Fast changes in the functional status of release sites during short-term plasticity: involvement of a frequency-dependent bypass of Rac at Aplysia synapses

Abstract

Synaptic transmission can be described as a stochastic quantal process defined by three main parameters: N, the number of functional release sites; P, the release probability; and Q, the quantum of response. Many changes in synaptic strength that are observed during expression of short term plasticity rely on modifications in P. Regulation of N has been also suggested. We have investigated at identified cholinergic inhibitory Aplysia synapses the cellular mechanism of post-tetanic potentiation (PTP) expressed under control conditions or after N has been depressed by applying lethal toxin (LT) from Clostridium sordellii or tetanus toxin (TeNT). The analysis of the Ca²⁺ dependency, paired-pulse ratio and variance to mean amplitude relationship of the postsynaptic responses elicited at distinct extracellular [Ca²⁺]/[Mg²⁺] elicited during control post-tetanic potentiation (PTP_cont) indicated that PTP_cont is mainly driven by an increase in release probability, P. The PTP expressed at TeNT-treated synapses (PTP_TeNT) was found to be similar to PTP_cont, but scaled to the extent of reduction in N produced by TeNT. Despite LT inducing a decrease in N as TeNT does, the PTP expressed at LT-treated synapses (PTP_LT) was characterized by exceptionally large amplitude and bi-exponential time course, as compared to PTP_cont or the PTP_TeNT. Analysis of the Ca²⁺ dependency of PTP_LT, paired-pulse ratio and fluctuations in amplitude of the postsynaptic responses elicited during PTP_LT or the variance to mean amplitude relationship of time-locked postsynaptic responses in a series of subsequent PTP_LT indicated that an N-driven change is involved in the early phase (1 s time scale) of PTP_LT, while at a later stage PTP_LT is composed of both N and P increases. Our results suggest that fast switching on of the functional status of the release sites occurs also during the early events of PTP_cont. The early N-driven phase of PTP_LT is likely to be a functional recovery of the release sites silenced by Rac inactivation. This effect did not appear to result from reversion of LT inhibitory action but from bypassing the step regulated by Rac. Altogether the data suggest that Rac and the intracellular pathway which allows the bypassing of Rac are key players in new forms of short-term plasticity that rely on fast, activity-dependent changes in the functional status of the release sites.

Short-term-plasticity (STP) is a transient modification of synaptic strength induced by bursting or sustained neuronal activity. STP refers to a variety of phenomena, expressed in the millisecond to the minute time scale and involving distinct molecular events at both pre- and postsynaptic compartments (for review: Zucker & Regehr, 2002). STP has been proposed to mediate temporal coding in cortical microcircuits (Silberberg et al. 2004), sensory adaptation (Chung et al. 2002), synaptic filtering (Dittman et al. 2000), and synchronization of neuronal activities during the execution of motor programmes (Sanchez & Kirk, 2000). Deciphering the mechanisms underlying STP remains a key issue for understanding brain computing properties.

Consistent with the quantal hypothesis for neurotransmitter (NT) release, postsynaptic responses evoked by synaptic vesicle (SV)-mediated exocytosis of NT can be described as a stochastic, quantal process defined by three parameters: Q, the amplitude of the quantum of response; P, the output probability that a release-ready SV will fuse with plasma membrane upon a [Ca²⁺]_i rise; and N, which has been proposed earlier to correspond to the number of independent release sites. Many forms of STP are related to changes in P (Zucker & Regehr, 2002). These include facilitation (Dittman et al. 2000; Oertner et al. 2002; Felmy et al. 2003) due to residual calcium, post-tetanic potentiation (PTP), which is thought to result from an increase in output release probability (Korogod et al. 2005), and/or recruitment of reserve SVs to refill the release-ready pool (RRP), under the control of the synapsins (Humeau et al. 2001a; Sun et al. 2006). Postsynaptic Q-related forms of STP have been described, due to receptor desensitization and/or saturation during intensive NT release, and they participate in synaptic depression (Foster & Regehr, 2004; Saviane & Silver, 2006). The possibility that N may rapidly change following modification of synaptic activity remains poorly documented. Recruitment of release sites has been proposed to participate in the expression of presynaptic long-term potentiation (LTP) at the mossy fibres in the hippocampus (Reid et al. 2004), and the 5-HT-mediated heterosynaptic facilitation of neurotransmitter release at Aplysia sensorimotor synapses is due to an increase in the number of release sites (Royer et al. 2000). However, the detailed molecular mechanisms involved remain as yet undeciphered.

Variance–mean analysis allows easy graphical distinctions for changes in N, P and Q (Humeau et al. 2001b, 2002; Silver, 2003; Foster & Regehr, 2004), but interpretation of the data obtained may suffer some difficulties. Indeed, possible stochastic steps in replenishment of release sites can introduce variability in the number of functional, release-ready sites at a given time, thereby contributing to variance (Quastel, 1997; Foster & Regehr, 2004). This introduces an ambiguity between the meaning of N and P determined by variance–mean analysis and the N and P defined in the quantal hypothesis for NT release. N deduced by variance–mean analysis is a functional parameter corresponding to the number of release sites equipped with a release-ready SV at a given time. P corresponds not only to the output probability, but also to the probability that a release-ready SV is present on the release site. Because variance–mean analysis is based on analysing the amplitude fluctuations of the postsynaptic response, it is also important to consider the rate of stimulation versus the rate of fluctuation in the functional status of the release site (see Discussion).

At Aplysia synapses, using variance–mean analysis of the fluctuations in amplitude of the evoked postsynaptic responses, we have previously demonstrated that the presynaptic inactivation of the small GTPase Rac by Clostridium sordellii lethal toxin (LT) blocks ACh release by reducing N without modifying either P or Q (Doussau et al. 2000; Humeau et al. 2002). We also found that when PTP-inducing protocols were applied to LT-poisoned synapses, in which few of the release sites remained active, PTPs of unusually large amplitude (PTP_LT, peaking at over 3 times the baseline) were observed (Doussau et al. 2000). Importantly, in these initial experiments, the extracellular [Ca²⁺]/[Mg²⁺] ratio was set to 0.42, which has been determined to correspond to an average release probability P of ∼0.4 (Humeau et al. 2002). Thus, because of the ceiling for P of 1, the possibility arises that the large changes in postsynaptic response amplitude that characterize PTP_LT are not due to modification of P. With the aim of better understanding the mechanisms underlying non-P-related forms of STP, we have investigated the cellular mechanisms of PTP_LT. To this end, we analysed the Ca²⁺ dependency, paired-pulse ratio and fluctuations in amplitude of the postsynaptic responses during the course of PTP elicited under control conditions (PTP_cont) or after LT injection (PTP_LT). Our findings indicate that PTP_cont is contributed by a change in the release parameter P, whereas PTP_LT is mostly contributed by a fast and transient increase in the release parameter N. Most likely, the change in N underlying PTP_LT is due to fast (i.e. within seconds) awakening of the release sites silenced upon Rac glucosylation. Altogether our finding suggest that Rac and the intracellular pathway which allows the bypassing of Rac are key players in new forms of short-term plasticity that rely on fast, activity-dependent changes in the functional status of the release sites.

Methods

Acetylcholine release and electrical recordings at Aplysia synapses

Aplysia californica were purchased from the University of Florida at Miami, USA. Electrophysiological experiments were performed at identified, chloride-dependent, inhibitory cholinergic synapses in dissected buccal ganglia of Aplysia as previously described (Doussau et al. 2000; Humeau et al. 2001a, b, 2002). In brief, two presynaptic cholinergic interneurons termed B4 and B5 and one postsynaptic neuron (either B3 or B6) were impaled with two glass microelectrodes (3 m KCl, Ag–AgCl₂, 2–10 MΩ). Action potentials were alternately evoked in B4 or B5 every 20 s (i.e. each presynaptic neuron was stimulated every 40 s; 0.025 Hz). The ensuing evoked, chloride-dependent IPSCs were recorded using the conventional two-electrode voltage-clamp technique (AxoClamp2B or Geneclamp-500, Axon Instruments) followed by digitalization (Digidata 1320A, Axon Instruments). The reversal potential of postsynaptic responses, V_rev, was determined every 5 min and IPSCs were recorded at a holding potential, V_h, maintained 30 mV hyperpolarized from V_rev. To express the IPSC amplitude as a value proportional to the amount of released ACh, but independent of the driving force for Cl⁻ ions, IPSC amplitude (I) was converted to apparent membrane conductance changes (G), according to the equation:

Although this has the dimension of a membrane conductance (nS), we refer to it in this paper as the IPSC amplitude.

When needed, trains of stimuli were generated by applying brief supraliminar depolarizing pulses of 5 ms separated by a repolarizing phase of adequate duration. Typically, post-tetanic potentiation (PTP) was initiated by two trains (50 Hz for 2 s) or four trains (50 Hz for 1 s), applied at 10 s intervals (Humeau et al. 2001a); 15 s after termination of the last 50 Hz train, the stimulation rate was returned to the initial rate (i.e. 0.025 Hz). The extent of PTP was expressed by normalizing the amplitude of the maximal IPSC during PTP to the ‘basal IPSC amplitude’ as determined by averaging the amplitudes of the 10 IPSCs preceding the application of the conditioning 50 Hz trains.

Extracellular media, intraneuronal injection, toxins and antibodies

Dissected buccal ganglia were maintained at 22°C using a Peltier-plate system and superfused continuously (10 ml h⁻¹) with a physiological medium containing (mm): NaCl 460, KCl 10, CaCl₂ 43.8, MgCl₂ 76.2, MgSO₄ 28, Hepes buffer 10, pH 7.5 as described (Humeau et al. 2001a, 2002). The very high concentration of divalent cations that we used was to prevent spontaneous neuron firing activity. This medium corresponds to an extracellular [Ca²⁺]/[Mg²⁺] ratio (abbreviated as [Ca²⁺/Mg²⁺]_e) of 0.42. When needed, [Ca²⁺/Mg²⁺]_e was modified, but the extracellular [MgSO₄] was kept unmodified, as previously described (Humeau et al. 2001a).

Intraneuronal injection of toxins was performed as detailed elsewhere (Humeau et al. 2001a, 2002). Since the injected volume was in the range of 1–2% of the cell body volume, we assumed that the final intraneuronal concentration was ∼1% of that in the injection micropipette.

Lethal toxin (LT) was purified from Clostridium sordellii IP82 (LT82) as previously described (Popoff, 1987; Humeau et al. 2002). Purified tetanus neurotoxin (TeNT) was a generous gift from Prof. Patrice Boquet (INSERM, Nice).

Data analysis

Description of neurotransmission based on the three quantal parameters: theoretical aspects

The variance–mean technique is a well-established method which allows distinguishing among changes in the quantal parameters (N, P or Q) by analysing the Var =f(I_mean) relationship (Silver et al. 1998; Reid & Clements, 1999; Humeau et al. 2001b, 2002; Scheuss et al. 2002; Clements, 2003; Silver, 2003; Foster & Regehr, 2004). The rationale for variance–mean analysis is the following. Consider a hypothetical synapse consisting of a single exocytotic site at which release of one SV with a given probability p produces a postsynaptic response of fixed amplitude q. According to the binomial statistics, at a release site, the average postsynaptic response amplitude is:

(1)

with variance:

(2)

Now, consider the N contacts between a presynaptic neuron and its postsynaptic target and assume that the release process at each site is independent of that at the other sites, and quanta sum linearly. The mean amplitudes and variances at the N sites add linearly, and the mean amplitude of the compound response is:

(3)

This can be rearranged as:

(4)

where P is the average release probability observed at the N sites and Q the average amplitude of the quantum of response at the N sites. The fluctuations of the responses around the mean have a variance of the form:

(5)

These simplified expressions do not allow the determation of actual N, P or Q because at real synapses the intra- and intersite variabilities in quantum amplitude and the heterogeneity in P at the N sites contribute to fluctuation in the amplitude of the postsynaptic response. For equations including these elements of variability see Silver et al. (1998); Reid & Clements (1999); Silver (2003); Clements (2003). However, the general form of the eqns (4) and (5) is preserved and therefore the graphical analysis of the relationship between Var and I_mean permits determination of which of the parameters N, P and Q is responsible for changing the amplitude of I_mean.

From eqns (4) and (5), it can be deduced that when only P is modified, Var =f(I_mean) takes the form of a simple parabola of initial slope Q:

(6)

whose initial slope and parabola extent allow determination of Q and N. When only N is changed, Var =f(I_mean) follows a linear function:

(7)

When only Q is modified, Var =f(I_mean) is a quadratic function of positive curvature:

(8)

Distinction between these three scenarios is possible by fitting the Var =f(I_mean) plots by eqns (6), (7) or (8). Practically, because of the spreading of the data, when P < 0.3, graphical discrimination of changes affecting N, P or Q is not possible. This is unambiguously feasible when the experiments are performed at an initial high release probability (i.e. P > 0.6 using very high [Ca²⁺/Mg²⁺]_e, Humeau et al. 2001b) provided that the changes in I_mean explore a high range of amplitude.

The linear expression is Var/I_mean=f(I_mean) (Heinemann & Conti, 1992; Royer et al. 2000; Humeau et al. 2001b).

When only P is modified:

(9)

It takes the form A−BI_mean in which the y-axis intercept and negative slope, B, allow estimation of Q and N by taking into account the other sources of variability.

When only N is changed:

(10)

which is independent of I_mean.

When only Q is modified:

(11)

It takes the form AI_mean with a positive slope, A.

The linear form Var/I_mean=f(I_mean) allows an easy graphical identification of the release parameters responsible for the observed I_mean changes, as does the analysis of the Var =f(I_mean) plots reported above. By normalizing Var to I_mean, this representation simplifies the pooling of data obtained from different experiments. Moreover, it allows discrimination between changes in N, P and Q from experiments performed under an initial release probability (i.e. [Ca²⁺/Mg²⁺]_e closer to the physiological condition than the very high probability required for analysing the Var =f(I_mean) plots; see comments above). Although these simplified expressions do not allow direct estimation of the release parameters, the graphical changes observed in the Var =f(I_mean) plots directly pinpoint the changes affecting the actual release parameters. For these reasons, we did not estimate the other sources of variability.

Practical determination of I_mean and Var

When average IPSC amplitude is not stable with time, the Var value determined for IPSC amplitude is contributed by both the fluctuations in amplitude of the IPSCs around their local mean and the changes in their mean amplitude during the time period considered. According to the previously described procedure (Humeau et al. 2001b, 2002), local mean and local fluctuations around it can be separated. Local fitting of a subset of n_mean subsequent I-values allows estimation of the corresponding local I_mean. Subtracting local I_mean from corresponding I generates ΔI values, which fluctuate around zero. Their Var is estimated for a subset of n_var subsequent ΔI values. This procedure is repeated along the whole data range to be analysed, with a step = 1. Then, the Var =f(I_mean) plots are constructed. The procedure we used did not allow estimating Var for the first and last n_mean/2 data. Because of the very fast onset of PTP_LT decay (see Results), calculation of local mean and Var using large n_mean and n_var led to information loss in the most prominent phase of PTP_LT. Therefore, n_mean and n_var were reduced to the smallest possible values, n_mean= 3 and n_var= 2 (note that in this case estimate of Var can be rewritten as (ΔI₁−ΔI₂)²/2). This introduced two problems: (i) reducing n_mean led to undervaluation of ΔI because local fitting included part of the local fluctuations and this undervalued Var; (ii) reducing n_var introduced a very large spreading of the Var estimates around the expected value (see also below). To facilitate the comparison of the Var =f(I_mean) plots made from the analysis of PTP_cont and PTP_LT, we used the same n_mean and n_var. When the inhibitory effects of toxins affecting NT release were analysed, the kinetics of IPSC amplitude changes were much slower than the PTPs and we used the optimized n_mean and n_var determined earlier (n_mean= 9 or 11 and n_var= 14–16; Humeau et al. 2002).

With the possibility in mind that Var was undervalued when the non-stationary analysis described above was applied to PTP_LT, we analysed, as an alternative, the fluctuations in amplitude of the ensembles of the same time-locked IPSCs in a series of subsequent PTPs. This approach has been successfully applied earlier by Scheuss et al. (2002) to separate pre- and postsynaptic contributions to synaptic depression at the calyx of Held synapse. According to eqns (6)–(8) for Var and eqns (9)–(11) for Var/I_mean, changing N, P or Q has different effects on Var and Var/I_mean and these differences may serve to identify which of N, P or Q is responsible for the observed I_mean change. As mentioned above, when the size n_var of the ensemble of data submitted to Var determination is small, the Var estimate displays high spreading around the expected Var. The problem has arisen of evaluating the significance of Var changes. The distribution of possible values of Var-estimate for samples of size n_var is a gamma-distribution similar to the probability density function of the reduced chi-square (${\chi }_V^2$) determined for the number of degrees of freedom v =n_var, but scaled to the mean Var. Practically, the probability density functions of the Var estimates, for samples of size n when N or Q are increased, were obtained by scaling the corresponding ${\chi }_V^2$ distributions by the expected Var (Var_N or Var_Q) calculated for the desired N or Q changes using eqn (5). However, in this calculation we neglected the contributions of the quantal intra- and intersite variability and heterogeneity in P among the release sites.

Pooling of the Var, I_mean, and Var/I_mean data

When Var =f(I_mean) displayed a parabolic shape, in order to pool the Var and I_mean data obtained from distinct experiments, individual Var =f(I_mean) plots were normalized to the maximum Var (Var_max) and corresponding I_mean (I_mean-to-Var_max) determined by fitting the Var =f(I_mean) plot from each experiment by a quadratic function of the form y =y₀+ax+bx², with all parameters left free, using the nonlinear regression procedure running under SigmaPlot 8 or 10 (Systat Software Inc.). Then, the normalized data from the n experiments were pooled and used to make the corresponding plots of normalized Var =f(normalized I_mean) and normalized Var/I_mean=f(normalized I_mean). When Var =f(I_mean) did not display a parabolic shape, I_mean and Var data were normalized to the average I_mean and average Var values determined under control conditions (i.e. prior to any treatment). In both cases, normalized Var/I_mean was calculated as the normalized Var/normalized I_mean ratio. Normalization of Var, I_mean and Var/I_mean data introduced loss of information on the actual amplitude of the release parameters, but preserved the determination of their relative changes. Indeed, when the normalized Var =f(I_mean) is a simple parabola, Var_max is reached for P = 0.5 as in non-normalized plots, allowing the determination of a corresponding P (e.g. I_mean= 0 →P = 0; I_mean=I_max/4 →P = 0.25; I_mean=I_max/2 →P = 0.5; …) at each point of the parabola. Initial slope and parabola extent do not allow the estimation of Q and N, respectively. Maximum extent of the parabola is twice the amplitude of I_mean-to-Var_max. In the corresponding normalized and linearized expression, Var/I_mean=f(I_mean), the y-axis intercept is obtained for twice the normalized Var_max/I_mean-to-Var_max.

Evaluation of the relationship between [Ca²⁺/Mg²⁺]_e and release probability

During numerous experiments, [Ca²⁺/Mg²⁺]_e was modified to manipulate the release probability, P. We observed that the relationship between P and [Ca²⁺/Mg²⁺]_e varies, possibly resulting from seasonal effects, ageing of Aplysia, or other factors. To allow comparison of the results obtained from a series of experiments performed in different periods of this long-term study, the P =f([Ca²⁺/Mg²⁺]_e) relationship was evaluated for homogeneous series of experiments as follows. During three to five experiments, [Ca²⁺/Mg²⁺]_e was modified (at least 3 different values). At each [Ca²⁺/Mg²⁺]_e, I_mean and Var were determined for ensembles of at least 20 subsequent IPSCs recorded when the plateau was reached, as previously described (Humeau et al. 2002). Then, I_mean and Var data were normalized to Var_max and I_mean-to-Var_max as described above. This allowed rough estimation of the average release P at any given [Ca²⁺/Mg²⁺]_e.

Data presentation

When needed, values obtained from various neurons were averaged and, unless otherwise stated, presented as the mean ±s.e.m. In several experiments, data were normalized using the mean value observed under control conditions (i.e. before any treatment). When appropriate (normally distributed data with equal variance), the significance between mean differences was tested by ANOVA or Student's t test. Comparisons of two estimates of Var were performed by the F quotient test. n.s. denotes no significant difference (i.e. P > 0.05). If not stated otherwise, the number n, mentioned when analysing the statistical significance of the comparison, refers to the number of independent sets of experiments (i.e. performed using different buccal ganglia) during which several identical STP protocols may have been generated. When needed, the number of the latter is also indicated.

Results

Post-tetanic potentiation at Aplysia synapse results from an increase in the release probability P

In each hemi-buccal ganglion of Aplysia, two identified presynaptic neurons (B4 and B5) make well-characterized cholinergic synapses with the same set of postsynaptic neurons (including B3, B6 and B8 cells; Doussau et al. 2000; Humeau et al. 2001a, 2001b, 2002). ACh release was monitored under control conditions by evoking single action potentials at a frequency of 0.025 Hz in either B4 or B5 neurons, and by measuring the amplitude of the IPSCs recorded in one of the postsynaptic neurons. The application of trains of stimulations at 50 Hz for a total duration of 4 s (4 × 1 s, Fig. 1Aa, or 2 × 2 s trains) elicited short-term enhancement in IPSC amplitude that we termed PTP_cont when elicited under control conditions. Typically, 15 s after the termination of the last 50 Hz conditioning stimulus, IPSC amplitude was 161 ± 5% of basal IPSC amplitude and then decayed with an average decay time constant τ= 11.5 ± 1 min (n = 32 experiments, Figs 1–2).

Figure 1.

PTP under control conditions and effects of manipulating [Ca²⁺/Mg²⁺]_e on PTP amplitude and PPR A, PTP induced under control conditions (PTP_cont). a, the PTP protocol. b, amplitude of IPSCs (n.s.) evoked at 0.025 Hz were plotted against time. At the times indicated by arrows, PTP protocols were applied. B and C, in a series of 6 experiments, paired IPSCs (interpulse time interval = 50 ms) were elicited at 0.025 Hz, except when the PTP protocols were applied. During the course of the experiments, [Ca²⁺/Mg²⁺]_e was increased from 0.42 to 2.1. In this series of experiments, these [Ca²⁺/Mg²⁺]_e were determined to correspond to an average P of ∼0.26 and 0.61, respectively. Ba, average amplitude of the first IPSCs of each pair was expressed as a percentage of basal values (i.e. before PTP_cont) recorded under [Ca²⁺/Mg²⁺]_e= 0.42. b, the extent of the PTP_cont recorded under [Ca²⁺/Mg²⁺]_e= 2.1 was normalized against that observed under [Ca²⁺/Mg²⁺]_e= 0.42, n = 6 experiments, 12 PTPs, P < 0.05. Ca, the average PPR (i.e. IPSC₂/IPSC₁) during basal stimulation and the course of PTP_cont under [Ca²⁺/Mg²⁺]_e= 0.42 and 2.1. b, average IPSC amplitude and PPR were determined under basal conditions with [Ca²⁺/Mg²⁺]_e= 0.42 (stimulation rate 0.025 Hz, no PTP). These values serve as reference for the subsequent normalization of data. The PPR and IPSC amplitude were also determined during the course of PTP_cont elicited in the presence of [Ca²⁺/Mg²⁺]_e= 0.42 or 2.1 and normalized against the reference values. Data (±s.e.m., filled symbols) were averaged by IPSC amplitude bin intervals of 5%. Open symbols: in a distinct series of n = 6 experiments, after reference values were determined under [Ca²⁺/Mg²⁺]_e= 0.42, [Ca²⁺/Mg²⁺] was increased up to 2.1 and progressively decreased to 0.14 (open symbols) and the IPSC amplitude and PPR were determined and normalized to the reference values.

Figure 2.

Analysis of Var to I_mean relationships during control PTP elicited under different [Ca²⁺/Mg²⁺]_e A and B, during a series of 4 experiments, PTP_cont were induced at 3 distinct [Ca²⁺/Mg²⁺]_e. The corresponding I_mean and Var were determined by using non-stationary analysis of the IPSCs elicited before (baseline) and during expression of PTP_cont. For each of the 4 experiments, the Var =f(I_mean) plots were made and the data were normalized to the Var_max and I_mean to Var_max determined by fitting each Var =f(I_mean) plot by the equation $y=y_0+ax+b{x}^2$ (which is a simple parabola when y₀= 0). Then the data of the 18 analysed PTP_cont (4 experiments) were pooled. A, the pooled normalized-Var to normalized-I_mean-to-Var_max plot. Small symbols denote data from PTP_cont. The filling colour refers to [Ca²⁺/Mg²⁺]_e: white = 0.14, grey = 0.42, black = 0.84. Continuous line denotes the fit of the data using the equation $y=y_0+ax+b{x}^2$ (the fitted parameters $(y_0=-2.8,P=0.6;a=2.05,P<0.0001;b=-0.01,P<0.0001)$ indicated a simple parabola of extent =−a/b≈ 2 times the I_mean to which Var is maximum). Large symbols denote the average Var and I_mean determined from the analysis of 8–10 baseline IPSCs (i.e. before PTP_cont). B, the pooled normalized-Var/I_mean to normalized-I_mean-to-Var_max plot. Same presentation as in A. Continuous line denotes the fit of the data using the equation $y=y_0+ax$. The fitted parameters $(y_0=1.97,P<0.0001;a=-0.01,P<0.0001)$ indicate a linear function starting at Var/I_mean≈ 2, with negative slope −1/(normalized N) =−0.01, which is expected to occur when only P changes; see eqn (7).

The PTPs analysed in various model systems are induced by a [Ca²⁺] rise inside the nerve terminal which occurs during, and sometime after, application of the tetanic conditioning stimulus. As a result of this, the release probability P is increased in a sustained manner (for a review see Zucker & Regehr, 2002). A first indication that this general rule applies to the studied synapses was the observation that relative PTP amplitude was decreased (Fig. 1B) when raising [Ca²⁺/Mg²⁺]_e. Indeed, increasing [Ca²⁺/Mg²⁺]_e results in enhancing P, and when initial P is set to high values (i.e. close to 1), P-mediated forms of plasticity occlude because of the ceiling for P of 1 (Humeau et al. 2003). To confirm that PTP_cont at Aplysia synapses was driven by an increase in P, we also analysed the paired pulsed ratio (PPR, i.e. IPSC₂/IPSC₁ amplitude) from IPSC pairs elicited at an interpulse interval of 50 ms during the whole duration of a subset of experiments. Indeed, PPR mainly depends on the residual intraterminal [Ca²⁺] and the number of available SVs at the time of the second stimulation (i.e. the SVs not consumed during the first stimulus added to the SVs replaced by refilling). In a series of five experiments [Ca²⁺/Mg²⁺]_e was sequentially adjusted to 0.42 and 2.1, which were determined corresponding to average P = 0.26 and 0.61, respectively. Under the [Ca²⁺/Mg²⁺]_e= 0.42 condition, PPR during PTP_cont was significantly diminished (at the peak of PTP_cont to 85 ± 4% of basal PPR; n = 6; P < 0.05) as compared to the basal condition (i.e. the PPR recorded immediately before the 50 Hz train application) (Fig. 1Ca). Moreover, independent of the initial value of [Ca²⁺/Mg²⁺]_e (0.42 or 2.1), the relationship between IPSC amplitude and PPR during PTP_cont was superimposable on the one observed by changing [Ca²⁺/Mg²⁺]_e from 2.1 to 0.14 (Figs 1–2). This indicates that PTP_cont is likely to result almost entirely from a transient increase in P.

To confirm the above deductions, we graphically analysed the relationships between variance and mean taken from the IPSCs elicited during PTP_cont. If indeed PTP_cont is due to an increase in P, according to eqn (6) (see Methods), the Var =f(I_mean) relationship should be a segment of parabola whose starting point depends on the initial P (i.e. [Ca^2+/Mg²⁺]_e). Thus, we performed a series of four experiments, during which [Ca²⁺/Mg²⁺]_e was sequentially adjusted to 0.14, 0.42 and 0.84, which corresponded to average P estimates of 0.10, 0.37 and 0.60, respectively. Under each [Ca²⁺/Mg²⁺]_e condition, when a plateau was reached, one or two PTP_cont were elicited. For each of the analysable 18 recorded PTP_cont, the local I_mean and Var were determined by non-stationary analysis. To allow comparison with the PTP elicited under other experimental conditions (see above sections in this paper), an estimate of Var was calculated from a set of two subsequent IPSCs, as described in Methods, at the price of a large scattering of the Var data. For each of the four experiments, the corresponding Var =f(I_mean) plot was made with the data taken from the PTP_cont recorded at the various [Ca²⁺/Mg²⁺]_e. The pooled normalized Var =f(I_mean) PTP_cont plot was well fitted by eqn (6) (continuous line in Fig. 2A), which is a simple parabola (fitting parameters are provided in the legend to Fig. 2). The data obtained for the PTP_cont recorded at a given [Ca²⁺/Mg²⁺]_e (denoted by different filling colour of the symbols in Fig. 2) explored only a fraction of the parabola extent but with a characteristic tangent at each [Ca²⁺/Mg²⁺]_e. For example, Var increased with I_mean at the low initial P, but decreased when I_mean increased at high initial P. The Var =f(I_mean) PTP_cont data were superimposed on those obtained from analysing the baseline IPSCs recorded before each PTP_cont (large symbols in Fig. 2). This indicates that Q and N are not significantly different during PTP_cont, as compared to baseline conditions. Altogether, these findings indicate that the changes in IPSC amplitude during PTP_cont result mainly, if not uniquely, from an increase in P.

Large amplitude and short-lived post-tetanic potentiations are elicited after Rac inactivation

After control recordings of IPSCs were made, purified LT was pressure-injected at a final intraneuronal concentration of ∼50 nm to inhibit the presynaptic functions of Rac GTPase. In agreement with earlier reports (Doussau et al. 2000; Humeau et al. 2002), ACh release started to decrease a few minutes after LT injection (Fig. 3A), and was completely blocked within 10 h. Non-stationary analysis of the fluctuations in amplitude of the IPSCs was performed all along the time course of LT action, and the corresponding Var/I_mean=f(I_mean) plots were made as described in Methods. Despite the strong decrease in I_mean, no significant change of Var/I_mean was induced by LT. Thus the Var/I_mean to I_mean relationship did not display any slope (see eqn (10) in Methods). Figure 3B summarizes the data from n = 12 neurons submitted to the action of LT. This analysis graphically indicates that the blockade of ACh release is due to a decrease of quantal parameter N, pinpointing a reduction in the number of active release sites when Rac is glucosylated. For further discussion on the functional meaning of a reduction in N, see Discussion. To ascertain for the discriminative properties of the Var/I_mean=f(I_mean) representation, a series of experiments were performed during which preparations were superfused with extracellular medium in which [Ca²⁺/Mg²⁺]_e was progressively decreased to modify the release probability, P. As expected, the Var/I_mean values were raised upon lowering [Ca²⁺/Mg²⁺]_e. Figure 3C shows the data pooled from n = 6 experiments and the corresponding Var/I_mean=f(I_mean) was best fitted by linear relationship with negative slope (see eqn (9)).

Figure 3.

PTP of very large amplitude at LT poisoned synapses A, representative experiment during which the amplitude of the IPSC was plotted against time, before and after pressure injection (arrow) of lethal toxin (LT; 50 nm final concentration). B, the IPSCs recorded during 12 experiments performed as in A were submitted to fluctuation analysis. I_mean and Var/I_mean values were normalized to the values determined during the control period (i.e. before LT injection) and the corresponding Var/I_mean=f(I_mean) plots were constructed. To indicate how Var/I_mean varies when N, P or Q are modified to change I_mean, three continuous lines are plotted using eqns (9)–(11) described in Methods. Unmodified Var/I_mean when I_mean decreases indicates a reduction in N as a result of LT action. C, same representation as in B, except that the IPSCs were recorded during experiments in which ACh release was depressed by lowering [Ca²⁺/Mg²⁺]_e from 2.1 to 0.14 (summary of 6 experiments). The negative slope of Var/I_mean=f(I_mean) when changing [Ca²⁺/Mg²⁺]_e indicates that P was diminished. D, same kind of experiment as in A, except that PTP protocols were applied at the times denoted by the small arrows. E, same experiment as in D but the amplitude of PTP_LT was normalized to the average basal IPSC amplitude between the PTP_LT, as described in text. F, average (±s.e.m.) relative amplitude of the PTPs recorded during 11 experiments performed as described in D. PTPs recorded during the control period (control) or 180 min after LT injection (180′ LT); were averaged after normalization to the average amplitude (100%) of 10 basal IPSCs recorded before each PTP (dashed line). The arrow denotes the application of the PTP protocol. ***P < 0.001; **P < 0.01; *P < 0.05.

When the 50 Hz conditioning stimulations were applied during the course of the LT-induced blockade of ACh release, PTPs of remarkably large relative amplitude were observed (Fig. 3D). We will refer to them as PTP_LT. To determine the relative amplitude and analyse the time course of the PTP_LT while the blockage of ACh release was not stabilized, the IPSC =f(t) plots were fitted (dashed line in Fig. 3D), but omitting the time periods corresponding to the PTP_LT (i.e. each for ∼30 min). Then, the relative IPSC amplitude was expressed by dividing the ensemble of the data points by the corresponding, time-locked, fitted values. On average, 180 min after LT injection, a time at which ACh release was reduced by about 75% (ranging from 66% to 90%, average: 74.3 ± 1.9%, n = 11 experiments), PTP_LT (i.e. first IPSC elicited 15 s after 50 Hz train) peaked at 439 ± 48% over baseline (n = 11 experiments, Fig. 3E and F, with PTP amplitudes ranging from 312 to 1127%), a value that was much higher (P < 0.001) than that recorded under control conditions (in the same set of 11 experiments, the PTP_cont peaked at 102 ± 19% over mean amplitude of basal IPSCs, Fig. 3F). In addition to the increase in the relative amplitude of PTP_LT, a prominent change in the time course of PTP was also noted: in contrast to most of the PTP_cont, none of the PTP_LT could be satisfactorily fitted with a monoexponential function. All PTP_LT were biphasic, with the appearance of very distinct fast and slow decay components – termed fPTP_LT and sPTP_LT, respectively – with τ_fast= 0.51 ± 0.05 min and τ_slow= 14.1 ± 2.8 min (n = 11, Fig. 3F) and weights w_fast= 322 ± 31% and w_slow= 117 ± 32%, accounting for 73% and 27% of the total PTP_LT amplitude, respectively (n = 11, Fig. 3F). It is noteworthy that by its time constant (τ_slow) and its relative amplitude (w_slow), sPTP_LT is reminiscent of PTP_cont. The decay time of PTP_cont (τ= 11.5 ± 1 min, n = 32, determined above) was not significantly different from that of sPTP_LT determined here (τ_slow= 14.1 ± 2.8 min, n = 11; P = 0.28). Thus, a likely possibility is that sPTP_LT corresponds to PTP_cont, but scaled by the reduction in N (i.e. to the fraction of the release sites that remain active after LT intoxination).

No fast PTP is observed when N is decreased by tetanus toxin

To determine whether the appearance of the fPTP_LT component was specific for the inactivation of Rac by LT, we performed a similar set of experiments using another potent bacterial toxin, the tetanus neurotoxin (TeNT). TeNT is a metalloprotease that specifically cleaves vesicle-associated membrane protein-2 (VAMP-2, also termed synaptobrevin-2), thereby preventing SV fusion (Schiavo et al. 1992). Consistent with our previous findings (Schiavo et al. 1992), presynaptic injection of TeNT (∼10 nm, final intraneuronal concentration) produced a fast and potent inhibition of ACh release (Fig. 4A). Similarly to the LT experiments, the Var/I_mean=f(I_mean) plots, obtained from non-stationary analysis of IPSC amplitude fluctuations during TeNT blocking action, showed no change in Var/I_mean (Fig. 4B) (see eqn (10)), indicating that TeNT induced a decrease in N. The interpretation of this finding will be considered in Discussion.

Figure 4.

Post-tetanic potentiation at TeNT-poisoned synapses A, the amplitude of the IPSC was plotted against time, before (open symbols) and after (filled symbols) pressure injection (arrow) of tetanus neurotoxin (TeNT) to reach a final concentration of 10 nm. B, the Var/I_mean=f(I_mean) plot made from 6 experiments during which TeNT blocked ACh release. Continuous lines are as described in Fig. 1B. Unmodified Var/I_mean when I_mean decreases indicates a reduction in N as the result of TeNT action. C, similar experiment as in A, except that PTP_TeNT were elicited (small arrows) during the TeNT induced blockade of ACh release. Note that, in contrast to the situation found after LT injection (Fig. 1D), the relative amplitude of PTP_TeNT was not increased. D, relative amplitude (%) of PTP_TeNT recorded during 6 experiments during the control period (open circles) or 100 min after TeNT-injection (100 min TeNT, filled circles). PTP_TeNT amplitudes were averaged after normalization to mean amplitude of the 10 basal IPSCs recorded before each PTP (dashed line). *P < 0.05.

We then induced PTPs during the course of TeNT-induced blockage (Fig. 4C). We will refer to these as PTP_TeNT. In contrast to the modification of PTP amplitude and shape observed after LT injection, the relative amplitude and time course of PTP_TeNT appeared grossly similar to those of PTP_cont (Fig. 4C and D). To allow the comparison with LT-poisoned synapses, we averaged the PTP_TeNT obtained ∼100 min after TeNT injection, at a time when ACh release was inhibited by 66–82% of the control value (72.3 ± 2.8%, n = 6), a level of blockade comparable to that observed 180 min after LT injection (∼75%, see above). However, instead of being potentiated, the relative amplitude of PTP_TeNT was slightly, but significantly, reduced as compared to the PTP_cont recorded before TeNT injection (PTP_{cont prior to TENT}; peak amplitude: 57.1 ± 6.2% over basal IPSC amplitude; PTP_TeNT: 36.6 ± 3.2%, n = 6, P < 0.05), and its time course remained strictly monoexponential. In this series of six experiments the constant of decay for PTP_{cont prior to TENT} was τ= 9.5 ± 0.9 min (not significantly different from that determined for the 32 PTP_cont analysed above; P = 0.38). The constant of decay of PTP_TeNT (τ= 9.7 ± 1.2 min, n = 6) did not significantly differ from that of PTP_{cont prior to TENT} (P = 0.89), other PTP_cont (P = 0.46) or the slow component PTP_LT (sPTP_LT, P = 0.28). These data indicate that PTP recorded at TeNT-poisoned synapses is analogous to PTP_cont subjected to a scaling effect determined by the extent of N blockade. Moreover, PTP_TeNT bears similarities to sPTP_LT. Since the TeNT-induced blockage of ACh release was characterized by a decrease in N similar to what occurs following LT injection, but without inducing large PTP, it is likely that the important changes in PTP_LT amplitude and time course (Fig. 3) are a specific consequence of Rac inactivation.

fPTP_LT results from an increase in N and sPTP_LT in P

To investigate the possible mechanisms underlying the fast and slow phases of PTP_LT (i.e. fPTP_LT and sPTP_LT), we first analysed the effects of manipulating [Ca²⁺/Mg²⁺]_e on the relative amplitude of these two PTP_LT components. Due to its fast kinetics, we assumed fPTP_LT to be prominent at the first IPSCs elicited 15 s after the 50 Hz train (denoted as b in Fig. 5A), whereas sPTP_LT was expected to dominate since the second IPSC elicited 55 s after the 50 Hz train (denoted as c in Fig. 5A). Experiments were performed under two [Ca²⁺/Mg²⁺]_e conditions, 0.42 and 2.1, which, in this series of experiments, we determined to correspond to an average initial P of ∼0.35 and 0.71, respectively, under control conditions. These values also applied after LT had blocked ACh release since LT does not affect P (Humeau et al. 2002). When [Ca²⁺/Mg²⁺]_e was changed from 0.42 to 2.1, the amplitude of baseline IPSCs was increased by a factor of 1.92 ± 0.17, but this did not induce any significant difference in the relative amplitude of fPTP_LT as deduced from the analysis of the first IPSC after the 50 Hz train (b in Fig. 5A). This is clearly distinct from the situation observed with PTP_cont (compare Fig. 5A with Fig. 1C). The relative amplitude of sPTP_LT phase, as deduced from the analysis of the second IPSC after the 50 Hz train, was clearly diminished upon raising [Ca²⁺/Mg²⁺]_e (c in Fig. 5A), similarly to what was observed since the first IPSC during PTP_cont (compare Fig. 5A with Fig. 1C). The latter effect is characteristic of the occlusion of P-driven STP when raising [Ca²⁺/Mg²⁺]_e, due to the ceiling for P of 1. This confirms our above deduction that sPTP_LT is likely to be similar to PTP_cont, but scaled at the extent of reduction in N induced by LT.

Figure 5.

Effects of manipulating [Ca²⁺/Mg²⁺]_e on amplitude of PTP_LT and PPR A, no occlusion of fPTP_LT amplitude when the initial P increases. In a series of experiments, after LT had inhibited ACh release by ∼70–80%, a first PTP_LT was elicited under [Ca²⁺/Mg²⁺]_e= 0.42 and a second after raising [Ca²⁺/Mg²⁺]_e to 2.1. Typical examples are illustrated in the inset. The average relative increase in amplitude of the first and second IPSC after the 50 Hz train, which are, respectively, dominated by fPTP_LT (b) or sPTP_LT (c), determined under [Ca²⁺/Mg²⁺]_e= 2.1 were normalized to the corresponding values determined under [Ca²⁺/Mg²⁺]_e= 0.42 and plotted (±s.e.m., n = 7 experiments in each condition, *P < 0.05). Error bars are smaller than the symbols. B, PPR changes during PTP_LT. In the same series of experiments, under [Ca²⁺/Mg²⁺]_e= 0.42, paired stimuli (50 ms time interval) were elicited before and after induction of PTP_LT (arrow, experimental protocol in the upper panel). Upper graph, a representative experiment during which the amplitude of the first IPSC of each pair (IPSC₁) was expressed as a percentage of the basal IPSC amplitude and plotted against time. Lower graph, the corresponding relative PPR amplitude changes, expressed as a percentage of the basal PPR. Traces illustrated in the upper inset correspond to the IPSC pairs denoted as a, b and c. Scale bars: 300 nS and 50 ms. In lower inset, the same IPSC traces were scaled to their respective IPSC₁. Note that, despite the large increase in IPSC amplitude at the peak of fPTP_LT (b), the extent of change in PPR is similar to that observed later during the slow phase of PTP_LT (c).

We also analysed PPR during fPTP_LT and sPTP_LT. We noticed that at the first IPSC pair obtained 15 s after the 50 Hz conditioning train (i.e. b in Fig. 5B) – that is when fPTP_LT is likely to be prominent – PPR, despite the strong increase in IPSC amplitude, was similar to that observed during the subsequent stimuli (c in Fig. 5B), a period which is dominated by the sPTP_LT phase. In fact, PPR at the first IPSC after the 50 Hz train was comparable to that of the corresponding time-locked pair in PTP_cont (compare Fig. 5B with Fig. 1B). Thus, we did not observe the large PPR fall that one would have expected if the IPSC amplitude rise, which characterizes fPTP_LT, was entirely due to an increase in P. The mild decrease in PPR that we observed during fPTP is likely to be due to the overlapping sPTP_LT phase. Indeed, during the subsequent sPTP_LT phase (Fig. 5B), PPR was diminished as compared to the basal condition (i.e. the PPR recorded immediately before the 50 Hz train application): sPTP_LT to 79 ± 5% of basal PPR, and this extent was close to what is observed during PTP_cont (85 ± 4% of basal PPR) or during the PTP elicited at TeNT-treated synapses (PTP_TeNT: to 86 ± 3% of basal PPR) (n = 3–6; all conditions versus basal: P < 0.05; other comparisons: n.s.). This indicates that P is similarly increased during the slow decaying phase of these three PTPs, whatever they had been elicited under control conditions or after neurotransmission had been blocked by LT or TeNT. To summarize, fPTP_LT largely escapes from the ceiling for P of 1, and the PPR changes during fPTP_LT do not follow the large increase in IPSC amplitude. Thus, fPTP_LT is unlikely to be mediated by a change in P. On the contrary, based on similar analytical criteria, sPTP_LT appears to be analogous to PTP_cont and likely to be driven by a transient increase in P.

If indeed fPTP_LT does not involve a change in P, the remaining possibilities are a strong increase in N, Q, or both. Although it may appear difficult to envisage that the quantum size may increase by an important factor (at least 4–10 to account for by PTP_LT), this possibility needs to be considered because Ca²⁺-dependent transfer of the exocytosis mode from a ‘kiss and run’ to full vesicular fusion has been described in chromaffin cells (Elhamdani et al. 2006) and central synapses (Aravanis et al. 2003). To confirm the deduction that fPTP_LT is not driven by a P increase and distinguish between the N versus Q possibility, we also analysed the relationships between Var and I_mean during the PTP_LT. This was addressed following two distinct approaches. The first one was similar in essence to that used for the PTP_cont (Fig. 2) and was based on non-stationary analysis of the fluctuations of IPSC amplitude during the course of several PTP_LT. Because the very fast fall in IPSC amplitude during PTP_LT may generate errors in estimating I_mean and Var by non-stationary analysis, we also used an alternative approach based on analysing the fluctuations in amplitude of time-lock IPSCs taken from a series of PTP_LT elicited during the course of an experiment.

Var to I_mean relationship during PTP_LT deduced from non-stationary analysis of the fluctuations of IPSCs during the course of PTP_LT

Four scenarios were considered.

(i) If PTP_LT is due to an increase in P, the Var =f(I_mean) should depict a parabola (as observed with PTP_cont) with the pooled Var =f(I_mean) data at a given [Ca²⁺/Mg²⁺]_e characterized by distinct tangent. Notably, Var should decrease when I_mean increases starting from a high initial P.

(ii) If PTP_LT is due to an increase in N, Var should linearly increase with I_mean (eqn (7)) independent of the initial [Ca²⁺/Mg²⁺]_e and Var/I_mean=f(I_mean) should be constant and independent of I_mean (eqn (10)). Moreover, because inhibition of ACh release is due to a fall in N (Humeau et al. 2002; Fig. 3), an increase in N would be functionally equivalent (not in terms of molecular mechanisms) to reverse the LT-induced blockage. The fit of Var =f(I_mean) taken from PTP_LT should tend to the I_mean and Var values determined before LT, while Var/I_mean should remain unmodified, at the same value determined before LT application.

(iii) If PTP_LT results from an increase in Q, Var =f(I_mean) should depict a quadratic function with positive curvature, pointing to a Var value much over the Var determined under control conditions (eqn (8)). Here also, this should occur independently of the initial P.

(iv) We also envisaged a scenario in which a substantial increase in P contributes to PTP_LT albeit not accounting for all the change in IPSC amplitude. The combination of an increase in both N and P should induce some negative curvature in the Var =f(I_mean) relationship, distinct from that of PTP_LT elicited under different initial P conditions.

Because of the impossibility of distinguishing between the initial slope of the parabola (i) and a linear function (ii) when P is low, only two [Ca²⁺/Mg²⁺]_e conditions were considered, [Ca²⁺/Mg²⁺]_e= 0.42 and 2.1 (corresponding to P∼0.35 and 0.71 in the following experiments). After ACh release had been blocked by 80–90% (i.e. at a stage in which the relative increase in the extent of inhibition during a PTP_LT is negligible with respect to the release changes due to PTP_LT mechanisms), one or two PTP_LT were induced. The use of small windows of data to calculate local I_mean and Var allowed analysis of most of the prominent fast decaying phase of PTP_LT. I_mean and Var were also determined before LT application. The pooled normalized Var =f(normalized I_mean) and normalized Var/I_mean=f(normalized I_mean) plots are shown in Fig. 6. They were fitted by linear and quadratic functions (y =y₀+ax and y =y₀+ax+bx², all parameters free). No improvement was obtained by fitting the data with a quadratic function (the fitting parameters are provided in the legend to Fig. 6). Extrapolation of the fits tends to reach the values determined before LT. Thus, during PTP_LT, and independently of the [Ca²⁺/Mg²⁺]_e used (i.e. of the initial P), Var linearly increased with I_mean (eqn (7)). This indicates that major changes in N underlie PTP_LT, at least in its most prominent phase (fPTP_LT). The contribution of the P driven component of sPTP_LT, deduced above (Fig. 4) could not be revealed by this analysis (but see next paragraph), either because of the high spreading of the Var data, or because the Var =f(I_mean) plot was dominated by fPTP_LT (N-driven phase) which, in contrast to sPTP_LT, explores a large range of IPSC amplitudes.

Figure 6.

Analysis of Var to I_mean relationship of IPSCs during the course of PTP_LT elicited under different [Ca²⁺/Mg²⁺]_e Several LT-poisoning experiments were performed at [Ca²⁺/Mg²⁺]_e of 0.42 (A) or 2.1 (B). During the course of these experiments, PTP_LT were elicited when ACh release was blocked by > 80% in order to minimize the impact on measurements of the fall in I_mean due to the LT-blocking that continues to develop during the course of PTP_LT. I_mean and Var were determined for baseline IPSCs elicited before injection of LT or during expression of the PTP_LT. The Var =f(I_mean) plots were made and the data were normalized to the Var and I_mean determined for the IPSCs evoked before LT injection. The data (A, 6 PTP_LT under [Ca²⁺/Mg²⁺]_e= 0.42 and B, 8 under [Ca²⁺/Mg²⁺]_e= 0.8) were then pooled. Aa and Ba, pooled normalized-Var to normalized-I_mean plots. Small circles denote PTP_LT data; large diamonds denote average (±s.d.) I_mean and Var determined from the analysis of IPSC before LT injection. The error bars of the baseline Var data illustrate how large is the spreading of the Var estimates when performed on series of 2 subsequent IPSCs. Continuous and dashed lines denote the fits, extended to the axis, of the pooled PTP_LT data using linear or quadratic functions, respectively. Simple positive slopes provided the best fits of the data (linear fitting: Aa: y₀=−6.06, P = 0.15; a = 1.24, P < 0.0001; Ba: y₀=−5.48, P = 0.12; a = 1.28, P < 0.0001; quadratic fitting: Aa: y₀=− 4.68, P = 0.53; a = 1.14, P = 0.02; b =+0.001, P = 0.82; Ba: y₀=−14.10, P = 0.13; a = 1.95, P = 0.006; b =−0.009, P = 0.32). This indicated that during PTP_LT, Var linearly increases with I_mean, with N being the only variable (eqn (7)). Ab and Bb, corresponding pooled normalized-Var/I_mean to normalized-I_mean plots. Continuous lines denote the fit of the data using the equation $y=y_0+ax$ (fitted parameters: $Ab:y_0=0.94,P<0.0001;a=0.002,P=0.72;Bb:y_0=0.80,P<0.0001;a=0.008,P=0.2$). Thus Var/I_mean is constant when varying I_mean (slope a is ∼nul) pinpointing the main contribution of the changes in N during PTP_LT.

Var to I_mean relationship during PTP_LT deduced from analysis of the fluctuations in amplitude of time-locked IPSCs in several PTP_LT

According to the binomial model for NT release, an increase in P or Q or N should result in similar I_mean (see eqn (4) in Methods), but not the same Var (eqn (5)). For example if I_mean increases by 10-fold following a 10-fold change in N or Q, the resulting Var and Var/I_mean will differ by a factor of 10 (according to eqn (5), Var_N increases linearly with N but Var_Q with Q². In Fig. 7A, we illustrate a computer simulation (binomial distribution of P, N) in which changes in I_mean and corresponding Var were modified following enhancement of N, P or Q using the initial average quantal parameters (N = 400; Q = 3; P = 0.38) that we previously determined at the same Aplysia model preparation using a [Ca²⁺/Mg²⁺]_e= 0.42 medium (Humeau et al. 2002). The blocking action of LT by 90% was mimicked by a 10-fold drop in N. According to eqn (10), simulated Var/I_mean was similar to under control conditions (Fig. 7Ab). Recovery of IPSC amplitude to initial values – as this occurs at the peak of fPTP_LT– was simulated by a 10-fold increase in N or Q. Full recovery of IPSC amplitude cannot be simulated by an increase in P because, starting from initial P = 0.38, ceiling for P of 1 does not allow an increase in P over than 1/0.38 (≈2.5). The simulated recovery of I_mean by enhancing Q leads to much larger spreading of the simulated IPSC amplitudes than when simulated by increasing N (Fig. 7Aa) and this is associated with a 10-fold higher Var and Var/I_mean than that calculated by increasing N 10-fold (Fig. 7Ab).

Figure 7.

Analysis of Var to I_mean relationship of time-locked IPSCs during the course of PTP_LT A, simulation of IPSC spreading, I_mean and Var/I_mean when N, P or Q is modified. Aa, ensembles of 50 IPSC amplitudes were simulated according to a binomial distribution of (P, N) and with quantum size Q. The control (Ctrl) average quantal parameters N = 400, P = 0.38 and Q = 3 used for simulation were taken from real experiments reported by Humeau et al. (2002). LT blocking action (LT) was simulated by a 10-fold drop in N. Three scenarios were considered to simulate a recovery of ACh release of similar amplitude to that observed during fPTP_LT: (i) 10-fold increase in N (open symbols) or (ii) in Q (filled symbols), and (iii) because of the ceiling of P at P_max= 1, the increase in amplitude of the simulated IPSC cannot recover control values. Note that the spreading of the simulated IPSC amplitudes is clearly distinct in each scenario. Ab, the corresponding simulated I_mean and Var/I_mean are plotted under control, blockade and recovery conditions. The simulated Var/I_mean allows clear distinction between N- or Q-based recoveries, although the corresponding I_mean are identical. B, analysis of real experiments. Ba, representative experiment during which the amplitude of the IPSCs was plotted against time, before (Ctrl) and after injection of LT (long arrow, 50 nm final concentration) had blocked ACh release by ∼90%. Then, 8 PTP_LT were elicited at 10 min time intervals (short arrows). I_mean, Var and Var/I_mean were determined from 10 min continuous recording epochs under control conditions (Ctrl) and after LT had blocked ACh release (b), or were calculated (b′) from the ensemble of the last IPSCs (each denoted by an open symbol) immediately preceding the 50 Hz trains (each denoted by an arrow). Bb, the amplitudes of the same time-locked IPSCs (i.e. IPSC₁, IPSC₂, …, IPSC_n) were reported. The panel displays the data from the 8 PTP_LT plotted in Ba. Bc, no significant difference in Var/I_mean determined from Ctrl and b periods indicates that LT blocked transmission by reducing N. The absence of difference between I_mean and Var/I_mean between b and b′ ensembles indicates there was no significant drift in basal release during the experimental period analysed. Bd, amplitudes of the ensembles of IPSC_n were normalized to the amplitude of the ensemble of the basal IPSCs recorded before the PTPs (i.e. b′). Then, Var and Var/I_mean values were calculated and averaged for 3 different experiments. Note that despite large changes in amplitude of the IPSC_n, the corresponding Var/I_mean remained similar. Dotted line denotes 100% (i.e. I_mean and Var/I_mean in b′ ensembles).

In the real experiments, IPSCs do not plateau at the peak of the PTP_LT. However, according to the procedure established earlier by Scheuss et al. (2002) at the calyx of the Held, an equivalent situation is provided when considering the ensemble of the same time-locked IPSCs taken from series of subsequent PTP_LT. Therefore, we analysed the fluctuations in amplitude of the ensembles of the same time-locked IPSCs elicited after the termination of the conditioning 50 Hz stimulus, in a series of PTP_LT elicited during the course of the same experiment. The limited life span of the experiments did allow only few PTP_LT to be analysed per experiment. From the distribution of the possible values of Var estimates (see Methods) we determined that a series of eight PTP_LT was sufficient to distinguish between changes in IPSC amplitude due to at least 10-fold increases in N or Q from analysing Var of the IPSC recorded at the peak of the PTP_LT. Indeed, using the average quantal parameters mentioned above, mean Var_N = 846.16 and mean Var_Q = 8481.6, and we calculated (see Methods) that if n = 8, Prob(Var_N < 2500) > 0.995 but Prob(Var_Q < 2500) < 0.05. In three different experiments, when LT had induced blockade of ACh release by more than 90% and reached a near stable level (i.e. when relative release changes due to LT action were marginal over a long period of time as compared to PTP_LT) (Fig. 7Ba), we elicited a series of PTPs. Possible error due to drift in the baseline (due to progress in LT action) was evaluated by comparing the Var/I_mean obtained from 24 subsequent IPSCs recorded after LT injection (denoted b in Fig. 7Ba) to that determined by analysing the ensembles of the last basal IPSCs acquired immediately before application of the 50 Hz trains (denoted b′ in Fig. 7Bb). Var/I_mean(b) and Var/I_mean(b′) were found not to differ significantly from the Var/I_mean determined under control conditions (denoted as Ctrl in Fig. 7Bb; Var/I_mean: Ctrl: 110 ± 56% of b-value; b′: 89 ± 38% of b-value; n = 3; n.s.,Fig. 7Bc). Then, Var/I_mean was determined for the ensembles of the same time-locked IPSCs (IPSCs of rank 1_, IPSCs of rank 2, etc.; Fig. 7Ba and b). On average, at the peak of the PTP_LT, amplitude of the IPSCs of rank 1 was increased ∼12-fold from basal level (1232%± 340%, 24 PTP_LT analysed taken from n = 3 experiments, P < 0.01, Fig. 7Bd), but the corresponding Var/I_mean remained not significantly different from that determined before LT injection (control) or after LT had induced blockade of ACh release (133 ± 39% of LT Var/I_mean value, n = 3 and 24 PTPs, n.s.). These findings clearly indicate that the main and earliest component of PTP_LT (i.e. fPTP_LT) is attributable to an increase in N. For the IPSCs of rank 2–5, during which both fPTP_LT and sPTP_LT overlap, we found an increase in Var/I_mean, as compared to the basal values. This pinpoints that a substantial increase in P also occurs during the course of PTP_LT. However, this effect develops later or more slowly than the increase in N. This is fully consistent with the above findings showing that significant saturation at high [Ca²⁺/Mg²⁺]_e occurs since the second IPSC during PTP_LT and is associated with PPR decrease (Fig. 5A and B). Collectively, the above-described data suggest that PTP_LT is a combination of a fast prominent early increase in N (fPTP_LT), combined with a P-driven component which develops later or more slowly (sPTP_LT). The sPTP_LT component is likely to correspond to PTP_cont, but scaled to the observed changes in N.

Timing and induction properties of fPTP_LT

To determine the time courses of fPTP_LT and the time point of initiation of the N-changes that drive fPTP_LT, ACh release was monitored at higher frequency after the conditioning 50 Hz train, by evoking IPSCs at 2 Hz (Fig. 8A). Because some fPTP_LT can induce very early, we also reduced the duration of the 50 Hz train to 1 s. We found that fPTP_LT was significantly induced since the first IPSC after the train and peaked at the second to third IPSC after the 50 Hz train (i.e. 1–1.5 s after it). Fast synaptic depression can be induced at 2 Hz at this synapse (Humeau et al. 2001a). Thus, to evaluate whether the fPTP_LT time course determined at 2 Hz was altered by the overlapping depression, we performed a set of three experiments during which only single IPSCs were elicited after each 50 Hz train, but at different time intervals from the end of the train (see the typical experiment illustrated in Fig. 8B). Again, we found fPTP_LT peaking at about 1–2 s after the 50 Hz conditioning train, indicating that monitoring fPTP_LT at 2 Hz does not greatly alter its kinetics. It is also possible that synaptic depression affects P-driven changes in synaptic efficacy, but not those due to a modification in N. We also explored the post-50 Hz train period at control or TeNT-treated synapses by monitoring transmission at 2 Hz and found that, on average, the 50 Hz trains were followed by a short-lived synaptic depression preceding the onset of sPTP_TeNT phase (Fig. 8A, TeNT), as observed under control conditions (Fig. 8A, Ctrl).

Figure 8.

Timing and induction properties of fPTP_LT A, in a series of experiments, a modified PTP protocol was applied to induce PTP under control conditions or after ACh release was blocked by about 80% following injection of LT (n = 10) or TeNT (n = 5). Conditioning 50 Hz trains were of 1 s duration, and ACh release during the post-tetanic period was monitored at 2 Hz for 8 stimuli. The amplitudes (mean ±s.e.m.) corresponding to IPSC₁ to IPSC₅₀ during the 50 Hz train and to the ensuing 8 post-tetanic IPSCs were normalized to the average IPSC amplitude determined under control period (i.e. before injection of LT or TeNT). B, alternative experiment during which a single IPSC was elicited at various time intervals (protocol in upper inset) after termination of the conditioning 50 Hz train at LT-poisoned synapse. Graph: IPSC amplitude (mean ±s.e.m.; n = 3 experiments) expressed as a percentage of the IPSC amplitude measured before the 50 Hz train. C, representative train responses (50 Hz for 2 s) recorded before or after LT injection. Scale bars: left traces, 200 nS and 0.5 s. D, average amplitudes of the IPSCs during 50 Hz train responses were normalized to the average amplitude of the first IPSC in the trains recorded before LT injection (n = 12 experiments, ±s.e.m.). Error bars can be masked by the symbols. E, the IPSCs recorded during and after the application of trains of 50 stimuli at the indicated frequencies (10, 25 and 50 Hz). The 50 Hz train was also administered for shorter durations (0.25 and 0.5 s). After application of each train, 8 IPSCs were monitored at 2 Hz. F, average amplitudes (±s.e.m.; n = 6) of the IPSCs during the train responses elicited at the indicated frequencies. Before averaging, the amplitude of IPSC_n was normalized to that of IPSC₁ in the train. G, average amplitude of IPSC₅₀ in trains produced at 3 frequencies were normalized to that of IPSC₅₀ at 50 Hz and plotted against the stimulation frequency. H, average amplitude at the peak of fPTP_LT (i.e. amplitude of IPSC₂ or IPSC₃ after train) plotted against duration of the 50 Hz train.

Examination of Fig. 8A and B suggested that fPTP_LT was the temporal continuation of a process initiating during the 50 Hz conditioning train. When the amplitude of the IPSCs in the 50 Hz train response recorded at LT-poisoned synapses are normalized to the corresponding time-locked IPSCs in train responses recorded in the control period, full recovery is observed after ∼1 s (Fig. 8A, LT, C and D): 180 min after LT injection, the amplitude of the first response in the train was 28 ± 3% of the corresponding response in control trains (P < 0.01, n = 12 experiments), but the amplitude of the 50th response in the train was 100 ± 6% of control (n.s., n = 12 experiments). No similar recovery effect occurred at TeNT-poisoned synapses (Fig. 8A, TeNT) or after inactivation of the PLD1 function (Humeau et al. 2001b), despite the fact that the blockade of ACh release by these treatments manifests as a decrease in N. Thus, similar to the presence of the fast PTP phase (fPTP_LT), the ability to recover from blockade during a 50 Hz train appears to be specific for the synapses in which Rac has been inactivated by LT.

Next, we determined the minimal neuronal activity required to induce release recovery from blockade during a 50 Hz train and ensuing fPTP_LT. To this end, we first compared the relative efficacies of 50 stimuli trains elicited at distinct frequencies to elicit release recovery (Fig. 8E, 3 left recordings, and F). Frequencies far over 50 Hz could not be tested because of the fast accommodation of action potentials and excitation failures. When the relative average amplitudes of the 50th responses in the train were plotted against the stimulation frequency, the x-axis intercept of the linear extrapolation of the data indicated that 3 Hz was the minimal frequency required to trigger a detectable recovery of release (Fig. 8G). We also compared the efficacies of 50 Hz trains of various durations to induce fPTP_LT. When the relative amplitudes of fPTP_LT were plotted against duration of the 50 Hz stimulations, the x-axis intercept of the linear extrapolation of the data indicated that a minimal duration of 150 ms at 50 Hz (i.e. 7–8 stimuli) is needed to trigger significant NT release recovery at the LT-poisoned synapses (Fig. 8H).

Do fast changes in N may underlay STP under control condition?

The question has arisen of whether fPTP_LT is irrelevant to ‘normal’ synaptic physiology or, on the contrary, the over-expression of a new form of STP, based on transient changes in N. To this end, we examined the possibility of that the counterparts of the fPTP_LT and recovery from LT blockade during 50 Hz trains may be expressed under control conditions. To focus on the early synaptic events occurring just after the 50 Hz trains elicited under control conditions, and then analysing later events, we used a modified protocol: ACh release was monitored after the conditioning 50 Hz train by evoking exocytosis at 1 Hz for four stimuli (expected to be the optimal time window for expression of a fast-PTP phase analogous to fPTP_LT, while minimizing synaptic depression), then, 15 s after the 50 Hz train, the stimulation rate was returned to 0.025 Hz for a further four stimuli (to monitor the peak of the ensuing PTP_cont) (Fig. 9A). The experiments were performed under two [Ca²⁺/Mg²⁺]_e conditions, 0.42 and 2.1, respectively. As observed above (Fig. 9A), the 50 Hz trains were followed by a very short-lived synaptic depression, then potentiation developed (grey area in Fig. 9A). Importantly, the relative changes in amplitude of the IPSCs recorded during the first 4 s after the train were unaffected by manipulating [Ca²⁺/Mg²⁺]_e. The relative amplitude of the fourth IPSCs (i.e. elicited 4 s after the train), which is significantly potentiated as compared to the basal IPSCs recorded before the 50 Hz trains, was unmodified under low [Ca²⁺/Mg²⁺]_e= 0.42 (152 ± 27%) or high [Ca²⁺/Mg²⁺]_e= 2.1 (144 ± 21%, n = 6, comparison low versus high [Ca²⁺/Mg²⁺]: n.s.; see Fig. 9A). This was no longer the case 15 s after the 50 Hz stimulation train, and significant occlusion in potentiation was detected (relative amplitude of IPSC at 15 s under low [Ca²⁺/Mg²⁺]_e= 0.42: 145 ± 9%; under high [Ca²⁺/Mg²⁺]_e= 2.1: 120 ± 7%, n = 12 neurons in each condition, P < 0.05, Fig. 9A). These results indicate the presence of an early phase of potentiation, independent of P, and suggest the existence of a functional counterpart of fPTP_LT under control conditions. This fPTP_cont may be mediated by a change in N; afterwards, since the 15th second after train, the P-driven slow PTP_cont becomes prominent as pinpointed by its occlusion at the higher [Ca²⁺/Mg²⁺]_e.

Figure 9.

A fast PTP (fPTP) is expressed under control conditions A, lack of occlusion in amplitude of early phase of PTP_cont when [Ca²⁺/Mg²⁺]_e is elevated pinpoints the presence of an fPTP phase analogous to fPTP_LT. Upper inset, the stimulation protocol. Experiments were repeated at low (0.42) and high (2.1) [Ca²⁺/Mg²⁺]_e. Left graph, the IPSC amplitudes, normalized to the mean IPSC amplitude observed before application of the 50 Hz train, were averaged and plotted against time. Right graph, the corresponding ratio IPSC_high-Ca/Mg/IPSC_low-Ca/Mg. Note that despite clear potentiation, amplitude of PTP during the early phase (i.e. for 4 s) did not occlude with the high [Ca²⁺/Mg²⁺]_e. *P < 0.05; n = 6. B, left, representative train response (50 Hz for 1 s) before (black), and 180 min after (grey) LT injection. Scale bars: 200 nS and 0.5 s. Right graph, same data as in Fig. 8A, but after normalization to the average amplitude of the first peak (arrow) in the trains recorded before or after LT had blocked ACh release by ∼80% (n = 10 experiments, ±s.e.m., errors bars may be masked by the symbols). Bottom, the difference between the two plots. C, similar experiments were performed under control conditions (Ctrl). However, the epochs (train response + 8 post-tetanic IPSCs) were sorted on the basis of the presence of postsynaptic depression (black symbols) or no depression/presence of fast potentiation (grey symbols). This correlated with absence of rebound (Rbd⁻) or presence of rebound of ACh release (Rbd⁺, arrow) during the 1 s at 50 Hz train, respectively. Bottom plot, the difference between the two plots. Typical recordings are displayed at left. Scale bars: 200 nS and 300 ms.

To end, we wanted to determine whether a counterpart of the ‘recovery from LT-blockade’ exists also under control conditions. The two main characteristics of the 50 Hz train responses and ensuing IPSCs (elicited at 1 Hz) recorded after LT poisoning are the marked release recovery phenomenon occurring within the 50 Hz trains and significant potentiation (i.e. fPTP_LT) since the first IPSC after the 50 Hz train. On average, under the control conditions, synaptic depression predominates immediately after the 50 Hz train. Importantly, the early post-tetanic period was characterized by a high scattering in the data (for example see IPSCs 1–4 in Fig. 9A). We found this was due to the fact that some neurons, instead of experiencing post-tetanic depression, exhibited a clear potentiation after the end of the 50 Hz train. Moreover, this was correlated with the presence of prominent ‘rebound’ of ACh release during the 50 Hz train response. This rebound was concomitant – on a time scale basis – with the recovery phase observed at LT-poisoned synapses (Fig. 9C; arrow in the left bottom trace). On the basis of the presence of an immediate potentiation or a synaptic depression after the 50 Hz train, we sorted the control experiments (i.e. neurons) into two groups. In the first one, the post-tetanic IPSCs were potentiated after the first IPSC after the train (to 120 ± 13% of baseline, n = 11 experiments), giving rise to fPTP_cont. This group was characterized by clear rebound (denoted as Rbd⁺) during the 50 Hz train. The second group (with the same size) was selected at random among the experiments displaying post-tetanic depression after the first IPSC after the 50 Hz train (77 ± 11% of baseline, n = 11, P < 0.05 as compared with the former group, Fig. 9C). In this second group, the rebound of ACh release was reduced (denoted as Rbd⁻).

Subtracting the average recordings (LT-treated minus control, or Rbd⁺ minus Rbd⁻, denoted as ‘Diff’ in Figs 9B, C) revealed the similarities in time course of the ‘recovery from LT’ and ‘rebound’ or fPTP_LT and fPTP_cont. This supports the notion that, at control synapses, there is a functional counterpart to the N-change-driven STP identified at LT-poisoned synapses.

Discussion

A new form of STP based on transient changes in N

We have analysed here an unusual form of PTP (i.e. PTP_LT) of very large relative amplitude and short lifetime, which is expressed after that transmitter release has been diminished by Rac inactivation, but not by other means such as SNARE cleavage by TeNT or reduction in [Ca²⁺/Mg²⁺]_e. Using PPR, saturation experiments, Var/I_mean analysis, we have shown that the PTP_LT elicited after Rac inactivation and PTP triggered under control conditions comprise two phases: an early one, fPTP_LT, driven by fast transient increase in N, and a slow decaying one, sPTP_LT, driven by delayed enhancement of P which is likely to correspond to the well-documented main component of PTP_cont. At LT-poisoned synapses, fPTP_LT is prominent whereas the slow decaying PTP phase is the main form at naïve synapses. A functional counterpart of fPTP_LT appears to exist in naïve synapses, at least in a subset of them, as identified by the presence of a modest fPTP_cont component. Overall, our data reveal a new form of activity-dependent and fast expressing STP (i.e. fPTP_LT and fPTP_cont), which is based on switching off and on the functional status of the release sites. This is reminiscent of the mechanisms proposed for 5-HT-induced facilitation at cultured Aplysia sensorimotor synapses (Royer et al. 2000), and recruitment of release sites during expression of presynaptic LTP as proposed at the mossy fibres in hippocampus (Reid et al. 2004). However, the main difference resides in the time scale for the changes in N: ∼1–10 s in the fPTPs that we described here, as compared with several minutes in the previous studies (Royer et al. 2000; Reid et al. 2004).

Physiological meaning of a change in N

The relevance to synapse physiology of the changes in N detected by fluctuation analysis awaits discussion. The N and P parameters obtained by fluctuation analysis do not fully fit their definitions in the quantal release hypothesis. Indeed, determination of P changes requires that the sampling (stimulation) rate is far below the temporal rate of stochastic fluctuations in P. This may not be the case for all components of P. P is compound and is the product of P_A (the chance that a release ready SV is available at the release site) and P_O (the release chance or output probability, if there is a SV ready to fuse; Quastel, 1997). The recent notion that the release site per se also needs maturation to become competent for mediating SV fusion (Humeau et al. 2001b; Bader et al. 2004) should be also taken into account by including another P term or extending the notion of P_A to the plasma membrane events. A high rate of temporal fluctuation in P_O is likely because the stochastic opening and closing of Ca²⁺ channels and Ca²⁺ binding to the Ca²⁺ detectors responsible for triggering SV fusion operate in a time scale faster that the sampling rate. This is likely not to be the case for all events referred to under the notion of ‘P_A’. Indeed, refilling and priming mechanisms have a relatively slow kinetics and operate in the second time scale (Wang & Kaczmarek, 1998; but see Saviane & Silver, 2006). Moreover, in the cascade of events that occur during priming, certain steps may reach a quasi-irreversible state (e.g. SNARE complex zippering), thus preventing stochastic fluctuations and ‘fixing’P_A at certain sites. The changes in ‘N’ detected by fluctuation analysis are likely to refer not only to the mechanisms regulating the number of morphological release sites (true N), but also some ‘P_A’-related molecular events that do not change their on-off status fast enough as compared to the sampling rate. N, as re-defined here, corresponds to the average number of entities (SV paired to a release site) capable of undergoing fusion (i.e. susceptible to P_O), which is close to the definition of the RRP. Therefore, the changes in N that we have detected are likely to refer to modifications of the RRP, but without distinguishing the vesicular events from those affecting the release site per se. Our finding that ACh release blockade by TeNT results in a decrease in N is consistent with the above interpretations of N. Indeed, when SVs with severed VAMP-2 enter the exocytotic process, they are likely to plug the fusion sites leading to their long-lasting functional inactivation (Humeau et al. 2000).

fPTP_LT is not due to transient removal of the release blockade exerted by glucosylated Rac

The blocking action of LT results in a decrease in N, whereas fPTP_LT is driven by an increase in N. Therefore, the question arises of whether the on and off phases of fPTP_LT are attributable to transient removal of the inhibition exerted by glucosylated Rac or to other mechanisms. First, a functional counterpart of fPTP_LT exists at naïve synapses (viz the fPTP), albeit we cannot exclude the possibility that this is coincidental. Second, glucosylation of Rac inhibits GTP hydrolysis stimulated by the GTPase activating proteins (GAPs), and Rac remains GTP-bound, unable to cycle. Interaction of glucosylated Rac with the guanine nucleotide dissociation inhibitor (GDI) is impaired, preventing its extraction from target membrane. Moreover, glucosylation of LT targets is irreversible in vitro (Just & Boquet, 2000; Schirmer & Aktories, 2004) and in vivo. Indeed, the intracellular application of UDP-mannose, which acts as a competitive inhibitor for LT, freezes the blocking action of LT, without recovery of ACh release, at least during the subsequent 3 h (Humeau et al. 2002). Application of PTP_LT protocols at LT-poisoned neurones subsequently injected with UDP-mannose (1 mm) did not produce any significant ACh release recovery or changes in PTP_LT amplitude or kinetics (data not shown). Therefore, once glucosylated, Rac appears to lock off the downstream pathways (we discuss their identity here below) in a long lasting manner and fPTP_LT is unlikely to result from a transient removal of the release inhibition. Possibly, during the 50 Hz train, a functional bypass of the step blocked by glucosylated Rac is activated, allowing SV–release site priming at the silenced release site. In other words, the 50 Hz trains are able to awake the release sites which have been silenced by inactivation of Rac by LT toxin.

How does Rac control functionality of release sites?

Our data provide insights on the molecular mechanisms involved in the regulation of N, as defined above, and pinpoint a key role for Rac. This GTPase is present on SV and plasma membranes and plays a role in exocytosis in various cell types including neurons (Doussau et al. 2000; Humeau et al. 2002; Morciano et al. 2005; Takamori et al. 2006; this paper) and neuroendocrine cells (Li et al. 2003). Whether SV-bound or plasma membrane-associated Rac plays a role in exocytosis remains unclear. In cells, Rac, as the Rho-GTPases, is largely distributed in many subcellular domains, where, depending on their upstream and downstream interactors, it regulates a variety of cell functions. At nerve terminals, the interactor(s) and effector(s) underlying Rac function in exocytosis are still largely unknown. One such candidate is the GIT/PIX complex identified in spine morphogenesis and synapse formation (Premont et al. 2004; Zhang et al. 2005). In this complex, PIX acts as a guanine nucleoside exchange factor (GEF) and catalyses GDP to GTP exchange on Rac, while GIT is a GTPase-activating protein (GAP) of the ARF6 GTPase that controls production of fusiogenic phosphatidic acid by phospholipase D1 (PLD1) at the release sites (Vitale et al. 2001; Premont et al. 2004; Meyer et al. 2006). Thus, Rac participates in one of the pathways converging on PLD1, which is established to control membranes fusion (Vitale et al. 2001; Humeau et al. 2001b; reviewed by Bader et al. 2004).

The most important consequence of the glucosylation of Rac by LT is its uncoupling with downstream effectors. Moreover, glucosylation inhibits its interaction with GAPs needed for GTP hydrolysis and termination of the Rac cycle, and with GDI, which allows its extraction from membrane into cytosol (for review see Schirmer & Aktories, 2004). We propose that the blocking action of LT results from the silencing of the above-described pathway converging on PLD1. This is supported by the report that PLD1 activity is diminished following LT application (El Hadj et al. 1999).

Like all Ras-related GTPases, Rac cycles between a GDP-bound inactive and GTP-bound active state. The lifetime of the GTP-bound state of Ras-related GTPases is believed to serve as a timer that determines the activation time of the biological function they control including membrane fusion (Rybin et al. 1996; Li & Qian, 2002). For example, Rab5–GTP has been proposed to act as a timer that determines the frequency of membrane docking/fusion events (Rybin et al. 1996). Likewise, Rac cycling may act as a rate limiting timer for priming of the SV–release site and to regulate N this way. The possibility arises that when the stimulation frequency is over the cycle rate of Rac or its regulators, priming of the release sites should not occur fast enough to sustain exocytosis in such a way that a fraction of the release sites become silent, giving rise to a decline in the amount of released neurotransmitter. Consistently, synaptic depression induced by increased stimulation rate at cultured sensorimotor Aplysia synapses is due to a fall in N as assessed by examination of Var/I_mean (Royer et al. 2000). We also observed that both N and P decline during synaptic depression at the synapse studied here (our unpublished observations). In this line, the small weight fPTP_cont component driven by an increase in N that we observed at naïve synapses may be due to the awakening of the release sites that are silenced early during the 50 Hz trains.

Possible mechanisms involved in fPTP_LT and fPTP

High-frequency stimulation of the presynaptic terminal can accelerate (in the order of 100 ms or less) recovery from synaptic depression via a Ca²⁺- and calmodulin-dependent process (Dittman & Regehr, 1998; Wang & Kaczmarek, 1998; Dittman et al. 2000; Sakaba & Neher, 2001). Consistent with such a possibility, we previously observed that intraneuronal application of EGTA prevents recovery of ACh release during the 50 Hz trains elicited at LT-treated synapses (Doussau et al. 2000). The fact that at LT-treated synapses, manipulation of extracellular Ca²⁺ in ranges that clearly affect ACh release efficacy does not alter recovery from blockade during the 50 Hz train (Doussau et al. 2000) or induction of fPTP_LT indicates that the molecular actors involved in the bypass of Rac are distinct from the Ca²⁺-dependent release trigger(s).

With regard to the residency time of ∼2–5 s needed for a SV to be fully primed at active mammalian synapses (Klingauf et al. 1998; but see Saviane & Silver, 2006) and the fast kinetics of the onset of recovery during 50 Hz trains (Doussau et al. 2000; this paper), it is tempting to speculate that the SVs contributing to recovery and fPTP_LT are recruited from an already docked pool, and that the LT-awakening mechanisms are likely to operate during a late stage of the exocytotic process.

To summarize, the molecular mechanisms involved in the bypassing of Rac are likely to be involved in the SV–fusion site priming, and to exhibit a Ca²⁺-dependent activity. At this stage, only hypotheses can be considered. Ral GTPase is present on the SVs and plays a role in Ca²⁺-dependent NT exocytosis, possibly by regulating RRP refilling (Polzin et al. 2002; Owe-Larsson et al. 2005). It can be activated by Ca²⁺–calmodulin (Park, 2001) and participates in the same multicomplex as Rac, and regulates PLD1 activity (Vitale et al. 2005). However, Ral serves as a substrate for LT (Just & Boquet, 2000; Humeau et al. 2002; Schirmer & Aktories, 2004) and its functioning is likely to be severed at LT-poisoned synapses. Another potential candidate is synapsin, whose functions are Ca²⁺ and calmodulin dependent through phosphorylation by CaM kinases or protein kinase A (reviewed by Hilfiker et al. 1999). Indeed, at the same Aplysia synapses used here, neutralization of synapsin suppresses the rebound of ACh release during the 50 Hz train, induces fast and marked postsynaptic depression (i.e. suggesting abolition of fPTP), and also prevents expression of PTP (Humeau et al. 2001a). While the latter effect is attributable to to the well-documented role of synapsin in the Ca²⁺-dependent control of reserve SV trafficking (reviewed by Hilfiker et al. 1999), the two former effects imply a release ‘boosting’ function of synapsin, operating at the plasma membrane level during the postdocking steps of exocytosis (Hilfiker et al. 1998, 2005; Humeau et al. 2001a; Sun et al. 2006; Fassio et al. 2006). Moreover, synapsin activates PI 3-kinase and regulates the RRP (Cousin et al. 2003); it is worth noting that some of the lipid products of PI 3-kinase are known activators of ARF6/PLD1 (Vitale et al. 2001).

In conclusion, we have identified a new form of STP based on the fast regulation of the on/off switching of the functional status of the release sites, which implicates Rac and as yet undeciphered actors. This finding extends to STP the concept of recruitment of silent synapses or release sites that has been proposed as an expression mechanism for long-term potentiation at mammalian synapses (Voronin & Cherubini, 2003; Reid et al. 2004; Groc et al. 2006). With regard to synapse physiology, a likely consequence of any change in N (without precluding any cause for it) is the scaling of P- or Q-based STPs to the same extent as the N change during the time period when N is reduced. An example is provided here with the PTP recorded at TeNT-poisoned synapses, which decreases N. The above-mentioned possibility that, depending on the stimulation rate, release sites can become silent because of the rate limiting role of small GTPases needs further investigation with the perspective of identifying new rules in synaptic functions. The presence of STP forms driven by change in N should affect the dynamic range, and possibly the computational properties of the synapse.