Correcting acetylcholine receptor biophysical defects in congenital myasthenia

Abstract

Voluntary muscle contraction is triggered by the neurotransmitter acetylcholine binding its receptors on the postsynaptic membrane of the neuromuscular junction, opening ion channels that allow cation influx and initiate depolarization^1-3. Mutations in muscle acetylcholine receptors disrupt this process by either impairing (fast-channel) or prolonging (slow-channel) channel openings^1,4. These defects cause congenital myasthenic syndromes (CMS), characterized by severe muscle weakness often present at birth, and in some cases progressing to paralysis and death^5,6. The structural mechanisms underlying these pathogenic defects and their pharmacological correction remain unknown. Using cryo-electron microscopy, chemical biology, and electrophysiology, we determined structures and functional consequences of representative CMS mutants with and without drugs. In fast-channel disease, we discovered a cryptic allosteric site targeted by positive modulators that restore gating in a mutation-specific manner. In slow-channel disease, quinidine, fluoxetine, and reboxetine act as pore blockers; notably the antidepressant reboxetine selectively blocks desensitized receptors in a mutation-independent fashion, suggesting repurposing potential. Mechanistically, fast-channel mutations uncouple agonist binding from gating, whereas slow-channel mutations stabilize an abnormally widened, desensitized-like pore. These findings reveal unifying principles of CMS pathogenesis and provide a framework for precision therapies.

Introduction

Skeletal muscle contraction is initiated by nicotinic acetylcholine receptors located on the postsynaptic muscle membrane, which respond to acetylcholine (ACh) released by motor neurons at the neuromuscular junction⁷. ACh binding triggers muscle ACh receptor channel opening, allowing cation influx, depolarization of the motor endplate, and ultimately muscle fiber contraction. Genetic mutations in muscle ACh receptor subunits can disrupt contraction by decreasing receptor at the membrane, called primary receptor deficiency, or by altering how efficiently the ACh receptor’s ion channel opens and how long it remains open, leading to a group of inherited disorders collectively known as congenital myasthenic syndromes (CMS)^8,9. These kinetic abnormalities give rise to two major clinical subtypes of CMS: slow-channel syndrome and fast-channel syndrome^1,10.

While both CMS types cause similar clinical symptoms of muscle weakness, their underlying pathogenic mechanisms and treatments are fundamentally different. Fast-channel CMS results from loss-of-function mutations leading to abnormally brief and infrequent ACh receptor channel openings, thus resulting in impaired synaptic transmission¹¹. Unfortunately, at present no therapeutics directly target fast-channel ACh receptor mutants to correct the kinetic defect. Treatments focus on increasing synaptic ACh concentration using acetylcholinesterase inhibitors^12-14 and promoting ACh release from motor neurons^14,15; however, these therapies are ineffective or even detrimental in some patients due to broad enhancement of cholinergic signaling across the nervous system^16,17. Positive allosteric modulators (PAMs) were proposed to be an ideal treatment for fast-channel CMS patients, as these could rescue ACh receptor defects without directly activating the channel^3,18,19. However, muscle ACh receptor PAMs remain unexplored both mechanistically and in the clinic. By contrast, slow-channel CMS is caused by gain-of-function mutations that prolong channel opening episodes, leading to sustained depolarization, calcium-mediated endplate damage, and progressive myopathy^9,20. Quinidine and fluoxetine are first-line treatments in slow-channel CMS^21-23; these act as channel blockers^21,22,24, have severe side effects^8,25,26, and their structural mechanisms for correcting defective receptors remain largely unknown.

The muscle ACh receptor is a pentameric ligand-gated ion channel composed of five homologous transmembrane subunits: two α1 subunits, and one each of β1, δ and ε. Each subunit contains an extracellular domain (ECD), a transmembrane domain (TMD, M1–M4), and an intracellular domain (ICD). The receptor contains two ACh-binding pockets at its α-ε and α-δ ECD interfaces^2,27,28. Currently, more than 50 missense mutations that alter channel kinetics have been discovered, spanning all receptor subunits, though most are found in ε^2,18,29. Structural information for the human muscle ACh receptor emerged only in the past year^18,30. CMS mutations interestingly cluster within distinct structural domains based on kinetic defect: fast-channel mutants concentrate in the ECD and ICD, while slow-channel mutants concentrate in the TMD, and especially in the M2 helices that line the channel pore^2,18,31. While the locations of pathogenic mutations are known, how they give rise to disease has not been directly studied at the level of receptor structure.

Here, we define CMS disease and therapeutic mechanisms through comprehensive structural and electrophysiological analyses. We determined 12 high-resolution structures of five representative fast- and slow-channel CMS ACh receptor mutants. We discovered how a first-in-class PAM rescues fast-channel mutant function through a previously unknown binding pocket. Further, we defined structural mechanisms by which both quinidine and fluoxetine correct the overly long channel opening in slow-channel CMS and revealed the structural and functional basis of block by reboxetine, which is already approved for treating depression, and could potentially be repurposed to treat CMS patients. Integrating these findings, we uncovered how specific drugs do or do not correct the defects caused by specific patient mutations, offering a route to personalized medicine for CMS. Finally, we present a universal structural mechanism by which CMS mutations cause their pathogenic kinetic defects.

Fast channel CMS defect correction

Fast channel CMS mutations reduce the incidence of channel openings, their durations, or a combination of the two, producing a reduced post-synaptic response too small to elicit an action potential, thereby impairing muscle contraction¹. Thus, a central therapeutic challenge is to identify strategies that promote channel openings without directly activating the receptor³². PAMs represent a promising approach, as they can promote channel opening without altering the time course of ACh signaling^19,33. To explore this therapeutic route, we synthesized and tested two candidate PAMs, XG-590 and its derivative EC-216 (Fig. 1a, Supplementary information). XG-590, previously called AS3513678, increased ACh-induced current in human myoblasts³⁴. We substituted the fluorophenyl group of XG-590 with a pyrimidine to generate EC-216 in an effort to assess whether or not subtle structural differences would influence PAM activity (Fig. 1a). We first evaluated the effects of both compounds on the wild-type muscle receptor using whole-cell patch-clamp electrophysiology. We found that both XG-590 and EC-216 increased ACh-evoked currents by 2-3-fold, confirming their ability to act as PAMs (Fig. 1b and Extended Data Fig. 1a).

Fig 1. CMS correctors potentiate the muscle ACh receptor through a membrane pocket.

a, Chemical structures of XG-590 and EC-216. Red arrows indicate differences.

b, Representative whole-cell electrophysiology of XG-590 potentiation on the wild-type muscle receptor and the statistics of XG-590 and EC-216 potentiation; the unpaired two-tailed t-test was used; data are represented as mean ± s.e.m. of biological replicates (n = 4 cells).

c, Single channel currents in the presence of 300 μM ACh and the indicated modulators (8 kHz Gaussian filter). O, open; C, close.

d, Logarithmically binned burst-duration histograms fitted by the sum of exponentials. Drug concentrations: 0.3 μM for XG-590 and 3 μM for EC-216. Fractional potentiation, FP, is the fraction of the open time within the prolonged exponential component (Extended Data Fig. 7 and Supplementary Information). In multiple FP determinations yielded the following means and SD: ACh alone, 0.27 ± 0.02 (n = 2 biological replicates).; ACh plus 3 μM EC-216, 0.72 ± 0.07 (n = 4), ACh plus 0.3 μM XG-590, 0.79 ± 0.03 (n = 3).

e, Channel reopening probability of εP121L mutant with increasing modulator concentration. Plots of channel reopening probability are fitted by a single exponential in the absence of drug, and a double exponential in the presence of drug. For both XG-590 and EC-216, comparing profiles, using the extra sum of squares F-test, without versus with drug yielded P < 0.0001, indicating a significant difference. Predicted reopening profile for the wild-type receptor as described in the Receptor Activation section of Supplementary information.

f, Cryo-EM density map of εP121L ACh receptor mutant bound to XG-590. Mutation site and XG-590 density details are shown.

g, XG-590 binds to a pocket in the β subunit transmembrane region.

h, XG-590 interactions in the β TMD binding site. Black dashed lines indicate polar interactions.

i, Superposition of εP121L structures in bound by ACh alone (white) and with ACh plus XG-590 (colored by subunit) reveals how drug binding affects receptor conformation.

j, The β subunit TMD binding pocket is occluded by a lipid tail in the absence of PAMs.

To test whether these PAMs correct fast-channel defects, we focused on a representative patient mutation εP121L, the first mutant in this class characterized in mechanistic detail³⁵. This variant is highly prevalent among CMS patients¹⁴ and produces very small miniature end plate potentials due to reduced ACh binding affinity and a markedly decreased rate of channel opening³⁵ (Fig. 1c). Patch-clamp recordings revealed that both PAMs prolonged individual channel openings (Fig. 1c and Extended Data Fig. 4a, 8b) as well as bursts of openings (Fig. 1d and Extended Data Fig. 4a), defined as a series of openings occurring in rapid succession (Extended Data Fig 7a-c). To quantify the effects on burst kinetics, for each burst we measured the number of channel reopenings following the initial closing event. Both PAMs increased channel reopening in a dose-dependent manner, with EC-216 exerting a greater effect than XG-590 (Fig. 1e). Comparison of these reopening profiles with those predicted for wild type indicated that EC-216 best approaches wild type reopening kinetics (see “Receptor activation mechanism and the effects of modulators and inhibitors” section in Supplementary Information). As a further measure of potentiation, we classified bursts as either potentiated or un-potentiated and found that both PAMs increased the fraction of potentiated (FP) bursts in a dose-dependent manner (Extended Data Fig. 7d). Together, these findings demonstrate that EC-216 and XG-590 ameliorate several functional deficits of the fast-channel mutant receptor.

To understand how these PAMs correct εP121L channel defects, we determined the structure of the εP121L mutant muscle ACh receptor bound to XG-590 in the presence of ACh at 2.4 Å resolution (Fig. 1f, Extended Data Fig. 2, 3, Table 1). Strikingly, we observed two inequivalent binding sites in the transmembrane domain. The primary, well-defined site is located in a previously unknown pocket formed by the extracellular ends of β M1, M3, and M4 helices together with the β subunit Cys-loop (Fig. 1f-h). A weaker, homologous site is present in the ε subunit (Extended Data Fig. 1b, c). At the β site, XG-590 is stabilized by both polar and extensive hydrophobic interactions. The sulfonamide of XG-590’s 1,2-benzothiazine moiety interacts with the M1 helix backbone through an H-bond, while the fluoroarene moiety is oriented deeply into a proximal pocket surrounded by nine hydrophobic residues from β TMD helices (Fig. 1h). There, F137 on the Cys-loop forms a π−π interaction with the fluoroarene group. In the analogous ε site, residue substitutions around the pocket lead to fewer interactions with XG-590 (Extended Data Fig. 1b, c), which may result in the weaker density corresponding to its fluoroarene moiety.

To further clarify the modulatory mechanism, we determined a companion structure of the εP121L receptor bound only to ACh at 2.5 Å resolution (Extended Data Fig. 2, 3). While the overall conformation was similar (RMSD < 1 Å), local rearrangements were evident upon PAM binding: the β Cys-loop, M1 helix, and the adjacent δ M2-M3 loop shift downward, and β M4 F464 flips outward to accommodate the fluoroarene moiety, giving access to the cryptic site (Fig. 1i). These conformational changes suggest that XG-590 enhances activation by promoting coupling between extracellular and transmembrane domains; large changes in interdomain contacts were not observed but these may be more evident in an activated state. Interestingly, in the ACh-only structure, the β site was partially occluded by a lipid tail (Fig. 1j), suggesting both a drug-induced opening of the pocket for PAM binding as well as competition with bound lipid.

Modulators differ in fast channel CMS

Muscle weakness in fast-channel CMS typically begins in infancy^3,9. Disease severity often increases with age and generally correlates with the degree to which a mutation impairs channel gating efficiency¹¹. More severe gating defects are associated with earlier onset and more pronounced respiratory or bulbar involvement^3,5. To assess the generality of PAM activity across diverse channel defects, we examined five additional fast-channel CMS mutants located in distinct regions of the receptor (Fig. 2a and Extended Data Fig. 8).

Fig 2. CMS correctors act in a patient mutation-specific manner.

a, Structural illustration of five fast-channel mutant sites in different regions of a receptor subunit. Mutations shown as yellow spheres.

b-f, Single channel currents in the presence of 300 μM ACh and the indicated modulators (8 kHz Gaussian filter). O, open; C, closed. b and c, plots of channel reopening probability are fitted by a single exponential in the absence of drug, and a double exponential in the presence of drug. d-f, plots of channel reopening probability are fitted by a single exponential in the both the absence and presence of drug. For the mutants αV285I, αV132L, εD175N, comparing profiles, using the extra sum of squares F-test, without versus with drug yielded P < 0.0001, indicating a significant difference. For αV188M, comparing without versus with drug P = 0.0001 for EC-216 and P = 0.0004 for XG-590. For εW55R comparing without versus with drug P = 0.0025 for EC-216 and P = 0.411 for XG-590.

Patients harboring the αV285I mutant in the M3 transmembrane helix present with moderate weakness from birth, attributed to a decreased channel opening rate and increased channel closing rate³⁶. For this mutant, EC-216 exerted the strongest effects, markedly prolonging mean burst duration and increasing channel reopening, whereas XG-590 produced only minor enhancement of these parameters (Fig. 2b, Extended Data Fig. 4b, 8c).

In contrast, XG-590 was more effective for two mutations in the extracellular domain. The εD175N mutation, located in loop F at the α-ε interface, was identified in patients with delayed motor development and rapid fatigability on mild exertion³⁷. Here, XG-590 prolonged mean open and burst durations and enhanced channel reopening, whereas EC-216 exhibited modest enhancement of these parameters (Fig. 2c, Extended Data Fig. 4c). Similarly, the highly disabling αV132L mutation within the Cys-loop that couples ACh binding to channel gating was identified in patients unable to hold their head upright, stand, or walk⁵. XG-590 exhibited a dose-dependent profile in this mutant. At low concentrations (0.6 μM), XG-590 increased mean open and burst durations and enhanced channel reopening. However, at higher concentrations (6 μM), these effects reversed, a phenomenon not observed in the other mutants tested (Fig. 2d, Extended Data Fig. 4d, 8d).

We examined two additional mutations near the ACh binding site: αV188M located in loop C on the principal face of the site³⁸, and the frequently fatal εW55R mutation, identified in patients with profound congenital weakness⁶ and located on the complementary face of the site; both mutations severely reduced both agonist affinity and channel gating efficiency. In these cases, EC-216 and XG-590 showed only weak rescuing effects, modestly increasing mean open duration for αV188M but not εW55R, while EC-216 slightly increased channel reopening for both mutants (Fig. 2e, f, Extended Data Fig. 4e, f).

Together, our findings demonstrate that XG-590 and EC-216 differ not only in potency but also in the spectrum of fast-channel CMS mutations they can partially rescue. The mutation-specific profiles indicate that no single PAM is universally effective; instead, different compounds may be optimal for different variants. These results underscore the importance of tailoring therapeutic strategies to the molecular basis of each patient’s disease.

Drug inhibitions of slow channel mutants

Quinidine and fluoxetine are widely used to treat slow-channel CMS^{8,21,23,25,39}, where gain-of-function mutations prolong channel openings and drive endplate myopathy^9,20. Both drugs are understood to act as channel blockers, but their binding sites and structural mechanisms of action have remained speculative. To begin, we tested their effects on wild-type muscle ACh receptors by whole-cell electrophysiology. Quinidine inhibited ACh-evoked currents at micromolar concentrations (Fig. 3a, b), consistent with earlier work on both quinidine and fluoxetine^22,23,40.

Fig 3. Quinidine and fluoxetine act via pore block in slow channel CMS mutants.

a, Quinidine chemical structure.

b, Representative whole cell electrophysiology of quinidine inhibition of the wild-type muscle receptor.

c, Cryo-EM density map of the βV266M ACh receptor bound to quinidine. Experimental densities for quinidine and mutation sites are shown.

d, Close-up view of quinidine bound to the βV266M ACh receptor pore region. Black dashed lines indicate polar interactions.

e, Fluoxetine chemical structure.

f, Density map of fluoxetine in the εL269F mutant structure.

g, Interactions of fluoxetine in the pore region of the εL269F ACh receptor.

h, Superposition of quinidine and fluoxetine structures in the pore.

To understand their structural mechanisms, we examined two representative slow-channel CMS mutants, βV266M and εL269F, both located in the M2 helices that line the channel pore. βV266M disrupts the 13′ valine ring, producing pathogenic channel openings in the absence of ACh^41,42, ultimately leading to calcium overload and endplate myopathy. We determined the structure of βV266M bound to quinidine and ACh at 2.5 Å resolution, revealing the presence of both the point mutation and the bound quinidine (Fig. 3c, Extended Data Fig. 2, 3, Table 1). The distinctive shape of the quinolone double ring and the quinuclidine cage in quinidine facilitated ligand fitting into the unambiguous density (Fig. 3a, c). Surprisingly, the well-defined quinidine was resolved in the lower pore, positioned between the 9′ leucine ring and 2′ residues. Its quinuclidine cage is sandwiched by hydrophobic 9′ leucines from ε and α_ε subunits, while the quinolone nitrogen orients to form an H-bond interaction with the 6′ serine on the α_δ subunit (Fig. 3d). For comparison, we determined the structure of εL269F, a severe leaky channel mutant⁴¹, bound to fluoxetine and ACh, also at 2.5 Å resolution (Fig. 3e, f, Extended Data Fig. 2, 3). Fluoxetine occupied a nearly identical position in the channel pore, with its trifluoromethyl group oriented intracellularly. Its phenyl ring is clamped by 9′ leucines from β and δ subunits, while its propylamine group positions to form an H-bond, again with the α_δ 6′ serine (Fig. 3g).

Despite their distinct chemistries, both quinidine and fluoxetine bind deep in the pore, bridging the 9′ and 2′ rings (Fig. 3h), and are stabilized by a shared set of hydrophobic and polar interactions. These findings contrast with earlier molecular docking studies (based on the Torpedo muscle-type ACh receptor) that predicted fluoxetine binding closer to the 13′ to 9′ region^43,44; this discrepancy may arise from differences in species and disease mutation backgrounds. Our cryo-EM data instead demonstrate that both drugs stabilize a widened, desensitized-like pore conformation, and directly occlude ion conduction (see Fig. 5l, m, and the related section for more analyses).

Fig 5. Structural basis of CMS pathogenic gating.

a, Structural comparison of the ACh binding pockets of the εP121L (colored by subunit) and wild type ACh receptor (PDB 9DMH; white). Two ACh molecules in the α-ε and α-δ pockets of εP121L are shown.

b, Conformational changes in the α-ε interface comparing the εP121L and wild type ACh receptor viewed from ECD periphery. Black dashed lines indicate an H-bonding interaction.

c, The coupling region differences of the εP121L vs. wild type ACh receptor, viewed from inside the channel pore.

d, The TMD region differences of the εP121L vs. wild type ACh receptor from the synapse.

e, Plots of pore diameters comparing the εP121L ACh receptor with wild type apo (PDB 9DMG), and ACh-bound (PDB 9DMH) states.

f, Structural comparison of the TMD regions of the εT264P ACh receptor (colored by subunit) and wild type receptor (PDB 9DMH; white color) viewed from TMD periphery. βF259 rotates out of the pore to avoid clashing with reboxetine.

g, TMD differences in the βV266M versus wild type ACh receptor (white) from the synapse.

h, βV266M ACh receptor pore conformation; ε and β pore-lining residues are labeled. Red color indicates the mutated residue.

i, As in h, but for the εL269F mutant.

j, The αV249F ACh receptor is compacted compared to the wild type receptor (white) viewed from TMD and ICD periphery.

k, Conformational differences in the TMD of the αV249F mutant vs. the wild type ACh receptor from the synapse.

l, Structural comparison of all slow-channel CMS mutants (shades of blue) and wild type ACh receptor (PDB 9DMH; white color) viewed from inside the channel pore reveals similar TMD conformational changes.

m, Plots of pore diameters comparing all slow-channel CMS mutants with wild type apo and ACh-bound states.

Reboxetine rescues slow CMS mutants

Although quinidine and fluoxetine provide clinical benefit in slow-channel CMS, their side effects^8,25,26, including the pro-arrhythmic risk with quinidine⁴⁵ and the increased risk of suicide with fluoxetine, particularly in younger patients⁴⁶, highlight the need for safer therapeutic alternatives. Reboxetine, a norepinephrine reuptake inhibitor approved for the treatment of depression^47,48, has been reported to non-competitively inhibit ACh receptor-mediated Ca²⁺ influx⁴⁹. We therefore investigated whether reboxetine could serve as a potential treatment for slow-channel CMS.

Whole cell macroscopic current recordings from wild-type receptors confirmed that reboxetine inhibits ACh-evoked currents in a dose-dependent manner (Fig. 4a, b). At the single channel level, reboxetine had minimal effects at low ACh concentration (0.1 μM). In contrast, at high ACh concentration (300 μM), it markedly reduced the number of channel reopenings per burst, consistent with an inhibitory mechanism involving preferential binding to the desensitized state (Fig. 4c, d, Extended Data Fig. 5).

Fig 4. Structural basis of channel block by reboxetine in slow channel CMS.

a, Reboxetine chemical structure.

b, Representative whole-cell electrophysiology of reboxetine inhibition on wild-type muscle receptor and the inhibition statistics; the unpaired two-tailed t-test was used; data are represented as mean ± s.e.m. of experimental replicates (n = 4 cells).

c, Single channel currents for wild type or mutant receptors in the presence of 0.1 μM ACh without (upper) or with (lower) reboxetine (8 kHz Gaussian filter). O, open; C, closed.

d, Plots of channel reopening probability are fitted by a single exponential in the presence of reboxetine and a double exponential in its absence for WT, εL269F, and εT264P. For εL269F, pairwise comparison of profiles, using the extra sum of squares F-test, for 0.1 μM versus 300 μM ACh, each with reboxetine, yielded P = 0.96, indicating no significant difference. For εT264P, comparing profiles for 0.1 μM versus 300 μM ACh, each with reboxetine, yielded P = 0.85, indicating no significant difference. For WT, comparing profiles for 0.1 μM ACh versus 0.1 μM ACh plus reboxetine, yielded P = 0.038 (defined significance by P < 0.01), again indicating no significant difference.

e, Cryo-EM density map of the εL269F ACh receptor bound to reboxetine. Experimental density is shown for reboxetine in its 3 sites and for the mutation site.

f, Close-up view of reboxetine bound to the upper pore of the εL269F ACh receptor. Black dashed lines indicate polar interactions.

g, Two poses of reboxetine bound to the lower pore and their interactions.

h, Interactions of reboxetine bond to the lower pore of the εT264P ACh receptor.

i, Superposition of the lower-pore reboxetine-bound structures in reveals a conserved binding position across mutants.

We next tested two representative pore mutants, εL269F and εT264P⁵⁰, both of which display abnormally prolonged bursts of channel openings (Fig. 4c). In both mutants, reboxetine significantly shortened burst durations by reducing channel reopening (Fig. 4c, d). Notably, this shortening occurred at both low and high ACh concentrations. The action of reboxetine was steeply concentration-dependent, with little effect at 3 μM but strong inhibition at 6 and 10 μM (Extended Data Fig. 5e, f). At 10 μM, reboxetine corrected mean channel open duration to wild-type or near-wild-type levels of both mutants (Extended Data Fig. 5b-d). Given the preferential effects of reboxetine at desensitizing ACh concentrations in wild-type receptors, these results suggest that εL269F and εT264P promote transition from the open to the desensitized state (see “Receptor activation mechanism and the effects of modulators and inhibitors” section in Supplementary Information). Together, these findings indicate that reboxetine counteracts the pathological gating phenotype of slow-channel CMS and support its further evaluation as a therapeutic candidate.

To define its mechanism, we determined the structures of reboxetine bound to the εL269F mutant in the presence of ACh (Fig. 4e, Extended Data Fig. 2b, 3m). Surprisingly, we identified two different binding sites in different structures: one in the upper pore at the α_ε-ε junction near the 16′ and 20′ positions, and another in the lower pore at the 9′ to 6′ level. From the cryo-EM structure, in the upper pore site, reboxetine was stabilized by polar interactions with E262 on αM2 and Q272 on εM2 while its phenyl ring inserted into a local pocket surrounded by hydrophobic residues including εF269, the pathogenic residue itself (Fig. 4f). At the lower site, reboxetine adopted two distinct poses (Fig. 4g). In both, it binds centrally in the pore between the 9′ leucine ring and 6′ residues, which is higher than for quinidine and fluoxetine. The first lower-pore pose has both polar and hydrophobic interactions with its nearby multiple M2 helices. The 6′ N258 on εM2 and 6′ S262 on δM2 likely form H-bonds with the polar groups of reboxetine, while its phenyl ring is clamped between the 9′ leucines from α and ε subunits (Fig. 4g). We observed only hydrophobic interactions in the second reboxetine pose, involving 9′ leucines from all five M2 helices as well as the 6′ F259 on βM2, which orients into the pore, buttressing reboxetine binding in both poses (Fig. 4g).

To assess whether the upper site is mutant-specific, we determined the structure of εT264P-reboxetine in the presence of ACh at 2.4 Å resolution (Fig. 4h, Extended Data Fig. 2b, 3m). Here, reboxetine was found only in the lower pore site, with interactions similar to those in εL269F (Fig. 4h). These findings suggest that the lower site represents the primary, mutation-independent inhibitory mechanism, while the upper site may emerge only in εL269F due to steric changes in that area or increased hydrophobicity. In both slow-channel CMS cases, reboxetine acts as a pore blocker that selectively stabilizes a desensitized state to correct pathologically prolonged channel activation.

Pathogenic CMS gating mechanisms

While CMS mutations have long been proposed to alter receptor gating through diverse mechanisms^1,9,11, the structural basis for these kinetic defects has remained unresolved. Comparing high-resolution structures of representative CMS mutants with the wild-type receptor allows for deduction of how patient mutations destabilize normal gating transitions and lead to disease. We first examined the εP121L fast-channel mutant as a paradigm (Fig. 5a-e). This proline to leucine substitution lies immediately behind αW149, which is part of the core of the aromatic binding pocket that stabilizes ACh binding. The bulky leucine displaces αW149 outward, which shifts bound ACh approximately 1 Å out of its native pocket (Fig. 5a). Despite these changes, we observed strong ACh density at both binding sites, with loop C adopting the closed conformation characteristic of agonist engagement, but shifting subtly outward by < 1 Å relative to the wild-type α–ε ACh-binding pocket. Moreover, in the mutant, the entire α–ε interface undergoes rigid shifts relative to the wild-type receptor, particularly in the lower ECD of the ε subunit and the upper ECD of the α subunit, while the other subunits remain largely unchanged (Extended Data Fig. 6c). All these changes may prevent loop F from moving upward to stabilize loop C via its usual H-bond to εD175^18,30, leaving the α-ε site comparatively “loose” (Fig. 5b). These structural observations align with prior functional studies showing that εP121L selectively reduces ACh affinity at the α-ε site while sparing the α-δ site³⁵.

At the level of transmembrane coupling, εP121L produced further disruption. The α_ε transmembrane helices and the associated M2-M3 loop retained an apo-like conformation rather than undergoing the gating-related displacement seen in wild type (Fig. 5d). In particular, the β1β2 loop retained its original resting conformation and failed to swing into position to promote movement of the M2-M3 loop and the M2 helix of the α_ε subunit (Fig. 5c). By contrast, the other three subunits underwent a conformational change like that seen in the wild type receptor upon ACh binding. As a result, binding energy was not efficiently transmitted into channel opening, uncoupling ligand binding from gating (Fig. 5d). Functionally, this decoupling well explains the markedly slowed channel opening rate of the εP121L mutant³⁵ and the emergence of an intermediate state with a halfway-opened upper pore relative to the wild-type desensitized conformation (Fig. 5e).

In contrast, analysis of all four slow-channel mutants studied here revealed a strikingly convergent mechanism. Despite arising in different subunits (Extended Data Fig. 6), εT264P, βV266M, αV249F⁵¹, and εL269F all produced similarly large outward displacements of the M2 helices, compared to the wild-type desensitized state (PDB ID 9DMH) (Fig. 5l). The expansion was largest in the β subunit (average 3 Å), accompanied by more modest shifts in α_ε and ε M2 (Fig. 5f). The α_δ and δ M2 helices remained largely unchanged (Fig. 5g, k). These rearrangements yielded a consistently widened pore conformation, distinct from wild type (Fig. 5h, i, m). A key player in these conformational differences was the bulky β 6′F259, which adopted two distinct rotamers. In some structures, F259 projected into the pore, where it could directly stabilize drug binding (as in reboxetine-εL269F) (Fig. 4g, 5i). In other cases, F259 rotated out of the pore and nestled between βM2 and δM2 helices to avoid steric clashes with the drug (as in the reboxetine-bound εT264P structure) (Fig. 4h, 5f). The ability of F259 to toggle between these conformations was enabled by the extensive outward shifts of the surrounding helices, effectively reshaping the channel gate from the canonical 9′ leucine ring down to positions below the 6′ or even 2′ layers (Fig. 5h, i). This repositioning produced minimum pore diameters of 4.0 Å in εL269F and ~5 Å in βV266M and αV249F mutants, all wider than wild-type^18,30 (Fig. 5m). While these pores are not expected to support full ion conduction⁵², they are consistent with less stably-closed states that, by virtue of having a weaker gate, may underlie pathological spontaneous openings. Finally, the αV249F mutant exhibited an additional global contraction, spanning 2.5-4 Å from intracellular to extracellular domains (Fig. 5j, Supplementary Video 1, see Extended Data Fig. 6 for more details). This unique effect may arise from the presence of two copies of the mutation per receptor, amplifying its structural consequences. Together, these analyses reveal two unifying principles of CMS pathogenesis. First, fast-channel mutations disrupt the efficient coupling of agonist binding to channel opening, yielding shortened or failed activations. Second, slow-channel mutations drive the pore into an abnormally wide, desensitized-like conformation, reshaping the channel gate and potentially enabling a pathological cation leak arising from spontaneous openings.

Discussion

Current treatments for fast-channel CMS rely on broadly acting cholinergic drugs^8,10, which often have limited efficacy and substantial side effects^16,17. By contrast, PAMs represent a more targeted approach: they enhance receptor activity only in the presence of ACh, thus avoiding direct receptor activation and allowing safer dose control than agonist-based strategies^19,32. Here we revealed the structural basis of PAM action in muscle ACh receptors. XG-590 binds to a previously unknown pocket at the extracellular ends of the β and ε transmembrane helices, a site that is, in the absence of PAM, not fully present; most of the pocket is closed off and only a small tunnel in the same β TMD region is occupied by a lipid tail. The PAM binding appears to induce full opening of the pocket, leading to a large cavity in the β TMD region, which reshapes the coupling region, highlighting a previously underappreciated role of the β subunit in muscle ACh receptor channel gating.

Our functional studies further show that PAM activity varies across fast-channel CMS mutations. XG-590 and EC-216 rescued distinct subsets of patient variants, with efficacy depending strongly on the structural context of the mutation. This mutation-specific activity underscores the capability and necessity of tailoring therapies to individual patient genotypes. Our electrophysiology data indicate that for a given PAM-mutant combination, the rescue of receptor function is only partial. However, unlike current treatments for fast-channel disorders, such as pyridostigmine¹³ combined with 3,4-diaminopyridine¹⁵, which act indirectly, these PAMs directly modulate the muscle ACh receptors. Even a partial enhancement of ACh receptor function may lift enough of the endplate potentials over the threshold for action potential generation to trigger muscle contraction^53,54. Consequently, these PAMs represent potential targeted therapeutics, either alone or in combination with drugs that elevate and prolong synaptic ACh. More importantly, the high-quality structural data provide a clear framework for rational optimization. Indeed, a related compound (AS3580239), distinguished by an appendant diol (Extended Data Fig. 1k), has shown greater potency in a mouse model of myasthenia gravis (MG)³⁴. Our models suggest that enhanced interactions with the nearby M2-M3 loop may underlie its improved efficacy (Extended Data Fig. 1l). Together, these findings point toward a new class of allosteric drugs with potential application not only in fast-channel CMS but also in MG, where muscle ACh receptor function is impaired by autoantibodies^30,55.

For slow-channel CMS, we found that quinidine and fluoxetine, despite their very different chemistries, bind in overlapping positions deep in the channel pore. This shared mechanism explains their efficacy as open-channel blockers^21,22,24,56. Reboxetine, by contrast, binds slightly higher in the pore and can adopt distinct poses, and access a third mutant-specific binding site in the upper pore. Detailed single channel analysis revealed consistent inhibitory effects across multiple slow-channel CMS mutants where reboxetine selectively targets the desensitized conformation of the channel. Reboxetine’s ability to suppress pathological reopenings and burst durations, combined with its long-standing clinical use and safety profile in depression^47,48, positions it as a potential candidate for slow-channel CMS therapy. Nonetheless, known side effects^57,58 also warrant caution when evaluating its clinical use for CMS.

Finally, by analyzing structures of several representative CMS mutants, we uncovered unifying principles of pathogenesis. Fast-channel mutations weaken coupling of agonist binding to channel opening, reducing the incidence of channel openings and shortening their durations. In contrast, slow-channel mutations induce an abnormally widened, desensitized-like pore. This finding overturns the traditional view of the β subunit as merely a structural scaffold^2,59,60 and suggests instead that it plays a central role in the gating cycle². The outward rotation of the bulky β 6′ phenylalanine, observed across slow-channel CMS mutants, is likely conserved in the wild-type human muscle receptor.

In summary, our integrated structural and functional analyses define how CMS mutations perturb receptor gating, reveal the molecular basis of current and candidate therapies, and identify new druggable sites for intervention. These results not only clarify long-standing questions in the field but also provide a roadmap toward the development of safer and more effective, mutation-specific treatments for both fast- and slow-channel CMS.

Methods

Chemical synthesis

The detailed procedures for synthesis of XG-590 and EC-216 are presented in the Supplementary information. All reactions were conducted, when required, in oven-dried round-bottom flasks under an argon atmosphere, using anhydrous solvents and magnetic stirring. Anhydrous solvents were obtained from Fisher Scientific or Sigma-Aldrich, while all reagents were sourced from commercial suppliers without further purification, including AK Scientific, Fisher Scientific, Ambeed, Chem-Impex International, ChemScene, Sigma-Aldrich, Combi-Blocks, and TCI America. Solvents such as ethyl acetate (EtOAc), acetone (Ace), hexanes (Hex), and methylene chloride (DCM), methanol (MeOH), used for column chromatography and reaction work-ups, were purchased from VWR International, Sigma-Aldrich, and Fisher Scientific. Compounds were visualized with UV light (254 nm).

Column chromatography was conducted with a Teledyne NextGen 300+ flash chromatography system. Rotary evaporation was carried out at 30 °C or 40 °C under appropriate pressure considering the solvent used. Nuclear magnetic resonance (NMR) spectra for ¹H, ¹⁹F and ¹³C NMR were recorded on a Bruker Avance III HD 400 MHz spectrometer at 298 K, referenced to the residual solvent resonance peaks (CHCl₃: 7.26 ppm for ¹H and 77.00 ppm for ¹³C, DMSO: 2.50 ppm for ¹H and 39.52 ppm for ¹³C). Multiplicities are denoted with the following abbreviations: s = singlet, d = doublet, t = triplet, q = quartet, quin = quintet, m = multiplet, or combinations thereof. NMR spectra were processed by either Mestrelab Mnova or Bruker TopSpin. High-resolution mass spectra (HRMS) were acquired using a Waters Xevo G2-XS time-of-flight mass spectrometer. Compound purity and reaction progress were analyzed using a Waters Acquity UPLC-MS system. Microwaved reactions were performed using a Biotage^® Initiator+ Microwave.

CMS mutation selection and mutagenesis

Given the impracticality to study all known missense mutations (more than 50), we selected five representatives of human muscle ACh receptor CMS mutants (four slow-channel and one fast-channel) based on principles of their prevalence in CMS patients, their representation of different mutation subgroups, and their significant impact on channel function and structure. We previously described how slow-channel CMS mutations are mostly concentrated in the receptor’s TMD region, particularly in the pore-lining M2 helices². This location distribution suggests that these mutations regulate channel kinetics by altering the channel pore conformation. We categorized the slow-channel mutants into four subgroups based on their position in the M2 helices: 1) located at subunit interface and upper M2 helix: mutations in this region influence upper channel opening (like εL269F); 2) mutations that alter helix geometry and local pore conformation (εT264P); 3) pore-lining residue mutations: mutations that directly affect ion permeation (βV266M) and 4) located at subunit interface and lower M2 helix: mutations affecting lower channel opening (αV249F). The fast-channel CMS mutations are mostly located at the ECD and ICD regions of muscle ACh receptor^2,11. For structural study, we selected εP121L, one of the most common mutations¹⁴. This mutation is located at the α-ε extracellular domain interface and is found in many fast-channel CMS patients¹⁴. For electrophysiology analyses, we selected several additional fast-channel mutants located in other regions of the receptor: at the complementary face of the ACh binding site, εD175N and εW55R; at the principal face of the ACh binding site, αV188M; in the Cys-loop, αV132L; and on the TMD M3 helix, αV285I.

For mutagenesis, complementary primers containing the selected CMS mutations were designed using the QuikChange Primer Design program (Agilent Technologies). The methylated parental DNA template in the PCR reaction was digested using DpnI enzyme (10 U/μL) and the digested product was then transformed into E. coli DH5α competent cells. Mutations were confirmed by sequencing.

Mutant stable cell line generations

To generate the CMS mutant stable cell lines, either 1 μg mutant pSBtet-GP-α+δ, 1 μg wild-type pSBtet-RH-β+ε or vice versa, and 0.1 μg pCMV(CAT)T7-SB100 plasmids were transfected into HEK293S GnTI⁻ cells (ATCC CRL-3022) at 75% confluency using Lipofectamine 2000 (Invitrogen). We modified the receptor constructs by deleting the FLAG tag from the ε subunit and moved the twin strep tag from M3M4 loop to the N-terminus of the δ subunit. Both pSB Sleeping Beauty system vectors (Catalog #60495, 60500) and the SB transposase plasmid (Catalog #34879) were obtained from Addgene. We have deposited the full α+δ and β+ε plasmids in Addgene as well (#247177 and 247178, respectively). One day after transfection, the dual selection was started by adding 1.5 μg/mL puromycin and 100 μg/mL hygromycin B to the culture plate. The selection was maintained for two weeks until all cells exhibited both GFP and dTomato fluorescence. Cells were next cultured using 2 L suspension flasks, in FreeStyle 293 medium (ThermoFisher) + 2% fetal bovine serum (Sigma), for large-scale protein production.

Mutant muscle receptor and drugs complex preparations

The CMS mutant proteins expression and purification were similar with that of wild type receptor reported previously³⁰. Briefly, the stable cells were cultured at 37 °C to reach a density of 4 × 10⁶ cells per mL and then 2 μg/mL doxycycline was added for inducing expression of the mutant receptor proteins at 30 °C for two days. 10 L of suspension cells were collected by centrifugation at 5,000 × g for 15 minutes at 4 °C. The resulting cell pellets were resuspended in a lysis buffer containing 50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 2 mM EDTA, 1 mM PMSF (phenylmethylsulfonyl fluoride), and cOmplete protease inhibitor cocktail (Sigma). The cell suspension was passed through an Avestin C5 homogenizer four times at a pressure range of 5,000–10,000 psi. The cell lysate was clarified by centrifugation at 10,000 × g for 15 minutes at 4 °C. The resulting supernatant was ultra-centrifugated at 40,000 rpm for 2 hours at 4 °C using a 45Ti rotor to pellet cell membranes. These membrane pellets were stored at −80 °C for subsequent use.

For mutant receptor purification, 10 g of membrane were homogenized using a Dounce tissue grinder in a buffer containing 50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 1 mM EDTA, 1 mM PMSF, protease inhibitor cocktail as well as 10 μM soy polar lipids (Avanti Polar Lipids), cholesterol (Sigma), and DDM (n-dodecyl-β-D-maltoside, Anatrace) in a 4:1:1.84 weight ratio. GDN (glyco-diosgenin, Anatrace) was added to a final concentration of 2% to solubilize the cell membrane at 4 °C for 5 hours to increase the mutant protein yield. After solubilization, insoluble debris was cleared by centrifugation at 186,000 × g for 40 minutes. The solubilized protein was then loaded onto 3 mL of Strep-Tactin XT Flow high-capacity resin (IBA Life Sciences) using a peristaltic pump at a flow rate of 0.8 mL/min at 4 °C. After binding, the resin was washed with 200 mL of TBS containing 0.5 mM GDN and 10 μM soy polar lipids plus cholesterol to maintain receptor stability. The receptor proteins were then eluted using TBS buffer supplemented with 0.25 mM GDN, 10 μM soy polar lipids, cholesterol, and 50 mM biotin (Sigma). Eluted fractions were assessed by fluorescence-detection size-exclusion chromatography (FSEC) utilizing intrinsic tryptophan fluorescence. Fractions containing the receptor proteins were pooled and concentrated for preparative SEC on a Superose 6 Increase 10/300 GL column (Cytiva), in a buffer containing 50 mM Tris-HCl (pH 7.4), 150 mM NaCl, 0.25 mM GDN, and 10 μM soy polar lipids with cholesterol. SDS-PAGE was used to assess protein purity. We observed that the expression levels of some CMS mutants differed from the wild-type receptor, while others were comparable. Among these mutants, αV249F exhibited the lowest yield, reaching only ~30% of the wild-type level. Drugs including the PAM and blockers were diluted using the same SEC buffer and added into the purified mutant receptor proteins at a final concentration of 200 μM and pre-incubated for 1 hour on ice. 100 μM ACh was then added and incubated for 10 minutes. These samples were used for cryo-EM grid preparation.

Cryo-EM sample preparation

Three μL of freshly purified mutant ACh receptor proteins bound to drugs and/or ACh at a concentration of 0.2 mg/mL were added to a glow-discharged (at 30 mA for 10 s) 300-mesh copper grid coated with a 2 nm carbon film (R2/1, Quantifoil, Cat# Q2100CR1-2nm). Grids were waited for 20 s and blotted for 2 s under 100% humidity at 4 °C. Grids were plunge-frozen into liquid ethane using a Vitrobot Mark IV and the quality was checked on a 200 kV Talos Arctica microscope.

Cryo-EM data collection and image processing

The data processing strategies for all mutants bound to ACh as well as XG-590 were comparatively straightforward to process while the methods using focused classification for blocker-bound datasets were similar with each other. To limit redundancy, here we described the processing details for XG-590 and reboxetine-bound datasets as two representative examples; details for all datasets are summarized in Extended Data Table 1. For the ACh and XG-590-bound sample, 9,212 raw images were collected on the UCSD Titan Krios 2 at 300 kV with a total dose of 50 e⁻/Ų and a magnification of 130,000x, resulting in a pixel size of 0.935 Å/pixel. The defocus range was set from −1.0 to −1.8 μm for data collection. All data processing was done using cryoSPARC v4.7⁶¹. The gain-normalization, motion correction, and CTF estimation were done using default parameters in cryoSPARC. Processing started from approximately 1.1 million raw particles. After three rounds of 2D classification, 359,449 particles were kept and subjected to a 3D Homogeneous Refinement using an initial model generated from an ab initio reconstruction. The 3D Homogeneous Refinement provided a 3.86 Å density map at the bin2 pixel size and the resulting map was then submitted to a 3D classification (8 classes) with PCA mode and 1% convergence criterion. The classification only generated one class with clear ECD, strong TMD and ICD densities. The only receptor class was selected, re-extracted at full pixel size, and refined using cryoSPARC NU Refinement⁶², which generated a map at a resolution of 2.44 Å (210,863 particles).

For the εL269F ACh and reboxetine-bound dataset, 8,479 raw images were collected using the same parameter settings as described above on the UCSD Titan Krios 2. All data processing was done again using cryoSPARC v4.7. After removing junk particles via 2D classification, 566,119 good particles with clear secondary structural features were kept. Twenty thousand particles were randomly selected to generate a 3D model using ab initio reconstruction. All selected particles were then submitted into a 3D Homogeneous Refinement job using the ab initio model as the initial volume, which again generated a 3.86 Å density map at the bin2 pixel size. To remove bad particles, a 3D classification (8 classes) was performed and the particles with clear TMD and ICD were selected and re-extracted at full pixel size, and refined using cryoSPARC NU Refinement to a map at a resolution of 2.29 Å (281,324 particles). We noticed that two potential binding sites may exist, one is at the upper region of the pore and another is at the pore bottom center; however, both sites have weaker local density. To improve the local density quality, two soft focus masks were generated using Relion_mask_creation⁶³ based on the NU refinement map. These two masks were used to perform focused 3D classification (6 classes) to classify out the good particles that bind to reboxetine at two different sites respectively. In both focused 3D classifications, the mode was set to PCA, resolution was filtered to 3 Å, and the convergence criterion was changed to 1%. The good classes with strong drug densities in each classification were selected and submitted to another NU Refinement. The final NU Refinement generated a map at a resolution of 2.46 Å (95,226 particles) in the upper region, while at the pore bottom, two maps were obtained at resolutions of 2.36 Å (143,299 particles) and 2.68 Å (39,554 particles) respectively.

Model building, refinement, and validation

We previously reported the high-resolution structure of the ACh-bound human adult muscle receptor (PDB ID 9DMH)³⁰. This structure was used as the starting model, and roughly fitted into the ACh-bound CMS mutant density maps using UCSF Chimera⁶⁴. The most different regions including the channel pore and some loops in the coupling region fitted poorly and were firstly manually checked and rebuilt, and then subjected to several cycles of real space refinement in Phenix⁶⁵. All CMS mutated resides were confidently assigned because of the high-resolution quality of the maps. We observed that the outer leaflet of M4 helices in the α subunits are more flexible, with sometimes poorly defined densities. Lipids and glycosylation were manually checked and adjusted based on the individual mutant map. Finally, the improved model was adjusted manually residue by residue in Coot⁶⁶ based on the PDB validation report and in combination with several rounds of global real space refinement in Phenix with secondary structure and Ramachandran restraints. Model geometry and clash scores were checked with MolProbity⁶⁷. For the ACh and drug-bound mutant model building, the well-refined ACh-bound mutant structure was used as the starting model, and roughly fitted into the respective drug-bound density maps and then refined by Phenix in combination with Coot. We noticed that fluoxetine binds to the εL269F mutant pore in slightly different poses; we modeled the fluoxetine molecule based on the strong density shape at the pore. In addition, we included fluoxetine in the cryo-sample preparation of αV249F mutant; however, we did not detect clear density that could be assigned to fluoxetine in the cryo-EM map. Thus, we did not build a model for fluoxetine in this coordinate file. Drugs restraint files were generated using the Grade Web Server v2.0.3 (https://grade.globalphasing.org/) with default settings. Model geometry and clash scores were again checked with MolProbity. LigPlot v2.3 was used to generate of 2D ligand-protein interaction diagrams.

All model building was carried out using the final sharpened density maps obtained from cryoSPARC NU Refinement. DeepEMhancer⁶⁸ was selectively used to enhance the quality of regions with lower local resolution. Pore diameter measurements were conducted using HOLE2⁶⁹. The map and structural figures were generated using UCSF Chimera, ChimeraX⁷⁰, and PyMOL (version 2.5.5, Schrödinger, LLC). The main text figures were made from density maps after post-processing with DeepEMhancer.

Whole-cell electrophysiology

Whole cell voltage-clamp recordings were made from adherent HEK293S GnTI⁻ cells stably expressing doxycycline inducible human α1, β1, δ, and ε, or transiently transfected with human α, δ, ε and β in a ratio of 2α:1β:1δ:10ε. After 48 hours of induction with 1 μg/ml of doxycycline or transfection, cells were re-plated on 35 mm dishes and allowed to settle for at least 3 hours. Recordings were performed 48 to 96 hours after induction. The bath solution contained (in mM): 140 NaCl, 2.4 KCl, 4 MgCl₂, 4 CaCl₂, 5 HEPES and 10 glucose pH 7.3. Borosilicate pipettes were pulled and polished to an initial resistance of 2–4 MΩ, filled with the pipette solution containing (in mM): 100 CsCl, 30 CsF, 10 NaCl, 10 EGTA, and 20 HEPES pH 7.3. Whole cell currents were recorded with an Axopatch 200B amplifier, sampled at 20 kHz, and low-pass filtered at 2 kHz using a Digidata 1550B (Molecular Devices). Cells were held at -75 mV. Solution exchange was achieved using a gravity driven RSC-200 rapid solution changer (Bio-Logic). Whole-cell current data were collected with pClamp 11 and analyzed with Clampfit 11 software (Molecular Devices). 2 μM of ACh and 0.1–30 μM of ligands (XG-590, EC-216 and blockers) were prepared in bath solution from concentrated stocks, 1 M ACh in water, and 10 mM XG-590 and 25 mM EC-216 in DMSO, stored at −20 °C.

Mutagenesis and expression of adult human muscle ACh receptors for patch clamp recording

cDNAs encoding human α1, β1, δ, and ε subunits cloned into the mammalian expression vector pRBG4 were as described. The pathogenic mutations, εP121L, εD175N, εW55R, αV132L, αV188M, and αV285I, were constructed as described^5,35-38, and confirmed by sequencing prior to use. Sample cDNAs for the mutations εW55R, αV132L, αV188M were kindly provided by Dr. Xin-Ming Shen, Mayo Clinic, Rochester. BOSC 23 cells, an HEK 293 derived cell line⁷¹, were maintained in Dulbecco's modified Eagle's medium (DMEM, Gibco) containing 10% fetal bovine serum, and transfected using calcium phosphate precipitation, as described⁷². The α1, β1, δ, and ε subunit cDNAs were transfected in a 2:1:1:1 ratio, respectively, where 1 = 2.5 μg, per 35 mm culture dish of cells. A cDNA encoding green fluorescent protein was included in all transfections. Transfections were carried out for 4 to 6 hours, followed by medium exchange. Single-channel recordings were made 48-72 hours post-transfection.

Single-channel patch clamp recordings

Single-channel recordings were obtained using the cell-attached patch configuration with a membrane potential of −70 mV and a temperature of 20 °C, as described⁷³. For all experiments, the pipette and extracellular bathing solutions contained (mM): 142 KCl, 5.4 NaCl, and 10 HEPES, adjusted to pH 7.4 with NaOH. Concentrated stock solutions of ACh were made in pipette solution and stored at −80 °C until the day of each experiment. Patch pipettes were pulled from glass capillary tubes (No.7052, King Precision Glass) and coated with Sylgard elastomer (Dow Corning). Defined concentrations of ACh were placed on ice and added to the patch pipette just prior to recording.

Single-channel currents were recorded using an Axopatch 200B patch-clamp amplifier (Molecular Devices), with a gain of 100 mV/pA and the internal Bessel filter set at 100 kHz. The current output from the amplifier was sampled at 2 μs intervals using a PCI-6111E acquisition card (National Instruments) and recorded to the hard disk of a PC computer using the program Acquire (Bruxton Corporation). Just before establishing a cell-attached patch, the pipette offset potential was set to zero, and upon giga-seal formation, a command voltage of −70 mV was applied to the interior of the pipette to establish the membrane potential. Recordings from 2-3 patches were acquired for each combination of drug and mutant receptor.

Single-channel data and analysis

For each recording, channel opening and closing transitions were detected using the program TAC 4.2.0 (Bruxton Corporation), which digitally filters the data (Gaussian response, final effective bandwidth 10 kHz), interpolates the digitized points using a cubic spline function, and detects channel opening and closing transitions using the half-amplitude threshold criterion, as described⁷⁴. Dwell time histograms were plotted using a logarithmic abscissa and square root ordinate⁷⁵, with a uniformly imposed dead time of 20 μs, and the sum of exponentials was fitted to the histograms by maximum likelihood using the program TACFit 4.2.0⁷⁴.

Bursts of channel openings were identified as a series of closely spaced openings preceded and followed by closed intervals longer than a specified critical time (τ_crit). This time was determined from the closed time histogram as the point of intersection between the briefest exponential component and the succeeding longer component. Values of τ_crit were determined independently for each patch but typically were 100-200 μs. The duration of a burst is thus the total time of a series of openings and intervening closings briefer than τ_crit. The mean channel open and burst durations were obtained from recordings from multiple patches (2 to 4) under identical experimental conditions. Statistical analysis of mean open and burst durations was done using the Prism software package (GraphPad Prism RRID:SCR 002798).

Channel reopening probability

The effect of a drug on a wild type or mutant receptor channel’s ability to reopen was analyzed by plotting the fraction of bursts that reopened N times against the number of re-openings N per burst, where a burst was defined as a series of openings with intervening closings briefer than τ_crit. The decay in fractional channel reopening was fitted with single or a double exponential using a non-linear least squares function. Pairwise comparisons of reopening profiles for ACh alone versus ACh plus modulator were done by applying a least squares F-test for which P < 0.01 was considered significantly different. Fitting and statistical analyses were done using GraphPad Prism software.

Fractional potentiation

For a given mutant-modulator combination we fitted the burst duration histogram with the sum of two exponentials using the maximum likelihood criterion. The component with briefest mean duration, which was the major component in the absence of modulator, was ascribed to un-potentiated bursts. The component with longest mean duration, which was the major component in the presence of modulator, was ascribed to potentiated bursts. Each component’s relative area was then multiplied by the corresponding mean duration, giving the fraction of the total burst time spent in either un-potentiated or potentiated bursts. The fractional potentiation, denoted FP, was then computed as the ratio of potentiated burst time divided by the sum of un-potentiated and potentiated burst times.

Statistical analysis

The figures describing whole-cell electrophysiology results are presented as normalized peak currents. Statistical analysis was performed using GraphPad Prism software (GraphPad software, Inc, La Jolla, CA). The unpaired two-tailed t-test was used for electrophysiology in Fig 1, 4 and Extended Data Fig. 4, 5. Channel reopening kinetics were fitted using nonlinear least-squares fitting, and statistical comparisons between conditions were performed using an extra sum-of-squares F-test, which tests the null hypothesis that a pair of decay profiles can be fitted by the same decay rate constant (since Y₀ = 1 for both profiles). The null hypothesis is rejected if the test comes up with P < 0.05. The P-values are described in the related Figures and figure legends. Replicate numbers n in each figure are the number of independent cells that the recordings are taken from.

Extended Data

Extended Data Figure 1. Structure and functional tests of PAMs XG-590 and EC-216 on muscle ACh receptor.

a, Representative whole-cell electrophysiology of EC-216 potentiation on wild-type muscle receptor.

b, Cryo-EM density map of εP121L mutant bound to XG-590 at the ε binding site. XG-590 density was shown. The XG-590 density at the β site was stronger than the ε site, here shown at a lower threshold than that in Fig. 1.

c, Interactions to stabilize XG-590 binding in the ε binding site. Black dashed lines indicate the polar interactions.

d, e, Plots of two XG-590 binding sites show the interactions. Red dashes, hydrophobic interaction; gray arrow, backbone polar interaction; black arrow, side-chain polar interaction.

f, Structural comparisons of two binding sites of XG-590 on the εP121L mutant receptor. The β and ε subunits are superimposed to illustrate the similarities and differences between the two XG-590 binding sites. XG-590 inserts more deeply at the β subunit site due to the larger cavity compared with the ε subunit site. In the ε subunit, F454 would sterically clash with XG-590 positioned as in the β site, as they partially occupy the same cavity volume.

g, Comparison of potentiating effect between wild type receptor and the triple mutant in the 2 μM ACh and 0.6 μM XG-590. Data are represented as mean ± s.e.m. of biological replicates (n value is labeled in each bar graph).

h, Comparison of potentiating effect between wild type receptor and the quadruple mutant in the 2 μM ACh and different concentrations of XG-590. The unpaired two-tailed t-test was used; data are represented as mean ± s.e.m. of biological replicates (n value is labeled in each bar within the graph). In panels g and h, no substantial reduction was observed in our mutagenesis analysis, which may reflect a non-canonical binding mode in which XG-590 interacts within the transmembrane domain through extensive hydrophobic contacts in a dynamic lipid environment (Extended Data Fig. 1d, 1e, 1j), thereby rendering the modulation relatively resistant to few point mutations. We cannot exclude alternative possibilities, including XG-590 binding to a conformation-specific pocket in the activated state that may differ from that captured in our structures, or engagement of an additional weak binding site at the equivalent position in the δ subunit, as suggested by much weaker density features in our maps (Extended Data Fig. 1i), which could partially compensate for perturbations at primary sites.

i, Discontinuous densities observed at the equivalent position in the δ subunit.

j, Extensive lipid densities are observed adjacent to XG-590 at both sites.

k, Chemical structures of meloxicam (top panel), which is the head part of two PAMs and a modified compound AS3580239 (bottom panel,³⁴) from XG-590 and EC-216. Red dashed circle indicates the modified region.

l, Modeled structure of AS3580239 based on the XG-590-bound εP121L structure suggests more interactions result from the modification.

Extended Data Figure 2. Representative cryo-EM data processing workflows.

Representative cryo-electron microscopy workflow used to determine high-resolution structures of XG-590-bound εP121L (a) and reboxetine-bound εL269F (b) ACh receptor complexes. Each step is delineated by arrows and performed in CryoSPARC v4.7.

Extended Data Figure 3. Map quality evaluation for human CMS mutant muscle ACh receptor structures.

a-l, Each panel is for a different density map; left top are Fourier-shell correlations, with resolution determined at gold-standard cut-off of 0.143 (solid line across graph). Unmasked, loose, tight, and corrected masks are plotted, with resolution of each indicated in parentheses. The bottom-left graphs of each panel are angular distribution plots. The right side of each panel are maps color-coded for local resolution as indicated in scale bar on bottom of each map.

m, Summary of the mutant structures in different conditions determined in this study.

Extended Data Figure 4. Burst duration and mean open duration of fast-channel CMS mutants.

a-f, Comparison of the effects of modulators on fast channel CMS as measured by the mean duration of all bursts; ACh concentration was at 300 μM. The left of each panel is mean open duration analyses of each fast-channel CMS mutant. The right of each panel is mean burst duration analyses. Bursts are as defined in Methods and include bursts with zero or more channel reopenings. The unpaired two-tailed t-test was used; data are represented as mean ± s.e.m. of experimental replicates (n value is labeled in each bar graph).

Extended Data Figure 5. Single-channel records and analyses of wild type and slow-channel mutants.

a, Representative single-channel recordings of different muscle receptors, WT, εL269F and εT264P mutants with and without reboxetine under the 300 μM ACh concentration. O, open; C, close.

b-d, Mean open duration analyses of wild type receptor and each slow-channel CMS mutant. The unpaired two-tailed t-test was used; data are represented as mean ± s.e.m. of experimental replicates (n value is labeled in each bar graph).

e, f, The probability a burst reopens more than N times against the number of channel reopenings per burst, N. The smooth curves are exponential fits to the data; for ACh alone and ACh + 3 μM reboxetine the fit is a single exponential decay, whereas for the others the fit is a double exponential decay.

g, Open time histograms for εT264P and εL269F in the presence of the indicated concentrations of reboxetine and 300 μM ACh. Histograms are fitted by a single or double exponential. For each histogram the mean duration of all openings (τ_mean) is displayed.

h, The reciprocal of τ_mean against reboxetine concentration; filled circles indicate mean values from 2 independent measurements, except for both mutants at 3 μM (n = 3).

Extended Data Figure 6. Structural comparisons of CMS mutants with wild type receptors.

a, Structural differences at the coupling regions between the εP121L mutant and wild type receptor.

b, Structural difference at the upper M2 helix between the εP121L mutant and wild type receptor. Structures are in different colors as indicated in the figures.

c, Detailed differences at the ACh binding site between εP121L mutant and wild type receptor. Structural superimposition of the εP121L mutant and the wild-type receptor based on their δ subunits. The structures are shown in different colors as indicated in the figure. Black arrows indicate the directions of conformational changes.

d, Structural analyses of the contraction movement of αV249F mutant relative to the wild type receptor. Structural superimposition of the αV249F mutant and the wild-type receptor based on their entire backbones. Comparisons of the overall architecture and helical arrangements of the mutant and wild type receptors.

e, Comparisons of the ECD between the mutant and wild type using the α_ε subunit as an example. The structures are shown in different colors as indicated in the figure. Black arrows indicate the directions of conformational changes.

f, The bulky methionine mutation pushes the β M2 helix in the opposite direction from the stable δ subunit, and leads to local conformational changes of α M2 helix in the βV266M mutant structure.

g, Conformational changes in the εL269F mutant compared to wild type receptor (PDB ID 9DMH). Mutation leads to rearrangement of the interactions in the α-ε TMD interface. The mutated residue is shown in red. Dashed circle indicates the channel pore. The mutant structures are colored by subunits while wild type is in white in all panels.

h, Local conformational changes in the αV249F mutant. Mutation pushes the hydrophobic residues at the β-α TMD interface away from the pore.

i, The T264P mutation leads to the twisted conformation of ε M2 helix (as red arrow indicated) and slightly moves the M2 helix out of the pore, leaving space for rearrangements in nearby M2 helices.

Extended Data Figure 7. Concentration dependency and single channel dwell time analysis for εP121L in the presence of 300 μM ACh without and with EC-216 or XG-590.

a, Open time histograms obtained from recordings at 8 kHz bandwidth. Histograms are fitted by one or two exponential components (thin lines).

b, Closed time histograms fitted by two or three exponentials. Vertical black lines show the critical time (τ_crit) used to define bursts of openings occurring in quick succession; τ_crit is the point of intersection between the briefest and longest exponential components.

c, Burst duration histograms fitted by two exponentials, the longer of which corresponds to potentiated openings comprised of one or more openings plus intervening closings shorter than (τ_crit).

d, The fraction of channel openings classified as potentiated, as determined from the relative area of the slowest component of the open duration histogram, is shown for the indicated drug concentrations.

Data points show the mean fractional potentiation with the error bars indicating the SEM For the 0.06 and 0.1 μM concentrations of XG-590, n = 1 recording; for all other data points, n = 3 recordings from independent cells.

Extended Data Figure 8. Single channel recordings from wild type and three fast channel mutant receptors in the presence of the indicated concentrations of either EC-216 or XG-590.

For recordings on the 1 s time scale the bandwidth is 5 kHz, whereas for the expanded time scales the bandwidth is 8 kHz.

Extended Data Table 1.

Cryo-EM data collection, refinement and validation statistics

	εP121L +ACh (EMD-72844) (PDB 9YE6)	εP121L +ACh, +XG-590 (EMD-72845) (PDB 9YE7)	εL269F +ACh (EMD-72846) (PDB 9YE8)	εL269F +ACh, +Reboxetine upper pore (EMD-72850) (PDB 9YEH)
Data collection and processing
Magnification	130,000×	130,000×	130,000×	130,000×
Voltage (kV)	300	300	300	300
Electron exposure (e−/Å2)	50	50	50	50
Defocus range (μm)	−1.0 to −1.8	−1.0 to −1.8	−1.0 to −1.8	−1.0 to −1.8
Pixel size (Å)	0.935	0.935	0.935	0.935
Symmetry imposed	C1	C1	C1	C1
Initial particle images (no.)	360,104	359,449	510,194	566,119
Final particle images (no.)	127,259	210,863	361,350	95,226
Map resolution (Å)	2.47	2.44	2.22	2.46
FSC threshold	0.143	0.143	0.143	0.143
Refinement
Initial model used (PDB code)	9DMH	9DMH	9DMH	9DMH
Model resolution (Å)	2.67	2.60	2.46	2.76
FSC threshold	0.5	0.5	0.5	0.5
Map sharpening B factor (Å2)	−36.5	−42.9	−36.6	−29.7
Model-to-map, CCmask	0.81	0.80	0.80	0.77
Model composition
Non-hydrogen atoms	16,745	16,695	16,613	16,614
Protein residues	2,046	2,034	2,036	2,032
Ligands	24	27	23	25
Water	2	3	2	2
B factors (Å2)
Protein	64.97	48.67	60.08	68.86
Ligand	74.73	59.22	65.48	78.97
Water	44.19	33.78	36.41	44.72
R.m.s. deviations
Bond lengths (Å)	0.004	0.004	0.004	0.004
Bond angles (°)	0.771	0.774	0.655	0.666
Validation
MolProbity score	0.87	0.96	0.93	1.10
Clashscore	1.40	1.92	1.71	2.44
Poor rotamers (%)	0.43	0.00	0.27	0.22
Ramachandran plot
Favored (%)	98.56	98.10	98.10	97.65
Allowed (%)	1.39	1.85	1.85	2.30
Disallowed (%)	0.05	0.05	0.05	0.05
	εL269F +ACh, +Reboxetine lower pore 1 (EMD-72851) (PDB 9YEI)	εL269F +ACh, +Reboxetine lower pore 2 (EMD-72853) (PDB 9YEK)	εL269F +ACh, +Fluoxetine (EMD-72854) (PDB 9YEL)	εT264P +ACh (EMD-72860) (PDB 9YER)
Data collection and processing
Magnification	130,000×	130,000×	130,000×	130,000×
Voltage (kV)	300	300	300	300
Electron exposure (e−/Å2)	50	50	50	50
Defocus range (μm)	−1.0 to −1.8	−1.0 to −1.8	−1.0 to −1.8	−1.0 to −1.8
Pixel size (Å)	0.935	0.935	0.935	0.935
Symmetry imposed	C1	C1	C1	C1
Initial particle images (no.)	566,119	566,119	684,269	561,794
Final particle images (no.)	143,299	39,554	211,224	327,352
Map resolution (Å)	2.36	2.68	2.48	2.19
FSC threshold	0.143	0.143	0.143	0.143
Refinement
Initial model used (PDB code)	9DMH	9DMH	9DMH	9DMH
Model resolution (Å)	2.77	3.02	2.70	2.34
FSC threshold	0.5	0.5	0.5	0.5
Map sharpening B factor (Å2)	−31.5	−23.5	−39.3	−35.8
Model-to-map, CCmask	0.76	0.76	0.79	0.80
Model composition
Non-hydrogen atoms	16,645	16,649	16,737	16,595
Protein residues	2,037	2,036	2,047	2,034
Ligands	24	25	25	23
Water	2	2	2	2
B factors (Å2)
Protein	70.23	80.26	58.10	50.10
Ligand	78.15	94.43	65.57	58.62
Water	44.18	61.63	29.08	22.87
R.m.s. deviations
Bond lengths (Å)	0.004	0.003	0.004	0.004
Bond angles (°)	0.646	0.597	0.641	0.659
Validation
MolProbity score	1.14	1.16	0.96	0.70
Clashscore	2.76	2.73	1.91	0.60
Poor rotamers (%)	0.38	0.27	0.21	0.22
Ramachandran plot
Favored (%)	97.66	97.51	98.31	98.20
Allowed (%)	2.29	2.44	1.64	1.75
Disallowed (%)	0.05	0.05	0.05	0.05
	εT264P +ACh, +Reboxetine (EMD-72862) (PDB 9YET)	βV266M +ACh (EMD-72863) (PDB 9YEU)	βV266M +ACh, +Quinidine (EMD-72866) (PDB 9YEX)	αV249F +ACh, +Fluoxetine (EMD-72868) (PDB 9YF0)
Data collection and processing
Magnification	130,000×	130,000×	130,000×	130,000×
Voltage (kV)	300	300	300	300
Electron exposure (e−/Å2)	50	50	50	50
Defocus range (μm)	−1.0 to −1.8	−1.0 to −1.8	−1.0 to −1.8	−1.0 to −1.8
Pixel size (Å)	0.935	0.935	0.935	0.935
Symmetry imposed	C1	C1	C1	C1
Initial particle images (no.)	790,322	313,827	756,189	213,237
Final particle images (no.)	95,028	164,267	109,269	75,733
Map resolution (Å)	2.36	2.39	2.47	2.79
FSC threshold	0.143	0.143	0.143	0.143
Refinement
Initial model used (PDB code)	9DMH	9DMH	9DMH	9DMH
Model resolution (Å)	2.61	2.60	2.68	3.14
FSC threshold	0.5	0.5	0.5	0.5
Map sharpening B factor (Å2)	−25.0	−36.2	−31.7	−41.8
Model-to-map, CCmask	0.80	0.80	0.80	0.72
Model composition
Non-hydrogen atoms	16,729	16,606	16,662	16,408
Protein residues	2,047	2,035	2,039	2014
Ligands	24	23	24	21
Water	2	2	2	2
B factors (Å2)
Protein	62.14	58.24	61.64	66.81
Ligand	67.55	65.91	65.79	72.57
Water	45.79	37.73	37.31	37.46
R.m.s. deviations
Bond lengths (Å)	0.004	0.004	0.004	0.004
Bond angles (°)	0.657	0.669	0.688	0.701
Validation
MolProbity score	0.89	0.95	1.01	1.44
Clashscore	1.52	1.89	2.34	4.26
Poor rotamers (%)	0.16	0.05	0.48	0.11
Ramachandran plot
Favored (%)	98.31	98.00	98.21	96.47
Allowed (%)	1.64	1.95	1.74	3.48
Disallowed (%)	0.05	0.05	0.05	0.05

Supplementary Material

Supplementary Information is available for this paper.

Data availability

All atomic models and cryo-EM maps have been deposited in the Protein Data Bank and Electron Microscopy Data Bank: εP121L+ACh (PDB 9YE6, EMD-72844); εP121L+ACh, +XG-590 (PDB 9YE7, EMD-72845); εL269F+ACh (PDB 9YE8, EMD-72846); εL269F+ACh, +Reboxetine, upper pore (PDB 9YEH, EMD-72850); εL269F+ACh, +Reboxetine, lower pore 1 (PDB 9YEI, EMD-72851); εL269F+ACh, +Reboxetine, lower pore 2 (PDB 9YEK, EMD-72853); εL269F+ACh, +Fluoxetine (PDB 9YEL, EMD-72854); εT264P+ACh (PDB 9YER, EMD-72860); εT264P+ACh, +Reboxetine (PDB 9YET, EMD-72862); βV266M+ACh (PDB 9YEU, EMD-72863); βV266M+ACh, +Quinidine (PDB 9YEX, EMD-72866); αV249F+ACh, +Fluoxetine (PDB 9YF0, EMD-72868). The wild type muscle receptor in apo (PDB 9DMG), and ACh-bound (PDB 9DMH) states were used to perform structural analyses. All other materials are available upon request.