View all hidden authors and organizations

Edited by Peter L. Strick, University of Pittsburgh, Pittsburgh, Pennsylvania, approved on June 7, 2021 (review received on February 18, 2021)

Learning depends on synaptic plasticity. The signaling mechanism that controls the induction of plasticity determines the learning rules for the specific synapses involved. In addition, the relationship between the activity pattern of synaptic input and the type, direction, and level of plasticity may change during development. Here, we established a key link between receptor activation of Purkinje cells in the presynaptic cerebellum, downstream signaling mechanisms, and the ability of adult animals to learn cerebellar motor tasks.

Long-term synaptic plasticity is considered to be the cell matrix for learning and memory. Synaptic plasticity rules are defined by the specific complement of receptors in synapses and related downstream signaling mechanisms. In young rodents, at the cerebellar synapse between granular cells (GC) and Purkinje cells (PC), bidirectional plasticity is determined by the presynaptic N-methyl-D-aspartate receptor (NMDAR) Activation drives the balance between transcellular nitric oxide (NO) formation and postsynaptic calcium dynamics. However, the role and location of NMDAR activation in these pathways are still controversial in mature animals. Here, we showed in adult rodents that NMDAR exists and functions at the presynaptic terminals, where their activation triggers NO signaling. In addition, we found that selective genetic deletion of pre-synaptic but not postsynaptic NMDAR prevents the synaptic plasticity of parallel fiber PC (PF-PC) synapses. Consistent with this finding, the selective absence of GC NMDARs affects the adaptation of the vestibular ocular reflex. Therefore, bidirectional synaptic plasticity and cerebellar motor learning require NMDARs presynaptic to PC.

The ability of organisms to adjust their behavior according to environmental needs depends on their ability to learn and perform coordinated movements. The cerebellum optimizes the exercise program through trial and error, thus playing a central role in this process (1). In the cerebellum, the synaptic output from granular cells (GC) to Purkinje cells (PC) shapes the calculation operations during basic motor functions and serves as the basis for motor learning (2). Several forms of motor learning depend on changes in the strength of parallel fibers (PF) (GC axons) to PC synapses (3, 4).

In the mammalian forebrain, synaptic plasticity usually depends on the activation of the postsynaptic N-methyl-D-aspartate receptor (NMDAR), which changes the AMPA receptor (AMPAR) turnover at the postsynaptic site ( 5). However, this may not extend to the cerebellar synapse between GC and PC, because no functional postsynaptic NMDAR is found in young or adult rodents (6, 7). However, pharmacological methods have shown that both long-term inhibition (LTD) and long-term potentiation (LTP) induction are dependent on the activation of NMDAR at the synapses of PF-PC in young rodents (8⇓ ⇓ ⇓ –12). Therefore, an alternative mechanism of NMDAR-dependent synaptic modulation may involve presynaptic NMDAR activation [(12⇓ ⇓ –15); for review: reference. 16 and 17]. In fact, the cell-specific deletion of NMDARs in GC eliminates LTP in young rodents (12). In addition to NMDARs, PF-PC synaptic plasticity also requires nitric oxide (NO) signals (18⇓-20). Since nitric oxide synthase (NOS) is expressed in GC but not in PC (21), the activation of presynaptic NMDAR may allow Ca2+ influx, thereby activating NO synthesis, which in turn may act on PC. However, in the mature cerebellum, the existence of presynaptic NMDAR on PF and the role of NO in PF-PC plasticity remain controversial. Previously, we have proposed that activation of putative presynaptic NMDARs in young rodents is necessary to induce PF-PC synaptic plasticity without affecting transmitter release (8, 9, 11, 12). Recently, it has been shown that a part of PF expresses presynaptic NMDAR containing the GluN2A subunit, and that these receptors are functional (11, 12). Therefore, contrary to their role in other synapses, at least in young rodents, presynaptic NMDAR, which is part of the PF-PC synapse, may induce post-synaptic plasticity by producing NO without changing neurotransmitters Release (9, 11, 12, 18⇓ ⇓ ⇓ –22). However, the causal relationship between the activation of NMDARs, NO synthesis, and the induction of synaptic plasticity in PFs is still missing.

In the cerebral cortex, the expression of presynaptic NMDAR is developmentally regulated (23, 24). However, little is known about the existence and function of presynaptic NMDAR in adult tissues. In the adult cerebellum, PC only expresses postsynaptic NMDAR at its synapses with climbing fibers (CF) (25). It has been proposed that the activation of these receptors may have heterosynaptic effects during PF-PC LTD. This mechanism may explain why the LTD in adults depends on NMDAR. According to this model, presynaptic NMDAR will be a transient feature of developmental tissues and is not necessary for inducing synaptic plasticity and motor learning in adult animals (25).

Here, we combine electron microscopy, two-photon calcium imaging, synaptic plasticity experiments, and behavioral measurements to show that presynaptic NMDARs are not developmentally regulated, but are necessary for adult cerebellar motor learning. We demonstrate that presynaptic NMDAR exists and functions in PF of mature rodents. By specifically deleting the NMDAR subunit GluN1 in PCs or presynaptic cells (GC), we demonstrated that NMDAR activation in GC plays a key role in bidirectional synaptic plasticity and vestibular ocular reflex (VOR) adaptation, test An important example of cerebellar motor learning (26⇓ –28). In contrast, NMDAR in PC is neither involved in PF-PC synaptic plasticity nor in cerebellar motor learning.

First, we determined the presence of presynaptic NMDAR in adult rodent PF by performing pre- and post-embedding electron microscopy immunohistochemistry using antibodies against GluN1 or GluN2 subunits. Pre-embedded immunoperoxidase staining revealed many GluN2 (Figure 1A) and GluN1 (Figure 1B) response profiles in PF varicose veins facing small postsynaptic elements. The latter has the cytological characteristics of PC dendritic spines (29). The pre-embedded immunogold labeling of GluN2 (Figure 1C) and GluN1 (Figure 1D) and distance measurement (Figure 1F) indicate that the particles are mainly observed at the edge of the presynaptic active zone (82.5% outside the active zone, 50 distances from the edge) Less than the percentage of the total at 125 nm). The same results were observed using the post-embedding technique (Figure 1E). Next, we tested whether PF varicose veins showed activity-related NMDAR-related calcium transients. In order to track the calcium dynamics in PF, we injected adeno-associated viruses carrying Floxed GCaMP6f and td-Tomato into the cerebellar vermis of α6-Cre mice; α6 is a specific promoter for cerebellar GC (30). 3 to 4 weeks after virus injection, td-Tomato fluorescence was observed in the GC somatic cell dendritic compartment of the GC layer and the PF axon in the molecular layer (Figure 1G). We used extracellular electrodes to stimulate PF in transverse slices of the cerebellum and measured calcium transients from 50 to 250 μm from the stimulated site. This distance ensures that there is no perturbation of presynaptic calcium dynamics (12). We stimulate the PF in bursts every 15 seconds (25 pulses at 200 Hz) to show the presence of functional NMDAR in adult PF. PFs stimulate calcium transients of variable amplitude between varicose veins. After reaching a stable baseline, we blocked NMDAR by mixing D-APV (150 μM) and buffered Zn2+ (300 nM). In the presence of these drugs, the average calcium transient amplitude decreased (Figure 1 HJ, 95.3 ± 1.1% of baseline, P = 7.5e-5, n = 107 putative buttons detected). After the drug was eluted, the calcium transient amplitude returned to the baseline value (98.2 ± 0.8% of baseline, P = 4e-3 compared with NMDAR blocker; P = 0.02 compared with control). Only in the presence of the NMDAR blocker D-APV, the calcium transient amplitude showed a skewed distribution (Figure 1J), which is consistent with the heterogeneous expression of NMDAR in these varicose veins reported in young animals (11, 12) . Therefore, in mature PF boutons, the presynaptic NMDAR is located at the periphery of the active area and may play a functional role.

NMDAR exists and functions in PF varicose veins. (AF) Electron immunohistochemistry showed the presence of GluN1 and GluN2 in the presynaptic sites of PF-PC synapses. NMDAR uses antibodies that recognize the GluN2 (A, C, and E) or GluN1 (B and D) subunits to mark the outline of the presynaptic to the dendritic spines. GluN2 (A) or GluN1 (B) subunit immunoperoxidase deposits were detected in presynaptic elements of asymmetric synapses. Presynaptic particles related to the GluN2 (C) or GluN1 (D) epitope (arrow) are detected at the edge of the active zone (arrow) after the pre-embedded immunogold label. (E) Through post-embedding, immunogold-labeled GluN2-related particles were also found on the edge of the presynaptic active area, which indicates that the acquisition of epitopes is not restricted by the cytoskeleton of presynaptic differentiation. (F) Gold particle quantification of the distance from GluN2 to the edge of the active zone: The value along the x-axis represents the distance between the edge of the synaptic complex and the nearest immunogold marker (mean = 100.1 nm [red], median = 50 nm , SEM = 24.3, n = 41), negative values ​​indicate particles in the active area. Note: 82.5% of the gold particles are outside the active area (to the right of the dark vertical line). (GJ) Calcium imaging of PF varicose veins expressing GCaMP6f. (G) The acousto-optic deflector is based on the two-photon snapshot projection of the GC expressing td-Tomato (1 plane, 10 images). PFs are directly stimulated in the molecular layer (ml), and images are recorded at least 50 μm away from the stimulation point (gcl: granular cell layer, pcl: Purkinje cell layer). Zoom in: Examples of calcium images of varicose veins before (top), during (middle), and after (bottom) blocking of NMDAR (top to bottom, respectively). (H) Another example of calcium transients for marking varicose veins (25 to 30 PFs stimulation at 200 Hz). Baseline, APV and Zn2+ application, flushing and ([baseline]-[NMDAR blockade]) subtraction (purple dotted line). Note the effect of NMDAR blockade on calcium transients. (I) Normalized time course of ∆F/F under controlled conditions (n ​​= 107 varicose veins, 31 sections from n = 14 mice). From the 8th minute to the 28th minute, a total of 150 μM D-APV and 300 nM Zn2+ were used. (J) Normalized histogram of ∆F/F data. A box plot of the normalized data of the signal under the control condition, during the NMDAR block, and after flushing (blue). The average value (red dot) and median value (red line) are shown. The Wilcoxon test was used to test for statistical significance (*** P <0.001, ** P <0.01).

In young rodents, we have previously shown that PF-PC synaptic plasticity requires high-frequency burst activation of PF to recruit presynaptic NMDAR (11, 12). In addition, we have previously demonstrated that high-frequency bursts can also induce LTP in adult mice between 2 and 3 months of age (31). Therefore, using 5 stimulus bursts of 200 Hz per second (300 repetitions) of PF, we first induced a post-synaptic expression enhancement of 227 ± 19% compared to baseline (P = 4.4e-05; paired pulse ratio (PPR) ) = 98.8 ± 2.8% of baseline, P = 0.23; SI appendix, Figure S1 AD). In contrast, a burst of five stimuli with PFs at a frequency of 16.7 Hz (300 repetitions) per second did not result in a strong increase in excitatory postsynaptic current (EPSC) (125 ± 5% of baseline; P = 3e -04 compared with 200 Hz conditions), P = 0.09 compared with baseline, SI appendix, Figure S1 AD), as shown in LTD and LTP pups (References 11 and 12, respectively). 200 Hz LTP induction requires NMDAR activation, because enhancement is blocked in the presence of D-APV (111 ± 7% of baseline; P = 4e-04 compared to control, P = 0.12 compared to baseline) (SI appendix, Figure S1 Advertising). In order to determine the location of NMDARs involved in LTP induction, we crossed α6-Cre with GluN1-floxed animals [GC-GluN1 knock-out (ko)], Figure 2A, see Materials and Methods, and generated small cells lacking NMDARs in GC. mouse. Homozygous mice lacking the GluN1 allele are viable, and their cerebellum appears to be developing normally. The high-frequency burst stimulation of PF (5 200 Hz stimulations per second, 300 repetitions) showed that these animals had impaired PF-PC LTP (109 ± 7% vs. 200 ± 40%, GluN1 flox/ flox from control animals) Without Cre, GC-GluN1 wild type (wt), P = 3e-5; Figure 2 B and DF). Similarly, PF-PC LTD was also abolished in these animals (110 ± 3% vs. 70 ± 0.5% of the slices from the control animal GC-GluN1 wt, P = 0.0009; Figure 2 CF) using two stimulations at 200 hours PF bursts per second (300 repetitions) are paired with high-frequency CF bursts to induce LTD (reference 31, see Materials and Methods). These results indicate that NMDARs expressed by GC in adult animals are necessary for LTP and LTD induction of PF-PC synapses. To determine whether NMDAR expressed by PC also contributes to synaptic plasticity in mature mice, we crossed L7-Cre with GluN1 floxed mice (PC-GluN1 ko, Figure 2A) to generate mice lacking NMDAR in PC . L7 is a specific promoter for cerebellar PC (Ref. 32, see Materials and Methods). PF-PC LTP and control (206 ± 17% and 209 ± 26% PC-GluN1 wt slices of control animals, P = 0.56; Figure 2 B and GI) using PF's high-frequency burst stimulation can not be distinguished (5 times per second 200 Hz stimulation, 300 repetitions). Similarly, PF-PC LTD was not affected in the sections of these animals (70 ± 4% vs. 76 ± 2%, in the control animals GluN1 flox/flox does not contain Cre, PC-GluN1 wt, P = 0.36; Figure 2C and GI ) Two stimulus bursts paired with a high-frequency CF burst with a PF of 200 Hz (300 repetitions) per second are used to induce LTD (Ref. 31, see Materials and Methods). Therefore, the NMDAR required for LTP and LTD induction in the mature PF-PC synapse is located in the pre-synaptic GC, not in the post-synaptic PC.

Synaptic plasticity in animals that lack NMDAR in specific neuronal populations. (A) Schematic diagram of the strategy used to knock out the GluN1 gene in cerebellar GC (GC-GluN1, wt: blue; ko: green) or PC (PC-GluN1, wt: red; ko: black). (B and C) Representative recordings of LTP (B) and LTD (C) before (gray) and after (dashed line) sensing. The colored dots beside the trace are the same as in the DI. (D and E) The time course of normalized EPSC charge (top) and PPR (bottom) in GC-GluN1wt (blue) and GC-GluN1ko (green) of LTP (D) and LTD (E). (F) GC-GluN1wt (LTP, n = 8 cells; LTD, n = 7 cells) and all individual experiments in GC after plasticity induction (t = 30 to 35 minutes) normalized EPSC charge (top) and PPR (bottom)-GluN1ko (LTP, n = 10 cells; LTD, n = 6 cells). (G and H) The time course of normalized EPSC charge in PC-GluN1wt (red) and PC-GluN1ko (black) of LTP (G) and LTD (H). (I) Normalized EPSC charge (top) and PPR( Bottom) and PC-GluN1ko (LTP, n = 9 cells; LTD, n = 6 cells). PPR did not change after the induction of synaptic plasticity (GC-GluN1: DF, bottom; PC-GluN1: GI, bottom). The boxes represent the median (black), the upper and lower quartiles of the distribution. To induce LTP, we used 5 200 Hz stimuli per second with 300 repetitions, while using two 200 Hz stimuli per second (300 repetitions) paired with high-frequency CF bursts to induce LTD (see Materials and Methods). Wilcoxon test was used to test statistical significance (ns: P> 0.05; ** P <0.001).

Contrary to our results, Piochon and colleagues (25) observed that NMDAR blockade had no effect on LTP induction. Therefore, we tried to determine the experimental factors that caused this difference. In the immediate vicinity of the stimulation electrodes (up to tens of microns), pre-synaptic (and possibly postsynaptic) calcium dynamics may be disturbed (12). Since the stimulating electrode is usually located near the synapse recorded in the sagittal view, the requirement for presynaptic NMDAR activation can be bypassed under these conditions. In order to check for potential deviations in slice orientation, we tested the plasticity induction of sagittal slices. In fact, using the same conditions as Piochon and colleagues (25) (7 PF stimulations at 100 Hz, 300 repetitions at 1 Hz) and the LTP induction scheme we used in horizontal slices (5 times at 200 Hz per second) PF stimulation, 300 repetitions), we successfully induced LTP (142 ± 11% of baseline, P = 0.0002, SI appendix, Fig. S1 EG; and 149 ± 10% of baseline, P = 6e-4, SI appendix, Fig. S1 E and G, respectively). Compared with the horizontal slice orientation experiment, this form of LTP has not been abolished by the application of APV (148 ± 16% of baseline, P = 0.003, P = 0.95 and sagittal control conditions, SI appendix, Figure S1 EG; and 139 ± 14% of the baseline, P = 0.008, P = 0.43 with sagittal control conditions, SI Appendix, Figure S1 E and G, respectively). Therefore, we believe that the lack of NMDAR blockade reported by Piochon and colleagues (25) may be due to the direct perturbation of the presynaptic calcium dynamics at the PF terminal.

In order to link the requirement of high-frequency stimulation of PF to induce PF-PC plasticity with the dynamic characteristics of NMDAR, we studied the participation of presynaptic NMDAR containing GluN2A subunits in PF-PC synaptic plasticity. It has been shown that the presynaptic NMDAR of PF-PC synapses in young animals shows the pharmacological properties of GluN2A-containing receptors (11, 12, 33), which may determine receptor dynamics to a large extent (23, 24 , 34). To determine whether NMDAR containing GluN2A subunits is involved in LTP induction in adult mice, we studied LTP induction in global GluN2A-KO mice (35). In these mutants lacking the GluN2A subunit, our standard protocol failed to induce LTP: 5 PF stimulations at 200 Hz, 300 repetitions (106 ± 15% of baseline, P = 0.21, SI appendix, Figure S1 B And D), in contrast with their heavyweight litter (177 ± 18% of baseline). This result indicates that NMDAR containing GluN2A is required for the plasticity induction of adult PF-PC synapses. These data are consistent with earlier evidence in young mice (12), highlighting the sustained expression patterns of NMDAR subunits in PF-PC synapses during postnatal development and beyond. Therefore, the bidirectional synaptic plasticity of PF-PC in the mature cerebellum is driven by the NMDAR containing the GluN2A subunit.

Both the induction of LTP and LTD depend on the nitric oxide (NO) signal in the cerebellum of young mice (18⇓ –20, 36). However, this finding has not been confirmed in adult animals. To test the involvement of NO signaling in adult LTP, we applied our standard LTP induction protocol in the presence of L-NAME (100 μM), a specific NOS antagonist. Under these conditions, LTP was cancelled (101 ± 5% of baseline, P = 0.47; SI appendix, Figure S2 AC). NO synthesis during PF burst activity may be caused by the activation of presynaptic NMDAR (9, 11, 12) and/or the activation of receptors located in interneurons (10, 37). In order to understand the connection between NMDAR activation and NO synthesis and determine the cell types involved, we used electrochemical probes to measure NO production during high-frequency burst stimulation in the molecular layer (see Materials and Methods). We stimulated PF using the same protocol as the previous configuration used to image NMDAR-dependent calcium signals (Figure 1). This protocol resulted in a significant increase in the current recorded on the electrodes (SI Appendix, Figure S2 D and E; 112.8 ± 32.8 pA, P = 8e-5, GC-GluN1 wt). This signal is specific to NO production because it does not exist in the presence of L-NAME (SI Appendix, Figure S2 D and E; -33.5 ± 34.1 pA, P = 0.18). In addition, NO production is due to NMDAR activation, because it is also eliminated in the presence of APV (SI appendix, Figure S2 D and E; -68.1 ± 27.1 pA, P = 1.3e-4, APV and control). Finally, there is no signal in the slices of mice lacking GluN1 in the GC (SI appendix, Figure S2 D and E; -10.4 ± 27.2 pA, wt and ko, P = 0.003). Therefore, we conclude that the high frequency activity of PF leads to the activation of its NMDAR, and the subsequent production of NO is involved in the induction of plasticity in the molecular layer.

In order to test the specific contribution of NMDARs in GC and PC to basic reflex eye movements, we performed a cerebellar-dependent behavioral test involving compensatory eye movements in animals lacking NMDARs. It is known that vestibular and full-field visual input drive compensatory eye movements through VOR and optokinetic reflex (OKR), respectively (38, 39). Although VOR compensates for head movement through reverse eye movements, OKR allows the eyes to follow the moving field of view when the head is stationary. These reflections work together to produce a visually enhanced VOR (or VVOR), which can maintain a stable image on the retina when the animal moves in its environment. In order to maintain the best stability throughout the life process, VOR will be adjusted based on visual feedback, this process depends on the cerebellar cortex (3). Therefore, we use these different behavioral paradigms to evaluate the contribution of pre-synaptic and post-synaptic GluN1 receptors to athletic performance. We tested baseline OKR, VOR, and VVOR in wt, PC-GluN1 ko mice, and GC-GluN1 ko mice (SI appendix, Figure S3A). In PC-GluN1 ko mice, the eye movement caused by the sinusoidal rotation of the visual field (OKR) or the table (VOR) was compared with the control group (SI appendix, Fig. S3B, OKR and VOR: both P> 0.5). In VVOR, combining the two inputs, the gain is not affected (P = 0.55), but there is a small difference in phase (P = 0.010, Δphase = 1.1 ± 0.1° within the test frequency range). Similarly, removing GluN1 from the GC slightly affects compensatory eye movements (SI appendix, Figure S3C). The VOR phase of GC-GluN1 ko mice is lower (P = 0.043, Δ phase = 6.2 ± 2.2°), while all other parameters are not different from those of the control group (all P> 0.3), although the VVOR gain has a trend in GC -GluN1 ko is lower in mice (P = 0.056). In conclusion, removing GluN1 from PC or GC has little effect on the baseline performance of VOR and OKR, which indicates that NMDARs and NMDAR-dependent plasticity are not of high importance for basal cerebellar-related motor behavior.

Genetic interventions, especially those that target the components of the plasticity process, usually affect motor learning more than exercise performance (28, 40, 41). In order to study the role of GC and PC NMDAR in cerebellar motor learning, we first tested the contribution of GluN1 to short-term learning by subjecting mice to the VOR learning paradigm designed to reduce VOR gain. Five sinusoidal rotations of the mouse and visual input of the same amplitude (in-phase) in the same direction for 10 minutes resulted in a gradual decrease in the VOR gain of the control mice (recorded in the dark, P <0.001, compared with before) 50 minutes later ; Figure 3A and B). This type of short-term cerebellar learning was not affected in PC or GC-GluN1 ko mice (compared to the respective control, P> 0.14; Figure 3A and B). Next, we continued training with the same parameters on the second day (after the mice remained in the dark for 23 hours) to test the mice on the long-term learning paradigm, but the visual stimulation amplitude increased. This paradigm aims to reverse the direction of VOR (that is, the phase of VOR is adapted to 180°). After 3 days of reversal training, the VOR phase increased from a value of ≤30° to the average maximum value of 144 ± 15° in the control group and 132 ± 13° in PC-GluN1 ko mice (Figure 3A). This phase increase is comparable between groups (days 2, 3, and 4; all P> 0.3). Although the VOR gain increased again after the phase inversion, this gain increase was significantly less pronounced in PC-GluN1 ko mice (day 4, P = 0.011), indicating that there is a slight impairment in VOR learning (Figure 3A). In contrast, the effect on GC-GluN1 ko mice is more severe. These mice did not achieve the same phase shift as the control mice, and compared with 115 ± 16° (weight; Figure 3B), the average maximum was 58 ± 10° (GC-GluN1 ko). This difference was significant throughout the training period (P <0.015 on Days 2, 3, and 4). In addition to its role in VOR adaptation, the cerebellum is also very important for OKR adaptation. Long-term exposure to sinusoidal rotating visual input, combined with or without vestibular input conflict, will lead to an increase in the adaptability of OKR gain (40). The OKR gain after reversal training was significantly higher in all mice, but this increase was attenuated in GC-GluN1 ko mice (Figure 3C, P = 0.045), while it was not affected in PC-GluN1 ko mice. Loss (P = 0.27). There are obvious phenotypes in the cerebellar-dependent learning paradigm, VOR phase reversal, and the increase in OKR gain, confirming the functional relevance of NMDAR in GC in cerebellar-dependent learning.

Cerebellar-dependent VOR phase reversal is affected in mice lacking NMDAR in GC. (A, top) Schematic diagram of the VOR phase reversal experiment. Mice receive a mismatched combination of vestibular and visual input. The VOR phase reversal is caused by five 10-minute training sessions, during which visual stimulation and vestibular input (amplitude, 5°) increase in increments (day 1, 5°; day 2, 7.5°; From day 3 to day 3 to 4, 10°). (Bottom) VOR phase reversal training usually starts with a decrease in VOR gain (day 1), then an increase in VOR phase (for a phase value of 180°, day 2 to day 4), and when the VOR phase is sufficient Conversely, the VOR gain increased (day 4). PC-GluN1ko mice showed a phase reversal of VOR, and its phase value was similar to that of the control (both n = 12 mice), but the increase in VOR gain was impaired after the reversal (day 4, P = 0.011). (B) Mice lacking NMDAR from GC have more obvious defects because the phase increases more slowly (days 2, 3, and 4: P = 0.001, 0.013, and 0.006, respectively). Please note that the control litter piglets of GC-GluN1 mutant mice seem to be slower to adapt to VOR than the litter piglets of PC-GluN1 mutant mice, presumably due to subtle batch and inter-experimental differences. (C) VOR phase reversal training can also increase the gain of OKR. This cerebellum-dependent adaptive change was also impaired in mice lacking NMDAR in GC (n = 11 mice) (compared to 7 control mice, P = 0.045), but in PC-GluN1ko mice None (all n = 11 mice, P = 0.27). (D) VOR gain consolidation, day 2 (b) relative to day 1 (a) learning response percentage (see illustration), unaffected in PC-GluN1 mice (P = 0.38, both n = 12 mice, unpaired Student's t-test). In contrast, compared with the control group, the impaired long-term adaptation may be related to the consolidation defect of GC-GluN1ko mice (P = 0.011, n = 11 mice, unpaired Student's t-test). (D) Error bars indicate SEM (please note that the error bars may fall within the symbol), *P <0.05, **P <0.001.

Overnight consolidation, the stabilization of acute adaptive changes to long-term changes, is essential for sustained exercise adaptation (42, 43). To reveal whether the GC-GluN1 phenotype is related to the ability to adapt during training or the ability to store and maintain adaptation overnight, we tested the consolidation of adaptive changes (44). Since this measurement is relatively sensitive to noise, we compared the most stable and consistent change: the difference between the VOR gain reduction on the first day and the second day. The consolidation of GC-GluN1 ko mice, that is, the percentage of changes that lasted overnight is impaired (Figure 3D, 39 ± 8% vs. 71 ± 8% in the control group, P = 0.011), which means that at least part of the defect can be explained by the impairment merge. Not surprisingly, the consolidation of PC-GluN1 ko mice was not impaired (Figure 3D, 41 ± 6% vs. 52 ± 7% in the control group, P = 0.38). This selective effect on integration underscores the long-term effects of the GluN1-dependent mechanism assumed above.

Here, we present immunohistochemical and functional evidence that presynaptic NMDAR is present in the PF-PC synapse of the adult rodent cerebellum. Like young rodents, these receptors require two forms of postsynaptic plasticity that induce synaptic plasticity, LTP and LTD. Using cell-specific NMDAR deletions in GC or PC, we demonstrated that only GC NMDAR can effectively participate in PF-PC synaptic plasticity and VOR adaptation. Contrary to previous theories, our results support the view that these receptors are continuously expressed in the adult cerebellum, and the role of PC NMDARs in PF-PC synaptic plasticity and cerebellar learning should be reconsidered.

Presynaptic receptors are usually located in the active zone, where they regulate synaptic transmission and plasticity. This phenomenon can be seen in various types of presynaptic receptors (13⇓ –15) throughout the brain, including GABA receptors (45), NMDARs (16, 17) and kainate receptors ( 46), all of these are found at the end of the PF. 47). However, the presynaptic NMDARs of cerebellar PF-PC synapses control plasticity in a different way: these receptors participate in the induction of plasticity by promoting NO synthesis, while transmitter release remains unchanged (9, 11, 12, 37). This is in sharp contrast to other central synapses such as the hippocampus (48, 49) or cerebral cortex (49, 50). How the activation of calcium-permeable presynaptic receptors at the synapses of cerebellar PF-PC can lead to calcium-dependent signal transduction (NO synthesis) without interfering with transmitter release has always been a problem, leading to suspicion of such receptors The presence. Here, we propose an explanation for this paradox. Immunogold staining performed in adult rats showed that most of the presynaptic NMDAR is not in the active area, but in its periphery, so it can be in an active state without affecting the calcium dynamics at the release site. For comparison, this location exceeds the 30 nm coupling distance between the calcium influx through the voltage-dependent calcium channel and the release sensor (51). Although our immunogold staining was obtained from adult rats, these results are not only consistent with our mouse experiments using various other techniques, but also with several studies reporting PF-PC synapses in mice and rats The probability of no release after plasticity induction changes uniformly (8⇓ ⇓ ⇓ –12, 18⇓ –20, 23, 25, 31, 36⇓ ⇓ –39, 52⇓ ⇓ ⇓ –56). Therefore, it is likely that a similar distribution of NMDARs was observed in the periphery of PFs boutons in rats and mice.

As demonstrated in young rodents, we show that PF-PC synaptic plasticity does not depend on it. Although NMDARs-dependent NO synthesis has been shown in young rodents, the origin of NO is still a controversial issue (9⇓ ⇓ –12, 33, 37). Here, using the specific absence of NMDAR in PF, we demonstrate that NMDAR activation leads to NO production, which suggests that it is impossible to recruit indirect NMDAR-dependent signaling processes from other cell types. This result indicates that there is a close coupling between NMDAR and NOS, which can minimize the crosstalk between the entry of Ca2+ released by the transmitter and the Ca2+ signal induced by plasticity through the NMDAR-NOS system. These results are also consistent with the involvement of NO in PF-PC LTP described in this study. Previous in vitro studies reported the role of NO in LTD in adult rodents (57, 58), as well as visual movement showing learning disabilities Behavioral research in NOS knockout mice (59).

We have previously described that presynaptic NMDAR is required to induce LTD (11) and LTP (12) at PF-PC synapses in cerebellar slices from young rodents. Our data indicate that the induction of LTP and LTD in adult animals still depends on NMDAR. These results are different from previous reports, which indicated that other cell types play a role in PF-PC synaptic plasticity (18, 25). Using cell-specific deletion of GluN1 subunit in GC (α6-Cre mice) or PC (L7-Cre mice), we show that only NMDAR expressed in GC, not NMDAR expressed in PC, are involved LTP and LTD at the PF-PC synapse. We propose two explanations for this difference. First, in the parasagittal slice configuration used in other studies (18, 25), the proximity of extracellular stimulation electrodes to presynaptic elements may disrupt calcium dynamics, which may bypass the requirement for PF to activate NMDAR (12 ). Consistently, we also found that activation of NMDARs does not require the induction of PF-PC LTP in the sagittal direction (SI appendix, Figure S1). Interestingly, using the specific deletion of the GluN1 subunit in adult mouse PC, Kono and colleagues also reported that PF-PC LTD is not related to NMDAR expressed in PC (18). Instead, they proposed that NMDAR located on the molecular layer interneurons instead of GC is needed to induce PF-PC LTD (18). Although the NMDAR subunit profile is inconsistent with the role of molecular layer interneurons (MLI) in PF-PC plasticity in young rodents (11, 12, 33), it remains to be determined whether MLI may also be involved in PF-PC synapses in adulthood. Plasticity in rodents using more physiological experimental conditions. Second, CF in the body is excited explosively at a high frequency (up to 400 Hz: References 60 and 61), which is much higher than the stimulation frequency used by Piochon and colleagues (25). We used a high frequency burst of CF to induce LTD 100 ms after the PF burst, as previously described in vitro (31, 52) and in vivo (53, 61). This burst parameter may be critical because the propagation of calcium spikes in PC dendritic trees depends on the excitability of the membrane (54). Therefore, the mild hyperpolarization of dendrites due to the blockade of CF-NMDARs may reduce the probability of propagation, thereby impairing the induction of LTD (12, 62). This phenomenon has potential physiological significance because it may attribute the subtle role of dendritic excitability to the NMDAR present in the CF-PC synapse. Finally, although the experimental conditions such as age, recording temperature, slice orientation, and plasticity induction protocol vary greatly in various cerebellar studies, many of them reported relatively low PF-PC LTP amplitudes (~10% to 20%) ( 38, 63, 64). Since this LTP is 5 to 10 times smaller than our LTP under control conditions, we can also imagine that a small part of our LTP does not depend on NMDAR. In conclusion, pre-synaptic rather than post-synaptic, NMDARs are essential for inducing bidirectional synaptic plasticity in adult mouse PF-PC synapses.

The powerful cellular effects on synaptic plasticity that we described in vitro raise the question of the extent to which NMDAR ablation affects motor behavior. Here, we proved that 1) ablation of NMDARs from GC and PC does not significantly affect motor performance, and 2) consistent with the situation that has no effect on the synaptic plasticity of PF-PC, ablation of NMDARs from GCs, not PCs Medium ablation impairs VOR adaptation, and 3) the joint loss of LTD and LTP weakens, but does not completely prevent cerebellar learning. By revealing these behavioral phenotypes in GC-specific and PC-specific GluN mutants, respectively targeting two connected structures, our data extends the data obtained in global mouse mutants. In fact, it was also found that global deletion of GluN2A would weaken but not prevent VOR phase reversal learning (40), but this study did not allow any explanation for the brain regions or cell types involved in the learning process. In addition, explaining the effect on exercise performance also proves the additional value of using cell-specific mutants. Although we were able to prove in our GC and PC specific mutants that NMDARs play a small role in athletic performance, early global studies have found that double GluN2A/C knockout has serious athletic performance defects in the rotating rod test ( 55), may lead to misleading assumptions. Using a conditional mouse model that deletes GluN1 from neurons expressing paralbumin (PV) (including PC and MLI), Kono and colleagues (18) observed the defect of increased OKR gain, which was deleted only from PC GluN1 is not present in mice. However, it should be noted that deleting GluN1 from neurons that express PV may also affect other circuits (56, 65, 66), including, for example, the visual cortex, which is also known to be required for OKR plasticity (67). In the same study, deleting GluN1 from the GC did not cause significant defects on the training day (18), which is consistent with the absence of phenotypes in GC-GluN1 mutant mice on the first day of training. learn. As we have discovered in our current research, long-term training is impaired, and expanding the training paradigm may expose flaws. This defect in long-term learning can partly be explained by the defect of consolidation early in training, although compared with PC-GluN1 control littermates, GC-GluN1 control mice have no strong effect on the following days and are relatively high The consolidation indicates that other factors can also lead to deficits.

Although cell-specific mutants are not limited by the classical global mutants because they allow interpretation at the level of brain regions and cell types, they are also limited when it comes to revealing correlations at the subcellular level. Since our cell-specific operation in the GC-specific mouse line will ablate all NMDAR expressed in GC, the effect on its display of cerebellar motor learning ability may not only be caused by its axon (ie, PF terminal) defect, And partly because there is no NMDAR at the level of their dendrites. In fact, the connection between mossy fibers (MF) and GC dendrites is glutamatergic and may also undergo NMDAR-dependent LTP (68). Therefore, the learning deficits in the cerebellum that we observed in the study of α6Cre-GluN1 mice may in principle also be related to the impaired induction of GC LTP by MF. The exact contribution of these two forms of synaptic plasticity (ie upstream, NMDA-dependent MF plasticity to GC and downstream PF plasticity to PC) to motor learning deficits can only be determined at the subcellular level by techniques that allow specific deletions, respectively Targets dendrites and axons. Since such techniques are currently not available, we cannot reliably estimate the individual contribution of these two cellular processes to the learning behavior of the cerebellum. Even so, we still want to highlight other arguments why the presynaptic NMDAR of the PF-PC synapse may at least help the cerebellum to learn. First, the loss of GluN1 in GC not only significantly affects LTD, but also significantly affects LTP at the synapse of PF-PC. Second, the behavioral phenotype observed in mutants with global deletion of the GluN2 subunit (40, 68) may also be due to similar effects on PF-PC synaptic plasticity, as we found that GluN2A is also a PF-PC synaptic LTP Necessary. SI appendix, Figure S1 B and D). Third, VOR adaptation may be mainly driven by simple spike activity and increase in LTP (69), which is consistent with the finding that VOR adaptation is affected by the loss of LTP selectivity to a similar degree (31, 70) and is not affected by selectivity alone. Destroy LTD (39, 71). Finally, although the presence and absence of CF input induces the plasticity of PF to PC synapses to form an ideal matrix for specific input-guided learning (66, 72), there is a lack of such input controlled by teaching activities in MF to GC. Specific level synapses.

In conclusion, we cannot rule out the potential contribution of MF-GC synaptic plasticity in VOR adaptation, as evidenced by the defects observed after GluN1 loss in GC, but we can exclude NMDARs in this form of cerebellar motor learning. The important role of expression in PC. In addition, the current research provides a lot of supporting evidence for the contribution of NMDA-dependent PF to PC synaptic plasticity in adult VOR learning, emphasizing that GC NMDAR may at least partially express NMDAR to mediate its PF input into The synaptic plasticity of PC dendrites provides a variety of presynaptic mechanisms and provides sufficient means to fine-tune the output of the cerebellum for a long time (47, 72).

The cell specificity was verified by crossing L7Cre and BACα6Cre mice with a floxed fluorescent reporter mouse line. Please refer to the SI appendix, SI materials and methods for detailed information.

Use specific anti-GluN1 mouse monoclonal antibody and anti-GluN2A rabbit polyclonal antibody for pre-intercalation immunoperoxidase or immunogold and post-intercalation immunogold detection. Please refer to the SI appendix, SI materials and methods for detailed information.

Whole-cell patch clamp recording in cerebellar slices from C57Bl6 mice (8 to 20 weeks old) is the same as the reference. 31. Detailed information is provided in the SI Appendix, SI Materials and Methods.

A NO-selective amperometric microprobe (World Precision Instruments, WPI) with a 7-micron tip was used to monitor the NO outflow of cerebellar slices, similar to previous work (37). Use an Axon Multiclamp 700B amplifier to apply a constant voltage of 0.9 V. The probe is cleaned regularly in 0.1 M H2SO4 to remove debris. The microprobe is at least 30 μm away from the stimulation electrode to avoid false Ca-NO signals (12). Before use, the probe is equilibrated in the culture medium for 30 minutes.

For the experiment, C57BL/6 Bac6-cre mice were used for 2 to 3 months, in which the recombinase was expressed only in the cerebellar GC. A mixture of curved GCaMP6f and td-tomato adeno-associated virus was injected stereotactically into the cerebellar cortex. Please refer to the SI appendix, SI materials and methods for detailed information.

All experiments were performed using a customized random access two-photon laser scanning microscope. Detailed information is provided in the SI appendix, SI materials and methods.

At least 5 days after preparation for surgery, the heads of adult male mutant mice and control mice are fixed in the middle of a turntable with a surrounding screen. Compensatory eye movements are caused by rotational vision and/or vestibular input, and cerebellar motor learning is tested by the mismatch of these two inputs. The eye movement recorded in the video is calibrated, and the gain and phase, which represent the magnitude and time of the movement, respectively, are determined. Detailed information is provided in the SI appendix, SI materials and methods.

The data set/code generated in the current research can be provided upon reasonable request.

This work was supported by the French government’s "Investissements d'Avenir" program, which was implemented by Agence Nationale de la Recherche. References: ANR-10-LABX-54 MEMOLIFE, ANR-11-IDEX-0001-02 PSL * Research University. MS is funded by Nederlandse Organisatie voor Wetenschappelijk Onderzoek (NWO) Aarden Levenswetenschappen (ALW-Veni) and the European Research Council Startup Grant (ERC-Stg). CIDZ has received FP7-C7 European Commission, ZonMw, NWO-Exacte en Natuurwetenschappen (ENW-Klein), European Research Council (Advanced Grant and Proof of Concept Grant), Medical NeuroDelta Project, Topsector Life Science and Health (Social Innovation Neurotechnology) Support) and Albinism Vriendenfonds Netherlands Institute of Neuroscience. GB is funded by Région Ile de France, Fondation pour la Recherche Medicale (FRM) and Labex MEMOLIFE. AEG is supported by the National Science Foundation (NSF) Graduate Research Fellowship Program (GRFP) and the Discovery Researcher Program at the University of California, San Francisco. The funder had no role in research design, data collection and analysis, publication decision or manuscript preparation. We thank R. de Avila Freire and L. Post for their technical assistance, and thank B. Barbour, RS Larsen, PL Reeson, E. Jones, M. Mukundan and C. Wang for their comments on the manuscript.

↵1 Current address: Department of Neurobiology, Harvard Medical School, Boston, Massachusetts 02115.

↵2 Current address: HHMI, Harvard Medical School, Boston, Massachusetts 02115.

↵3 Current address: Department of Physiology, University of California, San Francisco, CA 94143.

↵4 Current address: HHMI, University of California, San Francisco, CA 94143-2811.

Author contributions: MS, MC, CIDZ and GB design research; MS, AEG, PR, CM-H., AA, MC and GB research; AT contributed new reagents/analysis tools; MS, AEG, MC and GB analyzed the data; MS, AEG, ABN, MC, CIDZ and GB wrote this paper.

The author declares no competing interests.

This article is directly contributed by PNAS.

This article contains online support information at https://www.pnas.org/lookup/suppl/doi:10.1073/pnas.2102635118/-/DCSupplemental.

This open access article is distributed under Creative Commons Attribution 4.0 (CC BY).

Thank you for your interest in advertising on PNAS.

Note: We only ask you to provide your email address so that the people you recommend the page to know that you want them to see it and that it is not spam. We do not capture any email addresses.

Feedback privacy/legal

Copyright © 2021 National Academy of Sciences. Online ISSN 1091-6490. PNAS is a partner of CHORUS, COPE, CrossRef, ORCID and Research4Life.