Optical Postsynaptic Measurement of Vesicle Release Rates for Hippocampal Synapses Undergoing Asynchronous Release during Train Stimulation

A switch to asynchronous release during trains of stimulation

We have used autaptic microisland cultures to directly measure the rate of glutamate release and its postsynaptic action at individual synapses after trains of stimulation. In 3 mm extracellular Ca²⁺/1 mm Mg²⁺, neurons showed robust depression (>90%) of evoked synaptic responses after only 0.5 sec of 20 Hz stimulation (Fig. 1A). After trains of evoked synaptic stimuli, a large inward current and decaying tail were observed that were composed of asynchronously released quanta. These developing neurons in autaptic culture (4-10 d in vitro) undergo robust asynchronous release without the requirement to use Sr²⁺ or other agents to prolong time course of release (Abdul-Ghani et al., 1996). Previous studies as well as our own work indicate that asynchronous release is dependent on elevated intraterminal residual Ca²⁺ concentration because it is selectively blocked by treatment with the relatively slow chelator EGTA-AM (Cummings et al., 1996; Atluri and Regehr, 1998; Otsu et al., 2004). Given that the individual AMPAR mEPSCs making up the tail have relatively rapid decay kinetics (<5 msec), the decay of the tail current reflects the time course of asynchronous release frequency. We assume that the tail current is derived primarily from the summation of individual quanta with decay constants much shorter than the time course of asynchronous release (Otsu et al., 2004). The mean amplitude of the tail current correlated linearly with its variance over the first 1.5 sec after train simulation (Otsu et al., 2004), indicating that the tail current was composed of individual events that have a similar mean amplitude over time. If the tail current is made up of linearly additive elementary components that occur at random intervals, the amplitude of the elementary components is equal to 2 × (variance)/(mean) (Katz and Miledi, 1972). Therefore, on a plot of mean versus variance the slope should be constant and a measure of quantal amplitude. Although the analysis of mean and variance can be used to derive relative changes in mEPSC rates just after train stimulation (0.05-1.5 sec) (Otsu et al., 2004), we observed enhanced mEPSC rates well after train stimulation that continued for up to 15 sec (Fig. 1B)(y-axis greatly expanded to emphasize the mEPSCs). At periods >1.5 sec after the train, mEPSC rates were generally too low to be accurately measured from the mean amplitude of the current (and to a lesser extent, variance after 6 sec) because baseline offsets could be significant compared with the amplitude contributed by mEPSCs. At these later times we used a semiautomated mEPSC analysis routine that identified individual mEPSCs by their amplitude, rise, and decay times (see Materials and Methods). This analysis revealed that elevated mEPSC rates persisted for even up to 15 sec after a train (decay time constant 9.1 ± 4.3 sec for slow component; 11% amplitude) (Fig. 1C). Because there was overlap in the mEPSC rate time courses (derived from the different methods, variance, and semiautomated direct detection) (see Fig. 4), we were able to append the time courses measured using different approaches to produce a full description of the mEPSC rate decay after train stimulation (see Fig. 4). Our approach to assessing changes in release frequency over time was to use current variance for the first 6 sec and semiautomated direct AMPAR mEPSC detection for later times when rates were relatively low and superposition of events was minimal. As described previously, we estimated the peak rate of mEPSCs in units of RRPs per second (Otsu et al., 2004). In this case the RRP was defined as the cumulative postsynaptic response attributed to a high-frequency stimulation without a constant resupply of releasable quanta. With a 0.5 sec train delivered at 20 Hz (the stimulus used for subsequent imaging experiments), we estimate that neurons release 2.1 ± 0.2 RRPs per second of transmitter at peak (n = 5 cells). By using the electrophysiological approaches described above, we can get an accurate description of relative changes in mEPSC rates after train stimulation within a population of synapses; however, these methods do not permit determination of the actual quantal release rates at individual synapses.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1.

Asynchronous release continues well after phasic release depression. A, Averaged whole-cell current (n = 3) induced by train stimulation (20 Hz, 10 pulses). B, Delayed mEPSCs after train stimulation in four consecutive trials in the same cell in A are shown (time scale compressed compared with A). C, mEPSC event histogram plotted for 70 total trials in six cells. Bin size is 0.8 sec. Time 0 indicates the end of train stimulation. We did not count mEPSC events within 1.33 sec after train stimulation because individual events could not be unambiguously detected at this time. These experiments were done under conditions (100 μm d,l-APV and 3 mm CaCl₂ and 1 mm MgCl₂) to maximize AMPAR mEPSCs, which are easier to detect than NMDAR mEPSCs because of their greater amplitude and faster kinetics.

Imaging of asynchronous postsynaptic activity after train stimulation

By using individual hippocampal neurons with autaptic synapses in microisland culture, we monitored the postsynaptic rate of asynchronous activity at individual dendritic sites after train stimulation by using fluorescent Ca²⁺ indicators to image NMDAR mEPSCs as described previously (Umemiya et al., 1999, 2001; Murthy et al., 2000). An advantage of autaptic synapses is that all release events are mediated by a single axon, and therefore synaptic depression is not mediated by selective loss of impulse conduction between axons. A complication of using the autaptic microisland cultures, however, is that the action potentials (APs) that trigger synaptic release propagate to postsynaptic compartments leading to Ca²⁺ signals attributed to both the AP and the postsynaptic effect of the transmitter. Because the signal resulting from the AP should be relatively constant throughout trials, we subtracted it to resolve signals attributed to individual quanta that were released asynchronously. To differentiate between Ca²⁺ signals mediated by phasic and asynchronous release, we applied the rapidly dissociating NMDAR antagonist d-AA (1 mm) before and during the 10 stimuli delivered at 20 Hz to produce phasic release (in the presence of 1 mm Mg²⁺); then the NMDAR antagonist was rapidly washed out (over ∼400 msec), and the solution was replaced by one lacking Mg²⁺ to efficiently view the NMDAR-mediated postsynaptic Ca²⁺ component of asynchronously released quanta (Fig. 2A) [compare videos 1 and 2 (supplemental material, available at www.jneurosci.org), taken under conditions that block and unblock, respectively, NMDARs after train stimulation (site shown in Fig. 3A).] Under these conditions we expected that Ca²⁺ signals were caused by both the residual effects of the APs and asynchronous transmitter release. Ca²⁺ transients from control trials in which NMDAR antagonists were present during and for 17 sec after train stimulation were subtracted from all records to isolate an NMDAR-mediated optical signal attributed to asynchronous release (Fig. 2Aa,C) For an example of subtracted images demonstrating only NMDAR-dependent asynchronous release, see video 3 (supplemental material, available at www.jneurosci.org). In video 3 (supplemental material, available at www.jneurosci.org), very robust asynchronous release-mediated Ca²⁺ changes are observed within the first 3 sec after train stimulation; after ∼5 sec, individual MSCT (bright flashes) events can be readily observed at particular sites. In the presence of NMDAR antagonist we expect Ca²⁺ signals to arise from the slow decay of AP-associated Ca²⁺ entry (video 1, available at www.jneurosci.org as supplemental material). For example, in Figure 2Ab, a record is shown that depicts a Ca²⁺ transient produced in the presence of Mg²⁺ and d-AA both during train stimulation and over a 17 sec period that follows. A robust increase in intracellular Ca²⁺ followed by an exponential decay was observed and attributed to AP-mediated opening of voltage-gated Ca²⁺ channels alone (Fig. 2Ab, thin line). If the stimulus was repeated and the NMDAR antagonists were washed out just after the train, we observed that the Ca²⁺ transient now decayed in a more complex manner (Fig. 2Ab, thick line). Examination of records from individual sites revealed MSCT-like peaks that were attributed to individual quantal transmitter release events. Subtraction of the two traces (NMDAR unblocked minus NMDAR blocked) revealed the NMDAR antagonist-sensitive, late asynchronous release component (Figs. 2C, 3A). In the same cells we also recorded whole-cell currents under conditions expected to isolate NMDAR activity and observed a slow average current apparently composed of many NMDAR mEPSCs that decay with a time course similar to the elevated rate of MSCTs observed (Fig. 2B). In five cells we plotted the appearance of these MSCT events 5-17 sec after train stimulation and observed a fall in rate with a time constant similar to that for the slow component of the NMDAR current, suggesting that these events were induced by the train stimulus (Fig. 2Bb). By averaging fluorescence signals from 8 trials and 29 sites, we were able to establish that washout of d-AA and Mg²⁺ (after a train stimulation) effectively reveals NMDAR-mediated changes in [Ca²⁺]_i attributed to elevated MSCT frequency (Fig. 2C). The average Ca²⁺ transient decays with a time course similar to that for the NMDAR current activated by asynchronous activity; however, one must consider that individual MSCTs have longer kinetics than the NMDAR mEPSCs.

• Download figure
• Open in new tab
• Download powerpoint

Figure 2.

Ca²⁺ transients mediated by NMDA receptors induced by asynchronous release. Aa, Two whole-cell current traces are superimposed. One is recorded with Mg²⁺ and d-AA (1 mm) present (NMDARs were blocked) (gray line). Just after train stimulation, the second trace was made by washing out the Mg²⁺- and d-AA-containing extracellular solution after the last pulse of train (10 stimuli at 20 Hz) (black line). CNQX (1-3 μm) was included in the solution to reduce AMPAR activity. Ab, Time course of two superimposed Ca²⁺ records (ratio of Fluo-4 to mag-fura-2 fluorescence, F₄₉₀/F₃₈₀) that were obtained under the two different conditions [NMDAR blocked (thin line) and unblocked (thick line) during asynchronous release period] for the cell described in Aa. MSCTs were clearly observed after train stimulation under conditions that unblock NMDARs. Ba, Time course of the average NMDAR-induced current associated with asynchronous release, obtained under Mg²⁺- and d-AA-free solution after train stimulus (n = 5 cells). Bb, Frequency histogram for MSCTs recorded under Mg²⁺- and d-AA-free solutions after train stimulation (data shown is from 80 synaptic sites in 5 cells and includes 375 events). The zero point is at the onset of the train stimulus. Data from 0 to 5 sec was not analyzed because of event overlap and Ca²⁺ indicator saturation (see Results and Discussion). Ca, Average fluorescence change (in units of F₄₉₀/F₃₈₀) associated with train stimulation under conditions in which NMDARs are blocked after train stimulation or unblocked by washout of Mg²⁺ and d-AA (average of 29 sites and 8 trials). Cb, Plot of the difference between the records in Ca indicating the NMDAR-dependent Ca²⁺ signal associated with asynchronous release.

Evidence that the late Ca²⁺ transients were MSCTs included their complete suppression in solutions with NMDAR blockers (Fig. 2Ab) (compare videos 1 and 2, available at www.jneurosci.org as supplemental material) and a close correspondence between the amplitude and time course of spontaneous MSCTs and those evoked asynchronously by train stimulation in the same cell (Fig. 3A). The asynchronous release-dependent Ca²⁺ transients have considerably slower decay kinetics than NMDAR mEPSCs reported previously (Umemiya et al., 2001). This is expected because the Ca²⁺ transient is slowed (with respect to the current that triggers it) by relatively slow intracellular extrusion mechanisms and both endogenous and exogenous (Ca²⁺ indicator) buffering (Sabatini et al., 2002). To measure asynchronous Ca²⁺ transients, we found it necessary to use a relatively high concentration of a high-affinity Ca²⁺ indicator (0.4 mm Fluo-4). The use of this indicator led to some saturation associated with the AP train and initial, very robust asynchronous release (Fig. 2Ab, thick unsubtracted record). Because of saturation we did not use the MSCT data for periods <5 sec after train stimulation for estimation of peak MSCT rates. In addition, MSCTs decay relatively slowly (∼2 sec decay constant), and it was possible to have overlapping events that would lead to an underestimate of MSCT frequency. Because we use a high-affinity dye, MSCT amplitudes may be affected by saturation, and thus we used the Ca²⁺ indicators to measure relative changes in event frequency and not amplitude. Lower-affinity indicators would solve many of these problems at the expense of considerably smaller fluorescence changes, raising the possibility of missed events. Although unlikely (Mainen et al., 1999), it is also possible that NMDAR-mediated release of Ca²⁺ from stores may increase the amplitude of the events that we observe, providing another reason for monitoring only event frequency and not amplitude (Emptage et al., 2003).

Although we are being conservative in only monitoring MSCT event frequency and not amplitude, Ca²⁺ or fluorescence signal spill-over between adjacent synapses may lead to inaccuracy in asynchronous Ca²⁺ transient frequency determination. To reduce the likelihood of signal spillover, we have avoided analyzing sections of dendrites in which multiple processes overlap. Furthermore, synaptic density in cultured cells is considerably lower than that found in vivo, reducing the likelihood of signal spill-over. In previous studies using dissociated cultures, we reconstructed sites of MSCTs and found that most Ca²⁺ transient initiation sites were associated with a single presynaptic terminal (Mackenzie et al., 2000); however, it is conceivable that some differences in Ca²⁺ transient frequency may be caused by differences in synapse or release site number. Line-plot analysis of Ca²⁺-induced changes in fluorescence during repeated MSCTs indicated little spatial variation (within the limits of light microscopy and our imaging conditions) with regard to Ca²⁺ transient initiation points during multiple events at presumably the same synapse (Fig. 3B). Arguing against significant overlap in MSCT initiation sites (leading to higher apparent event frequency) is the relatively small positive skew in the distribution of MSCT frequencies (Fig. 3C). Positive skew could arise through the relatively infrequent overlap of individual synaptic MSCT initiation sites, creating relatively low-incidence sites of relatively higher frequency. If sites with higher MSCT rates were caused by overlapping synapses, the distribution suggests that they are a minority (∼10% of the sites), with no more than two- to threefold the rate of the MSCTs observed in the general population presumably attributed to one synapse.

Estimates of peak release rates from asynchronous miniature synaptic calcium transients

To evaluate how the frequency of MSCTs at individual synapses corresponds to the decay in asynchronous release rates, we examined the temporal distribution of MSCT events after 10 synaptic stimuli given at 20 Hz by plotting the times at which MSCT events were initiated on the histogram shown (Fig. 2Bb) (0.8 sec bins). As mentioned above, Ca²⁺ concentration in dendrites was very high initially just after train stimulation for several reasons, including the activation of voltage-gated Ca²⁺ channels in dendrites (associated with backpropagating APs) and a robust amount of asynchronous release. After ∼5 sec, however, individual MSCTs could be reliably detected. During this time period (>5 sec to end of record), the relative decay kinetics for MSCT frequency matched that of mEPSC frequency (time course derived from analysis of AMPAR mEPSCs) (Fig. 4A,B). Our assumption was that the relative frequency of MSCT appearance (event initiation times) during this period should reflect the underlying mEPSC frequency. Therefore, by knowing the decay of MSCT frequency over a time period well after train stimulation and assuming a similar decay time course for appearance of AMPAR mEPSCs, we estimated what the frequency of MSCTs would have been at time = 0 (by extrapolation to the AMPAR mEPSC rates). Because MSCT frequency can be expressed in units of events per synapse per second, we could get an estimate of the peak rates of quantal release for times when event frequency was too high for direct measurement by Ca²⁺ imaging. For the synapses shown in the graph in Figure 3C, we observed an average rate of MSCTs of 0.045 ± 0.004 events per synapse per second (80 sites) over a period ranging from 5 to 17 sec after train stimulation. To determine how well the decay of MSCT rates agreed with whole-cell-derived mEPSC rates, we plotted the average MSCT rate per synapse alongside relative changes in AMPAR mEPSC frequency measured by variance or direct event detection (Fig. 4A,B). To align the MSCT frequency plots with these other measures of activity, we multiplied the AMPAR mEPSC current variance, the NMDAR average current (see below), and the decay in AMPAR mEPSC frequency by a conversion factor and added a small offset (ranging from 14 to 19% of the average MSCT rate) (Fig. 4, legend). Given that these measures can involve different recording conditions with different sources of noise and that some error is apparent in MSCT frequency determination, these small offsets are acceptable.

Over a 5-16 sec period after train stimulation, the frequency of AMPAR mEPSC appearance decayed with a time course similar to that observed for MSCT frequency. Assuming that the MSCTs appeared with a time course determined by fitting of asynchronous release rates (measured by the AMPAR mEPSCs), we calculated a peak rate of 21 MSCTs per second per synapse (Y-intercept 21 events per second per synapse, adjusted for a 0.8 sec bin). Based on electrophysiological data obtained using these conditions, the peak MSCT rate would correspond to ∼2.1 ± 0.2 RRPs per second. Therefore, we estimated that the RRP should contain ∼10 quanta (21 MSCTs per second per 2.1 RRPs per second) assuming that an MSCT is mediated by a single quantum. To check the results with another measurement of mEPSC event frequency, we also plotted the decay of average NMDAR current (for a 0.8-17 sec period after the train) measured during the actual imaging experiments and found that it decayed with a slow time course that approximated that of the MSCT rates: 4010 ± 470 msec (n = 5 cells) (Fig. 4). Given that individual NMDAR EPSCs decayed with a time constant of <100 msec in these cells, at time points much longer than the EPSC decay time course (three or more time constants), the time course of the NMDAR current should approximate the decay of mEPSC rate. We did not plot the NMDAR current at earlier time points (< 0.8 sec after train stimulation) because it is relatively inaccurate for several reasons: (1) NMDAR mEPSCs can be long lasting (up to several hundred milliseconds, in particular for extrasynaptic currents after trains), and their time course would not necessarily provide a measure of release rates but would reflect receptor biophysics (Dingledine et al., 1999), and (2) at early time points the NMDAR current time course also reflects the washout of the antagonist d-AA (∼400 msec) and its unbinding from the receptor; see slow rise of current after solution change (Fig. 2B). Thus, although Ca²⁺ imaging provides us with rates of postsynaptic activity at single synapses, it does not provide us with postsynaptic measures of release rates at early time points after a stimulus train (<5 sec). By using electrophysiological measures we can estimate relative changes in mini-rates (MSCTs) for periods in which direct measurements using imaging were not technically possible. In this case agreement between multiple measures of postsynaptic activity indicative of mEPSC frequency, such as NMDAR current and AMPAR mEPSC frequency, indicate that relative changes in MSCT rates well after a train can be used to extrapolate to peak rates just after train stimulation.

The relationship between evoked release probability and asynchronous release

Several reports have shown that the evoked release probability (P_r) (mediated by APs) is heterogeneous between terminals, even those derived from a single axon (Hessler et al., 1993; Rosenmund et al., 1993; Murthy et al., 1997; Chavis and Westbrook, 2001). The presence of heterogeneity in P_r raises the possibility that synapses undergoing asynchronous release may not be representative of the general population. Possibly, synapses exhibiting a high asynchronous release rate do so because they have a relatively low P_r and a large store of quanta and thus are resistant to depression produced by depletion of transmitter stores (Sakaba and Neher, 2001). To investigate whether P_r and asynchronous release rate are correlated at autaptic synapses, our strategy has been to compare the use-dependent block of NMDAR EPSCs by MK-801 as a measure of P_r under control conditions and after previously blocking ∼50% of the NMDARs that are activated by asynchronous release with MK-801. Presumably if synapses that undergo asynchronous release had a significantly different P_r (Chavis and Westbrook, 2001) or NMDAR composition (Nimchinsky et al., 2004), then we would expect the rate of EPSC block to be altered if 50% of the NMDARs were previously blocked using an MK-801 treatment that selectively targets asynchronous transmission (Fig. 5).

To isolate NMDARs undergoing asynchronous release (allowing their selective blockade by MK-801), we used train stimulation (100 stimuli at 20 Hz) that was found previously to completely block evoked (phasic) release, leaving only asynchronous release after ∼20 or 30 stimuli (Otsu et al., 2004) (Fig. 5Ba). During the phase of synaptic transmission in which only asynchronous release was present (Fig. 5A,Bb), we switched to a Mg²⁺-free solution containing blockers of AMPARs as well as 10 μm glycine (conditions that isolate and augment NMDAR responses). Under these conditions we observed postsynaptic NMDAR currents attributed to asynchronous release. During this period of NMDAR-dependent asynchronous release, we applied MK-801 during multiple trials (one to seven) of 20 Hz train stimulation (Fig. 5A) (train/MK-801 experiments). This antagonist treatment resulted in ∼50% reduction in the amplitude of evoked NMDAR EPSCs (determined after recovery from depression) (Fig. 5Bb,C). To control for possible train-dependent depression mediated by other means, we also examined trials of train stimulation without MK-801. In these experiments (train experiments), we actually observed a modest potentiation of NMDAR EPSCs (124.8 ± 8.0%) (Fig. 5C) and a modest depression of AMPAR EPSCs (Fig. 5C). In trials performed in the presence of MK-801, no additional reduction in the AMPAR EPSC was observed, whereas the NMDAR EPSC was markedly reduced (Fig. 5C) (compared with train experiments).

We then assessed the impact of blocking a large proportion of synapses that undergo asynchronous release (∼50% block with MK-801) on a subsequent determination of P_r, again using use-dependent MK-801 block (0.1 Hz low-frequency stimulation) under conditions designed to block phasic release-generated NMDAR EPSCs (Fig. 5Ba,D) as in previous studies (Chavis and Westbrook, 2001). Previous work indicates that the rapid depression of NMDAR EPSCs by MK-801 corresponds to high P_r synapses that the antagonist can preferentially access and block (Hessler et al., 1993; Rosenmund et al., 1993; Chavis and Westbrook, 2001). Although synapses with a lower P_r have a slower rate of MK-801 block because they are less likely to open and be accessed by the use-dependent antagonist, it is also possible that differences in NMDAR composition and receptor saturation between synapses may result in altered NMDAR blocking rate by MK-801 (Nimchinsky et al., 2004). Consistent with previous investigators using MK-801 treatment, we observed a bi-exponential depression of evoked synaptic transmission during low-frequency stimulation (0.1 Hz) with MK-801 (Fig. 5D). We next assessed whether blockade of 50% of NMDARs in synapses undergoing asynchronous release would affect the MK-801 blocking rate for phasic release. Examination of individual experiments indicated no significant change in the rate of the MK-801 block (analysis of weighted time constants p > 0.05, by unpaired t test; train 3.1 ± 0.2 and train/MK-801 3.4 ± 0.2) for cells in which ∼50% synaptic NMDARs undergoing asynchronous release were blocked previously by MK-801 or their controls (n = 7-12). To better assess whether the proportion of synapses with fast and slow MK-801-mediated blockade (0.1 Hz stimulation, indicative of high and low P_r) changed, we used normalized average data (across different cells) to determine whether the proportion of the decay described by a slow component changed. Previous data (Chavis and Westbrook, 2001) and our own (Fig. 6) indicated that if high P_r synapses are blocked, the proportions of fast and slow MK-801 block should change (slow component should increase), and not necessarily the time constants. The average time constants for cells without MK-801 application during conditioning 100 Hz train stimulation was 0.72 ± 0.09 stimuli for the fast component and 6.17 ± 0.60 stimuli for the slow component (n = 9). Therefore, to increase statistical power we constrained the fitting by assuming an average fast and slow time constant derived from control trains and determined whether the fraction of block mediated by the slow component changed by blocking a large fraction of NMDARs during asynchronous release. Using this analysis, no significant change in the percentage of slow component MK-801 block was observed after blocking ∼50% of the NMDARs with MK-801 during 100 Hz train-induced asynchronous release (Fig. 5E) (p > 0.05). If the synapses that were blocked during the asynchronous release were of relatively higher P_r, we would expect a higher proportion of low probability synapses when use-dependent MK-801 block of phasic release was performed (low-frequency stimulation, 0.1 Hz). Our results, however, showed that there was no correlation between the asynchronous release rate and P_r because blockade of 50% of synaptic NMDARs by MK-801 application during the asynchronous phase of release had no effect on subsequent measurement of P_r using low-frequency stimulation (Fig. 5D). These findings are consistent with synapses that undergo high rates of asynchronous release being drawn from a general population and not reflecting a subset of synapses that have high or low P_r (P_r defined by MK-801 block during phasic release, 0.1 Hz stimulation) or markedly different NMDAR composition.

• Download figure
• Open in new tab
• Download powerpoint

Figure 6.

Ifenprodil blocks high P_r synapses and reveals two components of use-dependent MK-801 block. Aa, Representative traces of NMDAR EPSCs under control conditions (0 mm Mg²⁺, 10 μm glycine, and 5 μm CNQX) and addition of 3 μm ifenprodil. The EPSCs were mediated entirely by NMDA receptors because progressive blockade by MK-801 (at 70th stimulus after ifenprodil application) could completely block them. Stimulus artifacts are removed. Ab, Normalized EPSCs recorded under control conditions or ifenprodil reveal that ifenprodil selectively blocks the slower component of EPSC. B, Ratio of NMDAR EPSC amplitude after to before ifenprodil treatment (Ifen) is plotted in second column at the left. Ratio of AMPAR EPSC amplitude after to before either ifenprodil treatment (Ifen) or nontreatment (Control) and progressive blockade of NMDAR EPSCs by MK-801 are plotted in two columns at the right. The numbers in the bars show cell number. C, Use-dependent blockade of NMDAR EPSCs by MK-801 (0.1 Hz stimulation). Antagonism of NR2B-containing NMDARs by ifenprodil leads to a change in the blocking rate for the EPSC by MK-801 (open circle; n = 9) compared with nontreatment condition (filled circle; n = 9). NMDAR EPSC amplitude was normalized to responses just before application of MK-801. Data shown are the mean ± SEM. D, The result indicates a significant increase in the slow component of block with ifenprodil treatment (p < 0.01). The fitting procedure for EPSC use-dependent block by MK-801 is described in Figure 5D.

The results raise the question of whether a small difference in P_r could be measured with our technique. Previous work by Chavis and Westbrook (2001) indicate that blockers of NMDARs containing NR1/NR2B subunits such as ifenprodil can be used to selectively reduce the postsynaptic contribution of high P_r synapses. This effect is because two populations of synapses are thought to exist and not because of a potential presynaptic effect of ifenprodil. We confirmed that high P_r synapses preferentially have NR1/NR2B NMDARs as shown by Chavis and Westbrook (2001). EPSCs resulting from NR2B-containing NMDARs have been shown to have a slower decay time constant than those arising from NR2A NMDARs (Hessler et al., 1993) (Fig. 6A). In the presence of the NR1/NR2B blocker ifenprodil, we observed that NMDAR EPSCs decayed more rapidly (64.8 ± 0.6 msec; ifenprodil, n = 9; vs 97.5 ± 4.5 msec control; p < 0.01 paired t test) (Fig. 6A). Ifenprodil treatment was relatively selective to postsynaptic sites (NMDARs) because it has no effect on AMPAR EPSCs (Fig. 6B). In the presence of ifenprodil using the same analysis as described for Figure 5, we found a significant increase in the slow component of MK-801 blocking rate during low-frequency phasic release (Fig. 6C,D). With ifenprodil present we propose that high P_r-NR1/NR2B bearing synapses are antagonized (Fig. 6A,B) and prevented from participating in use-dependent block by MK-801, leading to an apparently lower P_r that is reflected in the reduced MK-801 blocking rate. In total these experiments, which show expected changes in P_r by blocking high P_r synapses with ifenprodil but no change in P_r associated with synapses undergoing asynchronous release, support the proposal that NMDARs undergoing asynchronous release do not represent a unique group.

Postsynaptic determination of rate release at individual synapses during train stimulation

Our results provide estimates for peak rates of exocytosis at CNS synapses by imaging the postsynaptic effect of asynchronous release. We estimate that hippocampal synapses are able to release ∼21 quanta per second at steady state during trains. These results suggest that release must be maintained at these rates by rapid mechanisms and not necessarily continuous endocytosis and clathrin-mediated recycling of vesicles that elegant studies from the neuromuscular junction estimate to require tens of seconds to occur (Cochilla et al., 1999; Richards et al., 2000; Harata et al., 2001b). Recent studies imaging the stimulus evoked turnover of clathrin in hippocampal neurons suggest that it exhibits delayed kinetics (Mueller et al., 2004) that cannot account for the high peak rates of release that we observe. Given that the observed high rates of asynchronous release can be maintained for only a few seconds, it is likely that they reflect not a recycling mechanism but refilling from a reserve pool of finite size. Perhaps Ca²⁺-dependent mechanisms engaged during train stimulation contribute to these high rates of refilling (Stevens and Wesseling, 1998; Wang and Kaczmarek, 1998). Rates of quantal release and vesicle turnover are temperature dependent and likely to be higher at physiological temperature than room temperature (∼21°C) as we have done (Pyott and Rosenmund, 2002; Fernandez-Alfonso and Ryan, 2004); however, recent work indicates that the 20 Hz trains that we use produce depression at 37°C (Fernandez-Alfonso and Ryan, 2004).

Potential sources of error

One limitation of this approach is that we cannot actually resolve individual MSCTs during the peak rate of asynchronous release. There are several concerns with measuring MSCTs during early periods after train stimulation. (1) Ca²⁺ indicator saturation induces nonlinearity to Ca²⁺ signaling, and (2) many double and triple events can occur over the imaging sampling time (∼66 msec). Because of this limitation and missed events, the decay of MSCT rates may not strictly follow a single exponential. To get at the issue of missed events, we took an average fluorescent signal from a large number of individual dendritic sites with and without the removal of NMDA receptor blockers after train stimulation. Under these conditions, we observed a large and long-lasting NMDAR-dependent Ca²⁺ transient, indicating that many early MSCT events (at times <5 sec) were likely missed. Another problem with imaging MSCTs just after train stimulation is that antagonist must be applied to the entire neuron, limiting the solution exchange times to no faster than several hundred milliseconds. For this reason the NMDAR current triggered by asynchronous release that we measure does not reflect the peak rate of quantal release as assayed postsynaptically. To better assess the peak rates of asynchronous activity, we have relied on monitoring AMPAR activity that does not have the limitation of slow receptor kinetics or the requirement to switch solutions rapidly (to wash out NMDAR antagonists). By combining several measures of AMPAR activity with NMDAR currents at later time points, we are confident that significant rates of asynchronous release continue for up to 10 sec after stimulation, and these rates correlate with peak activity just after train stimulation. Given that we have used 400 μm Fluo-4 as our Ca²⁺ indicator, it is conceivable that the asynchronous release that we observe may be blunted in peak rate and prolonged in time course because Fluo-4 will buffer Ca²⁺ with characteristics determined by its Ca²⁺ binding kinetics and concentration (Helmchen and Tank, 2000); however, it is not really possible to monitor the presynaptic concentration of Fluo-4 in terminals. Clearly, the use of conventional synapses with separate presynaptic and postsynaptic compartments (instead of autapses) would be a way to address this concern in future studies. In studies with separate presynaptic and postsynaptic compartments, presynaptic buffering capacity could even be tuned to vary the time course of asynchronous release.

Using partial MK-801 to assess the impact of synapses undergoing asynchronous release on evoked P_r

The application of use-dependent MK-801 block indicated that synapses undergoing asynchronous release were indistinguishable from the general population with regard to their P_r. Evidence for this included no significant change in the MK-801 blocking rate during phasic release (triggered by 0.1 Hz stimulation) when 50% of synaptic NMDAR was previously removed by MK-801 treatment designed to target asynchronous release; however, the results also raise the question of whether the technique is sensitive enough to resolve small changes in P_r. Plotting the average time courses of phasic release block by MK-801 for cases with and without 50% of the NMDARs participating in asynchronous release blocked indicated that the datasets were virtually superimposable. To assess any statistical change we relied on fitting these plots to a model with two populations of synapses, either high or low P_r or perhaps varied NMDAR composition or saturation by agonist. It is conceivable that this is an oversimplification, but as mentioned above, simple inspection of the data reveals no major effect. Importantly, positive controls support the two-population statistical model in that ifenprodil treatment, which blocks high-probability synapses, produces a significant change in the time course of the MK-801 blocking rate that can be modeled by an increase in a slow component of block.

Postsynaptic measures of quantal release after train stimulation

Postsynaptic Ca²⁺ imaging complements monitoring presynaptic vesicle release rate and provides additional information. First, we are monitoring actual synaptic responses and not membrane recycling, which may not always reflect transmitter release. Second, the results draw attention to the fact that biochemical events triggered by Ca²⁺ influx continue to occur many seconds after train stimulation. Perhaps these continued biochemical signals mediated by mEPSCs elicit specific forms of signal transduction (Otsu and Murphy, 2003a,b). Imaging asynchronous release may also represent a technique to gain insight into the workings of the synapse during high-frequency stimulation when most electro-physiological approaches would be unable to resolve individual quanta. It could be argued that some of our experiments could be repeated with FM dyes; however, given that asynchronous release closely follows phasic release, it would be difficult to apply FM dyes in such a manner that they would selectively label terminals undergoing asynchronous release and not terminals showing delayed dye uptake triggered by phasic release. Previous investigators have examined the response of spines to high-frequency stimulation and observed persistent changes in spine [Ca²⁺]_i that may have been related in part to asynchronous release (Muller and Connor, 1991); however, the sampling rates used were not high enough to resolve individual quanta. By using extrapolation and monitoring the rates of asynchronous release, it is possible to image a reflection of high-frequency signaling that occurs during train stimulation. This high-frequency signaling is particularly relevant considering that plastic behaviors are often associated with high-frequency gamma and theta rhythmic activity in vivo (Buzsaki et al., 2002). If high-frequency presynaptic activity at a single terminal is prolonged for several seconds, phasic release could be depressed and marked asynchronous release would be apparent. Recent experiments using cultured hippocampal neurons suggest that synaptic transmission can be maintained presynaptically with up to 10 Hz stimulation at physiological temperature (Fernandez-Alfonso and Ryan, 2004). These authors propose that this occurs through rapid mechanisms of endocytosis and vesicle recycling, although they did not directly resolve the postsynaptic action of individual release events.

Comparison of asynchronous release measures of transmitter release with other studies

Previously, investigators using FM dye imaging, electrophysiological techniques combined with local application of agents to stimulate synaptic release, have estimated the RRP size and rate of quantal release at individual pyramidal neuron synapses (Liu and Tsien, 1995a,b; Stevens and Tsujimoto, 1995; Murthy and Stevens, 1999; Schikorski and Stevens, 2001). Our present results using a different approach that does not necessarily rely on stimulating only individual synapses with an extracellularly applied agent or monitoring potentially indirect vesicle recycling as a measure of release now suggest similar peak rates of quantal release. In total, our results and others indicate that CNS synapses are able to undergo rapid bursts of transmission averaging ∼20 quanta per second. The ability of the synapse to sustain this level of release will likely be dependent on the size of its RRP and the rate at which it can refill from a reserve pool, reuse quanta via kiss and run, or recycle vesicles via clathrin-mediated endocytosis. Elegant recent electron microscopy data from the frog neuromuscular junction (Rizzoli and Betz, 2004) indicate that recycling vesicles belonging to the RRP may actually be distributed throughout the synapse, suggesting that refilling processes may not be constrained by the number of vesicles resident at plasma membrane release sites. Given the potential for transitions in release mechanisms (Aravanis et al., 2003; Gandhi and Stevens, 2003), future studies will require investigations during which both presynaptic vesicle trafficking and the postsynaptic quantal response can be monitored simultaneously at an identified synapse.