Structural mechanisms of α7 nicotinic receptor allosteric modulation and activation

Summary:

The α7 nicotinic acetylcholine receptor is a pentameric ligand-gated ion channel that plays an important role in cholinergic signaling throughout the nervous system. Its unique physiological characteristics and implications in neurological disorders and inflammation make it a promising but challenging therapeutic target. Positive allosteric modulators overcome limitations of traditional α7 agonists, but their potentiation mechanisms remain unclear. Here we present high-resolution structures of α7-modulator complexes revealing partially overlapping binding sites but varying conformational states. Structure-guided functional and computational tests suggest that differences in modulator activity arise from the stable rotation of a channel gating residue out of the pore. We extend the study using a time-resolved cryo-EM approach to reveal asymmetric state transitions for this homomeric channel and also find that a modulator with allosteric agonist activity exploits a distinct channel gating mechanism. These results define mechanisms of α7 allosteric modulation and activation with implications across the pentameric receptor superfamily.

Graphical Abstract

In brief:

Structural interrogation of α7 nicotinic receptor positive modulation clarifies class specific differences, reveals unique properties of the α7 subtype, and supports a conserved potentiation mechanism among related ion channels.

Introduction

The α7 nicotinic acetylcholine receptor is expressed throughout the brain and is also found in non-neuronal cells in the periphery^1–3. It forms one of only two known homopentameric nicotinic receptors, along with the α9 subunit⁴, desensitizes more rapidly than other nicotinic receptor subtypes, and is exceptionally permeable to calcium ions^5–8. Additionally, it is strongly implicated in human disease with a decrease in receptor signaling linked to schizophrenia^9,10, Alzheimer’s disease^11,12, Parkinson’s disease¹³, and inflammation^1,14. This disease relevance has spurred the development of numerous compounds that enhance channel activation. However, α7’s rapid desensitization limits the utility of traditional agonists that mimic neurotransmitter activity^9,15. Conversely, this property makes α7 profoundly sensitive to positive allosteric modulators (PAMs). These compounds do not activate α7, but instead, enhance neurotransmitter activation^16–20, and often display increased subtype specificity¹⁵. This therapeutic potential prompted the development of many α7-targeted PAMs with a wide range of chemical diversity, pharmacological sensitivity, and efficacy^15,17,20. These compounds are coarsely classified as type I, type II, or allosteric agonist (ago) PAMs based primarily on how they affect neurotransmitter-activated macroscopic currents^17,20. Type I PAMs increase the magnitude of neurotransmitter-elicited currents, but do so without substantially altering the onset or extent of desensitization^16,18,21. Type II PAMs increase the magnitude of neurotransmitter-activated currents, but also dramatically slow the onset of desensitization and can rescue previously desensitized receptors^22–24. Finally, ago-PAMs differ from ordinary PAMs by both enhancing neurotransmitter activation and allosterically activating the receptor in the absence of neurotransmitter²⁵.

The differences in PAM characteristics²¹ and the identification of intermediate PAMs^20,26 point to complex mechanisms of α7 potentiation. PAMs have shown pro-cognitive effects in animal models^19,22,27,28, but how their varying modulatory characteristics relate to their effects on neuronal activity is controversial^19,29. These discrepancies and previous clinical disappointments^30–32 highlight the need for high-resolution structures to define modulator binding sites, understand ligand recognition, and interrogate modulation mechanisms. However, structural information for α7 has lagged behind that of other pentameric ion channels. The first structures of α7 were published in 2021^33,34 and our current structural understanding of its allosteric modulation is incomplete^34–36. To address these outstanding questions, we present a panel of α7-PAM structures complemented by electrophysiology and molecular dynamics (MD) simulations. We first discuss modulator binding sites and the residues responsible for α7 selectivity. We then contrast ion channel conformations and outline a mechanism underlying the difference between PAM classes. Next, we expand the study using a time-resolved cryo-EM approach to uncover asymmetric state transitions for α7. Finally, we investigate the structural basis of ago-PAM activity and propose an alternative gating cycle leveraged by these compounds that may hold special therapeutic potential.

Results

Biochemistry and receptor architecture

We streamlined receptor production using a stably expressed, modified human α7 gene and the chaperone nAChO³⁷ separated by a T2A self-cleaving peptide (Figure S1A). We tested both lipidic nanodisc and detergent preparations for obtaining cryo-EM structures of α7-modulator complexes. Reconstitution in lipidic nanodiscs did not allow us to resolve bound PAMs³³, but samples prepared in the mild detergent glyco-diosgenin (GDN) produced high-resolution PAM bound structures (this study, ³⁴). We interrogated the structural mechanisms of allosteric modulation using two type I PAMs, two type II PAMs, and one ago-PAM (Figures S2A–2D). In all but one case discussed below, density quality allowed for unambiguous positioning of the small molecule PAMs in an allosteric transmembrane site.

Pentameric ligand-gated ion channels including α7 are composed of five subunits surrounding an ion-conducting pore. The extracellular domain (ECD) contains the neurotransmitter binding site. It is followed by the transmembrane domain (TMD), which consists of four membrane spanning α-helices (M1-M4), with the M2 helix lining the pore. The poorly conserved and flexible intracellular domain (ICD) is located between the M3 and M4 helices (Figures S1A, S3A, and S3B)³⁸.

Type I and type II PAMs bind overlapping sites

Type I PAMs potentiate α7 activity by increasing the current amplitude with little or no effect on desensitization^17,18,39. To sample both chemical and functional diversity, we chose the large macrolide insecticide ivermectin (IVM), and the smaller, more potent NS1738 (Figures 1A–1E)^27,40. We determined cryo-EM structures of α7 with IVM and with NS1738, both in the presence of the traditional agonist epibatidine (epi), and each to an overall resolution of 2.3 Å (α7-epi/IVM, α7-epi/NS1738, Figures S1 and S2; Table S1). Both reconstructions show clear density for epi in the neurotransmitter binding site (Figure 1F). Its azabicyclo moiety points toward the principal (abbreviated +) subunit and interacts with the conserved aromatic cage. The chloropyridine ring faces the complementary (abbreviated −) subunit. An ordered water molecule bridges the hydroxyl group of T105, the backbone carbonyl of N106, and the chloropyridine nitrogen (Figures 1G–1I).

Figure 1: Type I PAM potentiation characteristics and binding sites.

(A-B) Chemical structure of (A) ivermectin (IVM) and (B) NS1738.

(C) Representative two-electrode voltage clamp (TEVC) trace showing response of WT α7 to a 10 second application of 100 μM ACh and a 20 second application of 100 μM ACh+30 μM IVM. IVM was pre-applied for 50 seconds.

(D) Representative TEVC trace showing response of WT α7 to a 10 second application of 100 μM ACh and a 20 second application of 100 μM ACh+10 μM NS1738. NS1738 was pre-applied for 20 seconds. (E) Bar graph displaying apparent fold potentiation of ACh+IVM (yellow) and ACh+NS1738 (pink) on WT α7. Data are represented as mean ± SEM (n=5 for IVM and n=6 for NS1738).

(F) Cryo-EM map of α7-epi/IVM complex depicting the neurotransmitter and modulator binding sites. The subunit on the left is the principal or (+) subunit and is colored dark blue. The subunit on the right is the complementary or (−) subunit and is colored light blue. The neurotransmitter binding site is boxed in red and the modulator binding site is boxed in yellow. IVM density is colored yellow.

(G) Top, chemical structure of the traditional agonist epibatidine (epi) and bottom, response of WT α7 to a 10 second application of 10 μM epi.

(H) Side view of the epi binding site. Epi is shown in purple and its density in transparent grey. Interacting residues are shown as sticks.

(I) View of (H), rotated 90°.

(J) Side view of the IVM’s binding site. IVM is colored yellow, and its density in transparent grey. Interacting residues are shown as sticks.

(K) Top view of (J).

(L) Bar graph showing the top 15 interaction energies during an MD simulation of the α7-epi/IVM complex. The energy value is averaged over 5 binding sites and 3 independent replicates. Residues to the left side of the dotted line represent 50% of the total interaction energy. M2 helix residues are colored in red.

(M) Side view of NS1738’s binding site. NS1738 is colored pink and its density is shown in transparent grey. Interacting residues are represented as sticks.

(N) Top view of (M).

(O) Same as (L), but for α7-epi/NS1738.

See also Figures S1, S2, S3, and S4.

The reconstructions also reveal density for the modulators bound at partially overlapping sites at subunit interfaces within the transmembrane domain (Figure 1F). Like a modulator site found in related pentameric channels^41–44, these molecules bind in the outer membrane leaflet, and interact with residues in the (+) subunit M2 and M3 helices and the (−) subunit M1 helix (Figures 1J–1O). Experimental density supports the modelling of IVM like its position in related anionic channels, and in a recent NMR structure of an α7 construct lacking the extracellular domain^{35,41,42,44,45}. Its lactone ring wedges into the transmembrane subunit interfaces and its disaccharide moiety orients away from the channel and points toward the extracellular domain. This binding pose enables hydrophobic contacts with residues in (−)M1, (+)M2, and (+)M3, as well as a possible hydrogen bond with the backbone carbonyl of L212 in (−)M1 (Figures 1J–1L).

In the α7-epi/NS1738 complex, the modulator is shifted extracellularly compared to the macrolide core of IVM and is nestled just beneath the M2-M3 loop (Figures 1M, 1N, and S3C). This binding site is consistent with earlier work implicating the M2-M3 loop in NS1738 binding⁴⁶. NS1738 binding triggers striking flexibility within the coupling region that links the extracellular and transmembrane domains (Figure S3A). It disrupts interactions between R132-Q272 and R204-D477 and causes a detachment of the latch helix that is well-ordered in all other modulator bound structures (Figures S3F–S3I). Due to the apparent flexibility within the NS1738 binding site, we used molecular dynamics (MD) simulations to guide modeling of NS1738 using multiple starting positions. The simulations confirm the flexibility implied by the diffuse density. The ligand likely samples multiple orientations, but is most stable with the trifluoromethyl group pointing intracellularly and away from the pore (Figures S4A–S4H).

Like IVM and NS1738, type II PAMs enhance the magnitude of agonist-elicited current. However, they also dramatically extend the lifetime of the activated state by slowing desensitization (Figures 2A–2E, S3O, and S3P) and can re-activate previously desensitized receptors^22,24,47. Recently, a structure of α7 bound to the type II PAM PNU-120596^22,48,49 (PNU) was published³⁴; we repeated this structural experiment to improve resolution and directly compare it to the other α7 complexes presented here. We also selected the type II PAM (−)-TQS. Racemic TQS has served as the chemical backbone for developing numerous α7-PAMs and ago-PAMs with diverse characteristics^23,24. For simplicity, we refer to the more active (−)-TQS isomer⁵⁰ as TQS.

Figure 2: Type II PAM potentiation characteristics and binding sites.

(A-B) Chemical structure of (A) PNU-120596 (PNU) and (B) (−)-TQS (TQS).

(C-D) Representative TEVC trace showing the response of WT α7 to a 10 second application of 100 μM ACh and a 20 second application of 100 μM ACh+10 μM PAM. For the potentiated response, PAM was pre-applied for 10 seconds before co-application of ACh+PAM. (C) PNU, (D) TQS.

(E) Superposition of WT α7 response to ACh alone (black) and ACh+PNU (green, top) or ACh+TQS (salmon, bottom).

(F) Cryo-EM map of α7-epi/PNU complex. The intersubunit binding site is boxed in yellow and PNU density is colored yellow. The + subunit is in dark blue and the − subunit is in light blue. (G) Side view of the PNU binding site. PNU is colored green with its corresponding density in transparent grey. The residues interacting with PNU are shown as sticks. Lipid molecules are hidden for clarity. See also Figures S3 and S4.

(H) Top view of (G).

(I) Bar graph showing the top 15 interactions energies throughout the simulation of the α7-epi/PNU complex. The energy value is averaged over 5 binding sites and 3 independent replicates. Residues to the left of the black line represent 50% of the total binding energy. M2 helix residues are colored red. (J) Side view of the TQS binding site. TQS is colored salmon with its corresponding density in transparent grey. The interacting residues are shown as sticks. Lipid molecules are hidden for clarity.

(K) Top view of (J).

(L) Same as (I), but for α7-epi/TQS

See also Figures S1 and S2.

We determined cryo-EM structures of α7-epi/PNU and α7-epi/TQS complexes to overall resolutions of 2.6 Å and 2.3 Å, respectively (Figures S1 and S2; Table S1). These type II PAMs occupy an intersubunit transmembrane binding site similar to the one seen for the type I PAMs (Figure 2F). PNU binds with its long axis perpendicular to the channel pore. It interacts with residues in (−)M1, (−)M2, (+)M2, and (+)M3 and is positioned to form a hydrogen bond with (−)M1 N213 (Figures 2G–2I). Unexpectedly, this orientation is flipped about its long axis compared to the previously published structure³⁴; the revised orientation is unambiguously supported by the higher resolution maps and stability in MD simulations (Figures S4I–4N). Finally, there is density for a putative lipid molecule that wraps around PNU forming the floor of the binding site (Figures S3J and S3K). The other type II modulator, TQS, is positioned in an overlapping site slightly below PNU and oriented parallel to the channel axis. TQS also makes extensive interactions with hydrophobic residues in (−)M1, (+)M2, and (+)M3. Its sulfonamide group forms the lone electrostatic interaction with (−)M1 N213 (Figures 2J–2L). Together, all four modulators bind to a transmembrane subunit interface analogous to a modulator site found in other pentameric receptors^41–44. Binding site variability exists both within and between modulator classes (Figures S3C–S3E), but both type II PAMs make additional interactions with the pore-lining M2 helices compared to the type I PAMs (Figures 1J–1O and 2G–2L).

Side chain volume and flexibility underlie α7 modulator selectivity

An often-cited benefit of allosteric modulators compared to traditional agonists is that they target less conserved sites allowing enhanced subtype selectivity^17,20,39. The compounds studied here only potentiate α7 among nicotinic receptors, however they have recently been found to act on related receptors^41,42,51. A comparison of α7’s PAM binding site with other nicotinic receptors shows that insensitive subtypes have larger, rigid residues replacing the smaller, more flexible ones in α7 (Figure S3U)³⁴. Structural analysis of the α7-epi/PAM complexes highlights three α7 residues that likely contribute to regulating modulator activity: N213, M253, and A275³⁴. N213 in (−)M1 is positioned to interact with all four modulators (Figures 1J–1O and 2G–2L). Consistent with this observation, an N213A mutation almost entirely eliminates the activity of the four PAMs in two electrode voltage clamp (TEVC) experiments (Figures 3A–3D). While this residue clearly regulates PAM activity, its conservation among nicotinic receptors means it is unlikely to be driving α7 selectivity (Figure S3U). On the other hand, M253 in (+)M2 is an outlier, and has long been implicated in α7 PAM activity. Replacing M253 with a leucine, the equivalent residue in most other nicotinic receptor subunits, has previously been shown to decrease or eliminate modulator activity for all four of these PAMs (Figure 3)^39,49,52,53. Consistent with these findings, M253 is positioned to interact with the aforementioned PAMs, but its rotameric conformation is strikingly modulator dependent (Figure S3L). We mutated M253 to alanine to better define its role in regulating modulator activity. The smaller alanine residue presumably disrupts the M253-modulator interaction without blocking the binding site. NS1738 and PNU recovered near full modulator activity on α7 M253A, and IVM potentiated α7 M253A to levels even greater than WT (Figures 3A–3C). TQS’s activity, however, did not recover (Figure 3D). These results suggest that NS1738, PNU, and IVM rely heavily on the binding site volume permitted by the flexible M253, whereas TQS relies more on its interaction with M253.

Figure 3: Binding site mutations differentially alter modulator activity.

(A-D) Bar graphs showing apparent fold potentiation from biologically independent replicates n≥3. 100 μM ACh was used for all conditions except N213A and M253A where 300 μM was used due to a decrease in overall channel activity. 10 μM NS1738, TQS, and PNU, and 30 μM IVM was used for all conditions. Statistical significance was determined using a one-way analysis of variance (ANOVA) followed by a Dunnett’s multiple comparison test comparing the mutant and WT. For comparison of two mutants, significance was assessed using an unpaired two-tailed student’s T test. Statistical thresholds were set as ns P>0.05, * P<0.05, ** P<0.01, *** P<0.001, **** P<0.0001. (A) IVM, (B) NS1738, (C) PNU, (D) TQS.

(E) Summary table showing apparent mean fold potentiation ± SEM.

See also Figure S3.

While M253 certainly plays a role in α7-PAM activity, previous results have shown it is not the sole determinant⁵⁴. Therefore, we also considered the (+)M3 residue A275. Like M253, A275 is an outlier in the nicotinic receptor family (Figure S3U); it is smaller than the L, V, or M residues at equivalent positions of the insensitive nicotinic receptor subtypes. This same position has been implicated in IVM activity on related pentameric receptors^55,56, thus we tested the role of A275 on the activity of the other modulators by making A275L and A275I mutations. A275L turns NS1738 and PNU into weak negative modulators, while TQS shows reduced potentiation ability. A275I also turns NS1738 and PNU into weak negative modulators, but TQS surprisingly retains some of its type II PAM characteristics suggesting it is more tolerant to larger residues at this position (Figures 3B–3D). This tolerance helps explain previous work that demonstrated TQS’s ability to potentiate mutated nicotinic receptors⁵⁰ and highlights an interplay between residues in (+)M2 and (+)M3 that helps control PAM activity. These results confirm the modulator binding sites observed above and underscore the roles of both side chain volume and flexibility of (+)M2 and (+)M3 residues in regulating modulator activity. Importantly, they also demonstrate the structural differences in how individual PAMs selectively potentiate α7.

Modulator classes stabilize distinct pore conformations

We turn our attention to the ion pore to annotate the functional states stabilized by type I and type II PAMs. Residues within the M2 helix define the transmembrane portion of the conduction pathway and are numbered from the cytosolic end (0ʹ) to the extracellular end (20ʹ), with the conserved activation gate in the middle of M2 at 9ʹ (Figure S3B,⁵⁷). Neurotransmitter binding, the first step in the gating cycle, is translated to the channel pore through the coupling region (Figure S3A). These changes disrupt hydrophobic interactions in the extracellular half of M2, including the L9ʹ activation gate, and result in pore hydration and ion permeation³³. Desensitization encompasses a subtler movement, generally the closure of a distinct gate at the intracellular end of the pore to disrupt ion flow^58–60. However, previous α7 desensitized-state structures show the L9ʹ side chains pointing into the pore and forming a hydrophobic constriction^33,34. These structures, paired with MD simulations⁶¹ and mutational studies⁶², imply a special role for L9ʹ in α7 desensitization.

To better understand the effect of PAM binding on pore conformation, we determined a reference structure of α7 with only epi bound to a resolution of 2.3 Å (α7-epi, Figures S1 and S2; Table S1). The α7-epi structure is nearly identical to previously determined desensitized states and has a constriction at L9ʹ (4.0 Å diameter, Figure 4A)^33,34. The pore of the α7-epi/IVM complex closely resembles that of α7-epi with the tightest point also found at L9ʹ (3.6 Å diameter, Figure 4B). Interestingly, the α7-epi/NS1738 pore most closely matches a resting state^33,34 with multiple tight hydrophobic constrictions at the top of the pore (L9ʹ: 2.4 Å, Figure 4C). However, all other regions of the receptor resemble a desensitized state. Therefore, the more resting-like pore profile of the α7-epi/NS1738 complex likely arises from local NS1738-triggered flexibility rather than a global state transition. MD simulations probing the hydration state of these structures find that the pore is dehydrated and impermeable throughout the simulation (Figures 4H–4J). Given the pore profiles, MD simulations, and the similar desensitization kinetics of currents in the presence of type I PAMs, we annotate the α7-epi/type I PAM complexes as desensitized states (Figure 4F).

Figure 4: Type I and type II PAMs stabilize distinct pore conformations.

(A-E) Permeation pathway depicting hydrophobicity and pore diameter of (A) α7-epi, (B) α7-epi/IVM, (C) α7-epi/NS1738, (D) α7-epi/TQS, (E) α7-epi/PNU. Top, surface representation of three subunits with two L9ʹ residues shown as sticks and colored tan. The approximate membrane is in light grey. Bottom, two M2 helices with pore diameters (Å) at points of interest indicated by dashed lines. In the α7-epi/TQS complex, the density for L9ʹ suggests multiple conformations, with one pointing into the pore and one rotated out of the pore, we modelled an intermediate conformation.

(F-G) Pore profile traces comparing α7-epi and (F) α7-epi/type I PAM or (G) α7-epi/type II PAM complexes with the activated state structure (7KOX)³³.

(H-L) Backbone restrained (50 kJ/mol/nm) MD simulations probing pore hydration and ion permeability of the α7 structures. (H) α7-epi, (I) α7-epi/IVM, (J) α7-epi/NS1738, (K) α7-epi/TQS, and (L) α7-epi/PNU. Blue circles represent water and red circles represent sodium ions. Simulations were repeated three times with similar results.

Consistent with their strong opposition of channel desensitization, the α7-epi/type II complexes are expanded at the top of the pore compared to α7-epi and the α7-epi/type I complexes (Figures 4D and 4E). The α7-epi/TQS complex L9ʹ pore diameter increases to 6.8 Å; its narrowest point is near the cytosolic junction at G-2ʹ (6.0 Å, Figure 4D). The α7-epi/PNU complex shows L9ʹ fully rotated out of the pore (diameter: 12.0 Å), but a similarly constricted diameter at G-2ʹ (5.2 Å, Figure 4E). Both type II PAM complexes have minimum pore diameters larger than previous desensitized states, but tighter than both the putative activated state (Figure 4G)³³ and the pore diameter suggested by permeation studies^63,64. MD simulations find the α7-epi/TQS structure has a mostly dehydrated pore (Figure 4K) but is more hydrated than the desensitized conformations above (Figures 4H–4J). The α7-epi/PNU complex shows a hydrated and partially ion-permeable pore (Figure 4L). These conflicting results suggest that the α7-epi/type II PAM complexes are partially activated or partially desensitized states; for simplicity, we refer to them as desensitized intermediates.

Time-resolved cryo-EM reveals asymmetry of the channel gate

To better define the conformational state of type II PAM complexes and potentially capture a fully activated state with PNU bound, we used a time-resolved cryo-EM approach. Recent approaches in this area include rapid spraying or photo uncaging immediately prior to plunge freezing^65–67. Instead, we simply altered our grid freezing procedure by preserving the long incubation with PNU but adding epibatidine just seconds before plunge-freezing (Figure S5A). While this approach did not yield a fully activated state (discussed further below), it produced four distinct conformations (Figures S5 and S6; Table S2). First, we found a symmetric, resting-like state, indicating that a portion of the receptors (~4%) did not activate before freezing. This resting state lacks density for PNU in the intersubunit site, suggesting PNU cannot bind that state, and the top of the pore is tightly closed (Figures S5C and S6S). We next found two PNU-bound asymmetric states with one or more L9ʹ residues pointing into the pore, and other L9ʹ residues rotated out (Figures S5C, S6T, and S6U). Finally, the majority of the particles were found in a symmetric, PNU-bound desensitized intermediate state (Figures S5C and S6V; Video S1). This state is virtually identical to the equilibrium α7-epi/PNU structure. Reanalysis of the equilibrium α7-epi/PNU dataset using the time-resolved data processing strategy (Methods) showed no evidence of a resting-like conformation or meaningful asymmetry within the pore. While this experiment did not clarify the state of the α7-epi/type II PAM complexes, it suggests that α7 transitions may occur in an asymmetric rather than in a fully synchronous and concerted manner. We expand on the consequences of the asymmetric transition in the discussion and use the highest-resolution α7-epi/PNU desensitized intermediate structure (2.2 Å) to probe potentiation mechanisms.

L9′ rotation underlies PAM activity and channel activation

A major goal of this study was to investigate the mechanistic distinctions among α7 PAM classes. The complexes show a class dependent increase in M2 helix tilting and L9ʹ rotation (Figures 5A–5D). The experimental density for L9ʹ supports modeling this residue pointing into the pore in the α7-epi and α7-epi/type I PAM structures but rotated out in the α7-epi/type II PAM structures (Figures 5E–5H). This difference is not just a rotameric shift but involves a larger conformational change within the M2 helix. Type II PAM binding results in a tilting of the M2 helix away from the pore and an overwinding of the helix causing it to deviate from standard α-helical geometry. This secondary structure breakdown spans two helical turns starting at 7ʹ and returning to an α-helix at 14ʹ (Figures 5G and 5H). The breakdown is most evident for PNU but is also seen in the TQS complex. For simplicity, we refer to this conformational change in M2 as L9ʹ rotation.

Figure 5: Type II PAM complexes reveal L9ʹ rotation and M2 helical backbone distortion.

(A) Overlay of α7-epi (white) and α7-epi/PNU TR desensitized-intermediate (blue) models.

(B) Pore view of α7-epi (white) and α7-epi/PNU TR desensitized-intermediate (blue) with TMD helices shown as cylinders and L9ʹ residues as sticks.

(C) Pore view of α7-epi/IVM (grey) and α7-epi/PNU TR desensitized-intermediate (blue) with TMD helices shown as cylinders and L9ʹ residues as sticks.

(D) Side view of two M2 helices of α7-epi/IVM (grey) and α7-epi/PNU TR desensitized-intermediate (blue).

(E)-(H) M2 helices shown as sticks (E) α7-epi, (F) α7-epi/IVM, (G) α7-epi/TQS, and (H) α7-epi/PNU TR desensitized-intermediate. Cryo-EM density is displayed in transparent grey. Right panel shows residues T6ʹ-L16ʹ with side chains and experimental density hidden for clarity. The backbone hydrogen bond distances are in Å.

See also Figure S5.

While the L9ʹ rotation clearly distinguishes type I and II PAMs complexes, how this conformational change relates to α7 activation and whether it is specific to type II PAM bound channels is unclear. In the previously published α7 activated state³³, L9ʹ was modeled pointing toward the pore, but the weak density for this residue suggests it adopts multiple conformations. Additionally, structures and MD simulations of related pentameric channels show L9ʹ rotated out of the pore and buried in an intersubunit pocket during activation^44,68–72. In α7, this pocket is primarily composed of hydrophobic residues but also contains two polar residues, T6ʹ and S10ʹ (Figures 6A and 6B). To test how the L9ʹ rotation relates to channel activation, we designed mutations aiming to stabilize the rotated-out conformation by replacing T6ʹ, S10ʹ, or both residues with alanine, thus increasing the hydrophobicity of the pocket. We reasoned that if type II PAMs are uniquely rotating L9ʹ out of the pore, these mutations should only affect type II PAM bound channels, whereas if the L9ʹ rotation is a general characteristic of channel activation, these mutations should affect all responses. We first performed MD simulations to confirm the activity of the designed mutants. Unrestrained simulations on in silico mutated channels and analysis of the L9ʹ rotameric state revealed that making this pocket more hydrophobic (T6ʹA+S10ʹA) indeed resulted in L9ʹ favoring the rotated-out conformation, whereas making the pocket more hydrophilic (I221S+L224S+I243S) resulted in L9ʹ favoring a rotated-in conformation (Figure S7A–S7H).

Figure 6: L9ʹ rotation underlies modulator activity and channel activation.

(A) View from the channel pore of the α7-epi/PNU model. L9ʹ is rotated out of the pore and buried between T6ʹ and S10ʹ. The – subunit is shown in light blue and + subunit is shown in darker blue. PNU is shown in green.

(B) Top view of (A) with PNU hidden.

(C) Representative traces of single channel currents (top) and corresponding cluster duration histograms fitted by the sum of exponentials (bottom) for WT, T6ʹA, S10ʹA, and T6ʹA+S10ʹA in the presence of 100 μM ACh.

(D) Representative traces of T6ʹA mutant with no agonist (spontaneous channel openings) and in the presence of 10 μM PNU alone (top). Corresponding cluster duration histograms are fitted by the sum of exponentials (bottom).

(E) Representative traces and cluster duration histograms for ACh + PAM fitted by the sum of exponentials for WT (top row) and T6ʹA (bottom row). 100 μM ACh + 10 μM NS1738, 10 μM TQS, or 10 μM PNU was used for WT channel recordings and 100 μM ACh + 10 μM NS1738, 10 μM TQS, or 1 μM PNU was used for T6ʹA channel recordings.

(F) State diagram inferred from single channel analysis with activated states (O) increased by the mutants in green, and subscripts denoting bound agonist (A) and/or PAM (P). Mutants are noted above equilibria that they likely alter.

See also Figure S7.

After testing the effect of these mutations on the L9ʹ rotation in silico, we performed single channel recordings and found that the T6ʹA and S10ʹA mutations stabilize the activated state, but T6ʹA does so more broadly (Figure 6). The T6ʹA mutant shows channel openings in the absence of ACh, prolongs the lifetime of ACh-elicited clusters of channel openings, and exhibits clusters of channel openings in the presence of PNU alone. These effects were not observed in WT or S10ʹA channels (Figures 6C and 6D). Combining these mutations (T6ʹA+S10ʹA) stabilizes the activated state more than T6ʹA alone, revealing synergy between the two mutations (Figure 6C). We next examined channel openings with ACh+PAM. Interestingly, neither mutant prolongs cluster duration lifetime in the presence of ACh+PNU (Figures 6E and S7I). Looking more macroscopically, but still with single channel resolution, we find that both mutants enhance activation by reducing the equilibrium extent of desensitization (Figure S7I). We therefore suggest that PNU exhibits a ceiling effect, where it achieves a maximal cluster duration for WT and cannot be further increased by either mutation. Due to this ceiling effect, we tested the intermediate efficacy PAMs NS1738 and TQS. As expected, the intermediate PAM cluster lifetimes fall between unpotentiated (ACh) and maximally potentiated (ACh+PNU, Figure 6E). These responses correlate with the modulator strength observed in TEVC experiments and both show an increase in cluster lifetime on the T6ʹA mutant (Figure 6E).

These results may be interpreted using a state diagram including association of both ACh and PAM to resting, activated, and desensitized states (Figure 6F). The S10ʹA mutant stabilizes the activated state with both ACh and PAM bound, while not having a strong enough effect to detect changes in the stability of other activated states. On the other hand, the T6ʹA mutant stabilizes all four activated channel states. Structurally, enhancing the rotated-out conformation of L9ʹ appears to stabilize the activated state at the expense of the resting and desensitized states. These functional and dynamic results suggest that L9ʹ is rotated out of the pore in all forms of channel activation and that PAMs function to stabilize this rotation.

Allosterically activated channels undergo a unique gating cycle

With a better understanding of type I and type II PAM modulation, we next explored the mechanisms of allosteric agonists or ago-PAMs. These compounds potentiate and allosterically activate α7 with kinetics differing from traditional agonists^25,39. The best studied α7 ago-PAM is 4BP-TQS, which is a derivative of the type II PAM TQS, but has a bromophenyl group replacing TQS’s naphthalene substituent^25,39. Of the two enantiomers, only (+)-4BP-TQS was found to be active, and was renamed GAT107 (Figure 7A)⁷³. To probe the gating cycle of allosterically activated α7, we determined the structure of α7-GAT107 (without epi) and α7-epi/GAT107 (Figures S1 and S2; Table S1). Both reconstructions show strong density for GAT107 only at the intersubunit PAM site in the transmembrane domain (Figures 7B and 7C). GAT107 is stabilized primarily by interactions with hydrophobic residues in the (−)M1, (+)M2, and (+)M3 transmembrane helices. Additionally, like TQS, the sulfonamide group is positioned to form a hydrogen bond with N213 in (−)M1. Unexpectedly, the α7-GAT107 and α7-epi/GAT107 structures reveal different pore conformations (Figures 7D–7H). α7-GAT107 adopts a resting-like conformation^33,34, with tight, hydrophobic constrictions in the pore at the L9ʹ, V13ʹ, and L16ʹ positions (Figures 7D and 7G). This differs from our time-resolved experiment in which the type II PAM PNU was unable to bind in the absence of traditional agonist. It also suggests that after GAT107 directly activates the channel, α7 reverts to a GAT107-bound resting-like conformation. α7-epi/GAT107, however, adopts a desensitized intermediate conformation similar to the α7-type II PAM complexes (Figures 7E and 7H). Accordingly, these structures suggest an alternative ago-PAM gating mechanism that reconciles observed phenomena: direct activation that is transient but potentiates the response to subsequent neurotransmitter binding (Figure 7I)⁷³.

Figure 7: Allosteric agonists exploit an alternative gating cycle.

(A) GAT107 chemical structure.

(B) Side view of the GAT107 binding site. + subunit is colored dark blue and the − subunit light blue. GAT107 is shown in dark grey and the cryo-EM density in transparent grey. The interacting residues are shown as sticks.

(C) Top view of (B).

(D-E) Pore permeation profiles showing hydrophobicity and pore diameter for GAT107 bound structures: (D) α7-GAT107 and (E) α7-epi/GAT107.

(F) Pore profile trace of α7-GAT107 (red) and α7-epi/GAT107 (blue) compared with an apo, resting conformation (7EKI) in grey³⁴ and the putative activated conformation (7KOX) in blue³³.

(G-H) Two M2 helices with pore diameters (Å) at points of interest are indicated by dashed lines. (G) α7-GAT107 and (H) α7-epi/GAT107.

(I) Simplified state diagram of ago-PAMs. Red arrows show state diagram for α7 activated by GAT107 alone.

Discussion

Here we sought to define structural principles of allosteric modulation for a receptor that is broadly expressed in the brain and periphery, has been targeted by many potent modulators with hopeful therapeutic outcomes, and yet has remained enigmatic in its response to agonists and modulators. We built a structural foundation by obtaining high-resolution complexes of α7 with a panel of allosteric modulators that differ tremendously in their chemistry and modulatory efficacies. Somewhat surprisingly, we found that all of the modulators share a common site (Figures 1, 2, and 7). This transmembrane pocket is emerging as a consensus site across the superfamily for important drug classes, including general anesthetics at the GABA_A receptor⁴³, the antiparasitic site in the invertebrate glutamate-gated chloride channel⁴¹, and the ivermectin site in the glycine receptor^42,44,45. While the modulator sites overlap, interactions outside their common locus and effects on receptor conformation diverge. Electrophysiological experiments and molecular dynamics bolster the structural findings and extend their value in testing hypotheses regarding modulator selectivity and desensitization mechanisms. Several questions emerging from the structure-function analysis merit deeper discussion. Below we highlight the consequences of this work on understanding selectivity of the modulators among closely related receptors and ways in which modulators can tune channel activity. We then discuss important challenges in associating static structures with physiological states, and how allosteric agonists may differ from both traditional agonists and simpler positive allosteric modulators.

PAM selectivity

A mystery in the field has been how α7 PAMs can be selective for α7 among nicotinic receptors but remain active on more distantly related receptors⁵¹. We show that while PAMs have slight variations in binding determinants for α7, all four are strongly regulated by a pair of amino acids in M2 and M3 and depend heavily upon side chain volume and flexibility (Figures 1–3). These findings underscore the importance of a small number of amino acids and suggest an interplay of residues in regulating PAM activity⁵⁰. Additionally, the presence of smaller and more flexible residues regulating binding to this site extends to other members of the superfamily^74–76 and may help explain the recently observed activity of some of these same modulators on synaptic GABA_A receptors^51,77 (Figure S3U).

α7 PAM potentiation mechanism

Our insights into PAM binding led us to interrogate how these molecules potentiate α7. The α7-PAM complexes show that PAM efficacy correlates with the number of M2 helix interactions (Figures 1 and 2), and with the rotation of L9ʹ away from the pore axis (Figures 4 and 5). We then connected the rotation of L9ʹ to channel activation (Figure 6) and showed that PAM-like behavior is conferred by mutations that stabilize the 9ʹ rotation either by weakening the hydrophobic interactions made with other subunits (L9ʹT⁶²) or increasing the intersubunit pocket’s hydrophobicity (Figures 6 and S7). These results suggest that PAMs enhance α7 activity by increasing the favorability and stability of the L9ʹ rotation that underlies channel activation. The extra interactions between type II PAMs and the pore lining M2 helices (Figures 1 and 2) provide further stabilizing effects resulting in longer clusters of channel openings compared to type I PAMs. Additionally, single channel recordings show that both PAM classes increase cluster durations, but to differing extents (Figure 6 and S7)²¹. Therefore, we suggest that the difference between PAM classes is a distinction in potentiation strength rather than a difference in the underlying mechanisms, analogous to partial versus full agonists acting through the neurotransmitter binding site^78–80. Put simply, both modulator classes function to stabilize L9ʹ rotated out of the pore, but to differing degrees.

While the type I and type II PAM classification is unique to α7, the structural mechanism by which PAMs potentiate α7 is not. In fact, it is strikingly similar to GABA_AR-PAM complexes that also show varying degrees of L9ʹ rotation out of the pore^43,81. The similar mechanism, however, does not result in the same dramatic difference in PAM characteristics as it does with α7 PAMs. Instead, the exceptional efficacy of type II PAMs acting on α7 likely stems from the primary role of L9ʹ in α7 desensitization (Figures 4 and 6)^24,33,61,62. α7 PAMs are able to stabilize the desensitization gate rotated out of the pore. In contrast, the GABA_A and glycine receptors have a desensitization gate at the intracellular mouth of the pore⁵⁹ limiting the efficacy of PAMs targeting these receptors.

Conformational state assignment and asymmetric transitions

With the recent boom of channel structures, accurately annotating pore conformation is a challenging but important task. Here we used pore diameter measurements, MD simulations, and time-resolved cryo-EM to help define conformational states (Figure 4). α7-epi and α7-epi/type I PAM complexes are assigned as desensitized states. Comparison of these structures with previous work supports a strong consensus for the conformation of an α7 desensitized state with similar results from multiple constructs, membrane mimetics, and research groups, promoting confidence in its physiological relevance^33–35. Assignment of the α7-epi/type II PAM complexes is more difficult. They represent an intermediate of an activated and desensitized conformation. Our time-resolved cryo-EM experiment failed to uncover an unambiguously activated state, which would have aided our assignment of the α7-epi/type II PAM complexes. A possible explanation is related to a previously described discrepancy in PNU potentiation of single channel (~100,000x) versus macroscopic currents (~30–100x)¹⁷. This discrepancy suggests that only a small fraction of receptors is potentiated at a time, but by a very large amount^17,82, meaning the population of activated states in the sample may be too small to detect in cryo-EM data processing. Alternatively, the desensitized-intermediate conformation could represent a distinct desensitized⁸² or partially activated state, but questions for each of those assignments remain. Importantly, a caveat for every membrane protein structure is that the membrane mimetic is an imperfect replica of the native membrane and may alter the protein’s conformation⁸³. Regardless, our time-resolved experiment reveals interesting asymmetry within the pore that underscores the importance of L9ʹ and its rotation out of the pore in channel activation. It also supports a partially concerted mechanism for α7 state transitions with one or more L9ʹs rotating out of the pore making the subsequent rotations more favorable. Finally, the most significant asymmetry occurs at the level of L9ʹ which suggests, speculatively, that its rotation out of the pore may be a final step in activation.

Ago-PAMs

To gain further insight into potentiation, activation, and desensitization, we challenged α7 with the ago-PAM GAT107. Allosteric agonists like GAT107 behave as PAMs at low concentrations and agonists at higher concentrations^17,39,73. Previous work on GAT107 proposed a secondary binding site for allosteric agonists with the TMD binding site responsible for PAM activity and a separate ECD site regulating allosteric agonist activity^17,73,84. While this hypothesis is intriguing, we did not find clear density for a second GAT107 site, suggesting that both PAM and allosteric agonist activity are accomplished through the transmembrane site²⁵. Furthermore, our investigation of GAT107 suggests it alters the receptor’s canonical gating cycle (Figure 7I). We propose that instead of long applications desensitizing the channel like a traditional agonist, GAT107-activated receptors slowly transition back to a GAT107 bound resting-like conformation. Only after the addition of a traditional agonist does it desensitize. This reduced desensitization is consistent with functional observations of both α7^23,25 and other pentameric receptor allosteric agonists^85,86. Additionally, this gating cycle explains the ability of α7 ago-PAMs to prime the receptor for future activation events⁷³. This ago-PAM activity, which we propose stems from GAT107 binding to a resting-like state in the absence of a traditional agonist, distinguishes it from PNU. The prototypical type II PAM does not bind in the absence of an agonist, as revealed by our time-resolved cryo-EM analysis (Figure S5). Excitingly, this mechanism overcomes a major downside of using traditional agonists to increase α7 activity as they result in the accumulation of desensitized receptors.

Summary

Our investigation of α7 positive allosteric modulators highlights their therapeutic promise and reveals the value of using α7 as a model system for this study. The benefits were reciprocal: α7’s unique characteristics revealed the basis of modulator activity and these modulators provided further insight into α7 channel function. Additionally, the varying efficacies of these modulators (type I < type II < ago-PAMs) present an opportunity to selectively target disease states with different levels of severity. Finally, the mechanisms of modulator selectivity and activity extend to the entire superfamily, and depict an exciting future in drug development of positive modulators targeting pentameric ligand-gated ion channels.

Limitations of the Study

To aid protein expression and purification, we altered the α7 construct. While this construct functions similarly to WT, we cannot completely rule out subtle differences. As discussed above, the exact physiological state of the type II PAM complexes is still unclear and will require further work to convincingly annotate this conformation. Our time-resolved cryo-EM experiment revealed asymmetry in α7 state transitions, but further work is needed to precisely determine intermediate states and smaller degrees of asymmetry related to channel activation. Lastly, our broad conclusion that all PAMs act, through varying degrees, by enhancing stability of the desensitization gate rotated out of the pore, is a simplification. The PAMs are chemically diverse, and while they bind a similar site, their contacts are different. The fine differences are of interest, however the results presented here shed light on the broader properties of allosteric modulation and gating mechanisms across the superfamily.

STAR METHODS

Resource Availability

Lead Contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Ryan Hibbs (rehibbs@ucsd.edu).

Materials Availability

All unique/stable reagents generated in this study are available from the lead contact without restrictions.

Data and Code Availability

• Cryo-EM density maps and atomic models for α7-epi, α7-epi/IVM, α7-epi/NS1738, α7-epi/PNU (equilibrium), α7-epi/TQS, α7-GAT107, α7-epi/GAT107, α7 resting (time-resolved), α7-epi/PNU asymmetric 1 (time-resolved), α7-epi/PNU asymmetric 2 (time-resolved), α7-epi/PNU desensitized-intermediate (time-resolved) have been deposited in the Protein Data Bank and Electron Microscopy Data Bank and are publicly available as of the date of publication. The PDB IDs are 8UT1, 8UZJ, 8UTB, 8V82, 8V80 8V86, 8V88, 8V89, 8V8C, 8V8D, and 8V8A. EMDB IDs are EMD-42526, EMD-42841, EMD-42537, EMD-43015, EMD-43012, EMD-43025, EMD-43030, EMD-43031, EMD-43034, EMD-43035, and EMD-43032.

• This paper does not report any original code.

• Any additional information required to reanalyze the data reported in this work paper is available from the Lead Contact upon request.

Experimental Model and Study Participant Details

HEK293S GNTI⁻ suspension cells were obtained from the ATCC (Cat# CRL-3022) and cultured at 37°C and 8% CO₂.

Method Details

Protein expression

To aid expression and purification, the human α7 gene (Addgene #62630) was modified by deleting 39 residues and adding the thermostable protein bRIL¹¹⁰ in the disordered M3-M4 loop to boost thermostability and expression³³. A twin Strep-II affinity tag (two Strep-II tags joined by a (GGGS)₂GGSS linker) was attached to the C-terminus of α7 and then fused with the chaperone protein nAChO³⁷ linked by a self-cleaving T2A sequence. This construct was then subcloned into pSBtet-GP (pSBtet-GP-α7_EM). The modified α7 gene was previously tested for function³³. The pSBtet-GP vector was a gift from Eric Kowarz (Addgene plasmid # 60495).

HEK293S cells lacking N-acetylglucosaminyltransferase (GnTI⁻) were cultured in DMEM supplemented with 10% FBS at 37°C with 8.0% CO₂. The Sleeping Beauty transposase system was used to generate a cell line stably expressing α7. HEK293S GnTI⁻ cells were transfected with plasmid DNA (pSBtet-GP-α7_EM + pCMV(CAT)T7-SB100) using Lipofectamine 2000. The pCMV(CAT)T7-SB100 plasmid was a gift from Zsuzsanna Izsvak (Addgene plasmid # 34879). Puromycin was added (1.0 μg/mL) to the media 24 hours post transfection to select for cells stably expressing α7. Selection was carried out for 5–7 days and terminated after >90% of the cells exhibited green fluorescence. After confirming the expression of the target gene, the cells were adapted to suspension culture in Freestyle media supplemented with 2% fetal bovine serum (Gibco) and penicillin streptomycin (Gibco). They were grown to 4.8 L at a density of 2.5–3.0 × 10⁶ per mL and then induced with 2.0 μg/mL of doxycycline. The cells were harvested 48–72 hours after induction by centrifugation. The cell pellet was resuspended in TBS buffer (20 mM Tris pH 7.4 + 150 mM NaCl) supplemented with 1 mM phenylmethylsulfonyl fluoride (PMSF). Cells were lysed with an Avestin Emulsiflex and centrifuged for 20 min at 10,000 g. The membrane fraction was collected by ultracentrifugation (2 hours at 186,000 g) and stored at −80°C before use.

Protein purification

The membrane pellet was thawed in TBS + 5 mM ethylene glycol tetraacetic acid (EGTA) + 1 mM PMSF and homogenized using a Dounce homogenizer. The membrane was solubilized in 40 mM n-dodecyl-β-D-maltoside (DDM, Anatrace) for 1-hour at 4°C. The solubilized membrane was centrifuged at 186,000 g for 40 mins at 4°C. The membrane containing supernatant was collected and passed through a Strep-Tactin Superose affinity resin via gravity flow. The resin was washed with 10 column volumes (CV) of TBS + 0.2 mM glyco-diosgenin (GDN, Anatrace) + 5 mM EGTA, and eluted with 6 × 0.5 CV TBS + 0.2 mM GDN + 5 mM EGTA + 5 mM desthiobiotin. Elution fractions were checked via fluorescence size exclusion chromatography (FSEC)¹¹¹ measuring tryptophan fluorescence and pooled. The sample was concentrated and injected over a Superose 6 Increase 10/300 GL column (GE Healthcare) using a mobile phase of 20 mM Tris (pH 7.4) + 150 mM NaCl + 0.2 mM GDN + 5 mM EGTA. Fractions corresponding to the properly sized peak were collected, analyzed via FSEC, and concentrated to ~7.0–10 mg/mL. Modulator was added at 400 μM for one-hour nutating at 4°C, then the traditional agonist epibatidine (200 μM, Tocris) was added for one-hour at 4°C. The sample was centrifuged and checked via FSEC before freezing grids.

Cryo-EM sample preparation

Copper 1.2/1.3 200 mesh holey carbon grids (Quantifoil) were glow discharged (PELCO easiGlow) for 80 seconds at 30 mA. To remove excess Ca²⁺ present in the blotting paper¹¹², the blotting paper was soaked in a 2 mM EGTA solution. The solution was changed three times and the blotting paper was vacuum dried overnight to remove residual moisture. The final sample (3 μL) was applied to the glow discharged grid at 4°C + 100% humidity. Excess liquid was immediately removed by blotting for 3.0 sec and the grid was then plunge-frozen in liquid ethane using a Thermo Fisher Vitrobot Mark IV.

Time-resolved cryo-EM sample preparation

PNU-120596 (Tocris) was added to 400 μM for one-hour nutating at 4°C. Aliquots (0.75 μL) of epibatidine (Tocris) were added to PCR tubes. α7-PAM solution (3 μL) was added to the epibatidine containing PCR tube and mixed briefly by pipetting, resulting in a sample with an epibatidine concentration of 200 μM. The α7-PAM-epibatidine solution (3 μL) was immediately applied to the grids and frozen under the same conditions as the equilibrium preparation. This process was repeated in a new PCR tube for each grid to prevent contamination. The estimated time from addition of epibatidine to plunge freezing in liquid ethane was 10–15 seconds.

Cryo-EM data collection and data processing equilibrium α7-PAM complexes

Dose-fractionated images for the α7-epi/PAM equilibrium complexes were collected on a Titan Krios (Thermo Fisher) at UT Southwestern using SerialEM⁹³. The microscope has a Gatan K3 camera and a Bio-Quantum energy filter (20 eV). The total exposure was ~44 e⁻/Ų and the defocus range was set to −0.8 μm to −2.2 μm. The number of micrographs for each data set ranged between 3,800 and 5,300 (Table 1).

All datasets were processed using RELION 3.1⁸⁷. The images were motion corrected, dose-weighted, and 2x Fourier binned (pixel size = 1.079 Å) using MotionCor2⁹⁴. GCTF⁹² was then used to estimate the contrast transfer function (CTF) parameters and defocus values. Particles were picked using crYOLO¹⁰¹, reimported to RELION, and binned two times. The particles were then subjected to two rounds of 2D classification to remove junk particles. In 2D classification, the resolution was limited to 16 Å during the first round and 6 Å for the second round of 2D classification, and ignoring CTFs until first peak was set to true. A subset of the particles after 2D classification were used to generate an initial model without symmetry imposed. 3D classification was then run with 8 classes, no symmetry imposed, resolution limited to 6 Å, and C1 symmetry. One C5 symmetrical class containing ~50% of the remaining particles was selected and refined before re-extraction to a pixel size of 1.079 Å. All subsequent processing steps imposed C5 symmetry. The particles were aligned using 3D refinement followed by per-particle CTF refinement and beam tilt estimation¹¹³ before a second round of 3D refinement. These particles were subjected to particle polishing¹¹⁴ and a third 3D refinement. A final round of CTF refinement and beam tilt estimation was performed and refined again. If necessary, the remaining particles underwent a second round of 3D classification with C5 symmetry using small angular sampling and local angular searches (0.5°) with 8 classes (Figure S1). Classes displaying strong density for the TMD were selected, refined, and sharpened yielding overall resolutions between 2.3 Å and 2.6 Å (Figure S2). The density for NS1738 and surrounding residues was poor compared to the other complexes. Specifically for the α7-epi/NS1738 complex, an additional round of focused 3D classification¹¹⁵ on the transmembrane domain and the lower third of the extracellular domain was performed. One of the eight classes containing 80% of remaining particles was selected. Further focused classification efforts to improve density in this region were unsuccessful. Structural biology software was compiled by SBGrid¹¹⁶.

Time-resolved data processing

6350 movies for α7-epi/PNU time-resolved sample were collected on a Titan Krios (Thermo Fisher) at the Pacific Northwester Center for Cryo-EM (PNCC), equipped with a Falcon 3 detector and a Bio-Quantum energy filter in super resolution mode. The total exposure was 40 e⁻/Ų and the defocus range was set to −0.5 μm to −2.0 μm.

The movies were imported to CryoSPARC v4⁸⁸ and motion corrected and dose weighted using Patch Motion Correction. CTF estimation was then performed using the Patch CTF job. Particles were initially picked via blob picker and subjected to 2D classification to generate templates for template-based picking. Template picking resulted in ~2 million particles that were extracted with a box size of 336 and Fourier cropped to 168. The particle stack was cleaned through two rounds of 2D classification followed by two rounds of 3D classification (ab initio coupled with heterogeneous refinement, 6 volumes per run, 4 good volumes and 2 junk volumes served as inputs). The remaining 490k particles were aligned using Non-Uniform (NU) refinement¹¹⁷, with C5 symmetry imposed and dynamic masking and windowing turned off¹¹⁸. This resulted in a consensus refinement with an overall resolution of 2.2 Å. To sort through the heterogeneity in the time-resolved sample preparation, 3D variability analysis¹¹⁹ was performed with a mask around the entire receptor (masking out most of the detergent belt). The resolution was filtered at 3.0 Å and 3 modes were solved. Each mode was analyzed using the 3DVA display job in simple mode and the particles were subsequently clustered (20 total clusters) on mode 0 (a mode appearing to describe a transition between a resting-like state and a desensitized-intermediate-like state). Two clusters contained a symmetric-looking resting-like conformation, but with some heterogeneity in the neurotransmitter binding site. These clusters were combined and refined without any symmetry imposed using NU refinement. After C5 symmetry was confirmed, they were refined with C5 symmetry imposed yielding a reconstruction at an overall resolution of 2.5 Å. The remaining clusters were analyzed and those showing strong density for the TMD were selected and combined for further processing. The 350k remaining particles were subjected to local refinement with a mask around the TMD and the lower part of the ECD. Then, another round of 3DVA was performed on the remaining particles with a mask around the TMD. The filter resolution was set at 3.0 Å and 4 modes were solved (multiple filter resolutions and numbers of modes, up to 12, were tried and analyzed). Each mode was inspected using a simple 3DVA display job and then the particles were separated into 8 clusters based on modes 0, 1, and 2. Three clusters appeared symmetric and strongly resembled the desensitized intermediate seen in the equilibrium sample. These clusters were joined and refined using NU refinement without symmetry imposed. Upon confirmation of C5 symmetry, they were refined in C5 symmetry using NU refinement with dynamic masking and windowing turned off and refined to an overall resolution of 2.2 Å. The remaining 5 clusters showed weaker density in the TMD consistent with the presence of local movement. These clusters were subjected to NU refinement, which yielded strong density for the ECD, but weak density in the TMD. Local refinement (using the same mask as the one used for the second 3DVA job and rotation and shift priors set to 5 degrees and 2 Å, respectively) on the same clusters was performed to improve the TMD density. This did however result in much weaker density for the ICD compared to the resting and desensitized states (Figure S5C). Composite maps were generated by aligning the NU refinement and local refinement outputs in UCSF Chimera⁹⁸ and using the vop maximum command. Density modification in Phenix¹²⁰ was used to sharpen the symmetric resting and desensitized maps. Density modification was tested for the asymmetric intermediates as well, but did not improve map quality.

A similar processing workflow was performed using the α7-epi/PNU equilibrium data and found no evidence for a resting-like state or meaningful asymmetry within the pore. Additionally, to further validate the presence of L9ʹ asymmetry, and confirm its presence was not simply due to the processing workflow, focused classification without image alignment was performed in RELION^87,121 and revealed a resting-like class, a desensitized-like class, and asymmetric classes.

Model building, refinement, and validation

PDB:7KOX³³ was used as the starting model for epibatidine alone, type I and type II PAM complexes. α7-epi/TQS was used as the starting model for α7-GAT107 and α7-epi/GAT107. α7-epi/PNU (equilibrium) was used as the starting model for all time-resolved complexes besides the resting-like state, where α7-GAT107 was used. The starting model was manually aligned and fit into the experimental density in UCSF Chimera using the Fit in Map tool⁹⁸. This was followed by iterative cycles of manual building in Coot⁹⁵ and global real space refinement in Phenix⁹⁶ with secondary structure restraints and Ramachandran restraints turned on. The final refinement iterations were performed without secondary structure restraints. Model geometry and clash scores were checked using Molprobity¹⁰⁰. Ligand and lipid restraint files were generated using the Grade Web Server (https://grade.globalphasing.org/) with default settings. Figures were made using UCSF Chimera⁹⁸, UCSF ChimeraX 1.5⁹⁹, PyMOL (https://pymol.org/2/), and CHAP⁹⁷.

In the α7-GAT107 structure, there is a weak tube-like density found near the neurotransmitter binding site. This site is sticky with density found in related receptors under conditions lacking a traditional agonist¹²². Additionally, mutations within the neurotransmitter binding site do not alter the allosteric agonist activity of GAT107^73,123. Therefore, this density more likely represents a detergent or lipid molecule weakly occupying that site and was left unmodeled.

Two electrode voltage clamp electrophysiology

WT α7 and nAChO were subcloned into the pGH19 vector. Mutations were introduced by site directed mutagenesis. cRNA was synthesized using the MEGAscript kit (Austin, TX). Xenopus laevis lobes were purchased from Ecocyte (Austin, TX). The oocytes were treated with 1–2 mg/mL of Collagenase type I (Gibco) for 1–1.5 hours in Barth’s solution (pH 7.4) containing (in mM): 88 NaCl, 1 KCl, 2.4 NaHCO₃, 0.82 MgSO₄, 0.33 Ca(NO₃)₂, 0.68 CaCl₂, 10 HEPES. The oocytes were subjected to hypertonic shock in 130mM KH₂PO₄ for 5–10 mins and then washed with ND96 solution (pH 7.4) containing (in mM): 96 NaCl, 2 KCl, 1 MgCl₂, 1.8 CaCl₂, 5 HEPES. Stage V-VI oocytes were isolated and stored in ND96 solution at 16°C before injection. 40 nL containing 2–4 ng of cRNA (1:1 ratio of α7:nAChO) was injected. Recordings were carried out 1–3 days after injection and conducted at room temperature. Ligand stocks were made in DMSO (PAMs) or H₂O (ACh and epibatidine) and diluted in ND96 solution. Current and voltage electrodes were filled with 1M KCl and showed resistances of 0.5–1.2 MΩ. The holding potential was set at −60 mV. Data were collected using an Axoclamp 900A amplifier (Molecular Devices) and digitized using a Digidata 1550B (Molecular Devices) digitizer. Each recording consisted of two, 10-second applications of ACh (100 μM) or Epi (100 nM) separated by an 80 second wash. PAM was then pre-applied for 10–20 seconds (NS1738, TQS, and PNU) or 50 seconds (Ivermectin) before co-application of PAM (10 μM-30 μM) + ACh (100 μM) or Epi (100 nM) for 20 seconds. 300 μM ACh was used for the N213A and M253A mutants due to their lower channel activity. Uninjected oocytes and oocytes injected with only nAChO were tested and showed no currents. We found that variations in the duration of PAM preapplication resulted in variability in agonist response. As expected, longer preapplications resulted in increased potentiation. The relative results for each preapplication time, however, were consistent (data not shown). Occasionally very high levels of potentiation were observed. Two values for IVM M15ʹA (fold potentiation: 16.6) and PNU WT (fold potentiation: 294) were flagged as outliers and were removed from the statistical analyses (Figure 3).

Peak currents were measured using pCLAMP v10. The peak currents from two initial applications of ACh or Epi were measured and averaged. Apparent fold potentiation was calculated by dividing the peak current from the PAM potentiated response by the average ACh or Epi response. Data are expressed as a mean +/− standard error of the mean from at least 3 separate oocytes per condition. Statistical analysis was performed in GraphPad Prism v9.0.0. Significance (set at α=0.05) was determined either by an unpaired two-tailed Student’s t test or one-way analysis of variance (ANOVA). If significant, the wildtype (WT) and mutant responses were compared with Dunnett’s multiple-comparison test. NS P>0.05, * P<0.05, ** P<0.01, *** P<0.001, **** P<0.0001.

Single channel electrophysiology:

Patch clamp recordings from BOSC-23 cells, transfected with cDNAs encoding the wild type human α7 receptor or the T6ʹA and S10ʹA mutants, were obtained in the cell-attached patch configuration at a temperature of 21 °C. Patch pipettes were pulled from glass capillary tubes (No.8250, King Precision Glass) and coated with Sylgard (Dow Corning). The bath and pipette solutions contained (in mM): 142 KCl, 5.4 NaCl, 1.8 CaCl₂, 1.7 MgCl₂, and 10 HEPES, with the pH adjusted to 7.4 by addition of KOH. Acetylcholine chloride (Sigma-Aldrich), with or without PNU-120596 (Tocris Biosciences), was added to the pipette solution. Single channel currents were recorded using an Axopatch 200B patch clamp amplifier (Molecular Devices) with the gain set at 100 mV/pA and the internal Bessel filter at 100 kHz. Before establishing a cell-attached patch, the pipette offset potential was manually set to zero, and after forming a giga-ohm seal, a command voltage was applied to the interior of the patch pipette to establish a membrane potential of −70 mV. The current output from the patch clamp was sampled at intervals of 2 μs using a National Instruments model BNC-2090 A/D converter with a PCI 6111e acquisition card and recorded to the hard disk of a PC computer using the program Acquire (Bruxton).

Detection of single channel openings and closings and analysis of their dwell times was performed using the program TAC 4.3.3 (Bruxton Corp.), which digitally filters the data (Gaussian response, final effective bandwidth 5 kHz), interpolates the digitized points using a cubic spline function, and detects channel opening or closing transitions using the half-amplitude threshold criterion.

Clusters of channel openings, all from the same channel, were identified as a series of closely spaced openings preceded and followed by closed intervals longer than a specified critical time (τ_crit). This duration was taken as the point of intersection between consecutive brief and longer exponential components in the closed time histogram. Values of τ_crit were determined independently for each patch and ranged between 1 and 4 ms. The duration of a given cluster therefore comprises the total open time of a series of openings plus the total closed time of the intervening closings briefer than τ_crit. Cluster dwell time histograms were plotted using a logarithmic abscissa and square root ordinate with a uniformly imposed dead time of 40 μs, and the sum of exponentials was fitted to the data by maximum likelihood using the program TACFit 4.2.0 (Bruxton Corp.).

Molecular dynamics simulations:

To perform all-atom MD simulations the systems were prepared using experimental structures presented in this study as starting coordinates. For each structure, the protonation state of titratable residues at pH 7 was predicted with PROPKA^105,106. The ligands were treated with Marvin version 21.2.0, ChemAxon (https://chemaxon.com/) to predict the protonation state at pH 7 and CHARMM-GUI implementation of General Force Field (CgenFF)^107–109 to generate force field parameters. Using CHARMM-GUI Membrane Builder^107,124 each structure was inserted into a lipid bilayer composed of POPC molecules and solvated in 0.15M NaCl solution with TIP3P water model. The mutant structures were modelled based on WT structure coordinates during the process of system preparation.

The simulations were performed with Gromacs2022 (https://doi.org/10.5281/zenodo.10017699)¹⁰² using the CHARMM36m force field¹⁰³. 10,000 steps of energy minimization with the steepest descent algorithm were followed by 12 ns of equilibration at a constant temperature of 300 K and gradually reducing positional restraints. Berendsen thermostat and barostat¹²⁵ were applied during equilibration. During production runs, temperature was controlled with a v-rescale thermostat¹²⁶ and pressure with a Parrinello-Rahman barostat¹²⁷. For each of the structures we performed 3 independent replicates of 300 ns long simulations without positional restrains to study the dynamic behavior of the structure, and 3 independent replicates of 50 ns long simulations with positional restraints applied to the protein backbone to study channel hydration and permeability of the conformation captured by cryo-EM. For the mutants, we performed 5 independent replicates of 100 ns simulations without positional restraints. The analysis of simulations was performed using MDAnalysis package version 2.2.0¹⁰⁴.

To study the stability of PAM molecules in their sites of binding we computed the root-mean-square deviation (RMSD) of each ligand. In this analysis, the trajectory was aligned to the protein structure such that the RMSD value of a ligand reflects not only the internal conformational changes of the molecule but also its position relative to the site of binding. For comparison, the previously published model of α7-PNU complex (7EKT)³⁴ was explored by running 3 simulation replicates of 100 ns using the protocol described above and the RMSD of the ligand from the initial binding mode analyzed.

To evaluate the stability and strength of interactions established by PAM molecules during the simulation, we first identified the residues that have < 4.5 Å distance to the ligand in the majority of the simulation frames. Then, the short-range interaction energy between the ligand and these residues was computed for each frame in the trajectory as a sum of Coulombic and Lennard-Jones contributions.

Cluster analysis was carried out to aid modeling of the NS1738 binding mode. For this purpose, we first superimposed the MD trajectory of the five ligand-binding interfaces over one and then clustered the coordinates of NS1738 using the Gromos algorithm¹²⁸ with an RMSD cutoff of 1.2 Å. Cluster centers with >5% population were considered as representative binding modes.

Pore hydration and ion permeability were assessed by monitoring both water and ion coordinates within the ion pore over time. These coordinates projected onto the pore axis display the distribution of water and ions inside the channel.

Analysis of the rotameric state of L9ʹ was performed by monitoring the time series of the χ¹ and χ² torsions of the side chain of L9ʹ along the simulation trajectories. Heatmaps collected on the Janin plot¹²⁹ were then used to evaluate the population of the different rotameric states. The analysis revealed two major rotameric conformations: one corresponding to L9ʹ rotated into the pore lumen (in), and one to L9ʹ rotated-out pointing away from the pore (out).

Quantification and statistical analysis

Statistical analysis for two electrode voltage clamp experiments was performed in GraphPad Prism v9.0.0. Data are presented as mean ± SEM. Statistical significance was assessed by either an unpaired two-tailed Student’s t test or one-way analysis of variance (ANOVA) with Dunnett’s multiple-comparison test. See figure legend of Figures 3A–3D for details. Reported cryo-EM map resolutions were determined using the gold standard Fourier shell correlation (FSC)=0.143 criterion and calculated from half maps in RELION v3.1⁸⁷ (equilibrium data) or cryoSPARC v4⁸⁸ (time-resolved data). FSC curves are shown in Figures S2 and S6 and detailed statistics in Tables S1 and S2. Local resolution cryo-EM maps were calculated in RELION v3.1⁸⁷ (equilibrium data) or cryoSPARC v4⁸⁸ (time-resolved data) and are shown in Figures S2 and S6.

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
DH5α	Thermo Fisher Scientific	Cat# C404003
Chemicals, peptides, and recombinant proteins
n-Dodecyl-β-Maltopyranoside (DDM)	Anatrace	Cat# D310
Glyco-diosgenin (GDN)	Anatrace	Cat# GDN101
Lipofectamine 200	Thermo Fisher Scientific	Cat# 11668500
Puromycin	Sigma	Cat# A1113802
Collagenase Type I	Gibco	Cat# 17100017
Ampicillin	Sigma	Cat# 11593027
Kanamycin	Sigma	Cat# 15160054
Penicillin Streptomycin	Gibco	Cat# 10-378-016
Gentamycin	Gibco	Cat# 15750060
Phenylmethylsulfonyl fluoride (PMSF)	Sigma	Cat# PMSF-RO
Fetal Bovine Serum	EMD Millipore	Cat# TMS-013-B
DMEM Medium	Corning	Catt# 10-013-CV
Freestyle 293 Expression Media	Thermo Fisher Scientific	Cat# 12338-018
NS-1738	TOCRIS	Cat# 2995
Ivermectin	TOCRIS	Cat# 1260
(-)-TQS	This paper	n/a
PNU-120596	TOCRIS	Cat# 2498
GAT107	This paper	n/a
Acetylcholine chloride	Sigma Aldrich	Cat# A6625
Epibatidine	TOCRIS	Cat# 0684
d-Desthiobiotin	Sigma Aldrich	Cat# D1411
Critical commercial assays
MEGAscript T7 transcription Kit	Thermo Fisher	Cat # AM1344
Strep-Tactin superflow high capacity resin	IBA Life Sciences	Cat# 2-1208-500
Sepax SRT-500 5μM 4.6×300mm	Sepax	Cat# 215500-4630
Superose 6 Increase 10/300 GL	GE Healthcare	Cat# 29091596
Deposited data
Coordinates of α7-epibatidine complex	This paper	PDB: 8UT1
Cryo-EM map of α7-epibatidine complex	This paper	EMDB: EMD-42526
Coordinates of α7-epibatidine and ivermectin complex	This paper	PDB: 8UZJ
Cryo-EM map of α7-epibatidine and ivermectin complex	This paper	EMDB: EMD-42841
Coordinates of α7-epibatidine and NS1738 complex	This paper	PDB: 8UTB
Cryo-EM map of α7-epibatidine and NS1738 complex	This paper	EMDB: EMD-42537
Coordinates of α7-epibatidine and PNU-120596 complex	This paper	PDB: 8V82
Cryo-EM map of α7-epibatidine and PNU-120596 complex	This paper	EMDB: EMD-43015
Coordinates of α7-epibatidine and (-)-TQS complex	This paper	PDB: 8V80
Cryo-EM map of α7-epibatidine and (-)-TQS complex	This paper	EMDB: EMD-43012
Coordinates of α7-GAT107 complex	This paper	PDB: 8V86
Cryo-EM map of α7-GAT107 complex	This paper	EMDB: EMD-43025
Coordinates of α7-epibatidine and GAT107 complex	This paper	PDB: 8V86
Cryo-EM map of α7-epibatidine and GAT107 complex	This paper	EMDB: EMD-43030
Coordinates of α7-resting (time resolved)	This paper	PDB: 8V89
Cryo-EM map of α7-resting (time resolved)	This paper	EMDB: EMD-43031
Coordinates of α7-epibatidine and PNU-120596 desensitized intermediate (time resolved)	This paper	PDB: 8V8A
Cryo-EM map of α7-epibatidine and PNU-120596 desensitized intermediate (time resolved)	This paper	EMDB: EMD-43032
Coordinates of α7-epibatidine and PNU-120596 complex asymmetric 1 (time resolved)	This paper	PDB: 8V8C
Cryo-EM map of α7-epibatidine and PNU-120596 complex asymmetric 1 (time resolved)	This paper	EMDB: EMD-43034
Coordinates of α7-epibatidine and PNU-120596 complex asymmetric 2 (time resolved)	This paper	PDB: 8V8D
Cryo-EM map of α7-epibatidine and PNU-120596 complex asymmetric 2 (time resolved)	This paper	EMDB: EMD-43035
Coordinates of α7-epibatidine (desensitized)	33	PDB: 7KOQ
Coordinates of α7-EVP-6124/PNU-120596	34	PDB: 7EKT
Coordinates of α7-epibatidine/PNU-120596 (activated)	33	PDB: 7KOX
Experimental models: Cell lines
HEK293S GnTI−	ATCC	Cat# CRL-3022
Xenopus Laevis ooctyes	Ecocyte Biosciences	https://ecocyte-us.com/
Recombinant DNA
pSBtet-GP-α7-NACHOEM	89	Addgene_60495
pCMV(CAT)T7-SB100	90	Addgene_34879
pcDNA3.1-CHRNA-YFP	91	Addgene_62630
pcDNA3.1-CHRNA7	This paper	n/a
pcDNA3.1-CHRNA7 mutants	This paper	n/a
pGH19-CHRNA7	This paper	n/a
pGH19-CHRNA7 mutants	This paper	n/a
Software and algorithms
pCLAMP v10.7	Molecular Devices	https://www.moleculardevices.com/products/axon-patch-clamp-system/acquisition-and-analysis-software/pclamp-software-suite
GCTF	92	https://www2.mrc-lmb.cam.ac.uk/download/gctf_v1-06-and-examples/
RELION 3.1	87	https://www3.mrc-lmb.cam.ac.uk/relion/index.php/Main_Page
CryoSPARC v4	88	https://cryosparc.com
Prism v9	GraphPad	https://www.graphpad.com/features
Serial EM	93	https://bio3d.colorado.edu/SerialEM/
MotionCor2	94	https://emcore.ucsf.edu/ucsf-software
Coot	95	https://www2.mrc-lmb.cam.ac.uk/personal/pemsley/coot/
Phenix	96	https://phenix-online.org/
CHAP	97	https://www.channotation.org/
Pymol	PyMOL	https://pymol.org/2/
Chimera	98	https://www.cgl.ucsf.edu/chimera/
ChimeraX	99	https://www.cgl.ucsf.edu/chimerax/
MolProbity	100	http://molprobity.biochem.duke.edu/
Cryolo	101	https://cryolo.readthedocs.io/en/stable/
Grade Web Server	n/a	https://grade.globalphasing.org/)
Acquire	Bruxton	https://www.bruxton.com/products.html
TAC 4.3.3	Bruxton	https://www.bruxton.com/products.html
TACFit 4.2.0	Bruxton	https://www.bruxton.com/products.html
GROMACS-2022	102	https://www.gromacs.org/
CHARMM-GUI	103	https://www.charmm-gui.org/
MDAnalysis	104	https://www.mdanalysis.org/
Marvin version 21.2.0	ChemAxon	https://chemaxon.com/)
PROPKA	105,106	https://server.poissonboltzmann.org/
CGenFF	107–109	https://cgenff.silcsbio.com/
Other
Quantifoil Holey Carbon Grids, 1.2/1.3 Cu 200 Mesh	Electron Microscopy Services	Cat# Q3100AR1.3

acHighlights:

• Positive modulators bind to a site between subunits in the transmembrane domain

• Modulator class differences correlate with the stable rotation of a key gating residue

• Time-resolved cryo-EM reveals asymmetric state transitions

• Allosteric agonists trigger a unique gating cycle