Coordinated regulation of endocannabinoid-mediated retrograde synaptic suppression in the cerebellum by neuronal and astrocytic monoacylglycerol lipase

The endocannabinoid 2-arachidonoylglycerol (2-AG) mediates retrograde synaptic depression including depolarization-induced suppression of excitation (DSE) and inhibition (DSI). 2-AG is degraded primarily by monoacylglycerol lipase (MAGL), which is expressed in neurons and astrocytes. Using knockout mice in which MAGL is deleted globally or selectively in neurons or astrocytes, we investigated the relative contribution of neuronal and astrocytic MAGL to the termination of DSE and DSI in Purkinje cells (PCs) in cerebellar slices. We report that neuronal MAGL plays a predominant role in terminating DSE at climbing fiber (CF) to PC synapses, while both neuronal and astrocytic MAGL significantly contributes to the termination of DSE at parallel fiber (PF) to PC synapses and DSI at putative Stellate cell to PC synapses. Thus, DSE and DSI at different synapses is not uniformly affected by global and cell type-specific knockout of MAGL. Additionally, MAGL global knockout, but not cell type-specific knockout, caused tonic activation and partial desensitization of the CB1 receptor at PF-PC synapses. This tonic CB1 activation is mediated by 2-AG since it was blocked by the diacylglycerol lipase inhibitor DO34. Together, these results suggest that both neuronal and astrocytic MAGL contribute to 2-AG clearance and prevent CB1 receptor over-stimulation in the cerebellum.

Immunohistochemical studies have shown MAGL is not only expressed in neurons, but also in astrocytes 17,18 . In an attempt to determine the relative contribution of different brain cell types to 2-AG metabolism and signaling, we recently generated conditional knockout mice in which MAGL was deleted globally or specifically in neurons or astrocytes 18 . We showed that neuronal and astrocytic MAGL coordinately regulate brain 2-AG content, and cell type-specific deletion of MAGL from either neurons or astrocytes produces similar, modest but significant prolongation of DSE and SSE at the parallel fiber-Purkinje cell (PF-PC) synapses in the cerebellum, while global knockout of MAGL produces more dramatic prolongation of DSE and SSE 18,19 . In addition, we found that both neuronal and astrocytic MAGL limit the spread of endocannabinoid-mediated synaptic depression among PF-PC or PF-Golgi cell synapses and confer synapse-specificity of endocannabinoid signaling in the cerebellum 19 . Thus, both neuronal and astrocytic MAGL contribute similarly to the termination of retrograde 2-AG signaling at PF-PC synapses. However, it remains unknown whether cell type-specific knockout of MAGL alters DSE and DSI at other synapses in the cerebellum.
PCs, the single output neurons of the cerebellar cortex, receive excitatory inputs from PFs and climbing fibers (CFs) 20,21 . Depolarization of PCs induces DSE at both PF-PC and CF-PC synapses 6 . PCs receive inhibitory synaptic inputs from Basket cells (BCs) and Stellate cells (SCs), two types of interneurons that provide feed-forward inhibition to PCs [22][23][24] . DSI can be induced at interneuron-PC synapses 17,25,26 . MAGL global knockout prolonged DSI at SC-PC synapses but was not found to affect DSI at BC-PC synapses 17 . Thus, we studied the effects of cell type-specific MAGL knockout on putative SC-PC synapses.
It remains unknown whether cell type-specific knockout of MAGL affects DSE at CF-PC and DSI at SC-PC synapses. In the cerebellum, not all CB 1 R-positive nerve terminals express MAGL, and MAGL is not uniformly expressed in different cell types and cell layers 17,27 . Here, we sought to examine the effects of global and cell type-specific deletion of MAGL on DSE and DSI at several types of synapses onto PCs. In addition, we investigated the extent to which chronic increases in 2-AG content in MAGL knockout mice 18 cause tonic activation and desensitization of CB 1 receptors at PF-PC synapses. Our results revealed distinct effects of global and cell type-specific knockout of MAGL on endocannabinoid-mediated retrograde depression at different types of synapses.

Results
Effects of global and cell type-specific knockout of MAGL on DSE. Depolarization of PCs induces DSE at both CF-PC and PF-PC synapses 6 . We first examined the extent to which global or cell type-specific knockout of MAGL altered DSE at CF-PC synapses in cerebellar slices. CF-EPSCs were evoked as described in Materials and Methods. One-way ANOVA showed that global knockout (MAGL-TKO) and neuron-specific knockout of MAGL (MAGL-NKO) substantially prolonged CF-DSE as indicated by the increases in the decay time constant (τ ) (F (3,35) = 11.45, p < 0.001; TKO vs. WT, p < 0.001; NKO vs. WT, p = 0.003), whereas astrocyte-specific deletion of MAGL (MAGL-AKO) did not significantly alter the duration of CF-DSE (p = 0.975, Fig. 1A-C). Global or cell type-specific knockout of MAGL did not significantly change the magnitude of CF-DSE (F (3,35) = 1.15, p = 0.346; Fig. 1D). Thus, MAGL from neurons, but not astrocytes, plays a predominant role in terminating CF-DSE.
The above results stand in contrast with findings obtained at PF-PC synapses, where neuronal and astrocytic MAGL contributes similarly to the termination of DSE and SSE 18,19 . The MAGL inhibitor JZL184 produces dramatic prolongation of DSE 14,28 and SSE 19 , and does not further alter DSE at PF-PC synapses in slices prepared from MAGL global knockout mice 16 . SSE induced at PF-PC synapses in MAGL-TKO slices is not significantly different from that induced in wild-type slices treated with JZL184 19 . These results suggest that MAGL global knockout mimics and occludes the action of the MAGL inhibitor. However, it was unclear whether JZL184 exerts its effects by inhibiting neuronal MAGL, astrocytic MAGL, or both. In the present study, we determined whether JZL184 affected DSE at PF-PC synapses in MAGL cell type-specific knockout mice. Consistent with our previous study 18 , we found that MAGL-NKO and MAGL-AKO substantially prolonged PF-DSE compared with wild-type control (τ , F (2,23) = 8.59, p = 0.002; NKO vs. WT, p = 0.037; AKO vs. WT, p = 0.002; Fig. 2A). Interestingly, JZL184 (0.3 μ M) significantly prolonged DSE in wild-type, MAGL-NKO and MAGL-AKO slices ( Fig. 2A,B) and abrogated the difference in the decay time constants (τ ) of the three groups (F (2,23) = 0.06, p = 0.938; Fig. 2C). The magnitude of DSE in these three groups was not significantly different in the absence (F (2,23) = 0.13, p = 0.885) or the presence of JZL184 (F (2,23) = 0.07, p = 0.931; Fig. 2D). Thus, JZL184 can inhibit either neuronal MAGL or astrocytic MAGL in MAGL-AKO and MAGL-NKO slices, respectively, to prolong DSE in MAGL cell type-specific knockout mice.
Effects of global and cell type-specific knockout of MAGL on DSI. SCs and BCs are GABAergic interneurons that receive excitatory inputs from PFs and provide feed-forward inhibition to PCs 22,23 . A brief depolarization of PCs could induce DSI at putative BC-and SC-PC synapses 17,25,26 . However, MAGL global knockout prolonged DSI at SC-PC synapses but had no effect on DSI at BC-PC synapses 17 . We therefore determined whether global and cell type-specific knockout of MAGL altered DSI at SC to PC synapses. The cell bodies of SCs are located in the outer molecular layer, whereas BC somata are found in the lower molecular layer just above Purkinje cells 29,30 . Putative SC-PC IPSCs were evoked by placing the stimulation electrode in the outer third of the molecular layer, which likely preferentially stimulates SC afferents 29,30 . However, we could not rule out a minor contribution of IPSCs from BC afferents in our studies. MAGL-TKO, -NKO and -AKO substantially prolonged DSI, as shown by significant increases in the decay time constant (τ ) of DSI (F (3,40)  MAGL-TKO, but not MAGL-NKO or -AKO, caused tonic activation and partial desensitization of the CB 1 receptor. MAGL-TKO causes a drastic increase in 2-AG content in the brain, and MAGL-NKO or -AKO also cause a significant increase in 2-AG content 18 . If the increase in 2-AG content causes tonic activation of the CB 1 receptor, CB 1 receptor blockade should enhance basal synaptic transmission. To test this possibility, we examined the effects of the CB 1 receptor antagonist AM251 on basal PF-EPSCs in PCs in MAGL global and cell type-specific knockout mice. Bath application of AM251 (2 μ M) significantly increased PF-EPSCs in MAGL-TKO slices (F (3,32) = 9.77, p < 0.001; TKO vs. WT, p < 0.001; Fig. 4A,D,E), but had no significant effects on PF-EPSCs in WT, MAGL-NKO and -AKO mice (NKO vs. WT, p = 0.581; AKO vs. WT, p = 0.746, Fig. 4A,B,C,E). These results revealed that global deletion of MAGL, but not cell type-specific deletion of MAGL causes persistent suppression of PF-EPSCs through tonic activation of the CB 1 receptor.
We determined whether 2-AG mediated the tonic activation of the CB 1 receptor in MAGL-TKO slices. 2-AG is synthesized by diacylglycerol lipase (DAGL), which converts diacylglycerol into 2-AG 31,32 . We have shown that the selective DAGL inhibitor DO34 was effective in blocking DSE and DSI in the cerebellum and hippocampus 33  To test possible CB 1 receptor desensitization, we examined whether the CB 1 receptor agonist WIN55,212-2-induced depression of EPSCs was altered in MAGL global and cell type-specific knockout mice. Bath application of a saturating concentration (2 μ M) of the CB 1 agonist WIN55,212-2 produced significantly less depression of PF-EPSCs in cerebellar PCs in MAGL-TKO slices (F (3,41) Fig. 5A,B,C,E). These results indicate that only global deletion of MAGL causes tonic activation and partial desensitization of the CB 1 receptor at PF-PC synapses.

Discussion
The endocannabinoid 2-AG mediates multiple forms of retrograde synaptic depression including DSE and DSI 3,4,34 . 2-AG is hydrolyzed primarily by MAGL 13 , which is expressed in both neurons and astrocytes 17 . The present study examined the effects of global and cell type-specific knockout of MAGL on DSE and DSI in the cerebellum. The results indicate that the impact of global and cell type-specific knockout of MAGL on DSE and DSI varied among PCs, the principal output neurons of the cerebellar cortex, receive excitatory synaptic inputs from PFs from granule cells and CFs from the inferior olive 20 . Both PF-PC and CF-PC synapses display DSE 6 . Pharmacological inhibition and global knockout of MAGL substantially prolongs DSE at both PF-PC and CF-PC synapses in cerebellar slices 14,16,17 . Global knockout of MAGL dramatically prolongs DSE and SSE at PF-PC synapses 16,18,19 , while neuron-and astrocyte-specific deletion of MAGL produces modest but significant prolongation of DSE and SSE 18,19 . Further, viral transduction of Bergmann glia (BG) with MAGL in a MAGL global knockout background significantly shortens PF-DSE 17 . We extended these studies in an important way by examining the effects of cell type-specific knockout of MAGL on DSE at CF-PC synapses, and for the purpose of comparison, on DSE at PF-PC synapses as well. Here we show that MAGL-NKO and -AKO produced similar, modest prolongation of PF-DSE. In contrast, MAGL-AKO did not significantly prolong CF-DSE, whereas MAGL-NKO and -TKO produced robust prolongation of CF-DSE. Thus, neuronal and astrocytic MAGL collaborate to terminate PF-DSE, whereas neuronal MAGL plays a predominant role in terminating CF-DSE.
Previous studies have shown that the MAGL inhibitor JZL184 produces dramatic prolongation of DSE and SSE at PF-PC synapses in cerebellar slices prepared from wild-type mice 14,19 . JZL184 does not further alter DSE at PF-PC synapses in slices prepared from MAGL global knockout mice 16 . We found that JZL184 significantly prolonged PF-DSE in wild-type, MAGL-NKO and MAGL-AKO slices, and there was no significant difference in the observed DSE among the three groups with JZL184. These results revealed that JZL184 can inhibit neuronal or astrocytic MAGL in MAGL-AKO and MAGL-NKO mice, respectively, to prolong PF-DSE.
Given that not all CB 1 R-positive nerve terminals express MAGL 17 , one may suspect that the impact of cell type-specific MAGL knockout on DSE is determined by selective distribution of MAGL at individual synapses. However, this appears to not always be the case. Immunohistochemical studies have shown that MAGL is highly enriched in the PF terminals in the molecular layer but is absent in CF terminals 17 . However, the CF terminals are surrounded by dense PF terminals. The degradation of 2-AG by MAGL expressed at heterosynaptic sites may explain why neuron-specific knockout of MAGL substantially prolongs DSE at CF-PC synapses. BGs are astrocytes in the cerebellum that have their cell bodies in the PC layer and processes that extend into the molecular layer 35 . MAGL is weakly expressed in the BG 17 . Both the CF and PF terminals are surrounded by processes of BGs. However, we found that astrocyte-specific knockout of MAGL substantially prolonged PF-DSE but did not have a significant effect on CF-DSE. It is unclear why astrocyte-specific deletion of MAGL produced different effects on PF-and CF-DSE.
It was not previously known whether MAGL cell type-specific knockout alters DSI in the cerebellum. We found that MAGL-TKO, -NKO and -AKO significantly prolonged DSI at putative SC-PC synapses. The stimulating electrode was restricted to the outer third of the molecular layer, which contains numerous GABAergic axonal terminals from SCs, the only neurons localized exclusively in the molecular layer 30 . However, the molecular layer does contain sparse axonal projections from BCs, the somata of which are located just above the PC layer 30 . There is a possibility that BC afferents make a minor contribution to IPSCs recorded in the present study. SCs provide feed-forward inhibition that temporally limits excitatory input to PCs and reduces spike generation by asynchronous inputs 22,24 . Immunofluorescent staining revealed that GABAergic nerve terminals in the molecular layer do not express MAGL 17 . However, GABAergic axonal terminals are surrounded by a dense network of PF terminals and BG processes that express MAGL, which may explain why neuronal and astrocytic deletion altered DSI at SC-PC synapses.
Previous studies have shown that genetic deletion or chronic blockade of MAGL caused elevations in 2-AG levels and CB 1 receptor desensitization 16,36,37 . We found that, compared with that of wild-type mice, the magnitude of WIN55,212-2-induced depression of PF-EPSCs was decreased in MAGL-TKO mice, but not in MAGL-AKO and MAGL-NKO mice. Thus global knockout of MAGL, but not cell type-specific knockout of MAGL, causes partial CB 1 receptor desensitization. This desensitization in MAGL-TKO mice is likely caused by 2-AG-induced tonic activation of CB 1 receptors. Indeed, we found that AM251 significantly increased basal PF-EPSCs in MAGL-TKO mice, and the effect of AM251 was blocked by the DAGL inhibitor DO34 33 . There was no detectable tonic activation of CB1 receptors in MAGL-NKO and MAGL-AKO mice. These findings may reflect the sufficiency of the remaining neuronal or astrocytic MAGL in MAGL-AKO or -NKO mice, respectively, to clear 2-AG from the vicinity of CB 1 receptors. These results support the idea that neuronal or astrocytic MAGL can clear 2-AG sufficiently to prevent over-stimulation and desensitization of CB 1 receptors, as only global knockout of MAGL causes tonic activation and partial desensitization of the CB 1 receptor.
Previously, CB 1 receptor desensitization was investigated in global and cell type-specific MAGL knockouts using both behavioral and biochemical assays 18 . In these experiments, it was found that global knockout but not cell type-specific knockout was associated with behavioral and biochemical desensitization to the CB 1 receptor agonist WIN55,212-2 18 . Our experiments are consistent with these findings, as only MAGL-TKO mice displayed evidence for tonic activation and desensitization of the CB 1 receptor. Thus, our results add electrophysiological evidence in support of CB 1 receptor desensitization in MAGL global knockout mice. However, the molecular mechanism underlying these findings remains unclear. As mentioned previously, CB 1 receptor desensitization likely arises from chronically elevated 2-AG content resulting in CB 1 receptor overstimulation. Indeed, chronic administration of cannabis or other CB 1 receptor agonists in mice and humans leads to tolerance to several of the effects of these agonists and an associated CB 1 receptor downregulation [38][39][40][41][42][43][44][45][46] , which may account for the partial desensitization seen in our studies.
MAGL accounts for about 85% of 2-AG hydrolysis in mouse brains. Additional serine hydrolases, including serine hydrolase alpha-beta-hydrolase domain 6 and 12 (ABHD6 and ABHD12), are responsible for the remaining 15% 13 . Selective ABHD6 inhibitors WWL123 and WWL70 did not alter DSE and DSI in cerebellar PCs in wild-type and MAGL global knockout mice 16,17 . Thus, ABHD6 likely does not play a significant role in 2-AG hydrolysis in the cerebellum, even in the context of MAGL global knockout. It remains to be determined whether ABHD12 contributes to the termination of DSE and DSI. However, this cannot be investigated until a specific ABHD12 inhibitor becomes available.
In summary, we have found that in most synaptic inputs to PCs, both neuronal and astrocytic MAGL contributed significantly to the termination of DSE and DSI. Neuronal MAGL, but not astrocytic MAGL, significantly contributed to the termination of CF-DSE. In addition, we found that MAGL-TKO, but not -NKO and -AKO, caused tonic activation and partial desensitization of CB 1 receptors at PF-PC synapses. These results suggest that neuronal and astrocytic MAGL collaborate to terminate endocannabinoid-mediated synaptic suppression at most synapses.
Data Analysis and Statistics. The EPSC and IPSC amplitudes were normalized to the baseline. The decay time constant (τ ) and magnitude of DSE and DSI were measured as described previously 14 . The change (%) of EPSCs by the CB 1 agonist/antagonist was calculated as follows: 100 × [mean amplitude of EPSCs at last 5 min of drug application/mean amplitude of baseline EPSCs]. Data are presented as the mean ± SEM. Results were analyzed with one-way ANOVA or Student's t-test. Results were considered to be significant at p < 0.05.