Moderate Changes in CO2 Modulate the Firing of Neurons in the VTA and Substantia Nigra

Summary The substantia nigra (SN) and ventral tegmental area (VTA) are vital for the control of movement, goal-directed behavior, and encoding reward. Here we show that the firing of specific neuronal subtypes in these nuclei can be modulated by physiological changes in the partial pressure of carbon dioxide (PCO2). The resting conductance of substantia nigra dopaminergic neurons in young animals (postnatal days 7–10) and GABAergic neurons in the VTA is modulated by changes in the level of CO2. We provide several lines of evidence that this CO2-sensitive conductance results from connexin 26 (Cx26) hemichannel expression. Since the levels of PCO2 in the blood will vary depending on physiological activity and pathology, this suggests that changes in PCO2 could potentially modulate motor activity, reward behavior, and wakefulness.


HIGHLIGHTS
The substantia nigra (SN) and VTA are key to movement and goaldirected behavior Activity in specific SN and VTA neurons is modulated by physiological changes in CO 2 The neuronal CO 2 sensitivity results from connexin 26 (Cx26) hemichannel expression Minute-to-minute changes in CO 2 could modify motor activity and reward behavior

INTRODUCTION
Carbon dioxide (CO 2 ) is a waste product of cellular metabolism with its concentration in blood a major regulator of breathing. In humans, PCO 2 in blood is normally $40 mm Hg but can be increased in conditions such as chronic obstructive pulmonary disease (COPD) and sleep apnea and can be decreased by hyperventilation and prolonged physical exertion. According to traditional consensus, CO 2 is detected via the consequent change in pH, and pH is a sufficient stimulus for all adaptive changes in breathing in response to hypercapnia (Loeschcke, 1982). pH-sensitive ion channels and receptors have been proposed to play a role in respiratory chemosensing in both the periphery (carotid body) and centrally in the medullary chemosensory areas such as the retrotrapezoid nucleus and the medullary raphe (Trapp et al., 2008;Kumar et al., 2015;Wang et al., 2013;Hosford et al., 2018). pH sensing via ventral medullary glial cells may also contribute to the CO 2 -dependent regulation of breathing (Gourine et al., 2010;Turovsky et al., 2016). However, there is considerable evidence that CO 2 can have additional independent effects from pH on central respiratory chemosensors (Eldridge et al., 1985;Shams, 1985). CO 2 directly binds to connexin 26 (Cx26) via a structural motif, which results in carbamylation of Lys125, thus increasing hemichannel opening probability (Huckstepp et al., 2010a;Meigh et al., 2013). The midpoint for the binding is $40 mm Hg, which, as indicated above, is the resting level in human blood, and thus small changes in PCO 2 will shift the open probability of Cx26 hemichannels. Pharmacological evidence suggests that Cx26 contributes to the CO 2 -dependent regulation of breathing (Gourine et al., 2005;Huckstepp et al., 2010b;Wenker et al., 2012), and this has recently gained support from genetic evidence that links binding of CO 2 to Cx26 to the adaptive change in breathing (van de Wiel et al., 2020).
Coupling between dopaminergic neurons (DNs) in the substantia nigra (SN) was first described by Grace and Bunney (1983) who showed that the injection of lucifer yellow dye into single cells could result in the filling of neighboring ''coupled'' cells, with the dye transferring through gap junctions. They confirmed this using electrophysiology. Vandecasteele (2005) validated that pairs of DNs in the SNpc are coupled by functional gap junctions and later went on to describe the connexin expression profile of SN DNs ( Vandecasteele et al., 2006). They reported that, in young rodents (postnatal day 7-10), these neurons express mRNA for Cx26 and Cx30, which are sensitive to CO 2 , but by P17-21 they only express mRNA for CO 2insensitive connexins (Vandecasteele et al., 2006). This observation led us to investigate whether the DNs in the SN of young rodents (P7-10) express CO 2 -sensitive hemichannels and thus have a CO 2 phenotype. We subsequently discovered an additional population of neurons, GABAergic, in the ventral tegmental area (VTA), which also appear to express Cx26 hemichannels and are sensitive to CO 2 . Unlike the SN DNs, these neurons appear to retain their sensitivity to CO 2 throughout development. Our findings reveal an unexpected role for CO 2 in regulating the activity of these key brain regions and demonstrate a mechanism by which autonomic state could alter complex movement-related and goal-directed behaviors. This would also be the first documentation of connexin 26 hemichannel expression in neurons.

RESULTS
To investigate whether dopaminergic neurons (DNs) in the SN from P7-10 mice are sensitive to levels of carbon dioxide (CO 2 ), as predicted from their connexin mRNA profile (Vandecasteele et al., 2006), we made whole-cell patch clamp recordings from DNs in acutely isolated slices. Putative DNs in the SN were identified by their electrophysiological profile. DNs were identified primarily by their position in the slice and characteristic currentvoltage relationship; most displayed a large sag in response to hyperpolarizing current steps (characteristic of Ih), rebound and tonic firing at rest, and a hyperpolarizing response to dopamine application (Grace and Onn, 1989;Neuhoff et al., 2002). A subset of recorded neurons were confirmed as dopaminergic when positive for the dopamine marker tyrosine hydroxylase using immunohistochemistry (Grace and Onn, 1989; Figure 1A). In order to test whether the DNs were sensitive to CO 2 , following their identification with standard step current injections and in some neurons also the injection of naturalistic current (to measure firing rates), the level of CO 2 (35 mm Hg, basal level) was increased to 55 mm Hg under isohydric conditions (compensatory changes in bicarbonate concentration to maintain constant extracellular pH during the CO 2 stimulus, see Methods). This increase in PCO 2 from 35 to 55 mm Hg (hypercapnia) produced a time-dependent reduction in the tonic firing rate and a reduction in the voltage change in response to hyperpolarizing current steps ( Figures 1B and 1C). Both of these effects are characteristic of an increase in resting conductance. At steady state, the response to the hyperpolarizing current steps had fallen to 70 G 9.6% of control (p = 0.0015, Figures 1B and 1C), the input resistance had fallen from 380 G 28.15 to 217 G 27.9 MU (p = 0.0027, n = 10), and the tonic firing was abolished. For a subset of recordings, we tested whether it was possible to get recovery when PCO 2 was returned from 55 to 35 mm Hg; this was not quantified, but an example showing partial recovery of firing rate and input resistance is illustrated in Figure 1D.
These observations were not an artifact of the dialysis of the cell following whole-cell breakthrough as the cells were first allowed time to equilibrate, then standard and naturalistic currents were injected to form IV curves and to measure firing rates. In a subset of neurons, pharmacological agents such as dopamine were applied to identify the cells ($30 min to apply and wash) prior to the alteration of CO 2 and similar effects of changing the CO 2 were observed. For initial controls, the experiment was first repeated without changing the PCO 2 (although the solutions were still exchanged to eliminate any artifacts due to the mechanical process of solution change), and under these conditions, the resting conductance and firing rate of the neurons did not significantly change over the time course of the experiment ( Figure S1). Second, the experiment was repeated with hippocampal CA1 pyramidal neurons and there was no significant change in the electrophysiological properties of these neurons with hypercapnia (voltage response was 100.3 G 1.64% of control, p = 0.68, Figure S2, n = 6).

Evidence That the Effects of PCO 2 on Cell Conductance Are due to Cx26 Hemichannel Expression
We then took a number of approaches to investigate whether the SN DN CO 2 sensitivity is the result of Cx26 hemichannel expression. First, as DNs in the SN of older mice (P17-21) do not express mRNA for CO 2 -sensitive connexins (Vandecasteele et al., 2006), they should therefore be insensitive to CO 2 if it is connexin hemichannel dependent. Whole-cell recordings from DN in SN from P17-21 mice showed the expected changes in electrophysiological properties that have been reported (Dufour et al., 2014) to occur during postnatal development ( Figure S3) but showed no significant response to increased PCO 2 (voltage response was 101 G 0.9% of control, p = 0.33, n = 4, Figures 1E and 1F). Second, the effects of increasing PCO 2 could be blocked by the hemichannel inhibitor carbenoxolone (Meigh et al., 2013) in P7-10 slices (100 mM Figures 1G-1I, n = 6). Carbenoxolone incubation did alter the electrophysiological properties of neurons (as previously reported in Tovar et al., 2009), but these changes would be expected to enhance the effects of hemi-channel opening rather than occlude them.
The midpoint for CO 2 -dependent opening of Cx26 hemichannels is around the basal level of PCO 2 used in these experiments (35-40 mm Hg, Huckstepp et al., 2010a). Thus, a reduction in PCO 2 should close Cx26 hemichannels leading to a decrease in resting conductance and a corresponding increase in firing rate. As predicted, in P7-10 SN DNs, decreasing PCO 2 from 35 to 20 mm Hg (hypocapnia) increased the voltage response to hyperpolarizing current steps (104 G 1.2% of control, p = 0.0078) and increased the firing rate (184 G 28.65% of control, p = 0.015, Figure 2) consistent with a decrease in conductance. These effects of reduced CO 2 were partially reversible (Figures 2A and 2B). Thus, small changes in CO 2, around normal resting levels ( We next used a different, non-electrophysiological approach to provide further evidence that P7-10 SN DNs express CO 2 -sensitive hemichannels. A characteristic of hemichannels is that, when they open, they allow entry of membrane-impermeant fluorescent dyes into cells. Once the hemichannels close, the dye becomes trapped inside the cells (this is termed dye loading) and can then be used as a marker for cells

OPEN ACCESS
iScience 23, 101343, July 24, 2020 3 iScience Article that express CO 2 -sensitive hemichannels (Huckstepp et al., 2010a;Meigh et al., 2013). Grace and Bunney (1983) had previously shown that lucifer yellow can demonstrate dye coupling (into neighboring neurons from intracellular injection into a single neuron). However, Vandecasteele et al. (2006) attempted to dye load SN DNs (extracellular bath application, as described above) with lucifer yellow by opening hemichannels with low levels of Ca 2+ ; this was unsuccessful. We decided to use the impermeant dye carboxyfluorescein (CBF) as we could be certain that it would pass through open Cx26 hemichannels as it has been shown in previous studies (Huckstepp et al., 2010a;Meigh et al., 2013) and confirmed that neurons in the SN of P7-10 mice ( Figure 3A) could be loaded with the dye following hypercapnia. No dye loading occurred if the PCO 2 was not increased. Dye loading did not occur in the SN of older mice (P17-21) or in CA1 hippocampal pyramidal cells ( Figure S2). To further confirm that early postnatal SN DNs express connexin 26, we used a highly specific monoclonal antibody to Cx26 (Sun et al., 2009, Huckstepp et al., 2010a, 2010b. In slices from P7-10 mice, Cx26 expression was present in tyrosine hydroxylase-positive (TH + ) neurons in the SN ( Figure 3B). However, in older mice (P17-21), Cx26 appeared not to be expressed in TH + neurons ( Figure 3C). Cx26 was still expressed in the leptomeninges of corresponding sections P17-21 providing a iScience Article positive control for the labeling protocol ( Figure 3D). At P17-21, Cx26 sensitivity appeared to shift from TH + cells to the neighboring glial cells (co-localizing with the glial marker GFAP, Figure 3E).

Changes in CO 2 Significantly Modifies the Excitability of Neurons in the VTA
During the CBF loading assay carried out in the P17-21 slices, although no dye-filled neurons were observed in the SN, unexpectedly a population of dye-filled neurons was observed in the neighboring VTA ( Figure 4A). This region is central to circuits controlling motivation, reward, and goal-directed behaviors (Morales and Margolis,

OPEN ACCESS
iScience 23, 101343, July 24, 2020 5 iScience Article 2017). Dye-loaded neurons in the VTA had a markedly different firing pattern and voltage response to current injection compared with SN DNs, were not hyperpolarized by dopamine, but were hyperpolarized by the opioid receptor agonist [Met5]Enkephalin, therefore, they could instead be GABAergic neurons (Johnson and North, 1992; Figure 4B). These VTA neurons showed electrophysiological changes similar to that observed for P7-10 SN DNs in response to changes in PCO 2 : increased PCO 2 (55 mm Hg) decreased input resistance (voltage response reduced to 71 G 13.2% of control p = 0.0055, 335 G 66.7 to 222 G 39.4 MU, p = 0.0446, before the sag n = 5) and firing rate ( Figures 4C-4E). Reducing PCO 2 to 20 mm Hg increased input resistance (voltage response increased to 118 G 6.1%, p = 0.0428) and firing rate (193 G 35%, p = 0.049, of control, Figures 4F-4H). To identify the phenotype of the CO 2 -sensitive neurons in the VTA, we carried out iScience Article immunohistochemistry. Cx26 was not expressed in TH + neurons in the VTA, so is not present in dopaminergic neurons ( Figure 3I). However, Cx26 immunoreactivity was present in GAD65/67 + neurons ( Figure 3J), a marker for GABAergic neurons in the VTA (Chieng et al., 2011), which fits with the electrophysiological properties of the CO 2 -sensitive neurons. Thus, the CO 2 -sensitive neurons in the VTA are GABAergic.

DISCUSSION
We have demonstrated an unexpected CO 2 -sensitive phenotype for neurons in the SN at P7-10 and in the VTA, with increases in CO 2 markedly increasing their resting conductance. This effect appears to occur in only specific subtypes of neuron, as for example, it was not observed in hippocampal pyramidal cells. It is well established that increases in PCO 2 can close gap junctions and that DN neurons in young animals are coupled (Connors et al., 1984;Bukauskas and Peracchia, 1997;Vandecasteele, 2005). However, this effect cannot account for the effects that we have observed. First, the increase in PCO 2 that is required to close gap junctions is large and is well above the range of PCO 2 changes we used to elicit effects on neuron electrophysiology. In addition, the closure of gap junctions would result in a decrease in whole-cell conductance and an increase in excitability, which is the opposite of what we observed in our study. Here we have provided several lines of evidence that suggest that our observations of an increase in CO 2 -sensitive conductance result from the opening of Cx26 hemichannels, whose open probability increases through the direct CO 2 -mediated carbamylation of lysine residues (Meigh et al., 2013). We have shown that the effects of CO 2 in SN DNs occurs over the same developmental period as they express Cx26 mRNA (measured in an independent study, Vandecasteele et al., 2006). The effects of increasing PCO 2 on resting conductance can be blocked by the hemichannel inhibitor carbenoxolone. Although carbenoxolone has neuronal and synaptic effects as well as blocking hemichannels, they would be expected to accentuate the observed increase in conductance rather than reducing it, therefore not obscuring the observations (Tovar et al., 2009). If the effects of PCO 2 on cell conductance are due to the opening of Cx26 hemichannels, it would be predicted that, since the midpoint of Cx26 hemichannel opening lies around the basal level of PCO 2 in our experiments (Meigh et al., 2013, 35 mM Hg), a decrease in CO 2 would close Cx26 hemichannels leading to a decrease in the resting conductance. Such a decrease in conductance could be observed for both SN DN and VTA GABAergic neurons when CO 2 was lowered.
SN dopaminergic neurons and VTA GABAergic neurons could be filled with a membrane-impermeant fluorescent dye (CBF) when PCO 2 was increased (dye loading). CBF will pass through Cx26 hemichannels when they are open and then become trapped inside cells when the hemichannels are subsequently closed (Huckstepp et al., 2010a;Meigh et al., 2013). Unfortunately, CBF cannot be fixed with paraformaldehyde, which prevents the dye-filled neurons from being subsequently labeled using immunohistochemistry. However, we can be confident that the dye-filled cells were either SN DNs or VTA GABAergic neurons, as patch clamp recording was carried out before the dye loading (to confirm the identity of the cells from their electrophysiological properties and pharmacology) and then the same cells were subsequently dye filled. We have also used immunohistochemistry to show that Cx26 protein is expressed in these neurons. The expression pattern of Cx26 across development in SN DNs matched that reported for Cx26 mRNA expression (Vandecasteele et al., 2006). This particular Cx26 antibody (13-8100) has been used extensively to study the role of Cx26 in breathing. There are many independent papers that demonstrate the specificity of this antibody in KO of Cx26. KO of Cx26 in the organ of Corti abolishes Cx26 immunoreactivity with this antibody (Sun et al., 2009), and our prior publications show correspondence for Cx26 immunostaining with a reporter driven from the endogenous Cx26 promoter (Huckstepp et al., 2010a(Huckstepp et al., , 2010b. To separate the effects of CO 2 from any effects of changing pH, we kept extracellular pH constant during our experiments by using isohydric solutions (an increase in PCO 2 under these conditions is termed isohydric hypercapnia). However, we did not measure intracellular pH and there will probably be transient changes in pH when the solutions are exchanged. It is well documented that intracellular pH will transiently acidify on application of the stimulus (raised PCO 2 ) and transiently alkalinize on its removal (Filosa et al., 2002;Putnam, 2001). A mild intracellular acidification would be expected to result in hemichannel closure and therefore a decrease in conductance. Therefore, we concluded that our observations were not the result of a change in intracellular pH. In addition, these transient changes in pH cannot explain the marked and sustained changes in conductance that only occur in these specific subtypes of neuron.
In this paper we have outlined the CO 2 sensitivity of specific neurons in the SN and VTA and provided a mechanism for this effect: Cx26 hemichannel expression. As far as we are aware this is the first documentation of neuronal expression of Cx26, which is usually found in glia ( iScience Article out no behavioral analysis of the effects of CO 2 sensitivity, it is interesting to speculate on its possible behavioral consequences. Since the mid-point of Cx26 opening lies around the resting level of CO 2 in humans (Huckstepp et al., 2010a(Huckstepp et al., , 2010bMeigh et al., 2013), small increases or decreases in CO 2 will modulate neuron excitability and thus could potentially modulate behavior. This CO 2 sensitivity switches from SN neurons to glia during early postnatal development but is retained in GABAergic neurons in the VTA. The switch from neuronal to glial expression could change the signaling direction from inhibitory to excitatory, as opening hemichannels in glia can allow the diffusion of molecules such as ATP, which could in turn excite SN DNs (through P2X or P2Y receptor activation). One speculated role for the CO 2 -mediated reduction in excitability of SN DNs in early postnatal life is, since the nest is likely to be hypercapnic, inhibition of movement may promote suckling behavior. The maintenance of CO 2 sensitivity in the VTA postnatally is particularly interesting given its role in reward, addiction, motivation (Volkow and Morales, 2015), and sleep-wake behaviors (Eban-Rothschild et al., 2016). Activation of GABAergic neurons in the VTA induces sleep, and their inhibition increases wakefulness (Yu et al., 2018). There are several contributing mechanisms to hypercapnic arousal. The orexinergic neurons of the lateral hypothalamus, known to promote wakefulness, can be activated by hypercapnia, although this is through a pH-dependent transduction mechanism (Williams et al., 2007). The histaminergic neurons of the tuberomamillary nucleus (TMN), which also promote wakefulness, are activated by CO 2 (Johnson et al., 2005;Anaclet et al., 2009). Neurons of the dorsal raphe, not involved in the control of breathing are pH/CO 2 sensitive and contribute to hypercapnic arousal (Smith et al., 2018). Furthermore, the parabrachial nucleus integrates chemosensory inputs during hypercapnia from the medullary nuclei such as the retrotrapezoid nucleus and the raphe magnus, which contain pH-sensitive neurons to mediate arousal (Kaur et al., 2017). However, even after silencing these key relay neurons, hypercapnia still results in arousal albeit at a longer latency showing that other parallel pathways are involved (Kaur et al., 2017). Given that inhibition of the VTA GABAergic neurons have been demonstrated to cause wakefulness (Yu et al., 2018) we hypothesize that inhibition of these neurons by modestly raised CO 2 could potentially contribute an additional parallel pathway of hypercapnic arousal.

Limitations of the Study
This report outlines the novel observation of CO 2 sensitivity in a specific subset of neurons with several lines of evidence that it results from Cx26 hemichannel expression. There is no data on the physiological significance of this CO 2 sensitivity, in particular regarding movement and reward behavior. This will be examined in future studies. There is no quantification of the expression pattern of Cx26 protein in either TH + or GAD + neurons. In the future, tools like fluorescence in situ hybridization (FISH) could be used to produce more accurate measurements of expression.   Input resistance measurements 'both before the sag' (A) and at 'steady state' (B) decreased during development in line with published studies (4). We saw no difference in the stability of recordings and there was no change to resting membrane potential (C).  Figure S4: Example raw data traces for the full timeframe of raised CO2 exposure, related to figure 1. A representative example from a P7-10 substantia nigra dopaminergic neuron. Membrane potential traces in response to -50 pA current steps in 35 mmHg CO2 (black), then switched over to 55 mmHg CO2 (red).

A B C
Each plot represents a timepoint from the graph in Figure 1B

A D C B
(Molecular Devices, USA) and digitised at 20 KHz. Data acquisition and analysis were performed using pClamp 10 (Molecular Devices). Recordings from neurons that had a resting membrane potential of between -55 and -75 mV at whole-cell breakthrough were accepted for analysis. The bridge balance was monitored throughout the experiments and any recordings where it changed by more than 20 % were discarded.

Stimulation Protocols Standard IV protocol
A standard current-voltage relationship was constructed by injecting step currents (3 s duration, every 5 s) starting at -200 pA and then incrementing by either 50 or 100 pA until a regular firing pattern was induced. A plot of step current against voltage response around the resting potential was used to measure the input resistance (gradient of the fitted line).

Naturalistic current injection
The naturalistic current was generated using the summed numerical output of two Ornstein-Uhlenbeck processes (Uhlenbeck & Ornstein, 1930) with time constants fast = 3 ms and slow = 10 ms. This naturalistic current waveform (as in Badel et al, 2008), which mimics the stochastic actions of AMPA and GABA-receptor channel activation, was injected into cells (40 s duration) and the resulting voltage recorded (as a fluctuating noisy trace). This voltage trace was then used to evaluate the frequency of action potential firing.

Current Injection to assess conductance changes
A hyperpolarising step of 50 pA (100 ms) was injected at a frequency of 1 Hz. This allowed the time course of changes in input resistance/conductance to be assessed. For analysis, averages were constructed for 10-minute periods in 35 mm Hg and 55 mmHg CO2 (when the effects of CO2 had reached steady state).
For each recording, once whole cell breakthrough had occurred cells were allowed to equilibrate for a few minutes. Following this a standard IV curve was constructed and naturalistic current traces were injected to enable the measurement of firing rate. After these measurements had been recorded (5-10 mins post-whole cell breakthrough), the hyperpolarising step current was initiated to look for voltage changes in response to altered levels of carbon dioxide (from 35 mm Hg to 55 mm Hg, isohydric).

Immunohistochemistry
Mice (P7-10 and P17-20) were cardiac perfused with 4% PFA and then post-fixed overnight at 4ºC. The tissue was washed with PBS and then sliced coronally (350 m). The slices were left to recover for 1 hour and then were blocked for an hour (1% BSA, 0.4% Triton 100X in PBS, 400 l per slice) then washed 3 times for 5 minutes with PBS. The primary antibodies against tyrosine hydroxylase, (1:1000, Sheep), GFAP (1:1000, Chicken) or GAD65/67 (1:1000, Rabbit) and the primary antibody against connexin 26 (1:200, Mouse) were added to the slices (400 l per slice) for an hour at room temperature and then kept at 4-8ºC overnight. Slices were washed 5 times for 5 minutes with PBS and the corresponding secondary antibody (anti-sheep 488, 1:500, anti-mouse 594, anti-sheep 594, antichicken 488 or anti-rabbit 488, 1:500, 400 l per slice) added for 4 hours at room temperature. The slices were then washed 5 times for 5 minutes with PBS, and then mounted on glass slides with Vectashield (Vector laboratories, Peterborough UK). All imaging was carried with confocal microscopy (Leica 710 and Zen Black for image acquisition and processing). Controls were carried out without the primary antibodies and showed no fluorescence.

Dye loading
The dye loading method is based on that described in Huckstepp et al (2010). Briefly, a slice was transferred to the recording chamber, submerged and perfused (2-3 ml/min -1 ) with control aCSF (35 mmHg CO2) at 30 °C. Slices were visualized using IR-DIC optics with an Olympus BX151W microscope (Scientifica, Bedford UK) and a CCD camera (Hitachi). To confirm the correct location for imaging, whole cell patch clamp recordings were used to identity DNs and GABAergic neurons in the SN and VTA ( Fig. 1 and 4). Slices were then allowed to equilibrate for 20 minutes. The control aCSF was then exchanged for 55 mmHg CO2 aCSF (hypercapnic) containing 5(6)-carboxy-fluorescein (CBF, 100 M) for 20 mins to allow the CO2 sensitive-hemichannels to open. The solution was then exchanged for 35 mmHg CO2 aCSF containing CBF (100 M) for 5 minutes to allow the hemichannels to close. Finally, the slice was washed with 35 mmHg CO2 aCSF for 3 hours to reduce the background staining before imaging. Images were taken using the CCD camera (Hitachi) with 488 nm fluorescence (CoolLED). As CBF rapidly bleaches, images were quickly acquired from regions of interest. CBF cannot be fixed using PFA (as it lacks the required groups for cross-linking).

Statistics
Data is represented as mean and standard error of the mean with individual experiments represented by single data points. Appropriate statistical tests were chosen based on sample size, whether there were repeated measures and whether the populations were paired or unpaired (Wilcoxon rank sum/ paired t-tests and Mann Whitney tests respectively). For tests of more than two variables, Kruskal-Wallis ANOVAs were run with Dunn's post hoc multiple comparisons. All tests were run to find significance at the level p < 0.05 and were performed on raw (non-normalised) data, available in supplementary data (figures 5-8).