Structural insights into the gating mechanism of human Cx43/GJA1 gap junction channel

Connexin family proteins assemble into hexameric hemichannels in a cell membrane, which dock together between two adjacent membranes to form gap junction intercellular channels (GJIChs). The most ubiquitously expressed connexin Cx43 plays important roles in numerous biological processes. Here we report cryo-EM structures of Cx43 GJIChs at 3.1–3.6 Å resolutions, which show dynamic conformational changes of N-terminal helices (NTHs) caused by pH change or C-terminal truncation. Cx43 GJIChs in a channel-closing condition contain 12 protomers in gate-covering NTH (GCN) conformation, while those in opening conditions have varying compositions of GCNs and pore-lining NTHs (PLNs) resulting in various pore dimensions and electrostatic surface potentials. GCN-to-PLN transition accompanies π-helix formation in the rst transmembrane helix (TM1), movement of TM2-4 that creates a side opening to the membrane, and structural stabilization of the cytoplasmic loop. Our study provides structural insights into the intercellular ion/metabolite transfer and the lateral lipid transport through Cx43 GJICh.


Introduction
sharply with lowering or rising pH until complete closure of the channel at pH 6.4 or 7.6 39,40 . A Cx43 derivative with deletion of residues from K258 to the C-terminus (Cx43-M257) is constitutively open at least in the pH range from 6.4 to 7.6 [40][41][42] .
Here we determined cryo-EM structures of wild-type Cx43 (Cx43-WT) GJICh at different pHs (pH 6.9 and 8.0) and Cx43-M257 GJICh at pH 8.0, revealing the conformational change of Cx43 by increased pH or Cterminal deletion. The detailed structural comparison between two different conformations of Cx43 reveals substantial conformational changes in not only NTHs but also TM helices and CL, including πhelix formation in the middle of TM1. Our conformational variation analyses focused on individual protomers identi ed various compositions of two different protomer conformations in Cx43 GJICh at pH 6.9, the condition leading to channel opening. Dye transfer assays and molecular dynamics (MD) simulations showed that the conformational change is crucial for the transport of molecules larger than ATP and the regulation of ion selectivity, respectively. We also found that the conformational changes open and close a small gap between two neighboring TMDs possibly for the export of lipids remaining inside the pore. These ndings suggest that the size and surface property of the pore pathway are regulated by individual conformational changes of NTHs, and the gating mechanism includes not only the conformational changes of NTHs in the cytoplasmic gate but also that of TMDs allowing the movement of lipid molecules in and out of the pore.

Results
Puri cation and structure determination of Cx43 GJICh.
To understand the gating mechanism of GJICh, high resolution structures in both open and closed states are necessary. Since previously reported high resolution structures of other GJIChs showed similar conformations of NTHs and TMDs which are thought to represent the open state, we rst aimed to solve the structure in the closed state. Based on previous electrophysiological studies 40 showing that Cx43 GJICh is inactive above pH 7.6 or below pH 6.4, we expressed human Cx43-WT using human embryonic kidney 293 EBNA (HEK293E) cells and puri ed the protein at pH 8.0 in a buffer containing lauryl maltose neopentyl glycol (LMNG) and cholesteryl hemisuccinate (CHS). Subsequent treatment with lambda-phosphatase 43 and a universal deubiquitinanse DUB (M48) 44 was necessary to overcome problems with sample heterogeneity (Extended Data Fig. 1a,b). Puri ed protein was concentrated to 1.4 mg/ml to increase the ratio of GJICh to hemichannel. The improved Cx43 sample dramatically reduced biased particle orientation in EM grids and yielded a nal EM map of Cx43-WT GJICh with D6 symmetry at 3.1 Å resolution (Fig. 1a,b, and Extended Data Figs. 1d and 2a,d, and Table 1). The nal structural model contains NTHs, TMDs and ECLs (Fig. 1c,d), but not CLs and CTLs, which remained unresolved due to poor EM density. In addition, since the map density for N-terminal methionine (M1) was completely invisible, we investigated N-terminal modi cation using mass spectrometric analysis. The results showed that Met1 was not detected at all, and the following G2 residue was partially The GCN conformation of Cx43 GJICh is mainly stabilized by the intramolecular interactions of NTH with TM2, as we also observed in the previous Cx31.3 hemichannel structure. L10 and V14 of the NTH participate in intramolecular hydrophobic interactions with Y92, L93, V96, F97, and M100 of TM2 (Fig.  2b). These residues are hydrophobically conserved (>80%) in connexin A and C classes, supporting the previous idea that GJIChs in A and C classes may have similar GCN conformations 9 (Fig. 2c). In contrast, we could not nd any close intermolecular interaction between NTH and TM2 except a salt bridge between D12 and R101 or any signi cant interaction between neighboring NTHs except an ionic interaction of D3 with the N-terminal amino group (Extended Data Fig. 3b).
We found signi cant structural differences between GCN conformations of Cx43 and Cx31.3. First, a superposition of Cx43 GJICh and Cx31.3 hemichannel led to poor alignment of TM2 and the NTH, in contrast to TM1 (Extended Data Fig. 3c). In particular, the cytoplasmic end of TM2 in Cx43 is ~6 Å away from the pore center compared to that of Cx31.3 (Extended Data Fig. 3d). Second, the interaction networks between hydrophobic amino acid residues differ signi cantly. For example, in Cx31.3, F5 of NTH closely interacts with L96 of TM2, whereas the corresponding residues of Cx43 (A6 and M100) are distant from each other ( Fig. 2b and Extended Data Fig. 3d). This conformational difference results in a large hole between two neighboring NTHs and TM2 at the cytoplasmic gate of Cx43 GJICh. Notably, this hole is lled with unidenti ed density features that we attribute to partially resolved detergent or lipid molecules (Figs. 1d, 2a and 2d). In addition, the map densities presumed to be CHS added during puri cation were also found in a deep hydrophobic pocket between NTH, TM1 and TM2 (Figs. 1d, 2a, 2d and 2e). These lipids bound to NTH, TM1, and/or TM2 of Cx43 likely contribute to stabilization of the GCN conformation by masking solvent-exposed hydrophobic surfaces.
To rule out the possibility that the GCN conformation is an artifact caused by detergents, Cx43 GJICh was reconstituted in lipid nanodiscs at pH 8.0 and the structure was determined at 3.6 Å (Fig. 1e, and Extended Data Fig. 1e and 2b, and Table 1). Although the resolution was not as high as that in detergents, it was su cient to con rm that the overall structure, including the NTH conformation, was almost identical to that in LMNG/CHS environment (Fig. 1e). In addition, this structure also showed strong map densities (presumed to be phospholipids and CHS) lling the gap between NTHs and the deep hydrophobic pocket, respectively (Fig. 1e).
To visualize the structure of the Cx43 GJICh in the open state, we puri ed Cx43-M257 GJICh that is reported to be constitutively open 40 , and analyzed it by cryo-EM (Methods). During the initial 3D classi cation with no imposed symmetry, we found two prominent classes, one with a clearly visible NTH and the other with an ambiguous one (data not shown). We attributed this to the latter class possibly having varying NTH conformations, and therefore performed a protomer-based conformational variation analysis using single-subunit-focused 3D classi cation ( Fig. 3a and Methods) 46 . The result showed that three classes (classes 3, 4, and 8) were in the PLN conformation, while the remaining ve were in the GCN conformation (Fig. 3a). Next, we traced back the GJICh particles to which the protomers in classes 3, 4, and 8 originally belonged, and examined the distribution of how many PLN protomers (protomers in the PLN conformation) were included in each GJICh particle. We found that GJICh particles (59,789 particles) had various PLN:GCN compositions from 0:12 (fully GCN GJICh, 2,183 particles, 3.7%) to 12:0 (fully PLN GJICh, one particle) (Fig. 3b). The number of GJICh particles with three PLN protomers was the largest (12,083 GJICh particles, 20.2%), and >70% particles contained 2-5 PLN protomers, suggesting that these particles might represent open states. Alternatively, the dynamic equilibrium between PLN and GCN states of individual protomers may be required for GJICh to be open.
It should be noted that two protomer classes (classes 4 and 8) have map densities of NTH in both GCN and PLN conformations (Fig. 3a). Therefore, the actual number of PLN protomers in the 3D classi cation data should be less than the number of protomers in classes 3, 4, and 8 (~30%) and may be close to that in classes 3 and 4 (~20%). Therefore, we also produced another distribution graph with only classes 3 and 4, which showed that fully PLN GJICh was not found and ~67% GJICh particles contain 2-5 PLN protomers (Fig. 3b).
Next, we suspected that some GJICh particles may have PLN protomers in the cryo-EM data of Cx43-WT GJICh at pH ~8, and thus we performed the same analysis with the Cx43-WT data. However, no protomer class with the PLN conformation appeared in the single-subunit-focused 3D classi cation (Extended Data Fig. 1d), suggesting that the partial deletion of CTL activated the GCN-to-PLN transition of Cx43.
Structure of Cx43-M257 GJICh composed of two different hemichannels in fully PLN and fully GCN conformations, respectively.
Since the majority of connexins in the Cx43-M257 dataset were in the GCN state, we could determine the structure of fully GCN GJICh at 3.2 Å, con rming that it is almost identical with the Cx43-WT GJICh structure (data not shown). However, it was not easy to solve any structures containing PLN protomers at high resolution, because it was di cult to accurately detect and classify the conformational changes of small NTHs in each GJICh particle. The most feasible scenario was likely to solve the structure of GJICh containing one hemichannel with six GCN protomers (fully GCN hemichannel) and the other hemichannel with six PLN protomers (fully PLN hemichannel), because the highest possible symmetry C6 can be imposed in the 3D reconstruction process. To this end, we performed "focused 3D classi cation and localized reconstruction" on hemichannels (Methods) and obtained one hemichannel class (Class 6, 7.3%) in fully PLN conformation (Extended Data Fig. 4d). With 8,046 hemichannel sub-particles in this class, we could determine fully PLN hemichannel structure with C6 symmetry at 4.3 Å (Extended Data Fig.  4e). 3D reconstruction was also performed without symmetry to con rm that all six protomers are in the PLN state (Extended Data Fig. 4e). Next, we traced back and gathered the original GJICh particles containing the hemichannel sub-particles in Class 6, and then performed 3D reconstruction with C6 symmetry (Extended Data Fig. 4f,g). The result showed the GJICh structure with fully GCN and fully PLN hemichannels at a fairly high resolution (3.6 Å) (Fig. 3c, and Extended Data Figs. 2c,e and 4, and Table 1). It should be noted that in the GJICh particles that we used for 3D reconstruction, most PLN hemichannels would contains 1-2 GCN protomer(s) and vice versa, although these errors seemed to be mostly averaged out during image processing.
Fully GCN hemichannel in this Cx43-M257 GJICh structure was nearly identical with that in the Cx43-WT GJICh structure at pH ~8.0. The superposition of the two structures showed the average Cα deviation of 0.3 Å, indicating that this GJICh structure of Cx43-M257 is reliable (Fig. 3d, left). In contrast, fully PLN hemichannel structure was rather similar to that of the Cx46/50 GJICh structure (average Cα deviation of 1.2 Å) (Extended Data Fig. 5a). For examples, W4, L7, and L10 residues of Cx43 NTH are bound to the hydrophobic surface formed by TM1 and TM2 (Extended Data Fig. 5b). Most notably, W4 residue, which is identically conserved in 18 human connexins, appeared to play a key role in stabilizing the PLN conformation by entering a deep pocket formed by two neighboring TM1s (Extended Data Fig. 5b). In addition, the loop structures between NTH and TM1 and the loosely packed unstable helix (π-helix) in the middle of TM1 are also similar between Cx43 and Cx46/50 in PLN conformations (Extended Data Fig. 5a).
GCN-to-PLN transition accompanies dramatic structural changes of TM1 including π-helix formation.
The Cα deviation between GCN and PLN hemichannels shows substantial changes in the cytoplasmic half of TMD, especially in TM1 (Fig. 3d, right) of which the middle part (residues 29-35; LFIFRIL sequence) forms an α-helix in the GCN conformation and a π-helix in the PLN conformation (Fig. 4a,b and Extended Data Fig. 5c-e). The α-to π-helix transition in TM1 necessarily causes an ~55° rotation of the cytoplasmic half, resulting in dramatic changes in the interactions of TM1 with other neighboring helices (Fig. 4c,d and Supplementary Video 1). First, F30 is dissociated from TM2 and forms new interactions with two NTHs, while R33 and F32 change side-chain conformations to maintain their original interactions (Fig. 4c,d, level 1). Second, as L29 and L26 move towards the pore, their interactions with TM4 are replaced by new interactions with TM2 and NTH (Fig. 4c,d, level 2). Third, V28, W25, and V24 of TM1, which participated in tight intermolecular interactions with TM2 of the neighboring protomer in fully GCN hemichannel, move away from TM2 and interact intramolecularly with TM4 in fully PLN hemichannel (Fig. 4c,d, level 2). This change results in a small gap between protomers, and the movement of S27 towards the gap slightly reduce the strong hydrophobicity of this region (Fig. 4d, level 2). Especially, W25, exposed to the membrane in the GCN state, moves into the hydrophilic center of TMD, participating in a hydrogen bonding with S22 and a cation-π interaction with K162 ( Fig. 4c,d, level 2). When we superposed GCN and PLN hemichannels, W25 partially overlaps with the map density presumed to be CHS. Therefore, the movement of W25 during the GCN-PLN transition would slightly push the lipid out of the pocket. Fourth, in the GCN state, L11 and Q15 of NTH respectively interact with F97 and Y98 of TM2 in a neighboring protomer, and V24 of TM1 also interacts with Y98 (Fig. 4c, level 3). F97 and L11 closely contact the map density presumed to be a detergent/lipid molecule. However, in the PLN state, these interactions are broken away and the unidenti ed map density disappears, and an opening towards the membrane is created between protomers (Fig. 4d, level 3).
The dynamic π-helix in TM1 is one of three most conserved regions in human connexins (Extended Data Fig. 5f) and almost identically conserved in classes A and B (Extended Data Fig. 5g). In addition, V24, W25 and S27 are also quite well conserved in human connexins (Extended Data Fig. 5g). Therefore, other connexins may undergo the same structural changes of TM1 to not only regulate the intercellular transport but also create the membrane opening.
It should be noted that the W4-binding pocket is located on the extracellular half of TMD and does not change signi cantly during the GCN-to-PLN transition. In addition, TM1 has another π-helix at the extracellular end, which is conserved in Cx26 and Cx46/50 GJICh structures (Extended Data Fig. 5a). However, this π-helix does not undergo structural changes during the GCN-to-PLN transition (Fig. 3e).
Conformational changes in the NTH-TM1 loop and TM2-4 during the GCN-to-PLN transition.
In the GCN-to-PLN transition, substantial structural changes were observed not only in TM1 but also in other TM helices. Cytosolic halves of TM2, 3, and 4 slightly rotate around TM1 (Extended Data Fig. 6a). In addition, the loop connecting NTH and TM1 (residues 15-20) moves ~3 Å towards the pore inside, and the interactions of some residues in the loop (Y17, S18, and T19) with TM helices largely change (Extended Data Fig. 6b,c and Supplementary Video 2). The hydrophobic interaction of Y17 with L151, L152, and Y155 in TM3 changes to that with Y92 and V96 in TM2. The hydrogen bonding of S18 with E227 in TM4, and the hydrophobic interaction of T19 with Y230 of TM4 are broken. Only the hydrogen bonding of T19 with E227 of TM4 is remained between the loop and TM4. Therefore, during the GCN-to-PLN transition, the main interaction of the NTH-TM1 loop with TM helices changes from a stronger interaction with TM3/TM4 to a weaker interaction with TM2 (Extended Data Fig. 6e).
MD simulations identify the penetration of phospholipids through the membrane opening.
A series of structural changes in the TM helices in the GCN-to-PLN transition, including ~55° rotation and 10° bending of TM1 by π-helix formation, results in the formation of V-shaped membrane openings between neighboring TMDs in Cx43 GJIChs (Fig. 4e, and Extended Data Fig. 6d and 6F-I, and Supplementary Video 3). The surface of this intermolecular opening is formed mostly by hydrophobic and basic residues (Fig. 4f). Remarkably, the cytoplasmic halves of TM helices appear to be relatively more exible in fully PLN hemichannel compared with fully GCN hemichannel (Fig. 4g). These structural analyses suggest that the membrane opening could be larger so that phospholipids in the inner lea et could pass through the interprotomer gap.
To validate whether lipids can move through the membrane opening in the PLN hemichannel, we performed MD simulations based on the CHARMM36m force eld 47 with corrections for charge-charge interactions 48,49 . Because our structural models of Cx43 GJIChs are missing the unstructured CLs (residues 111 to 147), we reconstructed the loops in a random con guration with no clash and no intertwining between chains. Then, we incorporated the channel into a double bilayer system of a 1palmitoyl-2-oleoyl-sn-glycero-3-phosphatidylcholine (POPC), including an explicit ionic solution at 150 mM NaCl (Methods). We simulated each of the assembled systems for 400 ns in total (200 ns without any external bias followed by 200 ns at a voltage bias of 200 mV. We could detect head groups of several phospholipid molecules spontaneously inserted into the membrane opening in fully PLN GJICh, but not in fully GCN GJICh (Fig. 4h,I and Supplementary Videos 4 and 5), suggesting that phospholipid molecules may diffuse into fully PLN GJICh in a head rst fashion. Some basic residues (K23, K109, and K234) as well as hydrophobic residues on the surface of the membrane opening are highly conserved in human connexins, suggesting that the lateral lipid transport mechanism may be similar in other connexins.
The constriction diameter in the pore pathway of fully PLN hemichannel was ~10 Å. Although it is bigger than that of fully GCN hemichannel (~7 Å), it is not su ciently large for e cient transport of ATP or bigger molecules. However, we could easily predict that a hemichannel with 2 or 3 consecutive PLN protomers would have a bigger pore diameter based on our Cx43 GJICh structures. Although we could not determine the Cx43 hemichannel structures with mixed PLN and GCN protomers, we could reliably build arti cial models of those hemichannels because of almost no structural difference in ECLs and the extracellular halves of TMDs between GCN and PLN protomers (Fig. 3e). We produced ve hemichannel models: one with only one PLN protomer, three containing 2-4 consecutive PLN protomers, and one with ve PLN protomers (Fig. 5a). In these models, we con rmed almost no steric hinderance at the interface of GCN and PLN protomers (Extended Data Fig. 6j,k).
Next, we analyzed the solvent-accessible pore diameters of the ve hemichannel models and compared with those of fully GCN and PLN hemichannels (Fig. 5b). We found that the constriction diameters of the hemichannel pores formed by 4:2, 3:3, and 2:4 ratios of GCN and PLN protomers are 15.1, 16, and 14.6 Å, respectively, which are larger than the hydrodynamic diameter of ATP (~9.8 Å) 45 .
Dynamic conformational transition of TM1 is required for the transport of large molecules.
To see whether dynamic structural transitions between GCN and PLN conformations are physiologically important for the transport of large molecules through Cx43 GJIChs, we performed intercellular dye transfer experiments 50 . We transfected HEK293E cells with an expression plasmid harboring the CFPfused Cx43-WT gene, con rmed frequent formation of Cx43 gap junction, and performed scrape-loading experiments using sulforhodamine B (~560 Da) uorescent dye. Next, we measured the maximum distances of dye transfer from a scrape line for 15 minutes and found that the distance in Cx43 expressing cells was three times longer than that in the control cells transfected with mock plasmid ( Next, we carefully selected mutations that could destabilize the GCN state but not the PLN state (L91A and V24R) and those with the opposite effect on the two states (I31A and W25R) (Fig. 5d). The side chain of L91 in TM2 is completely exposed to the membrane in the PLN state, while this residue participates in tight intermolecular interactions with V24, W25, and V28 of TM1 in the GCN state ( Fig. 5c, top panels). Therefore, the L91A mutation is likely to inhibit conformational change and increase the PLN/GCN ratio in GJIChs. Conversely, the side chain of I31 in TM1 is completely exposed to the pore inside in the GCN state, while this residue engages in tight intermolecular interactions with L7, L10, V87, and I83 in the PLN state (Fig. 5c, bottom panels). Therefore, I31A is expected to lower the PLN/GCN ratio. For every scrapeloading experiment for these Cx43 mutants, we con rmed that their expression levels and gap junction formation rates were comparable with those of wild-type Cx43 (Extended Data Fig. 7b). The results showed that both mutations almost completely inhibited the dye transport activity ( Fig. 5e and Extended Data Fig. 7c), suggesting that a dynamic conformational transition of TM1 is required for the transport of large molecules (Extended Data Fig. 7d and Supplementary Video 6).
In addition, mutations in V24 and W25 located near the head groups of membrane lipids were also expected to increase and decrease the PLN/GCN ratio, respectively, and thus we tested the effects of V24R and W25R mutations on the dye transport activity (Extended Data Fig. 7a). However, although these mutations completely blocked dye transfer, they also substantially lowered the gap junction formation e ciency ( Fig. 5e and Extended Data Fig. 7b,c). These mutations may also inhibit localization of Cx43 to the plasma membrane.
GCN-to-PLN transition switches electrostatic surface potential of the cytoplasmic gate from positive to negative.
Whereas the cytoplasmic surface of fully GCN hemichannel has an overall positive charge, that of fully PLN hemichannel shows a locally negative surface at the pore pathway (Fig. 6a). This change, which should affect ion transport e ciency and selectivity, is because the NTH not only bends but also rotates during the GCN-to-PLN transition, and therefore basic residues (K9 and K13) exposed on the cytoplasmic surface in the GCN conformation are partially buried at the interface between the NTHs in the PLN conformation, and acidic residues (D3 and D12) are exposed on the cytoplasmic surface of the funnellike pore (Fig. 6b). D3 forms the pore constriction sites in both PLN and GCN conformation (Figs. 4a and 6b). D12 is involved in close intermolecular interaction(s) with R101 and K105 in the GCN conformation and with K9 in the PLN conformation (Fig. 6b). In addition, ODDD-related mutations of these two acidic residues, D3N and D12S, have been reported to reduce gap junctional conductance and lower V j gating sensitivity, respectively 32,34,51 .

MD simulations show anion selectivity of fully GCN GJICh and cation selectivity of fully PLN GJICh.
To investigate the ion transport mechanism through Cx43 channels, we analyzed the position-dependent local concentration and current of ionic species in MD simulations of fully PLN, 3:3 mixed (3 consecutive PLNs and GCNs) and fully GCN GJIChs at a 200-mV transmembrane potential 52 ; see the gray heatmap for the local concentration and streamlines for the current (Extended Data Fig. 8a,b). We found that GCNs form a Clreservoir at the entrance of the channel (z ~ 7 nm) with a local Clconcentration at the Clentrance of GCN channel (>2 M) much higher than that of PLN channel (Extended Data Fig. 8b, left and right panels). Possibly due to the difference in Claccumulation at the entrance, the Clcurrent through fully GCN GJICh (45 pA) was much higher than that through fully PLN GJICh (2 pA) (Fig. 6c). However, Na + ions are accumulated not in the Na + -entering hemichannel but in the Na + -leaving hemichannel of fully GCN GJICh (Extended Data Fig. 8a, left), suggesting that six GCNs block Na + ions, resulting in a Clselectivity. In fully PLN GJICh, we could not observe signi cant accumulation of Na + ions, resulting in the Na current (13 pA) higher than the Clcurrent (i.e., selective to Na + ) ( Fig. 6c and Extended Data Fig. 8a,b, right).
We also examined the ion transport through our Cx43-WT GJICh structure with lipids that bind to the pore inside and block twelve holes between NTHs and TM2 in the GCN state. The result showed a sharp decrease of the Clcurrent, suggesting that lipids would play a major role in channel closing by blocking the six holes at the cytoplasmic gate ( Fig. 6c and Extended Data Fig. 8c-e). In contrast, high Clselectivity remained regardless of bound lipids ( Fig. 6c and Extended Data Fig. 8c-e), supporting our conclusion that fully GCN GJICh is Cl --selective and fully PLN GJICh is Na + -selective. In addition, only slight increase in the ionic current and no charge selectivity in the 3:3 mixed hemichannel suggests that electrostatics would be more important in ion transport than the pore size ( Fig. 6c and Extended Data Fig. 8a,b, middle). It should be noted that Cl --transport activity and selectivity of fully GCN GJICh may not be physiologically relevant, because the pore in the state may contain more lipids, which were not included in our simulations.
Since Cx43-WT GJICh exhibits the maximum H + ion transmission at pH 6.9 and the minimum at pH 6.4 or lower 40 , we aimed to unravel the structures at pH 6.9 and pH 6.4. For the cryo-EM imaging, we lowered the pH of the protein sample from ~8.0 to ~6.9 and ~6.4 just before vitri cation. While Cx43-WT GJIChs at pH ~6.4 were severely denatured and aggregated, those at pH ~6.9 were relatively stable (Fig. 7a) and their structures determined with D6 symmetry to 3.4 Å resolution. We then performed the same protomerbased conformational variation analysis as in the study of Cx43-M257 GJIChs ( Fig. 7b and Methods). We could nd a protomer class (Class 2; 7.8%) with the PLN conformation in single-subunit-focused 3D classi cation. The investigation of the original GJICh particles to which PLN protomers in class 2 belonged showed that ~44% GJICh particles were in the fully GCN state, ~33% contained one PLN protomer, and ~20% contained two or three PLN protomers (Fig. 7c). These data suggest that pH ~6.9 increases the structural exibility of Cx43 to cause various GCN and PLN protomer compositions in GJIChs, although the effect of pH ~6.9 is lower than that of CTL deletion (Figs. 3b and 7c).
Structural stabilization of the CL in the PLN state.
In the Cx43-M257 GJICh structure, we found a strong map density corresponding to the CL in fully PLN hemichannel, but not in fully GCN hemichannel (Fig. 7d,e). Although this map was not su ciently clear to correctly register the residues of the CL, we could conclude that it includes the total length of the CL (residues 111-147) based on the length of continuous map and the connectivity with TM2 and TM3.
This stabilization of the CL could also be detected in PLN protomers in the dataset of Cx43-WT at pH 6.9 (Fig. 7b). The CL map is in a large-area contact with the cytoplasmic protrusion of TM2 and the NTH-TM1 loop (Fig. 7e,f), suggesting that this interaction could stabilize the PLN conformation. The contact sites include a hydrophobic patch formed by Y17 of the NTH-TM1 loop, Y92 of TM2, and Y155 of TM3, which may interact with hydrophobic residues in the CL (Fig. 7e,f).

Discussion
We determined the structures of Cx43 in two different conformations (GCN and PLN) and revealed that the GCN-to-PLN transition includes not only the conformational change of NTH but also π-helix formation in the middle of TM1 and reorganization of the interaction network among all TM helices. Although high resolution structures were solved in a detergent environment, we con rmed the GCN conformation from a lower resolution structure in lipid nanodiscs ( Fig. 1e and Extended Data Fig. 1e), and the available Cx46/50 structure in lipid nanodiscs showed a very similar PLN conformation (Extended Data Fig. 4a). Based on these structures, we could also design several point mutations that would inhibit conformational transition without disturbing structural integrity. Our dye transfer study in human cells con rmed that those mutations indeed abolish the transport activity through Cx43 GJIChs (Fig. 5e).
These data strongly support the physiological relevance of two identi ed Cx43 conformations.
We took advantage of cryo-EM technology that can identify conformational variants, and found that puri ed Cx43 GJIChs in channel opening conditions (pH ~6. It should be noted that the distribution of PLN/GCN ratios in Cx43 GJIChs described in this study, especially that at pH ~6.9, may not represent those in their physiological membrane environment.
Although we con rmed that Cx43 at pH ~8.0 is mainly in the GCN state in both lipid and detergent environments, conformational preference of Cx43 may at least be partially dependent on its lipid environment. In addition, many factors in the cell membrane and cytoplasm could change the PLN/GCN ratio distribution of individual GJIChs in gap junction plaques.
We showed that basic pH induces the channel closing by decreasing the PLN/GCN ratio in hemichannel.
Basic pH is thought to be sensed by C298 residue at CTL 40 . Since the CL is located between NTH and CTL, some changes in CTL would be transferred to NTH not directly but via the CL. Therefore, it was hypothesized that CTL, which is dissociated from the CL at neutral pH, would be bound to the CL at basic pH resulting in the channel closing. This idea is supported by the strong map density of the CL bound to TM2 and the NTH-TM1 linker in the PLN conformation (Fig. 7e). Since the map density of the CL is not seen in the GCN conformation, we think that the CL is released from TM2 and the linker at a basic pH by the interaction with CTL, and thus the GCN conformation become more favorable. Similar to the mechanism of how pH regulates the gating, other regulatory factors may also not target NTH directly but other regions to eventually induce conformational change of NTH via that of TM1 or the CL.
The sequence analysis of the connexin family (Extended Data Fig. 5g) suggests that some of the structural changes shown in Cx43, such as π-helix formation in the middle of TM1 and the interaction of N-terminal tryptophan with the hydrophobic pocket formed by two TM1s and TM2, may be conserved in other connexin family members. However, the hydrophobic residues involved in the intramolecular interaction between NTH and TM2, which is crucial for the GCN conformation, are conserved only in A and C class connexins (Fig. 2c). Therefore, connexins in other classes may have a different NTH conformation when it is dissociated from TMD. This is consistent with the structures of Cx26 in B class, which showed α-helix in the middle of TM1 but exible NTHs 5,6 . None of Cx26 GJICh structures showed π-helix in the middle of TM1 and W3 (corresponding to W4 in Cx43) bound to the hydrophobic pocket formed by two TM1s, although the π-helix-forming residues, W3, and the pocket-forming residues are highly conserved in this channel. Cx26 may have the PLN conformation similar as shown in Cx43 for a short time or in an unknown condition.
Cx46 and Cx50 preferably adopt the PLN conformation in both detergent and lipid environments. These members may function only as fully PLN GJICh with the mission of ion transfer. Alternatively, they may be able to transport large molecules by PLN-to-GCN transition of 2-4 consecutive protomers in a hemichannel. Since the residues involved in the intramolecular interaction of NTH and TM2 in the Cx43 GCN conformation are highly conserved in Cx46 and Cx50, Cx46/50 likely undergoes the same conformational change as Cx43.
The pore inside in the fully GCN state has lipid molecules tightly bound to the hydrophobic surface, whereas that in the fully PLN state has hydrophilic surface without bound lipids (Figs. 2a and 6a, left and right). Therefore, lipids in the pore in the GCN state should be exported to the membrane during the conformational transition for gating regulation. We found the membrane opening between PLN protomers similarly as observed in innexin-6 GJICh structure in lipid nanodiscs 54 , and our MD simulation study showed lipid molecules can pass through the membrane openings of fully PLN GJICh (Fig. 4i).
These data suggest that the dynamic conformational changes of individual protomers in Cx43 GJICh facilitate not only the intercellular transport of ions/metabolites through but also the lateral lipid transport in and out of the channel pore.

Methods
Plasmid construction and mutagenesis.
A synthetic gene fragment encoding the full-length human Cx43 (GJA1) (UniProt ID # P17302) was purchased from Integrated DNA Technologies and inserted into pX vector as previously reported 55 . The human Cx43-WT proteins were expressed as fusion constructs with a human rhinovirus (HRV) 3C cleavage site, an enhanced yellow uorescence protein (eYFP) tag, a 10´His-tag and a rho-1D4 epitope tag (8 amino acid sequence of TETSQVAPA) at its C-terminus. To make a Cx43-M257 expression plasmid, the PCR-ampli ed gene fragment encoding 1-257 amino acids of human Cx43 was inserted in pX vector.
For dye transfer assay, we also produced the expression plasmid containing an enhanced cyan uorescence protein (eCFP)-fused version of Cx43-WT to avoid the interference with a uorescent signal from sulforhodamine B (SRB) dye. Using this plasmid as a template, we introduced point mutations by PCR to make the expression plasmids of Cx43 mutants.
Protein expression and puri cation of human Cx43-WT.
Cx43-WT expression plasmids were ampli ed in Escherichia coli DH5α strains, puri ed, mixed with linear polyethylenimine (PEI, Polysciences, Inc.), and transfected into human embryonic kidney 293E (HEK293E) cells 55,56 . When the HEK293E suspension cells (2 L) were grown at 37°C to a density of 0.6 × 10 6 cells/ml in Ca 2+ -free Dulbecco's modi ed Eagle's medium (DMEM) supplemented with 5% fetal bovine serum, the mixtures of DNA and PEI were added to the cells. Dimethyl sulfoxide (Amresco) was added to the nal concentration of 1%, and temperature was lowered to 33°C immediately after transfection. Tryptone (Amresco) was added 48 hours after transfection to the nal concentration of 0.5%. The transfected cells were harvested 96 hours after transfection.
All puri cation steps were carried out at 4°C unless indicated otherwise. The harvested cells were universal deubiquitinase DUB (M48) at a 15:1 (w/w) ratio of Cx43 to the enzymes. For the activation of λ-PPase, 1 mM MnCl 2 was added to the reaction buffer, and the reaction was stopped by adding 5 mM EDTA after 2 hours. Next, the proteins were concentrated to ~3 mg/ml using an Amicon Ultra centrifugal lter (molecular weight cutoff, 100 kDa), ltered with a 0.22 μm lter, and then loaded on a Superose 6 Increase 10/300 column (GE Healthcare) equilibrated with buffer C. Peak fractions were pooled, concentrated to ~3.2 mg/ml, ash-frozen in liquid nitrogen, and stored at −80°C for nanodisc reconstitution and EM grid preparation. Protein purity and quality were assessed by SDS-polyacrylamide gel electrophoresis (SDS-PAGE) and negative-stain EM.
Protein expression and puri cation of Cx43-M257.
Cx43-M257 was expressed in HEK293E cells in the same method as in the Cx43-WT expression. Protein puri cation was carried out at 4°C. The transfected cells were lysed using a Dounce homogenizer in Expression and puri cation of membrane scaffold protein.
The membrane scaffold protein (MSP1E3D1) was expressed and puri ed as previously described 57  Reconstitution of Cx43-WT in lipid nanodiscs.
Puri ed Cx43-WT was reconstituted into lipid nanodiscs as previously described 57 , with some modi cations. Soybean polar lipid extract powder (Avanti Polar Lipids) was dissolved in chloroform, aliquoted, and dried completely to be a thin layer using argon stream. Just before reconstitution, the lipid lm was solubilized in 5/0.5% (w/v) LMNG/CHS, and incubated at 60°C for 30 min to make a clear lipid stock solution at a concentration of ~10 mg/ml. All the reconstitution steps were carried out on ice. Cx43-WT sample (~3.2 mg/ml, 0.68 mL) was mixed with the lipid stock (~10 mg/ml, 0.26 mL) at the 1:66 molar ratio of Cx43-WT to lipid, and incubated for 20 min. Next, MSP1E3D1 (~3.6 mg/ml, 0.14 mL) was added to the Cx43-lipid mixture at ~1:0.3 molar ratio of Cx43-WT to MSP1E3D1, and incubated for 40 min. Bio-beads SM2 absorbent (120 mg per 1 ml mixture, Bio-Rad) pre-washed with buffer G were added to the Cx43-lipid-MSP mixture to remove detergents, and incubated overnight with slow rotation. Bio-beads were removed by centrifugation and fresh Bio-beads were added and incubated for 4 hours to remove residual detergents. After Bio-beads were removed, the reconstituted mixture was concentrated to reduce the sample volume up to 500 μl, ltered with a 0.22 μm lter, and loaded on a Superose 6 Increase 10/300 column equilibrated with buffer G. Peak fractions containing Cx43 in nanodiscs were pooled, and concentrated to ~2.2 mg/ml for EM grid preparation. The reconstitution quality was assessed by SDS-PAGE and negative-stain EM.
Cryo-EM grid preparation and data collection.  Table 1.
Cryo-EM data processing.
All datasets of Cx43 samples were processed with RELION 3.1 58 or cryoSPARC version 3.1 59 softwares. The overall work ow for data processing is presented in Extended Data Figs. 1 and 4. For the dataset of Cx43-WT in detergent solution at pH 8.0, RELION 3.1 was used for the processing. Beam-induced motion correction and dose weighting of raw movies were performed using MotionCor2 version 1.2.6 60 . Contrast transfer function (CTF) estimation was performed using Gctf version 1.18 61 .
Micrographs unsuitable for further image processing such as those containing ice crystals, large motion drifts, extremely low or high defocus, and large astigmatisms were removed by manual inspection and selection job in RELION. As a result, 4,344 micrographs were selected for further processing, and then 1,130,417 semiautomatically picked particles were extracted into 400 × 400 pixel boxes. Many falsepositive images, poorly aligned particles, off-centered particles, and undocked hemichannel particles were discarded by performing 12 rounds of two-dimensional (2D) classi cation. Using 293,152 sorted particles, a 3D initial model of Cx43 GJICh was generated by a stochastic gradient descent algorithm in RELION with C1 symmetry. After 4 rounds of 3D classi cation with C1 or D6 symmetry imposition, 26,451 selected particles in the classes with good quality side views of GJIChs were used in 3D re nement. Perparticle motion drifts of assorted particles were corrected using Bayesian polishing 62 and 3D re nement with D6 symmetry imposition was performed, generating a 3.4 Å map. After one more round of 3D classi cation without orientation search, 21,689 selected particles were subjected to 3D re nement with D6 symmetry imposition, resulting in a 3.3 Å map. For further improvement of the map, per-particle CTF estimation was performed using CTF re nement 63 in RELION. After 3D re nement with D6 symmetry imposition, the estimated resolution of the nal map was 3.1 Å based on the 0.143 Fourier shell correlation (FSC) criterion 64 . The local resolution of the density map was assessed by ResMap 65 .
For the dataset of Cx43-WT in nanodiscs at pH 8.0, cryoSPARC was used for the processing. Beaminduced motion correction and dose weighting of raw movies were performed using patch motion correction. Then, CTF parameters were estimated using patch CTF estimation for further CTF correction. 1,683,178 particles were picked from 6,760 micrographs using Template picker with the initial 2D templates, which were generated by randomly selected 500 micrographs using Blob picker, and extracted into 512 × 512 pixel boxes. After rst 2D classi cation, only 52,110 particles were selected due to the extremely high false-positive rates and the large numbers of atypical particles. In addition, the number of side-view particles were insu cient to make a reliable 3D reconstruction map. To increase the number of side-view particles, we used Topaz 66 in cryoSPARC for particle picking, resulting in 73,964 side-view particles. Top-view particles previously selected by Template picker and side-view particles selected by Topaz were combined for the further steps. Finally, 28,341 particles with various orientations were selected by two more rounds of 2D classi cation and used for local motion correction and 3D homogenous re nement with D6 symmetry imposition. The estimated resolution of the nal map was 3.6 Å based on the 0.143 FSC criterion. The local resolution of the density map was assessed by ResMap.
For the dataset of Cx43-M257 in detergents at pH 8.0, RELION 3.1 was used through the processing. Motion correction and CTF estimation were performed using the MotionCorr2 and the Gctf, respectively. After removal of micrographs unsuitable for image processing, 2,781 micrographs were selected, and 559,754 semiautomatically picked particles were extracted into 400 × 400 pixel boxes. After 5 rounds of 2D classi cation, several rounds of 3D classi cation were performed with C1 or D6 symmetry. In the rst 3D classi cation, the nal map of Cx43-WT GJICh in detergents at pH 8.0 was lowpass-ltered to 30 Å and used as a reference map. Selected 59,789 particles classi ed into different classes with good quality side views of GJIChs were subjected to 3D re nement. After Bayesian polishing, 3D re nement with D6 symmetry imposition were performed to generate 3.3 Å map.
The dataset of Cx43-WT in detergents at pH 6.9 was processed in the same protocol. Motion correction and CTF estimation were performed using the MotionCorr2 and the Gctf, respectively. After removal of micrographs unsuitable for image processing, 2,997 micrographs were selected, and 3,224,367 semiautomatically picked particles were extracted into 400 × 400 pixel boxes. After 16 rounds of 2D classi cation, several rounds of 3D classi cation were performed with C1 or D6 symmetry imposition. In the rst 3D classi cation, the nal map of Cx43-WT GJICh in detergents at pH 8.0 was low-pass-ltered to 30 Å and used as a reference map. Selected 8,861 particles classi ed into different classes with good quality side views of GJIChs were subjected to 3D re nement. After Bayesian polishing, 3D re nement with D6 symmetry imposition were performed to generate 3.5 Å map.

Focused classi cation on protomers and determination of the number of PLN protomers in GJICh.
In the datasets of both Cx43-M257 in detergents at pH 8.0 and Cx43-WT in detergents at pH 6.9, during 3D classi cation steps with C1 symmetry, we identi ed two noticeable classes with and without NTH densities. To distinguish possible different classes representing various conformations of NTHs, we used single protomer-based conformational variation analysis 67 . For this analysis, 3.3 Å map of Cx43-M257 in detergents at pH 8.0 and 3.5 Å map of Cx43-WT in detergents at pH 6.9 were used as consensus maps.
Using already well-aligned particles used for reconstructing the consensus maps, we ran a command "relion_particle_symmetry_expand" in RELION 3.1 to produce 11 copies of each particle by computationally rotating and tilting based on the D6 symmetry axis. This resulted in 12 times the total number of particles, which were subjected to Particle subtraction job to generate subtracted particles containing only the signal of one protomer. The subtracted particles were re-centered and re-extracted into 200 × 200 pixel boxes, and subjected to a focused 3D classi cation (K=8; the number of classes to classify is 8) with a mask covering one protomer and without orientation search. This resulted in two distinct classes of protomers with GCN and PLN conformations for both Cx43-M257 (pH 8.0) and Cx43-WT (pH 6.9) datasets (Figs. 3a and 7b).
To see the distribution of how many PLN protomers (protomers in the PLN conformation) are included in each GJICh particle, we analyzed the metadata les generated by focused 3D classi cation. The metadata contains all subtracted protomer particles labeled with the class number of each protomer and the identi cation (ID) number of each original GJICh particle. Because of D6 symmetry expansion, the same identi cation number of GJICh should be found 12 times in the metadata. The metadata was sorted by the protomer class numbers from focused 3D classi cation, and the GJICh ID numbers were collected only for the classes with the PLN conformation. How many times the same ID is repeated in the collection indicates how many PLN protomers are included in the corresponding GJICh particle. To sort particles in the metadata, count the number of repetitions of a GJICh ID, and calculate the number of GJIChs with each of 13 different protomer compositions (0:12 to 12:0 of GCN and PLN) at the same time, we used a shell command below.
Structure determination of fully PLN hemichannel by focused classi cation and localized reconstruction.
The Cx43-M257 dataset and its 3.3 Å consensus map were used for focused classi cation on and localized reconstruction of hemichannels. We followed the procedures reported by Gestaut et al 68 . The overall work ow is presented in Extended Data Fig. 4. We used two masks covering each hemichannel of the consensus GJICh map to generate the subtracted particles in two opposite orientations. After recentering and re-extraction into 300 × 300 pixel boxes, we subjected the subtracted particles to 3D classi cation (K=6) with C1 symmetry imposition to align into one orientation. We chose ve hemichannel classes (Class 1, 3, 4, 5, and 6) with good quality side views and performed focused 3D re nement (C1 symmetry) with a soft mask covering hemichannel to obtain a hemichannel map at 3.8 Å resolution. Using this map as a new consensus map, we ran 25 iterations of 3D classi cation (K=8, T=4; T is Regularisation parameter) without orientation search. The result showed three hemichannel classes (Class 2, 5, and 6) containing one or two PLN protomer(s). In additional 15 iterations of 3D classi cation, we increased the T value (T=40) and applied a soft mask covering the cytoplasmic half of hemichannel to focus more on the conformational changes of NTHs. This generated one unique class (Class 6) in the fully PLN state. 8,046 particles in Class 6 were subjected to 3D re nement with C1 symmetry imposition and 1.8° angular sampling, which produced a 7 Å map. Another round of focused 3D re nement with C6 symmetry imposition resulted in a 4.3 Å map. Next, we traced back the original GJICh particles from which the subtracted hemichannel particles came, and redundant particles were removed. Without changing particle's orientations from focused 3D re nement, total 7,446 GJICh particles were subjected to 3D re nement with C6 symmetry imposition and 1.8° angular sampling. The result showed a 3.6 Å GJICh map with fully PLN hemichannel on one side and fully GCN hemichannel on the other side. This map was not subjected to sharpening because it signi cantly weakens map densities of NTHs. The local resolution of the density map was assessed by ResMap.
Model building, validation, and structural analysis.
The structural model for the Cx43-WT GJICh in detergents at pH 8.0 was built manually in Coot program 69,70 and re ned using phenix.real_space_re ne 71 in PHENIX software suite with secondary structure and non-crystallographic symmetry (NCS) restraints. This structure does not contain Met1 and N-terminal acetylation (see the main text for details). The nal model also does not include exible cytoplasmic regions such as CLs (residues 111-147) and CTLs (residues 236-382) because the map densities of these regions are invisible. We also put acyl chain models into long map densities around TMDs, but did not put any lipid model into unidenti ed densities presumed to be a head group of phospholipids/detergents and a sterol because of ambiguous map densities. The geometric restraints for acyl chains were optimized using the eLBOW module 72 Table 1. The pore radii and the surface of the pore pathway were calculated using HOLE 76  For all MD simulations, we used the GROMACS 2020.2 package 78 and the CHARMM36m force eld 79 , combined with the CHARMM-modi ed TIP3P model and the CUFIX corrections for charge-charge interaction pairs 80 . All simulations were performed under a constant surface tension-constant temperature (NPgT) ensemble at zero surface tension (g = 0) 81 and 300 K temperature 82 . Van der Waals forces were computed using a 10-to 12-Å switching scheme. Long-range electrostatic forces were computed using the particle-mesh Ewald summation scheme 83 of a 1.2-Å grid spacing and 12-Å realspace cutoff. Covalent bonds to hydrogen in non-water and water molecules were constrained using the LINCS 84 and algorithms 85 , respectively. During MD simulations using a 2-fs time step, we saved atomic coordinates every 20 ps for analysis. The visualization scheme for the density-ux map is described in a previous report by Yoo and Aksimentiev 86 .
MD preparation of Cx43 GJICh embedded in a lipid bilayer.
We manually reconstructed the unstructured CL of each Cx43 chain (residues 111 to 147) missing in experimental GJICh models; the reconstructed loops were in a random con guration with no clash and no intertwining between chains. To create a double bilayer system, we placed a lipid bilayer of a 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphatidylcholine (POPC) near the transmembrane domain of each hemichannel and removed lipid molecules overlapping with the channel. We immersed the complex system of lipid molecules and the channel in an explicit solution of 150 mM NaCl. The nal system contained a channel, 734 POPC lipids, 88,000 water molecules, 300 Na ions, and 360 Cl ions in a periodic hexagonal box ( . We energy-minimized each system for 5000 steps and equilibrated it for 200 ns with the positions of experimentally determined heavy atoms harmonically constrained using a force constant of 1,000 kJ/mol; note that we did not constrain the positions of reconstructed CLs. For the measurements of ionic currents, we performed a 200-ns simulation under 200 mV without any constraints for each system, starting from the nal structure of the equilibration.
Scrape-loaded dye-transfer assays were performed according to the method described previously 87,88 , with some modi cations. Brie y, HEK293E cells were grown to 50-60% con uency in 24-well plates and transiently transfected with Cx43-eCFP expression plasmids. The cells were rinsed twice with Hank's Balanced Salt Solution (HBSS; Sigma) and added with 0.15% sulforhodamine B (SRB) (559 Da; Sigma) in HBSS. The cells were scraped using a pipette tip and incubated at 37°C for 15 min. After washing three times with HBSS, the cells were observed with an EVOS uorescence microscope (Thermo Fisher Scienti c). To determine the extent of dye transfer, the distance of dye spread from the scrape line was measured in ImageJ software. For each of wild-type, mutant, and mock plasmids, the dye spread distances in total 9 areas were measured and averaged to produce the graph represented in Fig. 5e. In each experiment, we rst turn on the blue uorescence detection mode (Excitation (Ex)/ Emission (Em) = 360/447 nm) in the EVOS FL microscope to select an area of con uent cells with high CFP signals around a scrape line, and then turn on the red uorescence detection mode (Ex/Em = 530/593 nm) to see SRB signals in the selected area. We took a picture of the SRB signals and measured the maximum distance of dye spread at the whole area using ImageJ program. We analyzed 3 areas for each experiment and performed three independent experiments.

Declarations Data availability
All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Information. Additional data related to this paper may be requested from the authors. The Cryo-EM density maps for Cx43-WT GJICh in detergents and nanodiscs, and Cx43-M257 GJICh in detergents have been deposited in the EM Data Bank (http://emdatabank.org/) under the accession codes EMD-31495 (fully GCN GJICh), EMD-31496 (fully GCN GJICh), and EMD-31497 (GJICh with fully GCN and fully PLN hemichannels), respectively. The atomic coordinates for Cx43-WT GJICh in detergents and nanodiscs, and Cx43-M257 GJICh in detergents have been deposited in the Protein Data Bank under accession codes 7F92 (fully GCN GJICh), 7F93 (fully GCN GJICh), and 7F94 (GJICh with fully GCN and fully PLN hemichannels), respectively.    Structural details of the GCN-to-PLN transition, π-helix formation in the middle of TM1 and the membrane opening between TMDs. a, Comparison between GCN and PLN conformations depicted with pore pathways. Only two protomers facing each other are shown for clarity. NTHs are colored in magenta. πhelix in the middle of TM1 is colored in black and labeled. b, Comparison between the solvent-accessible pore radii changing along the pore pathways of GCN and PLN hemichannels. c,d, Detailed structural  protomers are also shown as sticks and colored in magenta. c, Ionic currents and selectivities of three different GJICh conformations in a.

Figure 7
Identi cation of PLN protomers in Cx43-WT GJIChs at pH 6.9 and stabilization of the CL in the PLN conformation. a, Three representative 2D class averages of Cx43-WT GJICh at pH 6.9. b, Single protomerbased conformational variation analysis using the dataset of Cx43-WT GJICh at pH 6.9. The densities of NTH and CL in Class 2 in the PLN conformation are indicated by a black bold line and a green circle, respectively. c, Distribution of various compositions of PLN protomers within GJICh particles. d, Strong map densities of CLs observed in fully PLN hemichannel (orange) of Cx43-M257 GJICh, but not in fully