Depolarization-induced potentiation of inhibition

We studied inhibitory synaptic plasticity during early cerebellar development using whole-cell recording from Purkinje cells at postnatal days (P) 6–8. To induce plasticity, we applied a single train of depolarizing voltage steps (0.8-s total duration of depolarization; see Methods) to Purkinje cells voltage clamped at -70 mV. This induces large dendritic Ca²⁺ transients similar to those evoked by climbing fiber stimulation^{19, 20, 21} and is a key prerequisite for the induction of various forms of inhibitory synaptic plasticity^{19, 22, 23, 24}. To assess the sensitivity of postsynaptic GABA_A receptors and the extent of spontaneous presynaptic GABA release, we measured both the amplitudes and frequencies of miniature inhibitory postsynaptic currents (mIPSCs) in the presence of the Na⁺ channel blocker tetrodotoxin (TTX; 500 nM) and the AMPA receptor antagonist 2,3-dioxo-6-nitro-1,2,3,4-tetrahydrobenzo(f)quinoxaline-7-sulfonamide (NBQX; 10 M). Direct Purkinje cell depolarization potentiated the mean mIPSC amplitude to 154 14% (control mIPSC amplitude measured before depolarization = 100%), at 5 min after stimulus (n = 7), in accord with the development of rebound potentiation^{1, 19} (Fig. 1a–c). As expected, after cessation of the stimulus the mean mIPSC frequency was reduced by the induction of DSI^{3, 4, 7} (63 10%, control frequency 100%; n = 7; Fig. 1a,b,d). However, as DSI began to dissipate after the stimulus (t = 1 min), we observed a robust increase in the mean mIPSC frequency (159 22%, t = 6 min; n = 7), which returned to baseline 8 min after the stimulus (Fig. 1a,b,d). The increased mIPSC frequency appeared as a new plasticity phenomenon at this synapse and was subsequently defined as a depolarization-induced potentiation of inhibition.

Figure 1: Postsynaptic depolarization induces DSI, rebound potentiation and DPI in P6–8 cerebellar Purkinje cells.

(a) Continuous recordings of mIPSCs in control aCSF and after a train of conditioning stimuli (Dep, +70 mV, 100 ms, 0.5 Hz, eight stimuli). (b) High-resolution mIPSC recordings (selected from a) during control (before stimulation), during peak DSI and after onset of DPI, which occur with rebound potentiation (RP). (c,d) Normalized mean mIPSC amplitudes (c) and frequencies (d; n = 7). Dotted line and arrows here and in other figures represent the baseline levels (100%) before the stimulus and the point of stimulus induction (t = 0 min), respectively. Throughout, all data points represent means s.e.m.

Full size image (43 KB)

During the first 2 weeks of postnatal development, the cerebellum undergoes considerable maturation and synaptogenesis^{25, 26}. We assessed whether DPI was maintained during the latter stages of this period using P11–14 slices. Notably, the depolarizing protocol used in the P6–8 slices did not induce DPI at P11–14 (98 3%, t = 5 min; n = 6; Fig. 2a). However, we observed rebound potentiation in each cell (Fig. 2a, inset), indicating that rebound potentiation and DPI are discrete forms of inhibitory synaptic plasticity, probably involving separate signaling pathways. The failure to induce DPI may result from a reduced availability of intracellular Ca²⁺ owing to increased expression of Ca²⁺ binding proteins in older Purkinje cells. This explanation would also account for the lower magnitude of DSI in P11–14 (87 2%, n = 6; Fig. 2a) as compared to P6–8 Purkinje cells (Fig. 1d). Any increased Ca²⁺ sequestration at P11–14 was obviated by promoting Ca²⁺ entry with an increased depolarization of the Purkinje cell from 0.8 s to 2 s. This longer stimulus evoked increased DSI (63 4%) and a small DPI (120 6%, t = 5 min; n = 6) that lasted for 10 min (Fig. 2b,d). Increasing the stimulus duration still further, to 5 s, induced both robust DSI (61 7%) and increased DPI (155 15%, t = 5 min; n = 9), the latter persisting for more than 15 min (Fig. 2c,d). Concurrently with the modulation of mIPSC frequency, lengthening the stimulus duration also increased the magnitude of rebound potentiation from 144 4% (0.8 s) to 156 8% (2 s) and 170 11% (5 s), respectively, at t = 5 min (n = 6, 6 and 9; Fig. 2a–c, insets). These results indicated that cerebellar maturation enforces a strict control over the induction of DPI.

Figure 2: DPI in P11–14 cerebellar Purkinje cells.

(a–c) Normalized mean mIPSC frequencies (black circles) before and after the following conditioning stimuli: 100-ms duration, 0.5 Hz, repeated eight times (n = 6; a); 2-s duration, single pulse (n = 6; b); or 5-s duration, single pulse (n = 9; c). Open circles in b and c represent data taken from a for comparison. Insets in a–c show superimposed mean mIPSCs from 50 consecutive events recorded in control (Ctrl) and 5 min after stimulus cessation during rebound potentiation. (d) Bar graph of the relative mIPSC frequency increase (DPI, %) over controls (set to 0%) measured at 4–5 min after induction by 0.8, 2 and 5 s conditioning stimuli. ^*P < 0.05, ^**P < 0.01.

Full size image (28 KB)

We next investigated whether the induction of DPI could also affect action potential–dependent inhibitory synaptic transmission. In each P11–14 Purkinje cell, applying a depolarizing protocol (5 s) in the absence of Na⁺ channel blockers induced robust DSI, rebound potentiation and DPI, similar in their magnitudes to those observed with action potential–independent GABA release (Table 1). Taken overall, DPI affects both action potential–dependent and –independent forms of inhibitory neurotransmission, implying that similar signaling pathways are involved.

Table 1: DPI induction affects both action potential–dependent and –independent GABA release

Full table

Climbing fiber stimulation induces DPI

To determine whether DPI was physiologically relevant and could be evoked after synaptic activity, we recorded from P11–14 Purkinje cells in artificial cerebrospinal fluid (aCSF) without TTX and NBQX, using a K⁺-based pipette solution, while stimulating distinct climbing fiber synaptic inputs. We stimulated climbing fiber inputs at 2 Hz for 5 s (Fig. 3a) to reproduce their intrinsically low in vivo firing rate^{17, 18} and to match the stimulus duration that induced DPI in P11–14 Purkinje cells (Fig. 2c,d). As determined using current-clamp recording, each stimulated climbing fiber input induced a typical complex spike (Fig. 3b) with a profile characteristic of a large Ca²⁺ influx into the Purkinje cell dendrites^{19, 20, 27, 28}. Furthermore, voltage-clamped EPSCs showed substantial paired-pulse depression, reflecting the high probability of release at this synapse²⁹ (Fig. 3c). Repetitive climbing fiber stimulation of Purkinje cells resulted in the induction of DSI (81 4%) and DPI (140 4%, t = 5 min; n = 5; Fig. 3d,e). Both DSI and DPI were similar in magnitude and duration to those induced by direct Purkinje cell depolarization (5 s) using a Cs⁺-based pipette solution (compare Fig. 2c and Table 1). In all cells examined with climbing fiber stimulation, the mean sIPSC amplitude increased to 150 6% (t = 5 min; n = 5; Fig. 3e, inset). To address whether DPI, after climbing fiber stimulation, could have resulted from the direct 'spillover' of glutamate from climbing fiber axon terminals, we perfused the Ca²⁺ chelator BAPTA (30 mM) into Purkinje cells. This completely blocked the induction of DPI by climbing fiber stimulation, indicating that glutamate spillover cannot be involved in this process (Fig. 3f). Taken together, these data show that identical forms of DPI can be induced either using physiological climbing fiber stimulation or under the more controlled conditions of direct somatic depolarization, thereby increasing synaptic efficacy at the IN-PC synapse for several minutes.

Figure 3: Physiological climbing fiber (CF) stimulation induces DPI in P11–14 Purkinje cells.

(a) Continuous current-clamp recording of a series of complex spikes generated by repetitive CF stimulation (2 Hz, 5 s). (b) Higher time resolution of a typical complex spike. (c) Typical paired-pulse EPSCs (50-ms interval) evoked by CF stimulation. (d) Spontaneous IPSCs, recorded with a K⁺-based pipette solution, in control aCSF (Ctrl) and after CF stimulation (DPI; 2 Hz, 5 s). (e) Normalized mean sIPSC frequency before and after CF stimulation (2 Hz, 5 s; n = 5). Inset, superimposed mean sIPSCs (of 50 consecutive events) recorded in control and 5 min after CF stimulation during rebound potentiation. (f) Normalized mean sIPSC frequency before and after CF stimulation (2 Hz, 5 s) in Purkinje cells internally perfused with 30 mM BAPTA (n = 5).

Full size image (32 KB)

Calcium dependence and presynaptic DPI expression

Internal perfusion of Purkinje cells with BAPTA (30 mM) prevented the induction of both DSI and DPI in either P6–8 or P11–14 slices after direct depolarization (n = 5; Fig. 4a,b). This joint dependence on postsynaptic Ca²⁺ may underlie the significant correlation (r² = 0.93; n = 30) between the magnitudes of DSI and DPI in every Purkinje cell examined (Fig. 4c). The increased frequency of mIPSCs and sIPSCs during DPI implied that this is a presynaptic phenomenon, a conclusion supported by the paired-pulse ratios (50-ms interval) determined from the amplitudes of successively evoked IPSCs during DPI. A change in the paired-pulse ratio is presumed to reflect a modification to the probability of transmitter release at that synapse. Similar to previous findings^{16, 30}, evoked IPSCs from P11–14 slices showed paired-pulse depression during control recordings (0.73 0.08; n = 7). By comparison, the induction of DSI was accompanied by an increased paired-pulse ratio (1.11 0.07, t = 1 – 40 s; n = 7), which is in accord with a reduced release probability^{7, 31}. However, during DPI, the paired-pulse ratio was significantly lower than in the control recordings (0.5 0.07, t = 5 min; n = 7; Fig. 4d,e), reflecting an increase in the probability of release at the IN-PC synapse. The presynaptic nature of DPI was further emphasized by examining the coefficient of variation (c.v.) for evoked IPSC amplitudes. After direct Purkinje cell depolarization, the transformed coefficient of variation (1/c.v.²) for evoked IPSCs significantly decreased during DSI and increased during DPI (Fig. 4f; n = 5). The paired-pulse ratios and c.v. values for evoked IPSCs, together with the frequencies of mIPSCs and sIPSCs, strongly implied that DPI is presynaptic in origin.

Figure 4: Calcium-dependent release of a retrograde messenger and the presynaptic origin of DPI.

(a) mIPSC recordings from P6–8 and P11–14 Purkinje cells before and after depolarization (P6–8, 0.8 s; P11–14, 5 s) in the presence of 30 mM intracellular BAPTA. (b) Modulation of mean mIPSC frequencies in P6–8 and P11–14 slices during peak DSI (black bars) and peak DPI (open bars) in the absence (n = 7 and 6, respectively) and presence of 30 mM internal BAPTA (gray bars, n = 5). (c) Relationship between the percentage magnitudes of DSI and DPI, induced by different stimuli, and recorded from the same cells for P6–8 (0.8 s, filled stars; n = 5); P11–14 (0.8 s, open circles; n = 5); P11–14 (5 s, filled triangles; n = 9); P11–14 (climbing fiber (CF), 2 Hz, 5 s; open triangles; n = 5); and P11–14 slices (5 s, sIPSCs; filled circles; n = 6). (d) Examples of paired IPSCs (50-ms interval) recorded in P11–14 Purkinje cells before (Ctrl), immediately after (DSI), and 5 min after application of a 5-s conditioning stimulus (DPI). (e) Changes in paired-pulse ratio during DSI and DPI (n = 7). Data from the same Purkinje cells (gray circles) are connected by straight lines. Black circles, means s.e.m. (f) Bar graph showing changes in the transformed coefficient of variation (1/c.v.²) of evoked IPSC amplitudes during DSI and DPI (n = 5). ^*P < 0.05, ^**P < 0.01.

Full size image (44 KB)

Presynaptic NMDA receptors mediate DPI

The molecules mediating DPI are presumed to act on specific receptors located on presynaptic axon terminals of basket/stellate cells. Previously, both glutamate and endocannabinoids have been proposed as retrograde messengers in the cerebellum^{4, 5, 11, 31, 32}. We therefore examined their putative effects using selective receptor antagonists. As expected, the induction of DSI by direct somatic depolarization in P6–8 and P11–14 Purkinje cells was abolished by the CB1 receptor antagonist SR141716A (1 M; Fig. 5a; n = 5 and 6). However, DPI was unaffected by either SR141716A (1 M; n = 5 and 6) or the metabotropic glutamate receptor (mGluR) antagonist LY341495, applied at a concentration (200 nM) that is selective for presynaptic group II mGluRs³³ (n = 5 and 5; Fig. 5b). In contrast, the involvement of NMDA receptors was strongly and specifically implicated in DPI. Irrespective of whether DPI was induced by direct (5 s) depolarization or by climbing fiber stimulation, the selective antagonist D-APV (50 M) abolished the increased IPSC frequencies in P6–8 (0.8 s, 101 13%, t = 5 min; n = 5) and P11–14 slices (5 s, 101 3% and climbing fiber, 101 7%, t = 5 min; n = 5; Fig. 5c–e). In comparison, the induction of rebound potentiation was not affected by any of these antagonists, indicating that this plasticity phenomenon is discrete from DPI. Overall, this indicates that Purkinje cell depolarization may induce retrograde release of either glutamate or glutamate-like molecules, which then activate presynaptic NMDA receptors to enhance the release of GABA at the IN-PC synapse. If glutamate were released by Purkinje cell depolarization, it would rapidly be sequestered by surrounding excitatory amino acid transporters (EAATs), which might explain the lack of DPI observed in P11–14 Purkinje cells after modest depolarization (0.8 s). To examine this possibility, we isolated DPI by blocking DSI using SR141716A (1 M) and then simultaneously blocked EAATs with THA (200 M). The previously subthreshold stimulation (0.8 s) now resulted in the induction of a robust DPI (125 8%, t = 3 min; n = 5), which was blocked by D-APV (50 M; Fig. 5f). These results indicate that even with modest Purkinje cell depolarization, glutamate is released into the extracellular space before being rapidly sequestered by EAATs. Therefore, increasing the stimulus duration will presumably cause increased levels of released glutamate to transiently saturate the EAATs. This will then facilitate presynaptic NMDA receptor activation and the induction of DPI.

Figure 5: Presynaptic NMDA receptors mediate DPI in cerebellar Purkinje cells.

(a,b) Bar graphs of mean mIPSC frequencies in P6–8 and P11–14 Purkinje cells during DSI (a) and DPI (b; t = 5 min), in the absence (aCSF, n = 7 and 9) or presence of: 200 nM LY341495 (n = 5), 50 M D-APV (n = 5) and 1 M SR141716A (n = 5 and 6). (c–e) Normalized mean mIPSC (c,d) and sIPSC frequencies (e) before and after application of a conditioning stimulus to P6–8 (c) and P11–14 Purkinje cells (d,e) in control aCSF (open circles, n = 7, 9 and 5, respectively) and in the presence of 50 M D-APV (black circles, n = 5, 5 and 5, respectively). (f) Bar graph of the relative mIPSC frequency increase (DPI, %) caused by a subthreshold stimulus (0.8 s) to P11–14 Purkinje cells in the presence of 1 M SR141716A (open bar, n = 4); 1 M SR141716A and 200 M THA (gray bar, n = 5); or 1 M SR141716A, 200 M THA and 50 M D-APV (black bar, n = 5). ^*P < 0.05.

Full size image (52 KB)

The duration of stimulus required to evoke heterosynaptic DPI (and DSI²³) was of the order of several hundred milliseconds; however, whether brief activation of presynaptic NMDA receptors is sufficient to induce DPI was uncertain. Bath application of NMDA for 3–6 min increases mIPSC frequency¹⁶, but such long applications may overestimate the NMDA receptor–mediated effect. We therefore attempted to reproduce DPI by briefly activating presynaptic NMDA receptors through pressure application of NMDA (50 M) to the Purkinje cell dendritic field in P11–14 slices in the presence of 1 mM Mg²⁺ (Fig. 6a,b). This caused an immediate increase in mIPSC frequency (310 33%), which subsided (156 14%, t = 5 min) before slowly returning to control levels (105 5%, t = 13 min; n = 5; Fig. 6c). We discounted any NMDA receptor activation in Purkinje cells during the induction of DPI, because no NMDA-induced current was observed (Fig. 6b), emphasizing the lack of any functional NMDA receptors in these cells^{16, 34}. Our results indicated that brief activation of presynaptic NMDA receptors is sufficient to induce a sustained elevation in spontaneous GABA release at the IN-PC synapse.

Figure 6: Brief activation of presynaptic NMDA receptors enhances IN-PC inhibitory synaptic transmission.

(a) mIPSCs recorded from P11–14 Purkinje cells before, during (blue bar) and after a brief (4-s) focal application of 50 M NMDA. (b) Selected regions of a shown at higher time resolution for the control period (Ctrl), during NMDA application (blue bar) and 5 min after the NMDA pulse. The red dotted line represents the holding current level before NMDA application. (c) Normalized mean mIPSC frequencies before, during and after a 4-s pulse of NMDA (50 M, arrow) to P11–14 Purkinje cells (n = 5). (d) Upper, confocal images showing NR1 subunit (green), GAD_65/67 (red) and synaptophysin (blue) immunoreactivity throughout the soma and bifurcating axon of a cultured basket/stellate cell. Lower, high-power magnification of a section of the basket/stellate cell axon shown in the upper panels (white box) showing the abundance of NR1 subunits colocalized with both GAD_65/67 and synaptophysin. Extensive triple immunocolocalization was confined to the axon varicosities, which are presumed to be functionally mature transmitter release sites. (e) High-power magnification of a section of basket/stellate cell axon showing colocalization of NR2A subunits with both GAD_65/67 and synaptophysin. Scale bars, 10 m in upper panels of d and 5 m in lower panels of d and in e).

Full size image (97 KB)

Although the existence of presynaptic NMDA receptors per se has been reported^{14, 15, 16, 35}, their presence on cerebellar interneuron axon terminals has not been explored. We addressed this by using triple immunolabeling for the NMDA receptor subunits, NR1 and the NR2A–D splice variants, together with synaptophysin and glutamic acid decarboxylase (GAD_65/67) as markers for presynaptic release sites on inhibitory axon terminals. Examination of the basket/stellate cell axons showed evidence of triple colocalization, specifically on axon varicosities where high densities of NR1 (Fig. 6d) and all the NR2 subunits are present (NR2A is shown as the example; Fig. 6e). Our results support the hypothesis that functional presynaptic NMDA receptors, expressed on axon terminals of cerebellar basket/stellate cells, are key components for the induction of DPI.

Presynaptic calcium channels and stores in DPI

We next examined whether the sustained increase in GABA release required NMDA receptor activation alone or relied on voltage-dependent Ca²⁺ channels recruited after depolarization of the axon terminals. Although 100 M Cd²⁺ blocked action potential–evoked GABA release onto Purkinje cells (control, mean sIPSC + mIPSC amplitude, 360 41 pA and frequency 10 1 Hz; +Cd²⁺, 118 5 pA and 3 0.6 Hz), as expected³⁶, it did not affect the sustained increase in mIPSC frequency after NMDA application (147 6%, t = 5 min; n = 5; Fig. 7a), indicating that voltage-dependent Ca²⁺ channels have no role in DPI. Cadmium ions can partially block NMDA receptors³⁷, but this effect was relatively insignificant in our study, with 100 M Cd²⁺ inhibiting 50 M NMDA responses in interneurons by only 11 2% (n = 6).

Figure 7: Presynaptic ryanodine-sensitive Ca²⁺ stores are required for DPI.

(a) Bar graph of mIPSC frequencies during and 5 min after a 4-s pulse of NMDA (50 M) in the absence (black bars) and presence of 100 M Cd²⁺ (gray bars, n = 5). (b) Effect of inhibiting ryanodine-sensitive presynaptic Ca²⁺ stores (ryanodine, 100 M) on the modulation of mIPSC frequency caused by the focal application (arrow) of NMDA (50 M) in P11–14 Purkinje cells (black circles, n = 5). The effects of NMDA on mIPSC frequency in the absence of ryanodine (open circles) are reproduced from Fig. 6c for comparison. (c) Normalized mean mIPSC frequencies recorded in the presence of ryanodine (black bar, 100 M) before and after application of a conditioning stimulus to compensate for the postsynaptic effects of ryanodine. This consisted of a 5-s depolarization concurrent with the rapid photolysis of nitrophenyl EGTA (three 1-ms flashes). Inset, superimposed examples of mean mIPSCs (from 50 consecutive events) recorded in control and 5 min after the compensated conditioning stimuli (rebound potentiation, RP; n = 5). (d) DPI enhances synaptic efficacy at the IN-PC synapse. Comparison of the relative charge transfer measured during 60-s epochs in control, after induction of RP alone, and after induction of RP and DPI at the IN-PC synapse (n = 5–8).

Full size image (32 KB)

Given that presynaptic ryanodine-sensitive Ca²⁺ stores may mediate spontaneous transmitter release^{38, 39} and that residual presynaptic Ca²⁺ is crucial for some forms of short-term plasticity⁴⁰, it was plausible that axonal Ca²⁺ stores might also support DPI. To determine whether presynaptic ryanodine receptors were required, we pressure applied NMDA (50 M) to P11–14 Purkinje cells in the presence of 100 M ryanodine, a concentration sufficient to block ryanodine receptors^{38, 41}. During NMDA application, ryanodine markedly attenuated the mIPSC frequency peak potentiation to 140 10% (n = 5). Notably, the increased mIPSC frequency was not maintained, returning to baseline within 2 min (101 5%, t = 3 min; n = 5; Fig. 7b). Ryanodine also reduced the basal mIPSC amplitude (by 30%) and frequency (by 50%) in unstimulated Purkinje cells, demonstrating that presynaptic ryanodine receptors are required during spontaneous GABA release³⁸. These findings indicated that NMDA receptor–mediated entry of Ca²⁺ into axonal boutons may be insufficient to evoke a sustained increase in GABA release at the IN-PC synapse. The increased release may therefore be dependent on Ca²⁺-induced Ca²⁺ release (CICR) from presynaptic ryanodine-sensitive stores, which amplify the initial rise in bouton Ca²⁺, thus inducing the sustained release component that typifies DPI.

To explore this hypothesis, we depolarized Purkinje cells in the presence of ryanodine (100 M). However, local application of ryanodine blocks both pre- and postsynaptic Ca²⁺ release from intracellular stores. In consequence, we observed reductions in the postsynaptic Ca²⁺-dependent induction of DSI, DPI and rebound potentiation. Therefore, to differentiate the role of the presynaptic Ca²⁺ stores during DPI induction, we applied a depolarizing stimulus (5 s) to P11–14 Purkinje cells in the presence of ryanodine (100 M) while compensating for the blocked postsynaptic ryanodine receptors. This compensation involved dialyzing the Purkinje cell with caged Ca²⁺ retained in the photolabile chelator nitrophenyl-EGTA. Postsynaptic Ca²⁺ was uncaged by rapid flash photolysis in a process designed to mimic the release of Ca²⁺ from intracellular stores. Application of this compensated conditioning stimulus in the presence of 100 M ryanodine resulted in the induction of DSI and a rebound potentiation of similar magnitude to that observed under control conditions (DSI, 56 6%; rebound potentiation, 155 8%, t = 5 min; n = 6; Fig. 7c and inset), indicating that postsynaptic Ca²⁺ dynamics were largely unaffected by ryanodine. However, the same conditioning stimulus did not induce DPI (92 3%, t = 6 min; n = 6; Fig. 7c), indicating that presynaptic ryanodine-sensitive Ca²⁺ stores are likely to have a pivotal role in maintaining the increased GABA release during DPI.

DPI and synaptic charge transfer at IN-PC synapses

To understand the functional consequences of inducing DPI and rebound potentiation, we measured the synaptic charge transfer as an indicator of synaptic efficacy. By inhibiting DPI with D-APV, the increase in postsynaptic GABA_A receptor responsiveness caused by rebound potentiation^{19, 42} potentiated the mean charge transfer by 79 9% (n = 5) for mIPSCs and by 45 9% (n = 5) for sIPSCs, as compared to controls (Fig. 7d). When rebound potentiation and DPI were induced together, the mean synaptic charge transfer increased by 209 32% (n = 8) and 205 36% (n = 6) for mIPSCs and sIPSCS, respectively (Fig. 7d). These values indicate that DPI markedly enhances synaptic charge transfer and overall synaptic efficacy at the IN-PC synapse beyond that resulting from rebound potentiation alone.

DPI relies on Ca²⁺-dependent retrograde glutamate release

The requirement for increased stimulus durations to evoke DPI at later stages of cerebellar development (P11–14) revealed two important facets about this synaptic plasticity: its reliance on intracellular Ca²⁺ and on the extent of glial development. During early Purkinje cell development, Ca²⁺ binding markedly increased from P6 to P15 (ref. 43), which is attributable to an increased expression of calbindin D_28K⁴⁴. Therefore, using similar stimulation protocols in P6–8 and P11–14 Purkinje cells will result in substantially different free Ca²⁺ concentrations that would affect Ca²⁺-dependent synaptic plasticity. Presently, the induction of DSI (and DSE) requires postsynaptic Ca²⁺ to rise to 15 M for half-maximal retrograde inhibition to occur^{23, 24}. Although requiring increments in postsynaptic Ca²⁺ levels for induction, the discrete signaling pathways for DSI^{4, 23, 24, 32} and DPI implies that these two mechanisms are not mutually exclusive in regulating GABA release.

An additional influence on DPI is the development of Bergmann glia, which increasingly encapsulate the Purkinje cell dendritic arbor, forming a physical barrier to the diffusion of transmitter from the synaptic cleft⁴⁵. Bergmann glia express a high concentration of the glutamate transporter GLAST⁴⁶, which rapidly sequesters any released glutamate. This mechanism would limit the concentration of glutamate at presynaptic NMDA receptors. Thus, by increasing the stimulus intensity and facilitating glutamate release, we may transiently saturate the EAATs, allowing presynaptic NMDA receptor activation and consequently DPI induction. Therefore, by inhibiting glutamate uptake, we would expect—as we observed—to evoke a robust DPI after only modest Purkinje cell depolarization.

Presynaptic NMDA receptors

The absence of functional NMDA receptors in Purkinje cells, in conjunction with our observations that DPI is inhibited by D-APV and can be reproduced by applying NMDA, indicate that presynaptic NMDA receptors are involved in the induction of DPI. A presynaptic location for these receptors is suggested by three observations. First, to potentiate GABA transmission in the presence of TTX, the NMDA receptors mediating DPI are likely to be juxtaposed to the site(s) of glutamate release. This is probable because released glutamate will be rapidly sequestered by EAATs on Purkinje cell dendrites and Bergmann glia. Second, we observed a decrease in the paired-pulse ratio and an increase in 1/c.v.², consistent with an increase in the probability of transmitter release at the IN-PC synapse. Third, our immunocytochemical data indicate the presence of both NR1 and NR2 subunits at putative presynaptic release sites on cerebellar basket/stellate cells. Notably, the NMDA receptor activation occurs even in the presence of 1 mM Mg²⁺, without any pre-depolarization to remove the voltage-dependent Mg²⁺ block. There are two possible reasons why presynaptic NMDA receptors might be activated under such conditions. First, NMDA receptors composed of NR1 and NR2C or NR2D subunits are less sensitive to Mg²⁺ blockade than their NR1 and NR2A counterparts and are thus capable of activation at more negative membrane potentials⁴⁷. Second, the high input resistance and small volume of interneuron axon terminals implies that only a very small glutamate-activated current would be necessary to depolarize the terminal and relieve the Mg²⁺ block, allowing the induction of DPI. Therefore, DPI, with its reliance on presynaptic NMDA receptors, is a mechanism that can enhance inhibitory transmitter release even without the need for coincident presynaptic action potential activity.

Role of presynaptic Ca²⁺ stores in synaptic plasticity

Depolarization of Purkinje cells induced a strong enhancement of synaptic efficacy that affected both action potential–dependent and –independent forms of inhibitory transmission. Therefore, it is likely that DPI affects the GABA release process at the level of presynaptic Ca²⁺ stores by a CICR mechanism, rather than relying solely on Ca²⁺ influx through the NMDA receptors, given the ability of ryanodine to block DPI completely. The support of DPI by CICR in individual boutons may occur after either an increase in the frequency of spontaneous Ca²⁺ transients, NMDA receptor activation or an initial discrete presynaptic Ca²⁺ signal that is sufficient to trigger a sustained increase in the transmitter release process. Thus, although presynaptic Ca²⁺ stores can contribute to spontaneous³⁸ and evoked⁴⁸ GABA release, they are pivotal to the induction of inhibitory synaptic plasticity at the IN-PC synapse.

DPI enhances GABAergic synaptic efficacy

Our data suggest that Purkinje cells can release glutamate after depolarization, providing a new feedback mechanism to enhance inhibitory synaptic strength over several minutes (see Supplementary Fig. 1 online). An important effect of synaptic inhibition is to provide an increased membrane Cl^- conductance that 'shunts' excitatory currents. This synaptic GABA-activated conductance will depend on the charge transfer across the membrane. In this regard, the functional consequences of DPI were recognized by examining the synaptic charge transfer during DPI and rebound potentiation. The coincident activation of these two independent plasticity mechanisms markedly increased the charge transfer and overall inhibitory synaptic efficacy at the IN-PC synapse for several tens of minutes. Therefore, glutamate retrograde signaling provides a specific, distinctive mechanism whereby Purkinje cells can alter the strength and properties of their synaptic inputs after periods of postsynaptic stimulation. This mechanism of inhibitory synaptic plasticity may have more generalized implications for synaptic inhibition in the CNS.

Slices.

Thin sagittal cerebellar slices (200 m thick) were cut from P6–8 or P11–14 Sprague-Dawley rats using a Leica VT1000S vibroslicer. Slices were cut at 2–4 °C in artificial CSF solution (aCSF) containing (mM) 125 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 25 NaHCO₃, 2 CaCl₂, 1 MgCl₂ and 10 glucose bubbled with 95% O₂/5% CO₂. The slices were then incubated at 35 °C for 45 min before being allowed to cool to room temperature (22–25 °C) before use. Slices were perfused in the recording bath at 1 ml/min with aCSF at 32 °C.

Electrophysiology.

Membrane currents were recorded with a MultiClamp 700A (Axon Instruments) and filtered at 3 kHz (Bessel, 8 pole, -36 dB per octave). During most recordings, Purkinje cells were voltage clamped at -70 mV. For direct Purkinje cell stimulation, patch pipettes (3–5 M) were filled with a solution containing (mM) 150 CsCl, 1.5 MgCl₂, 10 HEPES, 0.1 cesium BAPTA, 2 sodium ATP, 0.4 sodium GTP, 5 QX-314 (to block Na⁺ action potentials) at pH 7.3. The Ca²⁺ buffer BAPTA (30 mM) was included to allow examination of the Ca²⁺ dependence of DPI. For current-clamp experiments or when using climbing fiber stimulation, the pipette solution contained (mM) 105 potassium acetate, 25 KCl, 10 NaCl, 2 MgCl₂, 0.16 CaCl₂, 0.5 EGTA, 20 HEPES, 2 sodium ATP, 0.4 sodium GTP at pH 7.3. For the photolytic Ca²⁺ release experiments, patch electrodes were filled with (mM):150 CsCl, 3 MgCl₂, 10 HEPES, 4 nitrophenyl-EGTA, 2.4 CaCl₂, 2 sodium ATP, 0.4 sodium GTP at pH 7.3. To uncage nitrophenyl-EGTA, cells were irradiated three times, separated by intervals of 2.5 s, with a 1-ms pulse of near-UV light (260–370 nm with a UV-2 filter) from a short-arc xenon flash lamp (JML-C2 system, Rapp-Optoelectronic, Germany).

The induction of DSI, DPI and rebound potentiation was initiated with a conditioning stimulus applied 5 min after the whole-cell configuration was attained. This entailed stepping the membrane potential from -70 mV to 0 mV for a total duration of 0.8 s (eight 100-ms pulses; 0.5 Hz); 2 s (2,000 ms, 1 stimulus); or 5 s (5,000 ms, 1 stimulus). Climbing fiber EPSCs were evoked by placing a stimulating electrode filled with aCSF in the granular layer 20 m from the recorded Purkinje cell and applying a short (0.2-ms) stimulus of 1–10 V in the absence of TTX and NBQX. All drugs were bath applied, except NMDA, which was pressure-applied. Ryanodine, NMDA and cesium BAPTA were obtained from Sigma and LY341495, D-APV, CdCl₂, TTX, THA and QX-314 from Tocris.

Synaptic current analyses.

Spontaneous, miniature and evoked IPSCs were measured and analyzed offline using MiniAnalysis version 6 (Synaptosoft) or Clampfit 8.2 (Axon Instruments). To calculate the mean paired-pulse ratio (PPR) from IPSC amplitudes (amp), 20–30 paired evoked IPSCs were recorded for DPI (8 for DSI). The PPR was then calculated as mean ampIPSC₂/mean ampIPSC₁. This method of calculating the PPR at inhibitory synapses prevented bias that could result from IPSC amplitude fluctuations⁴⁹. The c.v. of the IPSC amplitudes was calculated as c.v. = s.d./m, where m is the mean and s.d. the standard deviation of IPSC amplitudes.

DSI was calculated as the percentage change in IPSC frequency during the initial 20s after stimulus cessation. DPI was measured as the peak percentage change in IPSC frequency after termination of DSI (1–20 min). Rebound potentiation was calculated as the percentage change in IPSC amplitude over 1–20 min after stimulus cessation. The synaptic charge transfer was determined by integrating the total areas of 60-s epochs of IPSCs under control conditions and 5 min after stimulus during the peaks of DPI and rebound potentiation. The charge transfers were normalized to the controls. All values in the text are provided as mean s.e.m. Statistical significance was examined using Student's paired t-test or by a one-way ANOVA with a Bonferroni post-test where P < 0.05 was considered significant.

Immunocytochemistry.

Cerebellar cultures were fixed in phosphate-buffered saline (PBS) with 4% paraformaldehyde before being quenched with 50 mM NH₄Cl in PBS. Cells were permeabilized using 10% fetal calf serum (FCS), 0.5% bovine serum albumin (BSA), 0.1% Triton X in PBS. Cultures were incubated with primary antibody for 30 min, washed in FCS and BSA in PBS solution before being incubated at room temp for 30 min with a single secondary antibody conjugate or an antibody conjugate mixture. Presynaptic NMDA receptor distribution used a primary antibody mixture containing goat anti-GAD_65/67 (Swant), rabbit anti-NR1 or anti-NR2A-D, mouse anti-synaptophysin (Biomeda Corporation) and a secondary conjugate mixture containing anti-goat–TRITC, anti-rabbit–FITC and anti-mouse–Cy5 (Stratech Scientific). Mounted slides of cultures were viewed using an upright Leica DMRE laser-scanning microscope and confocal images were obtained from a Leica TCS SP spectrophotometer.

Note: Supplementary information is available on the Nature Neuroscience website.