Introduction
The point-to-point transfer of information between neurons by synaptic transmission depends on several factors, including the concentration of transmitter in vesicles, the probability of vesicle release (Pr), the geometry of the synaptic space, and the number and location of postsynaptic receptors. Varying any of these factors causes changes in synaptic efficacy. Synaptic strength depends strongly on Pr; until recently, the belief was that presynaptic G protein–coupled receptors are responsible for tuning neurotransmitter Pr, whereas ionotropic receptors (such as the NMDA, AMPA and kainate subtypes of glutamate receptors) act on the postsynaptic cell. However, the landscape has changed and there is now considerable evidence that these receptors can be found on the presynaptic end of the synapse. In this issue1, a provocative study by Duguid and Smart adds to this emerging view by convincingly showing that a new form of synaptic plasticity of inhibitory interneurons in the cerebellum is caused by the activation of presynaptic NMDA receptors (NMDARs).
A surge of recent studies has investigated the role of presynaptic NMDARs in synaptic transmission. Immunohistochemical evidence suggested the existence of presynaptic NMDARs2. These data are now accompanied by functional work suggesting that presynaptic NMDARs can be found on both excitatory and inhibitory terminals. For instance, in visual cortical layer 5 neurons, Sjostrom et al.3 demonstrated that application of APV, an NMDAR antagonist, reduced the frequency of mini EPSCs, a phenomenon usually ascribed to a presynaptic locus. These authors went on to show that presynaptic NMDARs act as coincidence detectors for the induction of long-term depression (LTD). Similar results have been obtained in entorhinal cortex, where putative presynaptic NMDARs increase the frequency of mini EPSCs4. More recently, Humeau et al.5 showed that in lateral amygdala, glutamate released by thalamic terminals following repetitive stimulation activates presynaptic NMDARs on cortical afferents, causing an increase in the probability of glutamate release and induction of a heterosynaptic form of long-term potentiation. Spinal cord primary afferents are another area where presynaptic NMDARs regulate the release of both substance P6 and glutamate7.
Axons from cerebellar granule cells that make up parallel fibers also appear to express functional NMDARs that are proposed to induce the synthesis of nitric oxide, leading to LTD8. In cerebellar slices, bath application of NMDA leads to an increase to the frequency of spontaneous IPSCs9. Now Duguid and Smart1 provide compelling evidence for a new form of synaptic plasticity mediated by these presynaptic NMDARs located on cerebellar inhibitory interneurons. They show that depolarization of Purkinje cells (PCs) increases the frequency of mini GABAergic IPSCs (mIPSCs), and provide evidence that the frequency increase is through glutamate acting as a retrograde messenger on the interneuron presynaptic terminals.
Duguid and Smart1 recorded from rat cerebellar slices and found that a short depolarization of PCs leads to the previously described phenomena of depolarization-induced suppression of inhibition (DSI; Fig. 1a)—a decrease of mIPSC frequency mediated by activation of presynaptic cannabinoid CB1 receptors10—followed by an increase in mIPSC quantal amplitude, a phenomenon termed 'rebound potentiation' (RP; Fig. 1b)11. Surprisingly, however, they also observed a large rebound increase of mIPSC frequency superimposed on the RP. They termed this increase in frequency 'depolarization-induced potentiation of inhibition' (DPI; Fig. 1c). The effect was also mimicked by repetitive CF activation. DPI was blocked by buffering PC calcium with the Ca2+ chelator BAPTA, indicating that DPI, like DSI and RP, is Ca2+ dependent. Critically, the authors show that DPI changes the paired-pulse ratio and the 1/CV2 of the evoked response, indicating that the site of action is presynaptic.
Figure 1: Three forms of calcium-induced plasticity at inhibitory interneuron–Purkinje cell synapses.
![Figure 1 : Three forms of calcium-induced plasticity at inhibitory interneuron|[ndash]|Purkinje cell synapses.](/neuro/journal/v7/n5/thumbs/nn0504-419-F1.jpg)
In each panel, the top trace plots the GABAA charge transfer in arbitrary units versus time, and the diagram below shows a Purkinje cell (center circle) surrounded by an interneuron terminal, a climbing fiber terminal and a Bergmann glial process. In each diagram, the climbing fiber has just been stimulated, activating voltage-gated calcium channels (VGCC). (a) In DSI, the Ca++ initiates endocannabinoid synthesis that activates presynaptic CB1 receptors, resulting in inhibition of GABA release for tens of seconds. (b) In RP, Ca++ activates a kinase that phosphorylates the GABAA receptor, enhancing GABAA receptor currents for a prolonged period. (c) In DPI, Ca++ induces the release of glutamate from the PC by an unknown mechanism. Glutamate activates presynaptic NMDARs leading to Ca++ entry and release of Ca++ from ryanodine-sensitive Ca++ stores, which, in turn, increases the rate of vesicular GABA release for 
15 min. RyR, ryanodine receptor.
Debbie Maizels
Full size image (47 KB)DPI is mediated by NMDAR activation, because it is blocked by APV, a selective antagonist at NMDARs. Importantly, although APV blocked the induction of DPI, it did not affect the induction of either DSI or RP, in agreement with previous observations that these phenomena are NMDAR independent. Consistent with the activation of NMDARs by glutamate, application of THA, a glutamate transporter antagonist, during a subthreshold stimulus caused DPI. As a further test of NMDAR involvement in DPI, brief pressure application of NMDA to the PC dendritic tree mimicked DPI. Finally, the authors showed that the NMDAR subunits NR1 and NR2A–D are colocalized with synaptophysin (a synaptic vesicle protein) and GAD65/67 (glutamic acid decarboxylase; a marker for presynaptic GABA-containing terminals) on cultured cerebellar interneurons. The authors present evidence that Ca2+ influx through presynaptic NMDARs causes Ca2+ release from intracellular ryanodine-sensitive Ca2+ stores in the interneurons, which then causes an increase in GABA Pr.
These elegant experiments leave some unanswered questions. First, how does glutamate get released from PCs and escape uptake by glutamate transporters? The postsynaptic Ca2+ dependence of DPI suggests that glutamate, which is typically released from vesicles at synapses, is released through a SNARE-type vesicular mechanism. Such a mechanism would also require the presence of a vesicular glutamate transporter in PCs. Although the labeling of the PC layer for VGLUT3 mRNA12 is encouraging, few vesicles have been reported in PC dendrites, unlike the olfactory bulb where dendritic glutamate release has been clearly established. Glutamate release mediated by reverse glutamate transport can be ruled out because, unlike DPI, such a mechanism is independent of Ca2+. Understanding how glutamate is released from PCs therefore requires further investigation.
The authors also suggest that glutamate liberated from PCs would saturate the transporters found on surrounding Bergmann glia and then activate presynaptic NMDARs. Although the results with THA are consistent with a role for transporters in DPI, THA can activate NMDARs directly13. Moreover, THA is also transported by glutamate transporters14; these results would need to be replicated with the use of another inhibitor such as TBOA, which does not have these problems. Additionally, aside from the glutamate transporters found on Bergmann glia, PCs express high levels of the neuronal glutamate transporter EAAT4. Thus, the released glutamate has to escape not only the transporters found on the Bergmann glial cells encapsulating the PC, but also the transporters located on the PC dendritic tree. However, to produce DPI the authors depolarized the PC to 0 mV and, because the ability of glutamate transporters to capture glutamate depends on membrane potential, the depolarization would lower the binding of EAAT4 to glutamate. Thus, the depolarization may have two roles: causing Ca2+ influx and reducing the ability of neuronal transporters to capture glutamate.
A compelling finding in the new study1 is that NMDAR subunits localize with GAD65/67 and synaptophysin. This strongly suggests that NMDARs are localized on the presynaptic terminal and not on the axonal regions. However, the immunohistochemical analysis was done on mixed cultured cerebellar interneurons, and a similar analysis in slice preparations would be more definitive. This is also important because the subunit composition of NMDARs controls the extent of Mg2+ block. It is noteworthy that in the presence of 1 mM extracellular Mg2+, no prior depolarization of the interneuron was needed to induce DPI. One explanation is that the presynaptic NMDARs have a very low Mg2+ sensitivity or, less likely, that the terminal is already partly depolarized. Finally, the authors focus their analysis entirely on the facilitation of spontaneous release of GABA. We wonder what happens to evoked release, as in many instances presynaptic NMDAR activation actually decreases evoked release7, 9.
What role might DPI provide in cerebellar function? Small changes in the balance between inhibition and excitation within the cerebellar neural network result in ataxia. Thus, DPI, as well as RP, may provide a way of damping down PC excitability following strong activation. Regardless of the mechanism of glutamate release from PCs or the NMDAR subunit composition, the present work argues forcefully that NMDARs localized on interneuron terminals in cerebellum induce a new form of inhibitory synaptic plasticity. It will be of interest to see if this form of plasticity is found in other brain regions and how it might govern the computational power of the cerebellum.
