PIP2 induces the conformational change in Cx43 GJCh from GCN to PLN
In our previous structural study of Cx43 GJCh, we identified one condition (condition 8) in which the ratio of GCN, FIN, and PLN protomers was approximately 2:1:2 (37.3%, 22%, and 40.7%, respectively) (Fig. 1a, b), whereas the protomers in other conditions were mainly in the GCN conformation4. Under condition 8, the Cx43-M257 construct with a partial deletion of CTD was used; GJChs were reconstituted into lipid nanodiscs mainly containing 1-palmitoyl-2-oleoyl-sn-glycero-3-phosphoethanolamine (POPE); and the CHS content was largely reduced during nanodisc reconstitution and purification. As Cx43 GJCh was in its most dynamic conformational equilibrium under condition 8, we chose this condition to test the effects of regulatory factors on equilibrium. Notably, aliquots of the GJCh-nanodisc sample in condition 8 were cryo-cooled, stored at − 80 ℃, and used for both previous and current studies. This excludes the possibility of variations that may have occurred during sample preparation.
To examine the effect of PIP2 on the structural equilibrium of Cx43 GJCh, we used a water-soluble PIP2 analog, dioctanoylglycerol-PIP2 (diC8-PIP2), because PIP2 cannot be delivered to lipid nanodiscs without detergents. Detergents should be avoided because they affect the structure of GJChs and lipid nanodiscs by binding to GJChs and extracting lipids from the nanodiscs. diC8-PIP2 has shorter hydrophobic tails but an identical head group compared to PIP2 and has been effectively used to study the regulation of various ion channels by PIP228, 31.
The GJCh-nanodisc sample (5.9 mg/ml) in condition 8 was treated with diC8-PIP2 at a final concentration of 1.3 mM, which corresponded to an approximately eight-fold molar excess of the Cx43 protomer. After 10 min of incubation, the sample (referred to as the Cx43-PIP2-8x sample) was loaded onto cryo-EM grids and cryo-cooled. The high protein concentration of 5.9 mg/ml was required to obtain various particle orientations. Several rounds of 2D classification were conducted to remove aggregated, overlapping, or false-positive (such as hemichannel) particles. Three-dimensional classification and refinement produced a reasonably good consensus map at 3.06 Å resolution (Supplementary Figs. 1a, c, 2a). Using 26,825 particle images to reconstitute the consensus map, we performed a protomer-focused 3D classification to determine the ratio of the different NTH conformations (Fig. 1a). Of the resulting five classes, three classes, accounting for 62.3% of the particles, showed the PLN conformation, whereas one class, including 20.4% of the particles, exhibited the GCN conformation (Fig. 1c). This indicates that 1.3 mM diC8-PIP2 increased the PLN population and decreased the GCN population by approximately 20%.
Because a substantial portion of the protomers were still in the FIN or GCN conformations in the presence of 1.3 mM diC8-PIP2, we almost doubled the diC8-PIP2 concentration to produce the Cx43-PIP2-16x sample and performed a similar cryo-EM experiment to obtain the consensus map at 3.01 Å resolution (Supplementary Figs. 1b, d, 2b). In the protomer-focused 3D classification conducted with 5,720 particle images, we found no protomer classes in the GCN conformation and three classes in the PLN conformation, accounting for 71.6% (Fig. 1d). Therefore, we concluded that diC8-PIP2 shifted the conformational equilibrium from GCN to PLN in a concentration-dependent manner. However, the percentage of protomers in the FIN conformation (FIN protomers) was not reduced by high diC8-PIP2 concentration, suggesting that Cx43 GJCh remains in dynamic equilibrium between PLN and FIN, even in the presence of PIP2.
In the previous structural data without PIP2, strong pore-occluding densities were observed in a 3D structural class that mainly contained GCN protomers (Supplementary Fig. 3e-g). In contrast, the GCN conformation almost disappeared in the Cx43-PIP2-16x data, and the C1 consensus map, reconstructed without symmetry imposition, showed much weaker pore-occluding densities (Supplementary Fig. 3a-d). This suggests that pore-occluding lipids diffuse out of the pores during conformational changes from GCN to PLN after treatment with diC8-PIP2.
Structure of Cx43 GJCh in complex with PIP
Because diC8-PIP2 increases the PLN protomer population, this small molecule likely binds to and stabilizes the PLN conformation. To identify the map density of diC-PIP2, we determined the structure of Cx43-PIP2-16x GJCh in full PLN conformation. Because the Cx43-PIP2-16x dataset contained approximately 30% FIN protomers, we conducted 3D classification using a mask covering the protein to increase conformational homogeneity. While one class, including 77.4% of the GJCh particles, showed clear PLN density, the other class did not, suggesting that GJChs containing more FIN protomers were effectively separated from those containing more PLN protomers. We selected the former class for the ensuing refinement process and obtained high quality map density at 3.01 Å resolution (Fig. 2a and Supplementary Figs. 1d, 2c, d).
In the refined map of the GJCh nanodisc with D6 symmetry, we identified strong densities with clear features of the PIP head group, including an inositol 1,4,5-trisphosphate and a glycerol backbone at large gaps (membrane openings) between neighboring protomers (Fig. 2a-c). Such densities were not observed at the same position in the previous structure of the PLN conformation under diC8-PIP2-free conditions, confirming that the densities originated from the diC8-PIP2 added to the purified GJCh nanodisc sample (Supplementary Fig. 4a, b). Based on the clarity and connectivity of diC8-PIP2 density, we built a ligand model that included the entire head and parts of the two tails.
Cx43 GJCh in the GCN conformation has no membrane opening between the protomers; thus, PIP2 does not bind to the channel in this conformation. The conformational change from GCN to PLN is accompanied by the α-π transition in the middle of TM1 that creates the membrane opening. Therefore, PIP2 binding to the channel likely inhibits the reverse transition (PLN to GCN) by preventing the closure of the membrane opening.
Structure of CL stabilized by PIP binding
We previously found that the structure of Cx43 GJCh in the PLN conformation in the absence of PIP2 showed a large density blob next to the cytoplasmic protrusion of TM2 (residues E103–K114). However, the density was too low to determine its identity. In the structure of Cx43-PIP2-16x, the corresponding density was stronger, was a long chain of approximately 25 amino acids, and showed a clear feature of an α-helix extending parallel to the cytoplasmic protrusion of TM2 (Fig. 2a-c). In addition, one end of this density was connected to the N-terminus of TM3, and the other end was located close to the C-terminus of TM2. Because the CL was a 36-amino-acid-long polypeptide connecting TM2 and TM3 and had an α helix region in the secondary structure prediction, we were confident that the density was a part of the CL (Fig. 2d).
The main-chain connectivity of the density was clear enough to build a structural model of the CL, although most of the side chain densities were unclear. Nevertheless, Y137 at the C-terminus of the α-helical region (residues from H126 to Y137; referred to as CL-α) showed a clear side chain density and thus was a landmark for correct amino acid registration (Fig. 2c and Supplementary Fig. 5). The final model exhibited reasonable stereochemistry. For examples, 1) the CL formed an extensive hydrophobic interaction network with TM2 including the interaction of L127, I130, Y137, I139, and V145 in CL with V96, V99, M100, K102, L106, and L113 in TM2 (Fig. 3a, b), 2) K134 in CL-α participated in ionic interactions with E103 and E110 in TM2 (Fig. 3c), and 3) M147 was buried in a hydrophobic pocket formed by T19, L152, Y155, and V231 (Fig. 3d).
We also performed computational prediction of a hexameric Cx43 structure using AlphaFold232, 33 to identify the evolutionarily conserved interactions of CL with TM2. As the CL structures in the five top-ranked models were almost identical, we selected the first model and examined its per-residue confidence score (pLDDT) (Supplementary Fig. 6a-c). Notably, pLDDT values for the residues in CL-α and the cytoplasmic protrusion of TM2 were more than 80, indicating that the interaction between these two helices in the predicted model has high confidence. These regions were superposed onto the corresponding region in the experimental model with the average Cα RMSD of 2.1 Å (Supplementary Fig. 6d), and the amino acid positions in the regions were identical between the two models. These data suggest that the interaction between CL-α and TM2 is evolutionarily important.
The loop structure for the residues G138-G160 in the CL (referred to as CL-β) showed substantial difference between the predicted and experimental models (Supplementary Fig. 6d). The pLDDT values for this region were relatively low, suggesting that the interactions of these residues with other residues in the protein are not evolutionarily conserved (Supplementary Fig. 6b, c). In the experimental model, these two lysine residues interact with the 5' phosphate group on the inositol headgroup of diC8-PIP2, accounting for how PIP2-binding contributes to the stabilization of the CL structure (Fig. 4c, d).
We found no interaction of the CL with NTH or the NTH-TM1 loop except a close interaction of M147 with T19 and a distant interaction of V145 with Y17 at the inter-atomic distance of 6.6 Å (Fig. 3e), suggesting that the CL might not contribute much to the formation of PLN. However, the location of CL-β in the PLN conformation was largely overlapped with that of NTH in the GCN conformation, and thus the structured CL could contribute to the maintenance of PLN by inhibiting the formation of GCN (Fig. 3f).
Detailed interaction of Cx43 GJCh with diC8-PIP
In the structure of Cx43-PIP2-16x GJCh, the PIP2-binding site is formed by TM1, TM4, and CL-β from a Cx43 protomer and TM2 from the adjacent protomer. The cytoplasmic region of the binding site had highly basic surfaces created by seven lysine residues and one arginine (Fig. 4a). Six lysine residues interact with 4' and 5' phosphate groups on the inositol head, which are the unique feature of PIP2 distinct from other phospholipids in human cells (Fig. 4c, d). Especially, K146 in CL-β and K105 in TM2, closely interact with 5′ and 4′ phosphates, respectively, within a distance of 3.2 Å. In addition, three amino groups from K102 and K109 in TM2 and K237 in TM4 could be located within approximately 5 Å from 4' phosphate and thus may contribute to the binding affinity for PIP2. The K102N and K144E mutations cause ODDD34, 35, suggesting that normal channel function might require a strong interaction between Cx43 GJCh and PIP2.
K109 and K234 are highly conserved basic residues in the human connexin family, and K105, K144, and K146 are moderately conserved (Fig. 4e), suggesting that other connexin GJChs may have similar PIP2-binding sites. In addition, although the length and amino acid sequence of the CL domain are highly diverse among human connexins, the conservation of K144 and K146 suggests that PIP2 may induce conformational changes in the CL domains of other connexins to regulate various GJChs.
The 1′ phosphate group on the inositol head interacts closely with R101 and distantly with K23 (Fig. 4c, d). The glycerol backbone and the second carbon atoms in the hydrocarbon tails participated in hydrophobic interactions with A20, Y98, and Y230 (Fig. 4b). As phosphate and glycerol groups are shared by all phosphatidyl lipids, these interactions are not specific to PIP2. Although R101 is completely conserved only in A-class connexins, the other four residues are highly conserved in the human connexin family: K23 residue as lysine or arginine residues, A20 as hydrophobic amino acids, and Y98 and Y230 as tyrosine or histidine residues (Fig. 4e). Therefore, the interaction of these residues with phosphatidyl lipids at the membrane opening may be important for the function or regulation of connexin GJChs.