Connexin-46/50 in a dynamic lipid environment resolved by CryoEM at 1.9 Å

Gap junctions establish direct pathways for cells to transfer metabolic and electrical messages. The local lipid environment is known to affect the structure, stability and intercellular channel activity of gap junctions; however, the molecular basis for these effects remains unknown. Here, we incorporate native connexin-46/50 (Cx46/50) intercellular channels into a dual lipid nanodisc system, mimicking a native cell-to-cell junction. Structural characterization by CryoEM reveals a lipid-induced stabilization to the channel, resulting in a 3D reconstruction at 1.9 Å resolution. Together with all-atom molecular dynamics simulations, it is shown that Cx46/50 in turn imparts long-range stabilization to the dynamic local lipid environment that is specific to the extracellular lipid leaflet. In addition, ~400 water molecules are resolved in the CryoEM map, localized throughout the intercellular permeation pathway and contributing to the channel architecture. These results illustrate how the aqueous-lipid environment is integrated with the architectural stability, structure and function of gap junction communication channels.


Supplementary Tables and Figures
Supplementary Table 1. CryoEM Statistics. Summary of CryoEM data collection, refinement and model validation statistics. The ensemble CryoEM dataset was used to obtain the 1.90 Å resolution reconstruction and atomic models for Cx46 and Cx50, including 396 water molecules and 150 lipid acyl-chains. 3D classification was used to obtain the three PC classes, and associated atomic models for Cx46 and Cx50 (PC Class 1-3). Pre-processed and post-processed maps and associated masks from all datasets have been deposited to the EM databank (EMD-22358, EMD-22382, EMD-22390, EMD-22391). The original multi-frame micrographs have been deposited to EMPIAR (EMPIAR-10480). Coordinates for Cx50 and Cx46 atomic models have been deposited to the Protein Data Bank (7JJP and 7JKC correspond to the high-resolution models, 7JLW and 7JMD correspond to the ~2.5 Å models from PC Class 1; 7JM9 and 7JN0 correspond to PC Class 2; and 7JMC and 7JN1 correspond to PC Class 3). and total dose of ~60 eper Å 2 . Scale bar = 50 nm. b) Representative 2D class averages. Scale bar = 10 nm. c) Image processing and 3D reconstruction workflow carried out in Relion 4,5 , with representative maps at different stages of the image processing pipeline.

Supplementary
Step 1) De-novo model generated in Relion (left) and initial 3D AutoRefinement with D6-symmetry (~8 Å resolution, 3.9 Å pixel size) (right), which was then filtered to 20 Å and used for 3D template auto-picking in Relion (resulting in ~1.2M particle picks, which were culled to ~228k "good" particles following multiple rounds of 2D classification and de-duplication).
Step 4) Particles were de-duplicated, resulting in a set of ~221k particles, and subjected to 3D classification (two classes). Class 1 contained 88% of the particles and was further refined to 2.2 Å resolution (left). Class 2 contained 12% of the particles and was further refined to 2.0 Å resolution (right, asterisk). It is noted that the major differences between these two classes appears to be the distribution of particle defocus values, were the higher resolution class contains particles with lower defocus range (mean defocus = 1.20 ± 0.29 µm (s.d.)).
Step 5) Particles belonging to Class 2 (~26k particles), were then subjected to multiple rounds of 3D Auto-refinement followed by per-particle CTF, aberration-correction and polishing, using successively larger box-sizes until no further improvement, resulting in a final reconstruction at 2.3 Å resolution (C1 symmetry) (left) and 1.9 Å resolution (D6 symmetry) (center, left). There were no clear conformational differences between the C1 and D6-symmetrized reconstructions. The the sequence of Cx46 and Cx50 differ, and where sidechain density is weaker and/or consistent with heterogeneity. This is presumably due to the heteromeric/heterotypic mixture of these isoforms 1,6,7 and the imposed averaging of two different sidechains in these areas, and/or to relative flexibility at these sites, as many of these same residues correspond to solvent/lipid exposed sidechains (e.g., R9/N9; E43/F43, R68/E68, I91/V91 and A206/S218).

Supplementary Figure 4
Supplementary Figure 4. Image processing and resolution assessment for 3D lipidclassification work-flow. a) Image process and 3D reconstruction workflow carried out in Relion for the analysis of PC lipid configuration/conformational heterogeneity, with representative maps at different stages of the image processing pipeline. Steps 1 and 2) are the same as described in Supplementary Fig. 2, which resulted in a 2.7 Å reconstruction from a dataset of ~228k "good" particles (right).
Step 3) These particles were unbinned and re-extracted (0.649 Å/pix, 400 pixel box), and subjected to 3D classification (eight classes) without image alignment. Two of the eight classes yielded maps in which the lipid configuration was unambiguously resolved: assigned as PC Class 1 (yellow box), containing 9,190 particles (~4% of the data) and PC Class 3 (blue box), containing 6,944 particles (~3% of the data). Overlapping configurations were resolved in two of the other 3D classes (grey boxes).
Step 4) The particles from these overlapping classes (grey boxes) were combined and subjected to a second round of 3D classification with two classes. This yielded one in which the lipid configuration was unambiguously resolved: assigned PC Class 2 (orange box), containing 6,075 particles (~3% of the data).
Step 5) Particles assigned to PC Class 1 (left), PC Class 2 (center) and PC Class 3 (right) were separately subjected to a final round of 3D refinement and per-particle polishing, with D6 symmetry applied, resulting in final in the cytoplasmic space, to match either cellular or in vitro conditions used for CryoEM studies, respectively. All simulations were conducted with NaCl in the extracellular space and using DMPC as the lipid system. Following minimization, all systems were equilibrated for 30 ns at 37° C, and multiple replicates (n = 2) of production (100 ns each) were acquired for analysis at 37° C. b) Ca root mean squared deviation (r.m.s.d.) analysis of equilibrium (0 -30 ns) and production phases