Anti-diabetic drug binding site in KATP channels revealed by Cryo-EM

Sulfonylureas are anti-diabetic medications that act by inhibiting pancreatic KATP channels composed of SUR1 and Kir6.2. The mechanism by which these drugs interact with and inhibit the channel has been extensively investigated, yet it remains unclear where the drug binding pocket resides. Here, we present a cryo-EM structure of the channel bound to a high-affinity sulfonylurea drug glibenclamide and ATP at 3.8Å resolution, which reveals in unprecedented details of the ATP and glibenclamide binding sites. Importantly, the structure shows for the first time that glibenclamide is lodged in the transmembrane bundle of the SUR1-ABC core connected to the first nucleotide binding domain near the inner leaflet of the lipid bilayer. Mutation of residues predicted to interact with glibenclamide in our model led to reduced sensitivity to glibenclamide. Our structure provides novel mechanistic insights of how sulfonylureas and ATP interact with the KATP channel complex to inhibit channel activity.

Discovered in the 1940s, sulfonylureas have been a mainstay of type 2 diabetes therapy for more than half a century (Sola et al., 2015). The medical importance of this class of drugs has led to its evolution into several generations of agents, including first-generation sulfonylureas such as tolbutamide and second-generation agents such as the high-affinity sulfonylurea glibenclamide (GBC) (Gribble and Reimann, 2003;Sola et al., 2015). All sulfonylureas stimulate insulin secretion to reduce plasma glucose levels by blocking the activity of β-cell KATP channels (Gribble and Reimann, 2003). More recently, they have also become the primary pharmacotherapy for neonatal diabetes patients carrying gain-of-function KATP channel mutations (Aguilar- Bryan and Bryan, 2008;Ashcroft, 2007;Sagen et al., 2004). Despite their clinical importance and decades of research, how sulfonylureas interact with and block KATP channel activity remains poorly understood.
To begin to address the structural mechanisms by which ATP and sulfonylureas such as GBC inhibit KATP channels to stimulate insulin secretion, we carried out single particle cryo-EM and determined the structure of the β-cell KATP channel complex in the presence of ATP and GBC (Martin et al., 2017). While our initial structure at a resolution of 5.7Å revealed the overall architecture of the channel and location of the ATP molecule, it was unable to clearly define the GBC binding site and the atomic details associated with ATP binding. In this study, we improved the resolution of the KATP channel structure bound to GBC and ATP to 3.8Å. The higher resolution structure not only clearly defines the GBC and ATP binding pockets but also provides novel insights into the mechanisms of channel inhibition by ATP and GBC.

Structure determination
To obtain a structure of KATP channels bound to GBC and ATP, channels comprising a rat Kir6.2 and FLAG-tagged hamster SUR1 (96 and 95% sequence identity to human sequences, respectively) were expressed in rat insulinoma INS-1 832/13 cells (Hohmeier et al., 2000), affinity purified, and imaged in the presence of 1mM ATP (no Mg 2+ ) and 1µM GBC, as described previously (Martin et al., 2017). To improve resolution, we adjusted sample and grid preparation parameters (see Materials and Methods) to optimize ice thickness and particle orientation distributions, which both increased the overall quality and quantity of single particles.
3D classification in RELION yielded one four-fold symmetric class, which reached an overall resolution of 3.8Å after refinement ( Fig. 1-figure supplements 1; Table 1). The local resolution, as estimated by Bsoft, varied from 3.2Å in the Kir6.2 transmembrane domain (TMD) to 5.2Å in the SUR1 nucleotide binding domains (NBDs) (Fig. 1-figure supplements 2). Overall, the map displays excellent connectivity to allow for model building (Fig.1). We have constructed a full atomic model for all of Kir6.2 minus disordered N-and C-termini, and for TMD0 of SUR1, as this part of the map was almost entirely at or below 3.7Å, with clear side-chain density for most residues. The ABC core of SUR1 displayed greater variability in resolution: the inner helices (relative to Kir6.2/TMD0) were also very well resolved (between 3.5 and 4Å resolution) to permit nearly complete atomic model building, while the most exposed helices (TMs 9,10,12,and 13) showed signs of flexibility, and were only built as polyalanine chains. This was also the case for the NBDs, for which we only refined our previously-deposited NBD models as rigid bodies (see Materials and Methods).

Structural overview
The KATP channel is built around a tetrameric Kir6.2 core with each subunit in complex with one SUR1 (Fig.1B-E), as observed previously (Martin et al., 2017). Each Kir6.2 has the typical Kir channel architecture of an N-terminal cytoplasmic domain, a TMD consisting of two TMs termed M1 and M2 interspersed by a pore loop and selectivity filter, and a "tether" helix that links the TMD to the larger C-terminal cytoplasmic domain (CTD) ( Fig.2A, Fig.2-figure supplement 1). In our new structure, constrictions in the selectivity filter (T130), bundle crossing (F168), and the Gloop (G295, I296) are clearly seen (Fig.2B, C) to indicate a closed pore, as expected given the channel is bound to ATP. SUR1 is one of only a handful of ABC transporters which possess an N-terminal transmembrane domain, TMD0, in addition to an ABC core structure comprising two TMDs of 6 helices each and two cytosolic NBDs (Tusnady et al., 2006;Wilkens, 2015) (Fig.2D). In the structure, TMD0 is a well-resolved 5-TM bundle (Fig.2D, Fig.2-figure supplement 2). A long intracellular loop L0 which tethers TMD0 to the ABC core is found to contain both cytosolic and amphipathic domains (Fig.2D, Fig.2-figure supplement 2). The C-terminal 2/3 of L0 is homologous to the "lasso motif" observed in CFTR and MRP1 (Johnson and Chen, 2017;Liu et al., 2017;Zhang and Chen, 2016), and indeed, the structures are very similar (Fig.2E). SUR1 is found in an "inward-facing" conformation, with NBDs clearly separated and the vestibule formed by TMD1/TMD2 open towards the cytoplasm. As we noted previously, the two TMD-NBDs show a ~15° rotation and ~10Å horizontal translation relative to each other (Fig.2F). This lack of symmetry is also seen in recently reported CFTR and MRP1 inward-facing structures. The separation between the two NBDs in our structure is similar to that seen in MRP1 bound to its substrate leukotriene C4 (see Fig.2F). Like SUR1 in which only NBD2 is capable of hydrolyzing ATP while NBD1 harbors a degenerate ATPase site, CFTR and MRP1 also have two asymmetric NBDs, suggesting the relative rotation and translation between the two TMD-NBD halves may be a common characteristic of ABC transporters with asymmetric NBDs.

Unique molecular interactions between SUR1 and Kir6.2
Among all Kir channels, Kir6.1/Kir6.2 are the only members known to couple to an ABC transporter; and among all ABC transporters, SUR1/SUR2 are the only ones known to couple to an ion channel. These proteins are also unique in that they are co-dependent for both expression and function (Inagaki et al., 1995;Zerangue et al., 1999). How SUR1 and Kir6.2 achieve this unique regulation has been a long standing question in the field.
In the structure, we find a series of hydrophobic and polar interactions mediated exclusively by TMD0 and L0 with Kir6.2 (Fig. 3). The extracellular N-terminus of SUR1 closely contacts the turret and pore loop of Kir6.2 ( Fig.3A-D), while TM1 of TMD0 and the M1 helix of Kir6.2 form a series of hydrophobic interactions running the length of the helices (Fig.3E). On the cytoplasmic side, the intracellular loops ICL1, ICL2, and the N-terminal portion of L0, prior to the "lasso motif," cluster around the Kir6.2 N-terminus which harbors the slide helix and key ATP-binding residues and also forms part of an intersubunit β-sheet (Fig.2B).
Of all the Kir channel family members which display a high degree of sequence conservation, only Kir6.1 or Kir6.2 co-assemble with SUR proteins, raising the interesting question of what molecular interactions confer this specificity. While many residue pairs in the interface are conserved in either the Kir or ABC transporter family, a couple residue pairs are unique to both Kir6.2 and SUR1. Among these, H70 within the M1 helix of Kir6.2 forms an edge-to-face πstacking interaction with W51 of TM1 of TMD0 (Fig.3E), and Q57 of the Kir6.2 slide helix contacts F132 of ICL2 (Fig.3F). F132 is a well-studied PNDM mutation which causes very high Po but also reduces physical interaction between TMD0 and Kir6.2 (Proks et al., 2007), supporting its role as a critical part of the interface. To our knowledge, mutational studies of Kir6.2 Q57 and H70, and SUR1 W51 have not been reported; it would be interesting to test the role of these residues in channel assembly and function in the future.
The ATP binding site Non-hydrolytic binding of intracellular ATP to the cytoplasmic domains of Kir6.2 induces rapid and reversible closure of the pore (Nichols, 2006). We have previously reported the location of the ATP-binding site at the interface of the cytoplasmic N-and C-terminal domains from two adjacent subunits (Martin et al., 2017), giving 4 equivalent sites for the Kir6.2 tetramer. In the current map, there is strong electron density for the ATP as well as surrounding residues ( Fig.4., Fig.4-figure supplement 1), allowing for detailed analysis of the mode of ATP binding as well as the mechanism of inhibition.
In the structure, the ATP is directly below the inner membrane leaflet and is partially exposed to solvent (Fig.4). The bound ATP appears to adopt a conformation similar to that found in other non-canonical, Mg 2+ -independent ATP-binding sites, such as the P2X receptor (Hattori and Gouaux, 2012), in which the phosphate groups are folded towards the adenine ring ( Fig.4B, C). This places the and -phosphates to interact with basic residues contributed by the N-and Ctermini. The pocket itself is formed by the overlap of three distinct cytoplasmic structures: an Nterminal peptide (binding residues I49 and R50; subunit A) immediately before the Kir channel "slide helix"; a C-terminal -sheet (I182 and K185; subunit B) immediately following the TMD-CTD tether helix (see Fig.2A, Fig.2-figure supplement 1); and a short, solvent-exposed helical segment (Y330, F333, G334; subunit B).
The phosphate of ATP is coordinated by the main-chain nitrogen of G334 and K185, while the and -phosphates are coordinated by side-chain nitrogens of K185 and R50, respectively. The ribose group is in close contact with the I182 and F333 side chains; the adenine ring stacks against the aliphatic portion of the R50 side chain as well as Y330, and is H-bonded to the main chain nitrogen of R50, and main chain carbonyls of I49 and Y330. The aforementioned residues have all been shown previously to reduce ATP inhibition when mutated to other amino acids Cukras et al., 2002;Drain et al., 1998;Li et al., 2005;Proks et al., 1999;Tammaro et al., 2005;Tucker et al., 1998), consistent with a role of these residues in ATP gating. Notably, sequence comparison reveals that a key difference between Kir6.2 and other Kir channels is G334, which in other Kir channels is occupied by larger amino acids. Substitution of glycine at this position by a larger amino acids such as histidine seen in Kir2 or Kir3 channels would create steric hindrance to prevent ATP binding. This may explain, at least in part, why Kir6.2 is the only Kir channel sensitive to ATP regulation.

Structural interactions around the ATP binding site and their relationship to the PIP2 binding site
A number of residues within the vicinity of the ATP-binding site such as E179, R201, and R301 have previously been shown to reduce ATP sensitivity and proposed to be involved in ATP binding . However, from the structure it is clear these residues contribute indirectly (Fig.5). E179 and R301 were both proposed to interact with the adenine ring . In our structure, neither residue forms direct interactions with ATP. E179 appears to interact with R54 from the adjacent Kir6.2 and may be part of the network that stabilizes the interaction between R50 and ATP (Fig.5C). R301 is found to interact with Q299 in the same β-strand that is part of a β-sheet in the Kir6.2 CTD ( Fig. 5A; see also Fig.2-figure supplement 1). Interestingly, R301 is one of the most highly mutated residues in congenital hyperinsulinism; mutation of R301, in addition to mildly reducing ATP sensitivity, results in rapid decay of channel activity that can be reversed by increasing PIP2 concentrations in the membrane (Lin et al., 2008). Based on our structure, it is likely that mutation of this residue disrupts structural integrity of the Kir6.2 CTD necessary for stable channel interaction with PIP2 and ATP. By contrast, R201 is one of the most highly mutated residues in neonatal diabetes (Ashcroft, 2005). It has been proposed that R201 coordinates the -phosphate of ATP . However, in the structure R201 is found on the -strand directly below that of I182 and K185, and is too distant to directly interact with ATP. Instead, R201 is sandwiched between the benzene rings of F333 and F315, forming a dual cation-π interaction that likely stabilizes the ATP-binding site (Fig. 5B). Mutation of R201 would therefore destabilize this interaction to indirectly reduce ATP inhibition.
Another interesting residue is Q52. The PNDM mutation Q52R causes extremely high Po and very low ATP sensitivity (Lin et al., 2006;Proks et al., 2004). In the structure, Q52 interacts with R50 which coordinates the γ-phosphate of ATP, but is also interacting with R54, which orients the R54 side chain towards the ATP site and away from the PIP2 site nearby (Fig.5C). In the PIP2 bound Kir2.2 structure, the Kir6.2 R54 equivalent arginine residue interacts with the tether helix near PIP2 binding residues (Hansen et al., 2011), suggesting that R54 may be important for Kir6.2 -PIP2 interactions. Interestingly, mutation of R50 or R54 to an alanine has been reported to reduce sensitivity to both ATP and PIP2 (Cukras et al., 2002). From our structure, it is easy to envision how mutation of any of these residues can disrupt the interaction network to affect gating by either ligand.
It is also important to note that in our structure, Q52 is in close proximity to E203 in the L0 of SUR1 immediately following TMD0. We have previously shown that engineered interactions between Kir6.2 residue 52 and SUR1 residue 203 via a Kir6.2-Q52E and SUR1-E203K ion pair increases channel sensitivity to ATP by nearly two orders of magnitude, and that crosslinking of the two residues via a Kir6.2-Q52C and SUR1-E203C mutant pair induces spontaneous channel closure in the absence of ATP (Pratt et al., 2012). In addition, our previous studies have shown that ATP binding involves residues from not only the N-terminus of Kir6.2 such as R50 but also residues in SUR1-L0 such as K205 (Martin et al., 2017;Pratt et al., 2012). Together these studies lead us to propose that the inhibitory effect of ATP is partially due to stabilizing the interaction between this N-terminal region of Kir6.2 and L0 of SUR1 (see Discussion).
The GBC binding site Sulfonylureas stimulate insulin secretion by inhibiting pancreatic KATP channels. GBC, also known as glyburide, is a second generation sulfonylurea that contains both a sulfonylurea moiety and a benzamido moiety, and binds KATP channels with nM affinity (Gribble and Reimann, 2003). Despite intense investigation the GBC binding site has remained elusive. Early studies using chimeras of SUR1 and SUR2A, which is known to have lower sensitivity to GBC than SUR1, suggest the involvement of TMs 14-16; in particular mutating S1238 in SUR1 to Y (note in some papers, this is numbered as S1237) as seen in SUR2A compromised GBC binding and block (Ashfield et al., 1999;Winkler et al., 2007). Subsequent studies using 125 I-azido-GBC photolabeling implicated involvement of L0 of SUR1; specifically, two mutations Y230A and W232A in L0 severely compromised photolabeling of SUR1 (Vila-Carriles et al., 2007). These studies led to a model in which S1238 and Y230/W232 constitute two ends of a bipartite binding pocket, each recognizing opposite ends of GBC (Winkler et al., 2007); however, whether one or both contribute directly to GBC binding is unknown.
In the current reconstruction we find well defined, non-protein density within the TMDs of SUR1, with a size and shape which closely matches that of a GBC molecule (Fig.6, Fig.6-figure supplement 1A). One end of the density is in direct contact with S1238 and resembles the cyclohexyl moiety long presumed to constitute the "A" site that is abolished by the S1238Y mutation (Ashfield et al., 1999;Bryan et al., 2004). We used this to guide the initial docking of GBC, which could then be readily refined into the density together with SUR1.
The binding pocket is contoured to precisely accommodate GBC, and the combination of polar and hydrophobic residues help explain the sub-nM affinity of SUR1 for this sulfonylurea (Fig. 6C, Fig.6-figure supplement 1B, C). A primary anchor is composed of two arginine residues, R1246 and R1300, which coordinate each oxygen of the sulfonyl group. Each nitrogen of the urea moiety is coordinated by T1242 and N1245, and the adjacent benzene and cyclohexyl groups (adjacent to the sulfonyl and urea groups, respectively) are stabilized by a series of hydrophobic interactions contributed by TM helices from both TMD1 (TM6, 7, 8) and TMD2 (TM16). As a secondgeneration sulfonylurea, GBC contains another lipophilic group adjacent to an amide linker, which is lacking in first-generation compounds like tolbutamide. This group, a 1-chloro-4-methoxybenzene, is encircled by a ring of hydrophilic and hydrophobic side chains. In particular, the Cl appears to hydrogen bond with the amino group of N437, while the methoxy is H-bonded to the hydroxyl group of Y377. Y377 also seems to contribute a π-π stacking interaction with the benzene ring. In the structure, the previously proposed sulfonylurea binding residue S1238 juxtaposes the cyclohexyl group of GBC, with only ~3Å separation between the C of S1238 and the 6-carbon ring of GBC. Mutation of this residue to a tyrosine may alter interaction with GBC to compromise high affinity binding of GBC.
In order to validate the proposed binding site, we mutated a subset of key GBC-binding residues listed above to alanine and tested their response to 100 nM and 1µM GBC with Rb + efflux experiments, which measure channel activity and response to GBC in intact cells. All six mutations, R306A, Y377A, N437A, T1242A, R1246A, and R1300A trafficked normally and responded as WT to metabolic inhibition (Fig.7A, B). Strikingly, all six mutations showed significantly reduced or complete absence of inhibition at 100 nM, and five mutations (R306A, Y377, N437A, T1242, and R1246A) showed significantly reduced sensitivity compared to WT even at 1µM GBC (Fig. 7C, D). The four most GBC-insensitive mutants, R306A, Y377A, N437A, and T1242A were further analyzed by inside-out patch-clamp recording. Although these mutants were still sensitive to GBC inhibition, the extent of inhibition at steady-state was less compared to WT channels at 10nM and 100nM (Fig.7, figure supplement 1). Also worth noting, while inhibition of WT channels was nearly irreversible, inhibition of mutants was more reversible, consistent with the mutants having reduced affinity for GBC ( Fig.7-figure supplement 1). Together these results provide strong functional evidence for the GBC binding pocket defined in our structure.

The role of SUR1-Y230 and W232 in GBC interaction
In the previously proposed bipartite binding model for GBC, the pocket was formed from two overlapping regions: at one end was S1238 of TMD2, and at the other was L0 involving residues Y230 and W232, part of the "lasso motif" observed in MRP1 and CFTR. Mutation of Y230 to an alanine has also been shown to reduce the ability of GBC to inhibit channel activity (Devaraneni et al., 2015;Yan et al., 2006). In our current structure, Y230 is too distant to interact directly with GBC. However, the binding pocket is close to the L0-TMD interface, where the L0 amphipathic helix forms a series of mostly hydrophobic interactions with transmembrane helices from TMD1/2 that line the GBC binding pocket. Here, we find that Y230 stacks closely against the aliphatic portion of the R1246 side chain, which in turn coordinates an oxygen of the sulfonyl group of GBC (Fig.6D). W232 appears to form a strong interaction with M233, which interacts directly with two alanines, A1243 and A1244, on the opposite side of TM16 where two GBC interacting residues T1242 and N1245 are located (Fig.6D). These observations indicate an important but clearly indirect role for Y230 and W232 in GBC binding.

Discussion
The structure presented in this study is the first to reveal in detail the ATP and GBC binding sites in the SUR1/Kir6.2 KATP channel complex. The clear density for ATP and GBC as well as all residues involved in binding of both ligands in the current EM map allowed us to present a detailed atomic interpretation of ATP and GBC binding to the channel. Importantly, the binding pockets we identified are supported by strong functional data. In addition, the structure uncovers many molecular interactions that indirectly impact ATP and GBC gating, and those that underlie SUR1-Kir6.2 interactions. The structural information gained offers key insights into possible mechanisms of how the two ligands both inhibit KATP channels to stimulate insulin secretion.
The ATP binding site and mechanism of ATP inhibition The Kir6.2 interfacial ATP binding site model was first proposed by Antcliff et al. (Antcliff et al., 2005) based on ligand docking, homology modeling of Kir channel crystal structures, and structure-function mutagenesis data. Although some interactions in the original model between Kir6.2 residues and ATP are observed in our structure, many others require new interpretations. First, the ATP molecule adopts a conformation with the γ-phosphate bent towards the adenine ring (Fig.4), which is reminiscent of that observed in P2X receptors, an ATP activated ion channel (Hattori and Gouaux, 2012). Second, our structure suggests that E179, R201, and R301 rather than contributing directly to ATP binding are critical for interactions with other residues that support the ATP binding residues or general structural integrity of the Kir6.2 CTD for stable channel interaction with PIP2 (Fig.5). Elucidation of the structural role of R201 and R301 helps us to understand the mechanisms by which mutation of these residues cause insulin secretion disease. Finally, the structure provides details on the intricate relationship between ATP binding residues and those which have been implicated in PIP2 binding or gating, shedding light on how the channel senses ATP and PIP2 (Fig.5C).
We propose that ATP, by binding to a pocket created by the N-terminus and CTD from two adjacent Kir6.2 subunits with contributions from L0 of SUR1, acts like a stopper that prevents movement of the Kir6.2 N-terminus relative to L0 that is necessary to open the channel (Fig. 8A). This model is consistent with our previous study showing that crosslinking of SUR1-L0 with the N-terminus of Kir6.2 near the ATP binding site locks the channel closed even without ATP (Pratt et al., 2012). One interesting question is whether the interaction network we observed in the present ATP-bound structure undergoes remodeling in PIP2-bound open state. An open state structure of the KATP channel bound to PIP2 will be needed to understand the full extent of conformational change associated with ATP and PIP2 gating.

Mechanistic insights of GBC binding and inhibition
The binding site of the high affinity sulfonylurea GBC has been studied by many groups. These studies have implicated the involvement of transmembrane helices in the SUR1-ABC core, L0, and the N-terminus of Kir6.2 (Ashfield et al., 1999;Bryan et al., 2004;Vila-Carriles et al., 2007). Yet, the precise binding pocket for this commonly used anti-diabetic drug has remained unresolved. In a recent 5-6Å resolution reconstruction of a pancreatic KATP channel complex imaged in the presence of GBC, Li et al. (Li et al., 2017) assigned extra density around Y230 and W232 of SUR1 as GBC, thus placing the sulfonylurea binding site within an amphipathic helix of L0. However, at low resolution modeling of L0 is challenging, as this is an inherently lower resolution part of the structure; moreover, because Y230 and W232 are found at the interface of the inner leaflet of the bilayer, bound lipids could easily be mistaken for ligand density. Contrary to the study by Li et al., we were able to clearly assign the GBC density in the TM bundle connected to the NBD1, with residues from TM6, 7, 8, 11 in TMD1 and TM16 and 17 from TMD2 contributing to GBC interactions. Importantly, our model is supported by functional data using both 86 Rb + efflux assays and electrophysiological recordings. Moreover, our structure clarifies how Y230 and W232, which have previously been proposed to contribute to GBC binding based on indirect biochemical or functional assays (Vila-Carriles et al., 2007), can affect GBC binding or gating indirectly by supporting residues that are directly engaged in GBC binding.
The distal N-terminal 30 amino acids of Kir6.2 have been shown to be important for the binding or effect of GBC in a number of studies (Devaraneni et al., 2015;Koster et al., 1999;Kuhner et al., 2012;Reimann et al., 1999;Vila-Carriles et al., 2007). In our map, there is a lack of strong density N-terminal to position 32 of Kir6.2, suggesting this region is flexible. However, it is worth noting that our previous study using engineered unnatural amino acid Azido-pphenylalanine placed at the distal N-terminus of Kir6.2 (amino acid position 12 or 18) has demonstrated that GBC increased crosslinking of Kir6.2 to SUR1 (Devaraneni et al., 2015). Accordingly, we propose that GBC binding to the TMD bundle next to the L0 amphipathic helix stabilizes SUR1 in an inward-facing conformation, and possibly also stabilizes the interactions between the N-terminus of Kir6.2 and SUR1, as we have demonstrated previously, to prevent the movement of Kir6.2 N-terminus that is needed to open the gate (Fig.8B).
The residues which play a specific and critical role in GBC binding are also very likely important for binding of other sulfonylureas. While we only tested GBC, we predict that R1246 and R1300, which coordinate the sulfonyl group, and T1242 and N1245, which coordinate the urea group, will also be critical for binding of other sulfonylureas such as tolbutamide (Gribble and Reimann, 2003). In addition to sulfonylureas, glinides such as rapaglinide and nateglinide which lack the sulfonylurea moiety (Gribble and Reimann, 2003), and a structurally unrelated compound carbamazepine (Chen et al., 2013b;Devaraneni et al., 2015) are also known to inhibit KATP channels. Elucidating the role of the various GBC binding residues in channel interactions with the different channel inhibitors will be important for understanding channel inhibition mechanisms and for rational design of new drugs with desired properties.

Conservation of GBC binding residues in other SUR proteins and ABCC transporters
Multiple sequence alignment of 15 SUR1 orthologs from diverse genera shows relatively high sequence identity throughout the sequence relative to human; from 95% (hamster) to 75% (seahorse). Interestingly, the segments of the helices from TMD1 and TMD2 which comprise the GBC binding site show exceptionally high conservation, with every one of the 12 residues which most closely line the GBC pocket absolutely conserved in 15 out of the 15 sequences. The high degree of conservation suggests the importance of the interface formed by these transmembrane helices. It would be important to determine in the future whether this interface is involved in conformational switch of the SUR1 ABC core and its communication with Kir6.2.
Interestingly, SUR2 (ABCC9), the closest homolog of SUR1 (67% sequence identity), while also shows high conservation of the GBC binding residues (10/12), differs in two positions that correspond to S1238 and T1242 (Y and S respectively in SUR2). SUR2 assembles with Kir6.1 or Kir6.2 to form KATP channel subtypes found in the heart, skeletal muscle, and vascular smooth muscle. These channels are known to have lower sensitivity to GBC inhibition than channels formed by SUR1 and Kir6.2 (Inagaki et al., 1996). Variations at these two key GBC binding residues likely explains their different pharmacological sensitivity to GBC (Ashfield et al., 1999;Inagaki et al., 1996).
In addition to targeting KATP channels, GBC has been shown to inhibit other ABC transporters within the C subfamily (ABCC), including MRP1 and CFTR, albeit at lower affinity (~30µM for both CTFR and MRP1) (Payen et al., 2001;Schultz et al., 1996). Of all the residues within the GBC binding pocket, only two appear to be highly conserved across different members of the ABCC subfamily: R1246 and R1300. In fact, R1246 is strictly conserved within 11 of 12 ABCC homologs (Gln in ABCC10), and R1300 in 10 of 12 (Asn in CFTR and Ser in ABCC10). In the cryo-EM structure of MRP1 bound to substrate LTC-4 (Johnson and Chen, 2017), R1196 (equivalent to R1246 in SUR1) forms a salt bridge with a carboxylic acid group of LTC-4 and is also in the same rotameric conformation as R1246 in SUR1. Further, F221 (equivalent to Y230 in SUR1) also seems to form the equivalent hydrophobic stacking interaction with R1196 as Y230 does with R1246 in SUR1; this phenomenon is also observed in the human CFTR structure (F17 and R1097) (Liu et al., 2017). Such structural conservation likely explains the GBC sensitivity in other ABCC homologues, and suggests a critical role for this pair of residues in the function and/or structure of ABCC proteins.
In summary, we presented a KATP channel structure with improved resolution that allowed us to definitively identify the ATP and GBC binding sites. The novel insight gained from this structure significantly advances our understanding of how these two ligands interact with the channel to exert an inhibitory effect. As inhibition of KATP channels by sulfonylureas remains an important therapeutic intervention to control type 2 diabetes and neonatal diabetes, and there is a need for drugs that specifically target a KATP channel subtype, our study offers a starting point for future structure-guided drug development to mitigate diseases caused by KATP channel dysfunction.

Materials and methods
Protein expression and purification. KATP channels were expressed and purified as described previously (Martin et al., 2017). Briefly, the genes encoding pancreatic KATP channel subunits, which comprise a hamster SUR1 and a rat Kir6.2 (94.5% and 96.2% sequence identity to human, respectively), were packaged into recombinant adenoviruses (Lin et al., 2005;Pratt et al., 2009); Both are WT sequences, except for a FLAG tag (DYKDDDDK) that had been engineered into the N-terminus of SUR1 for affinity purification. INS-1 cells clone 832/13 (Hohmeier et al., 2000), a rat insulinoma cell line, were infected with the adenoviral constructs in 15 cm tissue culture plates. Protein was expressed in the presence of 1mM Na butyrate and 5 M GBC to aid expression of the channel complex. 40-48 hours post-infection, cells were harvested by scraping and cell pellets were frozen in liquid nitrogen and stored at -80°C until purification.
Sample preparation and data acquisition for cryo-EM analysis. In our previous data sets, most micrographs were of ice that was either too thin, which tended to exclude channel complex from the hole and induce highly preferred orientation, or of ice that was too thick, which gave good particle distribution and good angular coverage, but had lower contrast. Thus the current data set was the result of efforts to optimize ice thickness in order to retain high contrast and particle distribution. This was achieved through varying blotting time and also through extensive screening of the grid in order to find optimal regions. Two grids were imaged from the same purification and were prepared as follows: 3 µL of purified KATP channel complex was loaded onto UltrAufoil gold grids which had been glow-discharged for 60 seconds at 15 mA with a Pelco EasyGlow ®. The sample was blotted for 2s (blot force -4; 100% humidity) and cryo-plunged into liquid ethane cooled by liquid nitrogen using a Vitrobot Mark III (FEI).
Single-particle cryo-EM data was collected on a Titan Krios 300 kV cryo-electron microscope (FEI) in the Multi-Scale Microscopy Core at Oregon Health & Science University, assisted by the automated acquisition program SerialEM. Images were recorded on the Gatan K2 Summit direct electron detector in super-resolution mode, post-GIF (20eV window), at the nominal magnification 81,000x (calibrated image pixel-size of 1.720Å; super-resolution pixel size 0.86Å); defocus was varied between -1.4 and -3.0 µm across the dataset (Table 1). The dose rate was kept around 2.7 e -/Å 2 /sec, with a frame rate of 4 frames/sec, and 60 frames in each movie, which gave a total dose of approximately 40 e -/Å 2 . In total, 2180 movies were recorded.
Image processing. The raw frame stacks were gain-normalized and then aligned and dosecompensated using Motioncor2 (Zheng et al., 2017) with patch-based alignment (5x5). CTF parameters were estimated from the aligned frame sums using CTFFIND4 (Rohou and Grigorieff, 2015). Particles were picked automatically using DoGPicker (Voss et al., 2009) with a broad threshold range in order to reduce bias. Subsequently, each image was analyzed manually in order to recover any particles missed by automatic picking and remove obviously bad micrographs from the data set. This resulted in ~250,000 raw particles as input for subsequent 2D classification using RELION-2 (Kimanius et al., 2016). After four rounds of 2D classification, ~160,000 particles remained in the data set, in which only classes displaying fully assembled complexes and high signal/noise were selected. These 160K particles were re-extracted at 1.72 Å/pix and were used as input for 3D classification in RELION-2.
Extensive 3D classification was performed in order to sample the heterogeneity within the data. Symmetry was not imposed at this step in order to select only true four-fold classes. Up to 4 consecutive rounds of classification were performed, specifying 4 or 5 classes per round. Individual classes and combinations of classes were refined independently and lead to very similar structures. The two best classes from round 2 were combined (~63,000 particles), and then particles were re-extracted from super-resolution micrographs with a box size of 600 pixels. A soft mask encompassing the entire complex was used during refinement in RELION, with C4 symmetry imposed, which resulted in a 3.8Å reconstruction using the gold-standard FSC cutoff. The map was B-factor corrected and filtered using RELION-2 Postprocessing procedure, with the same mask used for refinement. Local resolution was calculated on the unfiltered map with the Bsoft package, which showed the resolution was highest in the Kir.6.2/TMD0 core, as well as the SUR1 helices surrounding the GBC-binding pocket (between 3.3-3.7), and lowest in the NBDs and most external helices of TMD1/TMD2 of SUR1 ( Fig.1-figure supplement 2).
Model building. In our previous reconstruction, many side-chains were left out of the final model as there was not sufficient density to support their placement (Martin et al., 2017). In the current reconstruction, there is good density for nearly every side chain of Kir6.2, TMD0, and the inner helices of the ABC core structure of SUR1. Thus using our previous structure as the starting template, we rebuilt nearly all of the structure with RosettaCM (Song et al., 2013), using the density as an additional constraint. This region included Kir6.2, TMD0/L0, and TMD1 and TMD2. The lowest energy models were very similar to one another, thus the lowest energy model was selected for each region. The resulting model was then minimized once in CNS (Brunger et al., 1998), substituting in the RSRef real-space target function (Chapman et al., 2013), adding (φ,ψ) backbone torsion angle restraints, and imposing non-crystallographic symmetry (NCS) constraints. In the density map, NBD1 and NBD2 showed signs of disorder, so our previously deposited NBD models were left as polyalanine chains and only refined as rigid bodies with RSRef. The distal N-and C-termini of Kir6.2, as well as the linker between NBD1 and TMD2 in SUR1 were not observed in the density map, and thus were left out the model. The final model contains residues 32-352 for Kir6.2, and residues 6-615 (TMD0/L0 + TMD1), 678-744 and 770-928 (NBD1), 1000-1044 and 1061-1319 (TMD2), and 1343-1577 (NBD2) for SUR1. All structure figures were produced with UCSF Chimera (Pettersen et al., 2004) and PyMol (http://www.pymol.org). Pore radius calculations were performed with HOLE (Smart et al., 1996).

Sequence alignments
Multiple sequence alignment was performed with the T-Coffee server (Notredame et al., 2000). Output was saved in Clustal Aln format, and then imported and visualized in UCSF Chimera.
Functional studies of GBC binding mutants. Point mutations were introduced into hamster SUR1 cDNA in pECE using the QuikChange site-directed mutagenesis kit (Stratagene). Mutations were confirmed by DNA sequencing. Mutant SUR1 cDNAs and rat Kir6.2 in pcDNA1 were cotransfected into COS cells using FuGENE®6, as described previously (Devaraneni et al., 2015) and used for Western blotting, 86 Rb + efflux assays, and electrophysiology as described below.
For 86 Rb + efflux assays, cells were plated and transfected in 12-well plates. Twenty-four to thirty-six hours post-transfection, cells were incubated overnight in medium containing 86 RbCl (0.1 μCi/ml). The next day, cells were washed in Krebs-Ringer solution twice and incubated with metabolic inhibitors (2.5μg/ml oligomycin and 1mM 2-deoxy-D-glucose) in Krebs-Ringer solution for 30 min in the presence of 86 Rb + . Following two quick washes in Krebs-Ringer solutions containing metabolic inhibitors and 0.1% DMSO (vehicle control), 100nM GBC, or 1µM GBC, 0.5 ml of the same solution was added to each well. At the end of 2.5 minutes, efflux solution was collected for scintillation counting and new solution was added. The steps were repeated for 5, 7.5, 15, 25, and 40 min cumulative time points. After the 40 min time point efflux solution was collected, cells were lysed in Krebs-Ringer containing 1% SDS. 86 Rb + in the solution and the cell lysate was counted. The percentage efflux was calculated as the radioactivity in the efflux solution divided by the total activity from the solution and cell lysate, as described previously (Chen et al., 2013a;Yan et al., 2007). Note we used higher concentrations of GBC for these experiments than the electrophysiology experiments described below as in the latter the channels were exposed directly in isolated membrane patches to GBC, thus requiring lower concentrations. Experiments were repeated three-four times and for each experiment, untransfected cells were included as a negative control.
For electrophysiology experiments, cells co-transfected with SUR1 and Kir6.2 along with the cDNA for the green fluorescent protein GFP (to facilitate identification of transfected cells) were plated onto glass coverslips twenty-four hours after transfection and recordings made in the following two days. All experiments were performed at room temperature as previously described (Devaraneni et al., 2015). Micropipettes were pulled from non-heparinized Kimble glass (Fisher Scientific) on a horizontal puller (Sutter Instrument, Co., Novato, CA, USA). Electrode resistance was typically 1-2 MΩ when filled with K-INT solution containing 140 mM KCl, 10 mM K-HEPES, 1 mM K-EGTA, pH 7.3. ATP was added as the potassium salt. Inside-out patches of cells bathed in K-INT were voltage-clamped with an Axopatch 1D amplifier (Axon Inc., Foster City, CA). ATP (as the potassium salt) or GBC at 10nM or 100nM were added to K-INT as specified in the figure legend. All currents were measured at a membrane potential of -50 mV (pipette voltage = +50 mV). 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).

RMS deviations
Bond length 0.01 Bond angles 1.11         Electrostatic interaction between R301 and Q299, viewed from the interior of Kir6.2 and looking out toward the cytoplasm. These residues are found on an internal -sheet 12Å from ATP (C of R301 to ribose of ATP) (B) Dual interaction between R201 and F315, likely via cation-π, and hydrophobic stacking between aliphatic portion of R201 side chain and F333, again viewed from the Kir6.2 interior. These residues are found directly below ATP (~9Å from C of R201 to ribose of ATP). viewed from the extracellular side. Note juxtaposition of L0 to helices in ABC core directly interacting with GBC. (C) Close-up of GBC binding pocket, showing all residues which immediately line the pocket and seem to form direct contact with GBC; a subset of these residues were mutated to test their role in GBC binding (Fig. 7). (D) Magnified view in (B), highlighting indirect roles of Y230 and W232 (L0) in GBC binding. These both likely stabilize interaction between residues on helix 16 of TMD2 and GBC, at the same time anchoring this helix of L0 to the ABC core structure.   Examples of inside-out patch clamp recordings of WT and various mutant channels. Recordings were made at +50mV in symmetric K + solutions and inward currents shown as upward deflections. (B) Quantification of residual currents (expressed as percent of initial currents observed in K-INT solution) after exposure to 10nM or 100nM GBC, with values taken when the currents reached a steady level. The value for "recovery after washout" was taken at ~100 seconds after the patch was returned to K-INT solution following a brief exposure to 1mM ATP (to check baseline). Each bar represents mean±s.e.m. of 3-6 patches. * p<0.05 by one-way ANOVA with Newman-Keuls post hoc test. Hypothetical model of GBC gating. GBC binds to the TM bundle juxtaposing L0, which stabilizes the SUR1-ABC core in an inward-facing conformation. GBC also stabilizes interactions of the distal Nterminus of Kir6.2 with SUR1 thus restricting movement of the Kir6.2 N-terminus to inhibit channel activity. In both A and B, Kir6.2 transmembrane helices: dark green; Kir6.2 cytoplasmic domain: green; Kir6.2 slide helix and N-terminus from adjacent subunit: light green cylinder and thick blue line, respectively; SUR1-TMD0/L0: magenta; SUR1-TMD1/2: blue; SUR1-NBDs: orange; GBC: yellow; ATP: red; PIP2: cerulean. Note the different states shown are not meant to reflect the actual conformational transitions. Also, PIP2 was not included in the closed states; however, it remains to be determined whether ATP precludes PIP2 binding.