Inhibition promotes long-term potentiation at cerebellar excitatory synapses

The ability of the cerebellar cortex to learn from experience ensures the accuracy of movements and reflex adaptation, processes which require long-term plasticity at granule cell (GC) to Purkinje neuron (PN) excitatory synapses. PNs also receive GABAergic inhibitory inputs via GCs activation of interneurons; despite the involvement of inhibition in motor learning, its role in long-term plasticity is poorly characterized. Here we reveal a functional coupling between ionotropic GABAA receptors and low threshold CaV3 calcium channels in PNs that sustains calcium influx and promotes long-term potentiation (LTP) at GC to PN synapses. High frequency stimulation induces LTP at GC to PN synapses and CaV3-mediated calcium influx provided that inhibition is intact; LTP is mGluR1, intracellular calcium store and CaV3 dependent. LTP is impaired in CaV3.1 knockout mice but it is nevertheless recovered by strengthening inhibitory transmission onto PNs; promoting a stronger hyperpolarization via GABAA receptor activation leads to an enhanced availability of an alternative Purkinje-expressed CaV3 isoform compensating for the lack of CaV3.1 and restoring LTP. Accordingly, a stronger hyperpolarization also restores CaV3-mediated calcium influx in PNs from CaV3.1 knockout mice. We conclude that by favoring CaV3 channels availability inhibition promotes LTP at cerebellar excitatory synapses.


Figure 1. High frequency PFs stimulation induces GABA A receptor dependent LTP at PF to PN synapses.
A schematic representation of the cerebellar microcircuit and experimental setting is shown in panel a. Purkinje neurons (PN) receive excitatory inputs from climbing fibers (CF) and mossy fibers (MF) via granule cells (GC) activation. Parallel fibers excitatory synapses (green+ ) drive PNs and inhibitory (red − ) molecular layer interneurons stellate (SC) and basket (BC) cells. Traces elicited by paired-pulses PFs stimulation (Stim, a) in a voltage-clamped PN (Rec, a) at different time points are showed in the inset of panel b. The Baseline trace was obtained from averaging of all recordings during baseline while the t = 45 min trace is the average of three consecutive PF-mediated responses recorded every 20 seconds at the indicated time point. High frequency PFs stimulation induced a long lasting increase in PNs response (MLI dep -LTP, b: mean ± SEM, N = 5, RM ANOVA P < 0.001). Bath application of the GABA A receptor antagonist SR95531 prevented MLI dep -LTP (c, mean ± SEM, N = 5, RM ANOVA P < 0.001). A summary graph for SR95531-mediated effect on MLI dep -LTP is shown in panel e (bar represents normalized PF-Rsp at t = 45 min, mean ± SEM, data from panel b and c), *indicates a statistically significant difference among values (t-test, P = 0.011). Keeping GABAergic transmission intact only during high frequency PFs stimulation was sufficient to induce MLI dep -LTP (d). For these experiments, a 10 minutes baseline was established with SR95531 and the antagonist was washed out for at least 15 minutes before the induction protocol was applied; SR95531 was added back to the recording chamber immediately or 15 minutes after high frequency stimulation (d, mean ± SEM, N = 7, RM ANOVA P < 0.001; e, bar represents normalized PF-Rsp at t = 65 min, mean ± SEM, data from panel d). PPRs value (mean ± SEM) for the baseline (t = 10) and the post induction phase (t = 45) with or without bath application of SR95531 are shown in panel f. Scientific RepoRts | 6:33561 | DOI: 10.1038/srep33561 channels-mediated low threshold calcium spikes 23,24 . By investigating the molecular pathway leading to LTP in PNs, we demonstrate a tight cooperation between ionotropic GABA receptors and low threshold voltage-gated T-type calcium channels (Ca V 3) which promotes LTP at PF to PN synapses. We show that potentiation of transmission at PF to PN synapses following high frequency PFs stimulation requires calcium influx via Ca V 3 channels and GABA A receptor activation. Also, we provide evidence supporting Ca V 3 modulation by inhibitory inputs leading to an increase in channels availability. We then conclude that FFI can control LTP at PF to PN synapses.

Results
LTP at PF to PN synapses is GABA A receptor-dependent. The role played by inhibition in LTP induction at PF to PN synapses was investigated in acute cerebellar slices. In voltage clamped PNs (V h = − 60 mV), molecular layer electrical paired-pulses stimulation (20 Hz) induced PF-mediated fast inward currents with facilitation at the second response (Fig. 1b inset, Baseline). The induced response (PF-Rsp) was stable within 10 minutes from the beginning of the recording (Fig. 1b, average baseline value: − 781.31 ± 25.77 pA, mean ± SEM, N = 5). Once baseline was established, PFs were stimulated at high frequency (burst of 15 pulses at 100 Hz repeated every 3 seconds for a period of 5 minutes) while PNs were switched to current clamp mode. This induction protocol was chosen to mimic GCs physiological activity [25][26][27] and to ensure reliable PF to MLI transmission whose failure rate has been shown to decrease at high frequency rate 28 . Train of PFs bursts stimulation caused a long lasting potentiation of the PFs-induced response (Fig. 1b). Higher values were reached at an early phase (Fig. 1b, normalized PF-Rsp t20 = 1.68 ± 0.18, mean ± SEM, N = 5) with a later stable lower level maintained until the end of the recording (steady state) (Fig. 1b inset, t = 45 min; Fig. 1b, normalized PF-Rsp t45 = 1.34 ± 0.14, mean ± SEM, N = 5). Since LTP can be expressed at presynaptic and/or postsynaptic site 29 , the paired-pulse ratio (PPR) was investigated to discriminate between these possibilities. PPRs at steady state showed no statistically significant difference when compared to baseline (Fig. 1f, PPR t10 = 1.74 ± 0.15, PPR t45 = 1.6 ± 0.13, mean ± SEM; PPR t10 vs PPR t45 : one-way repeated measures (RM) ANOVA, Tukey's post hoc test, P > 0.05, N = 5) indicating that a change in the presynaptic probability of release was unlikely.
The role of GABAergic transmission in LTP induction was then pharmacologically investigated by bath application of the selective GABA A receptor antagonist SR95531 (5 μ M). PNs holding current (I holding ) and baseline PF-mediated response decay time constant (τ off ) displayed no statistically significant difference in the presence of the antagonist (I holding SR95531: − 346.74 ± 44.44 pA, mean ± SEM, I holding control: − 241.63 ± 55.46, mean ± SEM, P = 0.177, t test, N = 5; τ off SR95531: 7.14 ± 0.8 ms, mean ± SEM, τ off control: 5.49 ± 2 ms, mean ± SEM, P = 0.467, t-test, N = 5) suggesting that, under our experimental conditions, PNs were clamped at a membrane potential (V h = − 60 mV) close to the equilibrium potential of GABA A receptor-mediated currents (calculated E Cl = − 63 mV) and that the PF-mediated response was mostly mediated by AMPA receptors.
LTP at PF to PN synapses could therefore be induced by high frequency PFs stimulation provided that the ionotropic GABAergic transmission was intact, supporting a role of MLIs in long-term potentiation at PF to PN synapses; for simplicity we will refer to LTP established under our experimental condition as MLI dependent LTP (MLI dep -LTP) in the later part of the text.
Taken together our data support the requirement of GABA A receptor activation in MLI dep -LTP; MLI dep -LTP is likely postsynaptic and it is expressed at PF to PN synapses.
All three Ca V 3 isoforms are found in PNs with weak Ca V 3.2 channels staining 39 and more pronounced Ca V 3.1 and Ca V 3.3 channels expression 39,40 . PFs bursts-induced calcium transients in mature PNs are mostly mediated by Ca V 3.1 channels 21 and Ca V 3.1 KO mice showed LTP impairment 15 . We therefore investigated whether Ca V 3.1 channels play a role in MLI dep -LTP in acute cerebellar slices from Ca V 3.1 KO mice 41 . Under the same experimental condition which led to MLI dep -LTP in WT mice, Ca V 3.1 KO mice showed no MLI dep -LTP (Fig. 2b, N  Also, PPRs value showed no statistically significant difference to baseline (Fig. 2b inset, PPR t10 = 1.68 ± 0.1, PPR t45 = 1.8 ± 0.26, mean ± SEM; PPR t10 vs PPR t45 , RM ANOVA, Tukey's post hoc test, P > 0.05, N = 5).
In order to exclude any effect of Ca V 3.1 absence on MLI to PN transmission that could potentially interfere with MLI dep -LTP induction, the total PF-induced inhibition was measured in WT and Ca V 3.1 KO mice. Input/ output curves showed no significant difference in Ca V 3.1 KO mice compared to WT animals (Fig. 2c, minimum stimulus intensity: WT = 382.88 ± 152.41 pA, N = 8; KO = 748.95 ± 391.01 pA, N = 10, P = 0.379, t-test) supporting a normal MLI to PN transmission. The absence of MLI dep -LTP is therefore a direct consequence of impaired calcium influx in Ca V 3.1 KO PNs dendrites 15 rather than a secondary effect due to altered inhibition.
A functional coupling between mGluR1 receptors and Ca V 3.1 channels has been identified in PNs with the activation of the G-coupled receptor leading to the potentiation of calcium influx through T-type calcium channels 21 and therefore mGluR1activation may be required for the described long-term potentiation. Indeed, blocking mGluR1 by bath application of the specific antagonist JNJ16259685 (2 μ M) prevented MLI dep -LTP (Fig. 2d, N = 5) with no statistically significant change in PPRs when compared to baseline (Fig. 2d inset, PPR t10 = 1.59 ± 0.09, PPR t45 = 1.57 ± 0.03, mean ± SEM, N = 5; PPR t10 vs PPR t45 P > 0.05, RM ANOVA, Tukey's test). Strikingly, blocking the receptor also revealed a mGluR1-independent LTD (normalized PF-Rsp t45 = 0.7 ± 0.07, normalized PF-Rsp t1 = 1 ± 0.02, P = 0.025 RM ANOVA, Tukey's post hoc test). As expected by MLI dep -LTP impairment Figure 2. Molecular pathway to MLI dep -LTP. MLI dep -LTP was effectively prevented by the broad spectrum low threshold voltage-gated T-type calcium channels specific antagonist TTA-P2 (a, mean ± SEM, N = 5, RM ANOVA P < 0.001). To identify the Ca V 3 isoform required for the GABA-dependent potentiation, long-term plasticity experiments were performed in Ca V 3.1 KO mice. MLI dep -LTP is absent in Ca V 3.1 KO mice (b, mean ± SEM, N = 5, RM ANOVA P < 0.001) under the same experimental condition that leads to GABA A receptor-dependent potentiation of PF to PN transmission in WT mice and traces from a representative experiment are shown in the middle inset. PF-induced inhibitory responses were recorded in PNs from WT and Ca V 3.1 KO mice (c, inset) and input/output curves obtained (c: WT, black circle: N = 8; KO, white diamond: N = 10, mean ± SEM). No statistically significant difference was detected among the two curves (P > 0.05, t-test at all stimulus intensities). MLI dep -LTP also depends on mGluR1 and intracellular calcium stores; high frequency PFs stimulation failed to induce MLI dep -LTP when activation of the metabotropic glutamate receptor mGluR1was prevented by bath application of the specific antagonist JNJ16259685 (d, mean ± SEM, N = 5, RM ANOVA P < 0.001). Inclusion of the non-hydrolysable GDP analog GDPβ S (e, mean ± SEM, N = 3) or heparin (f, mean ± SEM, N = 6, RM ANOVA P = 0.006) in the intracellular recording solution also impaired MLI dep -LTP. PPRs value (mean ± SEM) for the baseline (t = 10 min) and the post induction phase (t = 45 min) under each condition are shown in the left insets in panel a,b,d.
Scientific RepoRts | 6:33561 | DOI: 10.1038/srep33561 caused by mGluR1 inactivation in PNs, the inclusion of the non-hydrolysable GDP analog GDPβ S (2 mM) in the patch pipette also interfered with long term plasticity (Fig. 2e, N = 3). mGluR1 activation induces release of calcium from intracellular stores which could therefore be implicated in MLI dep -LTP. When calcium release from intracellular store via IP3 receptors was prevented by inclusion of heparin (50 μ g/ml) in the patch pipette 42 , MLI dep -LTP was also impaired (Fig. 2f, N = 6) suggesting that calcium influx via Ca V 3 channels and internal stores might cooperate for MLI dep -LTP induction.
These experiments support the requirement of a postsynaptic molecular cascade for the expression of MLI dep -LTP at PF to PN synapses; we showed that MLI dep -LTP is Ca V 3 channels, mGluR1 and internal calcium stores dependent.

Ionotropic GABA receptors and Ca V 3 channels cooperation is required for MLI
channels open in response to small depolarization of the cell, quickly inactivate and recovery from inactive state depends on cell membrane re-hyperpolarization after channel opening 22 . The three Ca V 3 channel isoforms are characterized by different activation curves with Ca V 3.1 (half activation V a : − 60 ± 0.9 mV at 37 °C) and Ca V 3.2 channels (half activation V a : − 51.5 ± 1 mV at 37 °C) opening at more depolarized membrane potential compared to Ca V 3.3 channels (half activation V a : − 73.5 ± 1.3 mV at 37 °C) 43 which therefore requires stronger hyperpolarization to de-inactivate. Modulation of GABA A -mediated transmission could therefore impact Ca V 3 channels availability and affect MLI dep -LTP. We investigated this hypothesis by lowering the chloride concentration in the internal solution for MLI dep -LTP experiments in Ca V 3.1 KO mice to obtain a stronger GABA A receptor-mediated hyperpolarization in PNs as shown by the 13 mV negative shift of the IPSP reversal potential when compared to control condition (control, N = 7: − 63 mV; low chloride, N = 7: − 76 mV) (Fig. 3a); a stronger hyperpolarization might enhance Ca V 3.3 channels availability and therefore improve Ca V 3 channels-mediated intracellular calcium rise in the Ca V 3.1 KO during bursts of PFs stimulation and facilitate MLI dep -LTP induction. Indeed, under these experimental conditions, high frequency PFs stimulation generated a long lasting increase of transmission in Ca V 3.1 KO mice (Fig. 3b, t = 45, normalized PF-Rsp = 1.29 ± 0.17, mean ± SEM, N = 5). The recovery of LTP depended on GABA A activation (KO-MLI dep -LTP) since it was absent in presence of SR95531 (Fig. 3c, normalized EPSC t10 = 1.11 ± 0.01, normalized EPSC t45 = 1.05 ± 0.04, P > 0.05 RM ANOVA, Tukey's post hoc test, N = 5). More importantly, TTA-P2 bath application (Fig. 3d) also prevented KO-MLI dep -LTP (normalized PF-Rsp t45 = 1.04 ± 0.07, normalized PF-Rsp t10 = 0.99 ± 0.02, P > 0.05 RM ANOVA, Tukey's post hoc test, N = 5) indicating that the enhanced availability of an alternative PN-expressed Ca V 3 isoform could compensate for the lack of Ca V 3.1 channels.

Ca V 3 channels-mediated calcium transient is under the control of ionotropic GABA receptors.
The recovery of MLI dep -LTP in Ca V 3.1 KO mice supports a functional coupling between GABA A receptors and Ca V 3 channels important for LTP at PF to PN synapses. The influence of inhibition on Ca V 3-mediated calcium transient elicited by high frequency PFs stimulation was therefore tested in PNs from WT mice by calcium imaging (Fig. 4). For these experiments, the calcium indicator Oregon Green BAPTA 6F (400 μ M) was added to the low chloride internal solution and loaded into the recorded PN via the patch pipette. To prevent calcium release from intracellular stores heparin (50 μ g/ml) 42 was also included in the patch clamp intracellular recording solution. The variation of intracellular calcium concentration was monitored in current clamped PNs (Fig. 4aa-ba) while PFs were stimulated by a single 100 Hz burst (15 pulses). Stimulus intensity was set to induce PFs-mediated responses (Fig. 4ab) similar to those recorded during the LTP induction protocol. The raise in intracellular calcium concentration caused by high frequency PFs stimulation (Fig. 4aa, left panel) was largely Ca V 3 channels-mediated as shown by fluorescence attenuation following TTA-P2 bath application (Fig. 4aa, right  panel); the relative change in fluorescence (Δ F/F) was strongly influenced by TTA-P2 with a 64% reduction of its peak value by the antagonist (Fig. 4ac, Δ F/F Control = 1.37 ± 0.5, Δ F/F TTA-P2 = 0.47 ± 0.09, mean ± SEM, P = 0.041, ANOVA, Bonferroni post hoc test, N = 5). Interestingly, the Ca V 3 channels-mediated component of the calcium transient was completely lost when inhibition was blocked. In presence of SR95531 (Fig. 4ba, left panel), the induced increase in intracellular calcium was TTA-P2 insensitive (Fig. 4ba, right panel and Fig. 4bb, N = 5) and therefore mediated by high threshold voltage gated calcium channels.
These findings demonstrate that ionotropic GABA receptors activation is required for Ca V 3 channels-mediated calcium rise in PNs during high frequency PFs stimulation.
In agreement with the recovery of MLI dep -LTP observed in Ca V 3.1 KO mice, the calcium transient recorded in PNs from KO mice was largely Ca V 3 channels-mediated only when elicited by PFs stimulation at low intracellular chloride concentration (Fig. 4d, low chloride); bath application of TTA-P2 strongly reduced calcium influx in low chloride (normalized Δ F/F MAX -Control = 1 ± 0.27, normalized Δ F/F MAX -TTAP2 = 0.42 ± 0.13, mean ± SEM, N = 5, P = 0.017, paired t-test) while no statistically significant difference (normalized Δ F/F MAX -Control = 1 ± 0.23, normalized Δ F/F MAX -TTAP2 = 0.79 ± 0.08, mean ± SEM, N = 5, P = 0.446, paired t-test) was detected in the presence of the antagonist when the regular internal solution was used (Fig. 4d, normal chloride).
Taken together these results support the fundamental role of Ca V 3 channels-mediated calcium influx in MLI dep -LTP induction at PF to PN synapses and the requirement of GABA A receptors activation to ensure the availability of Ca V 3 channels for opening during bursts of PFs activation.

Discussion
In the cerebellar cortex, GCs activation by MFs input leads to excitatory (monosynaptic) and inhibitory (di-synaptic) events in PNs and their interaction in the induction of LTP at PFs excitatory synapses is described in this article. We showed that high frequency PFs stimulation caused LTP at PF to PN synapses only when MLIs-mediated GABAergic transmission was intact (Fig. 1).
MLI dep -LTP is induced by a postsynaptic mechanism as showed by its recovery in Ca V 3.1 KO mice (Fig. 3b) and its impairment by the intracellular block of GPCRs (Fig. 2e) and IP3 receptors (Fig. 2f); MLI dep -LTP is GABA A receptors (Fig. 1), Ca V 3 channels (Fig. 2a,b), mGluR1 receptors (Fig. 2d,e) and intracellular calcium store (Fig. 2f) Ca V 3 channels-dependent LTP has been previously reported to be induced at PF to PN synapses in the presence of the GABA A receptor antagonist bicuculline 15 in apparent discrepancy with our results. Nevertheless, PNs have been shown to express the ionotropic GABA receptor bicuculline-insensitive ρ subunits 44,45 suggesting that bicuculline might not be able to completely eliminate inhibition in PNs. Indeed, we were able to record a bicuculline-insensitive IPSC component in PNs following PFs stimulation (Supplementary Information); after bath application of bicuculline (20 μ M), 6% of the total IPSQs was still present (Supplementary Figure S1 panel a  and panel b) and it was completely blocked by the following application of SR95531 (5 μ M). The fact that LTP was successfully induced in the presence of bicuculline suggests that this residual bicuculline-insensitive component of the ionotropic GABA receptors-mediated response was sufficient to mediate Ca V 3 channels recovery from inactivation at least in WT mice. In agreement, MLI dep -LTP was unaffected by bath application of bicuculline (Supplementary Figure S1 panel c and panel d).
The MLI network can be modulated by long-term plasticity and PF to MLI synapses can be potentiated 32 at low frequency stimulation (2-8 Hz) while high frequency stimulation induces LTD 30,31 . Also, high frequency PFs stimulation causes LTP at inhibitory MLI to MLI synapses 34 via increase of GABA release 33 . High frequency PFs stimulation could therefore reduce inhibitory inputs to PNs via decreased MLIs excitation and/or by increasing While TTA-P2 bath application strongly affected the calcium transient recorded in control condition (aa, right panel), the antagonist showed no effect in presence of SR95531 (ba, right panel). Quantified relative change in fluorescence (Δ F/F) showed a large TTA-P2 sensitive component revealing that calcium influx is mostly mediated by Ca V 3 activation but only in control condition (ac, mean ± SEM, N = 5, ANOVA P < 0.001). Ca V 3mediated calcium influx is lost when inhibition was blocked by bath application of SR95531 (bb, mean ± SEM, N = 5, ANOVA P < 0.001). High frequency PFs stimulation in low internal chloride induced MLI dep -LTP (c, mean ± SEM, N = 6, RM ANOVA P < 0.001); PPRs (mean ± SEM) value for the baseline (t = 10 min) and the post-induction phase (t = 65 min) are shown in the panel c inset. The effect of different intracellular chloride concentrations on the PFs-induced calcium transient was evaluated in PNs from Ca V 3.1 KO mice (d). The TTA-P2 sensitive component of the calcium transient observed in low chloride (Low chloride: control = 1 ± 0.23, mean ± SEM, N = 5; TTA-P2: 0.42 ± 0.13, mean ± SEM, N = 5, P = 0.017, paired t-test) was lost when the normal chloride internal solution was used (Normal Chloride, control: 1 ± 0.23, mean ± SEM, N = 5; TTA-P2: 0.79 ± 0.08, mean ± SEM, N = 5, P = 0.446, paired t-test). Before cell averaging, the maximal Δ F/F value obtained before and after TTA-P2 bath application for each cell was normalized to the mean Δ F/F value obtained in control condition. *statistically significant difference, paired t-test. # statistically significant difference, t-test.
Scientific RepoRts | 6:33561 | DOI: 10.1038/srep33561 inhibition onto MLIs. The decreased inhibition in PNs could result in indirect potentiation of PFs-mediated responses in these neurons. No significant GABA A -mediated component in the PFs-mediated responses was observed under our experimental conditions and therefore, if present, indirect potentiation should have had little influence on our recordings. This was confirmed by investigating the specific impact of high frequency PFs stimulation on excitatory transmission by blocking GABAergic inputs before and after the induction protocol. Excitatory transmission at PF to PN showed postsynaptic potentiation to level comparable to what previously observed (Fig. 1d). GABA A receptor activation is therefore required for MLI dep -LTP at PF to PN synapses only during the induction phase. This and the intracellular pharmacological block of MLI dep -LTP (Fig. 2e,f) strongly argue against indirect potentiation and they support involvement of inhibition in LTP induction at PF to PN synapses.
By investigating the molecular pathway leading to MLI dep -LTP we have provided evidence for a tight cooperation among GABAergic transmission and low threshold voltage-gated calcium channels in PNs (Figs 3 and 4). We propose an active postsynaptic role for GABAergic transmission in MLI dep -LTP induction with inhibition modulating T-type Ca V 3 calcium channels availability (i.e. by favoring de-inactivation). This hypothesis is supported by the recovery of MLI dep -LTP in Ca V 3.1 KO mice by the intracellular modulation of the chloride electrochemical gradient in PNs (Fig. 3b). As shown by KO-MLI dep -LTP dependency on TTA-P2 (Fig. 3d), an alternative Ca V 3 channels isoform is recruited when GABA A receptor activation favored hyperpolarization toward more negative potentials (Fig. 3a). Accordingly, the TTA-P2-sensitive component of the calcium transient was restored in Ca V 3.1 KO mice when the intracellular chloride was decreased (Fig. 4d). Furthermore, Ca V 3-mediated calcium influx and MLI dep -LTP was detected in PNs from WT mice only when inhibitory GABAergic transmission was intact (Figs 4a,b and 1).
MLIs-mediated postsynaptic response causes membrane potential re-hyperpolarization in depolarized PNs both in the dendritic compartment and at the soma 46,47 . Since high frequency PFs stimulation is able to bring PNs membrane voltage to firing threshold (Fig. 4ab), the recovery of MLI dep -LTP (Fig. 3b) and Ca V 3 channels-mediated calcium transient (Fig. 4d) in Ca V 3.1 KO mice suggests that activation of GABA A receptors favor re-hyperpolarization toward a level suitable for Ca V 3 channels de-inactivation. Ca V 3 channels inactivation has been shown to be effectively removed by GABA A -mediated IPSPs in DCN neurons [35][36][37] ; the IPSP reversal potential in these neurons (− 75 mV) 37 is close the one obtained under our experimental condition in low chloride (Fig. 3a) supporting an effective Ca V 3 channels de-inactivation also in our experiments. Ca V 3.3 channels are highly expressed in PNs and they require a stronger hyperpolarization to recovery from inactivation; based on its expression pattern and biophysical characteristics, Ca V 3.3 channels are therefore the isoform most likely to be involved in the rescue of MLI dep -LTP in Ca V 3.1 KO mice although a role of Ca V 3.2 channels cannot be excluded at this time 48 . The reversal potential of GABAergic currents (E GABA : − 85/− 87 mV) measured in mature PNs 49,50 predict a strong hyperpolarization in PNs caused by GABA A receptors activation suggesting that Ca V 3.3 channels might be also recruited under physiological conditions and it might also participate with Ca V 3.1 in MLI dep -LTP induction. Ca V 3.3 channels sole requirement is nevertheless unlikely. Cerebellar long-term plasticity is a calcium-dependent mechanism 7 and PFs-mediated increment of calcium in PNs spines is mostly mediated by Ca V 3.1channels even though in Ca V 3.1 KO mice a residual T-type dependent influx is still present 21 . Furthermore, Ca V 3.1 channels-mediated calcium influx is potentiated by mGluR1 activation and MLI dep -LTP dependency on this metabotropic glutamate receptor (Fig. 2d,e) further supports the requirement for the Ca V 3.1 channel isoform in MLI dep -LTP. Interestingly, the impairment exhibited by Ca V 3.1 KO mice in long term VOR phase-reversal training seems less severe when compared to the one of WT mice systemically injected with TTA-P2 15 suggesting that Ca V 3.3 channels might also participate in cerebellar-mediated motor learning in Ca V 3.1 KO mice and slightly attenuate their phenotype.
When MLI dep -LTP was compromised, decrease efficiency in PFs transmission was also revealed unmasking a pathway leading to depression at PF to PN synapses. Interestingly, this long term depression was still present when mGluR1 was inactivated (Fig. 2d). At PF to PN synapses, mGluR1 plays a central role in LTD 51-55 but long term depression is nevertheless also reliably induced by nitric oxide (NO) uncaging when coupled to PNs depolarization 56 suggesting that this PF-released anterograde messenger 57 together with the depolarization-induced intracellular calcium raise in PNs might be sufficient to permanently decrease the synaptic transmission strength at these synapses. Indeed a NO synthase (NOS) dependent LTD has been previously described at PF to PN synapses 58 ; this LTD requires NMDA receptors activation 58 and it is mGluR1 independent 59 . NO is likely to be released under our experimental conditions and therefore it might play a role in the depression observed whenever MLI dep -LTP was impaired.
A simple model describing the first events leading to MLI dep -LTP at PF to PN synapses can be proposed (Fig. 5): high frequency PFs stimulation activates AMPA receptors and causes PN dendrites depolarization counterbalanced by the GABA A -induced hyperpolarization elicited by MLIs; membrane depolarization triggers T-type calcium channels activation and their availability for opening is controlled by inhibition. In order to induce MLI dep -LTP at PF to PN synapses, Ca V 3 channels require mGluR1 activation that potentiates Ca V 3.1 channels-mediated calcium influx via a PLC independent pathway 21 . Via the G q /PLC pathway, mGluR1 activation also leads to the release of calcium from the intracellular stores via activation of IP3 receptors; intracellular released calcium also seems to be required for MLI dep -LTP.
While high frequency stimulation of PFs induced a mGluR1-dependent LTP in vivo 18 , low frequency (1 Hz) stimulation of PFs in the molecular layer of cerebellar slices induced a long lasting enhancement of transmission at PF to PN synapses which was unaffected by the pharmacological inactivation of this metabotropic receptor 9 . Together with our results, these findings support the existence of two distinct pathways leading to LTP which are differently engaged by PFs activity. Low frequency stimulation induces a mGluR1 independent potentiation which is unaffected by GABAergic ionotropic transmission impairment 19 but it depends on NO release 6 and PP2B activation 10  and it is dependent on ionotropic GABA receptors and mGluR1 activation which cooperate to ensure a reliable activation of low threshold voltage-gated calcium channels. We therefore describe here a complementary pathway regulating PF to PN synaptic efficacy in a context of high frequency GCs inputs. Since GCs inputs recorded in vivo range from few to several hundred hertz 27 , both mechanisms might coexist in order to control plasticity in different conditions or they might underlie specific pathways expressed in different groups of GC to PN synapses as suggested by the discovery that zebrin band specific physiological mechanisms could regulate cerebellar information processing 60,61 .
Suggestions on the possible role played by MLI dep -LTP in cerebellar physiology come from previous findings obtained in vivo. When PFs in the cat forelimb movements-related C3 zone are electrically stimulated with the same paradigm used in this paper, a bidirectional change in PFs receptive field of PNs is induced depending on the co-activation of CF input to the recorded cell. While co-activation leads to depression, an enlargement of PFs receptive field is observed when PFs are the sole excitatory pathway stimulated. Following PFs stimulation, PNs in the C3 zone are driven by cutaneous stimulation from several parts of the body in agreement with what could be expected following awakening of silent connection between PFs and PNs 62 . In attempt to mimic protocols performed in vivo, LTP has been established in this study without pharmacological perturbation of synaptic transmission and intracellular signal transductions. Thus, our work might provide a detailed description of the initial molecular events which lead to the PFs receptor field enlargement observed in vivo. Consequently, it is tempting to speculate that the described MLI-dependent LTP could be the result of the summation of newly awaken PF to PN synapses rather than the sole increased membrane expression of AMPA receptors at activated synapses.

Methods
Ethical approval. All animal procedures were performed in accordance with the University of Strasbourg animal care committee's regulations and they were approved by the Ethical Committee of the University of Strasbourg (A67-2018-38).

Electrophysiology.
Patch clamp experiments were conducted on acute coronal slices from cerebellum of adult C57BL/6 male mice (P27-P46) and age-matched Ca V 3.1 knockout (KO) male mice from homozygous breeding. Mice were anesthetized by exposure to isoflurane, decapitated and the cerebellum dissected in ice cold bubbled (95% O 2 /5% CO 2 ) aCSF containing (in mM): NaCl 120, KCl 3, NaHCO 3  For patch clamp recordings slices were moved to a recording chamber at 34 °C and continuously perfused with bubbled aCSF eventually supplemented with antagonists as stated in the main text. Borosilicate glass pipettes were pulled using a vertical puller (Narishige PC-10: Narishige, Tokyo, Japan) to a final resistance of 4-4.5 MΩ and filled with the following internal solution (in mM): KGluconate 130, KCl 10, MgCl 2 1, HEPES 10, Na 2 ATP Molecular layer interneuron inhibitory synapse (MLI) is also shown in the scheme. Following high frequency parallel fibers (PF) stimulation, GABA A -mediated hyperpolarization (blue shadow) limits AMPA-induced depolarization (pink shadow) to a range suitable for Ca V 3 activation (light violet shadow). AMPA-mediated depolarization (red arrow) activates Ca V 3 calcium channels dependently on their availability regulated by inhibition (blue arrow) and Ca V 3-mediated calcium influx is enhanced by mGluR1 activation (yellow arrow). mGluR1 activation also leads to calcium release from intracellular stores via IP3 receptors. The described molecular steps initiate the cascade that leads to MLI dep -LTP and downstream events are still to be determined. 4, NaGTP 0.4, sucrose 16, pH 7.3. For experiments in low internal chloride concentration the following internal solution was used (in mM): KGluconate 136, KCl 4, MgCl 2 1, HEPES 10, Na 2 ATP 4, NaGTP 0.4, sucrose 16, pH 7.3. Once whole-cell configuration was established, a period of at least 20 minutes was waited before the start of the experiment. PNs were clamped at − 60 mV and PFs-induced responses elicited by electrical stimulation delivered by a patch pipette positioned in the molecular layer distant from the recorded cell. The stimulation pipette was filled with the following solution (in mM): NaCl 120, KCl 3, HEPES 10, NaH 2 PO 4 1.25, CaCl 2 2, MgCl 2 1, glucose 10, pH 7.3. The PF-induced response was monitored over time by a test protocol of paired stimulation pulses (20 Hz) applied every 20 seconds. Three consecutive induced responses were averaged to obtain a mean trace of the evoked response per minute of recording. The induction protocol was applied in current-clamp mode with cells held at − 68 mV.
Data were collected with a MultiClamp 700B (Molecular Devices, Sunnyvale, California), filtered at 2 kHz and digitized at 20 kHz. Data from each cell were normalized to the mean baseline value before cell averaging. Data are expressed as mean ± SEM.
A small hyperpolarize step (− 10 mV) was applied before each stimulation to follow series resistance during the experiment. Series resistance was compensated by 70-80% and cells were discarded if significant changes were detected.
For input/output curves, cerebellar coronal acute slices from adult C57BL/6 and Ca V 3.1 KO mice were prepared as previously described and the following intracellular recording solution was used in whole cell patch clamp experiments (in mM): cesium methanesulfonate 135, NaCl 6, MgCl 2 1, HEPES 10, MgATP 4, Na 2 GTP 0.4, EGTA 1.5, QX314Cl 5, pH 7.3. The glutamatergic transmission was kept intact during the recordings to allow the full recruitment of all MLIs contributing to the GABA-mediated inhibitory response; to minimize the EPSCs contribution to the induced response, the recorded PNs was clamped at − 10 mV and the PF-mediated inhibitory response was elicited by molecular layer electrical stimulation; the stimulation electrode was pulled with a vertical puller (Narishige PC-10) to a final 5 MΩ resistance when filled with the following solution (in mM): NaCl 120, KCl 3, HEPES 10, NaH 2 PO 4 1.25, CaCl 2 2, MgCl 2 1, glucose 10, pH 7.3. Input/output curves were obtained by progressively increasing the strength of stimulation by 0.1 mA increments. At each intensity, PFs-mediated responses were recorded for three consecutive stimuli (one every 10 seconds) and traces averaged before analysis.
IPSPs were recorded in current clamp mode in presence of NBQX 5 μ M. For IPSP/Vm curves, PNs were held at − 40 mV and the membrane potential progressively hyperpolarized by 22 steps with a 200 pA increment.
Calcium imaging. 300 μ m sagittal slices were prepared from adult (4-5 weeks) C57BL/6 and Ca V 3.1 KO male mice as previously described and calcium imaging experiments were performed at room temperature. The calcium indicator Oregon Green 488 BAPTA 6F (K D = ~3 μ M, Molecular Probes) and heparin were added to the low chloride or regular internal solution to a final concentration of 400 μ M and 50 μ g/ml respectively and they were loaded into PNs via the patch clamp pipette. After whole cell establishment, the calcium indicator was allowed to diffuse for at least 20 minutes before starting the experiment. PNs where held at a potential close to − 70 mV in current clamp configuration and PF stimulation was achieved by electrical stimulation via a glass stimulation pipette placed in the molecular layer. A single 100 Hz burst stimulation (15 pulses) was applied while calcium images were acquired every 50 ms by a sCMOS CCD camera (Xyla 5.5, Andor Technology Ltd, UK) with a 20 ms exposure time. Stimulation was repeated at long interval (at least 3 minutes between trials) to avoid plasticity in the recorded cell and collected images analyzed by using ImageJ 63 . Relative change in fluorescence (Δ F/F) was quantified in ROIs including the entire dendritic tree area in which PF-induced increase in calcium concentration was detected. Average Δ F/F value for each cell was obtained from 5 consecutive imaging sections before and after TTA-P2 500 nM bath application with or without SR95531 5 μ M.

Statistics.
For plasticity experiments, PF-induced responses and paired-pulse ratios obtained at different time point were compared by one-way repeated measure (RM) ANOVA followed by Tukey's post hoc test.