Activity-dependent CO₂ production in the axon triggers opening of Connexin32 in the Schwann cell paranode

Abstract

Loss of function mutations of Cx32, which is expressed in Schwann cells, cause X-linked Charcot-Marie-Tooth disease, a slowly progressive peripheral neuropathy. Action potential propagation causes Cx32 hemichannels in the Schwann cell paranode to open. As Cx32 hemichannels are directly sensitive to CO₂, we have tested whether CO₂ produced in the axon, as a consequence of the energetic demands of action potential propagation, might gate Cx32 hemichannels. Using isolated sciatic nerve from the mouse, we found that the critical components required for intercellular CO₂ signaling are present (nodal mitochondria, the source of CO₂; a CO₂-permeable aquaporin, AQP1; paranodal Cx32; and carbonic anhydrase). We have used a membrane impermeant fluorescent dye, FITC, to demonstrate the opening of Cx32 in Schwann cells in response to an external CO₂ stimulus or during action potential propagation in the isolated nerve. Pharmacological manipulations of AQP1 or carbonic anhydrase activity altered Cx32 gating during action potential firing. Expression of a modified Cx32 subunit, Cx32^DN, that coassembles with Cx32^WT, revealed that the activity-dependent dye loading of Schwann cells depended upon CO₂ binding to Cx32. CO₂ can, therefore, mediate neuron-to-glia signaling via connexins. CO₂ permeable aquaporins and carbonic anhydrase are key components of this signaling mechanism.

Introduction

Connexin32 (Cx32) is expressed in Schwann cells and oligodendrocytes, the myelinating cells of the peripheral and central nervous system, respectively. These cells wrap around axons to form the myelin sheath, an insulating barrier that restricts voltage-dependent ion fluxes to the nodes of Ranvier, thus enabling saltatory conduction (Huxley and Stampfli, 1949). Charcot-Marie-Tooth (CMT) disease is a slow progressing peripheral neuropathy that involves a loss of peripheral myelin integrity (Murakami et al., 1996). Typical symptoms of CMT include slowing of peripheral conductance velocity, loss of feeling in the extremities, pes cavus, and in some cases, muscle wasting. Mutations in the gjb1 gene, encoding Cx32, result in the X linked version of CMT (CMTX) (Bergoffen et al., 1993; Fairweather et al., 1994; Hattori et al., 2003; Record et al., 2023). Cx32-null mice reproduce CMTX phenotypes, indicating CMTX is caused by a loss of Cx32 function (Scherer et al., 1998). This phenotype can be rescued by selective re-expression of Cx32 in Schwann cells, highlighting how fundamental Cx32 is to the maintenance of myelin health (Scherer et al., 2005).

Cx32 is a β connexin and is closely related to Cx26 and Cx30. Hemichannels of these connexin isoforms can be opened by increases in PCO₂, at constant extracellular pH and physiological concentrations of Ca²⁺ (Huckstepp et al., 2010; Meigh et al., 2013; Dospinescu et al., 2019; Butler and Dale, 2023). CO₂ sensitivity of these β connexins is dependent on a ‘carbamylation motif’ (Meigh et al., 2013; Brotherton et al., 2022; Brotherton et al., 2024; Nijjar et al., 2025). While the mechanism of CO₂ sensitivity has been most thoroughly studied in Cx26, Cx32 possesses the same carbamylation motif that is required for CO₂ sensitivity (Dospinescu et al., 2019; Butler and Dale, 2023). In Cx32, CO₂ carbamylates the primary amine of Lys124 in the motif. As this carbamylated amine is now negatively charged, it can form a salt bridge with Lys104 of the neighbouring subunit. The resulting carbamate bridges are thought to bias the hemichannel to the open state. Like most connexins, Cx32 will form gap junction channels where two hexamers of Cx32 in opposing membranes can dock together. Cx32 gap junction channels, however, are insensitive to the changes in PCO₂ that can open Cx32 hemichannels (Dospinescu et al., 2019).

Myelin expresses Cx32 as both unopposed hemichannels in the paranodal membrane and also as reflexive gap junctions in the Schmidt-Lanterman incisures (Bergoffen et al., 1993; Meier et al., 2004; Bortolozzi, 2018). The reflexive gap junctions provide radial diffusion pathways through the layers of myelin. However, radial diffusion pathways still exist in Cx32-null mice, suggesting a mechanism of redundancy or compensation (Balice-Gordon et al., 1998). Nevertheless, as Cx32-null mice still reproduce CMTX (Scherer et al., 1998), the loss of Cx32 hemichannel function must be sufficient to induce CMTX pathology.

Cx32 hemichannels in the paranode are thought to gate open and release ATP during action potential propagation (Nualart-Marti et al., 2013). The mechanism underlying the opening of Cx32 hemichannels in the paranode in response to action potential propagation remains uncertain. Two hypotheses have been proposed: (i) because Cx32 is intrinsically voltage sensitive, hemichannel opening could be caused by transmembrane potential excursions during action potential propagation (Abrams et al., 2002); and (ii) a rise in intracellular Ca²⁺ within the Schwann cell paranode possibly downstream of activation of a G-protein coupled receptor could open Cx32 (Carrer et al., 2018).

In this paper we explore an alternative hypothesis: that CO₂, produced in the axon as a consequence of the energetic demands of restoring transmembrane ionic gradients following action potential propagation (via Na⁺/K⁺ ATPases), diffuses into the paranode to open Cx32. We have tested our hypothesis by careful consideration of the requirements of a CO₂-based signaling system: a means of production (mitochondria); a channel to allow CO₂ produced in the node to diffuse into the paranode as CO₂ does not readily cross biological membranes; and a mechanism to terminate the actions of CO₂ (carbonic anhydrase). We show that all of these components are present at the node/paranode and that their manipulation will alter the gating of Cx32 in ways that support our hypothesis.

Results

The components for CO₂ signaling mediated via Cx32 are present in myelin

A CO₂-based signaling system in myelin requires: a means of production; a channel to allow CO₂ to cross the nodal and paranodal membranes; and a mechanism to terminate the actions of CO₂ (Figure 1). It is already known that: mitochondria are present in the node (Ohno et al., 2011); Cx32 is expressed in the paranode (Bergoffen et al., 1993); AQP1, highly permeable to CO₂ (Endeward et al., 2006; Musa-Aziz et al., 2009), is expressed in Schwann cells (Gao et al., 2006; Segura-Anaya et al., 2015); and carbonic anhydrase is universally present in every cell. Here, we have used high-resolution microscopy to examine the precise subcellular localization of these components relative to each other alongside markers of the nodal and paranodal regions (Figure 2—figure supplement 1).

Figure 1. Hypothesized connexin32 (Cx32)-mediated CO₂ signaling cascade in peripheral myelin.

Three fingers of a myelin paranode have been used for illustrative purposes. Restoration of transmembrane ionic gradients following action potential propagation via the actions of Na⁺/K⁺ ATPases incurs a metabolic cost and increases production of ATP and CO₂. AQP1, permeable to CO₂, provides a pathway for CO₂ to leave the node and enter the paranode and bind to Cx32 on the intracellular loop. This triggers opening of Cx32 and release ATP. Carbonic anhydrase (CA) catalyzes the combination of CO₂ and H₂O and ultimately the production of HCO₃^- and H⁺ ions and effectively competes with Cx32 for CO₂.

AQP1 is present in the axonal node and Schwann cell paranode

AQP1, a CO₂ permeable aquaporin (Endeward et al., 2006; Musa-Aziz et al., 2009), was localized to the Schwann cell paranode and outer myelin membrane (Figure 2). AQP1 expression also colocalized with Caspr, showing that it was present in the axonal nodal membrane (Einheber et al., 1997). Interestingly, analysis of colocalization Cx32 and AQP1 in the node/paranode region showed that AQP1 was in close proximity to Cx32 in the paranode (M1: mean 0.400; 95% CI 0.254–0.546 and M2: mean 0.301; 95% CI 0.199–0.403). This subcellular localization of AQP1 would allow it to act as a conduit for CO₂ generated at the axonal node to enter into the Schwann cell paranode and interact with Cx32.

Figure 2. AQP1 localizes to both the Schwann cell paranode and also the axonal node.

(A, B) Representative confocal SIM images from a single optical plane showing the localization of Caspr, connexin32 (Cx32), and AQP1 in an isolated mouse sciatic nerve. Arrowheads indicate the node. Scale bars, 10 μm. (C) Boxplots showing degree of colocalization between Cx32 and AQP1 at the node/paranode. Control measurements (to right of dashed line) used these same images with one channel flipped 90° and the same thresholds as when measuring colocalization. Kruskal-Wallis ANOVA p<0.0001.

Figure 2—figure supplement 1. Markers for the node, paranode, and juxtaparanode in the isolated mouse sciatic nerve.

(A) Location of the axonal node (KCNQ2) and Schwann cell paranode (Caspr) within an isolated mouse sciatic nerve fibre. KCNA2 is expressed in the outer myelin layer. Arrowheads indicate the node. Scale bar = 15 µm. Confocal LSM image, single optical plane. (B) Schematic indicating the same locations within an isolated nerve fibre.

Figure 2—figure supplement 2. Connexin32 (Cx32) colocalizes with mitochondria in the Schwann cell paranode, alongside SFXN1.

(A, B) Representative SIM images in single optical plane showing the localization of Cx32, CytC, and SFXN1 in an isolated mouse sciatic nerve. Arrowheads depict the node. Scale bars – 10 μm. (C) Boxplots showing degree of colocalization between Cx32 and CytC and SFXN1 and CytC. M1 is the proportion of Cx32 (or SFXN1) that colocalizes with CytC and M2 is the reverse proportion of CytC that colocalizes with Cx32 (or SFXN1). Control measurements (to right of dotted line for each pair) used these same images with one channel flipped 90° and the same thresholds as when measuring colocalization. Kruskal Wallis ANOVA: Cx32 vs CytC, p=0.0001; SFXN1 vs CytC, p<0.0001.

Figure 2—figure supplement 3. The localization of AQP1 in relation to mitochondria, CytC, in isolated mouse sciatic nerve.

Representative SIM images in single optical plane showing the localization of AQP1 and CytC in an isolated mouse sciatic nerve. Arrows indicate the node. Scale bar – 10 μm.

Mitochondria localize to the axonal node and Schwann cell paranode and may be brought in close proximity to Cx32 via SFXN1

SFXN1 is a mitochondrial protein that also binds to Cx32 (Fowler et al., 2013). Using cytochrome C (CytC), as a mitochondrial marker, we found that mitochondria were localized in both the axonal node and Schwann cell paranode (Figure 2—figure supplement 2), in accordance with previous reports (Rydmark et al., 1998; Ohno et al., 2011). There was colocalization between Cx32 and CytC in the Schwann cell paranode (Figure 2—figure supplement 2, mean; 95% confidence interval, M1: 0.314; 0.198–0.431, and M2: 0.261; 0.165–0.357). There was also colocalization in the Schwann cell paranode between CytC and SFXN1 (Figure 2—figure supplement 2, M1: 0.568; 0.441–0.695 and M2: 0.462; 0.336–0.588). This suggests that SFXN1 may facilitate the association of Cx32 and mitochondria (Fowler et al., 2013). AQP1 also closely associated with CytC (Figure 2—figure supplement 3). Interestingly, SFXN1 was also observed in the absence of CytC (Fowler et al., 2013) suggesting that it has additional cellular roles unrelated to its mitochondrial function.

Carbonic anhydrase is present in the paranode

We observed strong expression of CAII in non-myelinated fibres (Figure 3). However, consistent with earlier reports (Cammer and Tansey, 1987), we also observed weaker but more localized expression in myelinated fibres, specifically at the axonal node and Schwann cell paranode (Figure 3).

Figure 3. CAII localizes to myelinating Schwann cells, in particular to the axonal node and the Schwann cell paranode.

(A and B) Representative confocal LSM images in single optical plane showing the localization of CAII and Cx32 in an isolated mouse sciatic nerve. Arrowheads indicate the node. Intense CAII staining, denoted by a white asterisk (*) is present in non-myelinated fibres. Scale bar applies to A and B: 10 μm.

CO₂-dependent dye loading of Schwann cells in sciatic nerve

We first examined whether Cx32 hemichannels in Schwann cells could be opened by application of hypercapnic aCSF. We exposed isolated sciatic nerves to FITC in aCSF at different levels of PCO₂. As FITC is membrane impermeant but can readily move through channels with large pores, such as Cx32 hemichannels (Butler and Dale, 2023), any CO₂-dependent dye loading would thus indicate gating of a CO₂-sensitive large pore channel.

At 35 mmHg, a level of PCO₂ that is too low to open Cx32 hemichannels (Huckstepp et al., 2010; Dospinescu et al., 2019), FITC loading was not observed (Figure 4). However, in the presence of hypercapnic aCSF (70 mmHg, sufficient to open Cx32 hemichannels) dye loading into the paranode and outer myelin layers was readily observed (Figure 4, p<0.0001, compared to 35 mmHg). Note that the axons did not load with FITC.

Figure 4. A membrane impermeable dye, FITC, loads into Schwann cell paranodes in a CO₂ dependent manner through a hemichannel.

(A) Representative images showing the FITC loading into mouse sciatic nerve bundles. Arrowheads indicate the node. Little FITC loading occurs in response to control (35 mmHg, 10 min) aCSF. FITC loading was greatly increased by 70 mmHg aCSF (10 min) and application of 100 μM FCCP. FITC loading in 70 mmHg aCSF was blocked by carbenoxolone (CBX) but not the TRPA1 antagonist HC030031. (B) Boxplot showing intensity of FITC fluorescence under the different conditions. Each point represents a separate region of interest (ROI) from five different nerves for each condition. Scale bar – 10 μm.

Figure 4—figure supplement 1. Cx31.3 does not open in response to hypercapnia.

(A) Representative images of GRAB_ATP fluorescence, encoded by a 16 color LUT in Cx31.3 transfected HeLa cells in response to control (35 mmHg), hypercapnic (70 mmHg), and depolarizing (50 mM KCl) aCSF. Scale bar 20 μm. (B) Top, traces showing the normalized GRAB_ATP fluorescence for cells transfected only with GRAB_ATP. Bottom, traces from the cells transfected with both Cx31.3 and GRAB_ATP as shown in (A). (C) Summary statistic box plots showing the [ATP] release, calculated as a ratio from the normalized fluorescence change evoked by a certain solution or stimuli and the fluorescence change evoked by 3 µM ATP. Cells expressing only GRAB_ATP do not exhibit changes in fluorescence to 70 mmHg or 50 mM KCl aCSF. Each data point represents a cell, with all cells coming from at least three transfections.

To demonstrate that mitochondrially produced CO₂ could gate Cx32, we used a mitochondrial uncoupler, FCCP, to maximize rates of endogenous CO₂ generation (Balboni and Lehninger, 1986). We found that FCCP, applied in 35 mmHg aCSF, caused significantly increased dye loading into Schwann cell paranodes and outer myelin layers (p<0.0001) compared to nerves loaded at 35 mmHg PCO₂ with no FCCP (Figure 4).

CO₂-evoked FITC loading was abolished in the presence of carbenoxolone, indicating FITC entry occurred through a carbenoxolone-sensitive hemichannel (p<0.0001, Figure 4). TRPA1 can open with intracellular acidification (Wang et al., 2010), however, CO₂-evoked FITC loading was not blocked by a specific TRPA1 antagonist, HC030031, supporting that dye entry occurred via a connexin rather than TRPA1 (p=0.8643, Figure 4).

Schwann cells express two connexins: Cx32 and Cx31.3 (also known as Cx29) (Jeng et al., 2006; Gerber et al., 2021). Cx31.3 lacks the carbamylation motif and is, therefore, unlikely to be CO₂ sensitive. To confirm this, we measured ATP release via Cx31.3 expressed in HeLa cells (Liang et al., 2011) in response to changes in PCO₂ and membrane depolarization, by means of a co-expressed genetically encoded sensor, GRAB_ATP. HeLa cells transfected only with GRAB_ATP but not Cx31.3 did not show any fluorescent changes in response to 70 mmHg PCO₂ or 50 mM K⁺ (Figure 4—figure supplement 1). However, in cells transfected with Cx31.3, 50 mM KCl induced ATP release (Figure 4—figure supplement 1). By contrast, a stimulus of 70 mmHg PCO₂ was ineffective at triggering ATP release (Figure 4—figure supplement 1). We have previously shown that this level of PCO₂ readily induces ATP release via Cx32 (Butler and Dale, 2023; Lovatt et al., 2025). This confirms that Cx31.3 is not sensitive to CO₂ and makes it most likely that the CO₂-dependent entry of FITC into the Schwann cells was via Cx32.

Activity-dependent loading of FITC into Schwann cells depends on CO₂ production

To test whether Cx32 might open and permit FITC entry into the paranode during action potential propagation, we bathed isolated nerves in control aCSF (35 mmHg) and stimulated them electrically at 30 Hz, while measuring the compound action potential (CAP). Upon electrical stimulation, FITC entry into myelin was observed (Figure 5). We confirmed that FITC loaded into paranodes by counterstaining with the paranode marker Caspr (Einheber et al., 1997; Figure 5). FITC did not load into the axons. FITC loading into myelin was correlated positively with the stimulus duration (Figure 6B). Electrical stimulation of the axon was required for dye loading as it did not occur in nerves that were exposed to FITC for 10 min in the absence of stimulation (Figure 4A and B).

Figure 5. Activity dependent loading of FITC into Schwann cell paranodes.

(A) Representative images showing the FITC loading into mouse sciatic nerve bundles in response to different lengths of stimulation (30 Hz). Arrows indicate the paranode. (B) An isolated mouse sciatic nerve fibre loaded with the membrane impermeable dye FITC (30 Hz, 5 min) and counterstained with Caspr, an axonal membrane protein which is expressed only in the paranodal region. White arrows indicate the paranode of interest. Note the lack of loading into the axon. Scale bars – 15 μm.

Figure 6. Activity dependent loading of FITC is sensitive to manipulation of carbonic anhydrase activity.

(A) Representative images showing the FITC loading into mouse sciatic nerve bundles in response to 1 min of electrical stimulation in the absence (left) or presence (right) of the carbonic anhydrase inhibitor, acetazolamide. Arrows indicate the position of paranodes. Brightfield inset shows the presence of the nerve fibre and the arrowhead in the fluorescence image indicates its position. (B) Summary plot showing how the pixel intensity of Schwann cell paranodes, and, therefore, FITC loading, vary in response to stimulus duration and the presence of acetazolamide (each point mean ± SD). (C) Representative images showing the FITC loading into mouse sciatic nerve bundles in response to 5 min of stimulation in the absence (left) or presence (right) of the carbonic anhydrase enhancer, L-Phenylalanine. Brightfield inset shows the presence of the nerve fibre and the arrowhead in the fluorescence image indicates its position. (D) Boxplot showing the effect L-Phenylalanine had on FITC loading into mouse Schwann cell paranodes. Scale bars – 15 µm. Each point represents a separate ROI from five different nerves. L-Phe vs control MW test: p<0.0001.

Figure 6—figure supplement 1. Acetazolamide and L-Phe do not alter the compound action potential (CAP).

(A) Current-voltage input-output curves for the CAP, comprising data from all nerves subjected to either 100 µM acetazolamide or 1 mM L-Phenylalanine. Curves were produced before application of the respective compound (green circles), at the end of the pre-incubation of the compound (red circles) and at the end of dye loading (black circles) to show that the compounds had no effect on the amplitude of the CAP. N=5 nerves for each condition. Insets show the averaged CAP during incubation with the compound. (B) Boxplots showing the half maximum CAP amplitude for each respective condition and compounds laid out in (A). Kruskal-Wallis ANOVA: Acetazolamide, p=0.4441; L-Phenylalanine, p=0.9917.

To test whether this activity-dependent FITC loading was also CO₂ dependent, we first manipulated the activity of carbonic anhydrase (CA). Inhibition of CA activity, via acetazolamide, should increase the local PCO₂ as the conversion of CO₂ to carbonic acid will be slowed. We found that acetazolamide (100 µM) greatly increased FITC loading into the Schwann cell paranode in response to 30 Hz stimulation for 1 or 3 min (p=0.001 and p=0.0121, respectively, Figure 6A and B).

L-phenylalanine (L-Phe) is an allosteric enhancer of CA activity (Temperini et al., 2006). Myelinating Schwann cells express SLC7A5 (Gerber et al., 2021; Karlsson et al., 2021), the gene that encodes the L-type amino acid transporter. As this transports L-Phe (Nguyen et al., 2021), bath application of L-Phe (1 mM) to isolated nerve should be effective in enhancing the activity of intracellular CA in Schwann cells. The accelerated conversion of CO₂ to carbonic acid in the presence of L-Phe would be expected to reduce activity-dependent dye loading. We indeed observed that treatment with L-Phe greatly reduced activity-dependent FITC loading into the Schwann cell paranode (p=0.0159, Figure 6C and D). Neither acetazolamide nor L-Phe altered the amplitude or the current-amplitude curves of the CAP (Figure 6—figure supplement 1) indicating that these drugs did not affect the excitability of the axon.

CO₂ is produced from the Krebs cycle during two steps of oxidative decarboxylation. The first step of the Krebs cycle requires isomerization of citrate to isocitrate, via the enzyme aconitase, to enable the first decarboxylation event (via isocitrate dehydrogenase). In neurons, aconitase can be selectively blocked by 50 µM H₂O₂ (Tretter and Adam-Vizi, 2000). We, therefore, tested whether this dose of H₂O₂ could reduce activity-dependent FITC loading into the Schwann cell. Application of H₂O₂ greatly reduced FITC loading (Figure 7). There was no effect of H₂O₂ on the CAP (Figure 7—figure supplement 1), supporting the notion that Cx32 gating depends upon the Krebs cycle and production of CO₂.

Figure 7. Activity dependent loading of the FITC is reduced by inhibition of the Krebs cycle.

(A) Representative images showing the FITC loading into mouse sciatic nerve bundles in response to 5 min of 30 Hz stimulation in the absence (left) or presence (right) of the 50 µM H₂O₂ which blocks aconitase and the Krebs cycle. Brightfield inset shows the presence of the nerve fibre and the arrows in the fluorescence images indicate position of paranodes. Scale bar – 15 μm. (B) Boxplot showing the effect 50 µM H₂O₂ had on FITC loading into mouse Schwann cell paranodes. Each point represents a separate region of interest (ROI) from five different nerves. H₂O₂ vs control MW test: p<0.0001.

Figure 7—figure supplement 1. 50 µM H₂O₂ does not alter the compound action potential (CAP).

(A) Current voltage input-output curves for the CAP comprising data from all nerves subjected to 50 µM H₂O₂. Curves were produced before application of H₂O₂ (green circles), at the end of the pre-incubation (red circles) and at the end of dye loading (black circles) to show that H₂O₂ had no effect on the amplitude of the CAP. N=5 nerves for each condition. Inset shows the averaged CAP during incubation with the compound. (B) Boxplots showing the half maximum CAP amplitude for each respective condition in (A). Kruskal-Wallis ANOVA: p=0.6808.

As a final test of our hypothesis that activity-dependent CO₂ production in the axon gates Cx32 in the paranode, we used a specific blocker of AQP1, TC AQP1-1 (80 µM, Ghosh et al., 2020). We found blockade of AQP1 greatly reduced FITC loading into the Schwann cell paranode following 5 min of stimulation at 30 Hz, compared to that of WT (p<0.0001, Figure 8). This supports our hypothesis and also indicates that AQP1 is a key conduit for CO₂ to diffuse from the axonal node to the Schwann cell paranode. TC AQP1-1 had no effect on the amplitude or the current-amplitude curves of the CAP (Figure 8—figure supplement 1) indicating that it did not affect either the excitability of the axon or its capacity for action potential generation.

Figure 8. Activity dependent loading of the FITC is reduced by inhibition of AQP1.

(A) Representative images showing the FITC loading into mouse sciatic nerve bundles in response to 5 min of 30 Hz stimulation in the absence (left) or presence (right) of the AQP1 blocker TC AQP1-1. Brightfield inset shows the presence of the nerve fibre and the arrows in the fluorescence images indicate position of paranodes. Scale bar – 15 μm. (B) Boxplot showing the effect TC AQP1-1 had on FITC loading into mouse Schwann cell paranodes. Each point represents a separate region of interest (ROI) from five different nerves. TC AQP1-1 vs control MW U test: p<0.0001.

Figure 8—figure supplement 1. TC AQP1-1 does not alter the compound action potential (CAP).

(A) Current voltage input-output curves for the CAP comprising data from all nerves subjected to 80 µM TC AQP1-1. Curves were produced before application of the TC AQP1-1 (green circles), at the end of the pre-incubation with TC AQP1-1 (red circles) and at the end of dye loading (black circles) to show that TC AQP 1–1 had no effect on the amplitude of the CAP. N=5 nerves for each condition. Inset shows averaged CAP during incubation with TC AQP1-1. (B) Boxplots showing the half maximum CAP amplitude for each respective condition in (A). Kruskal-Wallis ANOVA: p=0.2516.

Figure 8—figure supplement 2. GDPβS blocks ATP-induced increases in paranodal Ca2²⁺ but does not alter activity dependent FITC loading into the paranode.

(A) Recording of ATP-induced increase in intracellular Ca²⁺ at the paranode (blue bar). When ATP was applied after 100 µM GDPβS (red bar), no increase in Fluo4 fluorescence was seen. (B) Images of Fluo4 fluorescence showing effect of ATP with and without GDPβS (same recording as A). Scale bar = 5 µm. (C) Boxplots summarizing that GDPβS blocks the change in fluorescence induced by ATP. Each point is a single paranode from a single nerve. (D) Images of FITC loading into sciatic nerve induced by 5 min electrical stimulation with and without 100 µM GDPβS. Arrows point to paranode, insets show corresponding brightfield images. Scale bar 10 µm. (E) Boxplots showing that GDPβS has no effect on activity dependent (30 Hz stimulation) FITC loading. Each point represents a separate region of interest (ROI) from five different nerves.

Figure 8—figure supplement 3. NH₄Cl does not alter FITC loading into the paranode.

(A) Images of FITC fluorescence in isolated nerves stimulated for 5 min at 30 Hz, in the control (35 mmHg) and after treatment with 100 µM NH₄Cl. Arrows indicate paranode. Scale bar 15 µm. (B) Summary graph of FITC fluorescence in nerves in response to 5 min stimulation in each condition. Each point represents a separate region of interest (ROI) from five different nerves.

Figure 8—figure supplement 4. Effects of acetazolamide and NH₄Cl on the intracellular pH of Schwann cell paranodes.

(A) Representative trace showing the normalized BCECF fluorescence ΔF/F₀ in response to changing pH to 6.5 and 7.15, 100 μM acetazolamide (AZ), and 100 μM NH₄Cl. (B) Representative images for the trace for the BCECF fluorescence in each condition. Scale bar = 5 µm. (C) Summary graph of effect of AZ and NH₄Cl on intracellular pH. Each point represents a nerve.

While our data are consistent with activity-dependent CO₂ production in the node-gating Cx32 in the paranode, they do not eliminate the possible involvement of other signaling pathways, such as those mediated by G-protein coupled receptors (GPCRs). We, therefore, used GDPβS (100 µM) as a general blocker of G-protein mediated signaling. As a positive control we showed that application of GDPβS blocked ATP receptor-mediated increases in intracellular Ca²⁺ in the paranode (Figure 8—figure supplement 2). However, the application of GDPβS had no effect on activity-dependent FITC loading (Figure 8—figure supplement 2).

Enhancement of FITC loading by block of CA is not mediated by pH changes

Inhibition of CA by acetazolamide could plausibly lead to subsequent alkalosis, as the production of HCO₃^- and H⁺ ions will be reduced. We, therefore, tested whether alkalosis by itself was sufficient to enhance activity-dependent FITC loading by applying NH₄Cl (100 µM), but this had no effect (p=0.1257, Figure 8—figure supplement 3).

We quantified the changes in intracellular pH induced upon perfusion of acetazolamide or NH₄Cl by using the pH-sensitive dye BCECF (Figure 8—figure supplement 4). We found that NH₄Cl induced greater increases in intracellular pH (change (median; 95% CI): 0.1579; 0.119 to 0.1968), than did acetazolamide which had no significant effect on intracellular pH (change: –0.0147; –0.040–0.011). The enhancement of activity-dependent FITC loading by acetazolamide cannot, therefore, be explained by changes in intracellular pH.

Activity-dependent loading of FITC depends on CO₂ binding to Cx32

To directly address both the involvement of Cx32 and specifically binding of CO₂ to Cx32 via the carbamylation motif, we utilized a dominant negative subunit, Cx32^DN. Cx32^DN carries the K124R and K104A mutations and can thus neither bind CO₂ nor form a salt bridge with a neighbouring subunit that has bound CO₂. We have previously shown that Cx32^DN removes CO₂ sensitivity from cells that express Cx32^WT (Butler and Dale, 2023). Using acceptor depletion FRET (Gu et al., 2004; van de Wiel et al., 2020) we documented that Cx32^DN coassembles with Cx32^WT (Figure 9—figure supplements 1 and 2). Furthermore, Cx32^DN formed gap junctions that retained their permeability to small molecules (Figure 9—figure supplement 3).

We, therefore, transduced sciatic nerve with AAV-Mpz-Cx32^DN-IRES-mCherry. This construct design uses the Mpz promoter to restrict expression to the Schwann cell. The Cx32^DN sequence is not tagged at the C-terminus but, as there is an IRES-mCherry sequence, cytosolic expression of mCherry permits identification of the transduced Schwann cells (van de Wiel et al., 2020). Expression of Cx32^DN greatly reduced activity-dependent dye loading into Schwann cells that expressed Cx32^DN but not in those that did not in the same nerve (Figure 9).

Figure 9. Activity-dependent dye loading into Schwann cells depends on CO₂ binding to connexin32 (Cx32).

(A) Images of single axons dissected from a sciatic nerve transduced with AAV-Mpz-Cx32^DN-IRES-mCherry. Axons that do not express Cx32^DN do not exhibit mCherry fluorescence and show robust dye loading to 15 Hz stimulation. By contrast, axons that express mCherry are Cx32^DN+ve and do not show dye loading during stimulation. Scale bar 15 µm. (B) Summary graph showing the pixel intensity of axons that are Cx32^DN+ve versus the control Cx32^DN-ve. MW p<0.0001, each circle an individual paranode from n=5 nerves.

Figure 9—figure supplement 1. Cx32^DN coassembles with Cx32^WT.

(A) Images of Clover and mRuby2 fluorescence before and after bleaching of the mRuby2 acceptor. The Clover fluorescence becomes brighter following bleaching for the Cx32^DN and Cx32^WT or Cx32^WT and Cx32^WT pairs but not for the Cx32^WT and Cx43^WT pair. (B) Values for the FRET efficiency and a comparison of bleaching efficiency for the different pairings.

Figure 9—figure supplement 2. Cx32^DN coassembles with Cx32^WT.

The dependence of the FRET efficiency (E) on the acceptor level (A) and donor to acceptor (D–A) ratio. A negative correlation between E and D-A ratio indicates coassembly into hexamers as opposed to random association in the membrane.

Figure 9—figure supplement 3. Cx32^DN forms functional gap junctions.

(A) Images of cells expressing connexin32^DN (Cx32^DN) under brightfield (BF), the mCherry tag (red), and NBDG, a fluorescent glucose analogue (green). NBDG is present in the patch pipette and readily diffuses between the cells. Scale bar 20 µm. (B) Changes in NBDG fluorescence over time in the Donor and Recipient cells showing the ready transfer of NBDG via Cx32^DN gap junction channels (n=8).

A simplified model of the paranode supports CA as a key regulator of Cx32 gating

To gain further insight, we made a simplified model of the paranode (as a single cell that in effect incorporated the nodal mitochondrion) to explore the effects of CA activity on loading of FITC into Schwann cell paranodes (see Methods and Figure 10). The mitochondrion in this simplified ‘paranode’ was based on a model proposed by Matsuda et al., 2020. The Matsuda model, which accurately replicates the experimentally observed dynamics of ATP production in mitochondria of myotubes, incorporates the concept of mitochondrial priming: that electrical activity in the myotube enhances the rate of ATP synthesis.

Figure 10. Simple model of CO₂ signaling at the paranode reproduces experimentally observed patterns of activity-dependent FITC loading.

(A) Adaptation of the Matsuda et al model to incorporate CO₂ production and metabolism via CAII. Action potentials provide the input to the model as activity variable X, which determines variables Y and Z, which determine ATP production and consumption. CO₂ is proportional to ATP production, via the rate constant α. Code for the model is provided in the MATLAB files Source code 1 and Source code 2. (B) Incorporation of the modified Matsuda et al model into a single cell that possesses a K⁺ leak channel and connexin32 (Cx32) and represents the paranode, albeit including the nodal mitochondria. (C) Outputs from the model to show how the change in [CO₂] and the consequent FITC loading evoked by two different durations of stimulation (1 and 5 min) varies with the V_max of CAII. Reduction from the control value (15 mM/s) to 1 mM/s simulates the effect of acetazolamide, whereas an increase of V_max to 30 mM/s simulates application of L-Phe. (D) A summary graph showing how FITC loading varies with stimulus duration with the three different values for the V_Max of CAII.

Figure 10—figure supplement 1. Inhibition of carbonic anhydrase increases loading of FITC into paranodes and outer myelin in the absence of electrical stimulation.

The images show the FITC fluorescence in the control and presence of 100 µM acetazolamide (AZ). The boxplot shows quantification of the fluorescence and that the increase in background fluorescence is significant (control vs AZ, MW test, p<0.0001). Each point represents a separate region of interest (ROI) from five different nerves.

We added a rate of CO₂ production that was proportional to mitochondrial ATP production, endowed the ‘paranode’ with a K⁺ channel to give it a resting potential, Cx32 and carbonic anhydrase. The CO₂-sensitive gating of Cx32 was based on the published CO₂ dose response curves (Huckstepp et al., 2010). FITC was assumed to only permeate open Cx32 hemichannels and its transmembrane concentrations were calculated according to the GHK equation assuming that FITC had a net negative charge of –1. CA activity was modeled with Michaelis-Menten kinetics with K_M being based on literature values for CAII. The V_max of CA was a free variable that could be altered to mimic the effect of inhibition or allosteric enhancement of CA.

We altered the duration of electrical stimulation of the ‘paranode’ from 1 to 20 min and calculated the amount of dye loading. With a V_max of 15 mM/s, this gave a graph that was very similar to the experimentally obtained data (Figure 10D, compare to Figure 6B). To simulate the effect of acetazolamide, we reduced the V_max of CA to 1 mM/s, and found an enhancement of dye loading that was once again very similar to the experimentally observed enhancement (Figure 10D, compare to Figure 6B). L-Phe can enhance the activity of CA by up to threefold. We found that increasing the V_max of CA twofold to 30 mM/s gave a very substantial reduction of dye loading that was similar to the experimentally observed effect of L-Phe (Figure 10D, compare to Figure 6B).

Our simplified model of the paranode suggests that CA is a key regulator of the local PCO₂ and hence Cx32 gating. We also observed that when inhibition of CA was simulated by a reduction of V_max to 1 mM/s, the concentration of CO₂ increased to a steady state value of 0.48 mM and there was a steady increase in FITC loading reaching a concentration of 0.7 µM after 30 min. Under the ‘control’ conditions [CO₂] had a steady state value of 0.05 mM and the FITC concentration after 30 min was only 20 pM. There is some support for this prediction of the model, as we observed that acetazolamide did indeed increase the background FITC loading of nerve fibres by a small but significant amount (Figure 10—figure supplement 1).

The Matsuda model explicitly incorporates mitochondrial priming by electrical activity, and its use in our model reproduces the experimentally observed dye loading. This suggests that mitochondrial priming might also occur in the node/paranode, although this remains to be tested directly.

Activity-dependent entry of Ca²⁺ into the paranode is CO₂-dependent

Our evidence so far supports the hypothesis that Cx32 is gated during action potential propagation by activity-dependent generation of CO₂ at the node. During electrical activity Ca²⁺ accumulates in the paranode (Lev-Ram and Ellisman, 1995). As we have previously shown Cx32 to be Ca²⁺ permeable (Butler and Dale, 2023), we tested whether this increase in paranodal Ca²⁺ could be caused by entry via the CO₂-dependent opening of Cx32.

To measure intracellular Ca²⁺ we loaded isolated mouse sciatic nerve with Fluo4-AM. We found that exposure of the nerve to hypercapnic aCSF (70 mmHg) increased Fluo4 fluorescence in paranode-like structures indicating an increase in intracellular Ca²⁺ (Figure 11). The CO₂-evoked increases in Fluo4 fluorescence were blocked by carbenoxolone indicating that they were channel-mediated most likely via Cx32 (Figure 11, p=0.0087 CBX compared to control).

Figure 11. CO₂ dependent Ca2²⁺ influxes into Schwann cells are hemichannel dependent.

(A) Representative trace showing change in normalized Fluo4 fluorescence in response to 70 mmHg aCSF (red bar), 35 mmHg aCSF with the non-specific hemichannel blocker carbenoxolone (100 µM, orange bar) and 70 mmHg aCSF plus 100 µM carbenoxolone (blue bar). (B) Representative images showing changes Fluo4 fluorescence in response to hypercapnic aCSF. The circles in the first panel show the measurement region of interests (ROIs) drawn around the paranodes. Scale bar = 10 µm. (C) Boxplot showing the change in normalized fluorescence (ΔF/F₀) in Fluo4 loaded Schwann cell paranodes evoked by 70 mmHg aCSF in the presence and absence of 100 µM carbenoxolone (CBX). Each datapoint consists of a paranode, with all the data collected from four sciatic nerves. Control vs CBX, MW test, p=0.0087.

Having established the existence of CO₂-dependent Ca²⁺ entry into the paranode, we next determined whether we could observe Ca²⁺ entry into the paranode during electrical stimulation and whether this was also CO₂ dependent. To measure intracellular Ca²⁺ we expressed GCaMP8 under the control of the Mpz promoter to ensure Schwann cell-specific expression (Figure 12A). Stimulation of the isolated sciatic nerve evoked increases in intracellular Ca²⁺ as reported by GCaMP8 fluorescence that could be enhanced by AZ and blocked by TC AQP1-1 (Figure 12B–D). Use of Fluo4-loaded sciatic nerves replicated these data and showed an increase of Ca²⁺ into Schwann cell paranodes during electrical stimulation (Figure 12—figure supplement 1). Crucially, these transient increases depended upon CO₂ production: they were significantly enhanced by acetazolamide (p<0.0001) and reduced by block of AQP1 by TC AQP1-1 (p<0.0001; Figure 12—figure supplement 1). Thus, the Ca²⁺ entry into the paranode during electrical stimulation depends on CO₂ generated by the axon entering the paranode and most likely opening Cx32. This is consistent with earlier reports that show that Ca²⁺ accumulation in the paranode requires extracellular Ca²⁺ (Lev-Ram and Ellisman, 1995).

Figure 12. Activity-dependent increase of intracellular Ca2²⁺ in Schwann cell paranodes.

(A) Superimposed brightfield (BF) and fluorescence image of GCaMP8 transduced nerve. To show expression at the paranode. (B) Representative GCaMP8 traces showing change in normalized fluorescence in response to 15 Hz electrical stimulation (red bar) in the presence of acetazolamide (AZ, green bar) or TC-AQP1-1 (purple bar). (C) The fluorescence images show before (Baseline), during stimulation of the nerve (15 Hz), during stimulation in presence of AZ (15+Az) and stimulation in the presence of TC AQP1-1 (15+TC AQP1-1). Scale bar = 5 µm; black dashed line indicates the shape of an individual fibre, white arrow indicates hotspot of GCaMP8 fluorescence at the paranode. (D) Boxplot showing the change in normalized GCaMP8. fluorescence (ΔF/F₀) evoked Schwann cell paranodes in response to 15 Hz electrical stimulation in the control, with AZ and with TC AQP1-1. Kruskal-Wallis ANOVA, p<0.0001. Pairwise MW comparisons: control vs AZ, p=0.0142; control vs TC AQP1-1, p<0.0001. Each datapoint consists of a paranode, with all the data collected from four sciatic nerves.

Figure 12—figure supplement 1. Activity-dependent Ca2²⁺ influxes into Schwann cell paranodes are dependent on CO₂.

(A) Representative trace showing change in normalized Fluo4 fluorescence in response to 15 Hz electrical stimulation (red bar) in the presence of 100 µM acetazolamide (blue bar), in control aCSF (green bar), or in the presence of 80 µM TC AQP1-1 (purple bar). (B) Boxplot showing the change in normalized fluorescence (ΔF/F₀) evoked in Fluo4 loaded Schwann cell paranodes in response to high frequency electrical stimulation in each condition. Each datapoint consists of a paranode, with all the data collected from four sciatic nerves. Kruskal-Wallis ANOVA, p<0.0001. Pairwise MW comparisons: control vs AZ, p<0.001; control vs TC AQP1-1, p<0.001. (C) Brightfield and fluorescence images showing the activity-dependent Ca²⁺ increase in a paranode in the presence of acetazolamide (AZ) and the lack of an activity-dependent increase in Ca²⁺ in the presence of TC AQP1-1.

Activity-dependent slowing of conduction velocity is CO₂-dependent

We observed that hypercapnic aCSF, FCCP, and electrical activity consistently induced FITC loading into the outer myelin layer, suggesting the occurrence of CO₂-dependent gating of Cx32 in this outermost membrane. Were this to occur, it should increase the leakage of current across the myelin sheath. Saltatory conduction depends on local current circuits travelling down the core of the axon to depolarize that next node (Huxley and Stampfli, 1949). If more current were to leak through the sheath before reaching the next node, there should be a small but measurable slowing of conduction velocity (Huxley and Stampfli, 1949; Bakiri et al., 2011). We would, therefore, predict that during more intense electrical activity in nerve, there should be more CO₂ production and thus a slowing of conduction velocity.

To test this, we measured the CAP firstly under low frequency stimulation (1 Hz), exposed the nerve to a period of high frequency stimulation (15 Hz for 10 min, to elevate local PCO₂) and then remeasured the CAP under low frequency stimulation (1 Hz). We found high frequency stimulation increased the delay from the stimulus artefact to the peak of the CAP by 0.11 ms (median, 95% CI: 0.04–0.17) (Figure 13). To demonstrate that this slowing was CO₂-dependent we manipulated the components of the CO₂ signaling system. 100 µM acetazolamide significantly increased the delay to the peak of the CAP caused by the high frequency stimulation (p=0.0016, Figure 13). Conversely, 1 mM L-Phe or 80 µM TC AQP1-1 reduced the effect of high frequency stimulation on the delay to the peak of the CAP (respectively, p=0.0317 and p=0.0079, Figure 13). Note that once again, TC AQP1-1 had no effect on the amplitude of the CAP (Figure 13D).

Figure 13. CO₂ dependent slowing of conduction velocity following high frequency stimulation.

The compound action potential (CAP) was evoked at 1 Hz and then 10 mins of 30 Hz stimulation was given prior to remeasuring the CAP at 1 Hz stimulation frequency. Representative CAPs from mouse sciatic nerve prior to high frequency stimulation (Black trace), and after (Blue trace) for WT nerves: (A) in the absence of any compound; (B) with 100 µM acetazolamide; (C) with 1 mM L-Phenylalanine; (D) with 80 µM TC AQP1-1. (E) Boxplot showing the change in latency (time to peak of CAP) before and after high frequency stimulation for Control, Acetazolamide (AZ), L-Phe, and TC AQP1-1. Kruskal-Wallis ANOVA p=0.0016. Pairwise MW tests: control vs AZ, p=0.0317; control vs L-Phe, p=0.0159; control vs TC AQP1-1, p=0.0079.

Figure 13—figure supplement 1. Model of the compound action potential (CAP) to show how changing conduction velocity alters the rise time, time to peak, and peak amplitude of the CAP.

(A) Shows the profile of an individual action potential (AP, top), the distribution of conduction delays based on the distribution of diameters of fibres in the sciatic nerve (middle), and the CAP arising from summation of 2000 individual APs occurring with different conduction delays (bottom, light blue line) and three different slowing factors. (B) Shows how the rise time, peak time and amplitude vary with the slowing factor. To mimic slowing, every conduction delay is multiplied by the slowing factor, when this is 1.0, no slowing occurs, and when, for example, it is 1.2 each conduction delay is slowed by 20%. MATLAB code for the model is provided in Source code 3.

We noticed that the period of high frequency stimulation broadened the CAP and slightly reduced its amplitude. This was particularly exaggerated in the presence of acetazolamide (Figure 13—figure supplement 1). To understand this effect on the shape of the CAP, we made a simple model of the CAP based upon 2000 individual axons each having an identically shaped action potential. To reflect the distribution of fibre diameters reported in sciatic nerve (Assaf et al., 2008) the conduction velocities were given a normal distribution skewed to lower velocities (Figure 13—figure supplement 1). The CAP was simply the sum of all of the individual action potentials. We then slowed the velocity of each fibre by the same proportion and computed the CAP for different amounts of slowing and calculated the 10–90% rise time, the time to peak, and peak amplitude of the CAP (Figure 13—figure supplement 1). This showed that under these simplified assumptions, changes in the shape of the CAP of the type we observed experimentally would be expected from slowing the conduction velocities in all fibres by the same proportion.

Discussion

Activity-dependent gating of Cx32

In this paper, we have investigated the mechanism of activity-dependent Cx32 hemichannel gating in peripheral myelin. Previously, the opening of Cx32 has been posited to depend on either its intrinsic voltage sensitivity (Abrams et al., 2002) or as a downstream consequence of an increase in cytosolic Ca²⁺ within the paranode (Stauch et al., 2012; Carrer et al., 2018). Here, we have tested an alternative hypothesis: that cell-to-cell signaling mediated via CO₂ produced in the axon is the primary trigger for Cx32 gating in the paranode. Our hypothesis explicitly links Cx32 opening in the paranode to the energetic demands of action potential propagation in the node.

In our experiments, we assessed Cx32 gating via entry of the membrane impermeant dye, FITC. Our results support our new hypothesis in several respects. First, Cx32 gating in response to axon stimulation was greatly reduced by blocking AQP1, which is CO₂ permeable (Endeward et al., 2006; Musa-Aziz et al., 2009; Michenkova et al., 2021) and thus provides a route for passage of CO₂ from axon to paranode. The inhibition of AQP1 had no effect on the CAP, eliminating a possible alternative interpretation that blocking this channel directly affected NaV1.8 (Zhang and Verkman, 2010).

Second, inhibition of CA with acetazolamide greatly increased the activity-dependent gating of Cx32. Third, facilitation of CA activity by applying an allosteric enhancer, L-Phe, greatly reduced activity-dependent gating of Cx32. Fourth, the effect of FCCP showed that mitochondrially generated CO₂ was sufficient to gate Cx32. Fifth, our use of low doses of H₂O₂ to block aconitase greatly reduced activity-dependent dye loading, indicating its dependence on a functioning Krebs cycle.

Finally, application of GDPβS to block all GPCR-based signaling had no effect on activity-dependent gating of Cx32. Together, these results suggest that CO₂ is acting as a cell-to-cell signal and is the prime trigger for Cx32 opening during action potential propagation. In the light of these results, it is interesting that elasmobranchs, the first vertebrates to evolve a fully myelinated nervous system (Salzer and Zalc, 2016), have an orthologue of Cx32 that has identical CO₂ sensitivity to human Cx32 (Dospinescu et al., 2019).

As there are no selective pharmacological blockers for Cx32, our evidence that Cx32 is the conduit for activity and CO₂-dependent FITC loading into the paranode is indirect. Nevertheless, our combined evidence is compelling for the following reasons. Cx32 is the only known large-pored channel expressed in Schwann cells that is directly sensitive to gaseous CO₂. We know that FITC permeates Cx32 and the CO₂ dose dependence of FITC loading matches that of Cx32. We have eliminated both Cx31.3 (not CO₂ sensitive) and TRPA1 (unaffected by a selective blocker of this channel) as the conduit. However, the strongest evidence to support our hypothesis is that expression of Cx32^DN in the Schwann cell blocks activity-dependent FITC loading. This shows that dye loading depends upon both Cx32 and CO₂ binding to Cx32.

Possible localization of the components required for CO₂ signaling and their relation to the energetics of action potential generation

Our data and previously published studies (Toews et al., 2007) support the localization of Cx32 in the paranode and outer myelin layers. AQP1 also localizes to both the paranode and axon either in the node or close to the node in the paranodal region of the axon. Nevertheless, Cx32 and AQP1 are not restricted to these locations and are found, for example, in the internodal regions. Colocalization analysis (restricted to the paranode/nodal regions) shows that in these regions Cx32 and AQP1 show significant proximity as do AQP1 and a mitochondrial marker CytC. Cx32 also shows significant colocalization with CytC. Mitochondria are thus likely to be present in both the paranode and node. This is consistent with other studies suggesting mitochondrial localization to the axonal node (Zhang et al., 2010; van Hameren et al., 2019). It should be noted that mitochondria are not restricted to the axonal node (Zhang et al., 2010).

In myelinated axons, the voltage-gated Na⁺ influx and K⁺ efflux occurs at the node of Ranvier. The transmembrane ionic gradients at the node need to be restored via the actions of Na⁺-K⁺ ATPases. The nodes of Ranvier are likely, therefore, to be the major sites of ATP generation and consumption and thus the production of CO₂. However, as we cannot directly measure CO₂ production, the site of its production remains a matter of supposition.

The binding site for CO₂ on Cx32 is intracellular and CO₂ must, therefore, cross both the nodal and paranodal membranes. The localization of the key components (Cx32, AQP1, mitochondria, and CA) in the nodal/paranodal region will shorten the diffusion path between the source of CO₂ (mitochondria) and its ultimate target Cx32. This would potentially speed the dynamics of the CO₂ signal. CA, which provides an efficient removal mechanism for CO₂, shows restricted localization to the paranode. This supports our hypothesis of CO₂ entry through paranodal AQP1.

The important roles of AQP1 and carbonic anhydrase

Whilst there has been controversy over the role of CO₂ permeable channels in enabling transmembrane CO₂ fluxes, it is now accepted that biological membranes are only poorly permeable to CO₂ and a channel-mediated mechanism is required (Boron et al., 2011). Our data further support this idea, as blockade of AQP1 prevents the activity-dependent gating of Cx32. Given that our data also show that CA activity limits the gating of Cx32, the colocalization of AQP1 and Cx32 may be important. As AQP1 will be the entry point for CO₂ into the paranode, its colocalization with Cx32 may favour CO₂ binding to Cx32 over capture and conversion to carbonic acid by CA.

Our simplified model of the paranode sheds further light on the regulation of CO₂ signaling and the gating of Cx32. The Matsuda model (Matsuda et al., 2020) incorporates priming of mitochondrial ATP production by electrical activity and hence this is also implicit in our paranode model. Mitochondrial priming will make CO₂ production more rapid than if it depended on ATP depletion to occur first. This implies that CO₂ production could vary relatively quickly with activity patterns and thus report the dynamics of action potential firing. It will be important to directly test this prediction by measuring the dynamics of mitochondrial ATP production in the node relative to imposed electrical activity.

Our model also predicts that complete inhibition of CA will give some Cx32 gating under non-stimulated conditions. We observed increased baseline loading of FITC into the nerves during acetazolamide giving support for this prediction. This suggests that an important role of CA is to prevent basal rates of CO₂ production being sufficient to gate Cx32. Our model also suggests that CA activity regulates the extent to which activity-dependent production of CO₂ can gate Cx32. The effects of L-Phe and acetazolamide lend support to this prediction. Thus, the model suggests that CA controls the dynamic range of the CO₂ signal and is likely to be an important regulator of CO₂-mediated signaling. By keeping paranodal cytosolic PCO₂ low, CA not only reduces the basal gating of Cx32 but importantly also maintains a concentration gradient that favours entry of CO₂ into the paranode.

Physiological consequences of CO₂ dependent signaling

We have further shown that two other aspects of Schwann cell physiology and function depend on the CO₂-dependent gating of Cx32. First, the well-documented increase of intracellular Ca²⁺ into the paranode evoked by nerve stimulation appears to largely depend on the opening of Cx32 and can be modified in the same way as dye loading by manipulating CA activity or blocking AQP1. Second, given the localization of Cx32 in the outer myelin layer and the observation of activity- and CO₂-dependent dye loading into that outer layer, we hypothesized that there should be activity-dependent slowing of nerve conduction. This is because any opening of Cx32 hemichannels should reduce the resistance to current flow across the myelin sheath. We did indeed observe a small degree of activity-dependent slowing of nerve conduction velocity. Crucially, this too depended upon CO₂ production and could be altered by manipulating CA activity or blocking AQP1.

Links to CMTX

Given our evidence suggests that the CO₂ sensitivity of Cx32 is critical for its gating during action potential propagation, we might expect that any mutations that affect this sensitivity could precipitate CMTX. We have previously examined the effect of 14 CMTX mutations on the CO₂ sensitivity of Cx32 (Butler and Dale, 2023). We found that five completely removed its CO₂ sensitivity, three greatly reduced its sensitivity while the remainder had no apparent effect. It should be noted that two mutations of K124 (Bone et al., 1997; Fattahi et al., 2017) and K104 (Williams et al., 1999; Wang et al., 2015), the critical residues for detection of CO₂ (K104E, K104T, K124E, and K124N, not included in our published study), have also been identified as possible CMTX mutations.

These results, while supportive of our hypothesis, are not conclusive as several of the eight CMTX mutations that altered CO₂ sensitivity have been documented to also affect other facets of channel function. However, the CMTX mutation E102G stands out as causing moderate severity CMTX while still permitting the formation of gap junction channels and hemichannels with apparently normal voltage dependence and ATP permeability. Because E102G involves the loss of CO₂ sensitivity of the hemichannel in the absence of other known functional effects on the hemichannel (Abrams et al., 2003) it lends some support to the hypothesis that the CO₂-dependence of Cx32 may be important for the health of myelin and that its loss could precipitate CMTX. Further exploration of CO₂- and Cx32-dependent signaling in myelin may suggest new strategies to treat peripheral neuropathies and peripheral nerve injury.

Methods

Key resources table.

Reagent type (species) or resource	Designation	Source or reference	Identifiers	Additional information
Cell line (human)	HeLa DH	UK Health Security Agency	RRID:CVCL_2483
Chemical compound, drug	DMEM	Merck Life Sciences UK Ltd	CAT# D6046
Chemical compound, drug	Fetal Bovine Serum	Labtech.com	CAT#FCS-SA
Chemical compound, drug	NEBuilder HiFi DNA assembly Master Mix	New England Biolabs	CAT#FCS-SA
Chemical compound, drug	PEI Prime linear polyethylenimine	Merck Life Sciences UK Ltd	CAT#919012
Chemical compound, drug	EZ-PCR Mycoplasma detection kit	Sartorius	CAT#20-700-20
Chemical compound, drug	FITC	Merck Life Sciences UK Ltd	CAT #46950
Chemical compound, drug	FCCP	APExBIO	CAT #B5004
Chemical compound, drug	L-Phenylalanine	Merck Life Sciences UK Ltd	CAT #P2126
Chemical compound, drug	TC AQP1-1	Tocris	CAT #5412
Chemical compound, drug	GDPβS	Merck Life Sciences UK Ltd	CAT #G7637
Chemical compound, drug	ATP	Merck Life Sciences UK Ltd	CAT #A26209
Chemical compound, drug	Paraformaldehyde	Merck Life Sciences UK Ltd	CAT # 158127
Chemical compound, drug	Bovine Serum Albumin	Merck Life Sciences UK Ltd	CAT #A7030
Chemical compound, drug	Fluorshield mounting medium with DAPI	Merck Life Sciences UK Ltd	CAT# F6057
Chemical compound, drug	Acetazolamide	Merck Life Sciences UK Ltd	CAT# A6011
Chemical compound, drug	Ammonium chloride	Invitrogen	CAT# A15000.0B
Chemical compound, drug	BCECF-AM	Invitrogen	CAT #B1170
Chemical compound, drug	Fluo4-AM	Invitrogen	CAT #F14201
Chemical compound, drug	DMSO	Merck Life Sciences UK Ltd	CAT #D5879
Chemical compound, drug	Pluronic F-127	Thermo Fisher Scientific	CAT# P3000MP
Chemical compound, drug	Sylgard–184	Scientific Lab Supplied Ltd	CAT# 63416.5 S
Sequence-based reagent	Cx31.3 (Gjc3) Forward	IDT	PCR primers	TTTGGCAAAGAATTCGGTACCATGTGTGGCAGGTTCCTGC
Sequence-based reagent	Cx31.3 (Gjc3) Reverse	IDT	PCR primers	CCGGTGGATCCCGGGCCCGCGGTACCCCGGCATCTCTGGGTCCAACTG
Sequence-based reagent	Cx32 (Gjb1) Forward (for Clover)	IDT	PCR primers	TTTGGCAAAGAATTCGGTACCATGAACTGGACAGGTTTGTACACCTTGCTC
Sequence-based reagent	Cx32 (Gjb1) Reverse (for Clover)	IDT	PCR primers	CCATGAATTCGCAGGCCGAGCAGCGGTC
Sequence-based reagent	EF Clover Forward	IDT	PCR primers	CTCGGCCTGCGAATTCATGGTGAGCAAG
Sequence-based reagent	EF Clover Reverse	IDT	PCR primers	CTTGATACTTACCTGCGGCCTCGAGCTAGCATTTAGGTGACAC
Sequence-based reagent	Cx32 (Gjb1) Forward (for Ruby)	IDT	PCR primers	TTTGGCAAAGAATTCGGTACCATGAACTGGACAGGTTTGTACACCTTGCTC
Sequence-based reagent	Cx32 (Gjb1) Reverse (for Ruby)	IDT	PCR primers	CCATGAATTCGCAGGCCGAGCAGCGGTC
Sequence-based reagent	EF Ruby Forward	IDT	PCR primers	CTCGGCCTGCGAATTCATGGTGTCTAAGG
Sequence-based reagent	EF Ruby Reverse	IDT	PCR primers	CTTGATACTTACCTGCGGCCTCGAGTTACTTGTACAGCTCGTC
Sequence-based reagent	Cx43 (Gja1) Forward (for Clover)	IDT	PCR primers	TTTGGCAAAGAATTCGGTACCGCGGGCCCGGGATCCACC
Sequence-based reagent	Cx43 (Gja1) Reverse (for Clover)	IDT	PCR primers	CCATGAATTCGATCTCCAGGTCATCAGGCCGAGG
Sequence-based reagent	EF Clover Forward	IDT	PCR primers	CCTGGAGATCGAATTCATGGTGAGCAAG
Sequence-based reagent	EF Clover Reverse	IDT	PCR primers	CTTGATACTTACCTGCGGCCTCGAGCTAGCATTTAGGTGACAC
Sequence-based reagent	Cx43 (Gja1) Forward (for Ruby)	IDT	PCR primers	TTTGGCAAAGAATTCGGTACCGCGGGCCCGGGATCCACC
Sequence-based reagent	Cx43 (Gja1) Reverse (for Ruby)	IDT	PCR primers	CCATGAATTCGATCTCCAGGTCATCAGGCCGAGG
Sequence-based reagent	EF Ruby Forward	IDT	PCR primers	CCTGGAGATCGAATTCATGGTGTCTAAGG
Sequence-based reagent	EF Ruby Reverse	IDT	PCR primers	CTTGATACTTACCTGCGGCCTCGAGTTACTTGTACAGCTCGTC
Recombinant DNA reagent	pDisplay-GRAB_ATP1.0	Addgene	Plasmid#167582; RRID:Addgene167582
Recombinant DNA reagent	AAV9-MPZmini-LCK-GCaMP8	BrainVTA	This paper
Recombinant DNA reagent	AAV9-MPZmini-dnCx32-IRES-mCherry	BrainVTA	This paper
Software, algorithm	GraphPad Prism	https://www.graphpad.com/features	RRID:SCR_002798
Software, algorithm	ImageJ/FIJI	https://imagej.net/software/fiji/downloads	RRID:SCR_002285

All experiments were performed in accordance with the United Kingdom Home Office Animals (Scientific Procedures) Act (1986) with project approval from the University of Warwick’s AWERB and licence PP7458325.

Sciatic nerve isolation

All mice used were C57BL/6, aged at least 6 weeks. Sciatic nerves were isolated following the protocol described in Rich and Brown, 2018. Placing a few drops of ice-cold aCSF loosened the perineural membrane allowing its easy removal with forceps, beginning at one cut end of the nerve, and moving inward. Once the perineural membrane was removed, forceps were carefully placed in between the seams of the larger bundle of fibres before being teased apart with careful lateral movement. Removal of the perineural membrane and slight teasing was sufficient to obtain dye loading, with extensive dissection to small bundles or individual fibres only occurring post-fixation for immunohistochemistry and visualization.

Immunocytochemistry

Antibodies used:

Antibody	Supplier	ID	RRID	Dilution	Reference
KCNQ2 (Kv7.2)	Abcam	ab22897	AB_775890	1/500	Kapell et al., 2023
KCNA2 (Kv1.2)	Sigma Aldrich	MABN77	AB_10806493	1/500	Otani et al., 2020
Caspr	Sigma Aldrich	MABN69	AB_10806491	1/250	Sánchez-de la Torre et al., 2022
Connexin 32	Thermo Fisher	13–8200	AB_2533037	1/250	Fowler et al., 2013
Aquaporin-1	Bioorbit	orb10122	AB_10751997	1/250	Pelagalli et al., 2018
Sideroflexin-1	Proteintech	12296–1-AP	AB_2185814	1/1000	Fowler et al., 2013
Cytochrome C	Invitrogen	54205-RBM6-P0	AB_3678654	1/500	Validated by supplier
Carbonic Anhydrase 2	Invitrogen	PA5-78897	AB_2746013	1/500	Zigo et al., 2022

Sciatic nerves were first washed with PBS three times before being fixed in 4% PFA for 45 min. Nerves were then washed in PBS three times and blocked using PBS containing 4% BSA and 0.1% Triton X-100 for 24 hr. Nerves were teased prior to immunostaining. Primary antibody was diluted in PBS containing 4% BSA and 0.1% Triton X-100 and added to nerves and left to incubate, constantly shaking, for 48 hr at 4°C. Nerves were then washed using PBS containing 0.1% Triton X-100 six times at 10 min intervals. The appropriate secondary antibodies diluted in PBS containing 4% BSA and 0.1% Triton X-100 were added to coverslips and left to incubate, constantly moving, for 2.5 hr. The secondary antibody was washed using PBS containing 0.1% Triton X-100 six times at 10 min intervals. Nerves were again blocked for 24 hr. To co-stain with a further primary antibody from the same species, antibody conjugation was used (ProteinTech FlexAble corallite). Conjugated antibodies were diluted in PBS containing 4% BSA and 0.1% Triton X-100 and added to nerves and left to incubate, constantly moving, for 48 hr at 4℃. The conjugated antibody was washed using PBS containing 0.1% Triton X-100 six times at 10 min intervals. Nerves were then placed onto a glass slide and further dissected using the tips of hypodermic syringes, yielding individual nerve fibres. Nerves were then mounted using Fluorshield with DAPI mounting medium (Sigma-Aldrich, Cat# F6057), placing a glass coverslip on top. Nerve fibres were subsequently imaged using the Zeiss-880 and Zeiss 980 confocal LSMs, specifically using the 488, 561, and 630 nm lasers. FIJI software was used for further analysis. Images were also taken on a Nikon N-SIM S with dual camera, utilizing a 100 X oil immersion lens. 470, 561, and 640 nm lasers were used.

Solutions used

Control (35 mmHg PCO₂) aCSF: 124 mM NaCl, 3 mM KCl, 2 mM CaCl₂, 26 mM NaHCO₃, 1.25 mM NaH₂PO₄, 1 mM MgSO₄, 10 mM D-glucose saturated with 95% O₂/5% CO₂, pH 7.4.

Hypercapnic (70 mmHg) aCSF: 73 mM NaCl, 3 mM KCl, 2 mM CaCl₂, 80 mM NaHCO₃, 1.25 mM NaH₂PO₄, 1 mM MgSO₄, 10 mM D-glucose, saturated with ~12% CO₂ (with the balance being O₂) to give a pH of 7.4.

Depolarizing (35 mmHg PCO₂) aCSF: 77 mM NaCl, 50 mM KCl, 2 mM CaCl₂, 26 mM NaHCO₃, 1.25 mM NaH₂PO₄, 1 mM MgSO₄, 10 mM D-glucose saturated with 95% O₂/5% CO₂, pH 7.4.

Pharmacological agents

Compound	Supplier	Concentration (μM)
Acetazolamide	Sigma Aldrich A6011	100
Ammonium chloride	Invitrogen A15000.0B	100
GDPβS	Sigma Aldrich G7637	100
L-Phenylalanine	Sigma Aldrich P2126	1000
TC AQP1-1	Tocris 5412	80

Dye loading assay

Isolated nerves were obtained as described above and slightly teased apart with sharp needles, as this was found to produce more profound and reliable dye-loading.

CO₂-dependent dye loading

Isolated sciatic nerves were first washed in control aCSF before being superfused with either 35 mmHg aCSF or 70 mmHg aCSF containing 50 μM fluorescein isothiocyanate (FITC) for 10 min. To induce endogenous CO₂ production via mitochondrial uncoupling isolated nerves were superfused with 35 mmHg aCSF containing 10 μM FCCP (APExBIO) for 10 min. Following this, FITC was washed off by superfusion with 35 mmHg aCSF with no dye for 10 min. The nerves were then transferred through a series of vessels with 35 mmHg aCSF to remove any remaining FITC.

Dye loading triggered by electrical stimulation

Nerves were pre-incubated in 35 mmHg aCSF, and any desired pharmacological agent, for 10 min prior to recording. Polished glass suction electrodes wrapped with a silver wire and backfilled with aCSF were used for stimulation and recording. The ends of nerves were gently sucked up into the suction electrodes such that orthodromic recordings were made, described in Rich and Brown, 2018.

The recordings of the stimulus-evoked CAPs were controlled by a National Instruments A/D interface (Model PCIe 6321) using the Strathclyde electrophysiology software program, WinWCP. A stimulator, Digitimer model DS3, was used to stimulate the nerve. The signal was amplified 1000× by an A-M Systems Inc Model 3000 AC/DC differential amplifier (A-M Systems, Sequim, WA 98382, USA). The signal was filtered at 20 kHz and 1 Hz and acquired at 20 kHz. To assess the validity of the CAP, the nerve was crushed between forceps at the conclusion of the experiment, leaving only the transient artefact.

Once a recording of the CAP had been successfully established, nerves were exposed to FITC during electrical stimulation at 30 Hz of different durations (1–10 min). They were then washed to remove FITC as described above. As each mouse possesses two sciatic nerves, when pharmacological agents were used, one nerve from each animal would be stimulated in the absence of any pharmacological agents as a matched control. I-V curves were recorded prior to drug pre-incubation, during pre-incubation, and following stimulation, to assess any effects of the used compound on the CAP. From the CAP traces rise time, rate of rise and latency could be calculated using WinWCP.

Fixation and imaging of dye-loaded nerves

Nerves were then fixed using 4% PFA for 45 min. Nerves were subsequently imaged using Zeiss 880 or 980 confocal LSMs, specifically using the 488, 561, and 633 nm lasers. FIJI software was used for further analysis. The statistical replicate was a single region of interest (ROI) and these were obtained from five nerves for each condition.

Measurement of intracellular pH with BCECF

Mouse sciatic nerve was dissected as previously described. BCECF-AM dissolved in DMSO was diluted into 35 mmHg aCSF to a final concentration of 2.5 μM. A hypodermic needle was blunted and joined to a fine glass capillary via a short length of tubing. Etched tungsten wire was used to make a small incision in the middle of the nerve, from which the nerve was teased open slightly. Whilst holding the incision open, the capillary loaded with BCECF was inserted and injected. The nerve was placed into 35 mmHg aCSF to wash for 3 min. The nerve was then placed into a recording chamber, immobilized with a platinum wire harp, and superfused with 35 mmHg aCSF. The BCECF-loaded nerves were imaged by epifluorescence (Scientifica Slice Scope, Cairn Research OptoLED illumination, 60 x water Olympus immersion objective, NA 1.0, Hamamatsu ImagEM EM-SSC camera, Metafluor software). BCECF was excited using 470 nm LED, with fluorescent emission being recorded every 4 s between 507 and 543 nm. Once a stable fluorescence baseline was reached, the various test solutions were superfused onto the nerve. Intracellular pH was then calibrated using Nigericin (James-Kracke, 1992).

The statistical replicate was a single nerve and 5 nerves were recorded for each condition.

Adeno-associated viral (AAV) constructs used

All AAVs were commercially produced (brainVTA) using the AAV9 serotype. To direct gene expression to Schwann cells, a minimal P0 (Mpz-mini) promoter was used (Sargiannidou et al., 2015; Kagiava et al., 2021; Georgiou et al., 2023). GCaMP8 was tagged with an LCK sequence, tethering it to the inner Schwann cell membrane. Cx32^DN (Butler and Dale, 2023) was flanked by IRES-mCherry (van de Wiel et al., 2020), giving cytoplasmic mCherry in transduced Schwann cells. Aliquots were stored at –80 until used.

Transduction of sciatic nerve in vivo

Surgical procedures were performed under the authority of the UK Home Office Licence PP7458325. Anaesthesia was induced by inhalation of isoflurane (4%; Piramal Healthcare Ltd, Mumbai, India) in pure oxygen (4 L·min^–1). The mouse was then placed on a temperature regulated heating pad (TCAT-2LV, Physitemp) to maintain body temperature. A face mask was used to maintain anaesthesia (isoflurane, intranasal, 0.5–2.5% in pure oxygen 1 L·min^–1) throughout the surgery. Atropine was provided (subcutaneous, 0.05  mg/kg) before surgery to stop pleural effusion. Adequacy of anaesthesia was assessed by respiratory rate, body temperature, and pedal withdrawal reflex. Preoperative meloxicam (subcutaneous, 2  mg/kg) and postoperative buprenorphine (subcutaneous, 0.05  mg/kg) were provided for analgesia. Unilateral intraneural injection into the sciatic nerve were performed on six- to eight-week-old male C57BL/6 mice, using 600–900 nL of AAV. The injections were performed manually, using a graduated micropipette attached to a 1 ml syringe, at a rate of ~200  nl/min. The micropipette was left in the nerve for 5 min following injection before removal. If any animal showed signs of pain in the days following surgery, additional analgesia (oral meloxicam) was administered as required. Postoperatively, 4–5 weeks were allowed to achieve maximal AAV expression, before dissection, electrophysiology, and imaging as previously described.

Measurement of intracellular Ca²⁺ with GCaMP8 or Fluo4

Nerves were placed into the recording chamber and anchored with a platinum harp. The nerve was perfused with 35 mmHg aCSF until a stable baseline was reached. The desired solution was then perfused until a stable level had been reached before being washed.

For experiments that utilized Fluo4 imaging, Fluo4-AM dissolved in Pluronic F-127 (Thermo Fisher Scientific P3000MP) with constant sonication and vortexing and was diluted into 35 mmHg aCSF to a final concentration of 2.5 μM. Nerves were then incubated for 20 min before being washed in 35 mmHg aCSF.

To enable simultaneous electrical stimulation and imaging of GCaMP8 or Fluo4 fluorescence, the nerves were mounted between electrodes within a bespoke micro-perfusion chamber constructed of Sylgard–184. Proprietary software was used to control nerve stimulation at 15 Hz, record the CAP and perform offline analysis.

Nerves were imaged by epifluorescence (Scientifica Slice Scope, Cairn Research OptoLED illumination, 60 × water Olympus immersion objective, NA 1.0, Hamamatsu ImagEM EM-SSC camera, Metafluor software). GCaMP8 or Fluo4 were excited with a 470 nm LED, and fluorescent emission between 507 and 543 nm recorded every 4 s.

The statistical replicate was a single ROI (paranode) and these were obtained from 5 nerves for each condition.

Measurement of activity-dependent conduction velocity slowing

Using isolated nerves, CAPs were recorded as described above using the Strathclyde electrophysiology software. The baseline of CAP was recorded at low frequency (1 Hz) for 30 s. High frequency stimulation (30 Hz) was applied for 10 min. Following this, the nerves were then stimulated for 30 s at 1 Hz. The CAPs from before and after high frequency stimulation were averaged and compared. A recording from an isolated nerve was considered as a statistical replicate.

Cell culture and transfection

The Cx31.3 gene sequence were synthesized by IDT and subcloned into the pCAG-GS-mCherry vector. DNA gBlock was amplified using PCR with primers (IDT). Plasmids were generated using Gibson assembly. The presence of the correct assembly was confirmed by DNA sequencing (GATC biotech). The Cx31.3 construct was inserted upstream of mCherry, with a short 12 amino acid linker (GVPRARDPPVAT).

pDisplay-GRAB_ATP1.0-IRES-mCherry-CAAX was a gift from Yulong Li (Addgene plasmid # 167582; http://n2t.net/addgene:167582; RRID:Addgene_167582).

Parental HeLa DH cells (obtained directly for the study from UK Health Security Agency and authenticated by ECACC) were grown in low-glucose Dulbecco’s modified Eagle medium (DMEM) supplemented with 10% FBS and 50 μg/ml penicillin/streptomycin. Regular testing ensured that they were free from mycoplasma infection. The HeLa DH cells were plated onto coverslips at a density of 7.5×10⁴ cells per well of a six-well plate and transiently transfected using a mixture of 1 μg each of the Cx31.3 construct and GRAB_ATP and 3 μg PEI for 6 hr. Cells were imaged 48 hr after transfection. We used a protocol to measure ATP release from cells developed and described in our previous work (Butler and Dale, 2023).

Analysis of GRAB_ATP fluorescence

Analysis of GRAB_ATP was performed in ImageJ (Schneider et al., 2012). Cell recordings were corrected for any motion using the Image Stabilizer plugin (Li, 2008). For cells expressing both Cx31.3 and GRAB_ATP, an ROI was drawn around the GRAB_ATP expression and median fluorescence measured for each image. The fluorescence pixel intensity (F) was normalized to the baseline fluorescence (F₀). The change in normalized fluorescence (ΔF/F₀) evoked by each stimulus, CO₂ and 50 mM KCl, was recorded for each cell.

We converted changes in normalized fluorescence evoked by 70 mmHg pCO₂ and 50 mM KCl into the concentration of ATP released by normalizing them to the ΔF/F₀ produced by a 3 μM ATP calibration solution. Over this range, the calibration curve for GRAB_ATP is approximately linear (Wu et al., 2022; Butler and Dale, 2023). Statistical comparisons were performed considering each cell as an independent replicate. Five transfections were performed.

Analysis of immunohistochemical colocalization

The JACOP plugin (Bolte and Cordelières, 2006) was used to calculate the Manders’ coefficients M1 and M2. The convention we have used throughout the paper is that for colocalization of A with B, M1 represents the proportion of A pixels that overlap with B pixels, and M2 would represent the proportion of B pixels overlapping with A pixels. Colocalization analysis was restricted to the nodal/paranodal regions.

As a control, one channel was rotated 90° and analysis was re-run using the same thresholds. Each datapoint represents an ROI or a paranode with all data points coming from at least five nerves.

Modeling of paranode

The paranode was modeled as a cell with a mitochondrion, a K⁺ leak channel, Cx32, and CA. Mitochondrial ATP production was modeled as per Matsuda et al., 2020. This involves a time-dependent variable Y that stimulates ATP production, where X is a variable that is proportional to the duration of electrical stimulation, Y₀ is the steady state value of Y and relaxes to that value with a time constant of τ₁. K_d1 and n₁ are parameters for the Hill equation that determines how variable X alters the value of Y:

$${\displaystyle {\displaystyle \frac{dY}{dt}=\frac{X^{n_1}}{K_{d_1}\text{+}{X}^{n_1}}-\frac{Y-Y_0}{{\tau }_1}}}$$ (1)

A time-dependent variable Z determines the use of ATP:

$${\displaystyle {\displaystyle \frac{dZ}{dt}=\frac{X^{n_2}}{K_{d_2}+X^{n_2}}-\frac{Z-Z_0}{{\tau }_2}}}$$ (2)

where Z₀ is the steady state value of Z and relaxes to that value with a time constant of τ₂ and K_d2 and n₂ are parameters for the Hill equation that determine how variable X alters the value of Z.

The rate of change of ATP concentration is thus:

$${\displaystyle {\displaystyle \frac{dATP}{dt}=k_1.Y-k_2.Z.ATP}}$$ (3)

Where k₁ and k₂ are rate constants for synthesis and breakdown of ATP, respectively.

To adapt this model to the paranode, we first defined the rate of CO₂ production as proportional to ATP production, and used the Michaelis Menten equation to calculate CO₂ conversion to carbonic acid:

$${\displaystyle {\displaystyle \frac{dC{O}_2}{dt}=\alpha .k_1.Y-\frac{V_{max}.C{O}_2}{C{O}_2+K_m}-k_s.C{O}_2}}$$ (4)

Where V_max and K_m are the maximal velocity of CA and affinity of CA for CO₂, respectively, α is a rate constant for the production of CO₂. For completeness, we also allowed for spontaneous conversion of CO₂ to carbonic acid–determined by the first order rate constant, k_s. Variables Y and k₁ have the same meaning as in Equation 3.

The rate of membrane potential change of the paranode was calculated from:

$${\displaystyle {\displaystyle \frac{dV}{dt}=-C_m.\left(I_K+I_{Cx32}+I_{FITC}\right)}}$$ (5)

Where C_m is the whole cell membrane capacitance, I_K the K⁺ leak current, I_Cx32 the current through Cx32 and I_FITC the current carried by FITC.

I_K is described by the Goldman Hodgkin Katz (GHK) equation:

$${\displaystyle {\displaystyle I_K=P_K.\frac{z^{2}V{F}^2}{RT}.\frac{K_i-K_o.e^{-zVF/RT}}{1-e^{-zVF/RT}}}}$$ (6)

Where P_K is the maximal whole cell permeability to K⁺, z the valence. R is the universal gas constant, T the absolute temperature (in Kelvin) and F the Faraday constant. Ki and Ko are respectively the intracellular and extracellular concentrations of K⁺.

I_Cx32 is described by:

$${\displaystyle {\displaystyle I_{Cx32}=G_{Cx32}.\frac{C{O}_2^H}{C{O}_2^H+K_{Cx32}}.\left(V-V_{rev}\right)}}$$ (7)

Where G_Cx32 is the maximal whole cell conductance for Cx32, K_Cx32 is the affinity of Cx32 for CO₂, H is the Hill coefficient of CO₂ binding and V_rev is the reversal potential of the current through Cx32.

I_FITC (through Cx32) is described by:

$${\displaystyle {\displaystyle I_{FITC}=P_{Cx32}.\frac{C{O}_2^H}{C{O}_2^H+K_{Cx32}}.\frac{z^{2}V{F}^2}{RT}.\frac{FIT{C}_i-FIT{C}_o.e^{-zVF/RT}}{1-e^{-zVF/RT}}}}$$ (8)

Where P_Cx32 is the maximal whole cell permeability of Cx32 to FITC and z the valence of FITC. R is the universal gas constant, T the absolute temperature (in Kelvin) and F the Faraday constant. FITC_i and FITC_o are respectively the intracellular and extracellular concentrations of FITC. K_Cx32 and H have the same meaning as in Equation 7.

The rate of change of FITC_i with time is thus:

$${\displaystyle {\displaystyle \frac{dFIT{C}_i}{dt}=\frac{I_{FITC}}{F}}}$$ (9)

Where F is the Faraday constant.

These differential equations were coded with Matlab, and a fourth order Runge-Kutta ODE solver with adaptive step size used to numerically integrate them and thus calculate the production of CO₂ during electrical stimulation and the extent FITC loading. The MATLAB code along with a command line interface is presented in the files: Source code 1 and Source code 2.

Modeling of CAP

Using Matlab, a compound action potential was computed from summing 2000 individual action potentials based on the product of two Boltzmann equations to give a realistic shape (Source code 3). The action potentials were given a random delay (representing conduction velocity) based on a mean ± SD described by a Gaussian distribution with a skew factor. This was chosen to reflect the skewed distribution of axon diameters in the sciatic nerve (Assaf et al., 2008). Slowing could be introduced by slowing the conduction velocity of every individual action potential by a fixed proportion of its delay.

Statistical analysis

All quantitative data are presented as box and whisker plots where the box represents the interquartile range, the bar represents the median, and the whiskers represent 1.5 times the interquartile range, or the range if this is less. Individual data points are superimposed onto boxplots. Statistical analysis was via the Kruskal-Wallis one-way ANOVA (KW test) followed by pairwise Mann-Whitney U-tests with correction for multiple comparisons via the false discovery method (Curran-Everett, 2000) with the maximum rate of false discovery set at 0.05. For analysis of the GRAB_ATP recordings in which the CO₂ and 50 mM KCl stimuli were applied to the same cell, these data were considered to be paired and comparisons of the amount of ATP released by each stimulus was, therefore, performed with the Wilcoxon Matched Pairs Signed Rank test. All pairwise tests were two-sided and all calculations performed with GraphPad PRISM.

Funding Statement

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Funding Information

This paper was supported by the following grant:

• Biotechnology and Biological Sciences Research Council BB/T00746X/1 to Jack Butler.

Data availability

All data is included in the source data files for each figure.