cAMP−EPAC−PKCε−RIM1α signaling regulates presynaptic long-term potentiation and motor learning

The cerebellum is involved in learning of fine motor skills, yet whether presynaptic plasticity contributes to such learning remains elusive. Here, we report that the EPAC-PKCε module has a critical role in a presynaptic form of long-term potentiation in the cerebellum and motor behavior in mice. Presynaptic cAMP−EPAC−PKCε signaling cascade induces a previously unidentified threonine phosphorylation of RIM1α, and thereby initiates the assembly of the Rab3A−RIM1α−Munc13-1 tripartite complex that facilitates docking and release of synaptic vesicles. Granule cell-specific blocking of EPAC−PKCε signaling abolishes presynaptic long-term potentiation at the parallel fiber to Purkinje cell synapses and impairs basic performance and learning of cerebellar motor behavior. These results unveil a functional relevance of presynaptic plasticity that is regulated through a novel signaling cascade, thereby enriching the spectrum of cerebellar learning mechanisms.


Editor's evaluation
The cerebellum plays a critical role in motor learning, but exactly which forms of synaptic plasticity contribute to learning and the underlying molecular mechanisms remain poorly understood.In this study, Wang and colleagues show that presynaptic long-term potentiation at the parallel fiber to Purkinje cell synapse is required for one form of motor learning, and involves a previously-unknown signaling cascade, where EPAC activation leads to PKCε-dependent threonine phosphorylation of RIM1α.The evidence is compelling and convincing.This study provides fundamental and new insights into the underlying mechanisms and functional consequences of presynaptic LTP.

Introduction
The cerebellum has historically been viewed as a motor coordination center (Ito, 2005).Recent evidence implicates that the cerebellum is also involved in a variety of learning-dependent high-level behaviors, including motor precision (Wagner and Luo, 2020;De Zeeuw, 2021) as well as cognitive and emotional functions (Schmahmann et al., 2019).The unique capability of the cerebellum to govern fine-tuned motor and cognitive skills at a high temporal resolution critically depends on delicate coordination of multiple forms of plasticity (De Zeeuw, 2021).Indeed, recent studies indicate that, in addition to the renowned postsynaptic long-term depression (LTD) (Ito, 2005) and long-term potentiation (LTP) (Schonewille et al., 2010), other forms of synaptic or non-synaptic plasticity may also contribute to cerebellar motor learning (Raymond and Medina, 2018;De Zeeuw, 2021).Relatively speaking, the molecular underpinnings of presynaptic plasticity in the cerebellar cortex are less understood (Wang et al., 2021), although early studies have shown that presynaptic Ca influx, Ca-sensitive adenylate cyclase, and cyclic adenosine monophosphate (cAMP) production are required for presynaptic LTP (Byrne and Kandel, 1996;Salin et al., 1996;Storm et al., 1998).Moreover, the function of presynaptic plasticity on cerebellar motor learning remains to be elucidated (Le Guen and De Zeeuw, 2010;De Zeeuw, 2021), although it was suggested that adenylyl cyclase-dependent LTP participates in rotarod learning (Storm et al., 1998).
In particular, the function of cAMP-dependent protein kinase A (PKA) on transmission release has been the subject of debate.Lonart et al., 2003 found that RIM1α-Ser413 is phosphorylated by PKA, which is required for presynaptic LTP.However, the mice with dysfunctional RIM1α-Ser413 mutation exhibit normal presynaptic LTP in the cerebellum and the hippocampus (Kaeser et al., 2008;Yang and Calakos, 2010), questioning the role of RIM1α-Ser413 and PKA in presynaptic LTP.Thus, how RIM1α is activated during presynaptic plasticity needs to be revisited.
In this study, we identified a new presynaptic signaling module that comprises EPAC (exchange protein directly activated by cAMP) and PKCε (epsilon isozyme of protein kinase C).This signaling module controls threonine phosphorylation of RIM1α, initiates the assembly of a Rab3A-RIM1α-Munc13-1 tripartite complex, and thereby facilitates docking and release of synaptic vesicles at parallel fiber (PF) to Purkinje cell (PC) synapses, which is in line with previous work (Martín et al., 2020) showing β-adrenergic receptors/EPAC signaling modulates PF release using EPAC2 knockout mice.Importantly, presynaptic ablation of either EPAC or PKCε is sufficient to inhibit presynaptic LTP and impair motor performance and motor learning.These data unveil a new signaling cascade governing presynaptic LTP and demonstrate that presynaptic plasticity is essential to cerebellar motor learning.
Finally, several lines of evidence demonstrated the causal relationship between EPAC and PKCε on the phosphorylation of RIM1α.First, we applied 8-pCPT alone or with εV1-2 (a selective PKCε inhibitor) to WT synaptosomes.The addition of εV1-2 to the synaptosomes strongly attenuated RIM1 p-Thr induced by 8-pCPT (Figure 1-figure supplement 4D).In contrast, RIM1 p-Thr was not affected by co-application of Gӧ6976, a PKCα/β inhibitor (Figure 1-figure supplement 4D).Second, we administered phorbol 12-myristate 13-acetate (PMA), an activator of all PKC isoforms, alone or along with εV1-2 or Gӧ6976, so as to inhibit PKCε or PKCα/β, respectively.εV1-2, but not Gӧ6976, significantly suppressed RIM1α p-Thr in the synaptosomes (Figure 1-figure supplement 4E).Third, RIM1 phosphorylation was examined in Prkce-cKO synaptosomes, which were treated with either control saline or 8-pCPT.In this scenario, neither p-Thr nor p-Ser of RIM1 was changed (Figure 1L).Overall, these data strongly indicate that EPAC can trigger RIM1α p-Thr phosphorylation and that this activation requires PKCε.
EPAC-PKCε module is critical to vesicle docking and presynaptic release through acting on the Rab3A-RIM1α-Munc13-1 complex Our finding that the EPAC-PKCε module regulates RIM1 activity through phosphorylation leads to an interesting question: whether the EPAC-PKCε module functions on synaptic formation and function through acting on RIM1, which is known to be critical to organization of the presynaptic active zone and neurotransmitter release (Schoch et al., 2002;Han et al., 2011;Kaeser et al., 2011;Acuna et al., 2016;Persoon et al., 2019).
We next examined the effect of the ablation of EPAC or PKCε on synaptic transmission.Miniature excitatory synaptic currents (mEPSCs) at PF-PC synapses were recorded in cerebellar slices from Atoh1 Cre ;Rapgef3 f/f ;Rapgef4 f/f (Rapgef3;Rapgef4-cKO) and Prkce-cKO mice, the former of which caused specific deletion of Rapgef3 and Rapgef4 in granule cells (Figure 1-figure supplement 3I-L), while Atoh1 Cre and Prkce f/f mice were used as corresponding controls.We found that mEPSC frequency was reduced in PCs from Rapgef3;Rapgef4-cKO mice compared to PCs from Atoh1 Cre mice, whereas mean amplitude did not differ between two genotypes (Figure 2C).Similarly, the frequency but not the amplitude of mEPSCs was significantly lower in Prkce-cKO mice than corresponding Prkce f/f mice (Figure 2D).A decrease in mEPSC frequency may be due to a reduction in release probability (Pr).To determine if Pr is affected following deletion of presynaptic EPAC and PKCε, we used a repeated stimulation protocol to estimate the readily releasable pool (RRP) as well as Pr (Thanawala and Regehr, 2016;He et al., 2019).Compared to Atoh1 Cre and Prkce f/f mice, repeated stimulation (100 Hz) revealed significant reductions in Pr in Rapgef3;Rapgef4-cKO (Figure 2E) and Prkce-cKO mice (Figure 2F).Furthermore, we examined the evoked PF-PC EPSCs with different stimulation intensities (3-15 μA) in control and mutant mice.Our results showed that presynaptic deletion of either EPAC1/ EPAC2 or PKCε significantly decreased evoked EPSCs in response to all stimuli (Figure 2-figure supplement 1).These recordings, together with the EM experiment (Figure 2A and B), indicate that EPAC-PKCε module is important to presynaptic transmitter release at PF-PC synapses.
We continued to explore how exactly the EPAC-PKCε module modulates synaptic release.An essential process during neurotransmitter release is that Rab3A, RIM1α and Munc13-1 form a tripartite complex and act in concert to dock synaptic vesicles to a release-competent state (Betz et al., 2001;Wang et al., 2001;Dulubova et al., 2005).Thus, we investigated whether the EPAC-PKCε module acts on the Rab3A-RIM1α-Munc13-1 complex.By measuring the ratios of IP/input in co-IP assay of synaptosome extracts, we found that both Munc13-1 and Rab3A had significantly weaker binding ability with RIM1α in both Rapgef3/4-dKO (Figure 2G) and Prkce-cKO (Figure 2H) synaptosomes, as compared to WT and Prkce f/f respectively.In contrast, neither EPAC nor PKCε ablation changed the expression levels of Rab3A and Munc13 (Figure 2G and H).These data indicate that the deficiency of either EPAC or PKCε impairs protein interactions in the Rab3A-RIM1α-Munc13-1 complex.
In another set of experiments, we studied whether the EPAC-PKCε module is sufficient to boost protein interactions in the Rab3A-RIM1α-Munc13-1 complex.First, we treated WT synaptosomes with 8-pCPT and εV1-2, and measured the amount of Munc13-1 and Rab3A precipitated with RIM1.The quantification showed a significant increment of precipitated Munc13-1 and Rab3A when synaptosomes were incubated with 8-pCPT (Figure 2I).Second, we measured the amounts of precipitated Munc13-1 and Rab3A in WT synaptosomes treated with FR236924, a selective activator of PKCε.

Presynaptic PF-PC LTP depends on EPAC and PKCε
Repetitive stimuli of PF terminals result in an increased Pr of neurotransmitters, leading to the expression of presynaptic LTP (Salin et al., 1996;Kimura et al., 1998;van Beugen et al., 2013;Hirano et al., 2016;Kaeser et al., 2008;Yang and Calakos, 2010;Martín et al., 2020).If the EPAC-PKCε module determines transmitter release through regulating the phosphorylation level of RIM1α, it is reasonable to hypothesize that this cascade controls presynaptic PF-PC LTP.
To test this hypothesis, presynaptic LTP at PF-PC synapses was induced by a tetanus stimulation (8 Hz for 5 min) at voltage-clamp mode (-70 mV) (Figure 3A).The potentiation of EPSCs reached 131 ± 6% of baseline in WT mice (t=38-40 min; n=13; p<0.001; Figure 3B and C), consistent with previous work (Salin et al., 1996;Kaeser et al., 2008).Concomitantly, paired-pulse facilitation (PPF) ratio decreased to 84 ± 4% (t=38-40 min; n=13; p<0.001; Figure 3C), indicating a presynaptic contribution to this form of LTP (Salin et al., 1996).Next, we preincubated WT slices with forskolin for 20 min to ensure the effect of forskolin.In this condition, the tetanus stimulation for presynaptic LTP failed to induce synaptic potentiation in PCs (Figure 3-figure supplement 1A and B), indicating that presynaptic LTP at PF-PC synapses occurs upon a rise in the cellular level of cAMP.

EPAC and PKCε mediate cAMP-triggered EPSC potentiation
cAMP is also required for presynaptic LTP induced by electrical stimulation (Salin et al., 1996;Le Guen and De Zeeuw, 2010), and its agonists are enough to produce a prominent increase in glutamate release (Weisskopf et al., 1994;Salin et al., 1996).Next, we wondered which downstream effector, EPAC or PKA (Cheng et al., 2008), is responsible for cAMP-induced potentiation.The role of PKA in presynaptic LTP has been contradicted by the studies showing that presynaptic LTP is intact when serine phosphorylation of RIM1 by PKA is interrupted (Kaeser et al., 2008;Yang and Calakos, 2010; also see Lonart et al., 2003).Moreover, Martín et al., 2020 showed that EPAC2 regulates synaptic release at PF synapses and is required for presynaptic PF-PC LTP.These findings inspired us to investigate whether perhaps the EPAC-PKCε module mediates cAMP-triggered EPSC potentiation.
We made whole-cell recordings from PCs and evoked PF-EPSCs every 30 s in Atoh1 Cre , Rapgef3;Rapgef4-cKO and Prkce-cKO mice.In Atoh1 Cre control mice, external application of forskolin produced a long-lasting elevation in PF-EPSC amplitude (Figure 4A and B), with a peak potentiation of 366 ± 25% (at 48-50 min; n=15; Figure 4C).In contrast, simultaneous ablation of EPAC1 and EPAC2 at presynaptic sites prominently affected the synaptic potentiation induced by forskolin application (162 ± 18% at 48-50 min; n=12; Figure 4A-C).Next, we incubated Rapgef3;Rapgef4-cKO PCs along with PKA antagonist KT5720 (3 μM) and again examined forskolin-induced EPSC potentiation.In this case, we found that combined blockade of EPAC and PKA completely eliminated the action of forskolin on EPSC potentiation (106 ± 4% at 48-50 min; n=12; Figure 4A-C).We continued to examine the effect of PKCε on cAMP-triggered EPSC potentiation using Prkce-cKO mice.Similar to Rapgef3;Rapgef4-cKO mice, the forskolin-induced potentiation in Prkce-cKO PCs was significantly attenuated (198 ± 5% at 48-50 min; n=12; Figure 4A-C).Again, the remaining potentiation was further blocked by the addition of KT5720 (101 ± 3% at 48-50 min; n=12; Figure 4A-C).The inhibitory effect of KT5720 on forskolin-induced potentiation was also examined by applying it alone in Atoh1 Cre PCs.We found that KT5720 inhibited the potentiation by 15%, a smaller effect than that     ).Thus, these results indicate that EPAC, PKCε and PKA all mediate cAMP-induced potentiation of transmitter release.In parallel with the observation of EPSC amplitude, PPF was monitored during the whole cell recordings.Forskolin application led to a significant reduction in PPF ratio of PF-EPSCs in Atoh1 Cre mice (Figure 4C).However, this reduction was significantly less when presynaptic of both types of EPAC as well as PKCε were ablated and KT5720 was added (Figure 4C).These results highlight that EPAC and PKCε function synergically on the synaptic release at PF-PC synapses.
We next assessed the impact of the EPAC-PKCε module on the strength of PF-EPSCs by directly applying EPAC agonist 8-pCPT.In line with previous work (Kaneko and Takahashi, 2004;Gekel and Neher, 2008), the administration of 8-pCPT was sufficient to potentiate PF-EPSCs by 179 ± 18% and reduce their PPF ratio by 17 ± 3% in WT PCs (n=6; at 18-20 min) (Figure 4D).Two lines of evidence confirm that the potentiation of PF-EPSCs by EPAC is mediated by PKCε.First, 8-pCPT-induced potentiation of PF-EPSCs was diminished in Prkce-cKO mice, as shown by unchanged PF-EPSCs and PPF (Figure 4E).Second, co-application of εV1-2 effectively prevented the 8-pCPT-induced synaptic potentiation and change in PPF (Figure 4F).
In summary, we conclude that EPAC-PKCε module and PKA are both downstream effectors of cAMP, but the EPAC-PKCε module plays the most prominent role in cAMP-triggered EPSC potentiation.

Presynaptic EPAC and PKCε are not involved in postsynaptic forms of plasticity
The mechanisms for postsynaptic LTP and LTD at PF-PC synapses can be complicated, in that they may depend not only on postsynaptic processes, but sometimes also on presynaptic events (Le Guen and De Zeeuw, 2010;Wang et al., 2014;Schonewille et al., 2021).For example, an endocannabinoidtriggered reduction of synaptic release is required by the induction of postsynaptic LTD (Kreitzer et al., 2002).As both EPAC and PKCε regulate Pr of PF-PC synapses, we wondered whether the EPAC-PKCε module also regulates postsynaptic LTP and LTD.
Next, we investigated whether the expression of postsynaptic PF-PC LTD is affected by ablation of EPAC and PKCε.PF-PC LTD was induced by giving repetitive PF stimulation at 100 Hz for 100ms paired with a depolarization of the PCs involved (Figure 6A; Steinberg et al., 2006;Zhou et al., 2015).As shown by example responses (Figure 6B), Rapgef3/4-dKO PCs showed robust PF-PC LTD (t=38-40 min: 59 ± 4% of baseline; n=13; Figure 6C), while the PPF ratio was not changed (p=0.26 at t=38-40 min; n=13; Figure 6D).Likewise, PF-PC LTD could be successfully induced in Atoh1 Cre and Rapgef3;Rapgef4-cKO mice (Figure 6E and F), while PPF was not altered (Figure 6G).Moreover, we found that the same protocol could induce PF-PC LTD in Prkce f/f and Prkce-cKO mice (Figure 6H and I) without affecting PPF (Figure 6J).
Overall, our results suggest that presynaptic EPAC and PKCε are not required for the induction of postsynaptic forms of LTP and LTD.

The EPAC-PKCε module is essential for motor performance and motor learning
Even though plastic changes in the granular layer of the cerebellum have been suggested to contribute to procedural memory formation (Le Guen and De Zeeuw, 2010), the evidence thus far is limited (Andreescu et al., 2011;Galliano et al., 2013).Therefore, we investigated whether the EPAC-PKCε module, which is critical to presynaptic PF-PC LTP, contributes to performance and adaptation of compensatory eye movements mediated by the vestibulo-cerebellum (Schonewille et al., 2010;Grasselli et al., 2020).
Basic performance parameters included amplitude (gain) and timing (phase) of the optokinetic response (OKR), vestibulo-ocular reflex (VOR), and visually enhanced VOR (VVOR) (Figure 7A).We found that basic motor performance was impaired in Rapgef3/4-dKO mice in that they showed significant deficits in the amplitude and timing of their OKR (p=0.009 and p=0.004, respectively; ANOVA for repeated measurements) and VOR (p=0.001 and p=0.02, respectively; ANOVA for repeated measurements) (Figure 7-figure supplement 1A and B).In contrast, no significant differences were observed in the VVOR (p=0.66 and p=0.68 for gain and phase values, respectively; Figure 7-figure supplement 1C).

Discussion
In the current study we demonstrate that triggering EPAC induces PKCε activation and threonine phosphorylation of RIM1α, which in turn facilitates the assembly of the Rab3A-RIM1α-Munc13-1 tripartite complex and thereby docking and release of synaptic vesicles at active zones of PF-PC synapses (Figure 7-figure supplement 2).The form of presynaptic LTP at these synapses that requires activation of the EPAC-PKCε module can be induced by either tetanic stimulation or forskolin at PF terminals (Figure 7-figure supplement 2).Via its presynaptic actions, the EPAC-PKCε module contributes to adaptation of compensatory eye movements, a motor learning task that depends on the vestibulo-cerebellum.

Distinct roles of EPAC and PKA at synapses
cAMP-mediated signaling pathways that are mediated by EPAC and PKA regulate a multitude of physiological and pathological processes (Cheng et al., 2008).EPAC shares homologous cAMP-binding domains with PKA, but also possesses domains absent in PKA, such as the Ras exchange motif, the Ras association domain, and the CDC25-homology domain (Cheng et al., 2008).Indeed, the specific domains endow EPAC and PKA with different and even opposite functions.For example, in contrast to PKA, EPAC can activate small GTPase Rap1 (de Rooij et al., 1998) and increase PKB phosphorylation (Mei et al., 2002).Our current work bolsters the differences, showing that EPAC can phosphorylate PKCε and RIM1α threonine sites at synapses.This highlights the question as to how EPAC and PKA operate in an integrated manner to control the net physiological effect of cAMP-signaling pathways at synapses.Some studies indicate that presynaptic potentiation depends predominantly on PKA (Salin et al., 1996;Linden and Ahn, 1999;Lev-Ram et al., 2002), whereas others advocate a more critical role for EPAC (Kaneko and Takahashi, 2004;Fernandes et al., 2015;Martín et al., 2020).Our results highlight that ablation of either EPAC or PKCε by itself is not sufficient to block forskolininduced synaptic potentiation, but that supplementing this with a blockage of PKA causes a complete blockage.These results demonstrate that EPAC and PKA can conjunctively regulate synaptic potentiation.Even so, our results clarify that the impact of EPAC on cAMP-induced EPSC potentiation is dominant, as it has the strongest contribution to the forskolin-induced increase of EPSC amplitude.Alternatively, PKA warrants a minimum level of potentiation that may be required under particular circumstances when EPAC is not active.The EPAC-PKCε module regulates synaptic release and is required for presynaptic LTP Our EM analysis shows that the number of docked vesicles at the PF terminals of Rapgef3/4-dKO and Prkce-cKO mutants is reduced, whereas the general structure of PF-PC synapses is unchanged.As the ablation of either EPAC or PKCε attenuated protein interactions in the Rab3A-RIM1α-Munc13-1 complex, which is required for the docking and priming of presynaptic vesicles (Schoch et al., 2002;Sudhof, 2004;Ferrero et al., 2013), the reduction in docked vesicles in Rapgef3/4-dKO and Prkce-cKO mice can be readily explained.In parallel with our observations at the ultrastructural level, we found that mice with presynaptic deletion of EPAC and PKCε displayed obvious defects in synaptic release at the electrophysiological level.Although early studies have shown that EPAC1 and EPAC2 are involved in synaptic release in the hippocampus and the cerebellum (Yang et al., 2012;Zhao et al., 2013), which was further strengthened by Martín et al., 2020, our finding that PKCε acts as the downstream effector of EPAC and regulates presynaptic release is novel.Furthermore, we demonstrate for the first time that presynaptic PKCε is required for presynaptic LTP at PF-PC synapses.These findings expand the repertoire of forms of PC plasticity that are driven by cAMP signaling.
The role of the cAMP-PKA cascade in presynaptic LTP has been extensively debated.Early studies claimed that PKA and RIM1α serine phosphorylation are critical for the induction of presynaptic LTP at PF-PC synapses (Salin et al., 1996;Lonart et al., 2003).However, this conclusion was challenged by follow-up studies, demonstrating that RIM1α-S413A mutant mice exhibit normal presynaptic LTP in both cerebellum and hippocampus (Kaeser et al., 2008;Yang and Calakos, 2010).In our opinion, a couple of caveats must be considered regarding the function of PKA in presynaptic LTP.First, cAMP analogs (Rp-8-CPT-cAMP-S and Sp-8CPT-cAMP-S) used in two studies advocating that PKA mediates presynaptic PF-PC LTP (Salin et al., 1996;Lonart et al., 2003) are able to regulate Rap1 (Roscioni et al., 2009), which is a direct substrate of EPAC (de Rooij et al., 1998).Therefore, these cAMP analogs may also act through the EPAC-PKCε module.Second, KT5720 at 10 μm, a concentration used by Lonart et al., 2003, can alter a range of protein kinases, including phosphorylase kinase, mitogen-activated protein kinase kinase, PKBα, glycogen synthase kinase 3β, as well as AMP-activated protein kinase (Brushia and Walsh, 1999;Davies et al., 2000;Murray, 2008).Thus, KT5720 at this concentration has numerous side-effects next to its ability to inhibit PKA.In contrast, our results derived from cell-specific mouse lines consistently converge on the concept that presynaptic PF-PC LTP depends on the EPAC-PKCε module.More specifically, our data demonstrate that repetitive 8 Hz PF stimulation increases the level of cAMP and consequently activates EPAC and PKCε, which in turn induces threonine phosphorylation of RIM1α, suggesting a phospho-switch machinery that can tune presynaptic PF-PC LTP.
Our finding that the EPAC-PKCε module is a central component for synaptic release and presynaptic LTP may not stand on its own.In fact, EPAC is involved in cellular processes like cell adhesion, cell-cell junction formation, exocytosis and secretion, cell differentiation, as well as cell proliferation (Cheng et al., 2008), while PKCε is necessary for sperm exocytosis in the testis (Lucchesi et al., 2016).Together, these lines of evidence suggest that the EPAC-PKCε module might be a widespread mechanism controlling not only synaptic release in nerve cells, but also granule secretion in endocrine or proliferating cells.In addition, Gutierrez-Castellanos et al., 2017 showed that EPAC may regulate GluA3 conductance in PCs, suggesting that postsynaptic EPAC or PKCε may regulate the conductance of AMPA receptor subunits, and thereby postsynaptic LTP or LTD at PF-PC synapses.

Role of presynaptic LTP in motor behavior
Many studies have explored the potential functional role of postsynaptic plasticity at PC synapses, in particular that of PF-PC LTP and PF-PC LTD (Gao et al., 2012;Raymond and Medina, 2018).The picture emerging from these studies is that postsynaptic PF-PC LTP and PF-PC LTD play an important role in forms of learning that are mediated by the so-called upbound and downbound modules (De Zeeuw, 2021).Whereas VOR adaptation is mainly mediated by upbound microzones in the vestibulocerebellum that increase the simple spike frequency during learning (Gutierrez-Castellanos et al., 2017;Voges et al., 2017), eyeblink conditioning is predominantly regulated by downbound microzones in lobule simplex that decrease simple spikes during learning (ten Brinke et al., 2015;Wu et al., 2019).Yet, what is the role of presynaptic LTP at PF-PC synapses?Even though it has been suggested more than a decade ago that the functional role of presynaptic plasticity at PF-PC synapses during learning can be expected to align with that of postsynaptic plasticity (Le Guen and De Zeeuw, 2010), evidence has been largely lacking.
Here, we found that Rapgef3;Rapgef4-cKO and Prkce-cKO mice, which showed reduced PF-PC transmission and lack presynaptic LTP, exhibit deficits in basic motor performance, in the form of an affected OKR and VOR, as well as in gain-decrease and phase reversal learning of their VOR (Figure 7-figure supplement 2).Similarly, presynaptic ablation of EPACs or PKCε results in altered gain and phase values of their OKR and VOR.Interestingly, the impairments in OKR and VOR caused by deletion of EPAC1/EPAC2 or PKCε in granule cells were similar to those caused by global deletion of EPAC.This finding raises the possibility that presynaptic EPAC is in fact more critical for basic motor performance than postsynaptic EPAC.This possibility is compatible with previous work showing that mice with a PC-specific deletion of GluA3, which leads to a lack of postsynaptic LTP mediated by EPAC, have hardly any significant deficit in basic motor performance (Gutierrez-Castellanos et al., 2017).By the same argument, the contribution of presynaptic LTP to phase reversal learning might be more in line with that of postsynaptic PF-PC LTP in that Rapgef3;Rapgef4-cKO and Prkce-cKO mice showed similar deficits as PC-specific GluA3 knockouts.The prediction that the impact of presynaptic plasticity at PF-PC synapses during learning operates in a synergistic fashion with that of postsynaptic plasticity (Le Guen and De Zeeuw, 2010), does in this respect hold.Two caveats should be considered in the present studies.First, Atoh1 Cre -induced deletion of EPAC or PKCε might affect the function of unipolar brush cells (UBCs), which are involved in cerebellar ataxias (Kreko-Pierce et al., 2020).However, we believe that the EPAC-PKCε module regulates VOR learning through presynaptic plasticity mechanism at PF-PC synapses rather than UBCs, in line with the observations in other granulecell-specific mutations (Galliano et al., 2013;Schonewille et al., 2021).Second, presynaptic PF-PC LTP was performed in the cerebellar vermis in the present work, whereas VOR learning generally requires PC activity in the flocculus.Unfortunately, we found that PC-EPSCs in the flocculus were not suitable to record PC plasticity because they were unstable.
Although we observed only a difference in phase at the end of VOR phase reversal training, it should be noted that the gain was different on multiple days in both Rapgef3;Rapgef4-cKO and Prkce-cKO mice compared to their controls.VOR phase reversal training subjects to multiple days of changing training stimuli to test different aspects of adaptation.The first aim is to decrease the gain, followed by an increase in phase.Once the phase has increased above 120 o , the gain will increase again (Wulff et al., 2009).Therefore, the initial decrease of gain followed by the late-stage increase presumably underlies the absence of differences in gain between control and mutant groups in days 4 and 5.This does not imply that presynaptic LTP is more essential for the phase than the gain, as VOR gain decrease is affected during the first day of training.

Animals
Original breeding pairs of Rapgef3/4-dKO and Atoh1 Cre mice were obtained from Youmin Lu (Huazhong University of Science and Technology, Wuhan, China) and Wei Mo (Xiamen University, Xiamen, China), respectively.Rapgef3 f/f , Rapgef4 f/f and Prkce f/f mice were made by us with the assistance of GemPharmatech (Soochow, Jiangsu, China) and Nanjing Biomedical Research Institute (Nanjing, Jiangsu, China).The resulting offspring were genotyped using PCR of genomic DNA.Mice were kept at the Experimental Animal Center of Zhejiang University under temperature-controlled condition on a 12:12 hr light/dark cycle.All experiments were performed blind to genotypes in agematched littermates of either sex.

Co-immunoprecipitation
After measuring protein concentration using the BCA assay, a tenth of lysis supernatant derived from synaptosomes or cultured cells was used for input and the remainder were incubated with anti-RIM1 or anti-HA antibody, which was precoupled to GammaBind Plus Sepharose at 5-10 μg antibody/1 ml beads for 3 hr.Proteins on the beads were extracted with 2×SDS sample buffer plus protease inhibitors and boiled for 5 min before western blot.

Electron microscopy
After anesthetic mice (P21) were transcardially perfused with saline and ice-cold fixative, brains were removed and stored at 4 °C for 2.5 hr in fixative.Sagittal slices of vermis (200 μm) were prepared and rectangular molecular layer sections from lobules IV-V were dissected.The samples were dehydrated and embedded in an epoxy resin.Ultrathin sections (90 nm) were cut using an ultra-microtome (Leica), stained with 2% uranyl acetate and lead solution, and mounted on grids.EM images were captured at ×30,000 magnification using a Tecnai transmission electron microscope (FEI, Hillsboro, OR).PF-PC synapses were identified by asymmetrical and short contacts, which were distinct from GABAergic or climbing fiber synapses (Ichikawa et al., 2016).ImageJ was used to count the numbers of total and docked vesicles per bouton.

Golgi staining and spine density analysis
Golgi staining was performed using Rapid Golgi Stain Kit (FD NeuroTech Inc, Ellicott, MD) according to the manufactory's instruction.PCs at the apical region were imaged using a bright field microscope (Zeiss, Germany).ImageJ was used to count the spine number and dendrites length of PCs with manual assistant.
PCs were visualized under an upright microscope (BX51, Olympus) equipped with a 40×waterimmersion objective and infrared differential interference contrast enhancement.Whole-cell recordings were made on PCs from lobules IV-V with a MultiClamp 700B amplifier (Molecular Devices).Currents were digitized at 10 kHz and filtered at 3 kHz.Patch electrodes (3-5 MΩ) were filled with an intracellular solution containing (in mM) 135 Cs-methanesulfonate, 10 CsCl, 10 HEPES, 0.2 EGTA, 4 Na 2 ATP, and 0.4 Na 3 GTP (pH 7.3, OSM 290).PCs were held at -70 mV to prevent spontaneous spikes that might escape clamp.For PF stimulation, standard patch pipettes were filled with aCSF and placed in middle third of molecular layer.Presynaptic PF-PC LTP was induced by stimulating PF input 120 times at 8 Hz (Salin et al., 1996;Kaeser et al., 2008).Postsynaptic PF-PC LTP was obtained when PFs were stimulated at 1 Hz for 5 min in parallel with current-clamp of recording PC (Wang et al., 2014).PF-LTD was induced by a conjunction of 5 PF-pulses at 100 Hz and a 100 ms long depolarization of PC to 0 mV, which was repeated 30 times with an interval of 2 s (Zhou et al., 2015).mEPSCs were recorded in whole-cell configuration in the presence of tetrodotoxin (0.5 μM) and an offline analysis was conducted using a sliding template algorithm (ClampFit 10, Molecular Device) according to previous work (Zhou et al., 2017).To estimate RRP and Pr, a repeated 100 Hz train stimulation protocol was used to evoke 50 EPSCs.RRP was calculated by linear interpolating the linear portion of the cumulative EPSC amplitude plot to virtual stimulus 0. Pr was calculated as the normalized 1st EPSC during the train stimulations divided by RRP (Thanawala and Regehr, 2016;He et al., 2019).A temperature controller was used to elevate aCSF temperature in the recording chamber (TC-344C; Warner Instruments, Holliston, MA).

Compensatory eye movement test
Mice (P60) were surgically prepared for head-restrained recordings of compensatory eye movements.A pedestal was attached to the skull after shaving and opening the skin overlaying it, using Optibond primer and adhesive (Kerr, Bioggio, Switzerland) and under isoflurane anesthesia in O2 (induction with 4% and maintained at 1.5% concentration).Mice were administered xylocaine and an injection with bupivacaine hydrochloride (2.5 mg/ml, bupivacaine actavis) to locally block sensation.The pedestal consisted of a brass holder (7×4 mm base plate) with a neodymium magnet (4×4 × 2 mm) and a screw hole for fixation.After a recovery period of at least 3 days, mice were placed in a mouse holder, using the magnet and a screw to fix the pedestal to a custom-made restrainer, and the mouse was placed with the head in the center on a turntable (diameter 60 cm) in the experimental setup.A drum (diameter 63 cm) surrounded the mouse during the experiment.The recording camera was calibrated by moving the camera left-right by 20° peak to peak at different light levels.Compensatory eye movement performance was examined by recording the OKR, VOR, and VVOR using a sinusoidal rotation of the drum in light (OKR), rotation of the table in the dark (VOR), or rotation of the table (VVOR) in the light.These reflexes were evoked by rotating the table and/or drum at 0.1-1 Hz (20-8 cycles, each recorded twice) with a fixed 5° amplitude.In order to evaluate motor learning, a mismatch between visual and vestibular input was used to adapt the VOR.The ability to perform VOR phase reversal was tested using a 5 day paradigm, consisting of six 5 minute training sessions every day with VOR recordings before, between, and after the training sessions.On the first day during training, the visual and vestibular stimuli rotated in phase at 0.6 Hz and at the same amplitude, inducing a decrease of gain.On the subsequent days, the drum amplitude was increased relative to the table and induced the phase reversal of the VOR, resulting in a compensatory eye movement in the same direction as the head rotation instead of the normal compensatory opposite direction (all days vestibular 5° rotation, visual day 2: 5°; day 3, 7.5°; days 4-5, 10°).Between recording sessions, mice were kept in the dark to avoid unlearning of the adapted responses.
Eye movements were recorded with a video-based eye-tracking system (hard-and software, ETL-200; ISCAN systems, Burlington, MA).Recordings were always taken from the left eye.The eye was illuminated during the experiments using two table-fixed infrared emitters (output 600 mW, dispersion angle 7°, peak wavelength 880 nm) and a third emitter that was mounted to the camera, aligned horizontally with the optical axis of the camera, which produced the tracked corneal reflection.Pupil size and corrected (with corneal reflection) vertical and horizontal pupil positions were determined by the ISCAN system, filtered (CyberAmp; Molecular Devices, San Jose, CA), digitized (CED, Cambridge, UK) and stored for offline analysis.All eye movement signals were calibrated, differentiated to obtain velocity signals, and high-pass-filtered to eliminate fast phases, and then cycles were averaged.
Gain-the ratio of eye movement amplitude to stimulus amplitude-and phase values-time difference between eye and stimulus expressed in degrees-of eye movements were calculated using custom-made MATLAB routines (The MathWorks, Natick, MA).

Statistical analysis
Experimenters who performed experiments and analyses were blinded to the genotypes until all data were integrated.Data were analyzed using Igor Pro 6.0 (Wavemetrics, Lake Oswego, OR), Graphpad Prism (San Diego, CA), SPSS 16.0 (IBM, Chicago, IL), and MATLAB.No statistical methods were used to pre-determine sample sizes, which were based on our previous studies.All data sets were tested for the assumptions of normality of distribution.No data were excluded except electrophysiological recordings with ≥15% variance in series resistance, input resistance, or holding current.Standard deviations for control were calculated from the average of all control data.Statistical differences were determined using unpaired or paired two-sided Student's t test for two-group comparison, or one-way ANOVA followed by Tukey's post hoc test for multiple comparisons, or repeated measures ANOVA for repeated measures.The accepted level of significance was p<0.05.'n' represents the number of preparations or cells.Data in the text and figures are presented as mean ± SEM.The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

Figure supplement 1 .
Figure supplement 1.The percentages of PF and CF synapses among cerebellar synaptosomes.

Figure supplement 2 .
Figure supplement 2. Threonine phosphorylation of RIM1 by EPAC in vitro.

Figure supplement 2
Figure supplement 2-source data 1.The uncut gel of western blots in Figure 1-figure supplement 2.

Figure
Figure supplement 3-source data 1.The uncut gel of western blots and PCR in Figure 1-figure supplement 3.

Figure
Figure supplement 4-source data 1.The uncut gel of western blots in Figure 1-figure supplement 4.

Figure 2 .
Figure 2. EPAC and PKCε act on vesicle docking, synaptic release, and Rab3-RIM1-Munc13 complex.(A) Representative EM (23,000×) of PF-PC synapses of WT and Rapgef3/4-dKO mice.Scale bars: 200 nm.The inserts show docked vesicles.Unpaired t test.****p<0.0001.(B) Representative EM of PF-PC synapses of Prkce f/f and Prkce-cKO mice.Scale bars: 200 nm.Unpaired t test.****p<0.0001.(C) Example PC mEPSCs in Atoh1 Cre and Rapgef3;Rapgef4-cKO mice.Lower: statistics of inter-event interval and amplitude.Grey dots indicate individual data points.Frequency: 2.0±0.2Hz Figure 2 continued on next page Figure 2-source data 1 file for source data of western blots in this figure.The online version of this article includes the following source data and figure supplement(s) for figure 2: Source data 1.The uncut gel of western blots in Figure 2.

Figure supplement 1 .
Figure supplement 1. Input-output relationship of evoked PF-EPSCs in control and mutant mice.

Figure supplement 2 .
Figure supplement 2. The induction of presynaptic PF-PC LTP in 2-month-old mice.

Figure supplement 3 .
Figure supplement 3. The induction of presynaptic PF-PC LTP in elevated temperature.

Figure supplement 4 .
Figure supplement 4. The induction of presynaptic PF-PC LTP in lower Ca 2+ concentration.

Figure 3 continued
Figure3 continued Figure supplement 1. PKA inhibition has a modest effect in blocking cAMP-triggered facilitation.

Figure 4 continued
Figure 4 continued

Figure 5 continued
Figure 5 continued

Figure supplement 2 .
Figure supplement 2. Proposed schematic model for the function of EPAC-PKCε module in presynaptic LTP and motor learning.
Figure 7 continued