Axonal GABA_A receptors depolarize presynaptic terminals and facilitate transmitter release in cerebellar Purkinje cells

Abstract

Key points

• GABA_A receptors have been described in the axonal compartment of neurons; contrary to dendritic GABA_A receptors, axonal GABA_A receptors usually induce depolarizing responses.

• In this study we describe the presence of functional axonal GABA_A receptors in cerebellar Purkinje cells by using a combination of direct patch‐clamp recordings from the axon terminals and laser GABA photolysis.

• In Purkinje cells, axonal GABA_A receptors are depolarizing and induce an increase in neurotransmitter release that results in a change of short‐term synaptic plasticity.

• These results contribute to our understanding of the cellular mechanisms of action of axonal GABA_A receptors and highlight the importance of the presynaptic compartment in neuronal computation.

Abstract

In neurons of the adult brain, somatodendritic GABA_A receptors (GABA_ARs) mediate fast synaptic inhibition and play a crucial role in synaptic integration. GABA_ARs are not only present in the somatodendritic compartment, but also in the axonal compartment where they modulate action potential (AP) propagation and transmitter release. Although presynaptic GABA_ARs have been reported in various brain regions, their mechanisms of action and physiological roles remain obscure, particularly at GABAergic boutons. Here, using a combination of direct whole‐bouton or perforated patch‐clamp recordings and local GABA photolysis in single axonal varicosities of cerebellar Purkinje cells, we investigate the subcellular localization and functional role of axonal GABA_ARs both in primary cultures and acute slices. Our results indicate that presynaptic terminals of PCs carry GABA_ARs that behave as auto‐receptors; their activation leads to a depolarization of the terminal membrane after an AP due to the relatively high cytoplasmic Cl⁻ concentration in the axon, but they do not modulate the AP itself. Paired recordings from different terminals of the same axon show that the GABA_AR‐mediated local depolarizations propagate substantially to neighbouring varicosities. Finally, the depolarization mediated by presynaptic GABA_AR activation augmented Ca²⁺ influx and transmitter release, resulting in a marked effect on short‐term plasticity. Altogether, our results reveal a mechanism by which presynaptic GABA_ARs influence neuronal computation.

Key points

• GABA_A receptors have been described in the axonal compartment of neurons; contrary to dendritic GABA_A receptors, axonal GABA_A receptors usually induce depolarizing responses.

• In this study we describe the presence of functional axonal GABA_A receptors in cerebellar Purkinje cells by using a combination of direct patch‐clamp recordings from the axon terminals and laser GABA photolysis.

• In Purkinje cells, axonal GABA_A receptors are depolarizing and induce an increase in neurotransmitter release that results in a change of short‐term synaptic plasticity.

• These results contribute to our understanding of the cellular mechanisms of action of axonal GABA_A receptors and highlight the importance of the presynaptic compartment in neuronal computation.

Introduction

The presence of receptors for neurotransmitters and neuromodulators in the axonal compartment of neurons has been shown in a wide variety of preparations, and impacts neuronal communication by influencing axonal excitability and neurotransmitter release (Bucher & Goaillard, 2011). Since the seminal works by Dudel and Kuffler, and the Eccles group highlighting presynaptic inhibition mediated by Cl⁻‐conducting axonal ionotropic receptors (Rudomin & Schmidt, 1999; Willis, 2006), axonal GABA_ARs have been found at various excitatory and inhibitory neurons in the peripheral and central nervous systems, including the spinal cord, retina, calyx of Held, posterior pituitary, cerebral cortex, hippocampal and cerebellar granule cells, and cerebellar molecular layer interneurons (MLIs) (MacDermott et al. 1999; Trigo et al. 2008). While these studies show that the presence of axonal GABA_ARs is widespread, their physiological effects and mechanisms of action remain elusive because of technical limitations. To functionally dissect axonal GABA_ARs in detail, direct recordings of their local effects at presynaptic boutons are essential. However, the small size of presynaptic varicosities (∼1 μm) has hindered precise examination of axonal GABA_ARs’ function, except for large presynaptic structures such as the lamprey spinal cord (Alford et al. 1991), glutamatergic terminals of the calyx of Held (Price & Trussell, 2006; Turecek & Trussell, 2001, 2002) and the peptidergic terminals of neurohypothalamic neurons (Zhang & Jackson, 1995).

GABAergic neurons constitute a special case in relation to axonal GABA_ARs. Synaptically released GABA binds to receptors on the postsynaptic membrane and, concomitantly, to presynaptic axonal GABA_ARs, producing a so‐called auto‐receptor (auto‐R) response. GABA_A auto‐R responses have been almost exclusively studied in cerebellar interneurons because these cells have a relatively short axon and the auto‐R responses can be recorded directly from the soma (Mejia‐Gervacio & Marty, 2006; Pouzat & Marty, 1999; Trigo et al. 2007, 2010; Zorrilla de San Martin et al. 2015). However, the functional role of axonal GABA_A auto‐Rs on presynaptic membrane excitability and transmitter release remains unclear, notably because of the lack of feasible methods to study these. In this work we took advantage of the recently developed direct patch‐clamp recording technique from GABAergic boutons of cerebellar Purkinje cells (PCs; Kawaguchi & Sakaba, 2015) in order to directly assess the presence of axonal GABA_ARs and to study their functional role on these inhibitory boutons.

By using a combination of axon terminal recording and local laser photolysis of GABA on single axonal varicosities of fluorescence‐labelled PCs, we show that PC terminals possess functional GABA_A receptors behaving as auto‐Rs, which exert a depolarizing effect that travels passively along the axon due to a relatively abundant internal Cl⁻ and high electrical length constant. As a result, the axonal GABA_ARs exert a strong influence on the presynaptic Ca²⁺ influx and hence transmitter release, as assessed by paired recordings from single terminal and its postsynaptic partner. Notably, the post‐spike depolarizations produced by axonal GABA_ARs have a profound impact on short‐term synaptic plasticity. Finally, our whole‐bouton recording from a PC terminal in acute slices confirmed the presence of axonal GABA_ARs in situ.

Methods

Ethical approval

All animal experiments were performed in accordance with the principles of UK regulations, the guidelines regarding care and use of animals for experimental procedures of the National Institutes of Health, USA, Doshisha, Kyoto, and Paris Descartes Universities, and approved by the local committee for handling experimental animals in Doshisha University, Kyoto University, and the French Ministry of Research and the ethical committee for animal experimentation of Paris Descartes.

Culture preparation

The method for preparing cerebellar neuronal cultures was similar to that in previous studies (Kawaguchi & Hirano, 2007). Briefly, newborn rats (Wistar rat, obtained from Japan SLC, Inc.) of both sexes were decapitated and their cerebella were obtained, followed by incubation in Ca²⁺ and Mg²⁺‐free Hanks’ balanced salt solution containing 0.1% trypsin and 0.05% DNase for 15 min at 37°C. Cells were then dissociated by trituration and seeded on poly‐d‐lysine‐coated cover slips in DMEM/F12‐based medium containing 1% FBS. One day after the seeding, 75% of the medium was replaced with serum‐free BME, and PCs were infected with adeno‐associated virus (AAV) vector (serotype 2) carrying enhanced green fluorescent protein (EGFP) under the control of CAG promoter (AAV‐CAG‐EGFP) (Kawaguchi & Sakaba, 2015). PCs could be visually identified by their large cell bodies and thick dendrites, and EGFP fluorescence which preferentially labelled PCs by relatively specific infection of AAV serotype 2. The axon of a PC was clearly different from its dendrites, which have a high density of spines. Each week after seeding, half of the medium was replaced with fresh one containing 4 μm cytosine β‐d‐arabino‐furamoside to inhibit glial proliferation. Electrophysiological experiments were performed 3–5 weeks after seeding.

Slice preparation

Sprague Dawley rats aged 12–16 days old of both sexes (obtained from Janvier Laboratories, Le Genest‐Saint‐Isle, France) were used to prepare sagittal cerebellar slices following usual procedures. Briefly, the animal was placed under deep anaesthesia using isoflurane and decapitated. The cerebellar vermis was quickly removed and fixed on a stage with a special super glue (Topvalu, Japan). Parasagittal slices (202 μm thick) were cut with a vibroslicer (Leica VT1200S; Leica Microsystems, Wetzlar, Germany) in an ice‐cold artificial cerebrospinal fluid (ACSF) and then placed in an incubating chamber for 60 min at 34°C. Thereafter slices were kept at room temperature. The same bicarbonate‐buffered saline was used as the cutting and storing solution; composition (in mm): 115 NaCl, 2.5 KCl, 26 NaHCO₃, 1.3 NaH₂PO₄, 25 glucose, 2 CaCl₂ and 1 MgCl₂, 5 sodium pyruvate; osmolality adjusted to 300 mOsm (kg H₂O)⁻¹ and pH set at 7.4 by the continuous bubbling of a mixture of 5% CO₂ and 95% O₂.

Electrophysiology

Electrophysiological experiments on cultured PCs were performed similarly to previous studies (Díaz‐Rojas et al. 2015). Whole‐cell patch clamp recording from a cultured PC was performed with an amplifier (EPC10, HEKA, Germany) in a Hepes‐buffered extracellular solution containing (in mm): 145 NaCl, 5 KCl, 2 CaCl₂, 1 MgCl₂, 10 Hepes, and 10 glucose (pH 7.3, 300 mOsm (kg H₂O)⁻¹). Recordings from a PC terminal in the slice preparation was performed in a Hepes‐buffered extracellular solution containing (in mm): 128 NaCl, 4 KCl, 2.5 NaHCO₃, 10 Hepes, 25 glucose, 5 sodium pyruvate, 2 CaCl₂, and 1 MgCl₂, with pH 7.4 adjusted with 1 n NaOH and osmolality of 300 mOsm (kg H₂O)⁻¹. In some experiments, 2,3‐dioxo‐6‐nitro‐1,2,3,4‐ tetrahydrobenzo[f]quinoxaline‐7‐sulfonamide (NBQX, 10 μm, Tocris Biotechne, UK), TEA (2 mm), and tetrodotoxin (TTX, 1 μm, Wako, Japan) were used to inhibit glutamatergic EPSCs, K⁺ channels, and action potentials (APs), respectively. All the experiments except for slice recordings were performed at room temperature (22–25°C).

The postsynaptic potentials (PSPs) and currents (PSCs) were detected using TaroTools extensions (https://sites.google.com/site/tarotoolsregister/) implemented in Igor Pro (Wavemetrics, Lake Oswego, OR, USA), and the selection of each event was visually confirmed before subsequent analysis. For the amplitude and half‐width analysis of APs recorded by gramicidin‐perforated current‐clamp recordings before and after picrotoxin application: more than 30 representative APs were selected for each cell in each condition, and the mean amplitude and half‐width were calculated. It should be noted that we excluded APs when the prior inter‐spike interval was less than 10 ms or when the subsequent interspike interval was above 30 ms to avoid the contamination of voltage‐gated conductances with different timings. Phase plots shown in Fig. 4 B were constructed from 21 spikes in each condition, control and picrotoxin. The APs selected for this analysis fulfilled the following criteria: the AP was not preceded by another AP during the previous 20 ms; the AP was not followed by another AP for the next 30 ms; the onset V _m of the APs was within a 5 mV range (control = −69.4 ± 1.6 mV, range −71.5 to −66.6 mV; picrotoxin = −68.5 ± 1.2 mV, range −70.4 to −66.4 mV). Even after this careful selection, it has to be noted that with gramicidin‐perforated patch‐clamp recordings there are technical limitations in correctly measuring rapid membrane potential changes like APs because of the relatively high access resistance (∼100 MΩ). Thus, our analysis of APs may be failing to detect a slight change of fast components of APs.

Figure 4. Depolarizing GABA_AR‐mediated signals at PC terminals.

A, representative traces of a PC terminal membrane potential recorded with the gramicidin‐perforated patch technique in current‐clamp mode before (left) and after application of GABA_AR blocker picrotoxin (right). The resting membrane potential is around −73 mV (dotted grey line). Black circles show spontaneous events and grey arrowheads post‐spike depolarizations, which are both depolarizing. Both signals were blocked by picrotoxin. B, left: expanded average AP (from >50 traces) recorded in control conditions (black) and in the presence of picrotoxin (green). In the presence of picrotoxin the post‐spike depolarization disappeared, giving rise to a clearer after‐hyperpolarization. B, middle: membrane potential value measured 15 ms after the AP (dotted rectangle in the left panel) in control and during picrotoxin. B, right: phase plots of the APs recorded in control (black) and during picrotoxin (green), showing similar dynamics of membrane potential change during the APs in two conditions. Ca, representative traces of a perforated patch experiment in which GABA was photolysed on top of the varicosity at different holding potentials; the current reverses between −50 and −40 mV. The arrowhead indicates the timing of the laser pulse. Cb, average amplitude of the laser evoked GABA_A currents (normalized to the amplitude at −90 mV), as a function of holding potential. The line is the linear fit to the data points, and indicates an E _GABA of −46.4 mV (n = 4 terminals). Error bars represent SEM. D, representative membrane potential change recorded from a PC soma with the gramicidin‐perforated patch technique in current clamp. Spontaneous PSPs are depolarizing below −70 mV (grey arrowheads) and hyperpolarizing above −70 mV (black arrowheads). Ea, representative GABAergic postsynaptic currents recorded from a PC soma in voltage clamp at different holding potentials (gramicidin‐perforated patch recording). Eb, average amplitude of the GABA_A‐mediated spontaneous events as a function of the holding potential. The line is the linear fit to the data points, and indicates an E _GABA of −73.4 mV (n = 5 cells). Error bars represent SEMs.

When calculating the electrical length constant in a PC axon, λ, the V _m changes in response to current injections (10 pulses for 10 ms, with 20 pA steps) recorded in current‐clamp mode were corrected off‐line for the voltage error due to the series resistance.

In order to quantify the resting V _m in current‐clamp recordings (Fig. 4), the raw trace was smoothed with a built‐in box algorithm (15 points, 300 μs intervals) in Igor pro, and a V _m histogram constructed from the smoothed data. The histogram shows a skewed distribution because of the presence of PSPs and spikes, so that a Gaussian distribution was fitted to the unskewed portion of the histogram. The mean value of that Gaussian was considered to be the resting V _m.

Intracellular solutions (ISs)

Presynaptic terminals of cultured PCs axons were recorded with a patch pipette (14–17 MΩ) filled with one of the following ISs (in mm):

• (1)For the experiments of Figs 1, 5 and 6: 152 CsCl, 0.5 EGTA, 10 Hepes, 10.5 CsOH, 2 ATP, 0.2 GTP.
• (2)For the experiments of Figs 2 and 3: 147 KCl, 5 EGTA, 10 Hepes, 2 ATP, 0.2 GTP.
• (3)For perforated patch‐clamp recordings (Fig. 4): a KCl‐based solution containing 10–20 μg ml⁻¹ gramicidin was freshly prepared before the experiments.

Figure 1. Axonal GABA_AR‐mediated currents in cultured PCs.

A, fluorescence images of a cultured PC that expresses EGFP (green). The inset shows a magnified image of the patched terminal that makes a synaptic contact on a DCN‐like cell. B, representative voltage‐clamp recordings from a PC terminal before and after application of picrotoxin (50 μm). An arrowhead and an arrow highlight an AP and the associated PSC, respectively. Picrotoxin abolished synaptic current‐like events. C, amplitude vs. time‐to‐peak of the spontaneous synaptic‐like currents recorded at −70 mV. D, left: representative voltage‐clamp recording from a PC terminal where GABA was photolysed from 0.2 mm DPNI‐GABA with a 405 nm laser (1 mW power and 1 ms duration) in control conditions and in the presence of picrotoxin. D, right: pooled peak amplitude data of current responses to GABA photolysis from 5 different terminals recorded in control and in picrotoxin (means ± SEM: 25.8 ± 11.7 pA (control) and 0.5 ± 0.2 pA (picrotoxin). E, left: response of an outside‐out patch excised from a PC terminal to DPNI‐GABA photolysis at the end of the whole‐terminal recording. Grey sweeps are the individual responses (6 repetitions); black sweep is the average. E, right: pooled peak amplitude data of current responses to GABA photolysis in 6 different outside‐out patches (n = 6, mean ± SD = 10.6 ± 7.2 pA). This response is similar to that obtained in D, except that the mean amplitude is smaller. In D and E, the arrowheads correspond to the timing of the laser pulse.

Figure 5. Short voltage pre‐pulses control presynaptic Ca²⁺ currents and transmitter release.

A, recording configuration: a paired recording is performed between a PC terminal (green, left pipette) and its postsynaptic partner (right pipette). B, voltage command (V _com) waveforms used. C, the simultaneously recorded presynaptic Ca²⁺ currents (Ca) and the GPSCs (Cb) are shown. Each trace is averaged from 15 trials. The different colours correspond to the different voltage commands shown in B. D, peak Ca²⁺ current normalized to the value without the pre‐pulse plotted as a function of the voltage pre‐pulse. E, peak GPSC amplitude normalized to the value without the pre‐pulse as a function of the voltage pre‐pulse. F, normalized GPSC as a function of the normalized peak Ca²⁺ current. The dotted line corresponds to a fit with a power function (power: 3.5 ± 0.8). D, E and F show pooled data from 6 different pairs.

Figure 6. Effects of after‐spike depolarizations on short‐term plasticity.

A, two voltage command waveforms consisting of 5 APs at 50 Hz were applied: the control one (AP alone, black traces) and the test one, where each AP was followed by a post‐spike depolarization (AP+GABA dep, red traces) that mimics the GABA_AR‐mediated post‐spike depolarization (∼10 mV). The AP trains were applied from a holding membrane potential of −60 mV (left) or −70 mV (right). B and C, presynaptic Ca²⁺ currents (B) and corresponding GPSCs (C) in response to the AP trains shown in A, AP alone (black) and AP+GABA dep (red), both from either −60 mV (left) or −70 mV (right) holding potential. The insets in B show Ca²⁺ currents induced by the 2nd AP in an expanded time scale (scale bars: 400 μs and 30 pA). D and E, presynaptic Ca²⁺ current amplitudes (D) and GPSC amplitudes (E) are normalized to the response to the 1st stimulation. Control data in response to AP alone are shown in black, and those with the post‐spike depolarization in red. Data are shown for three different holding potentials: −60 (top), −70 (middle) and −80 mV (bottom). Error bars correspond to SEMs. n = 5 pairs for holding potential at −60 mV, and n = 7 pairs for −70 mV and −80 mV.

Figure 2. Determination of the length constant, λ, in PC axons.

A, left: paired recordings from two different terminals of the same PC axon. The pictures on the left show the fluorescence image of the recorded terminals, with the exact position of the recorded varicosity shown with the schematic pipettes. Two examples are shown: in the upper figure the terminals are separated by a straight axon collateral with a length of 21 μm, while in the lower picture the terminals are joined together by a collateral axon loop with a length of 136 μm. The inset in the bottom figure represents the portion of axon connecting the two recorded varicosities (their positions shown with circles). A, right: current‐clamp traces of membrane potentials in the recorded terminals that were alternately stimulated by different current steps. V ₍₁₎ and V ₍₂₎ correspond to the membrane potential of terminal 1 and 2, respectively, and I ₍₁₎ and I ₍₂₎ to the current injected to either terminal 1 or 2, respectively. B, the amplitude ratio between the voltages recorded in both terminals is plotted against the distance between them. In each pair of terminals, the amplitude ratio was calculated by averaging the ratio of the V _m changes caused by all current injections. The continuous line shows an exponential fit to the data points, with an estimated λ of 121 ± 30 μm.

Figure 3. Passive propagation of subthreshold GABA_A‐mediated axonal signals in PC axons.

A, fluorescence image of an axonal segment of a cultured PC expressing EGFP. The arrowheads indicate the locations of the different photolysis spots (shown in B). B, voltage‐clamp (Ba) and current‐clamp (Bb) recordings of the responses evoked by GABA photolysis at the varicosities shown in A; as the GABA uncaging was performed farther away, the electrical responses got smaller. C, laser‐evoked GABA current vs. distances between laser spot location and recording site. The continuous line corresponds to the exponential fit to the data points, showing a D _37% of 47 μm (n = 36 spots from 7 cells). D, voltage vs. current responses induced by GABA photolysis. The continuous line corresponds to the linear fit to the data points (n = 14 spots from 8 cells).

The postsynaptic deep cerebellar nuclei (DCN) neurons were voltage‐clamped with a pipette containing (in mm): 152 CsCl, 0.5 EGTA, 10 Hepes, 10.5 CsOH, 2 ATP, 0.2 GTP. All ISs used for recording from cultured neurons were adjusted to 310 mOsm (kg H₂O⁻¹) and pH 7.3.

Basal membrane capacitance (C _m) was 2–3 pF, and series resistance of the terminal recordings (typically 50 MΩ) was compensated by ∼30–80 %. Synaptic transmission specific to the patched PC terminal was triggered by applying short voltage pulses (Fig. 5) or action potential waveforms (Fig. 6) to the PC terminal in the presence of TTX (1 μm) and TEA (2 mm), and GABAergic postsynaptic currents (GPSCs) were recorded from the postsynaptic DCN cell. We analysed recordings in which series resistance for postsynaptic DCN was <30 MΩ and leak current <−300 pA.

Presynaptic terminals of slice PCs were recorded with a patch pipette filled with the following IS (in mm): 150 mm KCl, 1 mm EGTA, 10 mm Hepes, 0.1 mm CaCl₂, 4.6 mm MgCl₂, 4 mm NaATP, 0.4 mm Na₂GTP, and, 0.04 mm Alexa 594, osmolality adjusted to 300 mOsm (kg H₂O⁻¹) and pH set at 7.3.

GABA photolysis

GABA was photolysed from DPNI‐GABA (1‐(4‐aminobutanoyl)‐4‐[1,3‐bis(dihydroxy‐phosphoryloxy)propan‐2‐yloxy]‐7‐nitroindoline, purchased from Tocris Biotechne, UK) with a 405 nm laser (OBIS, Coherent, USA). For the experiments in cultured PCs, the laser beam was focused through an Olympus 40×, 0.9 NA objective, giving a spot of ∼3 μm in diameter. For slice experiments, the laser beam was focused through an Olympus 60×, 1.0 NA water immersion objective, giving a spot of ∼2 μm in diameter (as in Zorrilla de San Martin et al. 2015). For both slice and culture experiments, DPNI‐GABA was diluted in a Hepes‐buffered extracellular solution to a final concentration of 0.2 to 1 mm, and photolysis performed with 1–3 ms, 0.3–2 mW laser pulses.

Two‐photon reconstruction of PCs in slices

Fluorescence stacks were performed with a custom‐built 2P system constructed around an Olympus BX51W upright microscope and based on the design of Tan et al. (1999) and Ducros et al. (2011). Alexa 594 was excited at 910 nm with a MaiTai T‐Saphire laser (Spectra Physics, USA). The scanning and the acquisition of fluorescence images were performed with custom routines (https://github.com/brandonStell/Igor2p) written in Igor Pro.

Statistics

Data are presented as means ± SEM. Statistical significance was assessed by Student's paired t test, unpaired t test, Mann‐Whitney U test or ANOVA.

Results

Direct recordings of GABA_AR currents in PC terminals

In order to explore directly whether GABA_ARs occur in the axonal domain of PCs, we took advantage of the dissociated culture preparation, where PCs extend long axons (>1 mm) with several bifurcations and form many synaptic contacts on their target neurons. These terminals are relatively large (2–3 μm) and amenable to patch recordings (Fig. 1 A; Díaz‐Rojas et al. 2015; Kawaguchi & Sakaba, 2015). Cultured PCs were labelled with EGFP by an AAV vector serotype 2, which shows a relatively specific transfection of PCs, and the fluorescence was used to precisely position a patch pipette with a small tip at the selected axonal varicosity. When recording from a PC terminal with a Cl⁻‐based internal solution (152 mm CsCl), we observed spontaneous inward currents resembling synaptic responses (Fig. 1 B, control). These signals were completely blocked by picrotoxin (50 μm), a GABA_AR antagonist, indicating that they were dependent on the activation of GABA_ARs (Fig. 1 B). The sharp, spontaneous inward currents that reflect the propagation of APs from the soma, were not blocked by the inhibition of GABA_ARs. Some of the large amplitude spontaneous events were associated with unclamped spikes (arrowhead and arrow in Fig. 1 B). A similar association of spikes with large inward postsynaptic currents (PSCs) was previously described for auto‐R currents in cerebellar interneurons (Pouzat & Marty, 1999). In most of PC boutons (7 out of 8 boutons), such auto‐R currents can be observed after APs, although the amplitudes are variable. The average amplitude of the spontaneous events was 9.5 ± 4.0 pA, and their 10–90 % rise time 13.1 ± 5.7 ms (Fig. 1 C). The variability of the rise times suggests that spontaneous events arise at a wide range of distances from the recording site.

As shown in Fig. 1 A, the patched terminals were usually located several hundreds of micrometres away from the soma. Therefore, it seems likely that the spontaneous events recorded from PC terminals reflect the passive transmission of spontaneous GABA_A currents arising from neighbouring axonal, but not from somatodendritic regions. To investigate the presence of GABA_ARs in PC axon terminals, we turned to local laser photolysis of GABA. Laser photolysis with a minimized spot has a high spatial resolution, on the order of a few micrometres (Trigo et al. 2009; Zorrilla de San Martin et al. 2015). Once a terminal was recorded, DPNI‐GABA was added to the bath at a concentration of 0.2–1 mm and photolysed with a 405 nm laser (1 ms, 1 mW pulses). As shown in Fig. 1 D, GABA photolysis at a PC axonal varicosity caused an inward current that was fully blocked by GABA_A antagonists (group results in Fig. 1 D, right; control amplitude: 25.8 ± 11.7 pA; in bicuculline: 0.5 ± 0.2 pA; n = 5). The spatial resolution of our photolysis apparatus was 5.8 μm estimated by changing the laser spot location relative to the patched varicosity. In addition, the outside‐out recordings from patches of membrane excised from different PC boutons exhibited similar (although variable in amplitude) inward currents upon local GABA uncaging (Fig. 1 E, n = 6), further supporting the GABA_AR presence at PC terminals. Altogether, these experiments demonstrate that presynaptic terminals of cultured PCs possess GABA_ARs that are activated by local application of GABA and that give rise to synaptic like spontaneous events.

Passive propagation of local electrical signals in a PC axon

The above results raise the possibility that the signal originating from the activation of GABA_ARs in a varicosity may spatially spread and reach neighboring axonal regions. According to cable theory the extent of signal propagation in an axon depends on the axon length constant, λ. In spite of the importance of λ for subthreshold signal computation among presynaptic boutons, technical limitations have hindered direct determination of that basic parameter. In order to obtain a quantitative estimate of λ, we recorded simultaneously from two different varicosities of the same axon. Figure 2 shows such an experiment, where two different varicosities were recorded simultaneously in current‐clamp mode while injecting current in either of them. The local voltage changes produced by different amounts of current injection in one varicosity were transmitted passively to the other (Fig. 2 A and B). As expected from cable theory (Rall, 1969), the voltage changes decayed with distance following an exponential relationship (Fig. 2 B) with a length constant, λ, of 121 ± 30 μm (obtained from 10 pairs). Thus, in an axon region that is distant from the soma, subthreshold electrical signals spread over ∼100 μm, supporting the idea that GABA_AR‐induced currents can travel from one terminal to the others.

The value of the length constant λ estimated by DC signal coupling between two boutons characterizes a physical property of an axon for maximal passive travelling of electrical signals, in the absence of any conductance activation. In order to test whether axonal GABA_AR‐mediated currents indeed travel along an axon, we recorded from a single varicosity either in voltage‐ or current‐clamp mode while uncaging GABA either on top of it or at progressively increasing distances along the axon (Fig. 3 A). The maximal current or voltage change was obtained when GABA was photolysed on top of the recorded varicosity, and the amplitude progressively decreased as the photolysis was performed farther away (Fig. 3 B and C). In these experiments, we only targeted other varicosities (enlargements of the axonal trunk) in order to minimize the differences in GABA_AR density and/or number among the different axonal sections. Group results indicated that the response decayed exponentially with distance, with a length constant value of 47 μm (Fig. 3 C). We call this value D ₃₇ (distance at which the signal decays to 37% of its maximal value), in order to differentiate it from λ. In addition, we found a linear relationship between current and potential changes, recorded at each site in voltage and current clamp, respectively, suggesting that the transmission occurs only passively and does not involve any contribution of voltage‐dependent conductances (Fig. 3 D). The smaller D ₃₇ value estimated from Fig. 3 C compared with λ from Fig. 2 B might be ascribed to the following reasons: λ defined in Fig. 2 is for steady state, and transient signals (like GABA_A‐mediated responses) are more strongly filtered; the recordings of Fig. 3 were performed with a high Cl⁻ internal solution to maximize the driving force through the limited number of local GABA_ARs. Due to the gradual decline of Cl⁻ concentration with distance from the pipette, the local driving force for GABA_AR currents should have decreased as a function of distance from the recording site, reducing the apparent D ₃₇ value. Finally, the shunting effect produced by the opening of the GABA_ARs may also contribute to this difference. The present estimate of D ₃₇ is comparable to the value previously estimated from somatic recordings of GABAergic responses in the axon in cerebellar interneurons (47 μm vs. 56 μm; Zorrilla de San Martin et al. 2015).

Depolarizing effect of axonal GABA_ARs in a PC bouton

By using whole‐bouton recordings, we have demonstrated so far that presynaptic terminals of cultured PCs possess GABA_ARs and that the signal caused by their activation in a single bouton can propagate with a length constant of 50–100 μm. The next question we asked is whether axonal GABA_ARs in intact PC boutons induce depolarizing or hyperpolarizing responses. For this purpose, we performed perforated patch experiments from either a terminal or the PC soma with pipettes containing gramicidin, which forms Cl⁻‐impermeable cation pores in the plasma membrane without disrupting the intracellular Cl⁻ concentration. When a terminal was recorded with the gramicidin‐perforated patch technique in current‐clamp mode, we observed under resting conditions spontaneous, depolarizing synaptic potential‐like events (Fig. 4 A, black circles). These events were similar in time course to somatically recorded PSPs, and had a large amplitude (on the order of 10 mV). In addition, we observed spontaneous APs, where each AP was followed by a depolarizing post‐spike depolarization resembling the spontaneous PSP‐like events (Fig. 4 A, grey arrowheads). Both the spontaneous synaptic potential‐like events and the post‐spike depolarizations were sensitive to picrotoxin (Fig. 4 A, right). Interestingly, they were always depolarizing, suggesting a relatively depolarized reversal potential for GABA_AR‐mediated responses (E _GABA) in PC boutons. After GABA_AR blockade by picrotoxin, the membrane potential value after the spike was significantly hyperpolarized (the V _m 15 ms after the AP was: control −68.4 ± 1.1 mV; picrotoxin −74.8 ± 0.4 mV, n = 5, P = 0.003, Fig. 4 B, left and middle), suggesting that axonal GABA_A auto‐Rs depolarize the presynaptic membrane after an AP. On the other hand, we could not find any significant changes in the AP frequency and waveform after the GABA_AR blockade (ratio of frequency, peak amplitude, and half‐width between control and picrotoxin: 0.64 ± 0.16, P = 0.13; 0.95 ± 0.08, P = 0.74; and 1.09 ± 0.08, P = 0.32, respectively; n = 5 cells). Phase plots of APs arising from a similar V _m and interval after the preceding AP also showed similar AP profiles in both conditions (Fig. 4 B, right). However, we would like to note that a slight effect of presynaptic GABA_ARs on the AP waveform, if present, might be masked because of the technical limitations of the perforated current‐clamp recordings, which are accompanied by relatively high access resistance.

In order to measure the exact E _GABA value at PC terminals, we recorded the laser‐induced GABA_A‐mediated responses in perforated voltage‐clamp recordings (Fig. 4 C). As shown in the example in Fig. 4 Ca, the response reverses at around −50 mV. The estimated E _GABA obtained from the I–V relationship of the GABA_A‐mediated responses was −46.4 mV (Fig. 4 Cb, n = 4), confirming the results by the previous experiments.

In contrast to PC terminals, gramicidin‐perforated current‐clamp recordings from the soma failed to display large GABAergic depolarizing PSPs around the resting potential. As can be seen in Fig. 4 D, the spontaneous GABAergic synaptic events were of small amplitude, and were either depolarizing (grey arrowheads) or hyperpolarizing (black arrowheads) depending on the membrane potential. This is different from the axonal recordings, where all the responses were depolarizing (Fig. 4 A), again supporting the idea that the GABA_AR‐mediated responses at boutons were not originating from the soma. In the example shown in Fig. 4 D, the GABA_A reversal potential is close to −70 mV (dotted grey line); this was confirmed when the GABA_A‐mediated spontaneous events were recorded at different holding potentials (Fig. 4 E). In a group analysis of voltage clamp results, the somatic E _GABA was estimated at −73.4 mV (Fig. 4 Eb). This average estimate of E _GABA was close to the value obtained from Fig. 4 D, and is also close to previously published values in slice preparations (Chavas & Marty, 2003; Eilers et al. 2001). In summary, gramicidin‐perforated patch‐clamp recordings indicate that axonal GABA_ARs mediate a depolarizing response in primary cerebellar cultures. Furthermore, our experiments indicate a difference between the axonal and the somatic E _GABA, suggesting a Cl⁻ gradient between the axonal and the somatodendritic compartments, as previously shown in glutamatergic presynaptic boutons at calyx of Held and cortical neurons (Price & Trussell, 2006; Khirug et al. 2008).

Enhanced Ca²⁺ influx and transmitter release by a depolarizing effect of axonal GABA_ARs

Our results so far showed that presynaptic GABA_ARs exert a depolarizing effect in PC boutons, eliciting depolarizations lasting tens of milliseconds and occurring either spontaneously or after an AP. Next we went on to assess the physiological impact of these depolarizations on subsequent synaptic transmission. Changes in the AP waveforms (either changes in amplitude or width, Bischofberger et al. 2002; Kawaguchi & Sakaba, 2015), and the V _m value for hundreds of milliseconds before an AP (long pre‐pulses, Bouhours et al. 2011; Christie et al. 2011; Bialowas et al. 2014; Rama et al. 2015) were shown to impact on the presynaptic Ca²⁺ influx and hence transmitter releases at several synapses, such as the calyx of Held, hippocampal mossy fibre boutons, MLIs and PC boutons. Based on the duration of GABA_A‐mediated depolarizations, we hypothesized that even the shortest membrane potential change (tens of milliseconds) might affect the activation of voltage‐gated Ca²⁺ channels and/or synaptic transmission, and systematically explored the effects of short pre‐spike depolarizations or hyperpolarizations. For this purpose, we performed simultaneous recordings from a single terminal and the postsynaptic soma in the presence of TTX (1 μm) and TEA (2 mm) in order to measure simultaneously presynaptic Ca²⁺ currents and GABAergic postsynaptic currents (GPSCs) (Fig. 5 A). Presynaptic transmitter release was triggered by voltage waveforms consisting of a sharp depolarization step (to +30 mV for 0.3 ms) with an exponential decay (tau, 0.3 ms) preceded by 20 ms pre‐pulses to different potentials (Fig. 5 B). Under these conditions, presynaptic Ca²⁺ current increased linearly with the pre‐pulse increase (Fig. 5 Ca and D), while GPSC amplitudes followed a supra‐linear relationship, with a clear take‐off around −70 mV (Fig. 5 Cb and E). The GPSC amplitude plotted as a function of the Ca²⁺ current amplitude (Fig. 5 F) showed a non‐linear relationship characterized by a ∼4th power dependency, in line with a previous study (Díaz‐Rojas et al. 2015). Although we ignore the mechanism(s) of this facilitation, the short pre‐pulse depolarizations could induce a priming of the Ca²⁺ channels that would lead to a higher probability of opening once the AP gets to the bouton. Thus, even small and short terminal depolarizations like the ones induced by the activation of axonal GABA_ARs can have a profound impact on Ca²⁺ channel activation and hence on synaptic release.

Taking into consideration that PCs fire spontaneously at tens of hertz, GABA_AR‐dependent post‐spike depolarizations (Fig. 4 B, left) might affect the presynaptic potential just before the following APs. This suggests that the effects of presynaptic GABA_ARs on Ca²⁺ influx and transmitter release become evident during repetitive activation of a PC terminal. However, it is not possible to simply examine how the synaptic transmission is affected by the inhibition of presynaptic GABA_ARs, because GABA_AR blockers will also suppress postsynaptic responses. Therefore, to explore the impact of presynaptic GABA_ARs on a repetitive synaptic transmission protocol, we induced AP trains in the voltage‐clamped PC terminal (5 APs at 50 Hz) either with or without the GABA_AR‐mediated post‐spike depolarizations. The AP waveform recorded from a PC terminal was used as a template for the voltage command, and a small and slow depolarization (10 mV; time to peak, 10 ms) was added to mimic the effect of presynaptic GABA_ARs (Fig. 6 A). As expected, presynaptic Ca²⁺ currents showed facilitation upon repetitive stimulation, reflecting Ca²⁺‐induced facilitation of Ca²⁺ current through P‐type channels (Díaz‐Rojas et al. 2015). The post‐spike depolarization augmented the facilitation of Ca²⁺ influx during the train (Fig. 6 B and D), resulting in a significant enhancement of synaptic transmission (Fig. 6 C and E). The facilitation of Ca²⁺ influx and the resultant increase in synaptic release was more pronounced when the V _m value before the train was more depolarized (Fig. 6 D and E). As a result, the AP train with post‐spike depolarizations caused a radical change in the short‐term facilitation phenomena of PC synaptic outputs. Thus, the depolarization induced by presynaptic GABA_ARs during AP trains are sufficient to augment Ca²⁺ influx and transmitter release from PC terminals, giving rise to a form of synaptic facilitation that has not been described before.

Axonal GABA_ARs at PC boutons in slices revealed by direct whole‐cell recordings

We finally investigated whether PCs in a slice preparation also exhibit axonal GABA_ARs. In order to do that we prepared sagittal cerebellar slices from rats aged 12–16 days old (for details see the Methods section) and combined direct patch‐clamp recordings from PC boutons with GABA photolysis. We aimed at axonal varicosities of PC collaterals in the PC layer (not in the DCN region), because these are non‐myelinated and easily visualized. PC presynaptic terminals were first identified via visual inspection, and identification was confirmed using morphological or physiological criteria (see below). Figure 7 A shows an example of a terminal recording from a PC varicosity. The axon varicosity was patched with a KCl‐based internal solution (to maximize the driving force through the GABA_ARs, as in the culture experiments in Fig. 1), together with Alexa 594 for the morphological characterization. In two recordings, the identity of the varicosity was unambiguous because the axon was intact and the somatodendritic compartment (as well as the rest of the axon) was retrogradely back‐filled with the Alexa dye during the recording (see Fig. 7 A). In the other case, although the patched varicosity was not connected to the soma because the axon had been damaged during the slicing procedure, the axon could be followed into the granule cell layer for a long distance, where it was connected to the main axonal tract running from the PC layer into the white matter, a typical branching pattern of PC collaterals. Once a stable recording was obtained in voltage clamp, the bath perfusion was stopped and DPNI‐GABA was added to the bath at a final concentration of 0.5–1 mm. Local photolysis of DPNI‐GABA induced an inward current that was blocked by gabazine, a GABA_AR antagonist (Fig. 7 Ab). In the example shown, uncaging was performed in three locations (location 1, 2 and 3), yielding diverse response profiles. Location 1 gave the shortest rise time, probably indicating that the GABA_ARs activated by the photolysis were at the recorded bouton. When GABA was photolysed at other locations (like location 3 in the figure), GABA_A‐currents were observed with a slower rise time, in a similar manner to cultured PC boutons (see Fig. 3), suggesting the passive travelling of GABA_AR‐mediated signals in the PC axon. As shown in Fig. 7 B, amplitudes of laser‐evoked GABA_A currents in slices (12 different varicosities from 3 cells) were comparable to those in cultured PC terminals (see Fig. 1).

Figure 7. Direct recordings from axonal boutons of PCs in a slice preparation reveals the presence of axonal GABA_AR.

Aa, reconstructed two‐photon fluorescence image of a PC that was filled with Alexa 594 from the recorded terminal. The inset shows a detail of the axon and the locations where GABA was photolysed. The recording pipette is highlighted with yellow lines. The approximate locations of the PC layer (PCL), molecular layer (ML) and granule cell layer (GCL) are indicated. Ab, laser‐evoked GABA_AR currents upon uncaging of DPNI‐GABA at the locations shown in Aa. The currents are blocked by gabazine. Bottom traces show the timing (and intensity) of the laser pulse. B, amplitudes of axonal GABA_AR currents evoked when uncaging GABA at different boutons. The data correspond to the recordings performed from three boutons from three different cells. From a single patched bouton, the laser‐evoked GABA_A currents at other, neighbouring boutons, were also recorded (boutons 2 and 3 in A). C, voltage‐dependent K⁺ currents showing high sensitivity to 1 mm TEA, a characteristic feature of PC axon terminals.

Although presynaptic terminals from basket cells lay near those of PC axon collaterals, sensitivity to TEA, a K⁺ channel blocker, consolidated our recordings from a PC axon terminal, in addition to morphological confirmation. While both PC and basket cell terminals express the alpha Kv1.1 subunit, the basket cell terminals also abundantly express Kv1.2 which has low sensitivity to TEA (Sheng et al. 1994). As a result, in contrast to a high sensitivity of the PC terminals to low doses (1 mm) of TEA (Kawaguchi & Sakaba, 2015), basket terminals should show substantial voltage‐dependent K⁺ current even in the presence of low concentration of TEA. Accordingly, voltage‐dependent outward K⁺ currents were largely blocked by 1 mm TEA applied at the end of each experiment (Fig. 7 C). Taking all our results together, they confirm that presynaptic GABA_ARs occur in PC axons in slices, as in cultured PCs.

Discussion

Taking full advantage of the possibility to perform direct presynaptic terminal recordings, the present work reveals for the first time the presence and functional role of GABA_ARs in the axonal domain of cerebellar PCs. Contrary to somatodendritic GABA_A receptors, PC axonal GABA_ARs cause a depolarizing response in presynaptic terminals, suggesting a Cl⁻ gradient between the somatodendritic and the axonal compartments. Our paired recordings from two neighbouring terminals quantitatively demonstrated substantial propagation of subthreshold signals arising from single varicosities to neighbouring boutons. Paired recordings between a single terminal and its postsynaptic partner show that short depolarizations like those caused by axonal GABA_ARs can have a profound facilitating effect on presynaptic Ca²⁺ currents and hence, on transmitter release.

Axonal GABA_ARs in PC boutons

The presence of functional axonal GABA_ARs in PCs was demonstrated here by direct terminal recordings in combination with laser photolysis of DPNI‐GABA. The discovery of axonal GABA_ARs in PCs adds to the very extensive list of different neuronal types where axonal ionotropic receptors for neurotransmitters, specifically GABA, have been described (Ruiz et al. 2003; Trigo et al. 2008). From that list, we can speculate that GABA_ARs are generally located not only at somatodendritic compartments where synaptic inputs arrive, but also at axonal compartments to fine‐tune the neuronal output signals.

The source of GABA that activates axonal GABA_ARs varies among different preparations and constitutes an important issue because it determines the time course of GABA_AR activation. Our data indicate that multiple activation patterns exist for GABA_ARs located at PC boutons. As expected, GABA_ARs in PC axons are self‐activated (see Fig. 4 A: the APs are followed by a depolarization), in a similar manner to the ones in cerebellar interneurons (Pouzat & Marty, 1999) and hence behave as auto‐Rs. In addition, our data also demonstrate spontaneous synaptic‐like responses mediated by axonal GABA_ARs irrespective of own AP firing (see Figs 1 B and 4 A). This still indicates other activation modes, like GABA spillover from other cells or other synapses, as in hippocampal mossy fibre boutons (Alle & Geiger, 2007) and cerebellar parallel fibres (Stell, 2011; Dellal et al. 2012; Pugh & Jahr, 2013), or even GABA released from its own varicosity in an AP‐independent manner, like the ‘preminis’ observed in MLIs (Trigo et al. 2010).

Mechanism of presynaptic function regulation by axonal GABA_ARs

The effect of activation of GABA_ARs on membrane potential is difficult to assess because intracellular recording methods change artificially the intracellular Cl⁻ concentration and hence, E _GABA. This problem can be overcome by the perforated patch technique, where electrical access is attained by using gramicidin, which keeps the intracellular Cl⁻ concentration unperturbed. However, it is still challenging to apply the perforated patch‐clamp technique to an axon terminal far away from the soma and study the physiological action of GABA_ARs locally. To date, this attempt has been limited to the giant glutamatergic terminal of the calyx of Held synapse, which showed a high Cl⁻ concentration and hence a depolarized E _GABA (Price & Trussell, 2006). In this work we applied that technique to a GABAergic PC terminal, and compared the GABA responses at a terminal and that at the soma (see Fig. 4). Our data showed systematic depolarizing actions mediated by GABA_ARs at terminals, in clear contrast to the inhibitory GABAergic effects in the soma. From our perforated patch experiments both in the soma and in the terminals, which show different E _GABA values in the two compartments (−73.4 mV in the soma and −46.4 mV in the terminals), we can estimate a somatic intracellular chloride concentration of 5 mm and an axonal intracellular chloride concentration around 20 mm. Thus, the internal Cl⁻ concentration is not homogeneous among different neuronal compartments and is relatively high at axonal compartments in GABAergic PCs, consistent with recent studies suggesting that axonal GABA_ARs are depolarizing (see, for example, Price & Trussell, 2006; Khirug et al. 2008). Furthermore, we showed that the shape (amplitude and time course) of APs in the terminals is not affected directly by axonal GABA_ARs (see Fig. 4 B). It should be noted that the bouton depolarization just after an AP was sometimes accompanied by the occurrence of another spike with a smaller amplitude (see Fig. 4 A), which may be locally triggered by the depolarization mediated by GABA_A auto‐Rs at some axonal varicosities.

Axonal GABA_AR‐mediated depolarizations are expected to have biphasic effects (Trigo et al. 2008; Stell, 2011): strong depolarizations cause inhibitory effects, due to the inactivation of Na⁺ or Ca²⁺ channels or to shunting effects (Zhang & Jackson, 1995); small depolarizations cause excitatory actions by modulating Ca²⁺ influx or the AP width (this study; Ruiz et al. 2010). The functional consequences of the presynaptic membrane depolarizations are far from being understood mainly due to technical limitations to study quantitatively the relationship between local presynaptic membrane excitability and transmitter release. To overcome these obstacles, we systematically analysed the effects of membrane potential changes imitating the axonal GABA_ARs’ action on presynaptic Ca²⁺ currents and synaptic transmitter release, using paired voltage‐clamp recordings from a terminal and its postsynaptic partner. Short (20 ms duration) voltage steps before an AP, like the ones mediated by axonal GABA_ARs, change the amplitudes of presynaptic Ca²⁺ currents (see Fig. 5 Ca and D). As a result, GABAergic postsynaptic responses are modulated supra‐linearly, keeping the ∼3.5 power relationship between GPSCs and Ca²⁺ currents (Fig. 5 F). In addition, such an enhancement of Ca²⁺ currents by presynaptic GABA_A auto‐Rs resulted in an enhancement of short‐term facilitation of transmission during high frequency activation (see Fig. 6). Thus, axonal GABA_AR‐mediated post‐spike depolarizations contribute to setting the exact sign of short‐term synaptic plasticity, which is an essential component of information transfer in neuronal networks (Zucker & Regehr, 2002). In the future, it would be interesting to study the effect of the post‐spike depolarizations under different regimes of presynaptic activity. These findings provide a detailed mechanism of presynaptic GABA_AR action on synaptic transmission, and are consistent with previous studies suggesting effects of presynaptic subthreshold voltage fluctuations on transmitter release (Awatramani et al. 2005; Scott et al. 2008; Bouhours et al. 2011; Christie et al. 2011; Rowan & Christie, 2017).

At present, the precise mechanism of the linear change of Ca²⁺ current amplitudes by short subthreshold membrane potential changes is unclear. The PC boutons exhibit a clear Ca²⁺‐dependent facilitation of Ca²⁺ currents during high frequency activation (Díaz‐Rojas et al. 2015), indicating that not all of the Ca²⁺ channels are activated during each fast AP (half‐width, 400 μs). In such a case, depolarization at even very negative potentials may shift the states of a small population of Ca²⁺ channel subunits towards ‘open’ configuration (although still ‘closed’ as a single channel), resulting in higher probability of openings upon the arrival of an AP.

Analogue signalling in the axon by GABA_ARs and its functional implications

Our data demonstrated the substantial passive propagation of subthreshold electrical signals beyond 100 μm even in distal regions of an axon (see Figs 2 and 3) by direct measurements for the first time in the CNS. Such an analog signalling in the axon has been found in several preparations (Pouzat & Marty, 1999; Alle & Geiger, 2006; Shu et al. 2006; Paradiso & Wu, 2009; Trigo et al. 2010; Dellal et al. 2012; Pugh & Jahr, 2013; Zorrilla de San Martin et al. 2015; Zbili et al. 2016), and the axonal AP signalling is modified by subthreshold voltage changes, in a manner similar to a hybrid (analog and digital) transmission (Clark & Häusser, 2006; Debanne et al. 2011). PCs are the sole output neurons of the cerebellar cortex, and mainly target neurons in the DCN. An individual DCN neuron receives convergent inputs from 40 PCs (Person & Raman, 2011) and each axon forms many contacts on the same target. Because of the passive travelling of electrical signals in a PC axon, even a small number of GABA_ARs at each bouton could cooperatively act to change the local membrane potential in an axon (as shown in Fig. 4). Thus, axonal GABA_ARs may work as a monitor and integrator of the total level and pattern of GABAergic inhibition around the postsynaptic target, making it possible to locally adjust the level of presynaptic activities.

In our slice experiments we targeted the presynaptic varicosities of PC collaterals in the cerebellar cortex, around the PC layer, which showed clear GABA_AR‐mediated currents upon GABA photolysis (Fig. 7). Considering the apparent non‐selective observations of GABA_AR responses upon GABA photolysis along a PC axon (see Fig. 3), GABA_ARs are expected to be transported non‐selectively to boutons over long distances, also reaching synapses on DCN neurons in vivo. PC collaterals form synapses not only on other PCs (Orduz & Llano, 2007; Bornschein et al. 2013; Díaz‐Rojas et al. 2015; Witter et al. 2016), but also on MLIs (Witter et al. 2016), Lugaro cells (Witter et al. 2016), and granule cells (Guo et al. 2016). If axonal GABA_ARs are also present in PC boutons at these contacts, computation in the cerebellar cortex may be orchestrated through the local modulation of PC bouton activities by GABA, in addition to the control of cerebellar cortical outputs in the DCN region. An interesting issue to be addressed in the future is whether PC boutons show different target‐dependent GABA_A conductances and play different roles at distinct synapses. It would also be important to study whether there is a developmental change in the expression of axonal GABA_ARs in PCs, similar to what has been described for molecular layer interneurons of the cerebellar cortex (Pouzat & Marty, 1999; Trigo et al. 2010).

Additional information

Competing interests

The authors declare no competing financial interests.

Author contributions

The primary culture experiments were performed in Graduate School of Brain Science, Doshisha University, Kyoto, Japan and Graduate School of Science, Kyoto University, Kyoto, Japan. The slice experiments were performed in the Brain Physiology Laboratory, CNRS UMR8118 and Université Paris Descartes, Paris, France. J.Z.deS.M.: conception and design, acquisition and data analysis. F.F.T: conception and design, data analysis and interpretation, manuscript writing. S.K.: conception and design, data collection, analysis and interpretation, manuscript writing. All authors have approved the final version of the manuscript and agree to be accountable for all aspects of the work. All persons designated as authors qualify for authorship, and all those who qualify for authorship are listed.

Funding

This work was supported by grants from the JSPS/MEXT, Japan (KAKENHI Grants number 15K06722, 15KT0082, 16H01284) to S.K., JSPS and CNRS under the Japan–France Research Cooperative Program (PRC1283) to S.K. and F.F.T., the Naito Science Foundation to S.K., and JSPS Core‐to‐Core Program A Advanced Research Networks.