Introduction

Sensory information originates from the stimulation of specific receptors in different parts of the body. Beyond the five classic senses (touch, taste, smell, vision, and hearing), it has long been recognized that other sensory information—such as proprioception and visceral sensation—is fundamental for normal bodily function. Located at the interface between the spinal cord’s central canal (CC) and the spinal cord parenchyma, cerebrospinal fluid-contacting neurons (CSFcNs) are part of a sensory system that provides information about the internal environment, specifically the composition of cerebrospinal fluid (CSF), and are thus considered part of the interoceptive system^1,2.

Sensory cells transduce sensory stimuli into electrical activity and typically have distinctive morphological specializations. CSFcNs are no exception, with their somas located in the subependymal layer from which a short, thick dendrite arises, ending in a bulbous structure located within the CC^3,4 known as the “apical process” (ApPr). The axon of CSFcNs, on the other hand, projects rostrally trough the ipsilateral ventral spinal cord, and contacts other CSFcNs and neurons of spinal central pattern generators, including motorneurons and premotor excitatory neurons^5–7. By providing sensory information about CSF composition and the mechanical forces acting near the CC, CSFcNs participate in the control of posture and locomotion, both in lower vertebrates^1,5,8–10 and in rodents^6,7.

CSFcNs are established chemoreceptors. Huang et al. were the first to report that CSFcNs in the mouse spinal cord: i. express the PKD2L1 (or TRPP2) channel, a cation permeable channel of the TRP (Transient Receptor Potential) family that has a high Ca⁺⁺ permeability¹¹; ii. respond to acid with an increase in firing rate¹². Subsequent studies examined the pH sensitivity of CSFcNs in the rat³, lamprey^10,13 and mouse spinal cord^14,15. These collective findings suggested that the response of CSFcNs to acidification depends on the activation of an acid sensing ion channel (ASIC), rapidly desensitizing inward current, and the response to alkalinization depends on the activation of PKD2L1 channels.

Despite previous work, the precise mechanisms and subcellular localization of the channels responsible for the pH response in CSFcNs remain unknown. This gap is partly due to two main factors. First, earlier investigations relied on bath application or pressure ejection of solutions with varying pH, which lack the spatial and temporal precision required to resolve localized pH sensitivity in specific cellular compartments. Second, there are technical challenges associated with recording from a small neuronal compartment using electrophysiological approaches. Here, we overcome these technical barriers with a combination of whole-cell and outside-out recordings directly from the ApPr, along with laser photolysis of protons and immunohistochemistry. We reveal that PKD2L1 generates not only phasic but also a novel tonic current that critically shapes the membrane potential of CSFcNs. These currents exhibit high pH sensitivity near the physiological range. Importantly, we demonstrate that functional PKD2L1 channels are exclusively localized to the ApPr, whose high input resistance and pH sensitivity make it a highly specialized sensory hub for the integration of chemical signals.

Results

PKD2L1-mediated spontaneous activity in CSFcNs from Gata3^eGFP mice

In this work we used the Gata3^eGFP16 transgenic mice, where CSFcNs express eGFP under a GATA3 regulatory element. GATA3 is a transcription factor expressed in the spinal cord both by CSFcNs¹⁷ and V2b interneurons¹⁶. Figure 1A shows sagital (Aa) and transverse (Ab) slices of Gata3^eGFP animals where the typical morphology and location (around the CC) of CSFcNs can be readily appreciated. This, together with the fact that V2b interneurons are located well away from the CC, allows unambiguous identification of CSFcNs. As expected, GFP+ in the Gata3^eGFP mice express the PKD2L1 channel showing they are indeedCSFcNs (Figure 1Ac and d). We first characterised the basal electrophysiological activity of CSFcNs from Gata3^eGFP mice in voltage-clamp, as these neurons in the Gata3^eGFP transgenic have not yet been characterized. CSFcNs were recorded with a KGluconate-based intracellular solution (IS) at near physiological temperature (34 ± 1 °C) with the whole-cell configuration of the patch-clamp technique from either the soma or the ApPr; neurons were clamped at -60 mV. In these conditions, spontaneous single-channel events were recorded, confirming previous results (in mice^14,15, lamprey¹⁰ and zebrafish¹⁸). Figure 1Ba shows a 1 sec period of such a recording. The dotted horizontal green lines indicate 3 levels of channel activity (c = closed; o1 = 1 channel open; o2 = 2 channels open). Figure 1Bb shows the corresponding histogram, where the 3 peaks that define the 3 levels of channel activity are shown with arrowheads. From the analysis, the probability of each state (pc, po1 and po2) was calculated. In this example, the single channel current recorded at -60 mV holding potential was -17.5 pA, and po1 was 0.12. From the spontaneous activity of different CSFcNs recorded either from the ApPr or from the soma, we measured an average single channel current amplitude of -15.9 ± 1.8 pA (Figure 1Bc, n = 38, with minimal and maximal values of -12.1 and -19.3, respectively) and a highly variable average probability for 1 channel being open, po1, of 0.08 ± 0.08 at -60 mV (Figure 1Bd, n = 38, with minimal and maximal values of 0.0065 and 0.29, respectively). No difference was found for these two values between ApPr and somatic recordings (p = 0.06 for the single channel current and p = 0.53 for po1), and the results were pooled together (25 recordings from the ApPr and 13 from the soma). Figure 1Ca shows the single-channel activity recorded at different holding potentials from the same neuron shown in B. From 16 cells recorded at different holding potentials, we calculated a single channel conductance of 222 ± 7.8 pS (Figure 1Cb, similar to what has been described before in CSFcNs from different species^13,14 and for the PKD2L1 channel expressed in heterologous expression systems), with an extrapolated reversal potential very close to the expected value of 0 mV (-2.2 mV). These experiments indicate that the PKD2L1-dependent single channel activity recorded from adult CSFcNs from the Gata3^eGFP mice is very similar to previous recordings in other transgenic mice models^14,15.

Fig 1.

PKD2L1 channels mediate phasic and tonic currents

Dibucaine hydrochloride was described as a blocker of PKD2L1 channels in an expression system¹⁹. We decided to test whether dibucaine could also be used as a blocker of these channels in CSFcNs. As shown in Figure 2Aa, pressure-application of 100-200 µM dibucaine during 30 seconds strongly reduced the frequency of single channel openings, from 177 ± 163 to 5 ± 8 Hz (control vs dibucaine, respectively, Figure 2Ab, n = 12, p = 0.0005).

Apart from blocking PKD2L1 phasic activity, dibucaine had another remarkable effect. Pressure application of the drug decreased the holding current, as shown in the representative example in Figure 2Ba and b, where the holding current (HC) went from -23 to -10 pA. On average, dibucaine decreased the HC from -18.3 ± 9.5 to -10.5 ± 6.6 pA (Figure 2Ca, n = 13, p = 0.0002), a 44 ± 12 % reduction. Pressure application of the extracellular solution without dibucaine had no effect on either the single-channel current activity or the HC (data not shown). As expected from the effect on the HC, when the CSFcN was recorded in current-clamp, pressure application of dibucaine induced a marked hyperpolarization of the resting membrane potential (RMP, see Figure 2Bc), ffrom -66.8 ± 11.0 to -89.0 ± 14.0 mV (Fig 2Cb, n = 11, p = 0.001). Because dibucaine is a local anaesthetic it may act on other targets, like voltage-dependent sodium and calcium channels²⁰. To rule-out non-specific effects of dibucaine, we performed the following experiments. First, we repeated the dibucaine application in the presence of antagonists for voltage and ligand-gated channels that may be open near the -60 mV holding potential values (TTX 0.4 µM, TEA 1 mM, 4AP 1 mM, TTAP2 20 µM, NBQX 10 µM and gabazine 10 µM to target, respectively, voltage gated Na⁺, K⁺ and T-type Ca⁺⁺ channels, and the AMPA and GABA_A ionotropic receptors). Under these conditions, a similar decrease in the HC was observed, from -18.1 ± 8.1 pA in control conditions to -10.8 ± 5.4 pA in the presence of the drug (Figure 2Cc, n = 11, p = 0.001), representing a 41 ± 12 % reduction. To note, there was no difference between the control HC recorded in the two conditions (without or with blockers, p = 0.9, Wilcoxon–Mann–Whitney test). Second, to rule out the possible contribution of conductances sensitive to the calcium flowing into the cell through PKD2L1 channels, we patched the neurons with the same intracellular solution + 10 mM BAPTA. Under these conditions, a decrease in the HC was still observed when pressure applying dibucaine, from -13.0 ± 5.7 pA in control conditions to -7.8 ± 2.5 pA (Figure 2Cd, n = 11, p = 0.001), a 35 ± 15 % reduction. Again, there was no difference between the control HC measured in the 3 different conditions (without blockers vs BAPTA, p = 0.17; with blockers vs BAPTA, p = 0.19, Wilcoxon–Mann–Whitney test). Such a marked effect of dibucaine on the HC and the RMP can be explained by the fact that CSFcNs are electrotonic compact neurons with a high input resistance (IR). Indeed, the IR measured from 14 somatic recordings was 1.8 ± 0.46 GΩ (Figure 2Ce; consistent with previous published values in the turtle²¹, rat³ and mice^14,15). As we will show later in this work, PKD2L1 channels are concentrated in the ApPr. Therefore, a better estimation of the impact of a PKD2L1-mediated current on the neuron’s membrane potential should be obtained by measuring the IR directly from the ApPr. To do that we made whole-ApPr recordings, where we measured an average IR of 2.3 ± 0.5 GΩ (n = 29), statistically higher than the somatic one (p = 0.002; Figure 2Ce). Finally, we took advantage of the fact that during the slicing procedure some ApPrs are separated from the rest of the cell; these will be referred to as “isolated” ApPr (iAPr). We succeeded to record from those iApPr, where we measured an IR that was even higher than the previous values (4.4 ± 1.2 GΩ, n = 8; Figure 2Ce; p = 6 × 10^-6, iApPr vs soma, and p = 4 × 10^-7, iApPr vs whole-ApPr). Altogether, these results indicate that the spontaneous activity induced by PKD2L1 channels gives rise to both a phasic and a tonic current. Both phasic and tonic components of the PKD2L1-associated currents have marked effects on membrane potential as CSFcNs and, particularly, their ApPr, display very high IR values.

Fig 2.

It has been reported that calmodulin modulates PKD2L1 channels through a direct interaction and that blocking calmodulin increases the activity of the channel^11,19,22. In order to test whether this also applies for PKD2L1 channels in CSFcNs, we recorded channel activity under control conditions and in the presence of the calmodulin inhibitor calmidazolium. As expected, pressure-application of calmidazolium (10 or 20 µM) at the ApPr increased the channel activity, as shown in the representative recording in Supp. Figure 1Aa (top panel). The increase in channel activity is shown in the middle panel, where the slope of the plot of membrane charge as a function of time, which is dependent on the spontaneous openings of the channels, suddenly rises when calmidazolium is applied. The lower panel shows the normalized membrane charge (average ± SD) of 10 different CSFcNs before, during and after calmidazolium application. In all neurons tested (Supp. Figure 1Ab for an example), calmidazolium application also increased the HC (from -13.3 ± 5.3 to -16.0 ± 4.7 pA, n = 10, p = 0.008, Supp. Figure 1Ac), a 24 ± 20 % increase. This effect of calmidazolium on the HC had a dramatic effect on the RMP of the recorded cells. In current-clamp, calmidazolium induced a depolarization (Supp. Figure 1B), from -62.3 ± 3.3 in control conditions to -50.0 ± 5.2 mV in the presence of calmidazolium, n = 10, p = 0.002, as well as an increase in the spontaneous EPSP frequency (lower part of Supp. Figure 1Ba). This experiment shows that the increase in the activity of PKD2L1 channels impacts both phasic and tonic currents. This is consistent with our previous conclusion that blocking PKD2L1 channels likewise reduces both phasic and tonic component of PKD2L1-associated currents. Also, it highlights the importance of the intrinsic membrane properties, in this case the input resistance, in setting CSFcN excitability.

Effect of extracellular pH on PKD2L1-mediated phasic and tonic currents

We next asked whether physiological stimuli can also modulate PKD2L1 activity in a similar fashion as dibucaine and calmidazolium. To address this issue we performed changes in extracellular pH, as CSFcNs have been shown to be sensitive to changes in extracellular pH (see Introduction and further below). When a pH 6.5, 10 mM HEPES-buffered extracellular solution was pressure-applied to the ApPr, a decrease in PKD2L1 activity was observed, as shown in Figure 3Aa, top trace, and in the corresponding inset (traces “7.4” and “6.5”). The decrease in channel activity can be appreciated in a plot of the current charge as a function of time (Figure 3Aa, bottom trace). Figure 3Ab shows the normalized current integral calculated from 10 different cells before, during and after the pressure application of the acidic solution. The dotted line represents a linear fit to the first 3 seconds of the recording, which corresponds to the control period. The decrease in channel phasic activity was manifested as an increase in the closed probability of the channels (pc, Figure 2Ca, from 0.89 ± 0.09 to 0.99 ± 0.01, n = 10, p = 0.002) and a parallel decrease in the open probability of the channels (po1, Figure 2Cb, from 0.1 ± 0.08 to 0.01 ± 0.01, n = 10, p = 0.02). Meanwhile, the single channel current amplitude was not affected (-14.7 ± 1.34 vs -14.0 ± 1.7, ctrl vs pH 6.5, n = 10, p = 0.15). Apart from the effects on the phasic currents, application of the acidic solution was also accompanied by a parallel decrease in the recorded HC, from -16.8 ± 6.7 to -12.0 ± 5.2 pA (Figure 3Cc, n = 10, p = 0.002), a 28.6 ± 12.7% reduction. The effect on the holding current can be seen in the VC recordings shown in Figure 3Aa, top, but is more clearly seen when the recorded current is averaged over 100 ms time periods (Figure 3Aa, middle graph). When the cell was held in CC, pressure application of the pH 6.5 solution produced a marked hyperpolarization of the RMP (Figure 3B for an example), from -64.5 ± 5.2 to -76.0 ± 3.4 mV (n = 8, p = 0.008; Figure 3Bd). In 3 out the 8 experiments, a quickly desensitizing inward current was observed in VC at the onset of the application of the acidic solution (Supplementary Figure 2Aa), which produced a short-lasting depolarization in CC that was followed by the persistent hyperpolarization described above (Supplementary Figure 2Ab). As previously described¹⁴, this current probably reflects the opening of ASIC channels. This current was not further characterized.

Fig 3.

On the other hand, when a pH 8.4 HEPES-buffered extracellular solution was pressure-applied, a clear increase in PKD2L1 activity was recorded (Supplementary Figure 3A and B). This was manifested as an increase in the current integral (Supplementary Figure 3), a decrease in the closed probability of the channels (from 0.96 ± 0.036 to 0.76 ± 0.13, n = 8, p = 0.008, Supplementary Figure 3Da) and an increase in the open probability of the channels (from 0.042 ± 0.034 to 0.2 ± 0.01, n = 8, p = 0.008, Supplementary Figure 3Db). As expected, there was also a slight increase in the HC (from -11.7 ± 7.3 to -15.2 ± 7.9 pA, n = 8, p = 0.008, Supplementary Figure 3Dc) and a depolarization of the RMP (from -75.2 ± 7.0 to -66.2 ± 10 mV, n = 6, p = 0.03, Supplementary Figure 3Dd). The single channel current remained unchanged (-16.85 ± 0.95 vs -17.3 ± 1.0 pA, ctrl vs pH 8.4; respectively, n = 8, p = 0.45).

The experiments presented so far indicate that the basal activity of PKD2L1 channels plays a central role in regulating CSFcNs excitability. This regulation takes place at two levels: by a modulation of the phasic activity of the channels, as has already been described^14,15, and also by a modulation of a novel PKD2L1-dependent tonic current that has a dramatic effect in setting the RMP in CSFcN. In this sense, PKD2L1 basal activity determines an excitability set-point that can be up and down-regulated by small pH changes. This is clearly seen when plotting po as a function of pH (Figure 3Ce). The experimental data has been fitted by a Hill equation that shows a pH half-value of 8.1 ± 0.04, very close to physiological pH and to previously estimated values in expression systems²³ and computational model²⁴. A similar result is obtained when plotting the holding current change (in relation to the control value at pH 7.4) as a function of pH. Fitting the data with a Hill equation gives a pH half-value of 7.5 ± 0.02 (Figure 2Cf). In order to better appreciate the effect of the HC on the Vm, we calculated the Vm and the HC differences produced (in individual neurons) by the different experimental challenges tested here, and then plotted the Vm vs the HC difference. The plot, shown in Figure 3D, indicates a strong correlation between the 2 values. The fit of the data with a linear function (dotted gray line) indicates a slope value of 2.3 GΩ, very close to the IR values calculated for the ApPr (Figure 1Ee). Taken together, these data indicate that both the PKD2L1 channel po and the holding current are very sensitive to extracellular pH changes, that the maximal pH sensitivity occurs within the physiological range and that small changes in the tonic current can have a substantial effect on CSFcNs RMP (≈ 2.3 mV/pA).

The use of laser photolysis to assess the pH sensitivity of CSFcNs

It has been reported that the sensitivity of CSFcNs to pH changes depends on the activity of two different channels, PKD2L1 and ASICs. Those studies used a combination of electrophysiological methods with bath or pressure applications of drugs, but these strategies suffer from limited spatial resolution, preventing precise localization of channel activity within the CSFcN membrane. To overcome this limitation and to precisely map the sub-cellular sensitivity to pH of the different CSFcN compartments, we employed laser photolysis, as this approach provides high spatial resolution (the laser can be focused to a small spot, with x-y dimensions of 1.5 µm^25,26) as well as a high temporal resolution (the laser can be switched on and off very rapidly, thus producing almost instantaneous agonist changes). This is shown schematically in Figure 4A-B. Briefly, after the different CSFcN compartments (soma, dendrite and ApPr) were identified by fluorescence of Alexa 594, the focus of the objective was positioned in the targeted compartment (ApPr in the example in Figure 4A), where laser photolysis was evoked.

Fig 4.

To induce pH changes we used the glutamate cage MNI-glutamate, as its photolysis generates the stoichiometric release of one glutamate molecule together with one proton²⁷ (Figure 4B, top reaction). MNI-glutamate uncaging in the ApPr of CSFcNs elicited two currents with different characteristics. A first component had very quick rise and decay times (values of the example shown in Figure 4C: amplitude 38 pA, 10-90 % risetime 0.86 ms and decay time constant 2.6 ms; see inset for details). It appeared immediately after the laser pulse (magenta arrowhead in Figure 4C) and fluctuated little among repetitions. A second component was indistinguishable from the PKD2L1-dependent spontaneous activity recorded above. It followed the laser pulse with variable latencies (although always longer than the latencies of the first component) and fluctuated widely among repetitions. The first current was blocked by AMPA_R antagonists (NBQX or CNQX) and was therefore attributed to AMPA receptor activation. The second component of the response was spared by both AMPA or NMDA_R blockers (the experiments shown in Figures 4D and G were done in the presence of NBQX and APV) and was presumed to reflect the pH change. To confirm that this current resulted from the release of protons (and not glutamate), we used the caged compound MNI-γLGG which has the same photochemistry as MNI-glutamate except that it releases the inactive enantiomer of the AMPA_R antagonist γDGG²⁸ together with a proton (Figure 4B, bottom reaction). This failed to produce the short latency current, but produced the same long latency current as above, indicating that the late current is related to proton release rather than glutamate (Figure 4B for an example). Also, no current could be observed when the ApPr was illuminated in the absence of a caged compound (data not shown), indicating that the current is not due to a light artifact or to photodamage.

The proton-induced current is an off-current

Inspection of the experimental traces presented in Figure 4C suggests that the photolysis of an MNI-cage in the ApPr of CSFcNs increases the probability of channel opening with a characteristic long latency. The increase in channel activity was analysed in two ways. First, the membrane charge was calculated from spontaneous recordings (gray traces in Figure 4D) and from recordings where MNI-glutamate (or MNI-γLGG) was photolysed (black traces in Figure 4D). As already mentioned (Figure 3), the membrane charge calculated from spontaneous sweeps shows a linear increase as a function of time, whereas that calculated from laser-evoked sweeps shows a sudden slope increase induced by the laser pulse, that recovers later. Figure 4E shows the average normalized membrane charge (black trace) calculated from 14 different neurons where photolysis was evoked on the ApPr. The data was fitted with the sum of a linear and exponential functions. The time constant of the exponential function, τ, is equal to 258 ms. The magenta-shaded area in the graph (in Figure 4E) spans 500 ms (≈ 2 τ_s), representing the estimated lifetime of the photolysis-induced increase in channel activity. Second, we measured channel probabilities from spontaneous recordings and compared them to those during the magenta-shaded time-window, which corresponds to the photolysis effect (see above). This analysis indicated that the opening probability increased (Figure 4Fa) and the closing probability decreased (Figure 4Fb) as a result of the photolysis.

As previously mentioned, the increase in channel activity does not happen during the laser pulse but a few milliseconds afterwards. In this sense, the latency difference between the AMPA-dependent and proton dependent components is striking. This difference can be clearly appreciated in the inset in Figure 4C, where the AMPA-dependent component coincides with the laser pulse (magenta trace) whereas the proton-dependent component appears more than 30 ms later. The PKD2L1 current latency in sweep #4, which is the shortest among the 5 repetitions, is 11 ms. To characterize the latency distribution of the PKD2L1-mediated single currents relative to laser pulse timing, we measured the current onset times from idealized traces during 2-second recording periods (see Figure 4Ga for an example; experiments done in continuous presence of NBQX or with MNI-γLGG). To minimize contamination from spontaneous channel openings, we only considered double events (simultaneous opening of two channels) as the spontaneous occurrence of such events is very low. From the onset times of these double events, we constructed the latency distribution histogram shown in Figure 4Gb. The open probability gradually rises after the laser pulse, with a time constant of 30 ± 12 ms, peaking at 108 ms after the laser pulse. This analysis, together with our calculation that photolysis induces a transient pH drop of around 4 units (from 7.4 to ≈ 3.3) that returns to baseline in microseconds (see materials section), indicates that the photolysis-evoked PKD2L1 single-channel currents represent, in fact, an off-response. Indeed, it has been shown that PKD2L1 channels, which are activated by alkali and inhibited by acid, are also activated by acid removal^24,29,30, giving rise to the so-called off-response.

The off-current is mediated by the activation of PKD2L1 channels

The experiments presented in the previous section suggest that the off-current is mediated by PKD2L1 channels. To confirm this possibility, we performed the following experiments/analysis. Firstly, channel activity was blocked when the photolysis is performed in the continuous presence of dibucaine. This can be appreciated in the representative trace shown in Supplementary Figure 4A: in the presence of dibucaine (gray trace), both the spontaneous currents and the laser-evoked currents were blocked. The bottom panel shows the time-dependent current integrals calculated from the upper traces. Supplementary Figure 4B shows the average normalized membrane charge recorded under two conditions, control (n = 14 neurons, same graph as in Figure 4E) and dibucaine (n = 6 neurons). Some of the experiments in the presence of dibucaine were done in the absence of AMPA_R blockers. In these cases, a clear fast, inward, AMPA_R-mediated current can be appreciated which is not followed by any single channel event (Supplementary Figure 4D), indicating that the lack of response was not due to an unresponsive or damaged ApPr.

Secondly, the single channel current evoked by the laser pulse (calculated during the 2 τ_s time window after the laser pulse; see above) was indistinguishable from that obtained during the spontaneous current activity (Supplementary Figure 4C, -16.7 ± 2.0 to 17.0 ± 2.0 pA, n = 21, p = 0.54).

ASICs have been shown to be present in CSFcNs^13–15,31. To rule out any contribution of ASIC to the photolysis-induced response, we performed the experiment in the continuous presence of the ASIC blockers PsTx (100 nM) and ApeTx2 (100 nM). In these conditions, neither the single channel events nor the charge increase evoked by photolysis were affected (Supplementary Figure 4E), confirming that the increase in channel activity related to the release of protons by photolysis on the ApPr is dependent on the activation of PKD2L1 channels.

The PKD2L1-mediated current is generated in the ApPr

The next series of experiments was designed to assess the spatial sensitivity of CSFcNs to pH changes. To this aim, we photolysed MNI-compounds in different CSFcNs compartments. Figure 5A shows the result of an experiment where the photolysis of MNI-γLGG was performed either in the ApPr, the soma or 5 µm away from the ApPr (Figure 5Aa). As expected, photolysis in the ApPr induced the typical off-current, characterized by the increase in the unitary activity of PKD2L1 channels (Figure 5Ab, black traces) as well as in membrane charge; on the contrary, photolysis in the soma (Figure 5Ab, pink traces) or in the dendrite (data not shown) did not produce a sizable increase in the unitary currents. Figure 5Ac summarizes the photolysis-induced changes on the membrane charge. Photolysis on the ApPr elicited a pronounced increase in membrane charge, whereas photolysis in the soma produced no detectable change. Notably, the response to photolysis in the soma was indistinguishable from the photolysis 5 µm away from the ApPr (magenta vs green traces, respectively), which is more evident in the bottom panel showing the same data set in an expanded amplitude scale.

Fig 5.

To further assess the sensitivity of ApPrs to pH changes, we exploited the fact that during the slicing procedure some ApPrs are separated from the rest of the cell. We term these cell fragments iApPrs. The iApPr were filled with Alexa 594 in order to confirm that they were indeed detached from the rest of the neuron (Figure 5Ba). Interestingly, iApPrs show single-channel currents with the same single-channel conductance and voltage-dependence as non-isolated ApPrs, as can be appreciated in the representative example shown in Figure 5Bb. Furthermore, iApPrs respond to proton photolysis similarly to non-isolated ApPrs (Figure 5C) and both the spontaneous single-channel activity as well as the photolysis-induced charge increase are blocked by dibucaine (Figure 5Cb-c).

To confirm the presence of PKD2L1 channels in ApPrs, we performed outside-out recordings from patches of membrane taken from the ApPr (Figure 5D). We first recorded the ApPr in the whole-ApPr configuration and then slowly took out the pipette until no unclamped action current could be seen (as iApPrs do not show voltage-gated sodium currents; Figure 5Da-b). PKD2L1 channel openings were observed in the outside-out configuration (Figure 5Db), albeit with a lower opening frequency than in the whole-cell configuration (Figure 5Da). Such single-channel activity was observed in 6/10 outside-out recordings.

Finally, we performed immunohistochemical experiments to characterise the location of the PKD2L1 protein in Gata3^eGFP. As seen in Figure 5E, the PKD2L1-related fluorescence intensity (shown in magenta in Figure 5Ea) is >6-fold higher in the ApPr than in the soma (Figure 5Eb) or dendrite (not shown). The mean fluorescence in the ApPr was 84 ± 49 AU and dropped to 12.5 ± 8 AU in the soma. Altogether, our experiments indicate that the PKD2L1 channels are specifically enriched in the ApPr. Although we cannot rule out the presence of PKD2L1 channels in other compartments, our electrophysiological and immunohistochemical experiments suggest that functional PKD2L1 channels are topologically segregated to the ApPr.

Discussion

The pH sensitivity of neurons depends on both their anatomical features and the functional properties of their ion channels. As all neurons exhibit some degree of pH sensitivity³², investigating it needs the use of targeted techniques that can accurately assess both the anatomical and physiological properties of the channels involved in the response to pH. Our data, derived from the integration of whole-ApPr and outside-out recordings, along with laser photolysis of protons and immunohistochemistry, demonstrates that the unique ability of CSFcNs to function as precise CSF pH sensors arises from the distinct physiological properties of PKD2L1 channels and, crucially, their spatial segregation within the ApPr. These finding highlight how specialized anatomical and physiological features enable selective pH sensing.

From the anatomical standpoint, the functional evidence gathered here by combining whole-ApPr and outside-out recordings with laser-photolysis of protons and immunohistochemistry indicates that PKD2L1 channels are predominantly located in the ApPr. This is an important finding that highlights the role of the ApPr as a sensory compartment. In this work we show that when recording from CSFcNs a PKD2L1-mediated off current is evoked when uncaging protons on the ApPr, but not when uncaging on the dendrite or the soma. As expected, uncaging a few microns away from the ApPr is inefficient to trigger a response, which confirms the high spatial resolution of the technique²⁵. The other experiments presented in Figure 5 point to the same direction: i. outside-out recordings from the ApPr show PKD2L1-dependent single channel activity; ii. whole-cell recording from isolated ApPrs show single-channel currents that are indistinguishable from those recorded from intact ApPrs; iii. photolysis on isolated ApPr evokes off-currents as in intact cells; iv. immunohistochemical experiments show that PKD2L1 expression is higher in the ApPr than in the soma (see also³³). Although we have not attempted to do outside-out patches from somatic recordings, single channel recordings in this configuration has proved to be difficult by another group¹⁴. It is interesting to analyse this anatomical segregation of the receptors in the context of previous electron microscopy data, which shows that in the intact spinal cord a tight cytoskeleton ring (mainly adherens and tight junctions) formed by the ependymal cells^34,35 isolates the ApPr from the rest of the spinal parenchyma. Interestingly, it has been shown that the ionic composition of the CSF is different from that in the serum³⁶. This implies that the ApPr and the soma are exposed to different extracellular solutions, with a potential impact on the equilibrium potentials of the main ions involved in the CSFcNs electrical response. Although in our slice preparation the interphase separating the CSF from the spinal cord parenchyma is lost, it makes sense that the ApPrs are immersed in the CSF and physically separated, as they should if they are to sense CSF composition.

The physiological features of PKD2L1 channels need also to be taken into account to fully understand the role of CSFcNs as pH sensors: i. PKD2L1 channels are extremely sensitive to proton concentration in the physiological range. This has been quantified by González-Perrett et al.²³, who showed that the equilibrium constant for the related PKD2 channel expressed in lipid bilayers is 6.4 (see also²⁴). At this pH value the open probability of the channel is half of its maximum. We performed the same analysis (Figure 3Ce) and obtained a value close to 8.0. Interestingly, the holding current can also be used as a proxy of the channel sensitivity to pH changes (Figure 3Cf). The fact that physiological pH indicates the half-maximum of the conductance state of PKD2L1 channels activation curve defines their second physiological feature; ii. PKD2L1 channels are able to sense deviations of the pH in both directions, either by decreasing (acidification) or increasing (alcalinisation) the channel open probability. Enhanced excitability of CSFcNs resulting from increased PKD2L1 activity has been shown in mice¹⁵ and lamprey¹³, and has been attributed to augmented phasic activity of the channel. Remarkably, the fact that a decrease in channel activity can also provide meaningful information to the cell by inducing a strong hyperpolarization has not been described before. This exquisite sensitivity of the cell’s RMP to small changes in the membrane current can be explained by the high input resistance of the ApPr, which we have quantified here by performing direct recordings from that compartment; iii. finally, the physiological effects of PKD2L1 channels are achieved in part through the modulation of a tonic current. This makes the channel particularly well-fitted to act as a sensor of more or less slow pH variations. Indeed, the pH of the CSF does not change phasically within the millisecond time-range, as does the pH³⁷ or the neurotransmitter concentration³⁸ in the synaptic cleft, but rather in a slower fashion that requires special channels to transduce these slow and probably long-lasting changes.

The tonic current

In GABAergic systems, two different types of GABA_A receptors mediate phasic and tonic currents: one depends on the activation of low-affinity receptors by synaptically released GABA, while the other depends on the activation of high-affinity GABA_A receptors by low concentrations of ambient GABA³⁹. In this work we describe phasic, single-channel currents that depend on the spontaneous opening of PKD2L1 channels, and a type of tonic current that was made evident when PKD2L1 channels were blocked with dibucaine. The following experimental evidence indicates that both the phasic, single channel events, and the tonic current depend on the same channel: first, the effect of dibucaine persists when blocking voltage-dependent and ligand-gated channels that may also contribute to the appearance of a tonic current; second, the effect of dibucaine was still present when recording CSFcNs with an IS containing a high concentration of BAPTA; third, the tonic current was modulated by other exogenous (calmidazolium) and endogenous (pH) modulators that are known to affect PKD2L1 channels. These experiments suggest that, apart from the closed and open states of the channel, some other conducting state of the channel may exist as well. Alternatively, very short opening periods of the channels, not seen in the recordings but nevertheless present and blocked by dibucaine, could also explain the tonic current. The assesment between these two putative mechanisms is beyond the scope of this work.

The magnitude of the tonic current compared to the spontaneous, phasic currents, is substantial. The average single channel current measured in this work is ≈ -16 pA and po is ≈ 0.04, which determines a mean phasic current of 16 pA × 0.04 = 0.64 pA. On the other hand, the average tonic current (Figure 2Ca) is ≈ 8 pA. Altogether, the percentage of tonic and phasic currents are roughly 90 % and 10 %, respectively, which highlights the importance of the tonic current in the physiology of CSFcNs.

The off-response

It has been reported previously that PKD2L1 channels are responsible for an off-response when challenged by acidic solutions^24,29,30,40. A first indication of the presence of an off-current in CSFcNs was provided by Orts-del’Immagine¹⁴, who reported in some of the recorded neurons an increase in the frequency of single channel events after pressure applying a pH 2.8 solution. In this work we provide a detailed analysis of the off-current by using laser photolysis of protons and further show that the current is only evoked when uncaging at the ApPr, indicating that functional PKD2L1 channels are located there (see above). In the published literature the off-response appears with large-scale acidifications, in the order of several pH units. In this paper we have likewise induced substantial pH changes (to 3 or 3.5; see Methods) but have not attempted to quantify the pH dependence of the off-current. It is nevertheless interesting to highlight the fact that the off-current does not require long-lasting changes in the extracellular pH to develop. Instead, very brief pH changes are enough, which opens the possibility that the ApPrs may also respond to sudden ACSF acidifications, as may occur during neurosecretion⁴¹. It would be interesting in the future to explore with photolysis the concentration and time dependence of the off-current (for example, smaller pH changes but longer pulses) in order to assess its physiological significance.

ApPrs do not have sodium currents

The possibility of recording from iApPrs allowed us to make a serendipitous finding: ApPrs do not have active sodium conductances. This is not surprising as many CNS neurons do not have active conductances in their dendrites but is nevertheless interesting as the voltage deviations that occur in the ApPr need to propagate down the soma (and probably the axon) in order to have impact on the spike triggering zone. The fact that the ApPr and the soma are electrotonically close, as shown by Orts-Del’immagine and collaborators¹⁵, certainly facilitates this propagation. It would be interesting in the future to investigate specifically what are the voltage-dependent conductances to be found in the ApPr. In any case, the high IR of isolated ApPrs is probably an indication of the low density of active conductances in this neuronal compartment.

The involvement of ASIC

ASICs, also known as proton-gated channels, play important physiological and pathological roles in the nervous system^37,42. In CSFcNs, ASICs have been reported as mediating part of the response to acidification^13–15,31. In this study, we observed in a few cells a rapidly desensitizing inward current when pressure applying a pH 6.5 solution, and occasionally an inward current with similar characteristics when uncaging either MNI-Glu or MNI-γLGG at the soma (Supplementary Figure 2B). However, we never saw such an inward current when photolysing at the ApPr, and the off-response evoked by the photolysis of protons on the ApPr was not affected in the presence of ASIC channel blockers (supplementary Figure 3). As ASICs are widely expressed in many brain areas and neuronal compartments³², the most parsimonious explanation for these experimental results is that the ASIC-dependent current recorded in CSFcNs results from the activation of somatic ASICs, but that those channels are not present in the ApPr. ASICs may be suited to sense changes in the pH of the parenchyma by CSFcNs. The pH sensitivity of CSFcNs is thus a complex response that depends on different channels localized in different neuronal compartments. Our results indicate that the sensitivity of CSFcNs to changes in the pH of the ACSF is solely dependent on the modulation of PKD2L1 channels.

Materials and methods

Transgenic mice

Gata3^eGFP mice were generated using bacterial artificial chromosome (BAC) recombination as described⁴³. The eGFP gene was inserted by deleting 286 bp of coding sequences from the first exon. The sequences flanking the insertions were: 5’agccgaggac-eGFP–cgtggaccca3’. Four Gata3-eGFP founder lines were generated that expressed eGFP in an identical manner to GATA3 in the spinal cord. One of these lines was selected for this study. Animals of either sex were maintained in a 12 hs. light/dark cycle with food and water ad libitum, humidity between 50 and 60%.

Slice preparation for electrophysiological experiments

Spinal cord slices were obtained from Gata3^eGFP mice aged 30 to 45 days old. Slices were prepared as previously described³. Briefly, mice were decapitated under isoflurane anesthesia following approved ethical procedures (CEUA 001/01/2022a) and the spinal cord dissected in an ice-cold artificial ACSF of the following composition (in mM): 101 NaCl, 3.8 KCl, 1.3 MgSO₄.7H₂O, 1.2 KH₂PO₄, 10 HEPES, 25 Glucose, 1 CaCl₂, 18.7 MgCl₂, osmolarity 300 mOsm/kg H₂O and pH adjusted to 7.4. The spinal cord was included in a 4% low melting point agar in order to glue it in the desired orientation. 300 µm thick lumbar spinal cord slices were obtained following 2 dissecting planes, either transverse or with a 45 degrees angle. The latter orientation was used because we found that the probability of obtaining superficial, and hence easier to patch, ApPrs, was higher. Also, we found that the chances of obtaining “isolated” ApPrs (see Figure 5B) was also higher in the oblique preparation.

Electrophysiological recordings and epifluorescence

CSFcNs were recorded with the patch-clamp technique⁴⁴ at near physiological temperature (34 ± 1 °C, Peltier system; Luigs & Neurmann) on an upright Olympus microscope (BX51WI) equipped with a 60×, 1.0 numerical aperture objective. Once in the recording chamber, slices were continuously perfused with the following extracellular solution (in mM): 115 NaCl, 2.5 KCl, 1.3 NaH₂PO₄, 26 NaHCO₃, 25 glucose, 5 NaPyruvate, 2 CaCl₂.2H₂O and 1 MgCl₂.6H₂O, osmolarity 300 mOsm/kg H₂O and pH 7.4 with the continuous bubbling of a mixture of 95% O₂ and 5% CO₂. Recordings were performed with a KGluconate-based intracellular solution of the following composition (in mM): 165 KGluconate, 10 HEPES, 1 EGTA, 0.1 CaCl₂, 4.6 MgCl₂, 4 Na₂ATP, 0.4 NaGTP and 0.04 Alexa 594, pH 7.3, and osmolarity 300 mOsm/kg H₂O. With this solution, pipette resistance was 6 to 7 MΩ for somatic and 10 to 11 MΩ for ApPr recordings; series resistance, on the other hand, was monitored during the whole experiment, but not compensated for. Recordings were performed with a HEKA amplifier (EPC10 USB) and the software Patchmaster either in voltage-clamp (-60 mV holding potential) or current-clamp. Recordings were acquired at a sampling rate of 20 kHz and low-pass filtered at 2.9 kHz. Recording and puffing pipettes were positioned in the slice with Luigs & Neumann micromanipulators. Epifluorescence excitation was by light-emitting diodes in a dual lamp-house (Optoled; Cairn Research). EGFP was excited at 470/40 nm and Alexa 594 at 572/35 nm. Fluorescence emission at 520/40 nm and 630/60 nm was detected with an EM CCD camera (Andor Ixon) and filters from Chroma Corporation (Vermont, USA).

Photolysis

Protons were photo-released either from MNI-Glutmate (4-Methoxy-7-nitroindoli-inyl-caged-L-glutamate) or MNI-γLGG (4-Methoxy-7-nitroindoli-inyl-caged-γ-L-glutamyl-glycine) with a 405-nm diode laser (Obis, Coherent, USA) that was focused through the ×60/NA1.0 water immersion objective to a 1.5 µm diameter spot in the sample plane^25,45. The cages were diluted in the recording ACSF to a final concentration of 0.5 or 1 mM and either pressure-applied with a Picospritzer III (Parker) for at least 10 seconds before applying the light pulses, or bath applied. Photolysis light pulses had durations of 0.5 to 1 ms and 1 to 3 mW laser power. The power of the laser-pulse was monitored on a calibrated photodiode. The laser spot was fixed, and the photolysis location is chosen by positioning the slice on the region of interest. Before each experiment, the exact location of the laser spot was verified by monitoring it with a fluorescent solution.

Salts were either from Sigma-Aldrich or Carlo Erba. TTX (tetrodotoxin; 0.4 µM), TEA (tetraethy-lammonium; 1.5 mM), 4AP (4 aminopyridine; 1 mM), NBQX (2,3-Dihydroxy-6-nitro-7-sulfamoyl-benzo[f]quinoxaline; 10 µM), D-APV (2-Amino-5-phosphonovaleric acid; 50 µM) and gabazine (10 µM) were from Tocris. MNI-glutamate (1 mM) was from HelloBio and MNI-γLGG (1 mM) was a generous gift from Dr. Céline Auger (Sppin, CNRS UMR8003). Dibucaine hydrochloride (100 to 200 µM) and calmidazolium chloride (20 µM) were from Sigma-Aldrich. Psalmotoxin 1 (100 nM), APETx2 (100 nM) and TTA-P2 (20 µM) were from Alomone Lab.

Immunohistochemistry

Animals were anesthetized with ketamine (100 mg/kg, i.p.), xylacine (10 mg/kg, i.p.), and diazepam (5 mg/kg, i.p.) and fixed by intracardiac perfusion with 4% PFA in 0.1 M PB. The spinal cord was sectioned with a vibrating microtome (50 - 70 µm thick), placed in PBS with 0.5% BSA for 30 min, and then incubated with the primary mouse anti-PKD2L1 antibody (Millipore-Sigma AB9084, 1:500) in PBS with 0.3% Triton X-100 (Sigma Millipore). Sections were then incubated in the secondary donkey anti-mouse Alexa-647 antibody (Thermo-Fisher A-21235, 1:1000) and mounted in 70% (v/v) glycerol pH 8.8. Nuclei were stained with 1µg/ml Hoechst 33342 (Thermo-Fisher).

Confocal microscopy and image processing

Line selections were manually drawn through either the soma or ApPr of GFP+ cells on single deconvolved confocal planes and fluorescence intensity values per pixel along the selection were obtained using FIJI/ImageJ⁴⁶. Descriptive statistical values were used to build the average and SD plots against position along the selection (x).

Images were acquired on a Zeiss LSM800 confocal microscope with a 63× oil immersion lens (NA = 1.4), pinhole set at 0.8 Airy units. Sampling interval was set to an oversample density according to the Nyquist criterion. Z-stacks were deconvolved with Huygens Essential 4.5 (Scientific Volume Imaging B.V., Hilversum, Netherlands) using an experimental PSF, SNR = 20, using an iterative Classic Maximum Likelihood Estimation (CMLE) algorithm set to a maximum of 40 iterations, with a quality threshold of 0.05. Composite image assembly and further processing (brightness/contrast adjustment), when needed, was performed using FIJI/ImageJ.

Analysis

Electrophysiological analysis was performed with Igor Pro (Wavemetrics) using Neuromatic⁴⁷, TaroTools (https://sites.google.com/site/tarotoolsmember/?authuser=1) and custom routines. Single-channel currents were analyzed with the software Nest-O-Patch (NOP; https://sourceforge.net/projects/nestopatch/). The currents were filtered with the built-in filter of the software (cutt-off filter of 1.4 KHz) and the single-channel current amplitude calculated from a multigaussian fit adjusted to the whole-cell current histogram. Closed and open probabilities were calculated from the idealized current traces. The obtained values were then imported to IgorPro (8.0 or 9.0, Wavemetrics, Lake Oswego) for further analysis.

Current charge was calculated as the integral of the whole-cell recordings after baseline subtraction. Holding current in different conditions was calculated as follows: histograms were constructed from current recordings during 2 to 3 seconds time epochs. Then, the first peak of the resulting histogram, which corresponds to the baseline current, was fitted with a Gaussian function (IgorPro built-in function); the mean value of the fitting corresponds to the tonic current (see Figure 2Bb for an example). In order to show visually the tonic current change in different conditions, current traces were averaged over 100 ms epochs, like what has been done for the analysis of GABA_A-mediated tonic currents⁴⁸. For the analysis of resting membrane potential in different conditions, average Vm values were calculated from 1 second long time periods.

Input resistance was measured from the slope of the I/V curve resulting from voltage steps from -110 or 100 to -60 mV in 10 mV increments.

Estimation of the pH drop induced by photolysis

The photolysis of MNI-based compounds leads to proton release. The pH drop induced by the photolysis reaction depends on how many protons are released, on the speed of the buffering system and on the diffusion of molecules in and out of the illuminated volume. In our experiments we used caged-compounds concentrations between 0.5 and 1 mM. Based on our previous callibrations²⁵ that indicate that the photolysis is complete with the energy used here (≈ 2 mW for 1 ms), we expect an added charge of 0.5 to 1 mM protons in the illuminated volume, corresponding to a drop of pH from 7.4 to ≈ 3-3.5 pH units. This acid pH is quickly buffered by bicarbonate. Given that the speed of protonation of bicarbonate is extremely fast^49–51, the pH returns to 7.4 within microseconds. This indicates that the late current evoked by the photolysis (Figures 4, 5 and Supplementary Figure 4) has very similar characteristics to the PKD2L1-mediated current that has been described in expression systems in that it is not a delayed current, but rather an off-current that is produced by the opening of the channel when the acidic stimulus is removed. The fact that this off-current is so clearly seen when performing photolysis experiments and not with other methods, like pressure-application, likely reflects the superior temporal resolution of photouncaging. This approach enables exceptionally rapid pH jumps. Similarly, it has been shown by Hu et al⁵² that using fast application methods is a necessary condition to induce a type of response, that the authors called Ca⁺⁺ spike, when applying extracellular Ca⁺⁺ stimuli.

Statistics

Data are presented as mean ± SD. Statistical significance was assessed with the Wilcoxon signed-rank test for paired and the Wilcoxon–Mann–Whitney test for non-paired data. The difference between groups was considered significant when p < 0.05; when significant, the exact p value is indicated in each figure and figure legends.

Data availability

Numerical data used to make the figures of this manuscript have been deposited at http://doi.org/10.60895/redata/39X8AV.

Acknowledgements

This work was supported by a Agencia Nacional de Investigación e Innovación grant (number FCE_1_2021_1_166464) to Federico F. Trigo and a grant from the Wings For Life Spinal Cord Research Foundation (number WFL-UY-13/23 #290) to Raúl R. Russo. Magdalena Vitar was supported by ANII through a Master Student fellowship. We thank Victoria Falco Pastorino and María Gabriela Fabbiani for their help with the maintenance of the Gata3^eGFP colony, Céline Auger for the generous gift of MNI-γLGG, John ET Corrie for initial discussions on the proton photolysis and Alain Marty for the critical reading of the manuscript.

Additional information

Funding

Agencia Nacional de Investigación e Innovación (FCE_1_2021_1_166464)

Wings For Life Spinal Cord Research Foundation (WFL-UY-13/23 #290)