AMPA (α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid)-subtype ionotropic glutamate receptors mediate fast excitatory neurotransmission throughout the central nervous system. Gated by the neurotransmitter glutamate, AMPA receptors are critical for synaptic strength, and dysregulation of AMPA receptor-mediated signalling is linked to numerous neurological diseases. Here we use cryo-electron microscopy to solve the structures of AMPA receptor–auxiliary subunit complexes in the apo, antagonist- and agonist-bound states and determine the iris-like mechanism of ion channel opening. The ion channel selectivity filter is formed by the extended portions of the re-entrant M2 loops, while the helical portions of M2 contribute to extensive hydrophobic interfaces between AMPA receptor subunits in the ion channel. We show how the permeation pathway changes upon channel opening and identify conformational changes throughout the entire AMPA receptor that accompany activation and desensitization. Our findings provide a framework for understanding gating across the family of ionotropic glutamate receptors and the role of AMPA receptors in excitatory neurotransmission.
Excitatory neurotransmission is tightly governed by the AMPA subtype of the ionotropic glutamate receptors (iGluRs)1,2. Gated by the neurotransmitter glutamate, AMPA receptors (AMPARs) activate on a sub-millisecond timescale and rapidly depolarize the post-synaptic membrane. AMPAR-mediated neuronal signalling has profound effects on synaptic strength, and many neurological diseases have been directly linked to changes in AMPAR synaptic signalling3,4,5.
AMPARs are tetrameric, two-fold symmetric membrane proteins with a three-layer, ‘Y’-shaped architecture6,7. Extracellularly, the amino-terminal domains (ATDs) form the top of the Y. Below the ATDs are the ligand-binding domains (LBDs) that bind the neurotransmitter glutamate. Two polypeptide stretches (S1 and S2) make up each of the four AMPAR subunit LBDs, which are tethered to the transmembrane domain (TMD) at the base of the Y by flexible linkers. The nearly four-fold symmetric TMD is made up of three membrane-spanning helices (M1, M3 and M4) and a re-entrant pore loop (M2) between helices M1 and M3. The cation-selective ion channel is lined by the M3 helices with their bundle crossing forming a tight gate to regulate permeation.
Since the first, closed-state intact AMPAR structure8, many studies have pursued different receptor conformations using both X-ray crystallography9,10,11,12 and cryo-electron microscopy (cryo-EM)13,14,15,16,17. However, the structure of the ion channel is yet to be resolved to high resolution, and the open-state of the receptor has yet to be seen. This is a fundamental barrier to understanding AMPAR biology and structurally informed design of therapeutic agents that target AMPARs in neuropathological conditions. Here, we use cryo-EM to solve the structure of an AMPAR ion channel at high resolution and resolve the opening of the AMPAR ion channel in response to binding of glutamate.
Ion channel structure
We used cryo-EM to analyse the structure of the homomeric GluA2 flip splice variant AMPAR in complex with the auxiliary subunit germline-specific gene 1-like (GSG1L), which we surmised would stabilize the closed-state structure of the ion channel as it lowers the ion channel open probability compared to the transmembrane AMPAR regulatory protein (TARP)-γ2 or stargazin (STZ)15,16. Indeed, using a covalent fusion construct (Methods) between truncated forms of GluA2 and GSG1L, we solved structures of the digitonin-solubilized receptor complex bound to the antagonist ZK200775 in two states, termed GluA2–GSG1LZK-1 (4.6 Å) and GluA2–GSG1LZK-2 (4.4 Å) (Extended Data Table 1; Extended Data Fig. 1), with local qualities of the maps showing higher resolution details in the ion channel pore (Extended Data Fig. 2).
The GluA2–GSG1LZK-1 complex has a typical Y-shaped GluA2 tetramer in the centre (Fig. 1a), with two GSG1L subunits assembling between protomers A/B and C/D on the periphery of the GluA2 TMD (Fig. 1b). The GluA2 TMD is completely resolved and allows building of the entire ion channel pore-forming region (Fig. 1c, d). The M3 bundle crossing, formed by T617, A621, T625 and M629, occludes cation permeation, as previously observed in the original GluA2 crystal structure8. Below T617, the M3 bundle crossing opens up into a hydrophobic cavity in the middle of the channel pore. Below this cavity is a second constriction formed by the extended regions of the M2 re-entrant loops. Similar to other tetrameric ion channels, this constriction might serve as a lower gate of the channel as well as a selectivity filter18,19,20. At the re-entrant loop tip, the glutamine/arginine (Q/R)-site glutamine residues (Gln586) protrude towards the centre of the ion channel pore, occluding the permeation pathway and forming a lower gate. Their location is consistent with the observation that channels composed of edited GluA2 subunits (Gln586Arg) have reduced Ca2+ permeation and polyamine block, probably due to electrostatic repulsion21,22. The pore loop, under the Q/R-site, appears to be more flexible in our closed-state structures, compared to the rest of the channel, but we expect it to become more ordered upon channel opening to form a selectivity filter along the permeation pathway.
The α-helical portion of M2 is largely hydrophobic (Fig. 1c) and contributes to cross-subunit interfaces within the TMD (Extended Data Fig. 3b). These hydrophobic interactions involve Leu540 and Tyr533 from M1, as well as Trp605 and Phe608 from the neighbouring M3, each from the same subunit as the M2, and Trp606 and Ile600 from M3 of the adjacent subunit. The M2 helix contributes to these hydrophobic interfaces through Leu577, Trp578, Leu581 and Phe584. For a helix that has often evaded structural studies, the M2 helix has quite an extensive network of hydrophobic interactions.
The second closed state structure, GluA2–GSG1LZK-2, has the same ion channel architecture as in GluA2–GSG1LZK-1, although with higher resolution details throughout the TMD (Extended Data Fig. 2). While the TMD structure is the same, the ECD in GluA2–GSG1LZK-2 is markedly different, being shorter by 6 Å (Extended Data Fig. 4). The ATD rotates 19° as a rigid body around the overall two-fold axis of symmetry (Extended Data Fig. 4c, d), and the LBD dimers rotate by 4°, creating a separation between the proximal subunits A and C (Extended Data Fig. 4e, f). An elongated density at the ATD–LBD interfaces inside subunits A and C perfectly matches a digitonin molecule (Extended Data Fig. 4b, inset). Thus, digitonin binding might help to trap GluA2 in a distinct closed state conformation. We hypothesize that the ATD–LBD interface represents a new binding site for hydrophobic molecules with a possible therapeutic importance and may bind to fatty acids that have recently been identified as non-competitive AMPAR inhibitors23,24.
Cryo-EM of activated GluA2–STZ complex
To understand how the native neurotransmitter glutamate (Glu) activates and opens the AMPAR ion channel, we used a fusion construct between GluA2 and STZ (Methods) that promotes GluA2–STZ complex formation and expression while maintaining its functional properties. In contrast to GSG1L, which stabilizes the AMPAR inactive states, STZ favours the open state25,26,27, as signified by increased steady-state current in response to Glu application in whole-cell patch-clamp recordings (Fig. 2a). Additional stabilization of the open state is provided by the positive allosteric modulator cyclothiazide (CTZ)28, which makes AMPAR responses non-desensitizing (Fig. 2a).
We solved a cryo-EM structure of the GluA2–STZ complex in the presence of Glu and CTZ to 4.2 Å resolution (Extended Data Fig. 1), with two-dimensional class averages showing the secondary structure features and diverse orientations (Fig. 2b). The reconstruction (Fig. 2c) shows the entire GluA2–STZ complex in the agonist-bound state, with high-resolution features in the GluA2 TMD core (Extended Data Fig. 2n, o) but lower local resolution for STZ. To achieve unambiguous side chain definition in the TMD, we performed directed refinement (Fig. 2c) that resulted in a 4.0 Å cryo-EM reconstruction (Fig. 2d) and distinct side-chain densities (Extended Data Fig. 5e). The TMD cryo-EM map shows clear differences in the central pore (Fig. 2e) compared to the closed-state GluA2–GSG1L complex (Fig. 1b) and four STZ molecules bound. Also, glutamate and CTZ are easily discernible in the cryo-EM density (Extended Data Fig. 5a, d), suggesting that we trapped the complex in an activated state.
The original, closed-state cryo-EM structures of the GluA2–STZ complex suggested that acidic residues in the STZ β4-TM2 loop mediate electrostatic interactions with positively charged residues in the lower GluA2 LBD and S1-pre-M1 linker to enhance receptor activation17,29. In our active state cryo-EM reconstruction, this region of STZ seems to be disordered. However, based on well-defined adjacent regions of STZ (Extended Data Fig. 5e), the same acidic residues are juxtaposed to the GluA2 basic patch. Perhaps rather than directly interacting in the activated state, the β4-TM2 loop has an important role in the structural transitions necessary for receptor activation.
Our density, and the corresponding GluA2–STZGlu+CTZ model (Fig. 3, Methods) show four STZ molecules bound around the AMPAR core (Fig. 2e, Extended Data Fig. 6c). We used the same fusion design to solve the first structure of the GluA2–STZ complex in the closed state15 but observed only one or two STZ molecules bound to the AMPAR core. Here we solubilized the complex in digitonin instead of dodecyl maltoside (DDM) detergent and observed maximum occupancy of STZ binding sites around the AMPAR core, similar to GluA2–STZ co-expression17. Accordingly, instead of the four-layer particle architecture in DDM15, we now see a clearly discernible three-layer architecture in the two-dimensional (2D) class averages (Extended Data Fig. 6a). Interestingly, the GluA2–GSG1L complex in digitonin, which showed a four-layer architecture when solubilized in DDM16, also showed the three-layer particle architecture (Extended Data Fig. 6b) but with only two, not four, molecules of GSG1L bound to GluA2 core (Fig. 1a, Extended Data Fig. 6d). As the nature of the fusion construct design does not preclude full occupancy of the auxiliary subunit binding sites, we hypothesize that STZ and GSG1L use different stoichiometries to assemble with AMPARs.
Open ion channel pore
As indicated by the markedly different architecture of the AMPAR core in GluA2–STZGlu+CTZ compared to the closed state, the width of the TMD increases by 10 Å from 55 Å in the closed state structures to 65 Å in GluA2–STZGlu+CTZ, and the ion channel pore undergoes widening (compare Figs 1b and 3b).We probed the ion conduction pathway in GluA2–STZGlu+CTZ (Fig. 4a) and found it much wider than in the closed state GluA2–GSG1L structures (Fig. 1d). Measurements of the pore radius (Fig. 4b) confirmed that in contrast to the pore-occluded conformations in the closed GluA2–GSG1LZK-1, GluA2–GSG1LZK-2 and desensitized GluA2–GSG1LQuis16 states, the GluA2–STZGlu+CTZ pore is wide open for ion conductance. The distances between the Cα atoms of diagonal Gln586, Thr617 and Ala621 pairs increased almost uniformly by about 3 Å in GluA2–STZGlu+CTZ (Fig. 4b, c), indicating nearly four-fold symmetric, iris-type opening of the main portion of the pore (Fig. 4d), reminiscent of other tetrameric ion channels19,30,31. In stark contrast, the distance between diagonal Thr625 pairs, which is similar for the A/C and B/D subunit pairs in the closed state (9.5 and 11 Å), increases to 16 Å for the A/C pair and to 32 Å for the B/D pair in GluA2–STZGlu+CTZ. This 6.5 Å versus 21 Å separation upon channel opening indicates that at the level of Thr625, the AMPAR pore loses its pseudo four-fold symmetry and becomes two-fold symmetric, consistent with previous functional experiments32. We also compared our recent molecular dynamics model of the open state33 and the GluA2–STZGlu+CTZ structure. The narrowest portions of the pore in the molecular dynamics model, estimated as Cα distances between the diagonal Gln586 (9.6 Å) and Thr617 (12.6 Å) pairs, are narrower than the corresponding portions of the GluA2–STZGlu+CTZ pore (10.2 Å and 14.1 Å, respectively). Yet, we clearly observed ion conducting events through our molecular dynamics model33, an observation strongly supporting the conducting conformation of the GluA2–STZGlu+CTZ pore. In addition, a clear density in the centre of the pore (Extended Data Fig. 5b, c), probably representing a sodium ion, also suggests that the selectivity filter of the channel is indeed in a conducting state.
The selectivity filter located at and below the Q/R site (Fig. 4a, b), which is not visible in the closed state, is clearly resolved in GluA2–STZGlu+CTZ and is approximately four-fold symmetric, similar to K+ channels30,31. Just below the Q/R site, the backbone carbonyls of Gln587, Gly588 and Cys589 line the selectivity filter, while the side chains of Gln587 and Cys589 point away from the central pore axis. At the bottom of the selectivity filter is Asp590 that creates a negative entry point into the cytoplasm from the channel pore, perhaps having a role in cation versus anion selectivity of AMPAR channels.
To understand the conformational changes that are associated with AMPAR gating, we compared the open-state GluA2–STZGlu+CTZ with the closed-state GluA2–GSG1LZK-1 and the desensitized-state GluA2–2×GSG1LQuis structures (Fig. 5, Supplementary Video 1). Since gating starts with binding of glutamate to an unliganded AMPAR, we first wanted to see whether GluA2–GSG1LZK-1 represents the resting, apo state of the receptor. We solved cryo-EM structures of the digitonin-solubilized GluA2–GSG1L complex in the absence of ligands, which, similar to the ZK200775-bound structures, are represented by two different states. The apo state structures, GluA2–GSG1Lapo-1 and GluA2–GSG1Lapo-2, turned out to be very similar to GluA2–GSG1LZK-1 and GluA2–GSG1LZK-2 (Extended Data Figs 3, 4), superposing on each other with root mean square deviation (r.m.s.d.) values of 0.526 Å and 0.701 Å, respectively. We therefore used the higher resolution GluA2–GSG1LZK-1 structure that is nearly identical to GluA2–GSG1Lapo-1 as an accurate representation of the closed resting state.
In the first approximation, activation, which includes ion channel opening, is represented by the GluA2–GSG1LZK-1 to GluA2–STZGlu+CTZ transition. At the level of a single LBD, glutamate binding causes 26° LBD clamshell closure (Fig. 5a–d). Consistent with studies on isolated LBDs34,35,36, this closure leaves the D1–D1 interface intact and leads to a 12 Å separation of the D2 lobes (Fig. 5e). The D2 separation results in widening of the LBD tetramer and expansion of the LBD gating ring in the middle of the LBD layer (Fig. 5g, h and Extended Data Fig. 7g, h), which translates to pore opening by the LBD–TMD linkers (Fig. 5j, k).
We monitored conformational rearrangements in the S1–M1, M3–S2 and S2–M4 linkers by measuring distances between Cα atoms of the diagonal Lys505, Ser635 and Gly771 pairs. The M3–S2 linker in subunits B and D has the central role in transmitting conformational changes in LBDs to opening of the ion channel, as signified by an 11 Å increase in Ser635 separation and a distinct kink in the M3 helix at Ala618 (Fig. 5j, k and Extended Data Fig. 5e), which turns the upper gate at the bundle crossing away from the central pore axis. The M3 helices in subunits A and C unwind and become one helical turn shorter, but the S635 separation in subunits A and C decreases by 1 Å, suggesting a weaker role in gating of the corresponding linkers. The S1–M1 linkers in subunits B and D make a much more notable contribution to channel opening than the S1–M1 linkers in subunits A and C, with separation between Lys505 increasing by 13 and 5 Å, respectively. Unexpected changes were observed in the S2–M4 linkers in subunits A and C, which included complete unwinding of the pre-M4 helices (Fig. 5j, k) and stretching these linkers towards the centre of the ion channel pore, while reducing the distance between G771 by 5 Å. These changes are in stark contrast to the S2–M4 linkers in subunits B and D, which maintain their relative positioning. The unprecedented conformational changes in the subunits A/C S2–M4 linkers that occur during channel opening support the importance of the M4 segments for iGluR function37,38,39 and suggest a possible contribution of these linkers to the open pore extracellular vestibule.
We next mapped the changes between GluA2–STZGlu+CTZ and GluA2–2×GSG1LQuis16 to assess conformational rearrangements during desensitization. Upon desensitization, individual LBDs maintain their closed clamshell conformation (Fig. 5a–d), while the LBD dimers undergo marked rearrangements (Fig. 5f), signified by rupture of the D1–D1 interface (increase in the S741 distance by 9 Å) allowing D2 lobes to come closer together (by 15 Å, measured as a change in the S635 distance) and permit channel closure. Rupture of the D1–D1 interface parallels loss of two-fold symmetry by individual LBD dimers, with 14° rotation of the A/C LBDs away from their B/D dimer partners and a more four-fold symmetric LBD tetramer arrangement16 signified by appearance of a cleft between local LBD dimers (Fig. 5i). Correspondingly, the D2 attachments to the LBD–TMD linkers in this new LBD tetramer arrangement acquire positions similar to the closed state, compressing the expanded open-state LBD gating ring (Fig. 5g–i). This results in the LBD–TMD linkers (Fig. 5l) returning to positions (Fig. 5j) that confer a closed, non-conducting ion channel. To return back to the resting state, the receptor must undergo conformational rearrangement from the desensitized to the closed, resting state, in a process termed recovery from desensitization16.
The ATD local dimers maintain their relative positioning between different gating states, resulting in rigid-body movements of the ATD tetramer (Extended Data Fig. 7d–f). During opening, the ATD tetramer rotates by 11° clockwise as viewed extracellularly (Extended Data Fig. 7e). Upon desensitization, the ATD further rotates clockwise by an additional 7° (Extended Data Fig. 7f). To recover from desensitization back to the resting state, the ATD must then rotate anticlockwise by 18°. Maintenance of the ATD tetrameric arrangement in different gating states29,40 emphasizes importance of the ATDs for iGluR assembly41,42,43,44,45, while enormous rigid body rotations provide a way to fine tune synaptic transmission via altering cross-synaptic interactions46.
The structures of AMPAR complexes allow better understanding of the iGluR gating mechanism (Extended Data Fig. 8). Channel opening in GluA2–STZGlu+CTZ is accompanied by the outward flipping of Gln586, which occludes the channel in the closed state, away from the central pore axis. Therefore, the Q/R site at the tip of M2 seems to function as a lower gate. We propose that individual AMPAR subunit contributions to the lower gate result in the sub-conductance states observed in single-channel AMPAR studies47,48. AMPARs occupy their highest conductance state in the presence of glutamate49,50 and STZ26. We expect that the highest conductance state results from all subunits being engaged by glutamate51,52 when channel occlusion is relieved at both gates. STZ also promotes the higher conductance states by stabilizing the open conformation of the lower gate through interaction with the M2 helices. We therefore believe that our GluA2–STZGlu+CTZ structure represents the maximum-conductance state of the channel.
In addition, the prominent placement of Gln586 of the Q/R-site in the ion channel pore explains why AMPARs containing the R-edited GluA2 subunits show markedly reduced ion channel conductance53, probably through adding a bulky, charged residue into the permeation pathway. At the level of glutamate receptors, this demonstrates how compartment-specific mRNA editing alters ion channel conductance and AMPAR-mediated excitatory neurotransmission2. Importantly, identification of the earlier-predicted32 M3 kink at Ala618 in subunits B/D upon channel opening suggests a wider role for Ala618 across gating in the iGluR family, as it is part of the highly conserved SYTANLAAF motif20. Also in this motif is the Lurcher mutation Ala622Thr, which notably changes function across iGluR subtypes, resulting in constitutively activated receptors and markedly different gating properties54. The proximity of this site to Ala618 suggests that the Lurcher mutation may change iGluR gating through altering the properties of M3 kinking upon channel opening. Better understanding of AMPAR gating and availability of the high-resolution open-state structure will help to overcome barriers in the design of therapeutic agents and the treatment of neurodegenerative diseases.
No statistical methods were used to predetermine sample size. The experiments were not randomized, and investigators were not blinded to allocation during experiments and outcome assessment.
The fusion constructs used for cryo-EM were prepared by introducing a single GT linker between rat GluA2flip construct described in literature11 and mouse GSG1L truncated after Phe23716 or STZ truncated after Leu20715, in the pEG BacMam vector55, with a C-terminal enhanced green fluorescent protein (eGFP) for fluorescence-detection size-exclusion chromatography (FSEC)56 profiling and monitoring during expression, in addition to a streptavidin affinity tag for purification purposes. The amino acids Thr-Gly-Gly were included between the C terminus of GSG1L or STZ and a thrombin cleavage site (Leu-Val-Pro-Arg-Gly-Ser) before the C-terminal eGFP and affinity tag.
DNA encoding wild-type GluA2 or GluA2–STZ was introduced into a plasmid for expression in eukaryotic cells that was engineered to produce green fluorescent protein via a downstream internal ribosome entry site11. Human embryonic kidney 293 (HEK293) cells (ATCC CRL-1573) grown on glass coverslips in 35 mm dishes were transiently transfected with 1–5 μg of plasmid DNA using Lipofectamine 2000 Reagent (Invitrogen). Recordings were made 24–96 h after transfection at room temperature. Currents from whole cells, typically held at a −60 mV potential, were recorded using Axopatch 200B amplifier (Molecular Devices, LLC), filtered at 5 kHz and digitized at 10 kHz using low-noise data acquisition system Digidata 1440A and pCLAMP software (Molecular Devices, LLC). The external solution contained (in mM): 140 NaCl, 2.4 KCl, 4 CaCl2, 4 MgCl2, 10 HEPES pH 7.3 and 10 glucose; 7 mM NaCl was added to the extracellular activating solution containing 3 mM l-glutamate (Glu). The internal solution contained (in mM): 150 CsF, 10 NaCl, 10 EGTA, 20 HEPES pH 7.3. Rapid solution exchange was achieved with a two-barrel theta glass pipette controlled by a piezoelectric translator. Typical 10–90% rise times were 200–300 μs, as measured from junction potentials at the open tip of the patch pipette after recordings. Data analysis was performed using the computer program Origin 9.1.0 (OriginLab Corp.).
Expression and purification
The GluA2–GSG1L and GluA2–STZ bacmid and baculovirus were made using standard methods55, and the following methods were applied to the expression and purification of both constructs. P2 virus, produced from Sf9 cells (GIBCO 12659017), was added to HEK293S GnTI− cells (ATCC CRL-3022) at 37 °C and 5% CO2. At 12 h after infection, 10 mM sodium butyrate was added the temperature was changed to 30 °C. At 72 h after infection, cells were collected by low-speed centrifugation (5,500g, 15 min). Cells were then washed with PBS pH 8.0 and pelleted with low speed centrifugation (5,500g, 15 min). The washed cells were lysed with a Misonix sonicator (18 × 10 s, power level 7) in 150 mM NaCl, 20 mM Tris-HCl (pH 8.0) containing protease inhibitors (0.8 μM aprotinin, 2 μg ml−1 leupeptin, 2 μM pepstatin A and 1 mM phenylmethysulfonyl fluoride); 50 ml was used per 1 l HEK293 cells. Next, the lysate was clarified after centrifugation (9,900g, 15 min), and the membranes were collected by ultracentrifugation (186,000g, 40 min). The membranes were collected and mechanically homogenized, then solubilized for 2 h in 150 mM NaCl, 20 mM Tris-HCl pH 8.0, 1% digitonin (Cayman Chemical Company, 14952). Insoluble material was removed by ultracentrifugation (186,000g, 40 min). Streptavidin-linked resin was added to the soluble material (1 ml resin per 1 l cells), and left to bind for 10–14 h. The resin was washed with 10 column volumes of 150 mM NaCl, 20 mM Tris-HCl pH 8.0, 0.1% digitonin (0.05% digitonin was used in the GluA2–STZ purification). Elution was done with the same buffer but supplemented with 2.5 mM d-desthiobiotin. The sample was then concentrated and digested with thrombin (1:200 mass ratio of thrombin to eluted protein) for 2 h at 22 °C. The sample was then loaded onto a size-exclusion chromatography column (Superose-6) equilibrated with 150 mM NaCl, 20 mM Tris-HCl pH 8.0, 0.1% digitonin (0.05% digitonin was used in the GluA2–STZ purification). The peak fractions were pooled for cryo-EM specimen preparation, and concentrated to 5 mg ml−1 (4 mg ml−1 for GluA2–STZ). All steps, unless otherwise noted, were performed at 4 °C.
HEK293 cells were used for patch-clamp electrophysiology. Cells were obtained from ATCC and were not further authenticated. Cells were grown and maintained using standard protocols at 37 °C with 5% CO2, in DMEM media (Corning) supplemented with 10% heat inactivated fetal bovine serum (Crystalgen). Before recordings, cells were plated on glass coverslips in 35 mm dishes and transiently transfected with 1–5 mg of plasmid DNA using Lipofectamine 2000 Reagent (Invitrogen). Cells were passed twice a week and were used until the 30th passage.
HEK293S GnTI− cells were used to express protein for cryo-EM. Cells were obtained from ATCC and were not further authenticated. Cells were grown and maintained at 37 °C and 5% CO2 using standard protocols, in Freestyle 293 media (GIBCO) supplemented with 2% heat inactivated fetal bovine serum (Crystalgen). Cells were passed twice a week and were used until the 30th passage.
Sf9 insect cells were used to make baculovirus for infecting HEK293S GnTI− cells. Cells were obtained from GIBCO and were not further authenticated. Cells were grown and maintained at 27 °C using standard protocols, in Sf-900 III SFM (GIBCO). Cells were passed twice a week and were used until the 30th passage. All cell lines were tested for mycoplasma contamination by the providers (mycoplasma test negative) and have not been retested again.
Cryo-EM sample preparation and data collection
C-flat (Protochips, Inc., Morrisville, NC) CF-1.2/1.3-2Au 200 mesh holey carbon grids were coated with gold using an Edwards Auto 306 evaporator. Carbon was removed using Ar/O2 (6 min, 50 W, 35.0 sccm Ar, 11.5 sccm O2) plasma treatment with a Gatan Solarus 950 advanced plasma system, making Au 1.2/1.3 holey on Au mesh grids57. To make the surface hydrophilic before sample application, the grid surface was plasma-treated with H2O2 (20 s, 10 W, 6.4 sccm H2, 27.5 sccm O2). Frozen-hydrated grids were prepared using a Vitrobot Mark IV (FEI). 3 μl of sample (5 mg ml−1 GluA2–GSG1L with or without 0.3 mM antagonist ZK200775) was applied to the plasma-treated grids using 3.0 s blotting time, 3 blot force, 30.0 s wait time and 100% humidity at 22 °C. For the open state, 4 mg ml−1 GluA2–STZ was incubated for 15 min with 100 μM CTZ, then 3 μl of this was quickly added and pipetted up-and-down in a mixture of 0.5 μl 700 mM glutamate (final concentration 100 mM). 3 μl of this mixture was added to the grid, blotted as described above.
The GluA2–GSG1L data (apo/ZK) were collected on a Tecnai F30 Polara (FEI) microscope operating at 300 kV, using Leginon58 with a Gatan K2 Summit electron detection (DED) camera (Gatan) in counting mode with a pixel size of 0.98 Å. Data were collected across 40 frames (0.2 s per frame), with a dose rate of 8 e− pixel−1 s−1 (total dose of approximately 67 e− Å−2), within defocus range −1.5 μm to −3.5 μm, Cs 2.26 mm. The GluA2–STZ data were collected on a Titan Krios microscope (FEI) operating at 300 kV, using Leginon with a post-GIF quantum energy filter (20 eV slit) and Gatan K2 Summit direct camera DED camera in counting mode operating at a pixel size of 1.08 Å pixel−1. A dose rate of 8 e− pixel−1 s−1 (total dose of approximately 55 e− Å2) was used across 40 frames (0.2 ms per frame), within the defocus range −1.0 μm to −3.0 μm.
All frame alignment and dose-weighting was carried out in MotionCor259. CTF correction, with gCTF60, was performed on non-dose-weighted micrographs, with all further processing on dose-weighted micrographs using RELION 2.061. From 2,938 micrographs for GluA2–GSG1L bound to ZK200775, 541,261 particles were picked. The data were binned to 3.92 Å pixel−1 and subject to 2D classification, which resulted in cleaning the particle pool to 513,406 particles. These particles were then subject to 3D classification without symmetry into ten classes and the GluA2–2×GSG1LZK model16 low-pass filtered to 40 Å as a reference. Three classes, with two showing different ECD conformations (that is, state 1 and state 2) and two auxiliary subunits bound were chosen for additional 3D sub-classification. This resulted in a group of 234,426 particles, which were unbinned to a pixel size of 1.96, and subject to classification with the GluA2–2×GSG1LZK model (40 Å filter) as the reference. This resulted in two well-structured groups of particles: GluA2–GSG1LZK-1 (26,971 particles) and GluA2–GSG1LZK-2 (41,926 particles). The particles were unbinned to the original 0.98 Å pixel−1 size, and refined with C2-symmetry. Of note is that we did not observe any stoichiometric heterogeneity, as was observed in cryo-EM performed on the GluA2–GSG1L complex solubilized in DDM16.
For the GluA2–GSG1L collection in the absence of ligand, 242,078 particles were automatically picked from 2,593 micrographs using six 2D classes as references (generated from manual picking as described above). The particle images were binned to 3.92 Å pixel−1, and 235,543 particles remained following 2D classification. Next, 3D classification was carried out without symmetry into ten classes with the 40 Å low-pass filtered GluA2–GSG1LZK-1 map. Six classes, numerating 115,120 particles showed structural details that warranted further image processing. The particles were unbinned to 0.98 Å pixel−1 and classified in 3D into 10 classes with the GluA2–GSG1LZK-1. Resulting were two prominent classes, GluA2–GSG1Lapo-1 (20,392 particles) and GluA2–GSG1Lapo-2 (18,926 particles), which were refined with C2-symmetry, and resemble the two states observed for GluA2–GSG1L in the presence of ZK200775 (Extended Data Figs 3, 4).
A total of 4,116 micrographs were collected for GluA2–STZ, and 595,889 particles were automatically picked, with manually picked particles used to generate reference classes as described above. The particles were binned twice to 2.16 Å pixel−1 and subject to 2D classification, which resulted in 581,495 particles being subject to further image processing. The original GluA2–2xSTZ antagonist-bound map15, low-pass filtered to 40 Å, was used as a reference model for 3D classification into ten classes without symmetry imposed. 3D classification resulted in 278,454 particles, from four classes, being chosen for further image processing, without observation of stoichiometric heterogeneity as in the DDM-solubilized complex15, and the particles were unbinned to 1.08 Å pixel−1 and classified in 3D to ten classes. Particles from three classes, numerating 69,207 particles and showing structurally similar features, were chosen for refinement. Initial refinement, with C2-symmetry, resulted in a 4.2 Å map, which showed side-chain features in the GluA2 core but was lacking details for STZ. To align particles better according to the GluA2 TMD and STZ, we generated a soft spherical mask around the detergent micelle (which encompassed all of the GluA2 and STZ TMDs and STZ extracellular head). This resulted in an overall 4.0 Å map, with distinct side-chain features throughout the TMD for model building of the GluA2 and STZ TMDs. We did not observe a digitonin-bound state of GluA2–STZGlu+CTZ, as we did for GluA2–GSG1L in the resting states (GluA2NGSG1LZK-2, GluA2–GSG1Lapo-2).
The resolution for all reconstructions was estimated using the Fourier shell correlation (FSC) = 0.143 criterion62 between independent half maps on corrected FSC curves in which the influences of the mask were removed63 (Extended Data Fig. 1). All maps were post-processed using a softmask in RELION, and B-factors for map sharpening were automatically estimated (Extended Data Table 1). All visualization of electron microscopy densities was done in UCSF Chimera64. Local resolution for each map (Extended Data Fig. 2) was calculated with unfiltered half maps using ResMap65.
To build the state 1 ZK200775-bound and GluA2–GSG1L models, we individually isolated the ATD and LBD dimers, as well as the TMD, from the GluA2–2×GSG1LZK complex structure16 using rigid-body fitting in COOT66. Side chains in the TMD were defined based on local resolution, as was the structure of the M2 helix and pore loop. The resulting model was refined against an unfiltered half map (work) in real space with constraints using PHENIX67. The refined model was tested for overfitting (Extended Data Fig. 2) by shifting the coordinates with shake in PHENIX and building a density in EMAN268 from the shaken model. FSC values were calculated between the densities from the shaken model, the map used in PHENIX refinement (work), second half map (free) and unfiltered sum map, using EMAN2. The resulting GluA2–GSG1LZK-1 model was used as a basis to build GluA2–GSG1Lapo-1. Correspondingly, the ATD and LBD local dimers were moved as rigid bodies to define the state 2 models (GluA2–GSG1LZK-2 and GluA2–GSG1Lapo-2), and features were adjusted manually according to high resolution features in GluA2–GSG1LZK-2. Secondary structure features of the M3 pore were further defined in the GluA2–GSG1LZK-2 M3 density. The resulting models were also real space refined in PHENIX and tested for overfitting (Extended Data Fig. 2). State 2 models were refined in the absence of digitonin.
For GluA2–STZGlu+CTZ, we rigid-body fit the high-resolution structures of the ATD45 and glutamate/CTZ-bound LBD35. We individually rigid body fit helices from the TMD of our GluA2–GSG1LZK-1 refined structure, and rigid-body fit STZ subunits from the GluA2–2xSTZZK structure15 in COOT for an initial model. After increasing the resolution in the TMD map, we built STZ and the GluA2 TMD de novo. The resulting models were also real space refined in PHENIX and tested for overfitting (Extended Data Fig. 2), of which there are no signs of over-fit models. Structures were visualized and figures were prepared in Pymol67.
Cryo-EM density maps have been deposited in the Electron Microscopy Data Bank (EMDB) under accession numbers EMD-8819 (GluA2–GSG1LZK-1), EMD-8820 (GluA2–GSG1LLZK-2), EMD-8821 (GluA2–GSG1Lapo-1), EMD-8822 (GluA2–GSG1Lapo-2), and EMD-8823 (GluA2–STZGlu+CTZ, including directed TMD map). Model coordinates have been deposited in the Protein Data Bank (PDB) under accession numbers 5WEK (GluA2–GSG1LZK-1), 5WEL (GluA2–GSG1LLZK-2), 5WEM (GluA2–GSG1Lapo-1), 5WEN (GluA2–GSG1Lapo-2), and 5WEO (GluA2–STZGlu+CTZ). All other data are available from the corresponding author upon reasonable request.
We thank H. Kao for computational support, A. des Georges, I. S. Fernandez, M. Fislage and A. K. Singh for processing advice. E.C.T. is supported by National Institutes of Health (NIH) F31 NS093838. A.I.S. is supported by the NIH (R01 NS083660, R01 CA206573), the Pew Scholar Award in Biomedical Sciences, and the Irma T. Hirschl Career Scientist Award. J.F. is supported by the Howard Hughes Medical Institute and the NIH (R01 GM029169). Cryo-EM data were collected at the Columbia University Medical Center cryo-EM facility and at the Simons Electron Microscopy Center and National Resource for Automated Molecular Microscopy located at the New York Structural Biology Center, supported by grants from the Simons Foundation (349247), NYSTAR, and the NIH National Institute of General Medical Sciences (GM103310).
Extended data figures
Extended data tables
A morph between GluA2 closed, open and desensitized stated represented by the GluA2-GSG1LZK-1, GluA2-STZGlu-CTZ and GluA2-2xGSG1LQuis structures, respectively. The GluA2 subunits A and C are colored purple and B and D green. The agonists Glu are represented by space-filling models. Note, the pre-M4 helices in subunits A and C unfold during channel opening, while the entire TMD becomes wider. The receptor undergoes a corkscrew twist and becomes shorter while progressing from the closed to open to desensitized states.