An electrostatic mechanism for Ca2+-mediated regulation of gap junction channels

Gap junction channels mediate intercellular signalling that is crucial in tissue development, homeostasis and pathologic states such as cardiac arrhythmias, cancer and trauma. To explore the mechanism by which Ca2+ blocks intercellular communication during tissue injury, we determined the X-ray crystal structures of the human Cx26 gap junction channel with and without bound Ca2+. The two structures were nearly identical, ruling out both a large-scale structural change and a local steric constriction of the pore. Ca2+ coordination sites reside at the interfaces between adjacent subunits, near the entrance to the extracellular gap, where local, side chain conformational rearrangements enable Ca2+chelation. Computational analysis revealed that Ca2+-binding generates a positive electrostatic barrier that substantially inhibits permeation of cations such as K+ into the pore. Our results provide structural evidence for a unique mechanism of channel regulation: ionic conduction block via an electrostatic barrier rather than steric occlusion of the channel pore.


Supplementary Figure 2. Comparison of Ca 2+ -bound, Ca 2+ -free, and 2ZW3 Cx26 gap junction channels.
Superposition of side views of the (a) GJC structures (represented in cartoon) and of (b) a single Cx subunit (represented in ribbon). In all panels, the Ca 2+ -bound model is shown in cyan, the Ca 2+ -free model in orange, and the 2ZW3 model in purple, with Ca 2+ ions shown as yellow spheres. Though the Cα RMSD of the TM -helices is modest (0.9 Å) and the pore dimensions are very similar, the largest divergence is found in the E1 and E2 loops, with maximum RMSDs of ~4 Å for the Cα atoms and ~8 Å for the side chains (averaged all-atom RMSD is 1.0 Å and 1.3 Å for the E1 and E2 loops, respectively). Specifically, there are two regions of significant backbone deviation between the models reported here and 2ZW3. One occurs near the Ca 2+ -binding site, at residues 42-46 (dashed box) (the other is residues 56-59, as described below). (c) A close-up of the dashed-box region from the overlay in (b) as viewed (top) from the pore or (bottom) from the extracellular gap. The Cα backbone is shown in ribbon representation, and the average main chain RMSD is 2.6 Å. Residue numbering is shown at the respective Cα atoms for each model. Although there are slight backbone differences between our Ca 2+ -bound and Ca 2+ -free structures in and around this region (mainly for residues 45-47), the average C RMSD is only 0.4 Å, substantially less than the deviation between the Ca 2+ -bound or Ca 2+ -free structures and 2ZW3. In addition, since this sequence forms part of the Ca 2+ binding site that we identified here and because there was a major side chain reconfiguration at E47, one expects main chain divergence in this region between the Ca 2+ -bound and Ca 2+ -free structures. (d) Overlay of Ca 2+ -bound, Ca 2+ -free, and 2ZW3 structures, viewed from the extracellular gap. Residues 56-59 in E1 show moderate backbone deviations between our structures and 2ZW3, with an average C RMSD of 1.6 Å (the C RMSD for the Ca 2+ -bound and Ca 2+ -free structures for this region is only 0.4 Å). In our structures the backbone bends inwards towards the pore and slightly down along the pore axis. L56 is at the apex of E1 at the hemichanneldocking interface and is shown as sticks. For the 2ZW3 model that was equilibrated via MD 4 , the apices of E1 at L56 turn inwards toward the pore. This creates a restriction at the hemichannel-docking interface not observed in any of the crystal structures determined so far. Perhaps this region of E1 may have greater conformational options in undocked hemichannels. (e) Overlay of the Ca 2+ -bound (cyan), Ca 2+ -free (orange) and 2ZW3 (purple) structures, viewed from the extracellular gap and shown at the level of the minimum pore diameter within the hemichannel, which is ~15Å between Lys41 residues. The side chains for the Lys41 residues are shown as sticks, with the minimum distance between Nζ atoms shown as a solid line.  indicate the approximate locations of the Ca 2+ binding sites in their respective hemichannels. In the absence of bound Ca 2+ (dotted curves, c-d), there is a slightly attractive potential for K + (c) and a slightly repulsive potential for Cl -(d) within the pore in the regions of the Ca 2+ binding sites. Though the pore structure is nearly identical in Ca 2+bound and Ca 2+ -free channels (a-b), upon Ca 2+ binding (solid curves, c-d) a large repulsive potential for K + is established throughout the pore (c) with a slightly attractive potential for Cl -(d). PMF calculations were based on the K + and Clion density distributions along the Ca 2+ -bound and Ca 2+ -free Cx26 GJC axes during 30 ns of the allatom MD simulation production phases. In the case of the Ca 2+ -bound simulation, a single K + ion entered the pore and passed the Ca 2+ binding sites of one hemichannel, where it became trapped ( Fig. 6c upper hemichannel). The presence of this trapped K + ion is responsible for the reduced K + PMF in the vicinity of the upper Ca 2+ binding site (at ~-30Å in (c)), and reflects the limitations of performing PMF calculations on relatively short simulations.

Supplementary Tables
Supplementary Table 1 The distances given are in Å and are listed as, in order, the RMS deviation, the average deviation, and the maximum deviation for a particular alignment pair.

Supplementary Discussion
The cytoplasmic domains of Cx26 The computational studies that we performed did not consider the roles of the NT, CT and CL in Ca 2+ -induced changes in permeation or gating, as these domains were not resolved in the crystal structures. To the extent that these regions shield, enter or interact with the pore, they would be expected to influence permeation in ways not addressed here. For instance, there is evidence that the NT senses voltage and may physically enter the pore 7-10 . If Ca 2+ binding alters the voltage profile, the occupancy of the voltage-sensing NT within the pore may also be altered, thereby affecting the pore-blocking effects of the NT. Intriguingly, several disease-causing mutations in the NT of Cx26 result in aberrantly open hemichannels with decreased sensitivity to extracellular Ca 2+ (ref. 11 ), empirically similar to some mutations in and near the identified Ca 2+ binding sites. This suggests an interaction between the NT and Ca 2+ sensitive processes, at least in hemichannels (see below), and perhaps in junctional channels as well.

Justification for independent structure determination
The first X-ray crystallographic structure of the Cx26 gap junction channel (PDB ID: 2ZW3) was solved in a heroic effort by the laboratory of Dr. Tomitake Tsukihara. The challenging and meticulous structure determination has been described in detail 12,13 . The obvious method for phasing our Cx26 X-ray crystallographic data was molecular replacement with the 2ZW3 model, and this is the method we used initially to solve the Ca 2+ -bound Cx26 structure. Ultimately, we were unable to model the cytoplasmic TM extensions or the N-terminus in either our Ca 2+ -bound or Ca 2+ -free structure.

Relevance of our results on gap junction channels to hemichannel physiology
Normal extracellular Ca 2+ keeps unapposed hemichannels in the plasma membrane in a predominantly closed/non-conductive state. We suspect that the Ca 2+ binding sites that we identified likely involve some or all of the same residues that regulate hemichannels. The basis for this assertion is that the Ca 2+ binding sites reside in a region that has been implicated across multiple connexin isoforms as a locus for hemichannel regulation by extracellular divalent cations, based on deafness-causing mutations that result in aberrantly open hemichannels and on experimental mutagenesis [14][15][16][17][18][19][20] . In addition, it has been proposed that a network of electrostatic interactions in this region 21 stabilizes a conductive state, and that specific electrostatic interactions are disrupted by extracellular Ca 2+ to destabilize it 17,18 . Furthermore, conformational changes associated with loop gating, a voltage gating process that can be modulated by divalent cations, map to this locus [22][23][24] . Altogether, this demonstrates the importance of the M1/E1 segment in the function and regulation of GJCs and hemichannels.
While the same or overlapping residues in GJCs and hemichannels may coordinate Ca 2+ , the structural and functional consequences of Ca 2+ binding may differ. For instance, it is possible that large-scale Ca 2+ -induced structural changes and/or localized effects on structural and electrostatic interactions in hemichannels are precluded upon docking of hemichannels. Previous AFM imaging of the extracellular surface of hydrated hemichannels showed significant Ca 2+induced narrowing of the pore 25 . However, we note that the hemichannels were created by force dissection of intact gap junction plaques, which may disrupt the protein conformation on the extracellular surface. In addition, metal-bridging experiments indicated no significant structural rearrangements and/or narrowing of the extracellular pore during loop-gate closure 4 , which involves much of the same segments of the protein. Admittedly, the relation between loop-gating and Ca 2+ gating is unclear 14 , and we realize that discernment of Ca 2+ effects on Cx hemichannels will require structures at high resolution.
Given these observations and provisos, we are not aware of any electrophysiological studies of wild-type Cx26 hemichannels that show evidence for a Ca 2+ -dependent transition to anion selectivity. Such a transition would be expected if the only effect of Ca 2+ binding were to decrease cation permeability. In the MD simulations reported here, we did observe residence of Clions in the Ca 2+ -free channel pore, and this population was enhanced in the Ca 2+ -bound channel. A logical extension from the simulations is that one might expect that the Ca 2+ -bound channel should allow Clconductance. In addition, due to the Ca 2+ -induced positive electrostatic surface potentials, the Clconductance may be greater than what is observed in the Ca 2+ -free channel. We note that in a junctional channel, the magnitude of the Clconductance would be very low due to the relatively low intracellular Clconcentration (~4 mM). However, in the case of hemichannels, the Clconductance might be significant because the extracellular end of the pore faces a milieu that includes ~120 mM Cl -. It is still possible that Ca 2+ binding to hemichannels results in occlusion of the pore by a physical gate, and state-dependent chemical accessibility studies suggest that this may be the case 22 to Ca 2+ at concentrations near millimolar. However, it is possible that there is much greater conformational flexibility in undocked hemichannels versus intact GJCs. Such flexibility may reduce the Ca 2+ binding affinity relative to that of the junctional channel. It is also possible that conformational flexibility in hemichannels may enable participation of additional nearby residues in Ca 2+ coordination.

Relation of our findings to other data on the location of sites for Ca 2+ binding and regulation
Given the experimental constraints of studying GJCs, all physiological and mutagenesis studies on the mechanism of direct regulation by Ca 2+ have been performed on unapposed hemichannels. Nearly all of these studies point to the pore-lining region of E1 that includes the Ca 2+ coordinating residues identified here (E47, G45, E42), or to the corresponding residues in other connexins, as the region at which Ca 2+ acts. It is notable that a near-contiguous group of deafness-causing point mutations resides in the Ca 2+ -binding region of Cx26 (Figs. 1a, 5f) 34 . We propose that such mutations may alter Ca 2+ binding and result in channelopathies.
In Cx26 hemichannels, electrophysiological studies combined with site-directed mutagenesis, chemical modification and thermodynamic mutant cycle analysis have identified residues and interactions that are involved in Ca 2+ sensing and the consequent gating changes.
The published data indicate that Ca 2+ sensitivity has a strong requirement for a negatively charged residue at position 50 (D50 in wild-type channels), which interacts with Q48 and/or K61 in an adjacent subunit in a Ca 2+ sensitive manner 17,18,35 . These studies do not identify these residues as directly coordinating the Ca 2+ but make clear that they are intimately involved in the molecular mechanism by which Ca 2+ gates the channels. Recent MD simulations have also implicated this region as a Ca 2+ binding site, identifying E42, D46, E47 and D50 as particularly important for coordination 36 .
An exception to the focus on residues within E1 as the site for Ca 2+ binding is the proposal that two aspartate residues (169 and 178) within E2 of adjacent subunits in Cx32 harbor the divalent cation-binding site; this is based on decreased Ca 2+ sensitivity of voltage dependence when the residues are mutated to asparagine 31 . However, this site could be specific to Cx32 since an anionic residue in the vicinity of position 169 is not present in most connexin isoforms, yet these are well regulated by extracellular Ca 2+ .