Introduction

There are 21 connexin genes encoded in the human genome. Connexins are transmembrane proteins that can form hemichannels and gap junction channels. The existence of 21 isoforms suggests functional specialisation across the gene family (Sohl & Willecke, 2004). On the basis of molecular phylogeny, the connexins have been divided into 4 broad clades -the alpha, beta, and gamma and delta groupings (Mikalsen et al., 2021).

Certain members of the β-clade, Cx26, Cx30 and Cx32, are directly sensitive to CO₂ (Huckstepp et al., 2010; Dospinescu et al., 2019). The mechanism of CO₂ sensitivity has best been studied in Cx26. While Cx26 hemichannels are opened in response to CO₂ (Huckstepp et al., 2010), CO₂ closes Cx26 gap junction channels (Nijjar et al., 2021). The CO₂-dependent gating of Cx26 hemichannels and gap junction channels involves carbamylation of a specific lysine, K125, which is part of a critical carbamylation motif, ₁₂₅KVRIEGS₁₃₁ (Meigh et al., 2013). This motif is essential for the CO₂-sensitivity of Cx26, as it allows the formation of a salt bridge between the carbamylated K125 and R104 of the neighbouring subunit (Meigh et al., 2013). The formation of this “carbamate” bridge induces conformational changes, ultimately leading to the opening of the hemichannel or closing of the gap junction channel (Meigh et al., 2013; Nijjar et al., 2021; Brotherton et al., 2022; Brotherton et al., 2024). Mutation of K125 to arginine, an amino acid that cannot be carbamylated, or R104 to alanine so that it cannot form a carbamate bridge, removes the ability of CO₂ to open Cx26 hemichannels (Meigh et al., 2013) or close Cx26 gap junction channels (Nijjar et al., 2021). Furthermore, the introduction of the carbamylation motif into a CO₂-insensitive connexin, Cx31, results in CO₂-dependent hemichannel opening of Cx31 (Meigh et al., 2013). Other members of the β-clade, Cx30 and Cx32, are also CO₂-sensitive, but have different sensitivity profiles (Huckstepp et al., 2010; Dospinescu et al., 2019). Evolutionary comparisons show that the carbamylation motif has been conserved over 400 million years (Dospinescu et al., 2019) and appears to be a necessary and sufficient feature for the CO₂-sensitivity of connexins.

Sequence comparisons of the β-connexins and α-connexins, revealed the presence of a potential carbamylation motif within the α-connexin, Cx43. Cx43 has a motif (₁₄₄KVKMRGG₁₅₀) homologous to the motif in Cx26 (₁₂₅KVRIEGS₁₃₁) suggesting that it might also be sensitive to CO₂. In this study we show that Cx43 hemichannels are indeed directly sensitive to changes in the partial pressure of CO₂ (PCO₂) over the physiological concentration range. Unlike Cx26, the CO₂-dependent opening mechanism of Cx43 is more complex, involving additional residues beyond the homologs of K125 and R104. Our discovery that Cx43 hemichannels are directly CO₂ sensitive has major physiological implications and may explain why Cx43 hemichannels appear to be open under physiological conditions but closed in vitro where they are usually studied in the absence of CO₂/HCO₃⁻ buffers.

Results

Identification of a potential carbamylation motif in Cx43

Cx43, an alpha-connexin, possesses a carbamylation motif homologous to that of the beta-connexins (Fig 1A). AlphaFold3 (AF3) predicts a structure very close to the experimentally determined structures for Cx43 (Lee et al., 2023a; Qi et al., 2023) but has the advantage of including portions of the structure that have not been experimentally resolved. This is a valid approach, as the de novo predictions by AF3 for the cytoplasmic loop of Cx26 have been supported by a recent experimentally-derived structure (Brotherton et al., 2024). By performing alignments of the Cx43 and Cx26 AF3 structures (Fig 1B), we found that two residues in Cx43, K144 and K105, aligned with the residues in Cx26 that are essential for hemichannel opening to CO₂, K125 and R104 (Fig 1C). The predicted 3D orientation of the alpha-helices involved are positioned in a similar manner to allow potential interaction between K144 and K105 following carbamylation. Cx43 also presents another feature that is similar to Cx26: the high incidence of Lys residues. In Cx26, 5 lysine residues in the cytoplasmic loop can be carbamylated, suggesting an environment favourable for this to happen. The presence of multiple Lys residues in the cytoplasmic loop of Cx43 implies that the local environment may be conserved to facilitate the required pKa for CO₂ binding and carbamate bridge formation.

Figure 1:

Cx43 hemichannels are opened by CO₂

We used an established dye-loading protocol, previously used for CO₂ connexin studies (Huckstepp et al., 2010; Meigh et al., 2013; Meigh et al., 2014; de Wolf et al., 2016; de Wolf et al., 2017; Cook et al., 2019; Dospinescu et al., 2019; van de Wiel et al., 2020), to test whether Cx43 hemichannels are CO₂ sensitive. We observed significant dye-loading in Cx43-expressing cells when exposed to a high CO₂ stimulus (70 mmHg) when starting from a baseline of 20 mmHg (MW test, PCO₂ 20 vs 70 mmHg: p = 0.008, n=5, Fig 2A). We used a zero Ca²⁺ aCSF to unblock hemichannels as a positive control to demonstrate that for WT Cx43, the zero Ca²⁺ stimulus and 70 mmHg PCO₂ gave a similar amount of dye loading (MW test: p-value = 0.726, n=5).

Figure 2:

Figure 2, Figure supplement 1.

To further support these findings, we next used whole cell patch clamp recordings from HeLa cells expressing Cx43 (Fig 2B). From a baseline of 35 mmHg PCO₂, an increase in an outward conductance was observed when switching to 55 or 70 mmHg aCSF (Fig 2B). On switching to 20 mmHg aCSF, a small conductance decrease was observed (Fig 2B). Untransfected parental HeLa cells did not display an increase in outward current to changes in PCO₂. However, some parental HeLa cells exhibited a slow increase in an inward current on transfer from 35 to 70 mmHg (Fig 2B). This was quite distinct from the outward current observed in Cx43 expressing cells (Fig 2B). Overall, the patch clamp data predicted a dose response that could be fitted by the Hill equation with a Hill coefficient of 6 and an EC₅₀ of 43 mmHg.

To assess whether the CO₂-dependent opening of Cx43 resulted in ATP release, we co-transfected HeLa cells with the genetically encoded ATP sensor - GRAB_ATP (Wu et al., 2022). CO₂ dependent increase in ATP release was indeed observed (n=22, Fig 2C). We observed an increase in GRAB_ATP fluorescence on going from 20 to 35 mmHg PCO₂ and further to 55 mmHg PCO₂ (Fig 2C). While ATP release at 70 mmHg was slightly less than that at 55 mmHg, it remained markedly higher than at 35 mmHg PCO₂ (Fig 2C). Cells transfected with GRAB_ATP only, showed no release in response to any concentration of CO₂ applied (n=14, Fig 2D). Overall, our results show through three independent assays that Cx43 hemichannels are opened by modest changes in PCO₂ around the physiological norm (Fig 2). Once opened by CO₂, Cx43 hemichannels display permeability to ATP (Fig 2C).

Gap Junction Channels (GJCs) of Cx26 are closed by CO₂ also acting via the carbamylation motif. We therefore tested whether Cx43 GJCs might also be CO₂ sensitive by using a dye transfer assay (Nijjar et al., 2021). Cx43 GJCs allowed equally fast permeation of dye at all levels of PCO₂ and therefore, unlike the hemichannel, were insensitive to CO₂ (Fig 2, figure supplement 1).

Identifying the residues required for CO₂-sensitivity of Cx43

Having established the dose-response characteristics of Cx43 to CO₂, we next examined the potential underlying mechanisms for CO₂-dependent gating via targeted mutagenesis of possible key residues, starting with the equivalents of those known to be involved in the CO₂-sensitivity of Cx26 (Fig 1).

Mutation of K144 to glutamine (K144Q), an amino acid that cannot be carbamylated, partially reduced the CO₂-sensitivity of the channel: the 70 mmHg stimulus resulted in significantly less increase in dye loading than the zero Ca²⁺ positive control (Fig 3A, MW test: p = 0.048, n=5). However, mutation of Lys105 to a non-charged residue (K105Q) resulted in hemichannels that gave a similar increase in dye loading to the zero Ca²⁺ positive control (Fig 3B, MW test: p = 0.111, n=5). In Cx26, mutation of the residues involved in carbamylation (K125 and R104) results in complete loss of CO₂ sensitivity (Meigh et al., 2013; Nijjar et al., 2021). The lack of such a complete effect on the CO₂ sensitivity of Cx43 implies that the mechanism is more complex and may involve further residues.

Figure 3:

We further inspected the AF3 structure prediction and identified two further lysine residues of interest (Fig 3E). K109 in Cx43 aligns with K108 of Cx26, a residue known to be carbamylated and required for Cx26 gap junction closure (Nijjar et al., 2025). AF3 also suggests that K109 could potentially interact with K234 following any carbamylation (Fig 3E). We therefore mutated both residues to Gln to test their possible involvement in CO₂ sensitivity. The dye-loading assay showed that K109Q remained CO₂ sensitive (Fig 3C MW test vs zero Ca²⁺ control, p = 0.274, n=5). However, the mutation K234Q displayed impaired CO₂ sensitivity (Fig 3D, MW test vs zero Ca²⁺ control: p = 0.016, n=5).

As the dye loading assay suggested that single mutations were on the whole rather ineffective in abolishing CO₂ sensitivity, we assayed ATP release via GRAB_ATP to gain further supporting evidence. We found that CO₂-evoked ATP release remained dose dependent and very similar to the WT Cx43 in the K105Q (n=23) and K144Q (n=12) mutants (Fig 4A, C, Figure 4, figure supplement 1). Cx43 hemichannels with the K109Q mutation appeared to have their sensitivity to CO₂ shifted to lower values of PCO₂ (n=20), appearing to give maximal ATP release at 35 mmHg (Fig 4B,E, Fig 4 figure supplement 1). K234Q Cx43 hemichannels had greatly reduced sensitivity to CO₂ (n=15, Fig 4D, F).

Figure 4:

Figure 4, Figure supplement 1.

CO₂ sensitivity of Cx43 involves multiple Lys residues

As single Lys mutations mostly had partial effects on the CO₂ sensitivity of Cx43 hemichannels, we next mutated pairs of Lys residues. Using the dye loading assay we found that all combinations of paired mutations abolished CO₂ sensitivity relative to the zero Ca²⁺ control (Fig 5): K109Q, K144Q (MW test p = 0.008, n=5); K105Q, K144Q (MW test p = 0.004, n=5); K105Q, K109Q (MW test p = 0.004, n=5); K105Q, K234Q (MW test p = 0.004, n=5); and K144Q, K234Q (MW test p = 0.004, n=5). Using the GRAB_ATP assay we confirmed that all of these double mutations also abolished CO₂-dependent ATP release (MW test for all mutations, CO₂ vs high K⁺ positive control p < 10⁻⁶, Fig 6).

Figure 5:

Figure 6:

The effect of introducing negatively charged residues into the carbamylation motif

We next sought to mimic carbamylation by mutating Lys to Glu. This introduces into the motif the negative charge that would occur upon carbamylation of Lys. In the dye loading assay the mutation K144E resulted in a complete loss of CO₂ sensitivity while dye loading to zero Ca²⁺ was retained (MW test, 70 mmHg vs zero Ca²⁺: p = 0.008 , n=5, Fig 7B). The mutation K105E seemed to disrupt both the CO₂-dependent and the zero Ca²⁺ evoked dye loading (Fig 7A). K234E displayed impaired CO₂-dependent dye loading compared to that evoked by zero Ca²⁺ (MW test, 70 mmHg vs zero Ca²⁺: p = 0.048 , n=5, Fig. 7C).

Figure 7:

Using the GRAB_ATP assay, we found that each of these single Lys to Glu mutations disrupted CO₂-dependent ATP release, but did not prevent depolarisation evoked ATP release (Fig 8). This disruptive effect on CO₂ dependence might arise because introduction of the negative charge changes the local environment and makes it harder to carbamylate the other Lys residues.

Figure 8:

As a change in PCO₂ would likely carbamylate more than one Lys residue simultaneously, we made the double mutation K105E and K109E. With the dye loading assay we found that even at a PCO₂ of 20 mmHg, the cells loaded with dye suggesting that the mutated hemichannels were constitutively open (Fig 9A, D). Increasing PCO₂ to 70 mmHg gave no further dye loading. To test whether the channels were open even at a PCO₂ of 20 mmHg, we used the general hemichannel blocker La³⁺ (100 µM) to show that this reduced the dye loading at 20 mmHg (Fig 9B, D, MW test: p = 0.016, n=5). We used the GRAB_ATP assay to demonstrate that at 20 mmHg, the K105E, K109E mutation caused continual release of ATP that could be blocked by La³⁺ (Fig 9C, D, MW test 20 mmHg vs 20 mmHg + La³⁺, p=7.7×10⁻⁷, n=16). This continual release of ATP at 20 mmHg was not evident from cells expressing WT Cx43, as application of 100 µM La³⁺ had no effect on the GRAB_ATP fluorescence (Fig 9C, D).

Figure 9:

Physiological role for CO₂-dependent ATP release via Cx43 from astrocytes

Cx43 hemichannels are CO₂ sensitive and are likely to be partially open at typical levels of PCO₂ in mammalian tissue. They could thus be able to release ATP under physiological resting conditions (35 mmHg PCO₂, Fig. 2). Considering the ubiquitous distribution of Cx43 in astrocytes (de Ceglia et al., 2023) and the likely resting PCO₂ in the brain (Hogg et al., 1984), we postulated that, barring external modulatory influences, Cx43 hemichannels could continually release low levels of ATP in the hippocampus. Indeed, there exists compelling evidence for astrocytic Cx43 hemichannels being partially open under resting physiological conditions in vitro in the hippocampus (Chever et al., 2014). In this study when Cx43 hemichannels were blocked by the mimetic peptide, Gap26, there was a decrease synaptic strength of about 30% that was occluded by targetted knock out of astrocytic Cx43 expression (Chever et al., 2014).

As many investigators state that Cx43 hemichannels are shut under physiological conditions, we tested whether the results of Chever et al (2014) could be explained by the CO₂ sensitivity of Cx43. We predicted that if PCO₂ were lowered sufficiently to largely close Cx43, when PCO₂ was returned to the physiological norm, Cx43 should re-open and result in an increase of synaptic strength, presumably downstream of Cx43-dependent ATP release (Chever et al., 2014). Because ATP can be rapidly broken down to adenosine in the extracellular space (Frenguelli et al., 2007; Wall & Dale, 2013), which could then act at A1 receptors to inhibit transmission and give confounding effects, we used 8-CPT (8-cyclopentyltheophylline) to selectively block A1 receptors.

We preincubated hippocampal slices in 20 mmHg PCO₂ aCSF for 30 mins prior to recording fEPSP in area CA1 evoked by stimulation of the Schaffer colaterals. At constant extracellular pH, in the presence of 8-CPT, elevation of PCO₂ from 20 to 35 mmHg resulted in a 26% increase in EPSP (n=6, Fig 10A). To test whether this effect was mediated via CO₂-dependent opening of Cx43, we used 100 μM Gap26 to see whether this blocked the increase in fEPSP induced by 35 mmHg PCO₂. Once again the fEPSP increased by about 30% on transfer to 35 mmHg aCSF (Fig. 10B), and application of Gap26 reduced the fEPSP to below its baseline value (Fig 10B, D, Friedmann 2 way ANOVA: p=0.006, n=6). This was an effect specific to the Gap26 peptide as 100 μM of the scrambled peptide did not have significant effects on the amplitude of the fEPSP (Fig. 10C, D, Friedmann 2 way ANOVA: p=0.55, n=5).

Figure 10:

Pathological mutations of Cx43 cause loss of CO₂ sensitivity

A prominent condition associated with mutations of Cx43 is occulodental digital dysplasia (ODDD) (Laird, 2014). This syndrome involves a range of conditions such as soft tissue fusion of the digits, abnormal craniofacial bone development, small eyes and loss of tooth enamel. Later in age conditions such as glaucoma, skin disease and neuropathies may become evident. To determine wether the loss of Cx43 CO₂-sensitivity may be a contributing factor to human disease, we studied two mutations, the Cx43 ODDD mutation L90V (Shibayama et al., 2005; Lai et al., 2006) and A44V which causes the skin condition, erythrokeratodermia variabilis et progressiva (EKVP) (Cocozzelli & White, 2019; Srinivas et al., 2019). We found that the L90V mutation removed CO₂ sensitivity from Cx43 by both the dye loading assay and the GRAB_ATP assay (Fig 11). The mutation A44V blocked CO₂ dependent ATP release (Fig 11).

Figure 11.

Discussion

Hemichannels of Cx43 are CO₂ sensitive

In contrast to Cx26 expression, which is restricted to specific, small regions of the brain (Nagy et al., 2001; Rash et al., 2001; Nagy et al., 2011), Cx43 serves as the principal astrocytic connexin, exhibiting widespread presence throughout the brain (Schulz et al., 2015; Yin et al., 2018; de Ceglia et al., 2023). This widespread distribution, combined with our results, suggests that Cx43 gives the capacity for CO₂ sensitivity across the entire brain.

Our data from three independent assays (dye loading, measurement of ATP release and whole cell recordings) show that hemichannels of Cx43 open to modest changes in PCO₂. However, Cx43 gap junction channels are insensitive to these same changes in PCO₂. At physiological levels of PCO₂ Cx43 hemichannels are partially open and able to release ATP. This observation potentially resolves a puzzling inconsistency in the literature. Many biophysical recordings of Cx43 hemichannels, performed in simplified HEPES buffered salines without appreciable dissolved CO₂ or HCO₃⁻, document Cx43 hemichannels as being shut. By contrast, under physiological conditions both in vitro and in vivo, which necessarily involve CO₂-HCO₃⁻ buffering, the evidence suggests that Cx43 hemichannels are constitutively open to some extent (Chever et al., 2014; Guillebaud et al., 2020; Turovsky et al., 2020). Continual ATP release from astrocytes via Cx43 has been shown to give a tone of ATP that enhances the strength of synaptic transmission in the hippoampus by about 20% (Chever et al., 2014). Our data is highly complementary to these observations as we find an enhancement of fEPSP amplitude of about 30% when increasing PCO₂ from 20 to 35 mmHg. It is important to note that our experiments were performed under isohydric conditions and differ from a prior study that examined hypocapnia with simultaneous alkalinisation. This manipulation enhanced fEPSP strength through a reduction in extracellular adenosine and consequently less inhibition via A1 receptors (Dulla et al., 2005).

While our data suggests that Cx43 hemichannels are sensitive to PCO₂ over the range 20-70 mmHg, there is an indication that there may be some CO₂-dependent inhibition of hemichannel opening above 55 mmHg. Slightly less ATP release was consistently seen at a PCO₂ of 70 mmHg compared to 55 mmHg. Furthermore, we cannot be sure that Cx43 hemichannels are completely shut at a PCO₂ of 20 mmHg, as it is not possible to reduce the PCO₂ further without changing to HEPES buffered salines. That Gap26 reduced synaptic transmission below the baseline amplitude recorded in slices bathed in 20 mmHg aCSF suggests that even under this condition there may still be a low level of Cx43 hemichannel gating at least in neural tissue which will continually generate CO₂ via oxidative phosphorylation.

The residues involved in CO₂ sensitivity and possible mechanisms

By exploiting the new structural data for connexins (Myers et al., 2018; Flores et al., 2020; Lee et al., 2020; Yue et al., 2021; Brotherton et al., 2022; Qi et al., 2022; Lee et al., 2023a; Lee et al., 2023b; Qi et al., 2023), the predictive power of AF3 and our understanding of carbamylation and CO₂-dependent gating of Cx26 hemichannels and GJCs (Nijjar et al., 2021; Brotherton et al., 2024), we identified a series of residues that could plausibly be involved in mediating the CO₂ sensitivity of Cx43: K144, the equivalent of K125 in Cx26; K105, the analogue of R104 in Cx26; K109, the equivalent of K108 in Cx26; and K234, an equivalent of R216 in Cx26. Both K125 and K108 in Cx26 are known to be carbamylated (Nijjar et al., 2025). Carbamylation of K125 is essential for hemichannel opening to CO₂, whereas carbamylation of both K125 and K108 is required for GJC closure to CO₂ (Nijjar et al., 2025). R216 in Cx26 has been tentatively identified as a possible interacting partner for K108 but this could be indirect (Nijjar et al., 2025). Crucially the spatial arrangements of K144, K105, K109 and K234 in AF3 predictions indicate that these residues are sufficiently close to interact across the subunit boundary.

In Cx26, K125 is proposed to interact with R104 following carbamylation, and interestingly mutation of either K125 or R104 is sufficient to abolish CO₂ dependent hemichannel opening and GJC closure. Mutation of K108 is also sufficient by itself to abolish GJC closure to CO₂ (Nijjar et al., 2025). By contrast single mutations to Gln of any of the 4 Lys residues identified in Cx43 as the equivalents to those involved in CO₂-dependent gating of Cx26 were unable to completely abolish CO₂ sensitivity. Similarly the mutations K125E and R104E in Cx26 create consitutively open hemichannels. However single Lys to Glu mutations in Cx43 did not have this effect.

Our data show that in Cx43 at least 2 of the 4 identified residues must be mutated to completely abolish CO₂ sensitivity or to give constitutively open hemichannels. For example mutation of any two Lys residues to Gln, completely abolishes CO₂ dependent ATP release and dye loading. Conversely the double mutation, K105E and K109E, is required for constitutively open hemichannels.

This suggests that in Cx43 there is some redundancy in the effect of CO₂. One possible interpretation is that two carbamate bridges are formed: one being between K144 and K105 (the direct equivalent of K125-R104 in Cx26); and the second being between K109 and K234. There is certainly some data that support this: the double mutations K109Q-K144Q, K105Q-K234Q and K144Q-K234Q being completely insensitive to CO₂. These double mutations would remove both of the proposed bridges. However our observation that the mutations K105Q-K144Q, K109Q-K234Q also completely remove CO₂ sensitivity is inconsistent with a simple two bridge hypothesis as these double mutations would only affect one of the bridges and should therefore leave the other intact to give some degree of CO₂ sensitivity. Instead, we favour a hypothesis in which any of these four residues could be carbamylated but that they all have potential to interact. Depending on which residues become carbamylated a bridge could form between K105 and K234 or between K109 and K144, and possibly even K144 and K234. There is some support for this idea from AF3 as K105 and K109 are only one turn distant in an alpha helix and both could therefore interact with either K144 or K234 following carbamylation. The residues in Cx43 that are involved in CO₂ sensitivity are equivalent to those identified in Cx26, where they participate in both hemichannel and gap junction channel gating to CO₂. However in Cx43 these residues are only involved in mediating the sensitivity of the hemichannel to CO₂. In this respect it is interesting that mutations of the different Lys residues alter the dose-sensitivity of Cx43 to CO₂ in different ways (Fig 4, figure supplement 1).

Physiological implications of the CO₂ sensitivity of Cx43

Cx43 is the most widely expressed connexin in the human body, being present in every organ system (Lucaciu et al., 2023). In metabolically highly active organs such as liver, kidney and brain, ouir data suggest that the physiological PCO₂ will be sufficient to substantially open Cx43 hemichannels (Hogg et al., 1984). In the context of the brain, Cx43 is the main astrocytic connexin and is expressed in all subtypes (de Ceglia et al., 2023). This would imply that CO₂-sensitivity mediated via Cx43 will extend to potentially all brain regions. As astrocytes are non-excitable, they are unlikely to become sufficiently depolarised for Cx43 hemichannels to open via voltage dependent gating. We suggest therefore, in non-excitable cells such as astrocytes, that variations in PCO₂ may be the most important physiological regulator of Cx43 hemichannel gating. For astrocytes, CO₂-dependent opening of Cx43 hemichannels is likely to result in the release of a mix of small, neurochemically significant molecules such as ATP, glutamate, D-serine and lactate, depending on their intracellular concentration.

Cx43 is also highly expressed in the heart, if a proportion of the population were to be in hemichannel form as opposed to GJCs (De Smet et al., 2021), the newly-discovered CO₂ sensitivity of hemichannels may have profound implications for heart function given that cardiomyocytes are highly active, and will thus generate high levels of CO₂ through oxidative phosphorylation. There is a long-established link between CO₂ and cardiac function with effects such as increased coronary blood flow in response to CO₂ as well as changes in myocardial contractile function (Crystal, 2015). CO₂ can even cause arrythmias (Zhang et al., 2019).

NF-kappa B and innate immunity exhibit sensitivity to CO₂ (Cummins et al., 2010; Keogh et al., 2017). Macrophages are CO₂-sensitive, although at least some of this is indirect and via pH changes mediated by carbonic anhydrase activity (Strowitzki et al., 2022). However, a direct effect of CO₂ on components of the immune system cannot be excluded. In cell culture, when carbonic anhydrase activity is blocked in THP-1 monocytes, the effects of CO₂ are significantly diminished but not abolished, suggesting the existence of a pH-independent pathway (Strowitzki et al., 2022). As Cx43 is expressed in monocytes it is a plausible candidate to mediate direct CO₂ sensitivity in these cells. Furthermore, Cx43 function in macrophages is deemed critical for various physiological and pathophysiological processes (Rodjakovic et al., 2021). The role of CO₂ on such pathways remains to be explored.

It is interesting that pathological mutations of Cx43 abrogate CO₂ sensitivity. This follows a pattern seen for Cx26, where a range of mutations linked to syndromic and non-syndromic hearing loss (Meigh et al., 2014; de Wolf et al., 2016; Cook et al., 2019), and for Cx32, where several mutations that cause X-linked Charcot Marie Tooth Disease (Butler & Dale, 2023) alter or completely block CO₂ sensitivity. Further studies are needed to see whether loss of CO₂ sensitivity from Cx43 does indeed contribute to pathology.

Evolution of CO₂ sensitivity in connexins

Our previous findings showed that a common carbamylation motif is prevalent across the beta-connexin clade and confers CO₂-sensitivity on the connexins that possess it (Dospinescu et al., 2019). This motif for example is present in shark Cx32 (which is CO₂ sensitive). Humans and sharks last shared a common ancestor about 400-450 MYA. Our discovery that a very similar carbamylation motif is present in Cx43 and confers CO₂ sensitivity onto their hemichannels, suggests that this property must have been present in the ancestral connexin gene that predated the establishment of the alpha and beta connexin clades. Further work is needed to understand the evolution of the carbamylation motif and CO₂ sensitivity, but our observations indicate that it is an ancient feature within the molecular phylogeny of the connexins, rather than being a more recently derived feature restricted to the beta connexins.

Methods

Recording Solutions

Hypocapnic (20 mmHg PCO₂): 140 mM NaCl, 10 mM NaHCO₃, 1.25 mM NaH₂PO₄, 3mM KCl, 1 mM MgSO₄, 10 mM D-glucose and 2 mM CaCl₂. This was bubbled with a mix of 95%O₂/5% and balanced with sufficient pure O₂ from an oxygen concentrator to give a final pH of ∼7.3.

Control (35 mmHg PCO₂): 124 mM NaCl, 26 mM NaHCO₃, 1.25 mM NaH₂PO₄, 3 mM KCl, 10 mM D-glucose, 1 mM MgSO₄, 2 mM CaCl₂. This was bubbled with 95%O₂/5% CO₂ and had a final pH of ∼7.3.

Hypercapnic (55 mmHg PCO₂): 100 mM NaCl, 50 mM NaHCO₃, 1.25 mM NaH₂PO₄, 3 mM KCl, 10 mM D-glucose, 1 mM MgSO₄, 2 mM CaCl₂. This was bubbled with sufficient CO₂ (∼9%, balance O₂) to give a final pH of ∼7.3.

Hypercapnic (70 mmHg PCO₂): 73 mM NaCl, 80 mM NaHCO₃, 1.25 mM NaH₂PO₄, 3 mM KCl, 10 mM D-glucose, 1 mM MgSO₄, 2 mM CaCl₂. This was bubbled with sufficient CO₂ (approximately 12%, balance O₂) to give a final pH of ∼7.3.

Zero Ca²⁺: 140 mM NaCl, 10 mM NaHCO₃, 1.25 mM NaH₂PO₄, 3mM KCl, 1 mM MgSO₄. On the day of recording, 10 mM D-glucose and 2 mM MgCl₂ was added, and the solution bubbled with O₂ with a final pH of ∼7.3.

Cell culture

HeLa DH cells were grown in Dulbecco’s Modified Eagle Medium (DMEM), supplemented with 10% fetal bovine serum, 50 μg/mL penicillin/streptomycin. HeLa DH cells were used for patch clamp studies on Cx43 and for the dye loading of all Cx43 variants. For dye loading experiments, cells were seeded onto coverslips at a density of 7.5 × 10⁴ cells per well and transiently transfected with the Cx43 constructs following the PEI Transfection Reagent protocol.

Dye-loading

Cell expressing the construct (48-72h) were given an initial 5-minute wash with control solution (artificial cerebrospinal fluid – aCSF, 20 mmHg PCO₂). Following this, the cells were exposed to either control solution, hypercapnic (70 mmHg PCO₂), or a zero Ca²⁺ control (20 mmHg PCO₂) solution, all of which contained 200 µM 5(6)-carboxyfluorescein (CBF) for 10 min. Next, cells were put into control solution with 200 µM CBF for 5 minutes to close any open channels and prevent dye loss in the next step. Then the coverslip was washed with control solution (without CBF) for another 30 min to remove extracellular dye. For each condition a different coverslip with HeLa cells was used. Cx43 was tagged with mCherry on the C-terminus using a short linker. Expression was verified using mCherry fluorescence. The experiments were replicated using independent transfections 5 times.

Following dye loading, HeLa cells were imaged by epifluorescence (Scientifica Slice Scope, Cairn Research OptoLED illumination, 60x water Olympus immersion objective, NA 1.0, Hamamatsu ImagEM EM-CCD camera, Metafluor software). CBFwas excited by a 470 nm LED, with emission captured between 504-543 nm. Cx43 constructs had a C-terminal mCherry tag, which was excited by a 535 nm LED and emission captured between 570-640 nm.

Following acquisition analysis was performed by a person blind to the experimental condition. ImageJ (Wayne Rasband, National Institutes of Health, USA) was used to measure the extent of dye loading by drawing a region of interest (ROI) around each cell, and subsequently, the mean pixel intensity of the ROI was determined. The mean pixel intensity of a representative background ROI for each image was subtracted from each cell measurement from the same image. At least 40 cells were measured for each condition per experiment, and five repetitions of independently transfected HeLa cells were completed. The mean pixel intensities were plotted as cumulative probability distributions, and these graphs show every data point measured. To assess the effect of the 70 mmHg and zero Ca²⁺ solutions on dye loading, the difference in the median pixel intensities between the 70mmHg and 20 mmHg conditions, and the zero Ca²⁺ and 20 mmHg conditions was calculated for each transfection (Figs 3, 5, 7 and 11).

GRAB_ATP recordings

Cells were transiently transfected with the pCAG-Cx43-mCherry construct and pDisplay-GRAB_ATP1.0-IRES-mCherry-CAAX (Addgene plasmid # 167582 ; RRID:Addgene_167582) (Wu et al., 2022) 48 hours prior to imaging. Cells were perfused with control aCSF until a stable baseline was reached, before perfusion with either hypercapnic or high K⁺ aCSF. Once a stable baseline was reached after solution change, cells were again perfused with control aCSF and when a stable baseline reached, recordings were calibrated by direct application of 3 μM of the corresponding analyte.

All cells were imaged by epifluorescence as above. The cpGFP in GRAB_ATP was excited by a 470 nm LED, with emission captured between 504-543 nm. As the Cx43 constructs were mCherry tagged, we selected only cells that expressed both cpGFP and mCherry for recording. GRAB_ATP fluorescence images were acquired every 4 seconds. For each condition, at least 3 independent transfections were performed with at least 2 coverslips per transfection.

Analysis of all experiments was carried out in ImageJ. Images were opened as a stack and stabilised. ROIs were drawn around cells co-expressing both sensor and connexin. Median pixel intensity was plotted as normalised fluorescence change (ΔF/F₀) over time to give traces of fluorescence change. Amount of ATP release was quantified as concentration by normalising to the ΔF/F₀ caused by application of 3 μM ATP.

Patch clamp recordings

Coverslips containing non-confluent HeLa cells were placed into a perfusion chamber at room temperature and superfused with control aCSF. An Axopatch 200B amplifier was used to make whole-cell recordings from single HeLa cells. The intracellular fluid in the patch pipettes contained: K-gluconate 60 mM, Cs-gluconate 50 mM, CsCl 10 mM, TEACl 10 mM, EGTA 10 mM, Na₂ATP 3mM, MgCl₂ 3 mM, CaCl₂ 1 mM, HEPES 10 mM, sterile filtered, pH adjusted to 7.2 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. Whole-cell conductance was measured by repeated steps to -40 mV. To allow for any drift in whole-cell conductance unrelated to the CO₂ stimulus, the maximal conductance during the CO₂ test stimulus was compared to the average of the conductance before the stimulus and after the stimulus had been fully washed off. Each whole cell recording was considered to be an independent statistical replicate.

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: K-gluconate 130 mM; KCl 10 mM; EGTA 5 mM; CaCl₂ 2 mM, HEPES 10 mM, pH was adjusted to 7.3 with KOH to give a resulting final osmolarity of 295 mOsm. Cells were imaged on a Cleverscope (MCI Neuroscience) with a Photometrics Prime camera under the control of Micromanager 1.4 software. LED illumination (Cairn Research) and an image splitter (Optosplit, Cairn Research) allowed simultaneous imaging of the mCherry-tagged Cx43 subunits and the diffusion of the NBDG into and between cells. Coupled cells for intercellular dye transfer experiments were initially selected based on tagged Cx43 protein expression and the presence of a gap junctional plaque (Figure 4, figure supplement 1). Dye permeation between cells was measured at PCO₂ levels of 20, 55 and 70 mmHg. After establishing the whole cell mode of recording, images were collected every 10 seconds. The assay was performed as described by Nijjar et al (Nijjar et al., 2021). The start of recording was taken to be the first image following the establishment of a stable whole cell recording. For each gap junction recording (considered as an independent statistical replicate), analysis of the cell images was performed in ImageJ using an ROI drawn on each cell to measure the median pixel intensity of the donor and acceptor cells. The time for the median fluorescence of the acceptor cell to reach 10% of the donor cell was then determined and used as the value for statistical comparisons.

Hippocampal slice preparation

Mice of either sex, 3-5 weeks old, were killed by cervical dislocation and decapitated in accordance with United Kingdom Animals (Scientific Procedures) Act (1986). The brain was dissected and kept on ice both the cerebellum and the rostral section of the brain were removed. 400 µm thick sagittal slices were cut with a vibratome (Microm HM 650V microsliver) in ice-cold cutting solution composed of (in mM: 85 NaCl, 2.5 KCl, 0.5 CaCl₂, 1.25 NaH₂PO₄, 24 NaHCO₃, 25 glucose, and 75 sucrose; pH was adjusted to 7.4, bubbled with 95% O₂ + 5% CO₂). Then slices were incubated in aCSF composed of (in mM: 127 NaCl, 1.9 KCl, 2 MgCl₂, 2 CaCl₂, 1.2 KH₂PO₄, 26 NaHCO₃, 10 D-glucose, pH 7.4, when bubbled with 95% O₂ and 5% CO₂) at 33 °C. For fEPSP recordings an individual slice was pre-incubated in 20 mmHg PCO₂ aCSF for 30 mins, following which the slice was transferred to a recording chamber submerged in aCSF at 32°C. For stimulation, pulses were delivered by a stimulator via concentric bipolar metal stimulating electrode placed on the surface of CA3. For extracellular recordings, an aCSF-filled microelectrode was placed in CA1.

Statistics

Data is plotted as either cumulative probabilities (showing every data point) or box and whisker plots where the box is interquartile rage, bar is median, and whisker extends to most extreme data point that is no more than 1.5 times the interquartile range. All individual data points are superimposed on the plots.

In the case of dye loading one transfection was considered one replicate. In the case of patch clamping and GRAB_ATP experiments, one cell was considered one replicate. For dye-loading, to assess the effect a mutation would have on the protein, the difference in dye loading evoked by 70 mmHg PCO₂ or 0 Ca²⁺ conditions (from the 20 mmHg baseline) for each construct were compared, using the median value for each condition from each transfection. The Mann-Whitney U one-sided test was used for the comparison on the basis that a priori we expected mutations to either reduce CO₂ sensitivity or have no effect. To compare the fEPSP size across different conditions, the non-parametric Friedman two way ANOVA was used. Statistical analysis was performed using the python language and SciPy library.

Cloning

Plasmids were generated using the Gibson Assembly method (Gibson et al., 2009). PCR fragments were amplified using Q5® High-Fidelity DNA Polymerase (New England Biolabs) (normal PCR or site-directed mutagenesis). The overlap region generated through PCR for Gibson Assembly was 20 base pairs, and the pCAG vector was used as the backbone. All PCR primers were purchased from Integrated DNA Technologies (IDT). All constructs were confirmed by DNA sequencing (Eurofins GATC sequencing or Plasmidasaurus).

Reagents used

Acknowledgements

We thank Prof Alexander Cameron for commenting on a draft of this paper.

Additional information

Funding

We thank the BBSRC (BB/T013346/1, ND) for support. JB was supported by the Biotechnology and Biological Sciences Research Council (BBSRC) and University of Warwick funded Midlands Integrative Biosciences Training Partnership (MIBTP) grant number BB/T00746X/1. VMD was funded by the Medical Research Council through the University of Warwick Doctoral Training Partnership, grant number MR/N014294/1