Uncover Inhibitory Synapse Secrets

GlyR-loop stabilizes GephE dimerization within full-length Geph

Given that the full-length structure of gephyrin remains unknown, we carried out single-particle analyses (SPA) of isolated wild-type gephyrin trimers (GephWT). For this, GephWT was overexpressed in E. coli and purified to homogeneity using affinity chromatography combined with analytical scale size-exclusion chromatography (aSEC) (Supplementary Fig. 1a), and then prepared for negative stain EM. Micrographs showed particles of varying sizes (Supplementary Fig. 2a) with a high degree of heterogeneity; however, common architectural principles could be identified after 2D classification (Fig. 1a, c). A subset of class averages, accounting for about 17% of particles, showed an elongated bilobed structure adjacent to a round structure. Based on the molecular dimensions of the GephG (PDB 1JLJ) and GephE (PDB 1JLJ) crystal structures, these shapes likely represent a GephG trimer and GephE dimer, respectively (Fig 1c1). A second subset of class averages, accounting for about 13% of particles, showed a single elongated bilobed structure, likely representing a GephE dimer without a resolved GephG trimer (Fig 1c2). A third subset of class averages showed structures of varying shapes and sizes that could not be confidently assigned to any gephyrin domain (Fig 1c3). Approximately 70% of all particles fell into this heterogeneous group of class averages, suggesting that the majority of gephyrin trimers adopted heterogeneous conformations, consistent with structural flexibility observed in previous SAXS and AFM studies20. The particle subset showing putative GephG and GephE densities in close proximity (Fig 1c1, green) was further processed in 3D (Supplementary Fig. 2b), yielding an intermediate-resolution negative stain EM density map that features two distinct, disconnected densities that were confirmed as GephE and GephG by rigid-body fitting known crystal structures (PDB 2FU3 and PDB 1JLJ) indicating the presence of stable GephE dimers and GephG trimers in our sample (Fig. 1d and Supplementary Fig. 2c). Neither density corresponding to the third copy of GephE nor density that might correspond to GephC could be identified.

We next investigated if the addition of GlyR-loop, as a mimic for receptor ICD binding, affects the architecture of full-length GephWT. For this we added a 3-fold excess of GlyR-loop to GephWT and subsequently isolated the complex by aSEC before staining for negative stain EM (Supplementary Figs. 1a and 3). Micrographs appeared similar to those obtained for the sample without the GlyR-loop (apo-Geph), showing particles of varying sizes (Supplementary Fig. 3a), which, after processing resulted in visually similar 2D class averages as before (Fig. 1e). However, in contrast to apo-Geph, the relative number of particles contributing to the heterogeneous group of class averages was decreased more than two-fold (Fig 1e1-4), implying that the presence of the GlyR-loop reduced heterogeneity by stabilizing Geph in defined conformations. Again, the particle subset showing GephG and GephE in close proximity (Fig 1e1) was further processed in 3D, revealing a structure composed of two distinct densities that was similar to the one obtained for apo-Geph (Fig. 1f and Supplementary Fig. 3b). However, while nominal at the same resolution and only 50% more particles contributed to the reconstruction, the structure was much better defined, particularly in the GephE region (Fig. 1d, f). Still, stain granularity prevented the direct observation of GlyR-loop binding.

Interestingly, 2D classification revealed an additional subset of 2D class averages that appeared to comprise linear assemblies of two or more GephE dimers (Fig 1e3). We also processed this particle subset showing linear assemblies (Fig 1e3) in 3D, resulting in an elongated negative stain EM density that allowed rigid-body fitting of two GephE dimers, which appeared to be linked via their respective SDIIs (Fig. 1g and Supplementary Fig. 3c).

In conclusion, our SPA analysis of full-length Geph revealed high structural heterogeneity, which was decreased following the incubation with GlyR-loop. The presence of GlyR-loop, moreover, triggered the formation of linear assemblies that appear to comprise two or more directly interacting GephE dimers.

Full-length Geph forms flexible filaments at high protein concentrations

To uncover the molecular basis for the linear assemblies of Geph observed in negative stain EM, we investigated full-length GephWT trimers complexed with the GlyR-loop by cryo-EM. Under identical sample preparation conditions as used for negative stain EM, except at a 100-fold increased protein concentration, we obtained micrographs that were densely packed with particles (Supplementary Fig. 4a). Subsequent unsupervised 2D classification resulted in a set of high-quality 2D class averages featuring distinct secondary structure elements (Fig. 2a). While some 2D class averages were centered on the GephE dimer, others were centered on a Z-shaped structural feature connecting two GephE dimers (Fig. 2a and Supplementary Fig. 4b). 2D class averages centered on the GephE dimer showed weak adjacent density, indicating that these GephE dimers are likely incorporated into larger linear assemblies, or filaments, in agreement with our negative stain EM data (Figs. 1e3 and 2a). Therefore, we re-extracted these particles using a box size twice as big as before and repeated the unsupervised 2D classification. The resulting 2D class averages confirmed the presence of slightly curved GephE filaments comprising GephE dimers interconnected by a Z-shaped interface (Fig. 2a and Supplementary Fig. 4c). We additionally observed 2D class averages showing GephE dimers in close proximity to a smaller bilobed density, which could be identified as a side view of the GephG trimer based on in silico 2D back-projections of the GephG crystal structure (Fig. 2a, b). However, subsequent 3D reconstruction of the GephE filament failed due to the crowdedness of these grids and a high degree of preferential orientation of the particles (Supplementary Fig. 4d).

Given that Geph complexed with the GlyR-loop eluted in SEC homogeneously at the expected elution volume of the trimeric Geph-GlyR complex (Supplementary Fig. 1a), we concluded that the observed GephE filaments must have been formed during the cryo-EM sample grid preparation. As the vitrification process itself has been shown to only minimally affect the sample, especially with regard to processes on the time-scale of protein oligomerisation37, we hypothesize that the blotting step might have induced the oligomerisation process, possibly due to a rapid increase in local protein concentration as observed for other proteins38.

Because in our negative staining SPA elongated assemblies of GephE dimers were only observed when Geph was complexed with the GlyR-loop, but not for apo-Geph, we also prepared cryo-EM samples in the absence of the GlyR-loop. Resulting micrographs were of comparable quality, also showing densely packed particles (Supplementary Fig. 5a). Interestingly, despite the absence of the GlyR-loop, data processing revealed 2D class averages that were similar to those observed in the presence of the GlyR-loop, showing GephE filaments alone or in close proximity to a smaller bilobed density (Supplementary Fig. 5b), implicating that potentially the increased concentration might overcome the necessety for GlyR-loop binding observed during negative stain EM.

GephE filaments are connected via an interlocking interface of neighboring SDIIs

In order to confirm that neither GephG nor the C-linker is involved in the formation of the filaments, and aiming to reduce the crowdedness of the sample on the grid, we prepared cryo-EM samples of isolated GephE (residues 327–736) using identical experimental conditions as for full-length Geph before (Supplementary Fig. 6). Again, 2D class averages revealed linear, but flexible assemblies of GephE dimers connected by a Z-shaped interface (Fig. 2c and Supplementary Fig. 6c).

To mitigate the effects of preferred orientation, we acquired an additional dataset at a 25° tilt of the sample stage (Supplementary Fig. 6b). In this dataset, several micrographs showed highly reduced particle crowdedness and clearly illustrated the distribution and proportions of filaments on the micrographs (Supplementary Fig. 6b1). Processing of combined datasets yielded extended particle orientations (Supplementary Fig. 6d), which allowed for an ab-initio 3D reconstruction that revealed a complete GephE dimer with additional density on one side, corresponding to the SDII of an adjacent GephE dimer in the filament (hereafter referred to as SDII‘) (Supplementary Fig. 6e). Further processing yielded a final map at a global resolution of 3.5 Å, showing clear side-chain density in most areas (Supplementary Fig. 6f). However, density for the SDII-SDII‘ interface remained fragmented, which we attributed to conformational flexibility in agreement with the various degrees of curvature observed for the 2D class averages of the filament (Fig. 2c and Supplementary Fig. 6c, d). Using a motion-based deep generative model for continuous heterogeneity39, the flexibility of SDII and the SDII-SDII‘ interface to the adjacent GephE dimer was confirmed (Supplementary Fig. 6g).

In order to improve the resolution in the SDII-SDII‘ interface region, we used in silico-generated 3D structures based on the deformation flow field from the flexibility analysis to initiate a 3D classification. This yielded one class with a poorly resolved SDII-SDII‘ interface (35% of particles) and two classes with improved resolution at the SDII-SDII‘ interface (65% of particles, Supplementary Fig. 6h), of which the latter subset of particles was further refined. Global refinement and subsequent local refinement focussing on the area encompassing the core part of GephE dimer including the SDII-SDII‘ interface yielded a map with continuous density at the SDII-SDII‘ interface and a resolution of approximately 3.0 Å in the core part of the GephE dimer and 3.5 to 4.0 Å in the SDII-SDII‘ interface (Fig. 2d; Supplementary Figs. 6i, j and 7a), allowing us to build an almost complete atomic model into the density map following additional deep-learning based40 map sharpening (Fig. 2e and Supplementary Fig. 6j). In our atomic model chain A comprises all four subdomains of GephE (Fig. 2e). Due to the highly fragmented density of chain B in the SDII region, the respective amino acid residues (Ala370–Pro459) were not build into our atomic model (Fig. 2e). For SDII‘ (chain C) we utilized a predicted structure of SDII as a starting template for building the atomic model41,42.

Overall, our structure of the GephE dimer highly agrees with previous crystal structures, especially in the regions of SDI, SDIII and SDIV4. The here identified SDII-SDII‘ interface features a prominent groove on the SDII side that interlocked with a complementary bulge on the SDII‘ side (Fig. 2e and Supplementary Fig. 7b, c).

Plasticity of the SDII-SDII‘ interface supports GephE filaments flexibility

While our 3D sorting strategy could alleviate the fragmentation of the SDII-SDII‘ interface density, likely caused by hinge-like motions at the two loops connecting SDI and SDII, the lower resolution of this area in our final reconstruction implied the presence of remaining structural flexibility. To further characterize potential heterogeneity of the SDII-SDII‘ interface we re-processed the particles from the initial 3D sorting (Supplementary Figs. 6h and 8). For this, particles that did not show a well-resolved SDII-SDII‘ interface region (“fragmented”), and particles that showed a well-resolved SDII-SDII‘ interface region (“defined”), were individually refined and then subjected to a motion-based deep generative model for continuous heterogeneity in order to visualize the main modes of SDII-SDII‘ interface movement (Supplementary Fig. 8a, b). Notably, the consensus position of the SDII‘ in the structure from the “fragmented” subset was shifted by approximately. 9.8° upward as compared to the previously identified, higher resolved, SDII‘ position (Fig. 3a and Supplementary Fig. 8c). SDII and SDII‘ from the “defined” subset showed a movement of up to 3.9° and 9.5° for SDII and SDII‘, respectively (Fig. 3b and Supplementary Fig. 9a). In contrast, the movement of the interface with the uptilted SDII‘ showed an increased movement of up to 9.9° for SDII and 20.9° for SDII‘ (Fig. 3c and Supplementary Fig. 9b). Potentially, the later “fragmented” particle population (Supplementary Fig. 8a) might represent a less stable state before SDII and SDII‘ fully interlock.

To follow up on the observed mobility of the SDII, we investigated whether the incorporation of GephE into filaments changes the position of the SDII in relation to the other domains of the protein. When comparing SDII positions in the crystal structure of human GephE either without or with bound GlyR-loop (PDB: 2FU3, 2FTS) we found that SDII in our structure is positioned further away from the core part of the protein compared to the GephE crystal structure without bound GlyR-loop (Fig. 3d). This difference was smaller if comparing to the GlyR-loop bound GephE crystal structure, implying a GlyR-loop dependent shift in the SDII positioning might be related to filament formation as it promotes a positioning of the SDII more similar to the positioning observed in GephE filaments (Fig. 3e and Supplementary Fig. 9d). A similar effect was observed in the crystal structures of the GephE plant orthologue, Cnx1E, that shows a shift of SDII towards the filament-promotion angle when the cofactor was bound (Fig. 3f, g and Supplementary Fig. 9e, f)43,44.

A single GlyR-loop is able to occupy both binding sites within one GephE dimer

Despite saturation of the sample with the GlyR-loop, the consensus structure showed only fragmented density in the receptor binding sites (Figs. 1b and 2d). We hypothesized that this might be caused by the GlyR-loop adopting multiple conformations within the binding pocket4,45. In order to better resolve the GlyR-loop density, we carried out a 3D classification of all particles of GephE irrespective their SDII conformation using a mask focusing on the core part of the GephE dimer, including both receptor-binding sites (Supplementary Fig. 6f; Sub-Fig. 10). The subset of particles showing the most continuous density for the GlyR-loop was further processed by local refinement excluding the SDII subdomains (Supplementary Fig. 10a). This yielded a final map that after low-pass filtration to 6 Å showed a continuous density spanning both binding pockets on the GephE dimer (Fig. 4a; Supplementary Fig. 10a). 3D classification resulted in classes that either showed GlyR-loop density in both binding sites or a complete absence of GlyR-loop density, suggesting that in the case of binding, a single GlyR-loop is simultaneously occupying both binding sites within one GephE dimer (Fig. 4a, b; Supplementary Fig. 10b). Importantly, this finding rationalizes previous biochemical data reporting the binding of one GlyR-loop to one GephE dimer4,23,31,45. Moreover, machine-learning-based structure predictions46 of the most probable position of a single GlyR-loop within the GephE dimer are in line with the observed density for GlyR-loop in our reconstruction (Fig. 4c–e and Supplementary Fig. 11), and predict an extension of the GlyR-loop core binding motive (R394-Y412) towards the C-terminal end, enabling the occupation of both binding sites simultaneously (D413-L426, Fig. 4d, e). This agrees with previous in vitro studies showing that residues 413-426 are required for high-affinity binding of the GlyR-loop to GephE dimer23.

SDII is crucial for Geph filament formation, LLPS, and clustering at synapses