Heterosynaptic GABAergic plasticity bidirectionally driven by the activity of pre- and postsynaptic NMDA receptors

Significance

The plasticity of inhibitory synapses is involved in regulating the balance between excitation and inhibition. Although a wide range of plastic mechanisms was analyzed in GABAergic synapses, a bidirectional plasticity engaged by NMDA signaling has never been observed. We report here that the synapse between Golgi cells and granule cells in the cerebellar granular layer can be modulated by heterosynaptic plasticity in a bidirectional manner, depending on NMDA receptors. Long-term potentiation is expressed in response to postsynaptic NMDA_R activation via nitric oxide retrograde diffusion. Conversely, long-term depression is triggered by presynaptic NMDA_R activation. The present results provide a plastic mechanism controlling signal processing in neural networks.

Abstract

Dynamic changes of the strength of inhibitory synapses play a crucial role in processing neural information and in balancing network activity. Here, we report that the efficacy of GABAergic connections between Golgi cells and granule cells in the cerebellum is persistently altered by the activity of glutamatergic synapses. This form of plasticity is heterosynaptic and is expressed as an increase (long-term potentiation, LTP_GABA) or a decrease (long-term depression, LTD_GABA) of neurotransmitter release. LTP_GABA is induced by postsynaptic NMDA receptor activation, leading to calcium increase and retrograde diffusion of nitric oxide, whereas LTD_GABA depends on presynaptic NMDA receptor opening. The sign of plasticity is determined by the activation state of target granule and Golgi cells during the induction processes. By controlling the timing of spikes emitted by granule cells, this form of bidirectional plasticity provides a dynamic control of the granular layer encoding capacity.

Inhibitory neurotransmission regulates the stability of neuronal circuits by tuning, timing, gain, and excitability of principal neurons (1, 2). In turn, the efficacy of inhibitory synapses can be regulated by long-term potentiation (LTP) or depression (LTD) (3). Inhibitory plasticity participates in the maintenance of circuit stability and circuit refinement by shaping the dynamic range of excitatory neurons (4). At the molecular level, a broad spectrum of mechanisms can activate inhibitory plasticity (5–8). Among these mechanisms, the diffusion of nitric oxide (NO) following postsynaptic NMDA receptor (NMDA_R) opening was shown to mediate LTP (9). Conversely, circuit disinhibition through GABAegic LTD has been observed to be promoted by presynaptic NMDA_Rs (10). However, the coexistence in the same synapse of potentiation and depression engaged by NMDA signaling has never been reported yet. The cerebellum coordinates sensory–motor function via the firing patterns of Purkinje cells (PCs), which are in turn modulated by the granular layer activity. Recent findings highlighted the role of single granule cells (GrCs) in processing cerebellar signals during locomotion, motor coordination, and sensory–motor integration (11, 12). Furthermore, the local computation in the granular layer is under control of Golgi cell (GoC) inhibitory activity through feedback and feedforward loops (13). GABAergic inhibition in fact regulates GrC output gain (14) and allows the tuning of first spike delay and precision (15) by implementing “time-windowing” operations (4). In addition, plastic mechanisms residing throughout the granular layer circuit are likely involved in shaping neural code. Among these mechanisms, the high-frequency discharge of mossy fibers (mfs) repeated in θ-rhythms potentiates glutamate release at the mf–GrC synapse through retrograde diffusion of NO (16). Here, we examined how a θ-burst stimulation (TBS) delivered to glutamatergic mfs influences the strength of neighboring GABAergic contacts. We found that GABAergic LTP (LTP_GABA) and LTD (LTD_GABA), which are engaged by NMDA signaling, modulate GrC firing and excitability, controlling signal processing in the granular layer.

Results

TBS Persistently Alters Inhibitory Neurotransmission in the Cerebellum Granular Layer.

The mfs of the cerebellum convey information through bursts of high-frequency action potentials (17). These bursts can be organized in θ-band rhythm during specific tasks, such as spatial exploration (18), or synchronized to other brain regions during associative learning (19). θ-Burst activity was shown to induce glutamatergic LTP (LTP_glu) in vitro and in vivo (20, 21). Because synaptic inhibition impacts LTP_glu (21, 22), we investigated how TBS affected GrC firing in the presence of ongoing inhibition. Changes in GrC firing were monitored in current clamp configuration. The expression of LTP_glu was monitored by recording excitatory postsynaptic currents (EPSCs) elicited by paired pulse protocol in voltage clamp at −70 mV (Fig. 1A). We initially evaluated the effects of TBS in cells responding with suprathreshold activity to a single impulse delivered to the mf bundle (Fig. S1A). In 6 of 10 recorded cells, TBS delivered in current clamp configuration induced a marked LTP_glu (15) (+42.6 ± 4.3%, n = 6, P < 0.01) (Fig. 1A). As expected in response to the expression of LTP_glu (22), these cells showed a decrease of first spike latency together with an increase of first spike precision and of the probability of eliciting spikes (Fig. 1 B and C). Unexpectedly, in these GrCs, we observed a decrease in the number of spike doublets (Fig. 1 B and C). In the remaining GrCs, TBS was unable to induce LTP_glu (EPSCs +4.6 ± 6.1%, n = 4 cells, P > 0.3) (Fig. 1D). Interestingly, these cells showed a marked increase of spike doublets (Fig. 1 E and F) and a nonsignificant increase of probability of firing, whereas first spike delay and first spike precision were not altered (Fig. 1 E and F). Similarly, we analyzed GrCs showing subthreshold activity in response to single mf impulses. Excitatory postsynaptic potentials (EPSPs) had an average time to peak of 7.9 ± 0.7 ms and a variable residual depolarization persisting for tens of milliseconds. A total of 5 of 11 cells showed LTP_glu (EPSCs +38.4 ± 6.1%, n = 5 cells, P < 0.01) (Fig. 1G). Accordingly, EPSPs showed amplitude increase and time-to-peak decrease (Fig. 1 H and I, see also ref. 22). In addition, despite the expression of LTP_glu, these cells exhibited a decrease of the total and residual depolarization (Fig. 1 H and I). In the remaining 6 cells, although EPSCs were unchanged (peak 3.8 ± 5.9%, n = 6 cells, P > 0.4) (Fig. 1J), TBS favored a small increase of all of the EPSP-related parameters (Fig. 1 K and L). To understand the origin of unexpected changes induced by TBS, we correlated values of spike- and EPSP-related parameters measured before TBS with the amount of glutamatergic potentiation. Cells undergoing LTP_glu initially showed higher probability of firing together with a larger number of spike doublets (Fig. S1B). Moreover, we observed a positive correlation between LTP_glu and EPSP amplitude and residual depolarization (Fig. S1D). Conversely, first spike jitter and delay poorly correlated with the amount of LTP_glu expressed (Fig. S1C). Because EPSCs evoked in cells expressing LTP_glu were not significantly different from the ones detected in nonpotentiated cells (47.2 ± 6.6 pA LTP_glu vs. 43.7 ± 7.3 pA non-LTP_glu for spiking cells, P > 0.8, 19.3 ± 2.9 pA LTP_glu vs. 18.1 ± 3.1 pA non-LTP_glu for nonspiking cells, P > 0.7), we hypothesized that excitatory inputs were similar among tested GrCs. We therefore speculated that diversities in inhibitory rather than excitatory inputs could produce the unexpected changes induced by TBS. We analyzed how the progressive reduction of GABAergic inhibition by SR9519 (gabazine) affected GrC responses. An increasing concentration of gabazine markedly enhanced the probability of firing and the number of doublets (Fig. S1 E and F), whereas poorly affecting first spike delay and jitter. Furthermore, gabazine increased EPSP amplitude and total and residual depolarization (Fig. S1 G and H). We concluded that parameters unexpectedly altered by TBS were tightly dependent on the amount of GABAergic inhibition. This evidence led us to hypothesize that besides LTP_glu, TBS could also alter the strength of inhibitory synapses.

Fig. 1.

Impact of TBS on GrC activity. (A, Upper) Average EPSCs obtained before (black) and after (gray) TBS in a GrC undergoing LTP_glu. (Scale bar: 10 ms, 20 pA.) (Lower) Time course of normalized EPSC changes (n = 6). (B) Voltage traces taken from the same GrC shown in A and generating suprathreshold responses following a single mf impulse before (Upper) and after (Lower) TBS. (Scale bar: 10 ms, 20 mV.) (C) Time course of spike-related parameters. Histogram summarizes average changes in cells undergoing LTP_glu (first spike delay −28.4 ± 4.3%, **P < 0.01; first spike SD −43.4 ± 7.2%, **P < 0.01; P firing +53.5 ± 19.4%, **P < 0.01; doublets −95.7 ± 2.3%, **P < 0.01, n = 6 cells). (D, Upper) Average EPSCs obtained before (black) and after (gray) TBS in a GrC in which LTP_glu was not expressed. (Scale bar: 10 ms, 20 pA.) (Lower) Time course of normalized EPSC changes (n = 4). (E) Voltage traces taken from the same GrC shown in D. (Scale bar: 10 ms, 20 mV.) (F) Time course of spike-related parameters. Histogram summarizes average changes in cells in which LTP_glu was not expressed (first spike delay −8.3 ± 3.3%, P > 0.1; first spike SD +4.5 ± 15.4%, P > 0.9; P firing +40.9 ± 22.7%, P < 0.08; doublets +322.2 ± 57.2%, **P < 0.01, n = 4 cells). (G, Upper) Average EPSCs obtained before (black) and after (gray) TBS in a GrC undergoing LTP_glu and showing subthreshold activity in response to a single mf impulse. (Scale bar: 10 ms, 10 pA.) (Lower) Time course of normalized EPSC changes (n = 5). (H) Voltage traces taken from the same GrC shown in G before (Upper) and after (Lower) TBS. (Scale bar: 10 ms, 5 mV.) (I) Time course of EPSP-related parameters. Histogram summarizes average changes on EPSP properties in cells undergoing LTP_glu [peak amp +45.3 ± 11.1%, *P < 0.05; time to peak −12.8 ± 2.5%, *P < 0.05; total depolarization (tot dep) −14.9 ± 2.4%, **P < 0.01; depolarization at 50 ms poststimulus (50 ms dep) −16.6 ± 4.2%, **P < 0.01, n = 5 cells]. (J, Upper) Average EPSCs obtained before (black) and after (gray) TBS in a GrC in which LTP_glu was not expressed and showing subthreshold activity in response to a single mf impulse. (Scale bar: 10 ms, 10 pA.) (Lower) Time course of normalized EPSC changes (n = 5). (K) Voltage traces taken from the same GrC shown in J. (Scale bar: 10 ms, 5 mV.) (L) Time course of EPSP-related parameters. Histogram summarizes average changes on EPSP properties in cells in which LTP_glu was not expressed (peak amplitude +13.9 ± 6.4%, *P < 0.05; time to peak +29.7 ± 9.1%, **P < 0.01; tot dep 26.8 ± 6.8%, **P < 0.01; 50 ms dep +39.5 ± 10.3%, **P < 0.01, n = 6 cells).

Fig. S1.

Properties of GrC firing. (A) Schematic drawing of the granular layer microcircuit. The stimulating electrode (stim) was initially positioned onto the mf bundle to indirectly activate feedforward inhibitory circuit. In GrCs responding with firing activity, a single stimulus generated a complex of EPSPs (46.5 ± 4.4% of total responses, n = 10 cells) (Materials and Methods and Fig. 1 B and E) and single spikes (53.8 ± 3.9%, n = 10 cells) (Fig. 1 B and E). Only occasionally, doublets of action potentials could be observed throughout recordings (14.5 ± 9.3%, n = 10 cells). In a subset of cells responding with suprathreshold activity, TBS induced LTP_glu, which was correlated with spike-related parameters measured before TBS. (B) Correlation of the amount of LTP_glu with the initial number of doublets (●) and the initial probability of firing (○). (C) Correlation of the amount of LTP_glu with the initial average first spike delay (●) and jitter (○). (D) Cells responding with subthreshold activity were analyzed and correlated with their response to TBS. Correlation of the amount of LTP_glu with the initial EPSP peak (●) and residual depolarization (○). (E) Ten traces from a GrC generating suprathreshold responses following mf stimulation obtained in the presence of increasing concentration of gabazine (0 µM, Top, 5 µM Middle, and 10 µM Bottom). (Scale bar: 20 mV, 10 ms.) (F) Relationship between spike-related parameters and the concentration of gabazine (n = 6 cells). Black circles, Left axis (P firing: probability of eliciting at least one spike 0.45 ± 0.02 with 0 µM gabazine, 0.98 ± 0.02 with 5 µM gabazine, and 1 in 10 µM gabazine; first spike delay 9.3 ± 0.4 ms, with 0 µM gabazine, 9.7 ± 0.6 ms with 5 µM gabazine, and 9.9 ± 0.9 ms with 10 µM gabazine); gray circles Right axis (doublets: number of times over 10 repetitions GrCs responded with at least two spikes following a single stimulus 0.08 ± 0.01 with 0 µM gabazine, 5 ± 1.3 with 5 µM gabazine, and 9.6 ± 0.04 with 10 µM gabazine; P < 0.01; first spike jitter: SD of first spike delay 1.1 ± 0.03 ms with 0 µM gabazine, 1.2 ± 0.1 with 5 µM gabazine, and 1.3 ± 0.09 with 10 µM gabazine). (G) Ten traces from a GrC showing subthreshold responses following mf stimulation obtained in the presence of increasing concentration of gabazine (0 µM, Top, 5 µM Middle). (Bottom) Average EPSPs obtained with 0 µM and 5 µM gabazine. Note that 10 µM gabazine is not shown because EPSPs were always converted into suprathresold responses. (H) Relationship between EPSP-related parameters and the concentration of gabazine (n = 6 cells). Black circles, Left axis (EPSP: peak amplitude 9.7 ± 0.5 mV with 0 µM gabazine, 17.8 ± 0.4 mV with 5 µM gabazine; mV*ms: EPSP integral between onset and 50 ms poststimulus 327 ± 16 mV*ms with 0 µM gabazine, 442 ± 18.3 mV*ms with 5 µM gabazine, in 10 µM gabazine cells responded with spikes), gray circles (ttp: time to reach EPSP peak 8.5 ± 0.6 ms with 0 µM gabazine, 9.6 ± 0.7 with 5 µM gabazine, in 10 µM gabazine cells responded with spikes; 50 ms: difference between membrane potential before the stimulus and the depolarization at 50 ms poststimulus 3.1 ± 0.4 mV with 0 µM gabazine 5.9 ± 0.9 with 5 µM gabazine, and 7.8 ± 0.5 in 10 µM gabazine).

LTP at GoCs–GrCs Is Heterosynaptic and Does Not Require GABAergic Activity.

To investigate long-term plasticity at the inhibitory connections between GoCs and GrCs, we simultaneously activated GoC axons and mf terminals (Fig. S2A). We initially evoked EPSCs, which showed properties compatible with the ones elicited by direct mf stimulation (Fig. S2 B and C, i). Inhibitory postsynaptic currents (IPSCs) were then isolated by voltage clamping GrCs at 0 mV (Fig. S2 C, ii). The block of rapid AMPA glutamatergic currents by 20 μM 2,3-dihydroxy-6-nitro-7-sulfamoyl-benzo[f]quinoxaline-2,3-dione (NBQX) allowed to unmask monosynaptic evoked IPSCs (eIPSCs) showing faster kinetics (Fig. S2 C, iii) than polysynaptic eIPSCs. The application of TBS while holding GrCs at −40mV to release the intracellular magnesium from NMDA channels (Fig. S2 C, iv and Materials and Methods) induced a marked increase of eIPSC first peak amplitude (Fig. 2A and Fig. S2 C, v). This potentiation persisted for up to 40 min after the induction (Fig. 2B). The increase of GABAergic currents was associated with a decrease of response variability [coefficient of variation (CV)] (Materials and Methods and Fig. S2C) and of paired pulse ratio (PPR) (Fig. S2D). On average, LTP_GABA was expressed as a persistent increase of eIPSC peak amplitude (+53.6 ± 7.3%, n = 19, P < 0.001) (Fig. 2B) together with a decrease of PPR (−20.2 ± 4.4%, n = 19, P < 0.001) (Fig. S2D) and of response CV (−29.6 ± 5.9%, n = 19, P < 0.001) (Fig. S2E). Notably, in 5 of 19 cells, after NBQX washout, GrC voltage clamp conditions allowed unmasking of a potentiation of EPSCs (+17.5 ± 1.8%, n = 5, P < 0.01), suggesting the concomitant expression of LTP_glu at the neighboring mf–GrC synapse. Besides changes of evoked currents, TBS induced an increase of spontaneous currents (sIPSCs) (Fig. S3A) elicited by the endogenous activity of GoCs (23). Furthermore, the frequency of sIPSC occurrence was unaffected (Fig. S3A), suggesting an increase of neurotransmitter release. The analysis of IPSC kinetics, revealed an increase of IPSC decay (Fig. S3A), indicating that alterations of postsynaptic receptor activity could not be ruled out. Moreover, the initial IPSC amplitude correlated with the initial EPSC amplitude (R² = 0.79, n = 22). Furthermore, in cells undergoing LTP_GABA (19 of 22; 86%), the amount of potentiation was proportional to the intensity of initial responses for both IPSCs (R² = 0.82) (Fig. S3B) and EPSCs (R² = 0.85) (Fig. S3B). Instead, in 3 of 22 cells (14%), TBS induced no significant changes, likely reflecting the absence of plasticity or, alternatively, the existence of compensatory mechanisms. A fundamental role in GrC neurotransmission is played by the tonic component of GABAergic inhibition, which is mediated by high-affinity extrasynaptic GABA_A receptors, activated by persistent GABA concentrations in the glomerulus (24). We thus investigated whether LTP_GABA also affected the tonic component of inhibition. The expression of LTP_GABA was not accompanied by changes in holding currents (1.9 ± 2.5%, n = 18 cells, P > 0.4) (Fig. S3C) or in peak-to-peak noise of the holding current (0.5 ± 1.3%, n = 18 cells, P > 0.4) (Fig. S3C), indicating that the tonic component was not affected by LTP_GABA. We then investigated whether the induction of LTP_GABA depended on the activity of excitatory synapses. When electrical stimulation did not elicit glutamatergic currents, LTP_GABA was not observed (six of six cells) (Fig. S3D), suggesting the importance of glutamate in the induction of GABAergic potentiation. Given its dependence on glutamatergic activity, we asked whether LTP_GABA also depended on GABAergic activity. The TBS was therefore delivered in the presence of 20 µM gabazine, whose transient application reversibly blocked spontaneous and evoked currents, confirming their GABAergic origin (Fig. 2C). Following drug washout, IPSCs recovered to potentiated levels (+84.1 ± 16.8%, n = 7, P < 0.01) (Fig. 2 C and D), demonstrating the heterosynaptic nature of LTP_GABA. A marked decrease of PPR and CV (Fig. S4 A and B) was also measured. Different types of glutamatergic signaling have been shown to mediate GABAergic LTP in other brain structures. We hypothesized that the opening of NMDA_Rs during TBS-dependent depolarization could trigger intracellular Ca²⁺ increase, in turn inducing LTP_GABA. Recordings performed in the presence of 50 µM d-(-)-2-Amino-5-phosphonopentanoic acid (d-APV) indeed showed that IPSCs were not altered by TBS (−2.6 ± 3.5%, n = 5, P > 0.5) (Fig. 2 E and F), confirming that LTP_GABA was critically dependent on NMDA_R activity.

Fig. S2.

The impact of GABAergic inhibition on granule cell neurotransmission. (A) Schematic drawing of the granular layer microcircuit. The stimulating electrode (stim) was positioned in the granular layer to directly stimulate GoC axon and mf terminals. (B) EPSCs evoked by positioning the stimulating electrode across mf bundle (Top: peak 31.4 ± 4.8pA, PPR 0.7 ± 0.06, n = 21) or in the granular layer (Middle: peak, 29.1 ± 2.7 pA, PPR 0.7 ± 0.05; n = 33). Average traces obtained after peak normalization (Bottom) show that EPSCs had similar kinetics. (C) Scheme of the experimental protocol. (i) EPSCs recorded from a GrC voltage clamped at −70 mV (black, 20 traces superimposed; gray, average EPSC). (ii) IPSCs obtained from the same cell by clamping GrC at 0 mV. IPSCs showed an average first peak amplitude of 53.4 ± 5.5 pA and current kinetics (r_10–90 2.6 ± 0.3 ms; τ 21.3 ± 1.1; n = 43). (iii) IPSCs recorded in the presence of NBQX to eliminate IPSCs activated by disynaptic pathway and eventual residual AMPA currents. Monosynaptic IPSCs showed an average first peak amplitude of 56.5 ± 4.9 pA (n = 32) and a PPR of 0.81 ± 0.04 (n = 32) without significant differences from control solution (P < 0.4). (Current kinetics r_10–90 1.7 ± 0.4 ms; τ 19.2 ± 0.7 n = 43). (iv) TBS was delivered while voltage clamping GrC at −40 mV. Traces show two 100-Hz bursts superimposed. (v) Monosynaptic IPSCs recorded post-TBS showed an average first peak amplitude of 80.4 ± 7.7 pA (n = 32). (D, Left) Time course of PPR of IPSCs taken from the same cell shown in Fig. 2A. (Right) Time course of normalized PPR changes (n = 18). Histogram in the Inset shows average PPR before (black) and after (white) TBS (0.083 ± 0.03 vs. 0.65 ± 0.04 n = 18; **P < 0.001). (E, Left) Time course of CV of IPSCs taken from the same cell shown in C. (Right) Time course of normalized CV changes (n = 18). Histogram in the Inset shows average values of CV before and after TBS (0.14 ± 0.03 vs. 0.07 ± 0.01 n = 18; **P < 0.01).

Fig. 2.

TBS potentiates the activity of GoC–GrC synapses. (A) Time course of IPSC first peak amplitude in a single GrC. Traces in the Inset show average responses before (black) and after (gray) TBS. (Scale bar: 30 pA, 20 ms.) (B) Time course of normalized IPSC changes (n = 19). Histogram shows average IPSC values before (white) and after (black) TBS (50.1 ± 8.3 pA vs. 80.5 ± 9.7 pA n = 19, **P < 0.01). (C) Time course of IPSCs first peak amplitude in a GrC transiently perfused with gabazine. TBS was delivered during gabazine perfusion (horizontal bar). Traces in the Inset show average responses before (black) and after (gray) TBS. (Scale bar: 40 pA, 20 ms.) (D) Time course of normalized IPSC changes in cells transiently perfused with gabazine (n = 7). Histogram shows average IPSC values (34.2 ± 2.6 pA vs. 62.1 ± 4.7 pA n = 7, **P < 0.01). (E) Time course of IPSCs first peak amplitude in a GrC perfused with d-APV (horizontal bar). Traces in the Inset show average responses before (black) and after (gray) TBS. (Scale bar: 40 pA, 20 ms.) (F) Time course of normalized IPSCs changes (n = 5 cells) obtained from cells perfused with d-APV. Histogram shows average IPSC values (56.4 ± 3.5 pA vs. 54.8 ± 3.9 pA n = 5, P > 0.6).

Fig. S3.

Properties of LTP_GABA. (A, Top) Traces showing sIPSCs before and after TBS. (Scale bar: 10 pA, 500 ms.) sIPSC peak 18.6 ± 0.4 pA, n = 900 events vs. 23.5 ± 0.3 pA, n = 900 events; P < 0.01; frequency +0.5 ± 3.5%, n = 18; P > 0.6. (Middle) Traces show a comparison between monosynaptic IPSCs evoked by single stimuli before and after TBS (20 traces superimposed). (Bottom) Normalized traces obtained from the averaging of Upper traces (black pre-TBS, gray post-TBS). Note the longer decay of post-TBS trace. Histogram summarizes average values of IPSCs decay (**P < 0.01). (B) Relationship between EPSC (●) and IPSC (○) peak amplitude with the amount of LTP_GABA. In three cells (EPSCs, light gray and IPSCs, dark gray), no significant changes were observed in response to TBS. (C) Histograms show changes of the steady state (Left) and of the peak-to-peak noise (Right) of the holding current recorded at 0 mV in response to TBS (n = 18 cells). (D) Time course of IPSCs peak amplitude obtained from cells in which the stimulation did not evoke glutamatergic responses (n = 6).

Fig. S4.

Properties of LTP_GABA. (A) Time course of normalized PPR obtained from cells transiently perfused with gabazine (n = 7). Histogram in the Inset shows average PPR before (black) and after (white) TBS (0.81 ± 0.04 vs. 0.69 ± 0.03 n = 7; **P < 0.01). (B) Time course of normalized CV in cells transiently perfused with gabazine (n = 7). Histogram in the Inset shows average values of CV before and after TBS (0.16 ± 0.04 vs. 0.08 ± 0.03 n = 7; **P < 0.01). (C) Time course of normalized PPR (n = 7) obtained when slices were incubated and perfused with l-NAME (10.8 ± 1.3%; n = 7, P < 0.01). Histogram in the Inset shows average PPR before (black) and after (white) TBS (0.7 ± 0.04 vs. 0.78 ± 0.03 n = 7; **P < 0.01). (D) Time course of normalized CV (n = 7) obtained when slices were incubated and perfused with l-NAME (15.5 ± 3.4%; n = 7; P < 0.01). Histogram in the Inset shows average CV before (black) and after (white) TBS (0.17 ± 0.02 vs. 0.2 ± 0.03 n = 7; **P < 0.01). (E) Histogram of PPR changes measured in different experimental conditions. (F) Histogram of CV changes measured in different experimental conditions (**P < 0.01).

The postsynaptic induction of LTP_GABA through a presynaptic maintenance mechanism requires the diffusion of a retrograde messenger potentiating GABA release. We investigated NO as a possible mediator of LTP_GABA. Recordings performed in slices preincubated and continuously perfused with the nitric oxide synthase (NOS) inhibitor, N^ω-nitro-l-arginine methyl ester hydrochloride (l-NAME) showed a complete block of LTP_GABA (Fig. 3A). Unexpectedly, IPSCs showed a significant and persistent peak amplitude decrease (-21.3 ± 4.1%; n = 7, P < 0.05) (Fig. 3B) along with an increase in the PPR and CV (Fig. S4 C and D). We termed this persistent depression in the efficacy of the GoC–GrC synapse LTD_GABA. To confirm the postsynaptic origin of LTP_GABA induction, we performed recordings with a high Ca²⁺ buffer concentration in the intracellular solution (10 mM BAPTA). LTP_GABA was blocked by 1,2-bis(2-aminophenoxy)ethane-N,N,N′,N′-tetraacetic acid (BAPTA) and IPSCs indeed showed depression (Fig. 3B) together with a persistent increase of PPR and CV (Fig. S4 E and F). These data confirmed that NO originated in the postsynaptic compartment following calcium entry through NMDA_RS. The guanylate cyclase (GC) and the subsequent cGMP activation are the main targets of NO. We therefore investigated the NO pathway by inhibiting GC with 5 µM 1H-[1,2,4]Oxadiazolo[4,3-a]quinoxalin-1-one (ODQ). Recordings performed on slices preincubated and continuously perfused with ODQ still showed a depression of IPSCs (Fig. 3C) and an increase of PPR and CV (Fig. S4 E and F). Accordingly, the perfusion of the membrane permeant cyclic GMP (cGMP) analog 8p-CPT-cGMP (25 µM) starting 10 min before TBS and in the presence of d-APV throughout recordings increased IPSC amplitude (Fig. 3C) and decreased PPR and CV (Fig. S4 E and F). These findings confirmed our hypothesis on the role of NO signaling in the expression of LTP_GABA.

Fig. 3.

Properties of LTP_GABA and LTD_GABA. (A) Time course of IPSCs first peak amplitude in a GrC incubated and perfused with l-NAME (horizontal bar). Traces in the Inset show average responses before (black) and after (gray) TBS. (Scale bar: 40 pA, 20 ms.) (B) Time course of normalized IPSC changes obtained from cells incubated and perfused with l-NAME (●) and cells recorded with high intracellular BAPTA (○). Histogram shows average IPSC changes (● −21.9 ± 3.9%, n = 7, **P < 0.01; ○ −37.1 ± 4.3%, n = 4, **P < 0.01). (C) Time course of normalized IPSC changes obtained from cells incubated (more than 2 h) and perfused with ODQ (●) and cells incubated and perfused with d-APV and transiently perfused with 8p-CPT-cGMP (○). Histogram shows average IPSC changes (● −24.7 ± 2.2%, n = 4, **P < 0.01; ○ +41.5 ± 13.4%, n = 4, **P < 0.01). (D) Time course of normalized IPSC changes obtained from cells perfused with CGP and l-NAME (●) and cells perfused with only CGP (○). Histogram shows average IPSC changes (● −44.1 ± 10.4%, n = 5, P < 0.01; ○ +78.4 ± 3.1%, n = 5, P < 0.01). (E) Time course of normalized IPSC changes obtained from cells with intracellular MK-801 (●) and cells receiving TBS at −70 mV (○). Histogram shows average IPSC changes (● −25.5 ± 6.1%, n = 5, **P < 0.01; ○ −35.9 ± 10.3%, n = 6, **P < 0.01) (F) Time course of normalized IPSC changes obtained with TBS at −70 mV and in the presence of d-APV (●) and cells with extracellular MK-801 and TBS at −40 mV (○). Histogram shows average IPSC changes (● −1.3 ± 4.2%, n = 5, P > 0.7; ○ −1.1 ± 3.7%, n = 4, P > 0.6).

We then investigated the mechanisms leading to LTD_GABA. The activation of metabotropic GABA_B receptors in response to a 100-Hz train of GoC action potentials is known to transiently depress GABA release. Recordings performed in the presence of the GABA_B receptor antagonist (2S)-3-[[(1S)-1-(3,4-Dichlorophenyl)ethyl]amino-2-hydro xypropyl](phenylmethyl)phosphinic acid hydrochloride (CGP) together with l-NAME showed that, in response to a TBS at −40 mV, IPSCs were persistently depressed (Fig. 3D), whereas PPR and CV were increased (Fig. S4 E and F). Furthermore, the presence of the sole CGP in the bath solution still showed IPSC depression (Fig. 3D) along with PPR and CV increase (Fig. S4 E and F). These data indicated that GABA_B receptors were not involved in the induction of both forms of plasticity. The evidence that, in the presence of d-APV, IPSCs were unchanged (Fig. 2 E and F) led us to hypothesize that NMDA_Rs could be involved in the induction of LTD_GABA. Extrasynaptic NMDA_Rs were in fact shown to mediate some forms of long-term plasticity in central synapses (25). Similarly, we supposed that presynaptic NMDA_Rs could be involved in the induction of LTD_GABA. TBS was therefore delivered while holding GrCs at −70 mV to limit the contribution of postsynaptic NMDA_Rs by virtue of the voltage-dependent Mg²⁺ block. LTP_GABA was indeed converted into LTD_GABA (Fig. 3E), whereas PPR and CV were increased (Fig. S4 E and F). Our hypothesis was confirmed by adding to the intracellular solution the NMDA ion channel blocker MK-801 (500 µM) to inhibit postsynaptic NMDA_R opening. The depression of IPSCs (Fig. 3E) and the increase of PPR and CV (Fig. S4 E and F) demonstrated that postsynaptic NMDA_Rs were essential to induce LTP_GABA, leaving space for a role of presynaptic NMDA_Rs in the induction of LTD_GABA. When TBS was delivered at −70 mV and in the presence of d-APV, IPSC depression was abolished (Fig. 3F). Furthermore, a metabotropic NMDA-dependent pathway has been shown to mediate hippocampal LTP (26). Thus, 100 µM (5S,10R)-(+)-5-methyl-10,11-dihydro-5H-dibenzo[a,d]cyclohepten-5,10-imine maleate (MK-801) was added to the extracellular solution to block ion influx through NMDA channels while delivering TBS at −40 mV. Recordings performed on slices preincubated and continuously perfused with MK-801 showed that NMDA channel opening is essential to trigger both forms of plasticity (Fig. 3F). The effect of LTP_GABA and LTD_GABA on membrane potentials was evaluated by measuring inhibitory postsynaptic potential (IPSP) changes (Fig. S5 A and B). Because these variations may derive from a shift of the Cl⁻ reversal potential (27), IPSC amplitudes were measured while holding GrCs at a different membrane potential. No significant changes were observed between values obtained before and after TBS in both cases (Fig. S5B). This evidence confirmed that changes in IPSPs depended on potentiation and depression of GABA release.

Fig. S5.

Properties of LTP_GABA. (A) IPSPs generated by a single pulse delivered to the GoC axon in the presence of NBQX. GrCs were injected with a constant depolarizing current to reach the membrane potential of −55 mV (−53.9 ± 1.1 mV, average injected current 4.5 ± 1.4 pA, n = 10). This value was chosen to increase the driving force of Cl⁻ based GABAergic currents while avoiding the activation of NMDA currents. (Upper) Average IPSPs obtained before (black) and after (gray) the induction of LTP_GABA. Histogram shows the effect of LTP_GABA on peak amplitude (46.1 ± 6.4%, **P < 0.01, n = 5), time to peak (30.2 ± 4.2%, n = 5, **P < 0.01), and total depolarization (28.3 ± 3.4%, **P < 0.01, n = 5). (Lower) Average IPSPs obtained before (black) and after (gray) the induction of LTD_GABA. Histogram shows the effect of LTD_GABA on peak amplitude (−24.5 ± 2.8%, **P < 0.01, n = 5), time to peak (−10.6 ± 2.4%, n = 5, **P < 0.01), and total depolarization (12.7 ± 1.8%, **P < 0.01, n = 5). (B) Correlation of holding membrane potential with eIPSC peak amplitude normalized to maximum currents. LTP_GABA and LTP_GABA did not affect Cl⁻ reversal potential. (CTR, black line, −71.4 ± 0.6 mV; LTP_GABA dashed line, −71.3 ± 0.5 mV, n = 4; and LTD_GABA gray line, −71.9 ± 0.8 mV n = 4). (C) Time course of spike-related parameters as in Fig. 1 obtained in the presence of d-APV to block GABAergic plasticity. (D) Time course of EPSP-related parameters as in Fig. 1 obtained in the presence of d-APV to block GABAergic plasticity.

Finally, GrC firing and subthreshold activity changes in response to TBS (Fig. 1) were evaluated when both LTP and LTD were abolished by d-APV. Both spike- and EPSP-related parameters were unaffected by TBS (Fig. S5 C and D), confirming that alterations on GrC response properties initially observed (Fig. 1) originated at least in part from GABAergic bidirectional plasticity.

In the classic view, NMDA_Rs act postsynaptically to control excitatory neurotransmission. In the granular layer, dendritic NMDA_Rs modulate glutamatergic neurotransmission from mf to both GrCs and GoCs. Our evidence on the induction of LTD_GABA implies the presence of presynaptic NMDA_Rs on GoC axonal terminals, which were never observed. We thus performed triple immunostaining against NR1 NMDA_R subunit (Fig. 4 C, G, K, and O), synaptophysin (Fig. 4 B, F, J, and N), and glutamic acid decarboxylase (GAD) (Fig. 4 A, E, I, and M) to identify NMDA_Rs, presynaptic boutons, and GoCs, respectively. As expected, a few scattered immunoreactive GoCs were found in the granular layer, whereas synaptophysin immunoreactive synapses were found concentrated in the glomeruli (Fig. 4 D, H, L, and P). As previously reported (28), NR1 immunoreactivity was scattered across the granular layer and richly decorated the large synaptophysin immunoreactive spots of the glomeruli. To assess the presence of presynaptic NMDA_Rs on GABAergic boutons of the granular layer, we singled out GAD/synaptophysin-stained structures of about 1 μm as putative GABAergic boutons and verified whether they were also stained for NR1 immunoreactivity. The analysis of 10 confocal images (70 × 70 µm, 1 µm thick) showed that roughly 80% (151/185) of putative GABAergic boutons express NR1 immunoreactivity (colocalization was considered positive when a NR1 immunoreactive spot was contained within a putative GABAergic bouton and covered ≥30% of its area). We also found several instances of ∼1-µm-thick structures positive for GAD and NR1 immunoreactivity but negative for synaptophysin that were interpreted as thin GABAergic dendritic processes of the GoCs or GABAergic axons of the Purkinje cells expressing postsynaptic NMDA_Rs. As a negative control we performed triple immunostaining against NR1, synaptophysin, and gephyrin (a protein associated with GABA_A receptors), and found no instances of triple colocalization of these markers in the granular layer (Fig. S6).

Fig. 4.

Confocal microscopy images of brain cryosections labeled with mouse anti-GAD (magenta), guinea pig anti-synaptophysin (red), and rabbit anti-NR1 (green), of the granular cell layer of cerebellum. (A–D) A GAD⁺ GoC (magenta, indicated in A) is represented, showing membrane staining for NR1 (green, C). (E–H) Magnifications of the GAD⁺ cell represented in A showing single (red, green, or magenta pixels for synaptophysin, NR1, and GAD staining, respectively) or multiple labelings marked with different labels: arrowhead for triple colocalization of GAD, synaptophysin, and NR1 signals; double arrowhead for colocalization of GAD and synaptophysin signals; and arrow for colocalization of synaptophysin and NR1 signals. (I–P) Taken from granular layer neuropil, different spots with high degree of labeling colocalization are indicated using the same label code used in E–H. [Scale bars: 50 μm (A–D) and 15 μm (E–P).]

Fig. S6.

Confocal microscopy image of brain cryosections labeled with mouse anti-gephyrin (red), guinea pig anti-synaptophysin (blue), and rabbit anti-NR1 (green), of the granular layer of cerebellum. (A) The image reports a cerebellar glomerulus (synaptophysin⁺, bright blue staining) on the Right side and glomerular cell bodies (green background staining) on the Left side. In all these structures, NR1 immunoreactivity (bright green spots) and gephyrin immunoreactivity (bright red spots) never colocalize. (B) Arrows indicate synaptophysin/NR1⁺⁺ spots in a glomerulus. No gephyrin⁺ structure is located close to synaptophysin/NR1⁺⁺ spots. A semiquantitative evaluation on 10 randomly chosen glomerular structures could not find any instance of colocalization (0/82) between synaptophysin/NR1⁺⁺ spots (i.e., presynaptic NR1⁺ structures) and gephyrin⁺ spots. (Scale bars: 15 μm.)

Discussion

We have shown a form of GABAergic heterosynaptic plasticity requiring the opening of pre- and postsynaptic NMDA_Rs. We reached this conclusion by examining the induction mechanisms. First, the presence in the extracellular solution of d-APV or MK-801 prevented both forms of plasticity. Second, LTP_GABA could not be induced when postsynaptic NMDA_Rs were blocked by GrC hyperpolarization, intracellular MK-801, or high concentration of Ca²⁺ buffer. Third, none of these conditions prevented the induction of LTD_GABA. Fourth, LTD_GABA was unmasked when presynaptic expression of LTP_GABA was inhibited by the block of NO synthesis and diffusion. We showed that this form of plasticity governs first spike precision and the number of spikes emitted by GrCs. The interplay between glutamatergic and GABAergic synapses dynamically controls GrC output patterns.

Bidirectional plasticity rules depending on glutamate receptors have been identified in GABAergic synapses of neonatal hippocampus (29) and in the basolateral amygdala (30). In these brain regions, the sign of plasticity depended on the ontogenetic developmental stage or, alternatively, on the activity of specific kainate receptor subtypes. Here, we describe a mechanism in which learning rules rather depended on the activation state of neurons involved in the induction processes [state-dependent plasticity, (SDP)]. The mechanisms underlying the sign of SDP are mediated by the activation of presynaptic and postsynaptic NMDA_Rs, whose gating is influenced by the corresponding neuron. As in other areas, in the cerebellar granular layer, the excitatory/inhibitory ratio (E/I) originates from the reciprocal interaction of excitatory and inhibitory afferences (21). Similarly to what was recently observed in the hippocampus (31) and in the auditory cortex (32), cerebellar inhibitory SDP compensates unbalanced E/I, implementing a negative feedback control of neuronal activity. LTP_GABA, predominantly observed in poorly inhibited neurons (Fig. 1 A and G), limits hyperexcited states arising from the potentiation of the mf–GrC synapse (21). Conversely, in poorly excitable GrCs (Fig. 1 D and J), the initial disinhibition triggered by high-frequency activity and mediated by presynaptic metabotrobic glutamate receptors and GABA_BR (33) can be reinforced by the expression of LTD_GABA. Central to these processes is the engagement of specific NMDA signaling. The role of GoCs in SDP induction is also important because (i) GoCs are spontaneously active at low frequency (23) and (ii) mf high-frequency discharge induces a firing reset in the refractory period (23). It is plausible that TBS is unable to induce a persistent depolarization capable of unblocking NMDA_Rs located on presynaptic GoC terminals. Nevertheless, among NMDA subunits expressed by GoCs, the low conductance NR1/NR2D complex (i) is preferentially located extrasynaptically and poorly contributes to EPSCs, (ii) is insensitive to voltage-dependent Mg²⁺ block, and (iii) shows slow deactivation rates (34). LTD_GABA could thus be subtly sensitive to GoC activity, and glutamate diffusion following TBS tends to depress GABAergic activity. This depression can be indeed converted into potentiation by mechanisms triggered by NO diffusion. The GoC-mediated inhibition is a fundamental player in controlling the temporal coding in cerebellar circuitry. GoC inhibition was in fact shown to operate as gain controller of GrC excitation and regulator of the temporal fidelity of mf–GrC neurotransmission (11, 15, 17, 35).

Recent findings also showed that when sensory information is conveyed to GrCs, the temporal precision of excitatory neurotransmission is weakened by GoC synaptic inhibition in favor of an increased response reproducibility across the GrC population (11). Our results demonstrate that the excess of excitation generated by LTP_glu in poorly inhibited GrCs is compensated by LTP_GABA improving time windowing (4). As a whole, the interplay between LTP_glu and LTP_GABA leads to an enhanced first spike precision and to a reduction of the total amount of emitted spikes. The result is thus a more efficient and reliable neurotransmission. A reduced spike discharging poorly correlated with synaptic inputs could in fact lower the noise in the transmission of information (36). Conversely, LTD_GABA occurring when inhibition excessively dampens GrC membrane potential helps GrCs to approach firing regime and widening the time window for spike transmission. Furthermore, LTD_GABA could act on the total inhibitory strength by reducing spillover inhibition activated by parallel circuits (37). As a whole, heterosynaptic SDP could increase the overall capability of the system to modulate signal transfer (36, 38). The potentiation of GoC–GrC would in fact maximize the temporal precision of mf–GrC neurotransmission by narrowing time windowing, whereas the depression of GABAergic efficacy following LTD_GABA could facilitate temporal summation of excitatory inputs during repetitive neurotransmission.

Finally, GrCs were recently shown to integrate inputs coming from different sensory pathways and generating distinct firing patterns representative of multimodal input combinations (12). Long-term plastic mechanisms separately occurring in single dendrites could contribute to potentiate input specificity and classification. The GoC–GrC connectivity has inspired the hypothesis that GoC inhibition also controls the spatial patterns of granule cell activity. Bidirectional GABAergic plasticity could provide a spatial modulation of granular layer activity, which in the granular layer is organized in center-surround structures (21, 39, 40). The lipophilic nature of NO, by facilitating diffusion across cell membranes, would generate a wide sphere of influence (41). NO diffusion from cores of activation could potentiate GABAergic inhibition in surrounding regions thus sharpening center-surround structures (21, 39). In particular, the correlation between EPSCs and the amount of LTP_GABA (Fig. S3B) suggests that GrCs receiving stronger excitatory inputs produce a larger amount of NO to also be spread to nearby glomeruli, thus activating “silent” synapses (42). Conversely, by virtue of its inability to cross cell membranes, glutamate remains confined inside glomeruli. We can speculate that the sign of inhibitory plasticity depends on the balance between the inactivation rate of presynaptic NMDA_Rs and NO (16). Experimental findings and theoretical predictions showed that the sphere of influence of NO from point sources has a limited radius (less than 100 µm) (41) and lasts for less than 1 s (16). These values are compatible with the size of the recently observed center-surround structures in the granular layer (21, 39, 40, 43). We can conclude that in response to heterosynaptic bidirectional SDP, GoC-mediated inhibition proportionally scales with excitatory input implementing a dynamic regulation of GrC responses (37). This form of heterosynaptic SDP provides an additional mechanism for the modulation of GrC response magnitude. A major contribution of this study is that we have demonstrated the existence of a form of bidirectional GABAergic plasticity dependent on the activity of neighboring glutamatergic synapses.

Materials and Methods

Experiments were performed by using Sprague-Dawley rats at postnatal days P17–P24 (internal breeding; Charles River Laboratories). All experiments were conducted in accordance with international guidelines from the European Community Council Directive 86/609/EEC on the ethical use of animals. Parasagittal cerebellar slices were obtained as described in ref. 40. Experiments were performed at 32 °C using the standard patch-clamp recording technique. Drugs were purchased from Tocris Bioscience and Sigma-Aldrich. All data are expressed as mean ± SEM.

Electrophysiological Recordings.

Whole-cell recordings from GrCs were obtained with patch-clamp technique by using an Axopatch 200B amplifier (Molecular Devices). Patch pipettes were filled with the following solution (in millimoles): 126 K-gluconate, 8 NaCl, 15 glucose, 5 Hepes, 1 MgSO₄, 0.1 BAPTA-4K, 0.05 BAPTA-Ca²⁺, 3 ATP, 100 µM GTP; pH adjusted to 7.2 with KOH. Simultaneous IPSCs and EPSCs were elicited by stimulating GoC axons and mf terminals through a bipolar tungsten electrode positioned in the granular layer close to the recorded GrC (Fig. S2A) (Clark Instruments; stimulation intensity ± 5–15 V, 100 μs). EPSP- and ESPS-spike complexes were elicited by positioning the bipolar electrode across the mf bundle (Fig. S1A). Paired pulse stimulation was generated at a frequency of 20 Hz and acquisition was repeated at 0.1-Hz repetition. The PPR was calculated as the ratio between the second and the first peak amplitude. The CV of IPSCs and EPSCs was calculated as the ratio between the SD and the average of 10 consecutive responses. Current relaxation induced by a 10-mV step from the holding potential was analyzed to ensure recording stability. The tonic inhibition was estimated by measuring the holding current recorded in voltage clamp configuration at 0 mV and the average noise peak to peak of the same holding current. GrCs showed average membrane capacitance (1.9 ± 0.4 pF, n = 38), input resistance (1.9 ± 0.1 GΩ, n = 38), and series resistance (17.3 ± 0.5 MΩ, n = 38). The induction of plasticity was estimated by evaluating changes between 10 min before TBS and a period of at least 20 min after TBS. Cells showing unstable responses or changes in series resistance larger than 20% throughout recordings were discarded. In current clamp configuration, GrCs had an average resting potential of −64.3 ± 0.7 mV (n = 20 cells), whereas EPSP potentials were generated starting from a membrane potential between −55 and −65 mV (−64.3 ± 0.7 mV, n = 20 cells). Spike- and EPSP-related parameters were calculated as follows: (i) first spike delay, (ii) first spike jitter (SD of the first spike latency), (iii) P firing (the number of times a GrC elicited at least one spike over 10 repetitions), (iv) spike doublets (number of times a GrC elicited two spikes over 10 repetitions), (v) EPSP time to peak, (vi) EPSP peak amplitude, (vii) EPSP total depolarization (integral of membrane depolarization between the onset and 50 ms from stimulus onset (44), and (viii) residual depolarization (difference between the membrane potential before stimulus and 50-ms poststimulus).

Immunofluorescence Analysis.

For histological processing, rats were prepared as in ref. 45. Briefly, rats were anesthetized and perfused with 4% (wt/vol) paraformaldehyde. Brains were then dissected out, postfixed, and frozen. Coronal 50-µm-thick section series were cut at a cryotome. Brain sections were processed for multiple immunofluorescence histochemistry. Incubation with anti-NR1 (rabbit, a kind gift of Cecilia Gotti, CNR Neuroscience Institute, Milan), anti-GAD (mouse monoclonal, Millipore), anti-synaptophysin, (guinea pig, a kind gift of Andreas Grabrucker, University of Ulm, Germany), anti-gephyrin (mouse, Synaptic Systems) diluted in 0.3% Triton X-100, 1% normal serum, and PBS 1× was performed overnight. Incubation with secondary antibodies (Alexa Fluor anti-rabbit 488, Alexa Fluor anti-mouse, and 546 Alexa Fluor anti-guinea pig 633) was carried out for 9 min at room temperature. Finally, brain sections were placed on gelatinized glass slides, dried and, after incubation with DAPI, mounted for confocal microscopy analysis (Leica, VTZ2000).