Image processing and structural analysis

Movie stacks were corrected for beam-induced motion (with twofold binning for super-resolution K3 data) in Relion 3.1 (Zivanov et al., 2018) yielding dose-weighted images with an image pixel size of 1.083 Å for the apo state dataset, and 1.096 Å for CNQX-bound and L-Glu-bound datasets. These images were used for contrast transfer function (CTF) estimation with CTFFIND4.1 (Rohou and Grigorieff, 2015). Reference-free particle auto-picking was done using Laplacian-of-Gaussian tool in Relion. The auto-picked particles were extracted with a box size of 416 pixels, binned to a box size of 64 or 128 pixels, and imported into cryoSPARC (Punjani et al., 2017). One round of 2D classification was performed for CNQX-bound and L-Glu-bound datasets but was not used for the apo dataset. Ab initio 3D reconstruction and several rounds of heterogeneous refinement with C1-symmetry were used for all three datasets. This process removed false positives from auto-picking and isolated interpretable particle sets (Figure 1—figure supplements 4 and ​and6,6, and Figure 4—figure supplement 1).

The resulting particles were re-extracted with a box size of 320 or 416 without binning and imported into cryoSPARC. Particles extracted with a box size of 320 pixels were first re-centered. Ab initio 3D reconstruction and several rounds of heterogeneous refinements were again performed with C1-symmetry to obtain the final set of particles. The final 3D reconstructions were obtained by non-uniform (GluK2/K5_em-apo) or homogeneous refinement (GluK2/K5_em-CNQX and GluK2/K5_em-L-Glu) in cryoSPARC. The global resolutions were calculated using the FSC as 7.5 Å (apo), 5.3 Å (CNQX), and 5.8 Å (L-Glu) (Figure 1—figure supplements 4 and ​and7,7, and Figure 4—figure supplement 2, Table 2). Local resolution was calculated using ResMap (Kucukelbir et al., 2014) and visualized in UCSF ChimeraX (Goddard et al., 2018).

The GluK2/K5_em-CNQX ATD layer and LBD-TMD assembly were independently refined starting with particles from homogeneous refinement of the full-length receptor. For the ATD layer, The LBD-TMD signal was subtracted from the particles in Relion, then the particles were processed with ab initio reconstruction, heterogeneous refinement, non-uniform refinement, and local refinement in cryoSPARC to a resolution of 3.6 Å (Figure 1—figure supplements 6 and ​and7,7, and Table 2). For the LBD-TMD assembly, the particles were subjected to 2D classification, ab initio reconstruction, and several rounds of heterogeneous refinements in cryoSPARC to further isolate a particle subset. A round of non-uniform refinement was performed followed by particle subtraction to remove the ATD layer. The signal-subtracted particles were then processed with 2D classification, ab initio reconstruction, and several rounds of heterogeneous refinements to isolate a well-resolved LBD-TMD class. Non-uniform refinement (C2 symmetry) was used in a final step to refine the assembly to 4.2 Å (Figure 1—figure supplements 6 and ​and7,7, Table 2). C1 symmetry was used throughout unless otherwise noted.

The GluK2/K5_em-L-Glu ATD layer and LBD-TMD assembly were separately refined starting with the particles used for homogeneous refinement of the full-length receptor, and which had been further processed by a round of 2D classification and non-uniform refinement in cryoSPARC. To refine the ATD layer, particles were subjected to signal subtraction to remove the LBD-TMD region, then processed with ab initio reconstruction and several rounds of heterogeneous refinements to isolate a uniform population of ATD particles. Non-uniform refinement (C2 symmetry) followed by local refinement were used to obtain an ATD structure at 3.8 Å resolution (Figure 4—figure supplements 1 and ​and2,2, Table 2). The LBD-TMD assembly was resolved by subtracting ATD signal from the particles, followed by 2D classification, ab initio reconstruction, and several rounds of heterogenous refinements to obtain a uniform particle subset, and then a final round of non-uniform refinement (C2 symmetry) gave a resolution of 4.3 Å (Figure 4—figure supplements 1 and ​and2,2, Table 2). C1 symmetry was used throughout unless otherwise noted.

Additional analysis was performed in an attempt to improve resolutions in the LBD and TMD regions of both GluK2/K5_em-CNQX and GluK2/K5_em-L-Glu, and we first attempted to use masked local refinement of these regions. This strategy did not improve the resolutions likely because of the small masses of the LBD and TMD layers, so we next applied conformational variability analysis in cryoSPARC. This showed no evidence for local conformational variability in either the LBD or TMD regions.