Hippocampal AMPA receptor assemblies and mechanism of allosteric inhibition

Abstract

AMPA-selective glutamate receptors mediate the transduction of signals between the neuronal circuits of the hippocampus¹. The trafficking, localization, kinetics and pharmacology of AMPA receptors are tuned by an ensemble of auxiliary protein subunits, which are integral membrane proteins that associate with the receptor to yield bona fide receptor signalling complexes². Thus far, extensive studies of recombinant AMPA receptor-auxiliary subunit complexes using engineered protein constructs have not been able to faithfully elucidate the molecular architecture of hippocampal AMPA receptor complexes. Here we obtain mouse hippocampal, calcium-impermeable AMPA receptor complexes using immunoaffinity purification and use single-molecule fluorescence and cryo-electron microscopy experiments to elucidate three major AMPA receptor-auxiliary subunit complexes. The GluA1-GluA2, GluA1-GluA2-GluA3 and GluA2-GluA3 receptors are the predominant assemblies, with the auxiliary subunits TARP-γ8 and CNIH2-SynDIG4 non-stochastically positioned at the B'/D' and A'/C' positions, respectively. We further demonstrate how the receptor-TARP-γ8 stoichiometry explains the mechanism of and submaximal inhibition by a clinically relevant, brain-region-specific allosteric inhibitor.

Extended Data Figure 1.. Biochemical characterization and cryo-EM analysis of hp AMPAR complexes.

a. Representative SEC profile of hippocampal AMPAR complexes. Inset shows an SDS-PAGE gel of AMPAR complexes and antibody fragments used for cryo-EM grid preparation, visualized by silver staining. The gel was repeated three times from different batches of purification with similar results. b. Western blot analysis of isolated AMPAR complexes using antibodies against GluA1, GluA2, GluA3, GluA4, PSD95, TARP-γ8 and Cornichon-2. The uncropped blot can be found in Supplementary Fig. 1 and blotting was repeated three times with similar results. c. A representative cryo-EM micrograph of hpAMPAR complexes. The experiments were repeated four times with similar results. d. Selected 2D class averages. Protrusions extending out of the detergent micelle are indicated by arrows, corresponding to the extracellular domain of TARP-γ8. Similar results were obtained from four repeated experiments.

Extended Data Figure 2.. Characterization of monoclonal antibodies 13A8 and E3

a. Octet measurements of the 13A8 mAb binding to TARP-γ8. Concentrations of the 13A8 mAb ranging from 25–200 nM were applied. b-f. FSEC profiles of recombinant GFP-tagged TARP-γ8 (b), TARP-γ2 (c), TARP-γ3 (d), TARP-γ4 (e) and TARP-γ7 (f) with 13A8 mAb (green traces) and without 13A8 mAb (black traces), detecting GFP fluorescence. Only the TARP-γ8 trace is shifted by the 13A8 mAb. g. Octet measurements of the E3 mAb binding to GluA4. h-k. FSEC profiles of recombinant mKalama-tagged GluA1 (h), GFP-tagged GluA2 (i), GFP-tagged GluA3 (j) and GFP-tagged GluA4 (k) with E3 mAb (green traces) and without E3 mAb (black traces), detecting mKalama/GFP fluorescence. Only GluA4 receptors are shifted by the E3 mAb.

Extended Data Figure 3.. A representative flow chart of data processing focused on the whole receptor and ATD layer using image processing strategy #1.

A total of 2,893,667 particles were picked from 46,927 motion-corrected micrographs in cryoSPARC v2.14. Classes showing clear receptor features were kept after several rounds of 2D classification, resulting in retention of 2,893,667 particles. Next, 3D classification with a large sampling degree was performed to further remove junk classes in RELION 3.0. Another around of 3D classification was carried out to sort receptors with the same Fab/scFv combination. Classes with the same ATD labeling and orientation were combined and subjected to ATD focused classification without alignment. For the A1A2A1A2-symmetric subtype, the ATD layer was classified into eight classes, of which one class occupying the largest population (55%) had the least well-resolved secondary structure features. Another round of ATD focused classification was performed on this class, producing a subtype with one unlabeled subunit, denoted as A1A2AXA2. Three remaining classes showing the most well-defined secondary structure features were selected for final refinement with C2 symmetry, producing a map at a resolution of 4 Å. ATD-focused refinement with C2 symmetry was carried out to improve map density in the ATD, yielding an ATD-A1A2A1A2-symmetric map at a resolution of 3.4 Å.

Extended Data Figure 4.. Data processing workflow to determine AMPAR subtypes using image processing strategy #2.

Motion-corrected micrographs were first curated based on ice thickness, motion correction, CTF fit and astigmatism. Template-based picking was used to autopick 6,002,517 particles in cryoSPARC v2.14. To remove junk particles and false positives, multiple rounds of 2D and 3D classification were performed, selecting only classes that showed discernible receptor features, resulting in a particle stack of 1,844,956 particles. To sort receptors based on subtype (AMPAR subunit stoichiometry and tilting), multiple rounds of 3D classification were performed without symmetry imposed or masking. Particles from classes showing clear labeling with antibodies were grouped into distinct subtypes. Each of the AMPAR subtypes were refined separately. This strategy elucidated three different heteromeric AMPAR subtypes comprised of both symmetric and asymmetric conformations.

Extended Data Figure 5.. 3D reconstructions of dimeric-GluA1/A2 and dimeric-GluA2/A3 complexes.

a, c, e, g, i. Local resolution estimates of the entire GluA1/A2-symmetric map (a), ATD-layer of the GluA1/A2-symmetric map (c), entire GluA1/A2-asymmetric map (e), the ATD-layer of GluA1/A2-asymmetric map (g) and the LBD-TMD layers of the GluA1/A2 map (i). d, h, k. FSC curves before and after masking and between the model and the final maps of ATD-layer of the GluA1/A2-symmetric (d), ATD-layer of the GluA1/A2-asymmetric (h) and LBD-TMD layers of GluA1/A2 (k). j. Angular distribution of the LBD-TMD layers of the GluA1/A2 map. l, n, p, r, t. Local resolution estimates of the entire GluA2/A3-symmetric map (l), ATD-layer of the GluA2/A3-symmetric map (n), entire GluA2/A3-asymmetric map (p), ATD-layers of the GluA2/A3-asymmetric map (r) and LBD-TMD layers of the GluA2/A3 map (t). b, f, m, o, q, s, v. FSC curves before and after masking maps of whole GluA1/A2-symmetric (b), entire GluA1/A2-asymmetric (f), entire GluA2/A3-symmetric (m), ATD-layer of GluA2/A3-symmetric (o), entire GluA2/A3-asymmetric (q), ATD-layer of the GluA2/A3-asymmetric (s) and the LBD-TMD layers of GluA2/A3 (v). u. Angular distribution of the LBD-TMD layers of the GluA2/A3 map.

Extended Data Figure 6.. 3D reconstructions of trimeric-GluA1/A2/A3 complexes and the LBD-TMD_mix map.

a, c, e, g, i, l, o. Local resolution estimates of the entire GluA1/A2/A3-asymmetric 1 map (a), ATD-layer of the GluA1/A2/A3-asymmetric 1 map (c), entire GluA1/A2/A3-asymmetric 2 map (e), ATD-layer of the GluA1/A2/A3-asymmetric 2 map (g), LBD-TMD layers of GluA1/A2/A3 without symmetry (i), LBD-TMD layers of GluA1/A2/A3 with C2 symmetry imposed (l), and LBD-TMD_mix map (o). b, d, f, h, k, n. FSC curves before and after masking maps of the entire GluA1/A2/A3-asymmetric 1 receptor (b), the ATD-layer of GluA1/A2/A3-asymmetric 1 (d), entire GluA1/A2/A3-asymmetric 1 (f), ATD-layer of the GluA1/A2/A3-asymmetric 2 (h), LBD-TMD layers of GluA1/A2/A3 without symmetry (k) and with C2 symmetry (n). j, m, p. Angular distribution of LBD-TMD layers of GluA1/A2/A3 maps with C1 symmetry (j) or C2 symmetry (m), and the LBD-TMD_mix map (p). q. FSC curves before and after masking and between the model and the final maps of the LBD-TMD_mix map.

Extended Data Figure 7.. Representative TIRF images for native AMPA receptor complexes captured with 15F1 mAb.

Fluorescence detection with (a) anti-GluA1-Alexa488 mAb (αGluA1) and anti-GluA3-Alexa594 mAb (αGluA3), (b) anti-GluA1-Alexa488 mAb, anti-GluA3-Alexa594 mAb, and anti-GluA4-Alexa594 mAb (αGluA4), (c) anti-GluA1-Alexa488 mAb, anti-GluA3-Alexa594 mAb, and anti-TARP-γ8 mAb (for each colocalization experiment), (d) anti-TARP-γ8 Fab-GFP (αTARP-γ8), and (e) anti-SynDIG4-Alexa594 mAb (αSynDIG4) and anti-TARP-γ8-Alexa647 mAb. Scale bar in each image represents 5 μm.

Extended Data Figure 8.. Structures of the dimeric GluA1/A2 receptor, trimeric GluA1/A2/A3 receptor and dimeric GluA2/A3 receptor complexes in symmetric and asymmetric conformations.

a, c. Cryo-EM structures of the GluA1/A2 subtype in symmetric (a) and asymmetric (c) conformations viewed parallel to the membrane. GluA1, GluA2, TARP-γ8, and CNIH2 are grey, red, green and blue, respectively. Antibody fragments 11B8 scFv and 15F1 Fab are pink and cyan, respectively. b. ATD layer analysis of symmetric and asymmetric conformations. ATD model of symmetric state is shown in upper panel, in which center of masses (COMs) of each subunit are indicated by black circles. The lower two panels show the distances (in Ångstrom) and angles determined by COMs of symmetric (left) and asymmetric (right) conformations. d-e. Close contacts between the ATD layer and LBD layer in the asymmetric conformations. Close-up views of the ‘left’ side (d) and ‘right’ side (e) of the ATD-LBD interfaces as indicated in the black and cyan rectangles. f-h. Cryo-EM structures of the trimeric GluA1/A2/A3 subtype in asymmetric conformations with different tilted angles and orientations viewed parallel to the membrane. GluA1, GluA2, GluA3, TARP-γ8 and CNIH2 are colored in grey, red, orange, green and blue, respectively. Antibody fragments 11B8 scFv, 15F1 Fab, 5B2 Fab are in pink, cyan and light yellow colors, respectively. i-j. Zoomed-in views of ATD-LBD interfaces in the asymmetric states (f and h) as indicated in the black and red rectangles. The distances are defined by the Cα atoms of indicated residues. k, l. Cryo-EM structures of the dimeric GluA2/A3 subtype in symmetric (k) and asymmetric (l) conformations viewed parallel to the membrane. m. ATD layer analysis of the symmetric and asymmetric conformations. An ATD model of the symmetric state is shown in the upper panel. The COMs of each subunit are in black circles. The lower two panels show the distances (in Ångstrom) and angles determined by COMs of symmetric (left) and asymmetric (right) conformations. n-o. Close contacts between the ATD layer and LBD layer in the asymmetric conformations. Zoomed-in views of left side (n) and right side (o) of the ATD-LBD interfaces as indicated in the green and cyan rectangles.

Extended Data Figure 9.. Flow chart of data processing for hippocampal AMPAR complexes focused on the LBD-TMD layers.

Particles corresponding to both the symmetric and asymmetric A1A2A1A2 subtype were combined and subjected to LBD-TMD focused 3D classification with alignment in RELION 3.0, resulting in three good classes with continuous transmembrane helical densities. Another round of classification without alignment was carried out for classes 1 and 8. Classes displaying strong density for TMD and auxiliary proteins were combined for refinement in cryoSPARC v2.14, yielding the LBD-TMD_A1/A2 map at a resolution of 3.63 Å.

Extended Data Figure 10.. Conformational differences in the LBD and TMD layers between native and recombinant AMPAR-auxiliary protein complexes.

a. Reference model and orientation of the hippocampal GluA1/A2-TARP-γ8-CNIH2 complex. GluA1, GluA2, TARP-γ8, and CNIH2 are in grey, red, green, and blue, respectively. b-e. Superposition of hippocampal GluA1/A2-TARP-γ8-CNIH2 with recombinant GluA1/A2-TARP-γ8 complexes (PDB code: 6qkc) to show the differences in the LBD (b, d) and TMD (c, e) layers. Recombinant GluA1/A2-TARP-γ8 is in blue. COMs of LBD and TMD layers of each subunit are shown in colored circles. The schematic diagrams illustrate the subunit arrangement differences in distance (Ångstrom) of the LBD (d) and TMD (e) layers of these two complexes. f-i. Superposition of the hippocampal GluA1/A2-TARP-γ8-CNIH2 structure with the recombinant GluA2-CNIH3 complex (PDB code: 6peq) to show the differences in the LBD (f, h) and TMD (g, i) layers. Recombinant GluA2-CNIH3 is in yellow. COMs of LBD and TMD layers of each subunit are shown in colored circles. The schematic diagrams illustrate the subunit arrangement differences in distance (Ångstrom) of the LBD (h) and TMD (i) layers of these two complexes. j. The B/C LBD dimers from the hippocampal GluA1/A2-TARP-γ8-CNIH2 structure and the GluA2-CNIH3 complex (PDB code: 6peq) were superimposed, exhibiting a 3.2 Å shift in the COM (black circles) between the opposing A/D LBD dimers. k. Superposition of the M1, M3, and M4 helices of the hippocampal GluA1/A2-TARP-γ8-CNIH2 structure with the recombinant GluA2-CNIH3 complex (PDB code: 6peq), highlighting the rotation and compression of the GluA2-CNIH3 TMD layer. Equivalent positions of the Cα atoms from the M1 (V538), M3 (I600), and M4 (L805) helices of the GluA2-CNIH3 structure are shifted by 4.5 Å, 5.7 Å, and 4.7 Å, respectively.

Extended Data Figure 11.. Representative densities of the LBD-TMD_A1/A2 or LBD-TMD_mix complexes maps.

a. The S1-M1, M2-pore loop, R/G site, and MPQX from GluA1 are isolated from LBD-TMD_A1/A2, contoured at 0.085. b. S1-M1, M2-pore loop, and R/G site from GluA2 are isolated from LBD-TMD_A1/A2, contoured at 0.085. c. Comparison of the differences by fit Arg and Gln into the GluA2 Q/R site density. d. Four transmembrane helices (TM1-TM4) in TARP-γ8 are isolated from LBD-TMD_mix, contoured at 0.15. e. Four transmembrane helices (TM1-TM4) in CNIH2 are isolated from LBD-TMD_mix, contoured at 0.13.

Extended Data Figure 12.. Electrophysiological recordings of GluA1-TARP-γ8 mutants.

a. Current responses of wild-type GluA1-TARP-γ8 complexes evoked by repeated application of 10 mM glutamate with 10 pulses, each for a duration time of 1s to reach a plateau of the steady-state current. 10 μM JNJ-555511118 was applied before and during glutamate application for 1 s to measure the inhibition of glutamate-induced currents. The lower two insets illustrate the inhibition effect of JNJ-55511118 on the steady-state current by overlaying currents without (the last application) and with JNJ-55511118 at time scale of 500 ms (left) and 20 ms (right). b-h. Representative recordings for the mutations derived from GluA1 (b-e) and TARP-γ8 (f-h) with the same recording condition as wild-type.

Extended Data Figure 13.. LBD-TMD_mix image processing strategy #2

Particles post 2D/3D classification cleanup from strategy #2 were combined into a single stack and refined, and unless otherwise specified, all subsequent processing was performed in cryoSPARC v2.14. Signal subtraction was implemented using the consensus refinement and a soft mask created around the ATD layer and all possible binding sites of the antibodies. Several rounds of 2D classification were used to remove false positives and particles still harboring the ATD layer. This cleaned stack of particles underwent 3D classification (C1 symmetry) resulting in a single class displaying continuous transmembrane density features. Particles from this class were subject to 2D classification to remove a small subset of junk particles. An iterative, sequential, refinement procedure consisting of 1) Homogenous refinement, 2) Non-uniform Refinement, 3) Local CTF Refinement and 4) Non-uniform Refinement, was used to improve the resolution of the stack of 151,141 particles. This procedure was iterated 2x until no resolution improvement was obtained, resulting in a 3.45 Å map. Particles from this map were then subjected to Ab Initio classification permitting removal of junk particles. A new stack of 132,427 particles was then subjected to the previously described 4-step refinement procedure for one iteration, before 3D classification was performed in RELION 3.0 to remove junk particles. This final particle stack was subjected to Non-uniform Refinement in cryoSPARC to obtain the LBD-TMD_mix map at 3.25 Å.

Figure 1.. Cryo-EM and single molecule fluorescence analysis of hpAMPAR complexes.

a-d. Cryo-EM maps of the four resolved complexes, viewed parallel to the membrane. Symmetric and asymmetric conformations within one complex are indicated. GluA1, GluA2, GluA3, GluAX, where ‘X’ represents an unidentified subunit, and potential auxiliary proteins are grey, red, orange, purple and blue/green, respectively. The anti-GluA1 11B8 scFv, the anti-GluA2 15F1 Fab, and the anti-GluA3 5B2 Fab are pink, light blue, and light yellow, respectively. Insets are cartoons showing the subunit arrangement and antibody fragment labeling of the ATD layer. e. The total number of GluA1, GluA3, and colocalized GluA1/GluA3 subunit-containing molecules detected by SiMPull are shown. See Methods for description of control experiments. We observed 37% colocalization of GluA3 spots with GluA1 spots. N=25 images examined over two independent experiments. The inset shows schematic depiction of AMPAR SiMPull. f. Distribution of AMPAR subunits from cryo-EM and SiMPull experiments. The number of each subunit in the single particle cryo-EM dataset was counted based on the presence of an identifying scFv or Fab fragment. Particle fractions were averaged across two cryo-EM datasets obtained with different data processing strategies (see Methods). ‘GluAX-EM’ particles are untagged subunits observed by cryo-EM. The subunit fractions from SiMPull experiments were calculated by probing immobilized hpAMPARs with a fluorescently-labeled, subunit-specific mAb. For SiMPull experiments, n = 120 images examined over two independent experiments. Data are represented as mean values +/− SEM.

Figure 2.. Architecture and subunit arrangement of the LBD-TMD_A1/A2 complex.

a. Three-dimensional reconstruction of LBD-TMD_A1/A2 complexes viewed parallel to the membrane. GluA1, GluA2, TARP-γ8, and CNIH2 are grey, red, green, and blue, respectively. Lipid-like densities are in cyan. b. The structures of LBD-TMD_A1/A2 complexes viewed parallel to the membrane. The JNJ molecule is in sphere representation. c. Interface between TARP-γ8 and receptor. Selected residues along the interface are shown as sticks. Possible hydrogen-bonds are indicated as black dashed lines. d. Interface between CNIH2 and receptor, highlighting key residues. e. Superimposition of the TARP-γ8 interface (c) and CNIH2 interface (d) to show the conformational differences in the A’ and B’ positions, viewed perpendicular to the membrane. For clarity, the solvent accessible surfaces of TARP-γ8 and CNIH2 are shown. f. Observed colocalization of TARP-γ8 with GluA1 (65%) and colocalization of GluA3 with TARP-γ8 (38%) from SiMPull experiments. See Methods for description of control experiments. N= 25 images examined over two independent experiments. g. Representative trace showing two-step photobleaching (blue arrows) of the 13A8 GFP-tagged anti-TARP-γ8 Fab. h. Summary of photobleaching step distribution for the 13A8 GFP-tagged anti-TARP-γ8 Fab. The photobleaching step distribution for anti-TARP-γ8 Fab (black bars) is consistent with a binomial distribution (grey bars) that assumes a dimeric protein and 80% GFP maturation. N = 600 spots were analyzed from three photobleaching movies (200 spots/movie) collected from two independent experiments. Each movie is represented by a blue dot. Data are represented as mean values +/− SEM.

Figure 3.. JNJ binding site and mechanism of inhibition

a-b. JNJ density and binding site viewed from the extracellular side (a) and parallel to the membrane (b). JNJ density is shown as a partially transparent blue surface. Possible hydrogen-bonds are indicated as black dashed lines. c-d. Current responses of GluA1-TARP-γ8 complexes evoked by glutamate and glutamate plus JNJ (cyan trace) at a time scale of 500 ms (c) and 20 ms (d), taken after steady-state responses reached a plateau. e. Box plot showing the extent of JNJ-induced steady-state current reduction from wild type and mutants (n=7 (wild type, A1-F527A/γ8), n=6 (A1-M523A/γ8, A1/γ8-I180A, A1/γ8-F205A) and n=5 (A1-Y519A/γ8, A1-C524A/γ8, A1/γ8-Y206A)). Boxes show the 25th and 75th percentiles, and whiskers down to the minimum and up to maximum values. The horizontal line in each box shows the median value. A one-way analysis of variance with Sidak’s multiple comparison test was used to determine the significance and p values are indicated above the boxes. No adjustment was applied for multiple comparison.

Figure 4.. Elucidation of putative SynDIG4 density and interaction sites

a. The LBD-TMD_mix reconstruction viewed parallel to the membrane. For clarity, only two AMPAR subunits are displayed in each panel. Cryo-EM density for SynDIG4 is shown in purple. Lipids are displayed in cyan. GluA2, TARP-γ8 and CNIH2 are red, green, and blue, respectively. AMPAR subunits in positions A (grey) and C (not shown) are undetermined. b. Observed colocalization of SynDIG4 with TARP-γ8 (61%) from SiMPull experiments. See Methods for description of control experiments. Inset: The dotted line represents the chromatography profile of hpAMPARs incubated with the anti-TARP-γ8 Alexa647-labeled 13A8 mAb. The solid line represents the profile of the hpAMPARs incubated with the Alexa647-labeled 13A8 mAb and the anti-SynDIG4 mAb. A shift in size represents detection of SynDIG4 from the hpAMPARs. c. Putative interaction sites of SynDIG4 with CNIH2 and an undetermined AMPAR subunit(s). View is perpendicular to the membrane. Red model = GluA2. Grey model = an undetermined AMPAR subunit (AX). d. Top down view perpendicular to the membrane displaying the overall stoichiometry and arrangement of the TMD layer. Cryo-EM density for putative SynDIG4 is opaque, density for all other proteins is transparent.