Structural basis for gating mechanism of Pannexin 1 channel

Dear Editor, Pannexin 1 (PANX1) plays extensive physiological roles across diverse fields of biology, including cell death, inflammation, cancer progression, and neurological disorders. As suggested by biochemical analysis, PANX1 forms an oligomeric channel for the facilitated diffusion of ions and large molecules across the plasma membrane upon activation. However, the underlying molecular gating mechanism of PANX1 and the structure of human PANX1 are still elusive. In the current study, we report the cryo-EM structures of full-length and carboxyl-terminal (CT) tail-cleaved human PANX1. Combined with single-channel electrophysiological study, we identified the key residues involved in the gating of PANX1. We purified full-length human PANX1 and solved the structure using cryo-EM (Supplementary information, Figs. S1 and S2). Among the classes of 2D particle images, the images representing the top views or bottom views exhibited seven obvious subunits in each detergent micelle. This observation differs from the previously proposed hexameric assembly of PANX1. Finally, the map was refined to 3.1 Å resolution (Supplementary information, Table S1 and Figs. S2–S4). The local resolution of the extracellular region reached ~2.8 Å, which allowed us to build residues with unambiguous assignment of side chains (Supplementary information, Figs. S2 and S4). The model we built contained an almost intact extracellular region, transmembrane domain, and partial intracellular region. Three fractions (N-terminal, 1–12; 158–193, between TM2 and TM3; and C-terminal, 355–426) were missing in the final model (Supplementary information, Fig. S5). The overall structure of human PANX1 presents a novel heptameric assembly (Fig. 1a), which is distinct from gap junction channels (connexins and innexins) and functionally related channels, such as CALHM and LRRC8. Interestingly, the protomer of PANX1 shares a similar fold of four transmembrane helices with the aforementioned channels (Supplementary information, Fig. S6). The distinctive structure of PANX1 implies its different substrate recognition and gating mechanisms. Caspase 3-mediated cleavage of the cytoplasmic CT tail irreversibly activates PANX1 and triggers the release of the “findme” signaling molecule for dying cell clearance. This cleavage-toactivation event was also observed in lipopolysaccharide-induced pyroptosis via caspase 11. Human PANX1 contains two identified caspase-cutting sites, site 1 (S1, DMRD) in the intracellular loop (IL) and site 2 (S2, DVVD) in the CT tail (Supplementary information, Fig. S1a). Cleavage at the S2 site is essential for the activation of PANX1. We generated CT-cleaved PANX1, named PANX1ΔCT, via Drosophila effector caspase (drICE) cleavage. As shown by SDSPAGE, PANX1ΔCT was partially cleaved by drICE at the S1 site, and this result is consistent with caspase 3 cleavage. However, the dramatically deferred monodisperse peak of PANX1ΔCT observed by gel filtration (~1–1.5 mL) indicated that PANX1ΔCT underwent a large conformational change or that reasonable disorder regions were removed (Supplementary information, Fig. S1b). To detect the activity of PANX1ΔCT, we performed protein reconstitution in a planar lipid bilayer and single channel electrophysiological assay in vitro (Supplementary information, Fig. S1c). Consistent with the patch clamp study, incorporation of wild-type PANX1 induced a small current opening of c.a. 3 pA, whereas incorporation of PANX1ΔCT resulted in the channel opening up to 177 pA under +100mV (Supplementary information, Fig. S1d). It is interesting that channel gating was observed in PANX1ΔCT (Supplementary information, Figs. S1d and S7), with a closure of c.a. 29% of the fully opened channel. This biochemical and electrophysiological investigation of PANX1 revealed that PANX1ΔCT and full-length PANX1 adopt two distinct states, active and inactive states, respectively. We therefore collected the cryo-EM dataset for PANX1ΔCT and solved its structure at 3.1 Å resolution following a similar image processing strategy as the full-length dataset (Fig. 1b; Supplementary information, Table S1 and Figs. S3 and S8). Surprisingly, the two structures revealed almost identical conformations except that the density of the intracellular region of PANX1ΔCT was weaker (Fig. 1b; Supplementary information, Figs. S5 and S8). Thus, the structure of full-length PANX1 was employed for the following structural analysis due to its clearer density map. The pore radius of PANX1 was calculated by HOLE. As the result shows, the radius of the narrowest constrictive site is ~4.7 Å, which is contributed by W74. The residue R75 also contributes to the positively charged ring (Fig. 1c, d). Interestingly, another constrictive site was uncovered at the extracellular site of the transmembrane domain. I58 from the tail of TM1 forms a larger hydrophobic ring (Fig. 1c, d). Importantly, the extracellular domain of each protomer provides an interface of the PANX1 oligomer (Fig. 1e). Each extracellular domain contains one helix, three βsheets and two conserved disulfide bonds (Fig. 1e; Supplementary information, Fig. S6). The interface between the two domains is constituted by extensive hydrophobic and hydrogen-bond interactions, which further stabilized the constrictive sites (Fig. 1e). The intracellular region of PANX1 comprises the N-terminus, the intracellular loop between TM2 and TM3, and the C-terminus (Supplementary information, Fig. S1a), which are essential for its regulation. We discovered a shrink ring formed by residues (T21/ E22/P23) in the N-terminus (Fig. 1f; Supplementary information, Fig. S4). To validate whether the ring is involved in gating, three residues (21–23) of PANX1 were deleted, resulting in PANX1Δ21–23. The single channel conductance of PANX1Δ21–23 was recorded (Supplementary information, Fig. S9). As the result shows, incorporation of PANX1Δ21–23 into the bilayer induced constantly open channels without effective gating. However, the channel conductance of PANX1Δ21–23 was smaller than that of PANX1ΔCT (Supplementary information, Fig. S9). Meanwhile, this result is consistent with gain-of-function mutations that cause aberrant PANX1 channel activity. Additional mutations of PANX1 also lead

effector caspase (drICE) cleavage. The drICE exhibits high homology and shares the same substrate specificity (DXXD) with mammalian caspases 1 . On-column digestion was carried out, which ensures that proteins were eluted from affinity column without the CT tail. For drICE-digested protein, drICE was added to resin suspension to a final concentration of 0.01 mg/mL and incubated at 4 °C for 2 h. Finally, drICE-digested PANX1 was eluted by 10 column volumes elution buffer without desthiobiotin. Eluted protein was concentrated and further purified by gel filtration (Superose 6 10/300 Increase column, GE Healthcare), buffered with 25 mM Tris pH 8.0, 150 mM NaCl and 0.01% LMNG. After identified by SDS-PAGE, peak fractions were collected for single channel recording or concentrated to approximately 8 mg/mL for cryo-EM sample preparation. All experiments were implemented at 4 °C.

Cryo-EM sample preparation and data acquisition
Aliquots of 3.5 µL concentrated protein were loaded onto glow-discharged holey carbon grids (Quantifoil Cu R1.2/1.3, 300 mesh). Grids were blotted for 3.5 s and plunge-frozen in liquid ethane cooled by liquid nitrogen using Vitrobot Mark IV (Thermo Fisher Scientific) at 8 °C with 100 percent humidity. Grids were transferred to a Titan Krios electron microscope operating at 300 kV and equipped with a Gatan Gif Quantum energy filter (slit width 20 eV). Micrographs were recorded using a K3 camera (Gatan) in super-resolution mode with a nominal magnification of 81,000x, resulting in a calibrated pixel size of 0.5435 Å. Each stack of 32 frames was exposed for 5.6 s, with an exposing time of 0.175 s per frame. The total dose for each stack was about 50 e -/Å 2 . AutoEMation was used for fully automated data collection 2 . All 32 frames in each stack were motion corrected with MotionCor2 3 , and dose weighting was performed 4 , with 2-fold binned to a pixel size of 1.087 Å/pixel. The defocus values were set from -1.0 to -1.5 µm and were estimated by Gctf 5 .

Image processing
The image processing procedures were shown in Fig S4 and Fig S8. A total of 2,541 and 1,828 micrograph stacks were collected, and 2,064,448 and 2,197,875 particles were auto-picked by Gautomatch (K. Zhang, www.mrc-lmb.cam.ac.uk/kzhang/) for full length and CT-cleaved dataset, respectively. All subsequent 2D and 3D classifications and auto-refinements were performed using RELION3.0 6 . After two rounds of reference-free 2D classification, 1,120,110 and 1,358,478 particles were selected for full length and CT-cleaved dataset, respectively. A subset of 100,000 particles was first applied to 3D classification with an initial model generated by RELION with C7 symmetry imposed. The best map from 3D classification was selected as the reference map for an auto-refinement of all selected particles. Then two rounds of local search 3D classifications were performed, each with several parallel runs in different class number (K=4 to 8). In the first round of local search 3D classification, an angular sampling of 1.8° was applied. The good particles from each run were selected and combined and duplicated particles were removed, yielding 397,740 and 381,458 particles for full length and CT-cleaved dataset, respectively. The selected particles were subjected to an auto-refinement, resulting in maps at 3.24 Å and 3.15 Å resolution, respectively. In the second round of local search 3D classification, an angular sampling of 0.9° was applied. A total of 251,945 and 133,330 selected particles from good classes were combined for auto-refinement, with local defocus values calculated with Gctf 5 for each particle. The final resolution of the 3D autorefinement after post-processing with an overall soft mask was 3.06 Å for full length dataset and 3.10 Å for CT-cleaved dataset.
To further improve the map quality of the intracellular region, the final selected particles of the full-length dataset were applied symmetry expanding for another round of local 3D classification (Supplementary information, Fig. S4). A soft mask that covers a monomer was applied during the 3D classification with C1 symmetry. The best class with 1,298,807 particles were selected for 3D auto-refinement, yielding a reconstruction with a resolution of 3.34 Å. Although the reported resolution of this map was not as high as the overall map, it shows better map quality in the intracellular region and addresses extra density at the N-terminals. Therefore, modeling building of the intracellular region was mostly based on this map.
Reported resolutions are based on the gold-standard Fourier shell correlation (FSC) 0.143 criterion. Before visualization, all density maps were corrected for the modulation transfer function of the detector and sharpened by applying a negative Bfactor that was estimated using automated procedures 7 . Local resolution variations were estimated using RELION.

Model building and refinement
The protomer of PANX1 was manually built in Coot 8 . Seven protomers were docked using PHENIX 9 . The model of full length PANX1 was docked in the density map of PANX1 ΔCT using PHENIX and followed by modification using Coot. The models were refined against the corresponding maps using PHENIX in real space (phenix.real_space_refine) with secondary structure and geometry restraints generated by ProSMART 10 .
Overfitting of the overall models was monitored by refining the models in one of the two independent half maps from the gold-standard refinement approach and testing the refined model against the other map 11 . Statistics of 3D reconstruction and model refinement can be found in supplementary Table S1.  Fig. S1 purification and characterization of PANX1 and PANX1 ΔCT . a Schematic of the full length PANX1. PANX1 contains two identified caspase-cutting sites (S1 in the ICL and S2 in the C terminus). b A representative size-exclusion chromatogram of purified full length PANX1 and PANX1 ΔCT (Superose 6 10/300 Increase). The peak fractions were applied to SDS-PAGE which was stained by Coomassie blue. In the SDS-PAGE, the proteins cleaved at only S2 and Both S1/S2 were labeled, respectively. c Diagram of single channel conductance recording setup. d Current recording traces of single