Opposing modulation of Cx26 gap junctions and hemichannels by CO2

A moderate increase in PCO2 (55 mmHg) closes Cx26 gap junctions. This effect of CO2 is independent of changes in intra‐ or extracellular pH. The CO2‐dependent closing effect depends on the same residues (K125 and R104) that are required for the CO2‐dependent opening of Cx26 hemichannels. Pathological mutations of Cx26 abolish the CO2‐dependent closing of the gap junction. Elastic network modelling suggests that the effect of CO2 on Cx26 hemichannels and gap junctions is mediated through changes in the lowest entropy state of the protein.


Introduction
The canonical function of connexins is to form intercellular junctions between cells -gap junctions -through the docking of hexameric connexons in the opposing membrane of each cell. However, the individual connexons -known as hemichannels -can also function on their own (Stout et al. 2004;Weissman et al. 2004;Pearson et al. 2005;Huckstepp et al. 2010b). Hemichannels act as plasma membrane channels, which, in addition to mediating transmembrane ionic currents, permit the transmembrane fluxes of small molecules such as ATP (Stout et al. 2002;Pearson et al. 2005;Kang et al. 2008;Huckstepp et al. 2010a).
By means of electrophysiological and dye-loading assays, we have studied the hemichannels of connexin26 (Cx26) and demonstrated that they are directly gated by CO 2 (Huckstepp et al. 2010a;Meigh et al. 2013Meigh et al. , 2014. CO 2 opens the hemichannel and permits the efflux of ATP, which can act as a neurotransmitter. This is particularly important in the CO 2 -sensitive control of breathing where Cx26 hemichannels in the medulla oblongata act as novel chemosensory transducers (Huckstepp et al. 2010a;Huckstepp & Dale, 2011). Cx26 hemichannels also impart CO 2 sensitivity to dopaminergic neurons of the substantia nigra and GABAergic neurons of the ventral tegmental area (Hill et al. 2020). The action of CO 2 on Cx26 has been proposed to occur via carbamylation of the residue K125 (Meigh et al. 2013). CO 2 carbamylation, sometimes called carboxylation, involves formation of a covalent bond between the nitrogen atom of a primary amine (in this case the lysine side chain) and the carbon atom of CO 2 . The resulting carbamylated amine bears a negative charge distributed over the oxygen atoms (and essentially converts the lysine side chain from a basic to an acidic moiety). This reaction differs from carbamylation involving cyanate, which results in a very stable modification of the amine. Unlike this latter reaction, CO 2 carbamylation is a highly reversible post-translational modification (Lorimer, 1983;Meigh, 2015).
The carbamylated lysine forms a salt bridge to R104 of the neighbouring subunit. These intersubunit carbamate bridges bias the hemichannel to the open configuration (Meigh et al. 2013). Extensive evidence supports a direct action of CO 2 on the hemichannel rather than an indirect effect via pH: (1) in isolated inside-out or outside-out patches, changes in P CO 2 at constant pH alter Cx26-gating (Huckstepp et al. 2010a); (2) insertion of the carbamylation motif into Cx31, which is insensitive to CO 2 , creates mutant Cx31 hemichannels that can be opened by CO 2 (Meigh et al. 2013); (3) mutation of the key residues K125 and R104 to respectively arginine and alanine destroys CO 2 sensitivity of the hemichannel; (4) the mutations K125E and R104E (in effect engineering the action of CO 2 into the subunit via the carboxy group of glutamate) create constitutively open hemichannels which are CO 2 -insensitive (Meigh et al. 2013); (5) the mutation K125C creates a hemichannel that can be opened with NO or NO 2 via a nitrosylation reaction on the cysteine residue at position 125 and subsequent salt bridge formation to R104 ; and (6) the double mutation K125C and R104C creates a redox-sensitive hemichannel presumably via disulfide bridge formation between the cysteine residues at positions 104 and 125 . Based on this understanding of how CO 2 modulates the Cx26, we have created a Cx26 subunit with the combined mutations K125R and R104A. This doubly mutated Cx26 subunit coassembles with wild-type Cx26 and removes sensitivity from the resulting heteromeric hexamer and therefore has a dominant negative action on the CO 2 sensitivity of wild-type Cx26 (van de Wiel et al. 2020). By expressing this doubly mutated subunit in glial cells of a confined target area of the medulla oblongata (the caudal parapyramidal area), we have demonstrated that CO 2 binding to Cx26 contributes about half of the adaptive ventilatory response to hypercapnia that is controlled by central chemosensors (van de Wiel et al. 2020). Cx26 also plays a role in mediating CO 2 -dependent modulation of the excitability of neurons in the substantia nigra and ventral tegmental area (Hill et al. 2020). We have also recently demonstrated that K125 is indeed carbamylated and produced new high resolution cryoEM structures of Cx26 at different levels of P CO 2 (Brotherton et al. 2020). Thus, carbamylation of Cx26 occurs and is involved in mediating the CO 2 -dependent control of physiologically important functions.
Although the actions of CO 2 on Cx26 hemichannels are well characterized, we have not studied the action of CO 2 on Cx26 gap junctions. When the two connexons dock to form a complete gap junction, there is likely to be significant conformational rearrangement and constraint of the resulting dodecameric complex. Therefore, we cannot assume that CO 2 will modulate a complete gap junction in the same way as a hemichannel. There has been a previous study on the closing effect of CO 2 on Cx32 and Cx26 gap junctions expressed in Xenopus oocytes (Young & Peracchia, 2004). This prior study used non-physiological conditions in which both the extracellular and intracellular pH would become very acidic: exposure to 30-100% CO 2 in the absence of bicarbonate in the extracellular medium. This study most likely reported an effect of pH on the gap junction. In this paper, we report the actions of much lower physiological concentrations of CO 2 (ß9%), in a CO 2 /HCO 3 − -buffered system at constant extracellular pH, on gap junctions formed between pairs of HeLa cells expressing Cx26. We find that modest increases in P CO 2 close complete gap junctions, and that this is most likely a direct effect mediated through CO 2 binding to the same residues that result in the opening of the hemichannel. This result reinforces the need to develop further high-resolution structures for Cx26 hemichannels and gap junctions with and without CO 2 bound.

HeLa cell culture and transfection
HeLa DH (ECACC) cells were grown in Dulbecco's modified Eagle's medium supplemented with 10% fetal bovine serum, 50 μg ml −1 penicillin-streptomycin and 3 mM CaCl 2 . For electrophysiology and intercellular dye transfer experiments, cells were seeded onto coverslips in six-well plates at a density of 2 × 10 4 cells per well. After 24 h the cells were transiently transfected with Cx26 constructs tagged at the C-terminus with a fluorescent marker (mCherry) according to the GeneJuice Transfection Reagent protocol (Merck Life Science UK Limited, Gillingham, Dorset, UK). The HeLa cells stably expressing mouse Cx26 were originally obtained from Dr K. Willecke (Elfgang et al. 1995).

Cx26 mutants
The mutations used in this study were introduced into the Cx26 gene by QuikChange site directed mutagenesis and have been described previously (Meigh et al. 2013;Cook et al. 2019).
Imaging assay of gap junction transfer 2-Deoxy-2-[(7-nitro-2,1,3-benzoxadiazol-4-yl) amino]-D-glucose (NBDG) was included at 200 μM in the patch recording fluid, which contained: potassium gluconate 130 mM; KCl 10 mM; EGTA 5 mM; CaCl 2 2 mM, Hepes 10 mM, pH adjusted to 7.3 with KOH and a resulting final osmolarity of 295 mOsm. Cells were imaged on a Cleverscope (MCI Neuroscience, Framfield, East Sussex, UK) with a Photometrics Prime camera under the control of Micromanager 1.4 software. LED illumination (Cairn Research, Faversham, UK) and an image splitter (Optosplit, Cairn Research) allowed simultaneous imaging of the mCherry-tagged Cx26 subunits and the diffusion of the NBDG into and between cells. Coupled cells for intercellular dye transfer experiments were initially selected on the basis of tagged Cx26 protein expression and the presence of a gap junctional plaque, easily visible as a band of mCherry fluorescence (e.g. Figs 1 and 2). After establishing the whole cell mode of recording, images were collected every 10 s.

Patch clamp recordings from coupled cells
Coverslips containing non-confluent cells were placed into a perfusion chamber at room temperature in sterile filtered standard aCSF. Two Axopatch 200B amplifiers (Molecular Devices, San Jose, CA, USA) were used to make whole-cell recordings from pairs of HeLa cells. The intracellular fluid in the patch pipettes contained: potassium gluconate 130 mM, KCl 10 mM, EGTA 10 mM, CaCl 2 2 mM, Hepes 10 mM, sterile filtered, pH adjusted to 7.3 with KOH. An agarose salt bridge was used to avoid solution changes altering the potential of the Ag-AgCl reference electrode. All whole-cell recordings were performed at a holding potential of −50 mV. Steps to −40 mV were applied to each cell in alternation to measure the whole cell and gap junction conductances. A similar protocol has been previously used by others (Wang et al. 2003). The current from the cell at −40 mV to the cell at −50 mV flows through the gap junction and can thus be used to calculate the gap junction conductance. The outward current during the +10 mV step in each cell represents the whole cell conductance, which is a combination of all current pathways out of the cell. These comprise any intrinsic conductances and also the gap junction conductance itself.

Elastic network modelling
Elastic network model (ENM) simulations were performed based on the regular implementation using PDB file 2ZW3, where all the Cα atoms in the protein within a given cut-off radius (8Å) were joined with simple Hookean springs (Tirion, 1996;Rodgers et al. 2013a). An 8Å cut-off radius was selected based on previous observations for the optimal correlation value for B factors between the calculated ENM and a crystal structure (Townsend et al. 2015). The spring constants were set to a constant value of 1 kcal mol −1Å−2 . The presence of CO 2 was represented in the ENM by the inclusion of an additional Hookean spring between residues K125 and R104 of each set of neighbouring monomers, following the same procedure previously used for ligand binding (Rodgers et al. 2013b). The mass-weighted second derivatives of this potential energy matrix were diagonalized to produce eigenvectors, e, which are the normal modes, and eigenvalues which, are the squares of the associated frequencies, ω, and can be used to calculate the free energy of each mode.
The first six modes, that is the lowest frequency modes, represent the solid body translational and rotational motions of the protein and are thus excluded from the analysis. The overlap of the modes in the unbound and CO 2 -bound states were calculated by comparison of the eigenvectors (Rodgers et al. 2013a). A value of 1 indicates that the motions are identical whereas a value of 0 indicates that the motions are completely different.

Statistics and reproducibility
Statistical analysis was performed with the R language. Data have been plotted as box and whisker plots, with individual points superimposed, where the box is interquartile range, bar is median and whisker extends to most extreme data point that is no more than 1.5 times the interquartile range. Each individual point is from a single patch clamp experiment and is counted as a replicate. Statistical comparisons were performed with the Kruskal-Wallis test (for multiple comparisons) or Mann Whiney U test.
Exact P values are shown for comparisons that showed a significant difference between samples. Figure 1. Dye transfer assay to assess the CO 2 sensitivity of gap junction coupling A, logic of the assay: a CO 2 -sensitive gap junction will display no or very little dye transfer from the donor to acceptor cell under conditions of high P CO 2 . This will occur once the cells are transferred to a low P CO 2 saline. A CO 2 insensitive gap junction will exhibit equally rapid dye transfer at low and high P CO 2 . B, images showing dye transfer between coupled cells at two different levels of P CO 2 . Numbers in lower right corner are recording time in minutes. Image at 0 min in both rows is a merge of differential interference contrast (DIC) and mCherry (for Cx26, red) and NBDG fluorescence (green) and is the beginning of the recording. The gap junction can be observed as a red stripe between the coupled cells (yellow arrows, both rows). Subsequent images show just the NBDG fluorescence. In 35 mmHg P CO 2 dye transfer to the acceptor cell is evident after 1 min (white arrows). Starting the recording at a P CO 2 of 55 mmHg and then transferring to 35 mmHg saline after 2 min greatly slows dye transfer; fluorescence in the acceptor cell is only seen after 10 min (white arrows). Scale bar is 20 μm. C, summary data for stably expressing (mouse Cx26, n = 9 and 18 for low and high CO 2 , respectively) and transiently transfected (human Cx26 tagged with mCherry, n = 12 for low and high CO 2 ) HeLa cells, showing the time for dye in the acceptor cell to reach 10% of the donor starting in 35 mmHg, and starting in 55 mmHg but transferring to 35 mmHg after 2 min. When the recordings are commenced in 55 mmHg the dye transfer time is much longer. Statistical comparisons with Mann-Whitney U test.

Rate of dye transfer between coupled cells shows CO 2 dependence of Cx26 gap junctions
Many investigators have used dye transfer assays to demonstrate the presence of gap junctions (Spray et al. 1991;Elfgang et al. 1995;Abbaci et al. 2007). We therefore adapted this type of assay to test whether increases in P CO 2 could alter the permeation of a fluorescent glucose analogue (NBDG) through Cx26 gap junctions formed between HeLa cells. Starting with HeLa cells that stably expressed untagged Cx26, we made whole cell recordings from a single cell in a coupled pair. After introducing NBDG, via the patch pipette, into one of the cells (the donor) of a coupled pair, we recorded the time taken for the dye to diffuse into the coupled (acceptor) cell and achieve 10% of the fluorescence intensity of the donor cell ( Fig. 1A). If the gap junction is sensitive to CO 2 , then permeation of NBDG from the donor to acceptor cell should be altered by elevated P CO 2 (Fig. 1A). By recording from pairs of HeLa cells that stably expressed Cx26, we found that NBDG rapidly permeated into coupled cells in solutions with a P CO 2 of 35 mmHg (Fig. 1C). When the recording was initiated in a saline with a P CO 2 of 55 mmHg, there was little permeation during the period of high P CO 2 and this only occurred once the P CO 2 was reduced to 35 mmHg (Fig. 1C). We refined this assay by performing it on cells that had been transiently transfected with Cx26 tagged with mCherry. This allowed us to directly visualize the gap junctions and thus select pairs of coupled cells for the assay (Fig. 1B). Once again, we found that the level of P CO 2 altered the time it took for NBDG to permeate from the donor to the acceptor cell ( Fig. 1B and C).
Having established the validity of the fluorescence assay to detect CO 2 -sensitive intercellular coupling, we next investigated whether the CO 2 sensitivity of Cx26 gap junctions depended on the same residues (K125 and R104) that are necessary for the CO 2 sensitivity of the Cx26 hemichannel (Meigh et al. 2013). We examined the effect of two mutations K125R and R104A that individually remove CO 2 sensitivity from the hemichannel. Cx26 K125R and Cx26 R104A formed gap junctions between HeLa cells ( Fig. 2A) that were readily permeated by NBDG. Our data showed that unlike wild-type Cx26 gap junctions, CO 2 made no difference to the rate of dye transfer between cells coupled via Cx26 K125R or Cx26 R104A ( Fig. 2A and B). Unlike the Cx26 WT , where elevated P CO 2 slows permeation of NBDG to the coupled cell, permeation of NBDG between cells coupled by Cx26 K125R or Cx26 R104A is apparently unaffected by P CO 2 . White arrows indicate appearance of NBDG in the coupled cell, yellow arrows highlight the gap junction structure; scale bar: 20 μm; numbers in bottom right corners indicate time of image in minutes after start of recording. B, summary data showing that rat Cx26 WT gap junctions (n = 4, 35 mmHg and n = 5, 55 mmHg) are CO 2 sensitive (55 mmHg P CO 2 delays the passage of dye across the gap junction), but those comprising rat Cx26 K125R (n = 6, 35 mmHg and n = 8, 55 mmHg) or rat Cx26 R104A (n = 6, 35 and 55 mmHg) show no CO 2 -dependence in the time required for dye transfer from the donor to acceptor cell. Kruskal-Wallis test χ 2 = 12.26, df = 5, p = 0.031; Mann-Whitney U test WT 35 mmHg vs. WT 55 mmHg, p = 0.008. J Physiol 599.1 CO 2 -dependent gap junction closure appears to require the same residues that are essential for hemichannel opening.

Electrical coupling via Cx26 gap junctions is sensitive to CO 2
To confirm the observations made during intercellular dye transfer assays, we made simultaneous whole cell recordings from isolated pairs of Cx26-expressing HeLa cells that were in close apposition, as these cells had a high probability of being electrically coupled. On establishing whole cell recordings from each cell, the cells were clamped at a holding potential of −50 mV. We used a simple protocol of repeated 10 mV steps, alternating in each cell (see Methods). In all the illustrations, the downward currents during the voltage steps are proportional to the junctional conductance. Untransfected parental HeLa cells sometimes exhibited gap junction coupling but this was insensitive to CO 2 (median change in conductance −0.03 nS (95% CI, −0.1 to 0.11), n = 10; percentage change shown in Fig. 3A and B; absolute conductances given in Table 1). In cells that stably expressed mouse Cx26, application of hypercapnic saline (P CO 2 55 mmHg) at a constant extracellular pH of 7.5 caused a reduction of junctional coupling ( Fig. 3A and B, Table 1). In some cases, this was a partial effect and in other cases the coupling between cells was completely abolished. Overall the median change of the gap junction conductance (from control to 55 mmHg P CO 2 ) was −5.6 nS (95% CI, −11.9 to −3.2 nS) (n = 29). The junctional current had a linear current voltage relation and reversed around 0 mV, as would be expected from identical recording solutions used for each cell of the coupled pair (Fig. 4, n = 3).
We next tested the requirement for the carbamylation motif for gap junction closure by CO 2 in the electrophysiological assay. Gap junctions formed between HeLa cells stably expressing Cx26 K125R tagged with mCherry were insensitive to a P CO 2 of 55 mmHg (Figs 3A and B; median change in conductance 0.01 nS (95% CI, −0.06 to 0.2 nS), n = 14, Table 1). Furthermore, gap junctions of Cx31 (stably transfected into HeLa cells and tagged with mCherry), which lacks the carbamylation motif (Meigh et al. 2013), were also insensitive to a P CO 2 of 55 mmHg ( Fig. 3A and B; median change in conductance 0.07 nS (95% CI, −0.29 to 0.44 nS), n = 12, Table 1).
We have therefore used two independent methods to demonstrate that Cx26 gap junctions are closed by In each case, both cells were voltage clamped at −50 mV. On top of the holding potential, alternating +10 mV (1.25 s in duration) steps were applied to each cell. The upward currents represent the whole cell current, which will include the gap junction current; the downward currents represent current flow from the cell at −40 mV to the cell at −50 mV and is thus current flow through the gap junction. Untransfected HeLa cells can exhibit coupling, but this is not sensitive to CO 2 . Gap junctions formed between HeLa cells stably expressing mouse Cx26 WT are rapidly and completely blocked by CO 2 . Gap junctions formed by rat Cx26 K125R are unaffected by CO 2 , as are rat Cx31 gap junctions. Bars represent the application of saline with 55 mmHg P CO 2 . B, summary data showing the percentage decrease caused by CO 2 on gap junction coupling between untransfected (parental) HeLa cells (n = 10), and HeLa cells expressing Cx26 WT (n = 29), Cx26 K125R (n = 14) or Cx31 (n = 12). Kruskal-Wallis test: χ 2 = 48.1, df = 3, p = 2 × 10 −10 . [Colour figure can be viewed at wileyonlinelibrary.com] The values are presented as the median (95% CI).
moderate changes in CO 2 (a P CO 2 of 55 mmHg) and this CO 2 -dependent closure depends upon the same residues (K125 and R104) that are required for hemichannel opening by CO 2 .

CO 2 -dependent and pH-dependent closure of Cx26 gap junctions can be dissociated by the K125R mutation
A possible mechanism by which CO 2 could close the gap junction is via intracellular acidification. CO 2 is known to permeate membranes and, by combining with water, can acidify the intracellular milieu. In prior studies, very high levels of CO 2 (30-100%) have been used as a pharmacological method to close a variety of gap junctions (Spray et al. 1991;Young & Peracchia, 2004). We therefore tested whether intracellular acidification, imposed independently of CO 2 by application of 30 mM propionate (Jahromi et al. 2002;Haussig et al. 2008), could close the Cx26 gap junction, and whether this could be altered by the K125R mutation, which eliminates CO 2 -dependent closure. This level of propionic acid will cause far greater intracellular acidification than that caused by the elevated CO 2 used in this study (Cook et al. 2012). Propionic acid treatment reduced the gap junction conductance of wild-type Cx26 by a median of 5.3 nS (95% CI, 2.8-8.3 nS), n = 12, Fig. 5). This represents a median reduction of 41.7% (95% CI, 26.6-53.7%). Propionate (30 mM) also reduced the gap junction conductance of Cx26 K125R by a similar amount: median 7.7 nS (95% CI, 4.1-11.0 nS), n = 13, Fig. 5). This represents a median reduction of 48.1% (95% CI, 28.0-86.3%). The actions of CO 2 and intracellular pH on the Cx26 gap junction conductance are thus of similar magnitude, but mechanistically independent. Specifically, pH-induced gap junction closure does not require K125, whereas CO 2 -dependent closure does require this residue. This in turn suggests that CO 2 has a direct action on Cx26, most probably via the carbamylation reaction that we have proposed for the opening of hemichannels (Meigh et al. 2013), to cause closing of the gap junction.

Effect of KID syndrome mutations on CO 2 -dependence of gap junction coupling
We have previously shown that KID syndrome mutations A88V, N14K and A40V (Fig. 6) abolish the ability of CO 2 to open the mutant hemichannels (Meigh et al. 2014;de Wolf et al. 2016;Cook et al. 2019). Recently, we discovered that certain KID syndrome mutations induce alternative splicing of the Cx26 mutation when this is tagged with a fluorescent protein ). This splicing results in poor expression and cell death but can be prevented by also mutating the 5 splice site ). However, the 5 splice site cannot be silently mutated, so we chose a conservative mutation, M151L (Fig. 6), which appeared to have little effect on Cx26 expression by itself. Cx26 M151L hemichannels are blocked normally by extracellular Ca 2+ and they retain CO 2 sensitivity . Furthermore, Cx26 A40V,M151L hemichannels are insensitive to CO 2 . When the M151L mutation was combined with the KIDS mutations, A40V, A88V or N14K (Fig. 6), in each case, the doubly mutated Cx26 was invariably capable of forming functional gap junctions through which NBDG could permeate (Fig. 7). For all three mutations, the Cx26 gap junction remained open in the presence of 55 mmHg P CO 2 (Fig. 7), a concentration that would normally close the wild-type gap junction (Figs 1-3). Thus, the KIDS mutations, when combined with M151L, not only prevent CO 2 -dependent opening of the hemichannel, but also prevent CO 2 -dependent closing of the gap junction. Interestingly, the effect of N14K on hemichannel sensitivity to CO 2 is less than those of the other KIDS mutations (de Wolf et al. 2016) and this parallels the trend in our data that permeation of NBDG through the Cx26 N14K,M151L gap junctions, while still occurring, may be slightly slowed at the higher level of P CO 2 ( Fig. 7C and D).
As the mutation M151L is a rare allele associated with non-syndromic hearing loss (Siemering et al. 2006), we tested whether this mutation by itself affected gap junctions. Cx26 M151L permitted gap junction formation as gap junction plaques were clearly visible between expressing cells (Fig. 8A). However, in 47 different recordings we did not observe either dye coupling (Fig. 8B) J Physiol 599.1 or the electrophysiological hallmarks of electrical coupling (slowed capacitive charging currents during a voltage step indicative of current flow through the gap junction). By contrast, cells expressing Cx26 A40V,M151L , Cx26 A88V,M151L and Cx26 N14K,M151L exhibited dye coupling in 14/15, 12/12 and 12/12 recordings respectively. These frequencies of coupling are statistically different χ 2 = 82.2 (3 degrees of freedom, p = 5.1 × 10 −18 ). We conclude that the mutation M151L prevents permeation through gap junctions that are apparently formed but either remain closed at the transjunctional potential studied here or are otherwise

. Junctional current through Cx26 is linear and reduced by CO 2 at all transjunctional voltages
A, raw current and voltage traces. A series of voltage steps from −80 to +10 mV in 10 mV steps were applied to cell 1 (V 1 ). The resulting currents were observed in cell 2 (I 2 ). Both cells were held at a holding potential of −50 mV. The current in cell 2 during the voltage steps in cell 1 represents the junctional current. Cell 2 also had periodic −10 mV steps to measure its whole cell conductance. A P CO 2 of 55 mmHg reduced the junctional currents (grey line) compared to the control (35 mmHg, black line). The whole cell conductance of cell 2 was also reduced in proportion to the reduction of the gap junction conductance by elevated P CO 2 (indicated by inward currents during regular −10 mV steps, V 2 ). This reduction in whole cell conductance occurs because the gap junction represents a resistance pathway out of the cell. B, plot of junctional current (I j ) versus transjunctional voltage (V j ) demonstrating a linear relation and the change in slope due to the action of CO 2 reducing the gap junction coupling. non-functional. This finding further implies that the KIDS mutations appear to compensate for the deficiency in gap junction permeability introduced by the M151L mutation.

Elastic network modelling
Previous experimental data point to the importance of carbamylation of K125 and the formation of a salt bridge to R104 in the adjacent subunit to facilitate Cx26 hemichannel opening in response to CO 2 . It is not clear from this mechanism alone why the dodecameric Cx26 gap junction would be closed by CO 2 . Previous work has used coarse-grained modelling to demonstrate a mechanism whereby CO 2 constrains the Cx26 hemichannel in the open state (Meigh et al. 2013). We therefore used further coarse-grained modelling to probe the difference in behaviour of the Cx26 hemichannel and gap junction. Coarse-grained modelling reduces protein atomistic complexity for efficient computational studies of harmonic protein dynamics and is particularly suited to examining the contribution of entropy to channel opening over millisecond time scales (Sherwood et al. 2008). While it is not possible to be certain that such calculated dynamics are true in the absence of experimentally determined structures, coarse-grained modelling has helped to support and explain structural data for membrane protein conformational changes (Shrivastava & Bahar, 2006;Sherwood et al. 2008;Zheng & Auerbach, 2011;Isin et al. 2012).

Figure 5. Intracellular acidification closes Cx26 gap junctions and is independent of the carbamylation motif
A, sample recordings of gap junctions formed between cells transiently expressing human Cx26 WT or human Cx26 K125R tagged with mCherry. Alternating +10 mV steps were applied to each cell; the downward currents represent the flow of current through the gap junction to the coupled cell. Application of 30 mM propionate (bar) caused a reduction in the gap junction currents, which was similar for Cx26 WT and Cx26 K125R  In an elastic network model (ENM) the Cα-atom coordinates of an atomic resolution structure are used to represent a protein structure. Global protein harmonic motions within the ENM consist of a defined number of modes, each of a characteristic frequency and representing a particular harmonic motion within the protein.
ENMs reproduce protein global low frequency modes well in comparison to experimental data (Delarue & Figure 6. Position of the Cx26 carbamylation motif relative to the KIDS residues mutated in this study A, view of the hemichannel as seen from the cytoplasmic face. Residues K125 and R104 of the carbamylation motif are shown in red and light green respectively. B, side view of a single subunit showing the positions of mutated residues (N14, A40 and A88 are mutated in KIDS; M151 was mutated to prevent alternative splicing of KIDS mutated mRNA). R104 from the adjacent subunit is shown aligned to K125 of the subunit. NTH: N terminal helix. C, additional view of the subunit after rotating it approximately 90°counterclockwise. Arrow indicates the loops in the subunit that dock to a hexamer in an opposed membrane to form a gap junction. Structure based on 2zw3 from the Protein Database.

Figure 7. KIDS mutations of human Cx26 alter the CO 2 sensitivity of Cx26 gap junctions
A, gap junctions (indicated by yellow arrows) formed by Cx26 A40V-5 (combining the A40V mutation with M151L, to eliminate alternative splicing) are highly permeable to NBDG and lack any CO 2 sensitivity. The images encompass 5 min of recording at a P CO 2 of 35 mmHg (top) and 55 mmHg (bottom). Permeation of NBDG to the coupled cells happens within 2 min in both conditions. B, gap junctions (indicated by yellow arrows) formed by Cx26 A88V-5 are insensitive to P CO 2 -rapid permeation occurs to the coupled cell even at a P CO 2 of 55 mmHg. C, gap junctions formed by Cx26 N14K-5 are not closed by CO 2 . Permeation of NBDG to coupled cells is shown at P CO 2 of 35 mmHg (top row) and 55 mmHg (bottom row). D, summary data showing the time for dye transfer to the coupled cell. While A40V-5 (n = 8, 35 mmHg; n = 6 55 mmHg) and A88V-5 (n = 6, 35 mmHg; n = 7 55 mmHg) show no difference in permeation time with P CO 2 , there is a tendency for the dye to permeate more slowly to the coupled cells for N14K-5 (n = 6 for both 35 and 55 mmHg).  Valadie et al. 2003). We therefore built coarse-grained ENMs (Tirion, 1996) to gain insight into the mechanism by which CO 2 maintains the Cx26 hemichannel in the open state but the Cx26 gap junction in the closed state. ENMs were constructed using the coordinates from high-resolution crystal structures for the Cx26 hemichannel and gap junction in the CO 2 unbound state. CO 2 was represented in the ENMs by the inclusion of additional Hookean springs between residues K125 and R104 of neighbouring monomers in both the hemichannel and the gap junction (Meigh et al. 2013).

Sanejouand
We examined the similarities between eigenvectors for the Cx26 hemichannel and gap junction to understand the changes in harmonic protein motion caused by CO 2 binding. The main open-close mode in the Cx26 hemichannel is defined as the lowest frequency mode that ignores the solid body translational and rotational motions. The solid body translational and rotational motions consist of six modes and so the open-close mode is mode 7 (Meigh et al. 2013). This open-close mode in the hemichannel in the absence of CO 2 (mode 7) becomes reordered as mode 15 in the presence of CO 2 ( Fig. 9A; red-bordered square). The main open-close mode in the gap junction in the absence of CO 2 (mode 10; modes 7-9 represent motions between the two hexamer rings) becomes reordered as mode 24 in the presence of CO 2 (Fig. 9B; green-bordered square). Analysis of the overlap between the main open-close mode in the gap junction and hemichannel in the absence of CO 2 reveals an almost complete overlap between mode of the hemichannel (mode 7) and the gap junction (mode 10) (Fig. 9C, blue-bordered square). The basic open-close dynamics are therefore likely similar between gap junction and hemichannel.
To understand the differing roles of CO 2 in the hemichannel and gap junction, we calculated a range of partially open and closed eigenvectors for both the hemichannel and the gap junction. We then calculated at each state, for both the hemichannel and gap junction, the free energy for CO 2 binding and the influence of CO 2 on the open/close eigenvector. This calculation provides a free energy for each state examined along the open/close eigenvector. Calculation of the free energy difference between the CO 2 -bound and -unbound state provides information on the stability (binding energy of CO 2 needed) for each state. In this case, the closer the value is to zero, the less energy is needed to bind CO 2 (Fig. 9D). The x-axis of the figure represents trajectory along the open/close eigenvector where the higher the value the more closed the hemichannel or gap junction. The y-axis of the figure represents the difference in free energy for the CO 2 -bound and non-bound state where the higher the value is hypothesized to be less preferable for CO 2 binding. On binding CO 2 , the hemichannel (6mer) or gap junction (12mer) will progress along the open/close eigenvector to make the difference in free energy more favourable for the CO 2 -bound state.
For the hemichannel, it is energetically favourable to bind CO 2 in the open-state and then energetically unfavourable to close. For the gap junction it is energetically favourable to bind CO 2 in the closed state and then energetically unfavourable to open. The differences in the open/closed state of the gap junction and hemichannel in response to CO 2 can therefore be explained through entropy contributions to free energy change.

Discussion
The key result from our study is that Cx26 gap junctions are closed by a direct action of CO 2 on the protein. Our prior publications have demonstrated the opening action of CO 2 on Cx26 hemichannels (Huckstepp et al. 2010a;Meigh et al. 2013Meigh et al. , 2014de Wolf et al. 2016de Wolf et al. , 2017Cook et al. 2019;Dospinescu et al. 2019). This paper further shows that these diametrically opposite actions of CO 2 Figure 8. Physical gap junction plaques form between HeLa cells expressing human Cx26 M151L tagged with mCherry, but are non-functional A, the gap junction plaques between cells are indicated by the yellow arrows. For clarity, the green channel (NBDG) has not been merged with the mCherry and DIC images. B, images of a cell loaded with NBDG that formed a gap junction plaque (yellow arrow) with a neighbouring cell ( * ). However, NBDG does not permeate to the neighbouring cell within 5 min even at a permissive level of P CO 2 (35 mmHg). Numbers represent minutes from start of recording.
on gap junctions and hemichannels depend on the same residues and presumably the same carbamate bridging mechanism.
Opposing modulation of gap junctions and hemichannels has been reported before. Lipopolysaccharide and basic fibroblast growth factor inhibit Cx43 gap junctions and open Cx43 hemichannels. While apparently similar to the results reported in this study, the modulation of these two entities is downstream of kinase activity and the signalling actions of arachidonate metabolites; i.e. these are indirect effects on Cx43 and are unlikely to involve the same residues in the protein (De Vuyst et al. 2007). Similarly, the cytokines interleukin 1β and tumour necrosis factor α also inhibit Cx43 gap junctions and open Cx43 hemichannels (Retamal et al. 2007a). Both of these effects are mediated through a p38 mitogen-activated protein kinase (MAPK)-dependent pathway. However, hemichannel opening triggered by these cytokines is sensitive to inhibition of nitric oxide synthase or changing redox state (Retamal et al. 2007b), whereas the gap junction closure is insensitive to redox state. Thus, for the two entities, the mechanisms of modulation evoked by the p38 MAPK pathway differ (Retamal et al. 2007a).

Independence of pH-and CO 2 -dependent modulation of Cx26 gap junctions
A possible alternative interpretation of our data is that CO 2 caused intracellular acidification and thus closed Figure 9. Analysis of the effect of CO 2 on Cx26 hemichannel and gap junction gating via elastic network modelling A, the main open-close mode in the hemichannel (6mer) in the absence of CO 2 (mode 7) is reordered to mode 15 (red square) in the presence of CO 2 . Therefore, this mode contributes less to the total motion of the molecule in the presence of CO 2 -the hemichannel spends more time open (Meigh et al. 2013). B, the main open-close mode in the gap junction (12mer) in the absence of CO 2 (mode 10) is reordered to mode 24 (green square) in the presence of CO 2 . CO 2 thus reduces the contribution of this mode to the total motion of the molecule, but it is not possible to state whether the gap junction is predominantly in the closed or open state from this analysis alone. C, correspondence of the modes for the hemichannel (6 mer) and gap junction (12 mer). Mode 7 in the hemichannel corresponds very closely to mode 10 in the gap junction (indicated by blue square). D, for a range of partially open and closed eigenvectors for both the hemichannel and the gap junction, we then calculated the change in free energy on CO 2 binding and thus the influence of CO 2 on the open/closed eigenvector. The closer the free energy value is to zero, the less energy is needed to bind CO 2 . The x-axis represents trajectory along the open/closed eigenvector going from fully open to fully closed. The y-axis represents the free energy for binding CO 2 where the higher the value the less preferable is CO 2 binding. For the hemichannel, CO 2 binding is less energetically favourable in the closed state, whereas for the gap junction CO 2 binding is favoured in the closed state. J Physiol 599.1 the gap junction. Under this interpretation, CO 2 would have no direct action on Cx26. This interpretation is unlikely for two reasons. Firstly, only modest increases in P CO 2 around the physiological norm were required to close Cx26 gap junctions. Such changes will only cause modest intracellular acidification. By contrast very profound acidification to pH values below 6.5 is required to close the Cx26 gap junction channel (Khan et al. 2020). Secondly, and more importantly, the mutation K125R prevents the CO 2 -dependent closure of the gap junction, but does not affect the closing effect of acidification induced by application of propionate (Fig. 4). The actions of CO 2 and pH on the gap junction are therefore mechanistically independent at the moderate levels of CO 2 used in this study. This independence of mechanistic action of pH and CO 2 is also true for Cx26 hemichannels as acidification causes hemichannel closure (Yu et al. 2007), whereas a modest increase in P CO 2 causes hemichannel opening (Huckstepp et al. 2010a;Meigh et al. 2013;de Wolf et al. 2017;Cook et al. 2019;Hill et al. 2020).
Although the carbamylation motif in Cx26 is required for both the CO 2 -dependent opening of hemichannels and the CO 2 -dependent closure of gap junctions, the link between the CO 2 -dependent modulation of hemichannels and gap junctions is not immutable. Cx26 hemichannels of the lungfish Lepidosiren are insensitive to CO 2 owing to the presence of an extended C-terminus, yet their gap junctions are still closed by CO 2 (Dospinescu et al. 2019). The contrary case is demonstrated by Cx32: hemichannels of this connexin can be opened by CO 2 , but Cx32 gap junctions are insensitive to the same doses of CO 2 (Dospinescu et al. 2019).

Implications for structural biology of Cx26
The crystal structure of Cx26 shows the molecule in the form of a gap junction (Maeda et al. 2009). The two connexons dock via interactions involving the two extracellular loops (E1 and E2) of each subunit. There are multiple hydrogen bonds formed between the opposing E1 and E2 loops of each connexon (Maeda et al. 2009). These give a tight interaction that is likely to alter and constrain the conformation of hemichannels docked in a gap junction versus undocked hemichannels. The opposing modulation of gap junctions and hemichannels by CO 2 must presumably arise from the conformational differences between free versus docked hemichannels.
Our ENM calculations of the interaction of CO 2 with Cx26 hemichannels used the docked hemichannel from the gap junction as a model for the structure of the free hemichannel (Meigh et al. 2013). This model allowed us to propose a plausible gating mechanism for CO 2 -dependent opening of the hemichannel, which we have supported with extensive mutational analysis (Meigh et al. 2013).
We exploited this knowledge to introduce novel gating mechanisms via the same intersubunit interactions by mutating Lys125 to Cys . This made the hemichannel NO-sensitive, via nitrosylation of the Cys125 and interaction with Arg104 of the neighbouring subunit. The double mutation K125C-R104C made the hemichannel redox sensitive (via redox modulation of disulfide bridges between subunits) . The mutated residues in these Cx26 variants are some considerable distance apart -far further than the short distances normally required for the ionic and covalent interactions between the respective residues. That these mutations were nevertheless effective at changing the gating of Cx26 implies considerable flexibility of the hemichannel that can bring apparently distant residues close enough to interact .
Despite this success in using a half gap junction as a model for the free hemichannel, the results in this study suggest the need for considerable caution in any future structural modelling of hemichannels. Not only may the isolated hemichannel be far more flexible than a hemichannel that is part of a gap junction, but it may also have a substantially different conformation. New structures of Cx26 hemichannels and gap junctions are therefore needed to probe their conformational differences, and provide firm mechanistic understanding of the differential modulation of hemichannels and gap junctions by CO 2 .

KIDS mutations and Cx26 gap junctions
There are relatively few reports of the effects of KIDS mutations on Cx26 gap junctions. Prior reports suggest that the A40V mutation prevents Cx26 gap junction formation (Montgomery et al. 2004). Whilst we were not able to study the KIDS mutations in isolation owing to alternative splicing of the mRNA of the mutated Cx26, we could study the effect of these mutations in combination with M151L to prevent this alternative splicing. Clearly our data suggest that, in HeLa cells and in combination with the mutation M151L, the A40V mutation does not prevent formation of functional gap junctions (Fig. 7A). Our data suggest further that N14K and A88V KIDS mutations (once again in combination with M151L) do not prevent gap junction formation and that these gap junctions are highly permeable to NBDG. It is possible, however, that without being combined with M151L these three KIDS mutations might by themselves prevent functional gap junction formation. However, this possibility is made less likely by the observation that in an oocyte expression system Cx26 A40V can form functional gap junctions (Sanchez et al. 2010). This same expression system has also been used to demonstrate that Cx26 N14K can form gap junctions (Lee et al. 2009). KIDS mutations are thought to cause the syndrome through gain-of-function. Several lines of evidence suggest that KIDS mutated Cx26 hemichannels are leaky: either they are less sensitive to Ca 2+ blockade (Sanchez et al. 2010;Mese et al. 2011;Lopez et al. 2013;Zhang & Hao, 2013;Sanchez et al. 2014) or have an altered voltage sensitivity such that they spend more time in the open state at transmembrane potentials closer to the resting potential (Sanchez et al. 2016;Valdez Capuccino et al. 2019). Our finding that three KIDS mutations studied here prevent gap junctions from closing to CO 2 is effectively a further gain of function caused by these mutations: they will remain open under conditions of elevated CO 2 . An important caveat is that we could only study the effect of these mutations on CO 2 sensitivity in combination with M151L. It is possible that by themselves these mutations might affect CO 2 sensitivity of the gap junctions somewhat differently. Interestingly, A40V hemichannels in oocytes are less sensitive to pH blockade than wild-type Cx26 hemichannels, indicating that this mutation has multiple effects on the gating of these channels (Sanchez et al. 2014). It is also interesting that the three KIDS mutations tested here could compensate for the loss of gap junction coupling induced by M151L. This observation is suggestive that these mutations modify channel gating in a dominant manner and is also consistent with the gain-of-function hypothesis.
Although not a KIDS mutation, M151L has been reported to be associated with deafness (Siemering et al. 2006). Our finding that it prevents functional gap junction communication suggests a possible underlying mechanistic cause of hearing loss with this mutation. The endocochlear potential is essential for hearing -it provides the driving force for K + to enter the hair cells through the mechanosensory ion channels present in their stereocilia. Cx26 gap junctions are a pathway for diffusion of K + between the fibrocytes, basal cells and intermediate cells of the stria vascularis (Wangemann, 2006) and have been proposed to form part of the pathway that recycles K + away from the hair cells and back into the endolymph, although this has been disputed (Zhao, 2017). Genetic deletion of Cx26 reduces the endocochlear potential by about half (Chen et al. 2014). Given that M151L prevents gap junction coupling (although not the formation of gap junctions per se), we predict that it would reduce the endocochlear potential and that this is the reason for hearing impairment.

Physiological implications
As Cx26 exists both as gap junctions and hemichannels, our findings have substantial physiological significance. In the context of breathing, Cx26 hemichannels are important and we have already explored their significance for chemosensory control (Huckstepp et al. 2010b;van de Wiel et al. 2020). However, there are reports of gap junctions in various nuclei implicated in the control of breathing (Dean et al. 2002;Solomon et al. 2003). CO 2 -dependent uncoupling of Cx26 gap junctions between cells may therefore be an additional mechanism that contributes to chemosensory control.
In lungfish and amphibia, hemichannels of Cx26 are insensitive to CO 2 despite possessing the carbamylation motif (Dospinescu et al. 2019). The extended C-terminal tail of Cx26 in these species interferes with hemichannel opening. However, the Cx26 gap junctions of these species can still be closed by CO 2 (Dospinescu et al. 2019). The ancestral function of the carbamylation motif in Cx26 was therefore most likely to close the gap junction and the new function of hemichannel opening only arose in the amniotes (Dospinescu et al. 2019). Given that it has been conserved over many hundreds of millions of years, the CO 2 -dependent closure of Cx26 gap junctions must have some important physiological function. What this may be remains open to question, but we speculate that if a single cell in a coupled network were to be excessively metabolically active, it would act as a sink for metabolites such as ATP, glucose or lactate from the coupled cells.
As the metabolically active cell would produce more CO 2 this could be a self-regulating mechanism to preserve network integrity by uncoupling the 'run-away' cell from the network thereby reducing its drain on the communal pool of metabolites. It is interesting in this context that the P CO 2 of renal cortex and liver (respectively 57 and 64 mmHg), two organs where Cx26 and Cx32 are abundantly expressed, is considerably higher than the P CO 2 of systemic circulation (39-45 mmHg) (Hogg et al. 1984). J Physiol 599.1 related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. All persons designated as authors qualify for authorship, and all those who qualify for authorship are listed.

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article.