Main

Endocannabinoids are lipophilic neuromodulators whose actions are in large part mediated by G-protein-coupled CB1 receptors1,2. Recent studies have described a new signaling system in which postsynaptic neurons release endocannabinoids that diffuse retrogradely to transiently decrease neurotransmitter release3,4,5. This retrograde inhibition of synaptic transmission has been described at synapses throughout the brain, suggesting that endocannabinoids represent a widespread mechanism of synaptic regulation6,7.

However, little is known about the role of endocannabinoid-mediated retrograde inhibition under physiological conditions. Most studies of endocannabinoid signaling have relied on bath application of drugs or large depolarizations of the postsynaptic neuron to evoke endocannabinoid release3,4,5,8,9,10,11,12,13. Although endocannabinoids can strongly inhibit synaptic transmission under such artificial conditions, only weak retrograde inhibition has been produced by synaptic stimulation, even with prolonged high-frequency trains rarely encountered physiologically10.

Thus, several fundamental issues regarding endocannabinoid signaling under physiological conditions are not resolved. First, it is not known whether the endocannabinoid system regulates synaptic transmission during normal ongoing synaptic activity or whether it simply transiently protects neurons during periods of intense synaptic stimulation. The prolonged synaptic activation required to evoke retrograde inhibition suggests that endocannabinoids may suppress synaptic transmission only under such extreme conditions10. Second, how endocannabinoid release is triggered by synaptic activity is not understood, although numerous studies have shown that either depolarization of the postsynaptic cell or activation of postsynaptic metabotropic glutamate receptors can evoke endocannabinoid release3,4,5,8,9,10,11,12,13,14,15. Third, it is not known whether endocannabinoid release globally regulates all of a cell's inputs or whether retrograde inhibition is synapse-specific.

The excitatory synapses between granule cell parallel fibers (PF) and Purkinje cells are well-suited for addressing these three issues. The presynaptic terminals of excitatory inputs onto Purkinje cells (PC) express cannabinoid CB1 receptors16,17, and agonists of these receptors inhibit neurotransmitter release18,19. Furthermore, both activation of postsynaptic metabotropic glutamate receptors and depolarization lead to endocannabinoid-mediated suppression of a Purkinje cell's inputs3,8,10,12. Finally, the anatomical configuration of parallel fiber inputs onto Purkinje cells allows spatially segregated subsets of a Purkinje cell's inputs to be activated experimentally20.

Here we report that brief bursts of synaptic activity produce robust endocannabinoid-mediated retrograde inhibition of the PF-to-PC synapse in rat brain slices at near-physiological temperatures. This inhibition was evoked by rates of synaptic activation within the ranges encountered in vivo21,22. Synaptic activation of either AMPA receptors or mGluR1 receptors triggered endocannabinoid release under these physiological conditions. Finally, we show that synaptically evoked retrograde inhibition is restricted to those inputs activated with a burst. There was no detectable spread of inhibition among synapses onto the same cell or among neighboring cells. Our findings suggest that neurons dynamically modulate their synaptic integration by releasing endocannabinoids to regulate specific inputs in response to brief bursts of synaptic activity.

Results

Bursts trigger endocannabinoid-mediated inhibition

We first tested whether short bursts of parallel fiber activity could elicit retrograde inhibition of the PF-to-PC synapse under physiological conditions. Experiments were conducted in rat cerebellar slices at 34–36 °C. Parallel fibers were stimulated extracellularly, and the resulting synaptic responses were measured with whole-cell current-clamp recordings from Purkinje cells. Spontaneous Purkinje cell activity was suppressed by injecting hyperpolarizing current, and the stimulus intensity was adjusted such that single EPSPs did not trigger action potentials.

Parallel fibers were activated with the following stimulus pattern: 0.5 Hz stimulation preceded and followed a train of ten stimuli at 50 Hz (Fig. 1a). This brief conditioning train reduced the magnitude of subsequent EPSPs. In a representative experiment (Fig. 1b,c), the cell responded to the conditioning train with a series of action potentials (Fig. 1b, top). The EPSP was reduced 2 s after the train, but recovered by 30 s to the level of the pre-train EPSP (Fig. 1b, bottom). To test whether this inhibition was mediated by endocannabinoids, we applied the cannabinoid receptor antagonist AM251. AM251 eliminated the inhibition of the EPSP following the train without significantly altering the response to the conditioning train (compare Figs. 1b and c). Average responses showed a depression of the EPSP amplitude (Fig. 1d, n = 9 control). AM251 eliminated the suppression (Fig. 1d, n = 6 AM251) and unmasked a small enhancement that likely reflects post-tetanic potentiation (PTP), a widely observed form of presynaptic enhancement23. This inhibition of synaptic transmission has a similar duration and pharmacological sensitivity to the inhibition after Purkinje-cell depolarization, previously termed depolarization-induced suppression of excitation (DSE; ref. 3).

Figure 1: Brief trains of parallel fiber stimulation inhibit EPSPs in Purkinje cells by evoking endocannabinoid release.
figure 1

(a) Parallel fibers were stimulated at low frequency except for a brief train of ten stimuli at 50 Hz at time zero (arrow). The response of the Purkinje cell to a train of ten stimuli at 50 Hz is shown for control conditions (b, top) and in the presence of AM251 (c, top). The vertical bars under the traces indicate the stimulus pattern of the train. The parallel-fiber EPSP 2 s before the train (left trace), 2 s following the train (middle trace) and 30 s after the train (right trace) are shown for control conditions (b, bottom) and in the presence of the cannabinoid receptor antagonist AM251 (c, bottom; 5 μM). The EPSP recorded 2 s before the train is superimposed in light gray for comparison. Each trace is the average of 2–3 trials. (d) The normalized peak EPSP is plotted under control conditions (n = 9) and in AM251 (n = 6). The arrow marks the start of the brief train.

Full size image

These results indicate that short trains of presynaptic stimulation elicit endocannabinoid-mediated inhibition of synaptic transmission at the PF-to-PC synapse. However, it was difficult to quantify this inhibition for several reasons. First, the spontaneous activity and active conductances of Purkinje cells obscured changes in synaptic strength measured in current clamp. To reduce these effects, we injected hyperpolarizing currents, but this clearly perturbed the Purkinje cell and may have diminished the extent of retrograde inhibition. Second, to mimic the conditions experienced in vivo, we did not block inhibitory currents, which contributed to variability in the postsynaptic response. Third, additional short-term, use-dependent plasticity such as PTP complicated our measurements.

Monitoring inhibition by measuring presynaptic Ca2+ influx

To overcome the difficulties associated with the current-clamp recordings, we next examined endocannabinoid-mediated retrograde signaling by optically monitoring presynaptic calcium levels in parallel fibers. Under our experimental conditions, the amplitude of presynaptic calcium transients following single stimuli provides a measure of presynaptic Ca2+ influx. Previous studies have demonstrated the feasibility of long-term monitoring of presynaptic Ca2+ influx at PF-to-PC synapses using optical techniques24. Furthermore, endocannabinoids have been shown to inhibit transmission at this synapse by modulating presynaptic calcium channels (ref. 24 and data not shown), suggesting that monitoring Ca2+ influx provides a more direct measure of the effects of endocannabinoids.

We first assessed the sensitivity and stability of this approach by stimulating parallel fibers at 1 Hz before and after a conditioning train of ten stimuli at 50 Hz. As shown for control conditions (Fig. 2a), short bursts (arrows) reliably reduced presynaptic Ca2+ influx. In this example, 12 trials were repeated over the course of an hour, illustrating the reliability and reproducibility of this synaptically evoked retrograde inhibition. We also monitored the presynaptic response to the train of stimuli and found that it was similar throughout the experiment (Fig. 2a, right).

Figure 2: Brief presynaptic bursts lead to reliable and reproducible endocannabinoid-mediated inhibition of presynaptic calcium entry.
figure 2

Presynaptic Ca2+ influx was measured in parallel fibers that were activated at 1 Hz before and after a conditioning train of ten stimuli at 50 Hz (arrows). (a,b) The amplitude of Ca2+ influx is plotted as a function of time for representative experiments consisting of 12 trials separated by 5 min in control conditions (a, left), following application of 1 μM AM251 (b, left) or 5 μM WIN55,212-2 (c, left) as indicated. Representative traces of the presynaptic Ca2+ influx monitored during the burst is shown under control conditions (a, right), following the application of 1 μM AM251 (b, right) or 5 μM WIN55,212-2 (c, right). (d) Summary of averaged Ca2+ influx in control conditions and in the presence of AM251 (1 μM, n = 5; 5 μM, n = 3). (e) Summary of control responses and responses in WIN55,212-2 (5 μM, n = 4).

Full size image

We next assessed the involvement of endocannabinoids in this transient inhibition of presynaptic Ca2+ influx. Application of a cannabinoid antagonist, AM251, prevented the inhibition of Ca2+ influx elicited by the short train under control conditions (Fig. 2b, left). Furthermore, AM251 eliminated the substantial retrograde inhibition without altering the Ca2+ influx during the presynaptic burst (Fig. 2b, right). Application of a cannabinoid receptor agonist, WIN55,212-2, occluded the effect of the short train on Ca2+ influx (Fig. 2c, left). Because WIN55,212-2 activates presynaptic CB1 receptors, WIN55,212-2 also significantly decreased presynaptic Ca2+ influx during the brief conditioning stimulus (Fig. 2c, right). In summary, the suppression of Ca2+ influx induced by the conditioning train was eliminated by AM251, whereas the effect of the brief train was occluded by WIN55,212-2 (Fig. 2d,e).

These experiments confirm that changes in presynaptic Ca2+ influx can be used to quantify endocannabinoid-mediated presynaptic inhibition. This optical approach has several advantages over electrophysiological techniques. First, because the primary sites of endocannabinoid modulation are presynaptic calcium channels, monitoring presynaptic calcium influx provides a more direct measure of modulation than the EPSP. These downstream measures are contaminated by factors including active conductances, inhibition and PTP. Second, by monitoring presynaptic Ca2+ influx, we can study retrograde inhibition without perturbing the Purkinje cells' responses, as measurements of presynaptic Ca2+ influx do not alter the spontaneous or evoked activity of postsynaptic cells. Third, as we discuss below (see Figs. 3 and 4), it is possible to monitor retrograde inhibition even when synaptic currents mediated by ionotropic receptors are blocked, thus allowing us to explore independently the roles of postsynaptic ionotropic and metabotropic glutamate receptors in evoking endocannabinoid release.

Figure 3: The roles of mGluR1s, AMPA receptors and postsynaptic action potentials in the production of retrograde inhibition.
figure 3

Postsynaptic EPSPs and presynaptic Ca2+ influx were measured while parallel fibers were activated before and after bursts of ten stimuli at 50 Hz. (a) The response evoked by the burst is shown on the left under control conditions (top trace), in the presence of the AMPA receptor antagonist NBQX (10 μM; middle trace) and in NBQX and the mGluR1 antagonist CPCCOEt (100 μM; bottom trace). On the right, the parallel-fiber EPSPs 2 s before the train (left traces) and 2 s after the train (right traces) are shown for control conditions (top), in NBQX (middle) and in NBQX and CPCCOEt (bottom). The EPSP recorded 2 s before the train is superimposed in light gray for comparison. Each trace is the average of 2–5 trials. (b) Presynaptic Ca2+ influx into parallel fibers was measured under the same experimental conditions as in a. On the left, Ca2+ influx during the stimulus train is shown under control conditions (top), in NBQX (middle) and in NBQX and CPCCOEt (bottom). On the right, presynaptic Ca2+ influx preceding and following the conditioning stimulus is shown. Each trace represents the average of five responses preceding the burst by 1–5 s and three responses following the burst by 2–4 s from 2–4 trials. (c) The Purkinje cell's response is shown under control conditions (top), after application of CPCCOEt (middle) and after co-application of CPCCOEt and NBQX (bottom). (d) Presynaptic Ca2+ influx into parallel fibers was measured under the same experimental conditions as in c. Four separate experiments are shown in a–d.

Full size image
Figure 4: Both AMPA and metabotropic glutamate receptors can trigger endocannabinoid release under physiological conditions.
figure 4

Presynaptic Ca2+ influx was measured in parallel fibers that were activated at 1 Hz before and after brief bursts of ten stimuli at 50 Hz. (a) The responses in control solution, after application of the AMPA receptor antagonist NBQX (10 μM) and after co-application of NBQX and the mGluR1 antagonist CPCCOEt (100 μM). (b) CPCCOEt was applied first, followed by a co-application of CPCCOEt and NBQX. Data are summarized for six experiments in a and seven experiments in b.

Full size image

Contribution of AMPA and mGluR to endocannabinoid release

The roles of ionotropic and metabotropic glutamate receptors in evoking endocannabinoid-mediated retrograde inhibition following presynaptic bursts are unclear. Parallel fibers express the metabotropic receptor mGluR4 (ref. 25), but a selective mGluR4 antagonist (100 μM CPPG, n = 4) did not affect retrograde inhibition, indicating that these presynaptic receptors are not involved in retrograde signaling. Purkinje cells express both ionotropic (AMPA) and metabotropic (mGluR1 subtype) glutamate receptors at synapses with parallel fibers26,27,28, and both classes of receptors may be involved in retrograde inhibition. Synaptic activation of AMPA receptors could lead to retrograde inhibition by depolarizing the Purkinje cell and opening voltage-gated calcium channels to trigger endocannabinoid release3,12. Previous experiments showing that bath application of an mGluR1 agonist results in robust endocannabinoid-mediated retrograde inhibition10 suggest that mGluR1s may help trigger synaptically evoked endocannabinoid release. Furthermore, blocking mGluR1s reduces the heterosynaptic, endocannabinoid-mediated retrograde inhibition elicited by prolonged parallel fiber stimulation10. In addition, blocking mGluR1s attenuates a transient homosynaptic depression of transmission at parallel fiber synapses following stimulus trains, although the role of endocannabinoids in this process was not explored29. Thus, how AMPA or mGluR1 receptors each contribute to synaptically evoked endocannabinoid release is not understood.

We therefore examined whether synaptic activation of metabotropic glutamate receptors in isolation was sufficient to trigger endocannabinoid release. We first measured EPSP amplitudes (Fig. 1) and tested the effect of blocking AMPA receptors. Under control conditions, the conditioning train depolarized the Purkinje cell and caused it to fire action potentials (Fig. 3a, top). The conditioning train also transiently inhibited the EPSP, as is apparent by comparing the EPSPs before and after the conditioning train. Blocking AMPA receptors with NBQX eliminated the spikes during the conditioning train (Fig. 3a, middle). Blocking AMPA receptors also eliminated the EPSPs, making it impossible to assess any remaining retrograde inhibition (n = 4). The results highlight the limitations of using electrophysiological approaches to assess the contributions of AMPA receptors and metabotropic receptors to endocannabinoid release.

By optically monitoring presynaptic Ca2+ influx instead of measuring postsynaptic EPSP amplitudes, we were able to quantify the extent of presynaptic inhibition triggered by mGluR1s, even when AMPA receptors were completely blocked. In control conditions, the brief train resulted in significant retrograde inhibition of presynaptic Ca2+ influx (Fig. 3b, top). Blocking AMPA receptors had little effect on the presynaptic Ca2+ influx during the conditioning train or the inhibition of Ca2+ influx following the burst (compare the top and middle panels of Fig. 3b). Co-application of NBQX and an mGluR1 antagonist, CPCCOEt, did not alter the presynaptic response to the conditioning train, but eliminated the retrograde inhibition of presynaptic Ca2+ influx (Fig. 3b, bottom). These results indicate that synaptic activation of mGluR1 alone is sufficient to evoke robust endocannabinoid-mediated retrograde inhibition similar to the inhibition seen under control conditions.

We next examined whether synaptic activation of AMPA receptors alone could also trigger endocannabinoid release. Electrophysiological studies showed that CPCCOEt had little effect on the Purkinje cell's response to the conditioning train (Fig. 3c; 92 ± 16% of control, n = 4). However, CPCCOEt greatly reduced the extent of endocannabinoid-mediated retrograde inhibition. In the presence of CPCCOEt, the EPSP often was not reduced, but was slightly enhanced after the conditioning train (Fig. 3c). We could not resolve any additional contribution of AMPA receptors to changes in synaptic strength because co-application of CPCCOEt and NBQX eliminated the postsynaptic response (n = 4).

However, by monitoring presynaptic Ca2+ influx, we found that AMPA receptor activation alone can also evoke endocannabinoid release. When CPCCOEt was applied first, approximately half of the retrograde inhibition of Ca2+ influx remained (Fig. 3d). This remaining inhibition was eliminated by co-application of CPCCOEt and NBQX (Fig. 3d, bottom). Although the contribution of AMPA receptors could not be appreciated by monitoring the EPSP (see Discussion), these results indicate that synaptic activation of AMPA receptors alone can also trigger endocannabinoid release.

A summary of these experiments is shown in Fig. 4. Eliminating the ionotropic component of synaptic transmission by blocking AMPA receptors with NBQX had little effect on the suppression of presynaptic Ca2+ influx elicited by activating parallel fibers with ten stimuli at 50 Hz (Fig. 4a). Co-application of NBQX and CPCCOEt eliminated the inhibition (Fig. 4a). When CPCCOEt was applied first, approximately half of the inhibition seen under control conditions remained, suggesting that considerable endocannabinoid release can be mediated through AMPA receptor activation (Fig. 4b). In the presence of the mGluR1 antagonist, the brief train suppressed presynaptic Ca2+ influx to 81 ± 3% of baseline influx versus 56 ± 3% in control conditions (n = 7).

These results indicate that activation of either AMPA or mGluR1 receptors can trigger retrograde inhibition. When AMPA receptors were completely blocked, thus eliminating the postsynaptic response of the Purkinje cell to single stimuli, activation of mGluR1 was sufficient to evoke near-maximal endocannabinoid release. However, activation of AMPA receptors in isolation evoked ∼50% of the retrograde inhibition seen under control conditions. These experiments highlight a dissociation between the response of the Purkinje cell to the train measured electrophysiologically and the retrograde inhibition evoked by the train. When AMPA receptors were blocked, robust retrograde inhibition remained, even when Purkinje cells no longer fired action potentials during the train (Figs. 3a,b and 4a). Furthermore, the number of action potentials fired during the conditioning train was minimally affected by CPCCOEt, even though the train evoked only half the retrograde inhibition observed under control conditions (Figs. 3c,d and 4b). Thus, our results suggest that synaptic activation of either AMPA or mGluR1 receptors during brief presynaptic bursts can contribute to endocannabinoid release, although the contribution of mGluR1 dominates under our conditions.

Dependence on the number and frequency of stimuli

We next explored the sensitivity of synaptically evoked endocannabinoid release by testing the effects of altering the stimulus number and frequency. First, we varied the number of stimuli while maintaining the frequency at 50 Hz (Fig. 5a, left). Bursts of as few as three pulses had small but measurable effects on presynaptic Ca2+ influx. The magnitude of the inhibition saturated at approximately ten pulses, although further increasing the number of stimuli in the conditioning train seemed to extend the duration of the inhibition. The magnitude of the inhibition for different numbers of stimuli is summarized in Fig. 5a (right).

Figure 5: Presynaptic modulation by endocannabinoids is highly sensitive to both stimulation number and frequency.
figure 5

(a,b) Presynaptic Ca2+ influx was measured in parallel fibers that were activated at 1 Hz before and after trains in which either the number of stimuli at 50 Hz was varied (a) or the frequency of a five-pulse train was varied (b). Traces on the left are the average responses before and after stimulation with a conditioning train at the time indicated by the arrows. Right, the modulation elicited 3 s after the burst. Six experiments are summarized in a, and four experiments in b.

Full size image

Next, we addressed the frequency-dependence of endocannabinoid inhibition by varying the stimulation rate while holding the number of stimuli constant. Because the effect of endocannabinoids saturated for ten pulses at 50 Hz (Fig. 5a), a train of five stimuli was used. Inhibition of presynaptic Ca2+ influx was detectable at 10 Hz and saturated by 100 Hz (Fig. 5b, left). The dependence of the inhibition of Ca2+ influx on stimulus frequency is summarized in Fig. 5b (right). These results show that bursts of 3–5 stimuli can lead to substantial retrograde inhibition by endocannabinoids.

Retrograde inhibition targets synaptically active inputs

It is not known whether endocannabinoids enable a Purkinje cell to globally regulate the synapses it receives or whether its synaptic inputs can be differentially regulated. Previous studies have shown that elevations of postsynaptic calcium can trigger endocannabinoid release3,4,5. Because widespread calcium increases in Purkinje cell dendrites have been observed under some experimental conditions30,31,32,33, Purkinje cell activation could lead to wide-ranging inhibition of synaptic inputs. Alternatively, the postsynaptic response to mGluR1 activation can be quite localized33,34,35, suggesting the possibility of synapse-specific modulation.

To determine whether subsets of excitatory inputs onto a Purkinje cell can be differentially modulated, we took advantage of the anatomical configuration of parallel fibers and Purkinje cells in the transverse slice. Parallel fibers run in a band in the molecular layer and make en passant synapses onto all Purkinje cells36. Two electrodes (Fig. 6a, red and green) placed within the band can be used to stimulate two independent subsets of parallel fibers and to activate different sets of synaptic inputs onto a Purkinje cell. We used a CCD camera to image presynaptic calcium transients evoked by stimulating each electrode independently, and verified that two different populations of parallel fibers were stimulated by the two electrodes (Fig. 6b). Using this configuration, short bursts of parallel fiber stimulation could be individually delivered to each of two subsets of parallel fibers making synapses onto Purkinje cells.

Figure 6: Endocannabinoid modulation is synapse-specific.
figure 6

(a) Diagram showing the placement of two electrodes in a band of parallel fibers labeled with the calcium indicator magnesium green. The electrodes were used to stimulate two separate populations of parallel fibers. The box outlines the region used (in b) to monitor presynaptic calcium transients. (b,c) Data from a single experiment. (b) The spatial location of fibers activated by each electrode was determined by activating the fibers with trains of 10 stimuli at 50 Hz and measuring the resulting change in fluorescence with a CCD camera. The fluorescence changes are shown in pseudocolor for clarity (red, electrode 1; green, electrode 2) and indicate the location of the two populations of fibers activated by electrodes 1 and 2. The ΔF signal is shown in cross-section to the right of the image. Scale bar, 20 μm. (c) Each population of parallel fibers shown in b was stimulated every second; stimulations of alternate electrodes were interleaved every 0.5 s while presynaptic Ca2+ influx was monitored. In the first trial, electrode 1 was activated with a 50-Hz, 10-pulse stimulus train (red arrow), and in the subsequent trial electrode 2 was activated with a 50-Hz, 10-pulse stimulus train (green arrow). (d) Data from eight experiments conducted in the same manner as in b and c. Using linear interpolation, the responses of these 16 pathways were averaged. The responses of the tetanized pathway and the non-tetanized pathway are shown.

Full size image

Ten stimuli at 50 Hz delivered with the first electrode (Fig. 6c, red arrow) inhibited presynaptic Ca2+ influx only in those fibers stimulated by that electrode (red circles) without affecting Ca2+ influx in fibers activated by the second electrode (green circles). Similarly, a conditioning stimulus delivered through the second electrode (Fig. 6c, green arrow) selectively suppressed Ca2+ influx only in that subset of activated fibers. We repeated these experiments for eight pairs of parallel fiber bands, and in each case, the conditioning train suppressed Ca2+ influx in the activated fibers (Fig. 6d, filled circles) without affecting the neighboring fibers (Fig. 6d, open squares). These findings show that, under our experimental conditions, the retrograde inhibition mediated by endocannabinoids was restricted to those inputs activated by a brief presynaptic burst.

Postsynaptic calcium signals are localized

Our results suggest that the signals that trigger endocannabinoid release are confined to regions of the Purkinje cell's dendrite. To examine postsynaptic factors governing the spread of endocannabinoid signaling, we combined current-clamp recordings with measurements of postsynaptic calcium levels (Capost). We first monitored changes in the spatial spread of Capost while varying the size of the EPSP. We recorded from Purkinje cells in sagittal slices with a pipette containing the indicator Fura-FF to measure Capost and Alexa-568 to visualize the cell (Fig. 7a–c). In this slice orientation, Purkinje cell dendrites are parallel to the surface of the slice, and parallel fibers are oriented vertically. After obtaining a stable EPSP of 1 mV following a single stimulus, the parallel fibers were activated with a train of ten stimuli at 50 Hz. The train produced Capost transients that were restricted to a small region of the dendritic arbor and caused the Purkinje cell to fire action potentials (Fig. 7d). Increasing the stimulus duration to 50 stimuli at 50 Hz increased the magnitude of Capost and prolonged the firing of the cell. Nonetheless, the large calcium signals were still spatially restricted to a small region of the dendritic arbor (Fig. 7e). After increasing the stimulus intensity to produce a 5-mV EPSP following a single stimulus, a train of ten stimuli at 50 Hz elevated Capost in a somewhat larger region of the dendrite and evoked sodium and calcium spikes (Fig. 7f). A comparison of the spatial extent of the postsynaptic calcium signal evaluated along the dendrite highlighted in Fig. 7c indicates that the Capost transients evoked by each of the stimulus intensities were restricted to a 20-μm-long portion of dendrite (Fig. 7g). The differences lay largely in the magnitude and the time course of the calcium signals (Fig. 7g,h).

Figure 7: Parallel fiber activation evokes localized postsynaptic responses.
figure 7

(a) Image of a Purkinje cell taken with a two-photon laser scanning microscope using 810 nm illumination to excite Alexa-568. Scale bar, 40 μm. (b) Enlarged view of the region in a outlined in blue. Scale bar, 10 μm. (c) An image of the same region taken with a CCD camera using 380 nm illumination to excite Fura-FF. The yellow line indicates the dendritic region analyzed in g. The arrows in c and g indicate the tip of the dendrite. Peak changes in fluorescence (top panels) and corresponding electrophysiological traces recorded in current clamp (bottom panels) for parallel fiber stimulation of different intensities: 10 stimuli at 50 Hz with an initial EPSP amplitude of 0.8 mV (d); 50 stimuli at 50 Hz with an 0.8 mV EPSP (e); and 10 stimuli at 50 Hz with a 5.1 mV EPSP (f). (d–f) The color bar in d indicates the fluorescence change (%ΔF/F). The timing of individual stimuli is indicated with tick marks below the traces. The spatial profile during the peak of the responses (g) and the time course of the calcium transients (h) corresponding to d (red), e (green) and f (blue). (i) The normalized peak EPSP is plotted for small (0.98 ± 0.03 mV; n = 9) and large (5.7 ± 0.3 mV; n = 7) stimulus intensities. The arrow marks the start of the train of 10 stimuli at 50 Hz.

Full size image

This remarkable spatial localization was a consistent finding. On average (n = 4), the extent of the calcium signal, measured using the width at half maximum (as in Fig. 7g) was 15.1 ± 1.0 μm (1.00 ± 0.02 mV EPSP, 10 at 50 Hz), 23.4 ± 3.7 μm (1.00 ± 0.02 mV EPSP, 50 at 50 Hz) and 31.1 ± 5.9 μm (5.10 ± 0.13 mV EPSP, 10 at 50 Hz) under the three conditions. Retrograde inhibition was observed following either low or high intensity stimulation (10 at 50 Hz), but the retrograde inhibition was more pronounced and longer lasting following high intensity stimulation (Fig. 7i). These findings suggest that the post-synaptic signals that trigger endocannabinoid release are quite local. Furthermore, our results indicate that retrograde inhibition is not restricted to exceptionally large stimulus strengths that activate a large number of inputs. Rather, even small synaptic inputs generate substantial retrograde inhibition.

Inhibition is restricted to active inputs

The spatially restricted Capost we observed in response to parallel fiber bursts (Fig. 7) is consistent with synapse-specific retrograde signaling, as suggested by data in Fig. 6. To further examine the extent to which retrograde signaling is localized, we next monitored Capost and the extent of retrograde inhibition while stimulating two separate sets of parallel fiber inputs onto a recorded Purkinje cell (Fig. 8). Two stimulus electrodes were placed 20 μm apart, and parallel fibers were activated with ten stimuli at 50 Hz. Stimulation with either electrode produced large calcium transients that were confined to non-overlapping regions of the dendrite (Fig. 8a–f). Again, retrograde inhibition was restricted to those synapses that were activated with the train (Fig. 8g,h). The retrograde inhibition is summarized for four such experiments during which the stimulus electrodes were placed 16–22 μm apart to stimulate adjacent regions of dendrite (Fig. 8i; eight pathways). No spatial spread of the retrograde signaling was evident.

Figure 8: Synapse specificity of the local signals triggering endocannabinoid release.
figure 8

(a) Image of a Purkinje cell taken as in Fig. 7a. Scale bar, 40 μm. (b) Enlarged view showing the dendritic regions activated by electrode 1 (red) and electrode 2 (green). Scale bar, 10 μm. (c,d) The spatial distribution of the peak calcium transient is shown following 10 stimuli at 50 Hz applied to either electrode 1 (c) or electrode 2 (d). The time courses of the calcium transients evoked by a train delivered to electrode 1 (e) or to electrode 2 (f) in region 1 (red traces in e and f, red outline in b) and region 2 (green traces in e and f, green outline in b). EPSPs recorded 1 s before the train (left traces), 1 s after the train (middle traces) and 15 s after the train (right traces) for electrode 1 (red) and electrode 2 (green) for a train delivered either to electrode 1 (g) or electrode 2 (h). (i) Normalized peak EPSP amplitudes for 4 cells (8 pathways) are summarized. The responses of the tetanized pathway and the non-tetanized pathway are shown. (j) Normalized peak EPSP amplitudes in control conditions and in Purkinje cells loaded with 40 mM BAPTA and 2 mM GDP-βS (n = 5). The arrows mark the start of the train in i and j.

Full size image

In view of the specificity of retrograde inhibition among synapses onto the same cell, we explored whether the synaptically evoked retrograde signal could spread among cells. Activating parallel fibers triggers endocannabinoid release from neighboring Purkinje cells as well as the cell under study. Can these neighboring cells affect the presynaptic terminals associated with the Purkinje cell from which we are recording? We examined this issue by including the intracellular calcium buffer BAPTA (40 mM) and the irreversible G-protein inhibitor GDP-βS (2 mM) in our recording pipette to prevent calcium-dependent and calcium-independent release of endocannabinoids from the recorded cell3,10. This manipulation prevented retrograde inhibition in the recorded cell (Fig. 8j) even though neighboring Purkinje cells continued to be synaptically activated and to release endocannabinoids. Thus, there was no detectable spread of endocannabinoid-mediated retrograde inhibition among cells or among synapses onto the same cell.

Discussion

We found that brief, high-frequency bursts of synaptic activity generate substantial endocannabinoid-mediated retrograde inhibition at PF-to-PC synapses. As few as 3–5 stimuli at 50 Hz were required to evoke half-maximal inhibition. This synaptic activity can trigger endocannabinoid release by activating postsynaptic AMPA or mGluR1 receptors and generating localized increases in postsynaptic calcium in the Purkinje cell dendrite. We found that endocannabinoid-mediated inhibition was restricted to synapses activated with a high-frequency burst. Our findings suggest that, under physiological conditions, endocannabinoids allow Purkinje cells to modulate subsets of synaptic inputs in response to brief bursts of granule cell activity.

Brief bursts of parallel fiber activity evoke retrograde inhibition

We first asked whether normal ongoing synaptic activity can trigger endocannabinoid release. Previous experiments suggest that endocannabinoid-mediated retrograde inhibition may suppress synaptic transmission only under extreme conditions. Depolarizing the postsynaptic cell or bath-applying drugs such as mGluR agonists mimics intense synaptic input onto a neuron to evoke robust retrograde inhibition3,4,5,8,9,10,11,12,13. When synaptic activation is used to trig-ger endocannabinoid release, prolonged stimulation is usually required to produce modest retrograde inhibition10. In view of the synapse-specific effects we observed (see below), it is not surprising that such strong stimulation was needed because this study10 assessed the heterosynaptic effects of parallel fiber activation on the climbing fiber to Purkinje cell synapse. Here, in contrast, we show that brief bursts of parallel fiber activity triggered sufficient endocannabinoid release to strongly suppress the parallel fiber to Purkinje cell synapse. We found that at stimulus frequencies of 20 Hz or higher, few stimuli are needed to evoke robust endocannabinoid-mediated retrograde inhibition. For example, a burst of five stimuli at 100 Hz inhibited Ca2+ influx to ∼45% of control. This corresponds to an even larger reduction in neurotransmitter release, which is steeply dependent on Ca2+ influx37. At 50 Hz, inhibition was half maximal after just 3–5 stimuli, and saturated after ∼10 stimuli. Although there are few descriptions of granule cell activity in vivo because these cells are small, closely packed and difficult to isolate, it is likely that granule cells fire at such rates21,22. Mossy fibers, which drive granule cells, often fire in short bursts of 150–350 Hz in awake behaving animals38,39, and mossy fiber–to–granule cell synapses can faithfully transmit high frequencies40. High-frequency bursts of activity are also observed at many central synapses during normal behavior and may have an important role in transmitting information41. Our findings suggest that the endocannabinoid system selectively responds to these brief, high-frequency bursts of synaptic input. Because synaptic transmission is selectively inhibited at those synapses, many central neurons may differentially regulate their synapses as a function of synaptic activity by releasing endocannabinoids.

AMPA and mGluR1 receptors trigger endocannabinoid release

We next assessed the roles of AMPA and mGluR1 receptors in evoking endocannabinoid release during synaptic activity. By monitoring presynaptic Ca2+ influx into parallel fibers, we tested whether activation of mGluR1 receptors alone can trigger endocannabinoid release. When AMPA receptors were blocked, we found that synaptic activation of mGluR1 receptors evoked robust retrograde inhibition. This inhibition was similar to the inhibition evoked under control conditions, even though the postsynaptic spiking and depolarization during the conditioning train were eliminated. Co-application of an mGluR1 antagonist and AMPA receptor antagonist eliminated this retrograde inhibition. Furthermore, changing the stimulus number and frequency affected the magnitude of synaptically evoked endocannabinoid release, much like the postsynaptic currents, calcium transients and transient post-tetanic depression mediated by mGluR1 receptors on Purkinje cells29,33,34,35,42,43. These results indicate that synaptic activity can trigger endocannabinoid release by activating postsynaptic mGluR1 receptors without relying on large postsynaptic depolarizations.

We also found that synaptic activation of AMPA receptors can play a role in evoking endocannabinoid release. In our current-clamp experiments, post-tetanic potentiation and the hyperpolarized potentials required to suppress the spontaneous activity of Purkinje cells may have obscured a contribution of AMPA receptors. By monitoring presynaptic Ca2+ influx, however, we uncovered a substantial contribution of AMPA receptors in triggering endocannabinoid release. In the presence of an antagonist of mGluR1s, ∼50% of the retrograde inhibition measured under control conditions remained. In this case, AMPA receptor activation likely depolarizes the Purkinje cell sufficiently to open voltage-gated calcium channels and to trigger endocannabinoid release. Thus, our findings indicate that synaptic activation of either AMPA or mGluR1 receptors at a small number of synapses can evoke significant endocannabinoid-mediated inhibition. It remains to be seen whether synaptic activation of mGluR1 or AMPA receptors at a single release site is sufficient to trigger endocannabinoid release.

Synapse specificity of endocannabinoid signaling

We next asked whether endocannabinoid release globally regulates a cell's inputs or whether retrograde inhibition is limited to active synapses under our conditions. Using two experimental approaches (Figs. 6 and 8), we found that retrograde inhibition of a Purkinje cell's synaptic inputs is restricted to those parallel fibers activated with a burst of high-frequency activity. Although we cannot rule out that an inactive presynaptic terminal interleaved among activated synapses may be affected, there was no detectable spread to synapses separated by less than 20 μm. Our results indicate that endocannabinoid release is restricted to a region of the Purkinje cell's dendritic arbor, and that there is little diffusion of endocannabinoids to neighboring synapses.

Previous studies of the spread of endocannabinoid signaling have largely focused on how endocannabinoids released from a depolarized cell influence synapses onto neighboring cells5,10,11,44. These studies have shown that the spread of endocannabinoid signaling following postsynaptic depolarization is influenced by many factors such as anatomy and temperature and can, under some conditions, have little effect on excitatory input onto neighboring cells10,11. Here we show that synaptically evoked endocannabinoid release does not affect neighboring cells (Fig. 8j), indicating that under these conditions, endocannabinoids are unable to affect parallel fiber synapses even a few microns from the site of endocannabinoid release.

Much less is known about the spread of endocannabinoid signaling among synapses onto a single cell. Synaptic activity could result in widespread increases in calcium30,31,32,33 and elicit a generalized release of endocannabinoids, thereby modulating all of the cannabinoid-sensitive inputs onto a cell. Prolonged synaptic stimulation revealed such a heterosynaptic spread of endocannabinoid-mediated inhibition from activated parallel fiber inputs to the climbing fiber input onto a Purkinje cell10. However, when we used brief synaptic activation, postsynaptic calcium signals33,34,35,45,46 and retrograde inhibition were both highly localized.

Functional implications

To date, it has been difficult to determine the function of retrograde inhibition by the endocannabinoids. One possibility is that the endocannabinoid system globally regulates a cell's cannabinoid-sensitive inputs, much like the homeostatic mechanisms that maintain the activity of a postsynaptic cell in a desirable range over hours or days47. The heterosynaptic inhibition observed following prolonged synaptic stimulation suggests that high levels of synaptic activity can elicit a generalized release of endocannabinoids and cause a widespread inhibition of a Purkinje cell's cannabinoid-sensitive inputs10. However, such global regulation would be invoked only during periods of extreme synaptic activity.

Here, we describe conditions that may routinely trigger endocannabinoid release. We found that brief bursts of synaptic activity evoke robust endocannabinoid-mediated retrograde inhibition. The patterns of synaptic activity that trigger endocannabinoid-mediated retrograde inhibition are commonly encountered during ongoing activity at many central synapses. Moreover, we show here that synaptic activation with high-frequency bursts inhibits parallel fiber synapses in a synapse-specific manner. The sensitivity and specificity of the endocannabinoid system provide a mechanism for a Purkinje cell to differentially regulate its synaptic inputs in response to their pattern of synaptic activation on the timescale of seconds. Furthermore, the induction of endocannabinoid-dependent forms of long-term plasticity14,15,18,48,49,50 could also be influenced by the spatially restricted endocannabinoid release evoked by brief trains that we observed. Our findings suggest that, at many central synapses, neurons use endocannabinoids to transiently inhibit specific synaptic inputs following bursts of activity, thereby dynamically modulating synaptic strength and fine-tuning their properties of synaptic integration.

Methods

Transverse or sagittal slices (300 μm thick) were cut from the cerebellar vermis of 12–19-day-old Sprague-Dawley rats as described previously3. All procedures involving animals were approved by the Harvard Medical Area Standing Committee on Animals. Experiments were performed in transverse slices except when monitoring postsynaptic calcium (Figs. 7 and 8). Trials were separated by 2–5 min. All chemicals were purchased from Sigma/RBI except for AM251, CPCCOEt, CPPG, L-AP4 and NBQX, which were purchased from Tocris Cookson.

Presynaptic Ca2+ imaging.

Parallel fibers were loaded with Magnesium Green-AM (Molecular Probes), and fluorescence intensity was measured with a photodiode as previously described24. In some experiments, fluorescence intensity was measured with a CCD camera (SensicamQE, Cooke Corporation). A monochromator set at 495 nm with a slit width of 24 nm (Photon Technology International) was used for excitation. The filter set used was 515DCLP dichroic and 530LP for emission (Omega Optical).

Electrophysiology.

Whole-cell recordings of visualized Purkinje cells were obtained as previously described3. Current-clamp experiments were performed using an Axoclamp 2B in bridge mode or a Multiclamp 700A in current-clamp mode (Axon Instruments). Glass electrodes (2–4 MΩ) were filled with an internal solution containing 130mM KMeSO3 or 130 mM potassium gluconate, 10 mM NaCl, 2 mM MgCl2, 0.16 mM CaCl2, 0.5 mM EGTA, 10 mM HEPES, 4 mM Na2ATP, 0.4 mM NaGTP and 14 mM Tris-Creatine phosphate; pH 7.3, 315 mosm. BAPTA (Molecular Probes) replaced potassium gluconate when a postsynaptic calcium buffer was used.

Postsynaptic Ca2+ imaging.

To measure postsynaptic calcium signals, the intracellular solution was supplemented with 20 μM Alexa-568 (Molecular Probes) and 500 μM Fura-FF (Molecular Probes). Fura-FF was excited at 380 nm with a Till Polychrome IV monochromator (Till Photonics). The filter set used was 455 dichroic (Till) and 455LP for emission (Omega Optical). Images were acquired with 18-ms exposures at 50 Hz using a Cooke Sensicam QE. Fluorescence excitation was restricted to a small area including the Purkinje cell dendrites and an adjacent cell-free region that was used for background correction. Photobleaching of indicators during imaging protocols was < 1%. For experiments measuring postsynaptic calcium transients, parallel fibers were focally stimulated using theta glass electrodes with a tip diameter of 2 μm.

High-resolution images of Purkinje cells.

Two-photon laser scanning images of Purkinje cells were obtained with a modified Olympus confocal microscope equipped with a Ti-sapphire pulsed laser (Coherent Inc.). Alexa-568 was excited at 810 nm. Data were collected using Fluoview software (Olympus). A 3×3 median filter was applied to individual confocal images and the z-stack was assembled by generating a maximal intensity projection of the confocal series. The confocal images were aligned with the CCD images using a linear conformal transform in Matlab (Mathworks).

Data acquisition and analysis.

All signals were digitized with a 16-bit A/D converter (Instrutech). EPSPs were filtered at 3–10 kHz and digitized at 10–50 kHz. Photodiode currents were digitized at 5 kHz and digitally filtered at 300–1,000 Hz. The Ca2+-dependent fluorescent signals were background-corrected and expressed as the change in fluorescence divided by the unstimulated fluorescence. All analysis was performed using custom macros written in Igor Pro (Wavemetrics), Matlab (Mathworks) or Vision Software (Till). Averages are given as mean ± s.e.m.