either the RPG-CTD-up or the RPG-CTD-down cryoEM density map by rigid-body fitting using Chimera’s ‘Fit in’ tool (Pettersen et al., 2004). Then different domains were combined using Chimera to form a composite model for further refinement. We used Coot to manually build and edit residues 32–78 and residues 79–361 at the interface of two Kir6.2 subunits (Emsley et al., 2010), we then copied those changes to each of the other four Kir6.2 subunits. Further edits and refinements were done independently to each chain without strict NCS restraints. The models were then iteratively built and refined in Coot (Emsley et al., 2010) and Phenix (Afonine et al., 2018), with Ramachandran restraints, secondary structure restraints, and side-chain rotamer restraints. The N-terminus of Kir6.2 (residues 1–31, KNtp) had sufficient continuous density and the density was sufficiently clear to allow modeling of several key interactions with SUR1, although accurate modeling of side chains was not possible for the entire KNtp. Similarly, the NBDs of SUR1 and particularly NBD2 had weaker density than most of the reconstructed map, imposing reliance on restraints and prior models. In addition to modeling the protein, ATP was modeled at the inhibitory ATP-binding site on each of the four Kir6.2 subunits, and at the nucleotide binding site in NBD1 of SUR1 which had sufficient cryoEM density. Phosphatidylserine, phosphatidylcholine, and phosphatidylethanolamine were modeled liberally into plausible lipid density; 17 lipids were modeled for the RPG-CTD-up model and 13 lipids modeled for the RPG-CTD-down model.

MD simulations and analysis

All MD simulations were performed at all-atom resolution using AMBER 16 (Case et al., 2016) with GPU acceleration. Initial coordinates were developed from the RPG/ATP-CTD-up model including four Kir6.2 (32–352) and four SUR1-TMD0 (1–193) without the SUR1-ABC core. ATP in the cryoEM structure was removed for simulations in the apo condition. For simulations in the presence of ATP and PIP₂, the ATP from the cryoEM structure was kept, and a PIP₂ molecule (DMPI24, di-myristoyl-inositol-(4,5)-bisphosphate) was docked in the PIP₂ binding pocket using Kir3.2-PIP₂ structure (PDB ID: 6M84) as a template.

The simulation starting structures were protonated by the H++ webserver (http://biophysics.cs.vt.edu/H++) at pH 7 and inserted in a bilayer membrane composed of 1-palmitoyl-2-oleoyl-phosphatidylcholine (POPC) lipids and surrounded by an aqueous solution of 0.15 M KCl. The optimal protein orientations in the membrane were obtained from the OPM database (Lomize et al., 2012). All systems contain 650–680 POPC lipids and ~87,000 water molecules, resulting in a total of ~385,000 atoms. They were assembled using the CHARMM-GUI webserver (Jo et al., 2008; Lee et al., 2016a; Wu et al., 2014), which also generated all simulation input files.

The CHARMM36m protein (Huang et al., 2017) and CHARMM36 lipid (Klauda et al., 2010; Pastor and Mackerell, 2011) force field parameters were used with the TIP3P water model (Jorgensen, 1983). Langevin dynamics (Pastor, 1988) were applied to control the temperature at 300K with a damping coefficient of 1/ps. Van der Waals (vdW) interactions were truncated via a force-based switching function with a switching distance of 10 Å and a cutoff distance of 12 Å. Short-range Coulomb interactions were cut off at 12 Å, long-range electrostatic interactions were calculated by the Particle-Mesh Ewald summation (Darden et al., 1993; Essmann et al., 1995). Bonds to hydrogen atoms were constrained using the SHAKE algorithm (Jean-Paul Ryckaert, 1977).

The atomic coordinates were first minimized for 5000 steps using the steepest-descent and conjugated gradient algorithms, followed by a ~2 ns equilibration simulation phase, during which dihedral restraints on lipid and protein heavy atoms were gradually removed from 250 to 0 kcal/mol/Ų, the simulation time step was increased from 1 fs to 2 fs, and the simulation ensemble was switched from NVT to NPT. To keep the pressure at 1 bar, a semi-isotropic pressure coupling was applied that allows the z-axis to expand and contract independently from the x-y plane (Martyna, 1994). The simulations were then run for over 1 μs with a time step of 4 fs enabled by hydrogen mass repartitioning (Balusek et al., 2019; Hopkins et al., 2015).

Analysis of ATP/PIP₂ occupancy

ATP and PIP₂ residue occupancies (Fig. 6D) were computed using the histogram of minimum hydrogen bond lengths between each residue and PIP₂/ATP, to show the amount of time spent at different distances. Summarized residue occupancies (Fig. S6D) were calculated as the fraction of time each residue spent in contact with ATP/PIP₂, where contact is defined as a minimum hydrogen bond length of below 4 Å. For both ligands, minimum bond lengths were used regardless of which pairs formed the bond. A 10 ns window average was used to smooth the minimum bond length time series data.

Functional studies

Point mutation SUR1-K134A was introduced into hamster SUR1 cDNA in pECE using the QuikChange site-directed mutagenesis kit (Stratagene). Mutation was confirmed by DNA sequencing. For electrophysiology, wild-type or mutant SUR1 cDNA and rat Kir6.2 in pcDNA1 along with cDNA for green fluorescent protein GFP (to facilitate identification of transfected cells) were co-transfected into COS cells using FuGENE®6, and plated onto glass coverslips 24 hours after transfection for recording, as described previously (Martin et al., 2019). Recording pipettes 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, pH 7.3. ATP was added as the potassium salt. PI4,5P₂ (Avanti Polar Lipids) was reconstituted in K-INT solution at 5 μM and bath sonicated in ice water for 20 min before use. Inside-out patches of cells bathed in K-INT were voltage-clamped with an Axopatch 1D amplifier (Axon Inc., Foster City, CA). Exposure of membrane patches to ATP- or PIP₂-containing K-INT bath solution was as specified in Fig.5 legend. All currents were measured at room temperature at a membrane potential of −50 mV (pipette voltage = +50 mV) and inward currents shown as upward deflections. Data were analyzed using pCLAMP10 software (Axon Instrument). Off-line analysis was performed using Microsoft Excel programs. Data were presented as mean ± standard error of the mean (S.E.M) and statistical analysis was performed using two-tailed student’s t-test, with p<0.05 considered statistically significant.

Accession numbers

Coordinates and cryoEM density maps for K_ATP channel models of the Kir6.2 tetramer core plus one SUR1 subunit have been deposited to the Protein Data Bank and the Electron Microscopy Data Bank with accession numbers as follows: RPG/ATP Kir6.2-CTD-up (PDB code 7TYS, EMDB code EMD-26193); RPG/ATP Kir6.2-CTD-down (PDB code 7TYT, EMDB code EMD-26194); RPG/ATP Kir6.2-CTD-up SUR1-in (PDB code 7U1Q, EMDB code EMD-26303); RPG/ATP Kir6.2-CTD-up SUR1-out (PDB code 7U1S, EMDB code EMD-26304); GBC/ATP Kir6.2-CTD-up (PDB code 7U24, EMDB code EMD-26307); GBC/ATP Kir6.2-CTD-down (PDB code 7U6Y, EMDB code EMD-26308); CBZ/ATP Kir6.2-CTD-up (PDB code 7U7M, EMDB code EMD-26309); CBZ/ATP Kir6.2-CTD-down (PDB code 7U2X, EMDB code EMD-26321); ATP-only Kir6.2-CTD-up (PDB code 7UAA, EMDB EMD-26312); ATP-only Kir6.2-CTD-down (PDB 7U1E, EMDB code EMD-26299); Apo Kir6.2-CTD-down (PDB code 7UQR, EMDB code EMD-26320).

Highlights.

• The activity of K_ATP channel is governed by the interplay between its ligands and its subunits, Kir6.2 and SUR1.

• Inhibitory ligands bias Kir6.2-cytoplasmic domain conformation towards one close to the membrane instead of one extended away into the cytoplasm via a corkscrew motion.

• SUR1 cooperates with Kir6.2 to stabilize inhibitor ATP and activator PIP₂ binding thereby enhancing channel sensitivity to both.

• ATP and PIP₂ compete with each other to close or open the channel, respectively, by having overlapping but non-identical binding residues.

• Ligand-dependent conformational dynamics of the Kir6.2-cytoplasmic domain offer insights into K_ATP channel gating mechanisms.