Insights into the structure and modulation of human TWIK-2

Abstract

The Tandem of pore domain in a Weak Inward Rectifying K⁺ channel 2 (TWIK-2; KCNK6) is a member of the Two-Pore Domain K⁺ (K_2P) channel family, which is associated with pulmonary hypertension, lung injury, and inflammation. Despite its physiological relevance, the structure, regulatory mechanisms, and selective modulators of TWIK-2 remain largely unknown. Here, we present a 3.7 Å single particle cryo-electron microscopy structure of human TWIK-2 and highlight its conserved and distinctive features. Using automated whole-cell patch clamp recordings, we demonstrate that gating in TWIK-2 is voltage-dependent and insensitive to changes in the extracellular pH. We identify key residues that influence TWIK-2 activity by employing site-directed mutagenesis and provide insights into the possible lipid-mediated mechanism of TWIK-2 regulation. Additionally, we demonstrate the application of high-throughput automated whole-cell patch clamp platforms to screen small molecule modulators of TWIK-2. Our work serves as a foundation for designing high-throughput small molecule screening campaigns to identify specific high-affinity TWIK-2 modulators, including promising- anti-inflammatory therapeutics.

TWIK-2 is an endolysosomal potassium channel implicated in inflammatory responses. Here, authors present a cryo-EM structure of human TWIK-2 and establish a high-throughput automated patch-clamp electrophysiology assay to investigate modulation of TWIK-2.

Introduction

Two-pore-domain K⁺ (K_2P) channels are a diverse family of potassium (K⁺) selective ion channels that supervise background K⁺ currents. They are critical to maintaining membrane resting potential and regulating cellular excitability^1–3. In mammals, the K_2P family comprises 15 members, classified into six subgroups based on their functional diversity: TWIK, TREK, TRESK, TASK, TALK, and THIK^4,5. Unlike Voltage-gated K⁺(Kv) channels, K_2P channels lack the canonical voltage-sensing domain. While the overall architecture of K_2P channels is evolutionarily conserved, there are multiple structural variations and sites of modulation, which make these channels susceptible to regulation by voltage-independent factors such as temperature, pH, pressure, bioactive lipids, and volatile anesthetics^5–7.

Among the 15 members of the K_2P family, channels from TREK and TASK subfamilies have been extensively structurally characterized^8–13, offering profound insights into their gating mechanisms, physiological functions, and pharmacological potential^8,14–24. TREK channels are known for their mechano- and thermosensitivity and role in sensory response^14,18,25,26, and TASK channels are well-characterized for their pH sensitivity and role in maintaining neuronal excitability^{12,13,27–29}. Nevertheless, the gating mechanisms, physiological modulation, and pharmacological potential of the TWIK family remain underexplored primarily because of the weak currents^30,31 and poor heterologous expression of these channels^32–34.

TWIK-1 (KCNK1) is the only structurally characterized mammalian K_2P channel member of the TWIK family^30,35. The poor currents in TWIK-1 were initially believed to be due to intracellular localization resulting from SUMOylation³⁶. However, mutations in the SUMOylation site and overexpression of a recombinant plasma-membrane-localized TWIK-1 demonstrated that the poor currents are characteristic of TWIK family members^37,38. Recent studies involving a combination of electrophysiology, site-directed mutagenesis, molecular dynamics simulations, and structural analysis of TWIK-1 at low (pH 5.5) and neutral pH (pH 7.4) revealed the molecular basis for pH-mediated regulation of gating and conformational dynamics of the selectivity filter in TWIK-1^39,40.

TWIK-2 (KCNK6) shares ~54% sequence similarity with TWIK-1 and exhibits ubiquitous tissue distribution^31,41. TWIK-2 is implicated in diverse pathological conditions such as pulmonary hypertension, acute lung injury, hearing loss, and NLRP3 inflammasome-induced inflammation^42–46. A significant degree of variation in the current recordings has been reported for TWIK-1 and TWIK-2 by various groups, underscoring the challenges of robust electrophysiological studies for the TWIK subfamily^31,33,41. The weak basal activity of TWIK-2 could be due to its subcellular partitioning between the plasma membrane and endolysosomes. Moreover, it has been reported that the Y308A mutation within the lysosomal-targeting YXXØ motif in TWIK-2 promotes plasma membrane expression of TWIK-2³⁴. Translocation of TWIK-2 from endosome to plasma membrane is implicated in NLRP3-mediated inflammasome activation in macrophages, with KCNK6 knockouts showing a suppressed inflammasome activation^45,46. Despite the high sequence similarity with TWIK-1, TWIK-2 is believed to be insensitive to pH and indifferent to activation by phosphorylation^30,31,38,40. The mechanism of regulation of TWIK-2 is poorly understood because of the lack of three-dimensional structures and due to inconsistent reports of its electrophysiological recordings^31,34,41. Additionally, although substantial progress has been made in developing K_2P channel modulators, particularly for the well-characterized TREK subfamily, there is a dearth of high-affinity, selective modulators of TWIK-2. The existing small molecule modulators of TWIK-2 activity, such as ML365 and NPBA, also cross-react with other K_2P channels^25,47–53.

To gain deeper insights into the molecular basis of channel regulation and to aid the pharmacological characterization of TWIK-2, we determine the structure of full-length human TWIK-2 by single particle cryo-electron microscopy (cryo-EM). We uncover conserved and distinctive features of this K_2P channel subtype and, using a combination of structure-guided site-directed mutagenesis and automated high-throughput whole-cell patch clamp electrophysiology, we identify key amino acids that regulate TWIK-2 channel activity. Finally, using the existing K_2P modulators, we develop an effective strategy for high-throughput screening of TWIK-2 modulators.

Results

Heterologous expression and functional characterization of recombinant human TWIK-2

We expressed full-length TWIK-2 (hereafter TWIK-2) in HEK293 GnTI⁻ and verified that it was functional by whole-cell patch-clamp electrophysiology (top panel of Fig. 1A, B, Supplementary Fig. 1, and Supplementary Table 1). When recorded in high potassium (100 mM) at pH 7.4, the TWIK-2 channels evoked larger outward potassium currents than the non-transfected (non-T) cells (Supplementary Fig. 1 and Supplementary Table 2). TWIK-1 is sensitive to pH^39,40. To assess whether TWIK-2 behaves similarly, we measured conductance at two pH levels (7.4 and 5.5) in high potassium (top panels of Fig. 1A, B, and Supplementary Fig. 1B, C). We did not observe significant changes in current densities or reversal voltage potential (V_rev) of TWIK-2 at pH 7.4 and pH 5.5 under these conditions (p = 0.9003; p = 0.1017), indicating that TWIK-2 is insensitive to changes in extracellular pH. To determine whether pH independence of channel activity is an inherent property of TWIK-2, we introduced a mutation at residue-equivalent position 111, known to affect pH dependence of TWIK-1 channels. In this position TWIK-2 has a tyrosine (Y) instead of a histidine (H) seen in TWIK-1 (Fig. 1C and Supplementary Fig. 2B). Functional analysis of the Y111H mutant at pH 7.4 and pH 5.5 showed no changes in current densities (p = 0.5142) or voltage reversal potential (V_rev) (p = 0.1394) related to pH dependence (bottom panels of Fig. 1A, B, and Supplementary Table 1). This result shows that TWIK-2 channels are pH-insensitive, indicating that the structural elements responsible for pH sensitivity in TWIK-1 differ from those in TWIK-2. To further explore structural differences within the TWIK family, we prepared a full-length human TWIK-2 construct for overexpression and structural analysis. We purified TWIK-2 using Decyl maltose neopentyl glycol (DMNG) at pH 7.5 and 150 mM KCl. We confirmed the protein homogeneity and purity by size-exclusion chromatography and SDS-PAGE analysis (Supplementary Fig. 3B). The peak fractions, corresponding to TWIK-2 dimer, that eluted at ~12.4 ml on a Superdex 200 Increase size-exclusion chromatography column were used for subsequent structural analysis (Supplementary Fig. 3B–F).

Fig. 1. Function and structure of TWIK-2.

A Functional analysis of TWIK-2 channel and Y111H mutant channel at various pH levels. Representative whole-cell current traces were recorded from cells expressing both wild-type TWIK-2 channels and the Y111H mutant channels. These traces were obtained using voltage-clamp steps that ranged from −90 to +100 mV in 10 mV increments, with pH levels set at 7.4 and 5.5. B A comparative analysis of peak current densities (pA/pF) at +100 mV and the reversal potential (V_Rev) was performed for TWIK-2 (n = 32; n = 35, respectively) and Y111H (n = 24; n = 28, respectively) at pH 7.4, as well as for TWIK-2 (n = 38; n = 45, respectively) and Y111H (n = 33; n = 54, respectively) at pH 5.5. The current–voltage (I–V) relationships recorded from both TWIK-2 and Y111H channels showed no significant differences across the voltage range of −90 to +100 mV, including at the reversal potential (VRev). The data in (B) are presented as mean ± standard deviation (SD), with some error bars smaller than the data points. Statistical significance was assessed by an unpaired t-test with Welch’s correction. C Topology and model of TWIK-2 protomer showing transmembrane helices (M1–M4), cap-forming helices (CH1 and CH2), pore-forming helices (PH1 and PH2), selectivity filters (SF1 and SF2). Critical residues along the ion-permeation pathway are highlighted. D Cryo-EM map and model of TWIK-2 dimer. The protomers are colored light magenta and light cyan. DMNG micelle density is transparent. Cartoon representation of TWIK-2 in an en face view parallel to the membrane. K⁺ ions are highlighted as blue spheres, and the cap domain (Cap) and extracellular ion pathway (EIP) are marked. E Intracellular view of cryo-EM map (top) and model of TWIK-2 (bottom) with pore domains (PD1: orange and PD2: blue). Note M1s in PD1s are domain-swapped.

Molecular architecture of TWIK-2

We determined the structure of TWIK-2 at ~3.7 Å (Fig. 1D, E; Supplementary Fig. 3C–F, and Supplementary Table 3). We were able to reliably model all the secondary structure elements of TWIK-2 except the following disordered-loop regions: M1-G4 (N-terminus), V75-P89, T150-W171, and L262-R313 (C-terminus) (Fig. 1C–E, and Supplementary Fig. 3F–H). Notably, the V75-P89 loop region harbors two predicted N-linked glycosylation sites, N79 and N85 (Supplementary Fig. 2A). For the overexpression of homogenous TWIK-2, we used HEK293 GnTI⁻ cells that lack N-acetylglucosaminyltransferase I activity, resulting in a minimally glycosylated recombinant protein. Perhaps complex native glycosylation at these sites is necessary for stabilization of this loop region. Each TWIK-2 protomer contains four transmembrane helices (M1–M4), two pore-forming helices (PH1 and PH2), two selectivity filter loops (SF1 and SF2), and two extracellular cap-forming helices (CH1 and CH2) (Fig. 1C). TWIK-2 assembles as a canonical domain-swapped homodimer with a pseudo-tetrameric central pore observed in all other K_2P channels. The cap-forming helices within a protomer are arranged in an inverted “V” shaped cap domain over the central pore of the channel. The cap domains from both protomers form an “arched dome,” creating a bifurcated extracellular ion pathway (EIP, Fig. 1C–E). As observed in other known K_2P structures, we notice that the cap domain is also responsible for pairing the M1 helix of one protomer in a three-dimensional module with the rest of the protein from the other protomer. Thus, the domain-swapping of the M1 helix leads to the formation of four pore-forming domains comprised of two PD1s that are non-identical to the two PD2s. PD1s are composed of the M1 helix from one protomer and PH1, SF1, and M2 from the other protomer. PD2s, on the other hand, are assembled by M3, PH2, SF2, and M4 of the same protomer (Fig. 1D, E). Three-dimensionally, each pore-forming domain consists of one pore-forming helix (PH) and one selectivity filter (SF) loop flanked by two transmembrane helices (Fig. 1C). The PH and SF regions from PD1 (PH1/SF1) and PD2 (PH2/SF2) of both protomers form the central ion-conducting pore. Our cryo-EM density map allows placement of three K⁺ ions in the central selectivity filter (Fig. 1D). M2 and M4 helices are positioned adjacent to the core, while M1 and M3 are located on the periphery. The domain-swapped M1 helices, the non-identical sequences of the SFs, and the unequal lengths of the SF1-M2 and SF2-M4 linkers, impart a pseudo-hetero-tetrameric symmetry to the elements surrounding the pore of the channel (Fig. 1E).

Ion conduction pore of TWIK-2

The ion conduction pathway of TWIK-2 extends from the hydrophobic cuff marked by a kink in the M2 helix within the vestibule, designed to funnel ions toward the central pore, to the bifurcated EIP on the extracellular side through a narrow selectivity filter (Fig. 2A, B). While the sequence within the selectivity filters of TWIK-2 is conserved mainly across different K_2P channels, the entry and exit of the ion conductance pathway are flanked by non-conserved M135 and Y111, respectively (Fig. 2A and Supplementary Fig. 2). We found that the narrowest constriction in the ion conductance pathway has a radius of 4.1 Å, which is sufficiently wide for K⁺ ions to pass through (Fig. 2C, D). Except for the two conserved threonines at the bottom of the SF, the backbone carbonyl oxygens regulate the flow of K+ ions within the SF, as opposed to side chains. While the SF sequences are primarily conserved across the K_2P family, TWIK subfamily members show some divergence. The canonical isoleucine (TIGY/FG) in SF1 is replaced by threonine in TWIK-1 and valine in TWIK-2. Similarly, the canonical phenylalanine (TI/VGFG) in SF2 is replaced by leucine in both TWIK-1 and TWIK-2 (Supplementary Fig. 2A). These variations could potentially govern the differences in the currents of the TWIK subfamily compared to other highly modulated K_2P channels, such as TREK and TASK.

Fig. 2. Ion conductance pathway of TWIK-2.

A A cross-section of TWIK-2 along the membrane plane, with hydrophobic and hydrophilic amino acids colored in orange and cyan, respectively. Critical amino acids flanking the entry and exit of the ion permeation pore, at the hydrophobic cuff and EIP, respectively, are highlighted. B Pore radius of the TWIK-2 channels without or with hydrocarbon tail (yellow) modeled as distance along the ion permeation pathway calculated by HOLE⁸⁰, shown in light blue (left) or light gray (right). C Cryo-EM density for the selectivity filters and K⁺ ions is highlighted in gray. The green (top) and black (bottom) dashed lines indicate the widest and narrowest pore diameters of the SF. D The cross-section views of the narrowest and widest planes of the SF as viewed from the extracellular cap domain. The distance between the amino acids indicates the pore diameters at that position. E TWIK-2 with lipid-like tubular density (yellow) and tetrahedral solvent or ion blobs (blue) highlighted. The center panel highlights the presence of lipid-like densities along the groove formed by M1, M4, and PH2 of one protomer with M2 of the other protomer.

Unidentified non-protein molecules in TWIK-2

In our cryo-EM sample preparation, we did not include any exogenous lipids or small-molecule ligands that bind to TWIK-2, with the intention of determining the structure of apo-TWIK-2. However, we observed tubular lipid-like densities below the SF in the canonical K_2P vestibule, the modulatory lipid site, and fenestration sites, which could be relevant to the mechanism of TWIK-2 action (Fig. 2E). Endogenous lipids are known to co-purify with the protein in the detergent micelle^54–56. While these tubular densities could not be assigned to the protein, the resolution of our cryo-EM map precludes their unambiguous identification. Alternatively, these densities could also be DMNG detergent. But we do not observe densities that can accommodate the DMNG head group. Hence, we modeled hydrocarbon tails in the tubular densities (Fig. 2E). We note that the tubular lipid-like densities occupy the groove created by the PH2 and M4 helices of one protomer with the M2 helix of the other. The lower end of this groove is believed to be a “classic” modulatory phospholipid binding site, and the site that opens to the bottom of SF has been termed a “fenestration” site⁵. It is surmised that the binding of lipids in this groove pushes the M4 helix down, allowing better access to SF. In a “down” conformation, the M4 helix traverses the lipid bilayer at a ~45° angle, creating a hydrophobic “fenestration site” between the bottom of the PH2 helix and M2 helix of the opposite protomer that enables binding of hydrophobic compounds at this pocket that inhibit the channel^5,57 (Supplementary Fig. 4A). In TWIK-2, we observe that M4 adopts a canonical “down” conformation, perhaps due to the presence of tubular densities in the “fenestration site” (Supplementary Fig. 4A). In addition, we also noticed three small tetrahedral blobs within the central cavity below the permeation pore that could accommodate an ion or water (blue, Fig. 2E). Molecular dynamics (MD) simulations of K_2P channels treated with negatively charged activators have suggested the presence of ions or water within the “cavity binding sites,” stabilized by the negative moiety of the small molecule or lipids⁵⁸. We have left these small tetrahedral densities unmodeled in our TWIK-2 structure because the precise identity of these densities remains unclear.

Automated patch-clamp recordings of variants of recombinant human TWIK-2

We observed that TWIK-2 channel activity was dependent on the applied voltage but not on the extracellular pH (Fig. 1A, B). Although TWIK-2 and TWIK-1 exhibit high structural similarity, including conserved residues in the SF and M2 domains, subtle differences in residue interactions may lead to distinct gating behaviors in the two channels. To identify and assess molecular determinants critical to K⁺ conductance in TWIK-2, we performed site-directed mutagenesis along the ion permeation pathway, near the hydrophobic cuff, and near the canonical K2P modulator pocket (Fig. 3A). Specifically, we generated alanine substitutions of F98 (PH1), T106 (SF1), Y111 (SF1-M2 loop), M135 (hydrophobic cuff), S213 (PH2), and T214 (SF2). We also introduced a charge into the hydrophobic cuff using the M135N mutation. Lastly, we generated the Y111H substitution in TWIK-2 to make it “TWIK-1-like,” as H122 in TWIK-1 (H98 in TASK-3) is believed to be a major pH-sensing amino acid^39,40.

Fig. 3. Effects of mutants along the ion permeation pathway on TWIK-2 conductance and channel activation.

A The residues in TWIK-2 along the ion-permeation pore (light blue) used for subsequent analysis are highlighted in yellow. A close-up view of the residues T106A, Y111H, M135A, and T214A is shown: side-view (middle panel), top-down and bottom-up views (right panel). B TWIK-2 and its mutants were tested under conditions of high extracellular potassium (K⁺) at a pH of 7.4. Representative whole-cell current recordings were obtained from cells expressing both wild-type TWIK-2 channels and various mutants. These recordings were generated through voltage-clamp steps ranging from −90 to +100 mV in 10 mV increments. C Activation time constants (τ) were calculated using a single exponential fit during a 350 ms pulse to +60 mV for TWIK-2 (n = 28), TWIK-1 (n = 15), and the TWIK-2 mutants (M135A, n = 18; T106A, n = 14; T214A, n = 18; Y111H, n = 32; M135N, n = 15; T106A/T214A, n = 15; F98A, n = 12; S213A, n = 15; Y111A, n = 15). Deactivation time constants (τ) were calculated using a single exponential fit during a 350 ms pulse to -90 mV for TWIK-2 (n = 15), TWIK-1 (n = 15), and the TWIK-2 mutants (M135A, n = 11; T106A, n = 24; T214A, n = 12; Y111H, n = 33; M135N, n = 15; T106A/T214A, n = 15; F98A, n = 13; S213A, n = 15; Y111A, n = 15). Results were expressed as mean ± SD. Statistical significance was assessed by comparing wild-type with each mutant using a one-way ANOVA followed by Dunnett’s multiple comparisons test.

We conducted whole-cell recordings on cells expressing either wild-type or mutant TWIK-2 channels (Fig. 3B, C, and Supplementary Table 4). Overall, there were no significant differences in current densities between the mutant channels and wild-type TWIK-2, except for the double mutant T106A-T214A, which showed an increase in the current density (Fig. 3B, Supplementary Fig. 5A, and Supplementary Table 5). However, notable differences in the activation profiles of the macroscopic currents were observed among the mutants, as evident by varying activation and deactivation time constants (τ) (Fig. 3C and Supplementary Table 4). Mutations in amino acids at the extracellular and intracellular openings of the ion-permeation pore, specifically Y111 and M135, delayed TWIK-2 activation. The Y111H mutant, located at the extracellular exit of the ion-permeation pathway, exhibited the most significant increase in activation time, indicating slower channel activation. While the Y111A mutant did not show a noticeable change in activation time, it did exhibit faster deactivation. Both M135A and M135N mutations, that lie at the entrance of the ion permeation pathway within the central cavity, caused only a slight delay in activation, compared to wild-type TWIK-2 (Fig. 3C). The F98A mutation, in the K_2P modulator binding pocket, did not affect the activation but led to a substantial increase in the deactivation time compared to wild-type TWIK-2 (Fig. 3C). Conversely, mutations at the bottom of the selectivity filter—T106A, S213A, T214A, and the double mutant T106A-T214A—all showed faster activation than the wild-type channel (Fig. 3B, C and Supplementary Table 4).

Measurements of activation rates at +60 mV showed that single and double (T106A and T214A) mutants have similar activation profiles (Fig. 3C). When examining the raw traces, the double mutant produced larger currents. However, the currents at higher potentials appeared to activate much more slowly. These results suggest that the double mutation enables a greater flow of potassium ions through the pore. Our findings indicate that alanine mutations of the conserved threonines in the selectivity filter (T106A and T214A), which coordinate the canonical site-4 K⁺ ion, directly affect the pore domain and enhance gating efficiency, favoring open states over closed ones. Mapping these variants onto the structural framework of TWIK-2 channels suggests that mutations near the GYG selectivity filter and adjacent pore helices affect the function of the channel by disrupting K⁺ coordination. Most mutants impacted gating, often changing the voltage-dependence of activation (V₅₀) by interfering with the coupling between regulatory sites and the pore. This led to hyperpolarizing shifts, indicating a gain-of-function due to easier opening (Supplementary Fig. 5B and Supplementary Table 6). These effects highlight that channel opening and closing are sensitive to disruptions in the selectivity filter and pore helices, which block K⁺ permeation, as well as disturbances of the M2/M4 helices at the bundle crossing that alter voltage dependence, and impaired coupling between the cytoplasmic regulatory domains and the pore, which slow or destabilize gating transitions.

Modulation of TWIK-2 by known K_2P modulators

The distinct pseudo-tetrameric architecture of the transmembrane region and the sequence asymmetry between the pore-forming domains of the K2P channels are believed to render them insensitive to existing K⁺ channel blockers. Moreover, the cross-reactivity of known modulators between different subfamilies of K_2P channels has made the efficient targeting of selective K_2P channels a bottleneck⁴⁷. To establish a high-throughput, automated approach for screening modulators of TWIK-2, we investigated the modulation of TWIK-2 using four known K2P modulators: ML335, BL1249, ML365, and NPBA (Fig. 4A). In agreement with previous reports, we observed that ML365 and NPBA inhibit TWIK-2 with IC₅₀ values of 1.5 and 5.7 µM, respectively (Fig. 4B, C)^48,49. However, we did not observe any modulation of TWIK-2 by ML335 and BL1249, which have been shown to modulate TREK and TRAAK family channels (Fig. 4B)^8,48,59,60. To better understand how ML365 and NPBA inhibit TWIK-2, we compared their effects on inhibiting TWIK-2 and its mutants (Fig. 4D, E, and Supplementary Table 7). We compared the IC₅₀ values of wild-type TWIK-2 against four mutants: one each at the entry (Y111H) and exit (M135A) of the ion permeation pore, as well as two at the bottom of selectivity filters SF1 (T106A) and SF2 (T214A). In all four mutants, NPBA showed negligible changes in IC₅₀ compared to wild-type TWIK-2. The IC₅₀ of ML365 remained unchanged for the M135A and T214A mutants. However, the Y111H (0.16 µM) and T106A (46 µM) variants of TWIK-2 exhibited increased and decreased susceptibility to ML365, respectively. Based on these findings, we used 50 µM of either ML365 or NPBA to assess maximum inhibition across all variants. We observed reduced maximum inhibition at the highest concentration of NPBA tested in Y111A, Y111H, M135A, S213A, and T106A-T214A double mutants (Fig. 4D, E, Supplementary Fig. 5C, and Supplementary Table 8). In the case of ML365, we noticed that the maximum inhibition for all the variants remained similar to wild-type TWIK-2, except the T106A-T214A double mutant, where we saw a significant reduction (Supplementary Fig. 5C). While the T106A-T214A double mutation caused a decrease in maximum inhibition for both ML365 and NPBA, the differences seen in IC₅₀ values and maximum inhibition observed in other TWIK-2 mutants underscore potential differences in the inhibition mechanisms of ML365 and NPBA.

Fig. 4. Small molecule modulation of TWIK-2 by known K_2P channel modulators.

A Chemical structure of ML365, NPBA, ML335, and BL1249. B Concentration-response curves of various K2P modulators on TWIK-2 channel inhibition. Data are presented as mean ± SEM: ML365 (n = 8), NPBA (n = 8), ML335 (n = 8), and BL1249 (n = 8). Data from non-T cells (traced lines, empty symbols) are also presented for ML365 (n = 7) and NPBA (n = 16). C Changes in TWIK-2 currents in response to ML365 or NPBA (50 μM) were measured over time using a voltage-step protocol, applying +60 mV for 350 ms after holding at −90 mV. Data are presented as the mean ± SEM: ML365 (n = 8), NPBA (n = 8). D Comparison of concentration-response curves of ML365 on wild-type and mutant TWIK-2 channels. Data are presented as mean ± SEM: ML365 (n = 8). Right panel. Scatter dot plots show the current inhibition percentage at 50 µM for ML365 for TWIK-2 (ML365, n = 9) and the mutant channels (M135A: n = 9; T106A: n = 9; T214A: n = 9; Y111H: n = 9). The data are presented as mean ± SD. A one-way ANOVA with Dunnett’s multiple comparisons test was performed to compare the wild-type channel against each mutant. E Comparison of concentration-response curves of NPBA on wild-type and mutant TWIK-2 channels. Data are presented as mean ± SEM: NPBA (n = 7). The right panel shows scatter dot plots of the current inhibition percentage at 50 µM NPBA for TWIK-2 (NPBA, n = 10) and the mutant channels (M135A: NPBA, n = 10; T106A: NPBA, n = 10; T214A: NPBA, n = 10; Y111H: NPBA, n = 10). The data are presented as mean ± SD. A one-way ANOVA with Dunnett’s multiple comparisons test was performed to compare the wild-type channel against each mutant.

Comparison of the TWIK family of K_2P channels

Given the notable sequence and structural similarities between TWIK-1 and TWIK-2, we compared their channel activity and modulation using our automated whole-cell patch-clamp setup. Our results indicated that TWIK-1 and TWIK-2 have similar current densities (Fig. 5A, Supplementary Fig. 5A, Supplementary Tables 5 and 9). However, TWIK-1 shows a significantly faster activation rate compared to TWIK-2 and exhibits much weaker inhibition by compounds such as ML365 and NPBA (Fig. 3C, Supplementary Fig. 5C, and Supplementary Tables 4 and 7). Furthermore, we observed a strong pH-dependent change in the voltage reversal potential for TWIK-1, aligning with previous reports (Fig. 5B and Supplementary Table 9). In contrast, this phenomenon was not seen in TWIK-2, even after introducing a TWIK-1-like Y111H mutation into its structure (Fig. 1A, B).

Fig. 5. Comparison of TWIK-1 and TWIK-2 functional and structural properties.

A TWIK-1 channels showed a pH-dependent function, unlike TWIK-2. Whole-cell current traces were recorded from cells expressing both wild-type TWIK-1 and TWIK-2 channels. These traces were obtained using voltage-clamp steps from −90 to +100 mV in 10 mV increments, with the pH held at 7.4. B A comparison of the reversal potential (V_Rev) was conducted for TWIK-1 (at pH 7.4, n = 18; at pH 5.5, n = 24) and TWIK-2 (at pH 7.4, n = 30; at pH 5.5, n = 36), confirming TWIK-1’s pH sensitivity. The current–voltage (I–V) relationships for both TWIK-1 and TWIK-2 exhibited significant differences across the range of −90 to +100 mV at the reversal potential (V_Rev). Data in (B) were expressed as mean ± SD, with some error bars smaller than the data points. An unpaired two-tailed t-test with Welch’s correction was used to assess statistical significance. C Superposition of TWIK-1 (pH 7.4, mint, PDB:7SK0 and pH 5.5, moss, PDB:7SK1) and TWIK-2 (pH 7.5, gray, this study). The extracellular cap and selectivity filter are highlighted by orange and blue dashed boxes, respectively. D Cap domain in TWIK-2 (gray) resembles the acid-inhibited TWIK-1 structure (moss, top). E The selectivity filter of TWIK-2 (gray) superposes well with the conductive TWIK-1 state (mint, bottom). In TWIK-1, the pH-sensitive histidine (H122) moves towards the pore at low pH (top), blocking K⁺ conduction, whereas in TWIK-2, the equivalent residue (Y111) adopts an “up” conformation without occluding the ion pathway (bottom). F TWIK-2 SF1 (gray) superimposed on pH 5.5 TWIK-1 SF1 (PDB: 7SK1, top, moss) and pH 7.5 TWIK-1 SF1 (PDB: 7SK0, bottom, mint). G TWIK-2 SF2 (gray) superimposed on pH 5.5 TWIK-2 SF1 (PDB: 7SK1, top, moss) and pH 7.5 TWIK-1 SF2 (PDB: 7SK0, bottom, mint).

TWIK-1 is a pH-sensitive K_2P channel whose structures at pH 7.5 and 5.5 have been previously reported³⁹. Upon comparing the TWIK-2 structure (this study) with that of TWIK-1, both at pH 7.5, we find that the differences in the structure are localized to the cap region (Fig. 5C, D, and Supplementary Fig. 6A). When compared with TWIK-1 at pH 5.5, the differences in TWIK-2 structure lie only in the SF region, specifically SF1-M2 linker. H122 in SF1-M2 linker of TWIK-1 (equivalent to Y111 in TWIK-2) is held towards the transmembrane region inside a hydrophobic pocket between SF1 and PH1 (Fig. 5E and Supplementary Fig. 6B). At low pH, protonation causes release of H122 from this hydrophobic pocket, flipping it into a “up” conformation over the ion permeation pore blocking the K⁺ ion exit (TWIK-1) (Fig. 5F, G). In addition, low pH displaces the cap domain upwards by ~4 Å to accommodate the “up” conformation of H122 (Fig. 5D). Despite the residue Y111 in TWIK-2 (H122 in TWIK-1) adopting this distinct “up” conformation, the SF1-M2 linkers themselves stay splayed apart and they do not close over the SF obstructing the ion permeation pathway (Fig. 5F, G). Coinciding with the up conformation of Y111, we notice that the cap domain in TWIK-2 superposes better with the acid-inhibited TWIK-1 structure than the TWIK-1 at pH 7.5 (Fig. 5D). Hence, TWIK-2, at pH 7.5, has the cap region that is similar to acid-inhibited TWIK-1 (pH 5.5) but has the SF loops positioned in a similar way as that of the active TWIK-1 (pH 7.5) (Fig. 5C–G and Supplementary Fig. 6A, B). As TWIK-2 is not pH-sensitive, it is likely that displacement of helical cap is not used to regulate K2P channel activity, as is also the case in other lipid-regulated TREK and TRAAK family of K2P channels. These findings emphasize the distinct functional and structural characteristics of TWIK-1 and TWIK-2.

Discussion

Our study employs cryo-electron microscopy (cryo-EM), site-directed mutagenesis, and automated whole-cell patch clamp electrophysiology to gain insights into the molecular determinants of TWIK-2 activity modulation. In addition, using known K_2P modulators, we established a robust assay to screen for inhibitors specific to TWIK-2.

Structure of the apo human TWIK-2

The ~3.7 Å cryo-EM structure of TWIK-2 reveals a canonical K_2P channel architecture, including the domain-swapped homodimer and a pseudo-tetrameric central pore (Figs. 1 and 2). However, we also notice M4 helix in a so-called “down” conformation and a non-canonical SF conformation. While this manuscript was under revision, PDB and cryo-EM map of apo TWIK-2 structure from another group became available⁶¹ (Supplementary Fig. 6C). We find no noticeable differences upon comparing their apo-TWIK-2 structure with ours. In pH-sensitive channels, such as TWIK-1 and TASK-3, the structures obtained in both open and acid-induced closed conformations show M4 helix in a down conformation, irrespective of bound lipid at the fenestration site^12,13,35,39 (Supplementary Fig. 4A). However, in lipid-modulated K_2P channels, such as TREK and TRAAK, only occupancy of the fenestration site has been shown to push the M4 helix into a down conformation^9,16. Presence of lipid at the classic modulatory phospholipid binding site, but not the fenestration site has been shown to retain M4 helix in an up conformation^8,57,62 (Supplementary Fig. 4A). Perhaps the molecular determinants of pH-induced and lipid-mediated modulation in the K_2P channels are different and complex, as molecular dynamics simulation studies of TWIK-1 have reported that the alkyl tails of surrounding lipids can enter the fenestrations but not far enough to occlude the inner pore⁶³. MD simulation studies to understand the role of C-type gating in K_2P channels have suggested that extensive conformational dynamics, especially in the SF2-M4 loop and around the fenestration site, are induced by the motion of the M4 helix^62,64. It is also believed that the movement of the M4 helix from a “down” to a “deep-down” conformation destabilizes the PH2 helix, causing a perturbation in SF2 and thereby inactivating the channel⁶⁴. In TWIK-1, the structures obtained in both open (pH 7.5) and acid-induced closed conformations show M4 helix in a down conformation, irrespective of bound lipid at the fenestration site (Supplementary Fig. 4A).

While this manuscript was under revision, a preprint by Mondal et. al. suggested that lipids could regulate channel activity in TWIK-2. Using cryo-EM, they propose that access to the ion permeation pore in TWIK-2 is obstructed by a lipid “plug” that, upon reorganization or removal, activates the channel by allowing access to the ion permeation pore. They demonstrate that mutations of residues near the hydrophobic cuff and near the C-terminus, that is, below the hydrophobic cuff, can increase channel activity, possibly by altering the lipid affinity at the central cavity⁶⁵. Notably, we too observed lipid-like densities in our cryo-EM map at the classic modulatory lipid binding site and the fenestration site (Fig. 2E). However, despite the “down” conformation of the M4 helix in our TWIK-2 structure, which is a result of occupation of fenestrations site by the lipid, we observed that the cytoplasmic entrance to the ion permeation pore is accessible. The access to the ion permeation pore from the intracellular environment is regulated by a seal formed by charged residues in the M4 helix (Supplementary Fig. 4B). In TASK-2, another pH-sensitive K_2P channel, an asparagine (N243) and lysine (K245) pull the M4 helices over the ion permeation entrance and form a hydrogen bonding network²⁷. The K245 side chain is held towards the hydrophobic M2 at basic pH, and at low pH, this residue flips out towards the N243, thereby sealing the entrance to the ion-permeation pathway. In TWIK-2, we see a similar residue arrangement where R257 is held towards M2, facing away from Q254, precluding the formation of a hydrogen bonding network, thus keeping the entrance to the ion permeation pore open (Supplementary Fig. 4B).

K⁺ ions exit the ion permeation pore at the space below the cap domain, guarded by SF1-M2 and SF2-M4 loops. In pH-sensitive channels like TWIK-1 and TASK-3, a conserved histidine residue (H122 in TWIK-1 and H98 in TASK-3) is positioned toward the transmembrane region within a hydrophobic pocket between SF1 and PH1. At low pH, protonation of this histidine causes it to be released from the hydrophobic pocket, rendering the SF1-M2 loop flexible. The histidine either comes together over the ion-permeation pore, blocking the K⁺ ion exit (TWIK-1), or pulls away, flipping the adjacent bulky aromatic residues into the path of the pore exit (TASK-3) (Fig. 5B–G and Supplementary Fig. 4C)^12,13,35,39. TWIK-2 has a tyrosine (Y111) in contrast to the pH-sensing histidine seen in TWIK-1 and TASK-3. Substitution of tyrosine with histidine at this position (Y111H) does not induce pH sensitivity in TWIK-2 (Fig. 1B). As noted by other groups, while a conserved histidine in the SF1-M2 loop of pH-sensitive K_2P channels is a major proton-sensing amino acid, it is not the only determinant of pH sensitivity in K_2P channels^{13,30,39,66,67}. Furthermore, we notice that while the conformation of SF1-M2 and SF2-M4 loops surrounding the pore in TWIK-2 match the conformations of open TWIK-1 and TASK-3, the amino acid Y111 itself is positioned facing up, away from the SF, as it is in the acid-inhibited structure of TWIK-1 (Fig. 5F). Two other groups reported structures of TWIK-2 with an distinct “up” conformation of Y111 and proposed that the pocket formed by the “up” conformation of Y111 could be a potential site for small molecule mediated regulation of TWIK-2, while this manuscript was under revision^61,65. Because we notice a “down” conformation of M4 helix, lipid-like tubular densities in the fenestration site, an open entrance and exit of the ion permeation pathway, but with only three K⁺ ions in an ion conduction pore surrounded by SF, we hypothesize that the TWIK-2 structure could be an intermediate conductive state.

Molecular determinants of K⁺ conductance in TWIK-2

In our automated whole-cell patch-clamp recordings using TWIK-2 expressed in HEK293 GnTI⁻ cells, we notice voltage- and time-dependence of currents, unlike previous observations involving TWIK-1 and TWIK-2^48,68. The currents observed in our experiments are similar to robust TWIK-1 currents recorded using giant inside-out excised oocyte patches, where the currents displayed time- and voltage-dependence of activation as seen in the TREK subfamily of channels⁶⁹. Perhaps, the lack of native complex glycosylation of TWIK-2 due to overexpression in HEK293 GnTI⁻ cells alters channel properties, including localization. Nevertheless, this experimental setup provided a quick, high-throughput method for measuring K⁺ outward currents in TWIK-2 (Figs. 1 and 3). Through site-directed mutagenesis, we identified that the conserved amino acids T106 and T214 act as critical determinants of channel gating in the selectivity filters. The corresponding alanine mutations of these threonines significantly reduced the time- and voltage-dependence of TWIK-2 channel activation (Fig. 3). T106A and T214A are the only conserved amino acids within the SF that coordinate K⁺ ions through their side chains, as opposed to carbonyl oxygens like the remaining amino acids in the SF. T106 in TWIK-2 is also stabilized by interactions with T214 and I215 at the SF2-PH2 interface (Fig. 3A and Supplementary Fig. 4D). T214 is stabilized by interactions with T106 and S213 around the base of the SF and with L246 of M2 near the hydrophobic cuff (Supplementary Fig. 4D). T106A and T214A mutations in TWIK-2 disturb these networks of hydrogen bonds and hydrophobic interactions that stabilize the base of the SF, resulting in a faster activation and making the channel more open (Fig. 3B, C). While the T106A-T214A double mutant shows a lower activation time constant compared to the wild-type, as in TWIK-2 and the threonine single mutants, the current density, however, shows a cumulative increase. Similarly, we notice a decreased activation time constant in the case of the S213A mutation at the bottom of the PH2. The decrease in time- and voltage-dependence of channel activation in these mutants indicates that recombinant TWIK-2 is voltage-gated, and that mutations at the bottom of the SF lower the threshold for opening of the channel. These findings provide a mechanistic framework for understanding TWIK-2 gating. Incidentally, in the TREK and TASK subfamilies of K2P channels, mutating the conserved threonine of SF1 and SF2 to cysteine abolished voltage gating⁶⁸.

The residues at the kink in M2 helix in juxtaposition with M4 helix create a “hydrophobic cuff” that regulates hydration of the central vestibule (Fig. 2A). It has been shown that mutation of a single non-polar amino acid at this hydrophobic cuff in K_2P channels to a polar or charged residue enhances channel activity by promoting hydration of the cavity below the SF^69–72. However, the introduction of a polar amino acid at this position in TWIK-2 (M135N) showed no drastic effect on channel activity, except for a mild delay in channel activation (Fig. 3). Next, we wanted to see if keeping the chemical nature of the side chain intact but reducing the size of the side chain would influence channel activity. The M135A substitution at the hydrophobic cuff in TWIK-2 behaved similarly to M135N. The M135 sidechain faces L253 of M4 at the hydrophobic cuff and is surrounded by a hydrophobic environment made of lipid-like densities (Supplementary Fig. 4D). As a result, only replacing the longer non-polar methionine with either a polar asparagine or a shorter non-polar alanine doesn’t drastically alter the channel activity (Fig. 3B, C).

A conserved histidine in the SF1-M2 linker has been shown to be a pH sensor in TWIK-1 and some pH-sensitive TASK family channels. In structures obtained at basic and acidic pHs, the histidine in the SF1-M2 loop is shown to induce conformational changes near the cap, constricting the pore on the extracellular side^{12,13,35,39,40}. While the residues surrounding this histidine are conserved in TWIK-2, histidine itself is replaced by a tyrosine (Fig. 5F and Supplementary Fig. 4C). We see that TWIK-2 is not pH sensitive. We compared TWIK-1 in our assay setup and observed pH-dependent changes in current; however, TWIK-2 remained pH-insensitive even with the introduction of a TWIK-1-like Y111H mutation. In TWIK-2, Y111H mutation shows increase in the time taken for activation, perhaps because of the interaction of the introduced histidine at this position with E224 of the SF2-M4 linker preventing the required conformational freedom of the SF1-M2 and SF2-M4 loops around the ion permeation exit (Fig. 3A and Supplementary Fig. 4D). A Y111A mutation shows no apparent difference in channel activation but shows a much faster deactivation. Interestingly, the F98A mutation in the K_2P modulatory pocket shows a slower deactivation. The F98 side chain is in proximity to Y109 and Y111 side chains, and perhaps the F98A mutation disturbs this stacking interaction and affects the integrity of the top of the SF (Fig. 3A). These observations suggest that amino acids at the top of the SF regulate channel deactivation.

Small molecule modulation of TWIK-2

High-throughput screening of K_2P channel modulators has been previously reported, especially by assaying the growth of engineered yeast whose survival depended on the expression of functional K_2P channels^22,73. Here, we confirm the feasibility of using an automated whole-cell patch-clamp electrophysiology approach as a high-throughput platform for screening TWIK-2 modulators in HEK cells. We used four known K_2P modulators in our trial—ML335, BL1249, NPBA, and ML365—and noticed that we could see inhibition by only NPBA and ML365 (Fig. 4A, B). In the case of NPBA, we do not see a prominent change in the IC₅₀ of inhibition, except for a mild decrease in the efficacy in T214A and M135A, both mutations that face fenestration site and central vestibule (Fig. 3A). Moreover, we notice a reduction in maximum inhibition for NPBA in M135A, S213A, Y111A, Y111H, and the T106A-T214A double mutants of TWIK-2 (Fig. 4 and Supplementary Fig. 5C). All of these amino acids, except the Y111, outline a pocket underneath the SF in the central vestibule (Fig. 3A). Based on these observations, we hypothesize that NPBA diffuses through the fenestration site and binds underneath the SF in the central vestibule.

In the case of ML365, we do not see any difference in the maximum inhibition for any of the mutants except the T106A-T214A double mutants, hinting at the bottom of the SF to be a potential site, similar to NPBA (Fig. 4 and Supplementary Fig. 5C). We see a variation in response to ML365 for T106A and Y111H, compared to WT. However, both of these mutations reside on either side of the SF. Interestingly, the change in ML365 IC₅₀ for T106A and Y111H is opposite - Y111H improves the efficacy of ML365 while T106A reduces it (Fig. 4D). Perhaps, the decreased efficacy of ML365 for T106A but not for T214A is because PD1 asymmetrically governs the binding of ML365 as opposed to PD2, as both Y111H and T106A are a part of PD1. As noted earlier, Y111H potentially minimizes the movement of the SF2-M4 and SF1-M2 linkers by interacting with E224 (Fig. 3A and Supplementary Fig. 4D). Combined with a reduced steric clash because of Y-H substitution, this reduced flexibility could allow for better binding of ML365. Conversely, the differential effect of ML365 in mutations that reside on either end of SF1 could indicate an allosteric coupling between the two sides of SF. As is also evident by observations made in the case of NPBA, where both Y111A and Y111H mutations reduce the maximum channel inhibition observed, although the potential binding site of NPBA is below the SF. Nonetheless, our studies suggest a potential difference in the mechanism of binding for ML365 and NPBA.

Methods

Cloning and overexpression of recombinant TWIK-2

The codon-optimized full-length wild-type human TWIK-2 (KCNK6) was synthesized by GenScript and was subcloned into the pEG BacMam expression vector (Addgene plasmid #160683) to be expressed with an N-terminal Strep-tag-II, followed by a GFP tag and an HRV-3C protease site. TWIK-2 mutations (F98A, T106A, Y111A, Y111H, M135A, M135N, S213A, T214A, T106A/T214A) were also made in this background by GenScript. Wild-type (TWIK-2) and variants of TWIK-2 were expressed in HEK293 GnTI⁻ cells (ATCC #CRL-3022). Large-scale protein production was done using a baculoviral expression system as per the protocol described by Goehring et al.⁷⁴. Briefly, the TWIK-2 expression vector was transformed into chemically competent DH10Bac cells (Thermo Fisher Scientific), and white colonies were selected on gentamicin-kanamycin-tetracycline LB agar plates. Isolated Bacmid DNA was used to transfect sf9 cells (ATCC #12659017) with Cellfectin-II reagent (Gibco) per the manufacturer’s protocol to make baculovirus P1. P1 was then amplified to make P2 by adding 200 μl of P1 to 200 ml culture of sf9 cells at a 1 × 10⁶ cells/ml density. Cells were grown at 27 °C for 96 h and harvested by spinning at 4000 × g for 20 min. The supernatant was filtered with a 0.22 μm filter and stored at 4 °C in the dark as a P2 virus. This was subsequently used for large-scale transduction of HEK293 GnTI⁻ suspension adapted cells grown in FreeStyle 293 media supplemented with 2% heat-inactivated fetal bovine serum (FBS, Gibco) at density of 2.5–3.0 × 10⁶ cells/ml. Transduced cells were incubated on an orbital shaker at 37 °C and 8% CO₂ at 120 rpm. Eight hours after transduction, 10 mM Sodium butyrate was added, and cells were grown for an additional 16–24 h at 37 °C, 8% CO₂. Cells were harvested by centrifugation at 4000 × g for 20 min, flash-frozen in liquid nitrogen, and stored at −80 °C.

Purification of recombinant human TWIK-2

Cell pellets from 2.4 L culture were thawed and lysed in 120 ml lysis buffer (25 mM Tris-HCl, 150 mM KCl, 0.8 μM aprotinin, 2 μg/ml leupeptin, 2 μM pepstatin A, 1 mM PMSF, pH 7.5) by sonication. Lysed cells were centrifuged at 2400 × g for 5 min to remove cell debris. The supernatant was then ultracentrifuged at 185,000 × g for 45 min. The pellet from ultracentrifugation was resuspended using a Dounce homogenizer in 60 ml of lysis buffer. To solubilize the membrane, 60 ml 2X solubilization buffer (25 mM Tris-HCl, 150 mM KCl, 0.8 μM aprotinin, 2 μg/ml leupeptin, 2 μM pepstatin A, 1 mM PMSF, 2% DMNG, pH 7.5) was added, and the solubilization was carried out for 90 min at 4 °C. The solubilized membrane was then centrifuged at 185,000 × g for 1 h. The supernatant was filtered through a 0.22 μm filter and was loaded onto a column containing 3 ml Strep-Tactin resin (IBA Life Sciences) at a flow rate of 0.5 ml/min. The column was then washed with 15 ml of wash buffer (25 mM Tris-HCl, 150 mM KCl, 0.1% DMNG, pH 7.5) and eluted with 12 ml elution buffer (25 mM Tris-HCl, 150 mM KCl, 0.1% DMNG, 5 mM Desthiobiotin, pH 7.5) in 1.5 ml fractions. The homogeneity of eluted fractions was analyzed by SDS-PAGE and subsequently pooled and concentrated to 7 mg/ml. In-house purified HRV 3C protease was added to purified TWIK-2 at a molar ratio of 1:20 and incubated overnight at 4 °C to remove the GFP tag. The digested protein was further purified by size-exclusion chromatography using Superdex 200 Increase 10/300 GL column (Cytiva) pre-equilibrated with SEC buffer (0.1% DMNG, 25 mM Tris-HCl, 150 mM KCl, pH 7.5). The fractions corresponding to GFP-tag-free TWIK-2 dimer were pooled and concentrated to 2.3 mg/ml for cryo-EM sample preparation.

Cryo-EM sample preparation and data collection

UltraAuFoil 300 mesh 1.2/1.3 grids (Quantifoil) were glow discharged (PELCO) for 60 s at 15 mA before use. 3.5 μl purified TWIK-2 was applied to the glow-discharged grids and blotted for 2.5–3 s at 4 °C and 100% humidity using a Vitrobot (Mark IV, Thermo Fisher Scientific). Grids were immediately plunged-frozen in liquid ethane. Grids were screened, and movies were collected at the University of Michigan cryo-EM facility on 300 kV Titan Krios G4i cryo-EM (Thermo Fisher Scientific) equipped with K3 direct detector camera (Gatan), and a post-BioQuantum GIF energy filter (Gatan), slit width set to 20 eV. 15,006 movies were collected using SerialEM 4.0 software in counting mode at a magnification of 105,000× (nominal pixel size of 0.83 Å/pixel). The total dose was 60 e⁻ Å⁻² with 60 frames with a defocus range between 0.8 and 2.5 μm.

Cryo-EM data processing

Data processing was performed in cryoSPARC (version 4.5.3)⁷⁵ Briefly, 15,006 movie stacks were motion corrected and patch-CTF estimated, followed by a manual curation to remove micrographs with relatively thick ice and CTF fit >7 Å. 10,629 micrographs were subsequently selected and used for reference-free particle picking by blob picker with minimum particle diameter of 80 Å and maximum particle diameter of 100 Å followed by extraction at a box size of 256 pixels at binning of 2. 5,447,554 particles were picked and subjected to two rounds of 2D classification that resulted in 1,387,475 relative particles. A subset of ~380,000 particles from 2D classes with clear secondary structure features were used to generate four ab-initio 3D classes with C1 symmetry. One “good” class with clear transmembrane features and one “bad” class representing empty micelle were selected as reference input for downstream heterogeneous refinement with the total relatively clean 1,387,475 particles. Multiple rounds of heterogeneous refinement were performed until no further improvement of the resolution. 301,000 selected particles from the best class after the last heterogeneous refinement were re-extracted with 256 pixels and subjected to one round of Non-uniform refinement with C1 symmetry. A 3D mask was generated using volume of the best ab-initio class in Chimera 1.17.1 by manually removing the micelle density and imported into cryoSPARC with a threshold of 0.064 and dilation radius of 3 pixels. This mask was used for a focused 3D classification of the clean re-extracted particles into 4 classes in cryoSPARC. 104,997 particles from the class with clear secondary structure features were selected for reference-based motion correction, and refined by non-uniform refinement with C1 symmetry to obtain a final map at ~3.7 Å.

Model building, refinement, and structure analysis

The sharpened map from the final non-uniform refinement in cryoSPARC was used to build a preliminary model in ModelAngelo 1.0⁷⁶. The preliminary model was analyzed, and improperly built regions were rebuilt manually in Coot 0.9.8.83⁷⁷. Final structure refinement was performed in Phenix 1.21⁷⁸. Figures were prepared using UCSF ChimeraX 1.17⁷⁹. The radii along the channel pore were calculated using HOLE⁸⁰. FoldX was used to generate mutant models of TWIK-2⁸¹.

Automated patch clamp recordings

The wild-type and mutant TWIK-2 expression constructs were used to transfect adherent HEK293 GnTI⁻ cells for automated whole-cell patch-clamp recordings. In brief, 1–7 million cells, grown in DMEM medium with 10% FBS, were transfected at a density of approximately 0.6–0.7 million cells/mL using 1–4 µg of DNA and TurboFect, following the manufacturer’s instructions (Thermo Fisher Scientific). After an 8-h incubation, the media was replaced with fresh DMEM containing 10 mM sodium butyrate, and the cells incubated for another 16–24 h before use. The cells were maintained at 37 °C, 8% CO₂, and 100% humidity. Before experiments, the cells were centrifuged at 200 × g for 5 min, and the resulting pellet was resuspended in an external recording solution (see below) to achieve a final density of 500,000 cells/mL. This cell suspension was transferred to the instrument cell hotel, kept at 10 °C, and shaken at 200 rpm until recordings began.

Automated patch clamp recordings were carried out using the Syncropatch 384 PE platform (Nanion Technologies, Munich, Germany). We used four-hole, 384-well recording chips with medium resistance between 2 and 4 MΩ. Pulse generation and data collection were managed with the PatchController384 V.1.3.0 and DataController384 V1.2.1 software (Nanion Technologies). Whole-cell currents were filtered at 3 kHz and sampled at 10 kHz. Access resistance and apparent membrane capacitance were estimated using the built-in protocols. Recordings were obtained from cells expressing both wild-type and mutant K2P channels at room temperature, with an internal solution containing 110 mM KF, 10 mM KCl, 10 mM NaCl, 10 mM HEPES, 10 mM EGTA, with pH adjusted to 7.2 using KOH. The external solution included 44 mM NaCl, 100 mM KCl, 2 mM CaCl₂, 1 mM MgCl₂, and 5 mM glucose, buffered with 10 mM HEPES (pH 7.4, adjusted with NaOH) or 10 mM MES (pH 5.5, adjusted with NaOH). Recordings were only accepted from cells that showed a seal resistance greater than 100 MΩ, a series resistance below 10 MΩ, and stable cell capacitance throughout the session. To assess the biophysical properties of wild-type and mutant TWIK channels, we measured both current density and potassium ion reversal potential (Vrev). Whole-cell currents were recorded from a holding potential of −90 mV and elicited with depolarizing steps (200 ms) from −90 to +100 mV in 10 mV increments, followed by a return step to −90 mV (tail currents, 100 ms), and current–voltage (I–V) relationships were plotted for each condition. The voltage reversal potential (Vrev) was identified from the I–V curve as the membrane potential where no net current was observed. Current density was calculated by normalizing the peak current amplitude at the highest voltage step to the cell’s membrane capacitance, which accounts for differences in cell size. Tail currents measured 1 ms after returning to −90 mV were normalized to their maximum amplitude and graphed against the test potential (−90 to +100 mV). To find the voltage at half-maximal activation (V₅₀) and the slope factor (k), normalized G–V curves were fitted with the Boltzmann equation:

$$G=1/\left(1+\exp \left[\left(V-V_{50}\right)/k\right]\right)$$ 1

The properties of each TWIK-2 mutant were compared to those of the wild-type (WT) channel measured in parallel, focusing on the difference (ΔV₅₀) in voltage-dependent activation V₅₀.

To evaluate the effect of small molecules on TWIK-2 channel activity, the test compound was applied to the extracellular face (cap domain) of the channel using an external solution. A depolarization step to +60 mV was then performed for 350 ms, starting from a holding potential of -90 mV, and this procedure was repeated every 5 s. During each depolarization, the peak outward current amplitude was carefully measured and normalized to the currents recorded before the addition of the compounds. For the concentration-response curves (CRCs), various concentrations of the compounds were applied across the four-hole, 384-well recording chip, or a single high concentration of 50 µM was used to achieve maximum blocking. Electrophysiological responses were analyzed in detail by calculating the ratios of peak currents in the presence (Icompound) and absence (I₀) of the test compounds. Data analysis and graphing were performed using DataController 384 V.3.2.0 (Nanion Technologies) and GraphPad Prism V.10.5.0, with concentration-response curves fitted to a three-parameter sigmoidal model. The rates of channel activation and deactivation were determined by fitting the outward currents elicited at +60 mV and the tail currents during -90 mV pulses to a single exponential function using DataController 384 from Nanion Technologies. The number of cells used for each experimental condition is listed in the figure legends. Detailed statistical methods were also described in the figure legends.

Drug and reagents

K_2P modulator ML365 (Sigma-Aldrich SML2643), NPBA (MedChemExpress), ML335 (MedChemExpress), and BL1249 (Tocris) were diluted to 20 mM in sterile DMSO (Sigma-Aldrich) and stored in -80°C before use.

Author contributions

V.N. and S.M. designed the study. V.N., J.Z., and J.K.R. optimized protein expression, purification, and cryo-EM sample preparation conditions. Q.M. and A.L. prepared the sample. Q.M. and J.K.R. collected the cryo-EM data. Q.M., V.N., and A.K. processed the cryo-EM data. A.K. and V.N. built the model and deposited the coordinates. C.C.H. performed all electrophysiology experiments. V.N., C.C.H., Q.M., and S.M. analyzed the data. V.N., Q.M., and C.C.H. wrote the original draft of the manuscript. All authors contributed to finalizing the manuscript.

Peer review

Peer review information

Nature Communications thanks the anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Data availability

The density map of TWIK-2 has been deposited in the Electron Microscopy Data Bank (EMDB) under the accession number EMD-47768. 3D coordinates for the TWIK-2 structure have been deposited in the Protein Data Bank (PDB) under the accession code 9E94 (cryo-EM structure of human TWIK-2 at pH 7.5). The TWIK-1 structures used for comparison in this manuscript were obtained from the Protein Data Bank (PDB) under accession codes 7SK0 (TWIK-1 in MSP1D1 lipid nanodisc at pH 7.4) and 7SK1 (TWIK-1 in MSP1E3D1 lipid nanodisc at pH 5.5). Source data are provided as a source data file. Source data are provided with this paper.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-69072-1.