We expressed the SpTx1 toxin recombinantly in E. coli and purified it to apparent homogeneity (Fig. S1A). Whole-cell patch clamp experiments demonstrated that purified SpTx1 can inhibit currents of the K_ATP channel formed by human hsKir6.2 and SUR1 from Mesocricetus auratus (maSUR1), indicating that the purified toxin is functional (Fig. 1A and B). We then crystallized SpTx1 and determined its crystal structure via x-ray crystallography at a resolution of 1.9 Å (Fig. 1C to E and Table S1). In 1 asymmetric unit, there are 6 SpTx1 monomers packed in a head-to-tail fashion, forming a hexametric ring (Fig. S1B). The formation of this hexametric assembly is likely due to the crystal packing effect, because SpTx1 migrates as a monomer on the size-exclusion chromatography (Fig. S1C). The structures of each monomer are highly similar (Fig. S1B and D). SpTx1 toxin exhibits a spearhead-shaped structure with an α-helix packing against 3 antiparallel β-sheets (Fig. 1C). Both the N-terminus and C-terminus of SpTx1 are located on the same side, at the tail of the spearhead (Fig. 1C). On the opposite side, a positively charged K15 is located at the tip of the spearhead (Fig. 1C). Two disulfide bonds are formed between C22 on α1 and C48 on β3, and between C26 on α1 and C50 on β3, further stabilizing the overall compact structure of SpTx1 (Fig. 1C). These 2 disulfide bonds were absolutely conserved in other centipede toxins that show inhibitory effect against hsKir6.2 [15,16] (Fig. 1F). Notably, the surface of SpTx1 is highly hydrophilic and rich in positively charged patches (Fig. 1D and E), suggesting that polar interactions might play a crucial role in the interactions between SpTx1 and hsKir6.2. Additionally, the overall structure of SpTx1 and the position of the Lys at the tip are also similar to another centipede toxin, SsTx from Scolopendra subspinipes mutilans, which specifically inhibits the KCNQ channel [17] (Fig. S1E and F).

To understand how SpTx1 binds the K_ATP channel, we sought to determine the cryo-EM structure of their complex. To express the K_ATP channel for structural studies with SpTx1, we fused hsKir6.2 to the C-terminus of maSUR1 and incorporated an PreScission Protease cleavage site upstream of hsKir6.2 (Fig. S2A). This design ensured the 4:4 stoichiometry between hsKir6.2 and maSUR1 and facilitated the correct assembly of the channel [7,8]. Additionally, cleavage of the maSUR1-hsKir6.2 fusion protein by PreScission Protease released the Kir6.2 N-terminal peptide (KNtp), which is crucial for conferring sensitivity to insulin secretagogues such as glibenclamide [7,8]. We also introduced mutations Q52E in hsKir6.2 and E203K in maSUR1 to enhance the ATP sensitivity of the K_ATP channel [18] (Fig. S2A). The resulting K_ATP fusion construct retains the sensitivity to SpTx1 inhibition (Fig. S2B), the same as wild-type channel (Fig. 1A). Following PreScission Protease cleavage of the expressed K_ATP fusion protein and fractionation on size-exclusion chromatography (Fig. S2C), we supplemented the protein with purified SpTx1 toxin for cryo-EM sample preparation. We also supplemented the channel protein with ATP and glibenclamide to stabilize the channel in the ATP-bound and glibenclamide-bound closed conformation to obtain a high-resolution structure (Fig. S3). During the cryo-EM image analysis, consensus refinement using C4 symmetry yielded a map with an overall resolution of 2.74 Å (Fig. S3); however, the SUR1 ABC transporter region (SUR1_ABC) appeared blurry, confirming its dynamic interface with the central Kir6.2 and SUR1 TMD0 region (K_{ATP CORE}) [12]. Additional focused refinement on SUR_ABC with C1 symmetry improved the resolution to 2.87 Å (Fig. S3). In the map of SUR_ABC, we observed the density of glibenclamide in the transmembrane domain, ATP on nucleotide-binding domain 1, and KNtp in the central cavity of maSUR1 (Fig. S4). In the map of K_{ATP CORE}, we identified toxin density at the extracellular mouth of the Kir6.2 channel (Fig. S3), which was absent in previous K_ATP maps with hsKir6.2 [19,20]. However, due to the C4 averaging, the density of SpTx1 remained uninterpretable. Further symmetry expansion and focused classification with C1 symmetry on SpTx1 resulted in 2 distinct 3-dimensional classes (Fig. S3): one with no SpTx1 bound and another exhibiting prominent SpTx1 density. The class with SpTx1 bound was refined to a resolution of 2.88 Å with C1 symmetry and was subsequently merged with the SUR1_ABC maps to generate the composite map for model building and interpretation (Fig. 2, Figs. S3 and S4, and Table S2). In the map, we observed all of the ligands added during sample preparation, including ATP bound in the Kir6.2 subunit, ATP bound in the SUR1 subunit (Fig. 2A), and glibenclamide bound in the SUR1 subunit (Fig. 2B). We also observed discontinuous density of KNtp in the central cavity of SUR1 (Fig. S4B and C), in agreement with its dynamic nature.

The structure reveals that SpTx1 binds to the highly negatively charged surface on the extracellular side of the K_ATP channel (Fig. 3A to C and Fig. S4) and primarily engages with hsKir6.2 subunits, in agreement with previous data showing that SpTx1 potently inhibits Kir6.2-ΔC26 [13]. SpTx1 positions its positively charged K15 at its tip into the extracellular mouth of the hsKir6.2 pore (Fig. 3B to D and Movie S1). The primary amine group of K15 is situated on the 4-fold symmetry axis and near the S1 site of the pore of hsKir6.2, where it interacts with the 4 backbone carbonyl groups of F133 located at the selectivity filter of hsKir6.2 (Fig. 3E and Fig. S4E). Surrounding K15 are 3 phenylalanines—F16 on the β1-α1 loop and F45 and F46 on the β2-β3 loop of SpTx1—which engage with the carbonyl groups of G134 at the selectivity filter of hsKir6.2 via the edges of their benzene rings and also form hydrophobic interactions with residues at the pore (Fig. 3E and Fig. S4F). Particularly, both F16 and F45 form hydrophobic interactions with M137 of 2 adjacent hsKir6.2 subunits (Fig. 3E and Fig. S4G). Additionally, R11 on the β1-α1 loop of SpTx1 forms 1 hydrogen bond with the carbonyl group of G135 at the mouth of the Kir6.2 pore (Fig. 3F and S4H). Meanwhile, R18 on the β1-α1 loop of SpTx1 interacts with E108 on the S2-S3 turret of 1 hsKir6.2 subunit (Fig. 3G and Fig. S4I), and R13 on the same loop further interacts with both E108 and the carbonyl group of G105 on the turret between inner helix and outer helix (OH) of the adjacent hsKir6.2 subunit (Fig. 3F and Fig. S4J). To investigate the functional significance of these hydrophilic interactions, we mutated the relevant charged residues to alanines and measured the inhibitory effects of 1 μM SpTx1 mutant proteins, which is close to the saturated concentration of wild-type SpTx1 (Fig. S5). We found that the K15A mutant of SpTx1 exhibited a great loss of inhibition of the K_ATP channel (Fig. 3H). Similarly, the R11A, R13A, and R18A mutations in SpTx1 all reduced its inhibitory effect to some extent (Fig. 3H). Additionally, the E108A mutation in hsKir6.2 conferred resistance of the K_ATP channel to SpTx1 inhibition (Fig. 3H). These findings confirm the crucial roles of these polar interactions and are largely consistent with previous mutagenesis results [14,15]. To understand the mechanism of high selectivity of SpTx1, we performed the sequence alignment of all Kir proteins as well as other potassium channel proteins such as KCNQ2, HCN4, and KCNH2 of Homo sapiens. We found that only hsKir6.2 has glutamic acid at 108 position (Fig. S4N), in agreement with the high selectivity of SpTx1 toward Kir6.2.

K_ATP channel conducts potassium currents by providing an ion permeation pathway through its pore formed by 4 Kir6.2 subunits (Fig. 4A). The structure of the Kir6.2 pore is highly similar to that of other potassium channels, exemplified by the KcsA [21]. In the high-resolution crystal structure of KcsA in high potassium condition, potassium ions bind at both the extracellular mouth and 4 positions (S1 to S4) within the selectivity filter [21] (Fig. 4B). However, in the structure of K_ATP in complex with SpTx1, SpTx1 physically occupies not only the potassium binding site at the extracellular mouth (S0 site) but also the S1 position in the selectivity filter, precluding the binding of potassium ions at these sites (Fig. 4A). Moreover, SpTx1 also occludes the ion permeation pathway of hsKir6.2 (Fig. 4A to D), preventing ion permeation (Fig. 4D). In the structure, the gate of hsKir6.2 is tightly closed (Fig. 4D), in agreement with the binding of inhibitory ATP (Fig. 2D). Furthermore, we found that there is little conformational change in the extracellular regions of Kir6.2 channel upon channel opening or the binding of SpTx1 (Fig. S4K to M). Given the fact that SpTx1 can block the open hsKir6.2 (Fig. 1A), we speculate that SpTx1 also interacts with hsKir6.2 in the open state as observed here.

It is previously found that the potassium channel blocker charybdotoxin (CTX), derived from scorpions, also exhibits the α/β defensin fold, characterized by 3 disulfide bonds. But the structure of CTX is largely different from SpTx1, especially the position of the helix in relation to the rest of the protein (Fig. S1G). Nuclear magnetic resonance studies [22] and crystallographic studies [23] have shown that CTX also binds to the extracellular mouth of the potassium channel pore, with a lysine side chain obstructing the ion permeation pathway (Fig. 4F and G), using a mechanism similar to the SpTx1 observed here. However, the orientation of CTX and detailed interactions with potassium are largely different compared to SpTx1 with Kir6.2 (Fig. 4E to G).

Based on our structural and functional analysis (Fig. 3), we identified key residues within the SpTx1 toxin that are essential for inhibiting hsKir6.2 (Fig. 1E). Using the SpTx1 sequence as a query, we performed a BLASTp search against the NCBI ClusteredNR database. Among the resulting hits, the toxin Sm3a from the centipede Scolopendra morsitans ranked top. Although Sm3a shares only 56% sequence identity with SpTx1, all key residues critical for Kir6.2 inhibition are highly conserved in Sm3a (Fig. 5A). To assess whether Sm3a also inhibits K_ATP channels, we recombinantly expressed and purified the toxin. Whole-cell patch-clamp recordings demonstrated that Sm3a strongly blocks currents from K_ATP channels composed of hsKir6.2 and maSUR1 subunits (Fig. 5B). Additionally, we compared the inhibitory effect of Sm3a with that of SpTx1 using 200 nM concentration, which is below the half-maximal inhibitory concentration of wild-type SpTx1. We found that the application of 200 nM Sm3a inhibited K_ATP channels to a similar extent as 200 nM SpTx1 (Fig. 5C), indicating comparable inhibitory potency. Given the substantial sequence similarity between SpTx1 and Sm3a, we propose that Sm3a binds to and inhibits hsKir6.2-containing K_ATP channels in a manner similar to SpTx1.

The coding sequences of the matured form of SpTx1 and Sm3a (P0DQC2.1) were synthesized and inserted into a pET-based N-terminally His-tagged vector (pPH) with tobacco etch virus protease cleavage site to generate the toxin expression vectors.

The coding sequences of maSUR1 (E203K) and hsKir6.2 (Q52E) are linked by a long linker including a PreScission protease cleavage site (​VDG​SGS​GSG​SAA​GSG​SGS​GSG​SGA​AGS​GSG​SGS​GSG​AAALELEVLFQGP). The initial methionine of hsKir6.2 after the linker was removed. The open reading frame of fusion protein was inserted into a modified BacMam vector with C-terminal green fluorescent protein and 2 Strep tags [28].

The expression vector for the toxins were transformed into Rosetta-gami 2(DE3) competent cells. Protein expression was induced by adding 1 mM isopropyl β-D-1-thiogalactopyranoside for 3 h at 37 °C in Luria–Bertani medium. Collected bacterial cells were resuspended in 20 mM tris (pH 7.5), 150 mM NaCl, 1 mM phenylmethylsulfonyl fluoride and were disrupted via sonication. The lysate was centrifuged at 30,000g for 1.5 h, after which the supernatant was applied to a Talon affinity column and eluted using 250 mM imidazole. The eluted protein was subjected to tobacco etch virus protease digestion to remove the purification tag. After cleavage, contaminant proteins were precipitated using methanol, yielding highly purified toxin protein, which was subsequently purified via SP ion-exchange chromatography to remove residual methanol. Final purification was performed using a Superdex 75 size-exclusion column running in 10 mM tris (pH 7.5) and 150 mM NaCl. Protein was concentrated to 20 mg/ml for crystallization.

Protein was mixed with 0.1 M succinic acid (pH 5.5), 0.05 M ammonium sulfate, and 30% (v/v) pentaerythritol ethoxylate (15/4 EO/OH) with a 1/1 ratio at room temperature in a hanging drop crystallization setup. Crystal was harvested using 0.1 M succinic acid (pH 5.5), 0.05 M ammonium sulfate, 30% (v/v) pentaerythritol ethoxylate (15/4 EO/OH), and 10% ethylene glycol as cryoprotectant and flash frozen in liquid nitrogen.

Native datasets of SpTx1 were collected at Shanghai Synchrotron Radiation Facility beamline BL17U at 0.97918 Å. Diffraction images were processed using HKL3000 [29] and converted to a mtz file using CCP4 [30]. SpTx1 structure predicted by AlphaFold2 [31] was used as the search model for molecular replacement using Phaser [32]. The model was rebuilt manually using COOT [33] and was refined using Phenix [34]. The final model contains 99.68% favored and 0.32% allowed residues but no outliers on the Ramachandran plot.

K_ATP channels were expressed as described previously, and the purification process was carried out with minor modification [7,35]. For protein purification, membrane pellets were homogenized in tris buffer saline (TBS, 20 mM tris-HCl [pH 7.5], and 150 mM NaCl) and solubilized in TBS with 1% digitonin (Calbiochem), supplemented with protease inhibitors (1 mg/ml leupeptin, 1 mg/ml pepstatin, 1 mg/ml aprotinin, and 1 mM phenylmethylsulfonyl fluoride), 5 μM glibenclamide, 1 mM ATP, 1 mM EDTA, 2.5 mM MgCl₂, and 2.5 mM CaCl₂, for 30 min at 4 °C. Insoluble materials were removed after centrifugation at 100,000g for 30 min, and the supernatant was loaded onto a 5-ml column packed with Streptactin 4FF resin (Smart Lifesciences). Strep column was first washed by buffer A (5 μM glibenclimide, 1 mM ATP, 1 mM EDTA, and 0.1% digitonin in TBS), then 1 mM ATP and 2.5 mM MgCl₂ in buffer A, and finally in buffer A again. K_ATP protein was eluted with 10 mM desthiobiotin in buffer A. Protein was then digested with PreScission protease overnight, concentrated, and purified using a Superose 6 column running in buffer A. Peak fractions were collected and concentrated to A₂₈₀ (absorbance at 280 nm) = 15 (estimated as 15 μM K_ATP octamers) for cryo-EM sample preparation.

K_ATP octamers were supplemented with 200 μM glibenclimide and 1 mM ATP, along with 100 μM SpTx1. The final K_ATP protein concentration was estimated to be 9 μM octamer in solution. Cryo-EM grids were prepared with GIG R1/1 holey carbon grids, which were glow-discharged for 120 s. A 2.5-μl K_ATP octamer sample was applied to the glow-discharged grid, and then the grid was blotted at blotting force in level 2 for 2 s at 100% humidity and 20 °C, before plunge-frozen into the liquid ethane.

Cryo-EM data collection was performed on a Titan Krios microscope (Thermo Fisher Scientific) operating at 300 kV. Images were acquired using a Gatan K2 direct electron detector mounted post a quantum energy filter with a 20-eV slit width. Super-resolution mode was employed with a calibrated pixel size of 1.324 Å at the specimen level, and data acquisition was controlled via SerialEM software. Defocus values were systematically varied between −1.3 and −1.8 μm to optimize contrast transfer. A dose rate of 8 e⁻/s/Ų was applied, with a cumulative electron exposure of 50 e⁻/Ų. Each 12-s exposure was dose-fractionated into 50 frames to facilitate motion correction during image processing.

A total of 5,655 movies were corrected for beam-induced motion, dose-weighted, and 2-fold binned to a pixel size of 1.324 Å using MotionCor2 [36]. The contrast transfer function parameters of micrographs were estimated using patch contrast transfer function in cryoSPARC [37]. Iterative 2-dimensional and 3-dimensional classification generated 548,390 particles with 256 box size which were refined to 2.74-Å overall resolution using nonuniform refinement function in cryoSPARC [37]. Symmetry expansion, particle subtraction, and local refinement of the SUR1_ABC regions yielded a local map of 2.87-Å resolution. Symmetry expansion and iterative focused classification of SpTx1 region using C1 symmetry resolved 2 classes: One has SpTx1 bound and the other has no SpTx1 bound. The class with SpTx1 bound was locally refined using C1 symmetry focusing on the K_{ATP CORE} region to obtain a map at 2.88-Å resolution. This map of K_{ATP CORE} was combined with the map of SUR1_ABC to obtain a composite map for model building and interpretation. The structures of SpTx1 and K_ATP channel in complex with repaglinide (Protein Data Bank [PDB] ID: 6JB1) were docked into the composite map. The model was rebuilt manually using COOT [33] and was refined using Phenix [34].

K_ATP constructs were transfected into FreeStyle 293-F cells using polyethylenimine at a cell density of 1 × 10⁶ cells/ml. Cells were cultured in FreeStyle 293 Expression Medium with 1% fetal bovine serum for 24 h before recording. Currents were recorded using whole-cell mode at −40 mV in the pipette and an Axon-patch 200B amplifier (Axon Instruments, USA). Patch electrodes were pulled by a horizontal microelectrode puller (P-1000, Sutter Instrument Co, USA) to tip resistance of 2.0 to 3.0 MΩ. Pipette solution containing 107 mM KCl, 1.2 mM MgCl₂, 1 mM CaCl₂, 10 mM EGTA, 5 mM Hepes (pH 7.2, KOH), and bath solution containing 40 mM KCl, 100 mM NaCl, 2.6 mM CaCl₂, 1.2 mM MgCl₂, and 5 mM Hepes (pH 7.4, NaOH) were used for measuring inhibitory effect of wild-type and mutant toxins. Application of 5 mM BaCl₂ in bath solution was used to block the K_ATP currents to determine the baseline. Data analysis was carried out using GraphPad Prism 8.0 or 10.4.0.

Global resolution estimations of cryo-EM density maps are based on the 0.143 Fourier Shell Correlation criterion [38]. The local resolution was estimated using cryoSPARC. The number of independent experiments (N) and the relevant statistical parameters for each experiment (such as mean or SD) are described in the figure legends. No statistical methods were used to predetermine sample sizes.