INTRODUCTION
The M-channel, a heteromeric voltage-gated potassium (K
v) channel composed of KCNQ2 and KCNQ3 subunits
1-3, governs neuronal excitability by generating a slowly activating, non-inactivating potassium current (
IM) that dampens repetitive firing
2, 4-6. First identified in sympathetic neurons for its suppression by muscarinic acetylcholine receptor signaling (hence "M"-channel)
4, it is now recognized as a master regulator of action potential threshold and spike-frequency adaptation across the central and peripheral nervous systems
6-12. The molecular identity of
IM was established with the cloning of
KCNQ2 and
KCNQ31, 2, 13-16, which form heteromeric channels with biophysical properties precisely matching native M-currents — distinct from their homomeric counterparts
1-3, 17. Dysfunction of these subunits underlies a spectrum of severe neurological disorders, including benign familial neonatal seizures (BFNS)
3, 5, 13-15, 18-27, developmental epileptic encephalopathies (DEE7)
22, 28-31, and autism-associated phenotypes
32-35, underscoring their critical physiological and therapeutic relevance
36.
Despite decades of research, fundamental questions regarding the M-channel’s architecture and gating mechanisms remain unresolved
34, 37. A central controversy concerns its heteromeric stoichiometry
34: while biochemical and functional studies suggested a 2:2 (KCNQ2:KCNQ3) assembly
2, 38, others proposed variable or asymmetric arrangements
39, 40. This uncertainty has obscured how subunit composition dictates the channel’s unique subthreshold activation — a hallmark feature enabling
IM to stabilize resting membrane potentials
2, 4, 6. Although recent structures of homomeric KCNQ channels have revealed their canonical architecture
37, 41-46, the absence of heteromeric M-channel structures has precluded mechanistic understanding of its distinct gating properties and subunit cooperativity
34, 47, 48. Resolving these questions is critical for both ion channel biology and drug discovery, particularly given the M-channel’s validation as a target for epilepsy and neuropsychiatric disorders
47, 49-52.
Therapeutic targeting of the M-channel has faced significant hurdles
49, 53. Retigabine, the first-in-class M-channel activator approved for refractory epilepsy, was withdrawn due to several adverse effects (
e.g., bladder toxicity and retinal discoloration) linked to its activity across multiple KCNQ subtypes or metabolism in the human body
51, 52, 54-60. Developing subtype-selective activators requires precise structural insights into the heteromeric channel’s drug-binding sites and activation mechanisms — a goal hindered by the lack of M-channel structures.
Here, we resolve these longstanding questions through cryo-electron microscopy (cryo-EM) structures of the human M-channel in multiple functional states. Our findings reveal unexpected stoichiometric flexibility in KCNQ2:KCNQ3 assemblies and identify a markedly depolarized conformation of the voltage-sensing domain (VSD) in KCNQ3 compared to KCNQ2, providing the first structural explanation for the channel’s characteristic subthreshold activation. Electrophysiological characterizations of engineered concatemers and chimeric constructs corroborate these structural insights. Leveraging this knowledge, we developed CLM142 (1), a potent and selective M-channel activator that facilitated determination of the open-state structure. Our work not only elucidates fundamental principles of M-channel assembly and gating but also establishes a framework for structure-guided development of targeted therapies.
RESULTS
Structure of the human M-channel reveals unexpected stoichiometric plasticity
We determined cryo-EM structures of the human M-channel using wild-type (WT) of both KCNQ2/KCNQ3 heteromers and engineered constructs where flexible N- and C-terminal regions and the disordered loop between HA and HB helices were removed to improve biochemical stability (Fig. 1a; Supplementary Figs. S1–S11 and Tables S1–S5). The functional integrity of the engineered constructs was confirmed through electrophysiological recordings. This approach enabled us to resolve four distinct structural states: the WT M-channel (Fig. 1a), the engineered construct in the apo state (apo M-channel; Supplementary Fig. S3a), the construct bound to the activator CLM142 (CLM142-bound M-channel; Supplementary Fig. S3b), and the construct in the open state stabilized by both CLM142 and PIP2 (open-state M-channel; Supplementary Fig. S3c).
All resolved structures adopted the canonical voltage-gated K
+ channel architecture, with four VSDs surrounding a central pore domain (PD; Fig. 1a–c)
61. However, unlike homomeric channels, the M-channel’s central pore and VSDs are formed by asymmetric arrangements of KCNQ2 and KCNQ3 subunits
37, 41-44, 46. The local resolution of the cytosolic domains, crucial for the tetrameric assembly
62, 63, was poor and precluded reliable modeling. Three key structural features enabled unambiguous discrimination between subunits: first, the VSD of KCNQ3 adopts a distinct orientation relative to the central pore compared to KCNQ2 (Fig. 1d, e); second, KCNQ3 possesses a significantly longer extracellular loop (ECL) which yields a pronounced density difference (Fig. 1f; Supplementary Figs. S1, S12); and third, the high resolution (~2.8 Å) of the central pore region allowed discrimination of non-conserved residues, including Y284 in KCNQ2 vs T323 in KCNQ3, F316 in KCNQ2 vs L355 in KCNQ3, and Y226 in KCNQ2 vs C255 in KCNQ3 (Fig. 1g; Supplementary Fig. S1).
Analysis of these preparations revealed that the M-channel exists in four distinct stoichiometric configurations (Fig. 1a; Supplementary Fig. S3). These include a 3:1 KCNQ2:KCNQ3 assembly (M2223), two 2:2 assemblies with either adjacent (M2233) or alternating (M2323) subunit arrangements, and a 1:3 assembly (M2333). Intriguingly, the relative proportions of these assemblies varied significantly between preparations. Under our standard co-expression conditions for WT subunits, the KCNQ2-rich M2223 configuration was the least prevalent (6.45%), indicating an inherent assembly preference under these conditions that favors incorporation of KCNQ3. This resulted in a population dominated by the balanced 2:2 stoichiometries (M2323, 38.38%; M2233, 32.6%) and the KCNQ3-rich M2333 configuration (22.57%).
In contrast, engineered constructs exhibited a pronounced redistribution, with the M2223 configuration becoming dominant (53.2% in apo M-channel) (Supplementary Fig. S3), correlating with higher expression levels of KCNQ2 relative to KCNQ3 (Supplementary Fig. S2b). This shift likely reflects a combination of altered expression levels and the enhanced biochemical stabilities of the truncated subunits compared to their WT counterparts. This stoichiometric plasticity was further quantified through systematic transfection experiments using defined ratios of KCNQ2-GFP and KCNQ3 plasmids (1:1, 1:4, and 1:9). By measuring the ratio of peak GFP fluorescence (reporting KCNQ2 incorporation) to the total UV absorbance of the purified tetramers (reporting total protein yield), we observed a progressive decrease in the relative KCNQ2 content (146, 128, and 54 mV per mAu, respectively) as KCNQ3 availability increased (Supplementary Fig. S13; see Materials and Methods for detailed protocols). Collectively, these data demonstrate that stoichiometric flexibility is an inherent property of M-channel assembly, where the final population distribution is determined by the relative availability and biochemical stability of the subunits within the expression system.
The high-resolution structure enables precise structural alignment of the subunit interface (S5 and S6 helices) of KCNQ2 and KCNQ3, revealing an exceptional degree of conservation in both sequence (> 90% similarity) and tertiary structure (RMSD = 0.437 Å; Supplementary Fig. S14). This near-perfect structural compatibility explains the M-channel’s remarkable stoichiometric plasticity, permitting all combinatorially possible configurations (M2223, M2233, M2323, and M2333) to assemble through favorable structural compatibility at the subunit interfaces.
Concatemeric constructs of KCNQ2 and KCNQ3 recapitulate electrophysiological properties of the WT M-channel
To validate the physiological relevance of the observed stoichiometric assemblies, we engineered concatemers with defined subunit ratios matching our structural findings: 3:1 (M2223), two distinct 2:2 arrangements (M2323 and M2233), and 1:3 (M2333) (Fig. 2a). Fluorescence size-exclusion chromatography confirmed that these concatemeric constructs form stable tetrameric assemblies with elution profiles matching the WT M-channel (Supplementary Fig. S15). These designs enabled precise control over subunit composition while maintaining native inter-domain connectivity.
We performed comprehensive electrophysiological characterization using both whole-cell and single-channel recordings (Fig. 2). WT M-channels (co-expressed KCNQ2/KCNQ3) served as functional benchmarks (Fig. 2b–d), while KCNQ2 homomers provided baseline references for homomeric channel properties. Whole-cell recording results demonstrated that all four concatemer configurations produced current-voltage (I-V) relationships superimposable with WT M-channels, with activation thresholds consistently shifted toward hyperpolarized potentials compared to KCNQ2 homomers (Fig. 2c, d). Quantitative analysis revealed that each concatemer exhibited half-maximal activation voltages (V
1/2) statistically indistinguishable from WT M-channels (
P > 0.05) yet significantly more hyperpolarized than KCNQ2 homomers (
P < 0.001), recapitulating the hallmark subthreshold activation profile of native M-channels
2. To further evaluate their steady-state and kinetic properties, we estimated the apparent gating charges and measured the voltage-driven activation kinetics (Supplementary Fig. S16). The apparent gating charges showed no significant differences across the WT M-channel, the four concatemers, and the KCNQ2 homomer. Furthermore, despite structural variations in their voltage-sensing domains, all concatemers exhibited macroscopic activation kinetics that closely matched the WT M-channel, which were significantly faster than those of homomeric KCNQ2.
Single-channel analyses provided mechanistic corroboration at the molecular level. All concatemer variants displayed unitary current amplitudes (Fig. 2e), single-channel conductance (Fig. 2f), and open probabilities (Fig. 2g) matching WT M-channels within experimental error (
P > 0.05 for all parameters across constructs). To uncover potential fine-scale kinetic distinctions among the different stoichiometric assemblies, we analyzed the dwell-time distributions for both open and closed states (Supplementary Fig. S17). Both closed and open events were typically best fit with a mixture of two exponential distributions. While we observed subtle variations in microscopic kinetics (
e.g., the bimodal open-time distributions showed varying degrees of dispersion among configurations), the distinct concatemers shared broadly similar overall dwell-time profiles with the WT channel. This functional equivalence across distinct stoichiometries demonstrates that each assembly configuration faithfully reproduces the fundamental biophysical properties of native M-channels
17.
The KCNQ3 VSD confers hyperpolarized activation kinetics
The M-channel’s physiological importance stems from its unique ability to activate at subthreshold membrane potentials, a property critical for regulating neuronal excitability
2,4,6. Despite its functional significance, the structural basis for this distinctive gating behavior has remained elusive.
In voltage-gated ion channels, VSD activation is defined by the positions of S4 arginine residues (R1–R6) relative to the charge transfer center, which comprises a conserved phenylalanine (F137 in KCNQ2; F167 in KCNQ3) flanked by two acidic residues (E130/E140 in KCNQ2; E160/E170 in KCNQ3; Fig. 3a; Supplementary Fig. S1)
64. Comparative structural analysis of the
apo M-channel revealed striking differences between subunits: while R4 (R207) in KCNQ2 forms cation–π interactions with F137 and sits slightly above it, KCNQ3’s R4 (R236) resides farther above F167, with R5 (R239) instead forming the cation–π interaction below (Fig. 3a). This conformational disparity suggests that the KCNQ3 VSD adopts a more activated state than the KCNQ2 VSD, indicating that the KCNQ3 subunit is more easily activated at hyperpolarized membrane potentials. These structural observations provide a plausible mechanism for the M-channel’s hyperpolarized activation threshold relative to KCNQ2 homomers.
To test whether KCNQ3’s VSD drives subthreshold activation, we engineered chimeric constructs by swapping VSDs between subunits: VSD2-PD3 (KCNQ2 VSD + KCNQ3 pore) and VSD3-PD2 (KCNQ3 VSD + KCNQ2 pore) (Fig. 3b; Supplementary Fig. S1). Electrophysiological characterization revealed that channels containing only KCNQ2 VSDs (VSD2-PD3 + KCNQ2) exhibited activation thresholds similar to KCNQ2 homomers (Fig. 3c, d). Conversely, channels with exclusively KCNQ3 VSDs (VSD3-PD2 + KCNQ3) activated at even more hyperpolarized potentials than WT M-channels. This finding is consistent with the hyperpolarized activation profile we observed for the homomeric KCNQ3 channel (V1/2 = −33.49 ± 1.87 mV; Supplementary Fig. S18), with the activation threshold of the native heteromer situated at an intermediate level between these two extremes (Fig. 3c, d). These results definitively establish that KCNQ3’s VSD is both necessary and sufficient for the M-channel’s subthreshold activation phenotype.
Identification and characterization of the selective M-channel activator CLM142
The withdrawal of retigabine (Fig. 4b), the first clinically approved M-channel activator, partially due to off-target effects across KCNQ subtypes (Supplementary Fig. S19), highlighted the critical need for selective therapeutics
47, 49, 51, 52, 54-59. Through structure-guided drug design informed by both the M-channel architecture and retigabine's activation mechanism, we developed CLM142, a next-generation activator identified via an integrated virtual screening and electrophysiological validation approach (Fig. 4a; Supplementary Fig. S20; see Materials and Methods for detailed screening procedures).
CLM142 features three key structural elements: (i) a fluorophenyl group linked via (ii) an amide bond to (iii) an indazole core with cyclopropyl and ethynyl substitutions (Fig. 4a). These strategic modifications yielded significant pharmacological improvements, demonstrating a 10-fold greater potency than retigabine on human M-channel (EC50 = 0.19 ± 0.09 μM vs 2.16 ± 0.15 μM for G-V curve shifts) (Fig. 4c–e; Supplementary Fig. S21a). At 1 μM concentration, CLM142 produced a ΔV1/2 shift of −32.49 ± 1.64 mV (Fig. 4c, d) and significantly modulated channel kinetics, accelerating activation (118.09 ± 9.21 ms vs 172.09 ± 17.19 ms at 0 mV; Supplementary Fig. S21b) while slowing deactivation (126.54 ± 12.98 ms vs 19.51 ± 2.72 ms; Supplementary Fig. S21c). Moreover, CLM142 achieves a remarkable subtype specificity (Fig. 4f; Supplementary Fig. S22). Even at a high concentration of 10 μM, CLM142 exerted negligible effects on KCNQ1 activation. For KCNQ4 and KCNQ5, noticeable but minimal hyperpolarizing shifts (~10 mV) were only observed at 10 μM, a concentration over 50 times higher than its EC50 for the M-channel. At 1 μM concentration, where CLM142 produced a massive activation shift for the M-channel, the compound showed negligible effects on these other subtypes. These properties establish CLM142 as a promising therapeutic candidate, combining submicromolar potency with the precise subtype selectivity required to avoid the historical pitfalls of retigabine.
Molecular mechanism for the potent activation of CLM142 on human M-channel
To elucidate its mechanism of action, we determined the cryo-EM structure of CLM142-bound M-channel at 3.1 Å resolution (Fig. 5a, b; Supplementary Figs. S3, S4, and S10). CLM142 occupies the inter-subunit pocket between S5 and S6 helices (Fig. 5c) — the canonical activator binding site shared with retigabine (Supplementary Fig. S19) — but establishes a unique interaction network through its distinct chemical architecture (Fig. 5d)
37, 43. The compound binds in a characteristic orientation where its fluorophenyl tail extends intracellularly while the cyclopropyl/ethynyl moieties project extracellularly (Fig. 5d). Within this pocket, the amide carbonyl forms a crucial hydrogen bond with S303/S342 on S6, while the central indazole core engages in robust π–π stacking with the highly conserved W236/W265 on S5. Simultaneously, the cyclopropyl/ethynyl groups are sequestered within a hydrophobic pocket formed by F240/F269 on S5 and L299/L338-F305/F344 on S6, while the fluorophenyl group makes complementary van der Waals contacts with F304/F343, L221/L250, V225/I254 and L312/L352. This sophisticated interplay of directional hydrogen bonding and extensive hydrophobic complementarity synergistically enables CLM142 to bind and activate the M-channel.
Comparative analysis with retigabine and Ebio2 reveals several distinct structural advantages that provide a rationale for the 10-fold increase in potency exhibited by CLM142 (Supplementary Fig. S23). Most notably, the rigid indazole-amide scaffold of CLM142 directs its extended fluorophenyl tail to anchor into a deep hydrophobic sub-pocket beneath the primary binding site. This critical anchoring space remains completely unutilized by the more flexible retigabine molecule. Furthermore, the highly aromatic indazole ring significantly strengthens the π–π stacking interaction with the S5 tryptophan compared to the simpler triaminobenzene core of earlier activators. This interaction also explains the improved subtype selectivity of CLM142. In KCNQ1, this critical tryptophan is replaced by a leucine residue (L266), which effectively abolishes the CLM142 binding site and explains the lack of drug activity on that subtype.
Despite these significant structural modifications and the superior shape complementarity of its binding pose, CLM142 activates the channel through a mechanism reminiscent of retigabine. The compound promotes PD opening by inducing a clockwise rotation of the S5 and S6 helices when viewed from the intracellular side (Fig. 5d)
37, 43, 65. This mechanical movement is evidenced by the significant displacement of W236 in KCNQ2 and W265 in KCNQ3 (Fig. 5d), confirming that both activators share a common downstream pathway for gate opening regardless of their distinct chemical geometries. This sophisticated interplay of directional hydrogen bonding and deep hydrophobic anchoring synergistically enables CLM142 to stabilize the open state with high potency and specificity.
Open-state structure of M-channel reveals its activation mechanism
To elucidate the structural basis of M-channel activation, we determined the open-state structure of the human M-channel (open-state M-channel) in complex with CLM142 and PIP
2 (Fig. 6a, b; Supplementary Figs. S3, S4, and S11). While maintaining the four CLM142 molecules bound between S5-S6 helices observed in the closed state (Fig. 6b), the intracellular gate formed by L318 in KCNQ2 (L357 in KCNQ3) of the open structure undergoes significant dilation, expanding from less than 1 Å to over 2 Å in radius (Fig. 6c, d). This pore opening results from a coordinated clockwise rotation of S6 helices (viewed intracellularly) initiated at the conserved G313 in KCNQ2 (G352 in KCNQ3) within the GSG motif, coupled with outward displacement of the gate-forming leucine side chains (L318/L357) (Fig. 6e; Supplementary Figs. S1, S24). These movements align with established activation mechanisms in voltage-gated channels
43, 65.
The structure captures four weak densities beneath the VSDs, which we assign as PIP
2 based on three lines of evidence (Fig. 6f): (i) their exclusive appearance in PIP
2-supplemented samples, (ii) the open conformation's dependence on PIP
2 in KCNQ channels
8, 48, 66-68, and (iii) congruence with PIP
2 binding sites in other open-state KCNQ structures
37, 42, 44-46, 69. These lipids engage multiple positively charged residues, including R87/R117 and R89/R119 in the VSD, R213/R242 and R214/R243 in the S4-S5 linker, and K327/K366 in the extended S6 helix through electrostatic interactions, consistent with previous studies
42, 44-46. These interactions support an activation mechanism whereby PIP
2 binding to this basic residue-rich interface couples S4 movement to S6 displacement, facilitating gate opening. The pivotal role of the conserved GSG motif (G313/G352) as a gating hinge is underscored by its position at the initiation point of these conformational changes (Supplementary Fig. S1).
DISCUSSION
Our systematic structural and functional characterization of the human M-channel resolves several long-standing questions in the field while providing new insights into its physiological regulation, pathophysiological mechanism and therapeutic targeting (Supplementary Fig. S25).
Our cryo-EM structures reveal unexpected stoichiometric plasticity in the M-channel, which can adopt all possible KCNQ2:KCNQ3 configurations (M2223, M2233, M2323, and M2333). This remarkable structural flexibility likely stems from the high degree of sequence conservation and structural similarity between KCNQ2 and KCNQ3 subunits. The relative abundance of these assemblies varies across preparations, correlating with differential expression levels of KCNQ2 and KCNQ3. This structural flexibility resolves previous conflicting reports about M-channel composition, where some studies proposed fixed 1:1 stoichiometry while others suggested variable ratios
2, 38-40. The discrepancy likely stems from unaccounted variations in subunit expression levels across experimental systems.
This plasticity may represent an important endogenous mechanism for tuning neuronal excitability, particularly given the established spatiotemporal expression patterns of KCNQ subunits
12, 16, 38, 70, 71. During development, KCNQ2 expression stabilizes rapidly while KCNQ3 shows gradual accumulation to its peak levels
16, 38, 70. Moreover, the relative proportions of KCNQ2 and KCNQ3 expression vary significantly across different brain regions
12, 71, 72. Such systematic variations in subunit availability would naturally produce different distributions of M-channel assemblies and KCNQ2 homomers, creating a spectrum of channel populations fine-tuned to specific physiological requirements across neuronal circuits and developmental stages.
Despite structural heterogeneity, all four configurations exhibited remarkably similar electrophysiological properties. This surprising functional homogeneity may reflect either: (i) dominant activation by the first responsive subunit as proposed previously, or (ii) limitations in detecting subtle gating differences with current methodologies
73. If confirmed, such functional equivalence would imply that stoichiometric variation primarily modulates channel density rather than biophysical properties — a possibility that warrants further investigation.
Through chimeric constructs and structural comparison, we established that the KCNQ3 VSD drives the characteristic subthreshold activation of the M-channel. While our structures reveal distinct conformational states for the KCNQ2 and KCNQ3 VSDs, the current local resolution of these peripheral regions limits a definitive atomistic comparison of the entire gating charge pathway. Future high-resolution studies or complementary computational approaches will be necessary to resolve the precise molecular determinants and side-chain interactions that drive these fundamental differences in voltage sensitivity.
Our development of CLM142 addresses the critical need for specific M-channel therapeutics following retigabine’s withdrawal
49-52, 54, 55, 57-60, 74, 75. This optimized compound demonstrates 10-fold greater potency and superior subtype specificity, potentially mitigating the bladder toxicity linked to the non-selective activation of KCNQ1, KCNQ4, and KCNQ5
57. Notably, the retinal discoloration that prompted retigabine’s withdrawal is widely attributed to metabolic liabilities rather than a lack of channel selectivity
60. While CLM142 achieves the necessary structural specificity to avoid common off-target effects, future clinical development must also address these metabolic risks to ensure long-term safety. Ensuring the success of next-generation activators will require a dual focus on both structural selectivity and metabolic stability.
Our open-state structure suggests that PIP
2 mediates VSD-pore coupling through a conserved basic residue interface (R87/R117, R89/R119, R213/R242, R214/R243, K327/K366), consistent with prior studies
42, 44-46. Notably, we resolved only a single PIP
2 density per subunit, which matches the P2 site identified in a previous KCNQ4 structure
44. While the current densities were relatively weak and precluded the identification of additional binding sites, future investigations using PIP
2-incorporated nanodiscs or liposomes may better capture these critical interactions and reveal further mechanistic details. Whether a secondary PIP
2 binding site exists for the heteromeric M-channel remains an important question for future research.
It is worth noting that our findings are subject to several limitations. First, both the structural observations and functional characterizations are based on recombinant proteins expressed in heterologous systems, which may not fully recapitulate the native neuronal environment. Second, the enhanced potency and subtype specificity of CLM142, while rigorously established under patch-clamp conditions, require further validation in physiological disease models and clinical studies to substantiate its therapeutic superiority over retigabine.
By elucidating the M-channel’s structural principles, we have: (i) reconciled conflicting stoichiometry models through demonstration of expression-dependent assembly, (ii) provided the first structural explanation for subthreshold activation, and (iii) developed CLM142 as a precision therapeutic candidate. These advances establish a framework for understanding M-channel physiology and developing targeted treatments for epilepsy and related disorders.
MATERIALS AND METHODS
Cell culture and transient transfection
Baculovirus was generated using Sf9 insect cells cultured in Sf900-II SFM medium (GIBCO) at 28 °C. Recombinant proteins were expressed in HEK293F cells maintained in SMM 293-T II serum-free medium (Sino Biological) at 37 °C in a humidified incubator with 5% CO2. Cells were infected when the density reached 2.0–3.0 × 106 cells/mL. HEK293T and CHO-K1 cells were used for electrophysiological recordings. Whole-cell patch-clamp recordings of the engineered concatemers and chimeric constructs were performed using HEK293T cells. To provide microscopic gating insights and cross-validate the whole-cell findings, single-channel recordings were conducted using CHO-K1 cells. Additionally, all pharmacological assays evaluating the effects of the activators CLM142 and retigabine were performed in the CHO-K1 expression system. Both cell lines yielded highly consistent results, ensuring the robustness and reproducibility of the functional data across different mammalian expression environments. HEK293T cells were cultured in Dulbecco’s Modified Eagle Medium (DMEM; BI) supplemented with 4.5 mg/mL glucose and 10% (v/v) fetal bovine serum (FBS; BI). CHO-K1 cells were maintained in DMEM/F12 medium (Gibco) supplemented with 10% FBS and 1% (v/v) penicillin–streptomycin (Gibco). All mammalian cells were incubated at 37 °C in a 5% CO2 atmosphere. Plasmids were transfected when the cells reached ~70% confluency. HEK293T or CHO-K1 cells were transiently transfected with 2.5 μg plasmid DNA per well using Lipofectamine 3000 (Invitrogen) following the manufacturer’s instructions.
Cloning, expression, and purification of M-Channel
Gene encoding human KCNQ2 (UniProt: O43526) and KCNQ3 (UniProt: O43525) were synthesized by GENEWIZ and confirmed by sequencing. Full-length sequences of KCNQ2 and KCNQ3 were cloned into a pEGBacMam expression vector with an N-terminal 8×His-GFP-tag and FLAG-tag, respectively
76. Expression products of the two plasmids were NHis-GFP-KCNQ2 (WT) and NFlag-KCNQ3 (WT). To improve stability and increase production, amino acids (353–533 AA) between HA and HB of the truncated KCNQ2 (64–674 AA) were replaced by a GS-linker (GGGSGGGS); KCNQ3 was also truncated and kept the 93–691AA. Structures, including
apo, CLM142-bound, and open-state M-channels, were obtained using plasmids engineered as described above. All plasmids were confirmed by DNA sequencing.
To express M-channels, recombinant baculoviruses were generated using the Bac-to-Bac system (ThermoFisher Scientific) as previously described
76. When cell density reached approximately 2.5 × 10
6 cells/mL, P2 viruses of KCNQ2 and KCNQ3 were co-infected into the cells at a ratio of 1:10 (v/v). After 8 h of culturing, 10 mM sodium butyrate (Sigma) was added to the medium to boost protein expression at 30 °C. Cells were harvested after ~48 h of culturing and frozen in liquid nitrogen before being stored in a −80 °C refrigerator for future use.
For the purification of the M-channel, 6 L cells were solubilized in extraction buffer containing 20 mM Tris-HCl, pH 8.0, 200 mM KCl, 1% DDM/CHS (10:1, w/w; Anatrace), 2 μM leupeptin, 1 μM pepstatin A, 1 μg/mL aprotinin, 1 mM PMSF for 2 h at 4 °C. Solubilized cell lysis were clarified by centrifugation at 13,000 rpm for 1 h. The resulting supernatant was applied to Anti-Flag G1 Affinity Resin (GenScript), and the loaded resin was washed with wash buffer containing 20 mM Tris-HCl pH 8.0, 200 mM KCl, 1mM PMSF and 0.03% GDN (Anatrace). The protein was eluted with 200 μg/mL FLAG peptide in wash buffer and then loaded onto High Affinity Ni-Charged Resin (QIAGEN). The resin was washed with wash buffer supplemented with 30 mM imidazole, and the protein was eluted with 300 mM imidazole. The eluent was concentrated with a 100 kDa MWCO Amicon Ultra-4 centrifugal filter (Millipore) and then applied to size-exclusion chromatography (Superose 6 Increase, 10/300 GL, GE Healthcare) in buffer containing 20 mM Tris-HCl pH 8.0, 150 mM KCl, 1mM PMSF and 0.03% GDN. The peak fractions containing the KCNQ2/KCNQ3 complex were concentrated to about 12 mg/mL for further experiments. To prepare the sample of the M-channel with CLM142, concentrated protein was incubated with 1 mM CLM142 for at least 30 min at 4 °C prior to cryo-EM sample preparation. To obtain the open conformation of the M-channel, the purified protein was incubated with 1 mM PIP2 and 1mM CLM142. CLM142 was synthesized by the laboratory of Professor Huaiyu Yang (East China Normal University). The PIP2 we used is 1,2-dioctanoyl-sn-glycero-3-phospho-(1’-myo-inositol-4’,5’-bisphosphate) (ammonium salt) purchased from Avanti.
Cryo-EM sample preparation and data collection
To prepare cryo-EM sample, the concentrated complex mixture (3.5 μL) was placed on glow-discharged holey carbon grids (Quantifoil Au R1.2/1.3), which were blotted for 3.5 s and flash-frozen in liquid ethane cooled by liquid nitrogen with Vitrobot (Mark IV, ThermoFisher Scientific). The grids were loaded onto a 300 kV Titan Krios (ThermoFisher Scientific) equipped with a K3 Summit detector (Gatan) and GIF Quantum energy filter. Automated data collection was performed using EPU software (ThermoFisher Scientific) in super-resolution mode at a nominal magnification of 81,000×, with a slit width of 20 eV on the energy filter. A defocus series ranging from −1.5 μm to −2.0 μm was used. Each stack was exposed for 2.56 s with an exposure time of 0.08 s per frame, resulting in a total of 32 frames per stack and the total dose was approximately 50 e
–/Å
2 for each stack. The stacks were motion corrected with MotionCor2 and binned 2-fold, resulting in a pixel size of 1.087 Å/pixel
77. Meanwhile, dose weighting was performed
78. The defocus values were estimated with Gctf
79.
Cryo-EM data processing
The data processing workflow for the M-channel is illustrated in Supplementary Figs. S5–S7. All steps were conducted in CryoSPARC v4.6.2
80. For WT M-channel, a total of 5,393 micrographs were collected. Using the template picker, 5,223,566 particles were automatically selected. After particle extraction, multiple rounds of 2D classification,
ab initio reconstruction, and heterogeneous refinement were performed, resulting in a dataset of 727,947 selected particles. These particles were subjected to non-uniform refinement with C1 symmetry. To distinguish between the KCNQ2 and KCNQ3 components, non-uniform refinement with C4 symmetry was performed to align the symmetry axis along the
Z-axis. A new particle stack was subsequently generated via symmetry expansion using C4 symmetry. The expanded dataset was classified into two distinct VSD states through multiple rounds of 3D classification, using a focused mask on the VSD region. Particles corresponding to each state were grouped based on their stoichiometric ratios by intersecting different VSD classifications. The resulting particle stacks were further processed using heterogeneous refinement followed by non-uniform refinement to enhance map quality and improve resolution. A total of 16,214, 15,162, and 27,141 micrographs were collected for
apo, CLM142-bound, and open-state M-channels, respectively. Two strategies were applied for the data processing. The first one followed the same procedure as described for the WT M-channel. The second conducted 3D classification using four separate masks, each enclosing an individual VSD domain. Particles of identical stoichiometries were combined and refined via non-uniform refinement. CryoSieve was used to exclude unwanted particles
81. The resolution was estimated with the gold-standard Fourier shell correlation 0.143 criterion with high resolution noise substitution
82, 83.
Model building and structure refinement
The AlphaFold-predicted models of KCNQ3 (AlphaFold DB ID: O43525) and KCNQ2 (AlphaFold DB ID: O43526) were initially docked into the final cryo-EM density maps using UCSF ChimeraX
84-86. Manual model building and adjustment were subsequently performed in Coot v0.9.8.1, with careful attention to the chemical properties of individual amino acid residues
87. Due to insufficient electron density, the N-terminal and C-terminal regions were not modeled. Several putative lipid molecules were tentatively placed into the observed densities.
Real-space refinement was performed using phenix.real_space_refine in PHENIX 1.20, with secondary structure and geometry restraints applied throughout
88. To avoid overfitting, gold-standard refinement was employed by alternately refining against one of the two independently calculated half-maps and validating against the other. Final refinement statistics and map quality metrics are summarized in Supplementary Tables S1–S5.
Design of KCNQ2 and KCNQ3 concatemers
Concatemeric constructs were created by the sequential insertion of individual subunits into a pEGBacMam expression vector with an N-terminal 8×His-GFP-tag. Every subunit of KCNQ2 or KCNQ3 was WT without deletions or truncations. Four concatemers were generated: M23, M2333, M2233, and M2223. Adjacent subunits were linked by a GS-linker (GGGGSSGGGGSSGGGGSS). The accuracy of the final sequence was confirmed by both insert digestion and third-generation long-read DNA sequencing.
Quantification of average M-channel stoichiometry
To quantify the average stoichiometry across varying subunit expression levels, we performed systematic transfection experiments using KCNQ2-GFP and KCNQ3 plasmids at molar ratios of 1:1, 1:4, and 1:9 while the total amounts remain constant. Following membrane protein extraction and purification, the resulting heteromeric assemblies were analyzed via size-exclusion chromatography with simultaneous monitoring of total protein concentration (UV absorbance at 280 nm) and KCNQ2 content (GFP fluorescence).
For the quantification of these signals, we utilized the ratio of the relative peak heights. To ensure accuracy, the baseline-corrected maximum height for both the GFP fluorescence and UV absorbance signals was determined at the peak elution volume. The resulting ratio of GFP fluorescence (measured in mV) to UV absorbance (measured in mAu) served as a relative proxy for the average number of KCNQ2 subunits incorporated per assembled tetramer.
Virtual screening and identification of CLM142
To identify novel and selective M-channel activators, we conducted a structure-guided virtual screening campaign using an in-house library comprising over 600 targeted KCNQ channel modulators. Molecular docking was performed by targeting the canonical activator binding pocket, which is structurally defined by the S4-S5 linker, the S5 helix, and the S6 helix of one subunit, along with the S6 helix of the adjacent subunit (Supplementary Fig. S20). Compounds were evaluated and ranked based on their docking scores and predicted binding conformations. Following this virtual screening, the 80 top-ranked candidates were selected for empirical validation using whole-cell patch-clamp electrophysiology. This functional screening successfully identified 9 distinct hit molecules that acted as M-channel agonists. Among these validated hits, compound 7 exhibited the highest potency for channel activation and was subsequently renamed CLM142 for comprehensive structural and pharmacological characterization.
Whole-cell electrophysiology
For the whole-cell electrophysiological experiments of the KCNQ2/KCNQ3 concatemers and chimeric constructs, the whole-cell K+ currents were recorded in HEK293T cells using an EPC-10 amplifier with Patchmaster 2x92 software (HEKA Elektronik) and glass micropipettes (2.5–4 MΩ) made by P-1000 pipette puller (Sutter Instrument). To record the KCNQ current, the pipette solution contained 100 mM KCl, 50 mM KF, 5 mM EGTA, 10 mM HEPES, adjusted to pH 7.2 with KOH, and the bath contained 130 mM NaCl, 20 mM KCl, 1 mM CaCl2, 2 mM MgCl2, 5 mM D-Glucose, 10 mM HEPES, adjusted to pH 7.2 with NaOH. Fitmaster 2x92 (HEKA Elektronik) and Prism 10.1.2 (GraphPad Software) were used for data analysis.
To record the currents, a series of 2,000 ms depolarizing steps (holding potential –80 mV) was applied from –90 mV to +80 mV in 10 mV increments. Then, the cells were stimulated by –120 mV for 250 ms to obtain tail currents. The linear component of leak current and capacitive transients was subtracted using the –P/4 procedure, and the voltage errors were minimized using series resistance compensation. The normalized tail current amplitude of the activation curves was plotted against step potentials and fitted by the Boltzmann Sigmoidal function:
Where I refers to the tail current, V1/2 refers to the voltage for half-maximal activation, Vm refers to the test potential, and k refers to the slope of the curve. The slope factor (k) derived from the fits is summarized to estimate the apparent gating charge (zapp) via the relationship zapp = RT/(F·k), where R refers to the gas constant, T refers to the absolute temperature, and F refers to the Faraday constant. The Shapiro-Wilk test and Bartlett’s test were used to verify the normality and homogeneity of our data. A one-way ANOVA was employed to assess statistically significant differences between groups.
Whole-cell K+ currents of the M-channel with CLM142/retigabine were recorded at room temperature using HEKA EPC-10 amplifier on CHO-K1 cells. The series resistance (Rs) was 70–80% compensated using the internal compensation program of the amplifier. The patch pipettes (World Precision Instruments) were pulled to a resistance of 3–5 MΩ. The pipette solution consisted of 140 mM KCl, 3 mM MgCl2, 5 mM EGTA, 5 mM K2ATP and 10 mM HEPES (pH 7.3, adjusted with KOH, all from Sigma). The extracellular solution contained 145 mM NaCl, 5 mM KCl, 1 mM CaCl2, 3 mM MgCl2, 10 mM D-glucose and 10 mM HEPES (pH 7.3, adjusted with NaOH).
To construct M-channel activation curves with CLM142, cells were held at –80 mV and currents were elicited by a series of 2,000-ms depolarizing steps from –90 mV to +60 mV in 10-mV increments, followed by –120 mV to record tail currents. Activation curves were fitted by the Boltzmann Sigmoidal function. Also, this protocol was used to study the M-channel channel activation kinetics. The M-channel deactivation kinetics were measured by tail currents elicited by a hyperpolarizing voltage of –120 mV for 1,200 ms after a 2,000-ms depolarized potential of +50 mV, with a holding potential of –80 mV. The activation and deactivation traces were fitted to a single exponential function: I = A × [1 – exp(–t/tau)] + I0, where I is the current, I0 is the steady-state amplitude of the current, A is the difference between the peak and steady-state current amplitudes; t is time; and tau is the time constant. The concentration-response curves were analyzed by three parameters logistics regression model. For recordings of KCNQ1, KCNQ4, and KCNQ5 channel currents, pipettes were filled with the intracellular solution of the following composition: 150 mM KCl, 3 mM MgCl2, 5 mM EGTA and 10 mM HEPES (pH 7.3, adjusted with KOH). Extracellular solution contained 145 mM NaCl, 5 mM KCl, 1 mM CaCl2, 3 mM MgCl2 and 10 mM HEPES (pH 7.4, adjusted with NaOH). The methods used to record KCNQ channels were the same as those used to record M-channel currents.
Processed by Clampfit 10.4, patch-clamp data were analyzed in GraphPad Prism 8.0.2. An unpaired two-tailed Student’s t-test was used to determine significance between groups. n.s. indicates no significance. *P < 0.05, **P < 0.01, ***P < 0.001, and ****P < 0.0001. All data are presented as mean ± SEM.
Inside-out single-channel recordings
Inside-out recordings were performed 48 h after transfection in CHO-K1 cells using EPC10-USB amplifier (HEKA). Patch recordings were digitized at 10 kHz and filtered at 2 kHz. The bath solution contained 175 mM KCl, 4 mM MgCl2, and 10 mM HEPES (pH 7.4, adjusted with KOH). The pipette solution was 150 mM NaCl, 5 mM KCl, 1 mM MgCl2, and 10 mM HEPES (pH 7.4, adjusted with NaOH). The patch pipette was pulled to a resistance of 8–12 MΩ. Single-channel statistical analysis was conducted by Clampfit 10.4 software. All-point histograms were fitted with Gaussian functions to obtain the mean single-channel current (i). Single-channel conductance (γ) was obtained by the equation γ = i/(V − VE), where V is the test potential and VE is the reversal potential of potassium. The dwell-time histograms were constructed by plotting the logarithm of the dwell times against the square root transformation of the event counts and were fitted with a mixture of two exponential distributions. Events shorter than 2 ms were ignored from analysis.
Synthetic scheme of CLM142
Chemical formula
Synthetic Route
Step 1: 7-cyclopropyl-1H-indazol-3-amine
7-bromo-1H-indazol-3-amine (2, 2.00 g, 9.43 mmol) was dissolved in toluene (20 mL) and water (6 mL). Cyclopropylboronic acid (5.67 g, 66.01 mmol), tricyclohexylphosphine (0.53 g, 1.89 mmol), potassium phosphate (6.01 g, 28.29 mmol) and palladium acetate (0.42 g, 1.89 mmol) were added sequentially. Under nitrogen protection, the mixture was heated to 100 °C and stirred for 16 h. TLC (PE:EA = 1:1, Rf = 0.1) indicated the complete consumption of the starting material. The reaction solution was cooled to room temperature, filtered through Celite, quenched with water (10 mL), and extracted with ethyl acetate (10 mL × 3). The combined organic phases were washed with saturated brine (20 mL × 3), dried over anhydrous sodium sulfate, and concentrated. The crude product was purified by silica gel column chromatography (PE:EA = 2:1–2:3) to afford the title compound 3 as a white solid (0.69 g, yield 42%).
LC-MS: m/z = 174.2 [M+H]+
Step 2: N-(7-cyclopropyl-1H-indazol-3-yl)-4-fluorobenzamide
Compound 3 (380 mg, 2.19 mmol) was dissolved in pyridine (2 mL), cooled to 0 °C, and 4-fluorobenzoyl chloride (347 mg, 2.19 mmol) was added dropwise. The mixture was stirred at this temperature for 30 min. TLC (PE:EA = 1:1, Rf = 0.4) showed the complete reaction of the starting material. The reaction was quenched with water (10 mL) and extracted with ethyl acetate (10 mL × 3). The combined organic phases were washed with saturated brine (20 mL × 3), dried over anhydrous sodium sulfate, and concentrated. The crude product was purified by silica gel column chromatography (PE:EA = 5:1–2:1) to obtain the title Compound 4 as a brown solid (325 mg, yield 50%).
LC-MS: m/z = 296.1 [M+H]+
Step 3: N-(7-cyclopropyl-1-(prop-2-yn-1-yl)-1H-indazol-3-yl)-4-fluorobenzamide
Compound 4 (80 mg, 0.27 mmol) was dissolved in N,N-dimethylformamide (2 mL). 3-bromopropyne (39 mg, 0.33 mmol) and potassium carbonate (75 mg, 0.54 mmol) were added at room temperature, and the mixture was stirred at room temperature for 16 h. TLC (PE:EA = 2:1, Rf = 0.4) indicated a small amount of starting material remaining. The reaction was quenched with water (10 mL) and extracted with ethyl acetate (10 mL × 3). The combined organic phases were washed with saturated brine (20 mL × 3), dried over anhydrous sodium sulfate, and concentrated. Purification by preparative HPLC (TFA) gave the title compound CLM142 (1) as a white solid (18 mg, yield 20%).
LC-MS: m/z = 334.2 [M+H]+
HPLC:99.10% purity, 220 nm
1HNMR (400 MHz, DMSO-d6) δ 10.88 (s, 1H), 8.15 (dd, J = 8.4, 5.6 Hz, 2H), 7.53 (d, J = 8.4 Hz, 1H), 7.39 (t, J = 8.8 Hz, 2H), 7.17 (d, J = 7.2 Hz, 1H), 7.04 (t, J = 8.0 Hz, 1H), 5.56 (d, J = 1.6 Hz, 2H), 3.44 (s, 1H), 2.49-2.44 (m, 1H), 1.10-1.05 (m, 2H), 0.89-0.83 (m, 2H).
DATA AVAILABILITY
Atomic coordinates and EM maps of M-channels reported in this study (WT M2223: 24VL and EMD-69843; WT M2233: 24VM and EMD-69844; WT M2323: 24VN and EMD-69845; WT M2333: 24VO and EMD-69846; Apo M2223 (no symmetry expansion): 24VI and EMD-69840; Apo M2223 (symmetry expansion): 24VP and EMD-69847; Apo M2233: 24VQ and EMD-69848; Apo M2323: 24VR and EMD-69849; Apo M2333: 24VS and EMD-69850; CLM142-bound M2223 (no symmetry expansion): 24VT and EMD-69851; CLM142-bound M2223 (symmetry expansion): 24VJ and EMD-69841; CLM142-bound M2233: 24VU and EMD-69852; CLM142-bound M2323: 24VV and EMD-69853; CLM142-bound M2333: 24VW and EMD-69854; Open-state M2223 (no symmetry expansion): 24VK and EMD-69842; Open-state M2223 (symmetry expansion): 24VX and EMD-69855; Open-state M2233: 24VY and EMD-69856; Open-state M2323: 24VZ and EMD-69857; Open-state M2333: 24WA and EMD-69858) have been deposited in the Protein Data Bank and the Electron Microscopy Data Bank.
The Author(s) 2026. Published by Higher Education Press. This is an Open Access article distributed under the terms of the CC BY license (https://creativecommons.org/licenses/by/4.0/).
