A frequency-dependent mobilization of heterogeneous pools of synaptic vesicles shapes presynaptic plasticity

The segregation of the readily releasable pool of synaptic vesicles (RRP) in sub-pool which are differentially poised for exocytosis shapes short-term plasticity at depressing synapses. Here, we used in vitro recording and modeling of synaptic activity at the facilitating mice cerebellar granule cell to Purkinje cell synapse to demonstrate the existence of two sub-pools of vesicles in the RRP that can be differentially recruited upon fast changes in the stimulation frequency. We show that upon low-frequency stimulation, a population of fully-releasable vesicles is silenced, leading to full blockage of synaptic transmission. A second population of vesicles, reluctant to release by simple stimuli, is recruited in a millisecond time scale by high-frequency stimulation to support an ultrafast recovery of neurotransmitter release after low-frequency depression. The frequency-dependent mobilization or silencing of sub-pools of vesicles in granule cell terminals should play a major role in the filtering of sensorimotor information in the cerebellum.


21
In neural networks, transfer of information largely relies on the ability of 22 presynaptic terminals to transduce information encoded by changes in action potentials 23 (APs) firing rate into changes in the release of neurotransmitters. During trains of APs, 24 the immediate tuning of synaptic efficacy is determined by the combination of multiple 25 parameters including AP frequency, past firing activities, number of release-competent 26 synaptic vesicles (SVs), also referred as the readily-releasable pool (RRP), probability of release (p r ), or the number of release sites (N). To date, the deciphering of cellular 28 mechanisms underlying synaptic efficacy is challenged by a non-unified view of the 29 identity of SVs belonging to the RRP (Neher, 2015). Depending on studies, the RRP  asynchronously did not participate in a detectable way during theses protocols (Fig. 1A). 180 This enables us to estimate the number of quanta released during these trains. Values

196
At 2 Hz, the synaptic responses failed to facilitate (Fig. 1A, C) and a sustained 197 stimulation of PF during hundreds of stimuli led to a near full blockage of synaptic 198 transmission (Fig. 1D). We named this rapid blockage of synaptic transmission "Low

216
While the onset and the plateau of LFD were highly variable from one experiment 217 to another (Fig. 1F), these parameters stay stable for a given PC as long as the 218 recording of EPSCs could be maintained. In a series of 8 experiments, two LFD (LFD#1 219 and LFD#2) separated by a resting period of 5 to 10 minutes were successively elicited.

220
As shown in figure 1G, a full recovery of EPSC amplitude was achieved at the first  100 Hz applied in control condition and after LFD induction were compared (Fig. 2D). 251 Results showed that a full recovery from LFD was achieved within approximately 50 ms  257 We then studied how the frequency of stimulation in the train influenced the ability be repeated several times after a recovery time (Fig. 1F), we systematically applied the LFD and LFD triplet protocols to each PC recorded. Strikingly, although the number of 279 stimuli was the same in both type of protocols (LFD classic or LFD triplet ; Fig. 3A, B), 280 recovery from depression was highly influenced by triplet frequency: the higher the 281 frequency inside the triplets, the lower the level of recovery was seen to be ( Fig. 3B). 282 For example, with triplets at 200 Hz during LFD triplet , synaptic transmission hardly 283 recovers (maximal recovery 38.9 % ± 7.19 n=6, compared to 196.7 % ± 8.7, after LFD, (triplets at 100 Hz) successively applied in the same cells were both followed by the 290 application of a 100 Hz recovery train (Fig. 4A, B). Cumulative EPSC amplitudes were 291 determined, and the approximate number of quanta released per bouton for the whole 292 protocol (LFD or LFD triplet + recovery train) was estimated (Fig. 4C) is another argument as to why this hypothesis cannot be retained. Ca 2+ -driven, five-site sensor from a fraction of releasable SVs (Fig. 6A, see Methods).

331
These SVs become replenished in two steps. The first step is a Ca 2+ -dependent priming 332 step (R 0 R 1 ), followed by a Ca 2+ -independent filling step (R 1 V), i.e. R 0 +R 1 333 corresponds to the reluctant pool and V to the fully-releasable pool in our experiments.

334
In the simulations, release was triggered by Gaussian shaped Ca 2+ signals with and a strong facilitation during sustained low-frequency stimulation (Fig. 6F,G). This is in 361 stark contrast to our experimental results that showed the opposite behavior of GC 362 synapses. Thus, although the latter two models successfully described several aspects shown in Figure 7A, full recovery after LFD is described by a single exponential function 382 (tau =153.8 s, R 2 =0.93), indicating a single-step process. However, recovery following 383 LFD triplet was biphasic: during an initial phase, release remained almost fully blocked.

384
This initial phase was followed by a second phase that was again characterized by a  Fig. 7D). The amplitudes of EPSCs 397 recorded in the test trains were compared to those obtained from the application of a 398 similar train in a control condition (before LFD). As shown in Figure 7D, the amplitude of 399 EPSCs during the test trains increased proportionally to the length of the resting period.

400
More interestingly, after 1 min of rest, the first responses in the test train were still fully blocked (Fig. 7E) whereas the amplitudes of the late responses (after stimuli # 11-12 in 402 the test train) were similar to the corresponding responses in the control train (Fig. 7F). 403 This demonstrates that after 1 minute of rest, no SVs from the fully-releasable pool were 404 ready to be released, despite partial reconstitution of the reluctant pool and the 405 replenishment of the fully-releasable pool by the reluctant pool.  conditions (Fig. 8A, B). At the opposite, recovery from LFD was strongly reduced 424 (maximal recovery 85.5 % ± 11.6 n=5, Fig. 8B showed, the application of 10 µM EGTA-AM did not affect the basal release of glutamate 430 (Fig. 8C) signed rank test, n=6) (Fig. 8D).

436
The present work demonstrates that the release of glutamate and presynaptic

444
Our results showed that reluctant SVs that cannot be recruited during single 445 stimuli at low-frequencies supports the release of glutamate during high-frequency train, 446 even after a full depletion of fully-releasable SVs (Fig. 2-4). At GC-PC synapses, the SVs differ by their Ca 2+ sensitivity and that 2 Ca 2+ sensors differentially shape glutamate 457 release during low-and fast-frequency stimulation (Fig. 8). In central synapses, the Ca 2+ 458 sensors synaptotagmin 1, 2 and 9 are involved in fast synchronous release triggered by Both experimental data (Fig. 7) and in silico simulation (Fig. 6)  indicate that both pools are recruited sequentially (Fig. 7), one cannot exclude additional  (Fig. 2). Finally, during high-frequency inputs, GC-PC synapses are able to 534 reset and standardized synaptic efficacy independently of the recent history of previous 535 events (Fig. 2).     94, paired t-test, n=8) in EPSC amplitudes between LFD#1 and LFD#2 reflects a full recovery from depression after the resting period that followed LFD#1. Right graph: Superimposition of mean EPSC amplitudes recorded during LFD#1 (green points) and LFD#2 (blue points). The similarity between LFD#1 and LFD#2 was tested using a paired t-test. The values were statistically different for the entire experiment (p<0.001, n=8) but the 60 firsts response were statistically identical (paired t-test, p=0.08, n=8).     In the corresponding mathematical model (right) the recycling / reserve pool is contained only implicitly by restoration of the reluctant pool (R 0 +R 1 ) from fused SVs (dashed arrow) and via a basal refilling rate (k basal ). Therefore, this model is referred to as sequential "two-pool" model. During highfrequency activation the residual Ca 2+ increases, resulting in recruitment of SVs from the reluctant pool into the RRP, i.e. a temporal increase in the RRP that causes substantial facilitation (red arrow). The residual Ca 2+ generates an additional moderate, short-lasting facilitation due to slow unbinding from the release sensor (dashed red arrow). During low frequency activation at 2 Hz, the residual Ca 2+ fully drops back to resting level between stimuli and SVs recruited to the RRP return to the reluctant pool (green arrow). B, Simulated time courses and amplitudes of residual Ca 2+ during high-(100 Hz, black) or low-frequency (2 Hz, red) activation starting from a resting Ca 2+ (Ca rest ) level of 50 nM. C, Fraction of Ca 2+ unoccupied SVs in the RRP (V) of the release sensor during the initial three activations of a 100 Hz (black) or 2 Hz (red) activation train. Note that during the first three activations at low frequency the RRP relaxes to its initial (V=1, i.e. 100%) from a transient overfilling (V>1) prior to the next pulse while it continues to increase in size during high-frequency activation due to the build-up in residual Ca 2+ and continuing recruitment of reluctant vesicles (cf. B). D, Transmitter release rates during three activations at high (black) or low frequencies (red), normalized to the first release process. E, Paired pulse ratios (PPRs) calculated as the ratio of release probabilities in the i-th (p r , i ) and the first pulse (p r,1 ) during 100 Hz (filled black) or 2 Hz (red) activations plotted against the logarithmically scaled stimulus number. Open circles show PPRs duringh100 Hz activations started in a previously depressed model. F, (left) One-pool model of Ca 2+ binding and release according to Wölfel et al (2007) consisting of the "allosteric" sensor model (Lou et al., 2005) supplemented with a reloading step of 2 SVs/ms. In contrast to our experimental findings, this model generates low-frequency facilitation and high-frequency depression (right). G, (left) as in F but for two parallel, non-interacting pools of SVs differing in their release rate constants thereby generating a "fast releasing pool of SV" (release rate I + as in F) and a "slowly releasing pool of SV" (release rate J + <I + as in F, Wölfel et al., 2007). Both models are restored via Ca 2+ independent reloading steps of 2 SVs/ms.
Note that similar to the model in F, this simulation generates a high-frequency depression and a low-frequency facilitation.  At the opposite, early responses were slightly reduced after application of EGTA-AM during recovery by 100 Hz train (p=0.031 at stimulus #5 and stimulus #6, signed rank test, n=6) (right panel).