Single-vesicle imaging reveals that synaptic vesicle exocytosis and endocytosis are coupled by a single stochastic mode

Abstract

The nature of synaptic vesicle recycling at nerve terminals has been a subject of considerable debate for >35 years. Here, we report the use of an optical strategy that allows the exocytosis and retrieval of synaptic components to be tracked in real time at single-molecule sensitivity in living nerve terminals. This approach has allowed us to examine the recycling of synaptic vesicles in response to single action potentials. Our results show that, after exocytosis, individual synaptic vesicles are retrieved by a stochastic process with an exponential distribution of delay times, with a mean time of ≈14 s. We propose that evidence for fast endocytosis, such as that proposed to support the presence of kiss-and-run, is likely explained by the stochastic nature of a slower process.

Once synaptic vesicles undergo exocytosis and release neurotransmitter, they are recycled for further rounds of reuse. Although significant evidence indicates synaptic vesicle endocytosis is dynamin- (1–3) and clathrin- (4) mediated, a number of studies have also implicated a mechanism of direct vesicle retrieval that would bypass the more classical clathrin-mediated route. In general, evidence for a kiss-and-run pathway is based on experiments that postulate a faster mechanism of retrieval than is usually reported for clathrin-mediated endocytosis. The first such evidence arose in the original experiments in bipolar nerve terminals, where electrical capacitance recordings revealed an endocytic time constant of ≈2 s (5). Experiments in small CNS nerve terminals include use of FM dyes reporting that, under certain conditions, dye does not completely destain during an exocytosis and retrieval event (6–9). These data have been interpreted as evidence for retrieval events happening in ≈1 s, much faster than has been reported for the clathrin-mediated route. More recent data have suggested that such events predominate in hippocampal neurons for a stimulus frequency at 0.1 Hz (10). In addition, electrical capacitance measurements at a giant calyceal synapse have provided evidence that synaptic vesicle retrieval can happen on a second or even subsecond timescale (11), however only for a relatively small fraction of events. Although these studies point to the possibility that some retrieval events might occur on a ≈1 sec timescale, it is unclear whether such events represent a separate pathway or simply one end of a distribution of times of a common process. We propose that the mere presence of faster events does not necessarily imply that a separate mode of retrieval is operational.

To determine whether multiple modes of retrieval are present, we sought to examine the distribution of retrieval times for all events after exocytosis. Here, we report an improved approach that combines specific targeting of pHluorin, a GFP with modified pH sensitivity (12, 13), with high collection-efficiency optical detection that allows one to follow exocytosis and endocytosis of individual synaptic vesicles. Targeting of pHluorin to synaptic vesicles by fusion to an intraluminal loop of the vesicular glutamate transporter vGlut1 (14) resulted in much lower surface expression. The resulting decrease in background fluctuations provided a much higher signal-to-noise measurement than has previously been possible. This approach allows for the direct detection of fusion and retrieval of single synaptic vesicles at individual nerve terminals and indicates that a stochastic mode couples exocytosis with synaptic vesicle retrieval.

Results

Improvements in Optical Recordings.

Detection of single-vesicle exocytosis using pHluorins has been limited by a combination of fluctuations in baseline fluorescence arising in part from a fraction of synaptic vesicle proteins residing on the synaptic surface (13, 15), photobleaching, as well as the Johnson noise and quantum efficiency of the photodetector. Of the approximately eight possible choices of synaptic vesicle proteins (16, 17) for targeting pHluorin to the synaptic-vesicle lumen, those tested to date have significant surface expression of the pHluorin-tagged molecule at the synapse [vesicle-associated membrane protein (VAMP) ≈15% (13), synaptotagmin I ≈22% (15), synaptophysin ≈9% (4)]. The vesicular glutamate transporter vGlut1 was recently tagged successfully with pHluorin, and expression of this construct does not appear to alter presynaptic function (14). To estimate the surface fraction of vGlut1-pHluorin (vGpH) after transfection into neurons, we used acute applications of NH₄Cl to rapidly alkalinize the acidic internal pools of the molecule without altering surface fluorescence (13). These experiments indicate that transfection of this construct results in only ≈2% surface expression at synapses (Fig. 1 A–C), indicating that vGlut1 resides largely in synaptic vesicles under resting conditions, and it is a good candidate for optimizing the detection of small signals using pHluorins. To optimize vGpH fluorescence detection, we used detectors suited for single-molecule imaging, with >95% quantum efficiencies in the visible spectrum and Johnson noise suppressed to well below one electron per readout as the result of on-chip electron multiplication (18). Our optical scheme consisted of coupling laser-excited fluorescence to this detector with minimal optical elements in the path. We determined the sensitivity of our optical system using a dilute solution of immobilized EGFP molecules on a coverslip where they appear as diffraction limited spots (Fig. 1 D and F). Analysis of the intensity distribution of these spots revealed a quantized distribution with a unitary size of ≈58 ± 3.3 units (Fig. 1G), which we attribute to the fluorescence of a single EGFP molecule. We examined the intensity changes after photobleaching by intense illumination (Fig. 1E) to verify that these spots corresponded to single molecules. As expected for single fluorophores, photobleaching occurred in an all-or-none fashion, where the histogram of fluorescence changes (Fig. 1H) was well described by the sum of distributions with peaks at zero intensity (no bleaching), ≈64 (single-molecule bleaching), and at ≈134 (two-molecule bleaching) fluorescence units. We attribute the center of the second peak as corresponding to the mean intensity of single EGFP molecules, which is in good agreement with the quantal size determined before bleaching (Fig. 1G Inset). In addition to demonstrating the overall sensitivity of the optical system, these measurements also provide a means of calibrating the experiments with appropriate gain and illumination corrections in terms of number of EGFP molecules.

Fig. 1.

Improvements in pHluorin detection. (A and B) Images of the vGlut1-pHluorin-transfected hippocampal boutons before and after addition of NH₄Cl. (C) The ensemble average of the fluorescence intensity at boutons indicates a ≈17-fold increase in fluorescence corresponding to a surface fraction of 2 ± 0.8% assuming an initial vesicular pH = 5.5 (nine neurons, ≈50 boutons each). (D) Fluorescence intensity of single EGFP molecules. Smoothened surface plot (3 × 3) of GFP in polyacrylamide gel before and after bleaching is shown in D and E, respectively. (F) Intensity profile of a fluorescence spot (circles). A Gaussian fit (line) yields a FWHM of 2.8, in excellent agreement with point spread function measured from 100-nm beads in our setup. (G) Intensity distribution of fluorescent puncta before bleaching is well fit (reduced χ² = 0.98) by a set of Gaussians whose spacings are exact integer multiples (Inset, linear fit, slope = 58 ± 3 fluorescence units per peak, R = 0.998). (H) Distribution of the change in fluorescence of the puncta after bleaching ΔF (F_prebleach − F_postbleach) (bars) shows peaks at 0, 64, and 134. Solid line is the overall fit, and dotted lines are the individual Gaussian fits.

Detection of Exocytosis and Endocytosis in Response to Single Action Potential (AP) Stimuli.

Under these experimental conditions, synapses expressing vGpH showed robust fluorescence responses to single AP stimulation (Fig. 2 A and B) with photobleaching rates of only ≈0.01% per frame (see Materials and Methods). The ensemble average of the fluorescence change driven by single stimuli (Fig. 2C) shows a sharp rise followed by a much slower decay. Signals derived from pHluorins are based on deprotonation of GFP during the fusion event and a pH transition from ≈5.5 to 7.4. After endocytosis, reacidification of the vesicle lumen quenches the fluorescence. The ensemble average shown here is well described by a simple kinetic model of two sequential first-order processes corresponding to endocytosis and reacidification (ref. 4; see Materials and Methods). Using rapid quenching approaches, vesicle reacidification has recently been estimated to occur with a time constant of τ_reacid ≈ 4 s (4, 16). The time constant for endocytosis derived from fitting the data in Fig. 2C indicates that τ_endo = 15 ± 1 s.

Fig. 2.

Optical detection of single-vesicle responses. Image of vGlut1-pHluorin-transfected hippocampal boutons before (A) and after (B) applying a one-AP stimulus. Arrowheads point to some of the responding boutons. (Scale bar: 10 μm.) (C) The gray line is the average of single action potential response over 120 events. The black line is the fit to a double exponential describing endocytosis followed by reacidification (see Materials and Methods). The fit yields a time constant for endocytosis of 15 ± 1 s. (D) In the presence of bafilomycin (red trace), repeated stimulation at 0.2 Hz results in a continuous staircase increase in fluorescence, whereas in the control run (black trace), endocytosis and reacidification balance the exocytic-driven increase. (E) The cumulative histogram of the response amplitudes to single AP in the absence (squares) and presence (circles) of bafilomycin (three neurons, 165 boutons) are identical (Kolmogorov–Smirnov test maximal D value =0.1, which corresponds to P = 1.000 that the distributions are the same), indicating all fusion events that release protons contribute to the signal.

During low-frequency repetitive stimulation, on-going endocytosis and reacidification prevent continuous increase in fluorescence (Fig. 2D, black trace). We verified that the fluorescence decay after each stimulus corresponds to endocytosis and reacidification, because these decays were completely blocked by the V-ATPase inhibitor bafilomycin (19) and instead showed staircase-like fluorescence increases in response to continuous stimulation (Fig. 2D, red trace). In addition, these observations allow us to determine whether any exocytosis events might be followed by endocytosis and reacidification that is faster than our time resolution. We reasoned that any vesicle that fuses and releases its protons would still require the V-type ATPase to reacidify the lumen, regardless of the speed of endocytosis. Thus, if such events occur, they would be trapped in the fluorescent state in the presence of bafilomycin. The amplitude of the average fluorescence response to a single action potential (ΔF_1AP) in the presence of bafilomycin should, in turn, be larger than that in the absence of bafilomycin. Comparison of the cumulative intensity histograms of ΔF_1AP values obtained with and without bafilomycin across many boutons showed that the distributions were indistinguishable (Fig. 2E), indicating those events must happen only very rarely. Thus, these measurements are not biased by missing exocytosis events that subsequently endocytose and reacidify too rapidly to be detected, and we conclude that all fusion events that lead to the release of protons contribute to the signal.

Analysis of AP-Driven Fluorescent Transients from Single Boutons.

We sought to characterize the distribution of signals associated with vGpH responses at individual boutons after single APs. Examples of fluorescence traces from individual boutons stimulated with single APs at 20-s intervals are shown in Fig. 3A. These data indicate that fluorescence increases at single boutons occur as sudden step-like changes, as would be expected for exocytosis (Fig. 3A). The distribution of fluorescence changes resulting from single APs (ΔF_1AP) derived from repeated trials (spaced >2 min apart) at many boutons (Fig. 3B) is well fit (reduced χ² = 0.97) by a set of Gaussians with different means and areas. The best fit, however, yielded a set of Gaussians with exact-integer multiple spacing (Fig. 3 B and C) that is insensitive to bin width (see Materials and Methods). The first peak of the distribution, centered about zero intensity, reflecting events that failed to show any exocytosis, is similar to the noise peak measured in the absence of stimulation (Fig. 4A). All other peaks represent multiples of a discrete quantal size. If this distribution truly reflects the variation from trial to trial of quantal responses, we presume that the area under each peak, and not the spacing, should vary as release probability is changed. We tested this hypothesis by comparing the event size distribution for trials performed in 1.5 mM external CaCl₂ ([Ca]_e) with that performed in 4 mM [Ca]_e. At lower [Ca]_e, the distribution shows a much larger peak around zero amplitude, whereas independent fits to the two data sets revealed identical quantal size (Fig. 4 A and B). Our data thus show that single AP responses at single boutons arise in discrete steps that are multiples of a basic size. It is possible that higher-order peaks reflect fusion events of vesicles with a greater number of vGpH; however, that these peaks appear only under conditions of elevated release probability argues against this possibility. The data obtained at lower release probability ([Ca]_e = 1.5 mM) allows one to set an upper bound on the number of events attributable to brighter vesicles by comparing the area under the first and second nonzero peaks. These data indicate that at least 80% of individual vesicles undergoing fusion have a fluorescence intensity that corresponds to the first nonzero peak in the distribution, and thus higher-order peaks largely correspond to the fusion of multiple vesicles at the same bouton. Examples of fusion events corresponding to the median of the first (one vesicle) and second peaks (two vesicles) compared with a noise trace are shown in Fig. 4C.

Fig. 3.

Distribution of instantaneous fluorescence changes. (A) Representative fluorescence time traces of individual boutons during repeated stimulation (indicated by arrows). An example of the amplitude of the first event (ΔF_1AP) is shown in the top upper left trace. (B) The distribution of instantaneous fluorescence change amplitudes from individual boutons associated with single-AP stimuli (ΔF_1AP) shows integer multiples of quantized fluorescence intensities. The data (540 events) were obtained from 30 boutons in 1.5 (four trials), 2 (four trials), 2.5 (three trials), 3.5 (three trials), and 4 mM (four trials) CaCl₂. The solid and dotted black lines are the overall and individual fits to multiple Gaussians, respectively (reduced χ_red² = 0.97). The coefficient of variation for the first nonzero peak is 0.37. (C) The peak positions obtained as a fit parameter in B are integer multiples indicating that ΔF_1AP are distributed as multiples of a fundamental quantal unit (r = 0.998).

Fig. 4.

Quantal size is invariant across different release probability. (A) Noise distributions (Top) obtained in the absence of stimulation. The solid line is a Gaussian fit (χ_red² = 0.6). Middle and Bottom show quantal histograms (bars) obtained by measuring the ΔF_1AP with 1.5 (224 events) and 4 mM (186 events) external calcium, respectively, taken from the same ensemble data as in Fig. 3. The distribution is fit [solid line, χ_red² = 0.5 (1.5 mM), χ_red²= 1.3 (4 mM)] to multiple Gaussians (dashed lines are the individual Gaussian components). The CV for the first nonzero peaks is 0.38 (1.5 mM) and 0.40 (4 mM). (B) Comparison of quantal size determined from 1.5- and 4-mM histogram, and histogram of all events (Fig. 3B) shows that the quantal size is invariant. (C) Illustrative example of time traces from single boutons that we classify as no response; one quantum and two quanta are shown in Bottom, Middle, and Top, respectively.

The quantal size expressed in fluorescence units can be directly compared with the intensity of single EGFP molecules (Fig. 1 G and H). Comparison of the fluorescence intensity of the quantal size using vGpH (Fig. 3C) with that of single immobilized EGFP molecules indicates that a single EGFP under these conditions corresponds to 1.23 ± 0.1 vGpH quanta. Thus, single vGpH quanta correspond to near single-molecule events. That a single vGpH quantum did not correspond to an exact integer number of EGFP fluorescence equivalents is possibly due to the difference in environment of immobilized EGFP compared with pHluorin spliced into vGlut1 and expressed on the cell surface. Thus, although the absolute calibration does not provide a perfect one-to-one correspondence between single EGFP signal amplitudes and vGpH quanta, the comparison indicates that the quantal size likely corresponds to very few, if not single, vGpH molecules. We have also obtained estimates of the total number of vesicles labeled by comparing the intensity of the maximal fluorescence obtained during sustained stimulation at high frequency in the presence of bafilomycin with that of our first nonzero peak in the quantal distribution. This type of stimulation results in the entire recycling pool becoming alkaline (19) and indicates that, on average, the labeled pool size was ≈64 ± 14 vesicles (n = 4 experiments, ≈300 total boutons). This is very similar to estimates obtained based on FM labeling (not shown), indicating that most recycling vesicles likely contain a single vGlut-pHluorin, implying the sorting process for insertion of the reporter into the vesicle is not random.

Endocytosis Is Well Described by a Single Poisson Process.

The fluorescence decay of individual endocytic events probed with pHluorin should correspond to two sequential steps: a dwell time on the cell surface (the time for endocytosis) followed by vesicle reacidification. We characterized the timescale of endocytic retrieval of single vesicles by restricting the analysis to fusion events whose intensity amplitude was within 1 SD of the center of the peak in Fig. 3B. This restriction is necessary, because multiple vesicles would potentially endocytose at different times, resulting in complex fluorescence decays. Examples of the time course of single-vesicle events are shown in Fig. 5A and reveal that the decay phase occurs asynchronously with respect to the stimulus. pHluorin molecules report a time-averaged ensemble of pH, because the dissociation time for protons is in the millisecond time range (20); thus, even for individual events, the time course of this step should be similar to that obtained from averages across many boutons where τ_r≈4 s (4, 16). We reasoned, therefore, that averaging the data over a timescale corresponding to τ_r/2 would improve the signal-to-noise ratio without significant loss of information. Inspection of events filtered in this manner (Fig. 5A) showed a broad range of retrieval times that varied among (i) prompt retrieval, i.e., events whose decay appears to be dictated solely by reacidification with a t_1/2 ≈2.7 s (the t_1/2 of a process whose time constant is 4 s) and therefore whose endocytosis occurred after exocytosis within the time resolution of the measurement, (ii) events that showed a measurable delay before the fluorescence decay, and (iii) events whose fluorescence failed to decay within the time-measurement window (t_1/2 > 20 s). We defined the endocytic dwell time as t_1/2−2.7. The frequency distribution of endocytic dwell-time values for ≈150 individual endocytic events indicates these values are distributed exponentially, well described by a single Poisson process with a constant transition probability per unit time k that results in a probability distribution P(t) = ke^−kt. The mean dwell time τ of the distribution P(t), is equal to 1/k. Here, τ = 13.4 ± 2.4 s is in close agreement with the single action-potential ensemble average value (Fig. 2C). Furthermore, the appearance of discrete dwell times before an exponential decay is also evident in ensemble average traces of many events drawn from individual times bins (Fig. 5C). These data indicate that, although the mean time for vesicle retrieval after exocytosis is ≈14 s, individual retrieval events clearly occur over a very wide range of timescales. The simple exponential distribution of dwell times, however, strongly implies they all represent a single stochastic process.

Fig. 5.

Single-vesicle endocytic dwell time follows an exponential distribution with a mean lifetime of ≈14 s. (A) Fluorescence traces obtained at individual boutons after single-vesicle exocytosis. Gray lines are raw time traces, and the black line is the running average over 13 points (≈2 s). The endocytic dwell time is defined as t_{dwell = t1/2} −τ_r ln(2), where t_1/2 is the time to decay to 50% of the peak amplitude, and τ_r is the reacidification time constant (4 s). (B) Frequency distribution of dwell times obtained from 150 single-vesicle events obtained from four neurons. Bars are the fraction of events decayed in that time bin, and error bar is the Poissonian noise estimated from the number of events in that bin. All events that did not decay within the observed time (such as the lower trace in A) are collected in the final bin (not shown) and included for calculating the fraction of events. The solid line is a fit to a Poisson distribution. The time constant for a Poisson distribution can be estimated from the exponential and gives 13.4 ± 2.4 s as well as reciprocal of the amplitude of the exponential divided by the bin width, which gives 15.0 ± 1.5 s. The fraction of events that remain undecayed within the observation time is in good agreement with that obtained by integrating the exponential fit (17.3%). (C) Average of five events from bins centered at 0.1, 2.85, and 8.35 s along with average of five events that did not decay during the observation time window is shown. The solid-gray exponential decays are fits to the reacidification time course. The average time constant from the fit is 3.8 ± 1.8 s.

Discussion

We report here a significant improvement in the ability to monitor exocytosis and endocytosis at the level of single vesicles at individual presynaptic boutons. The major technical advance was the combination of: (i) a pHluorin-tagged synaptic vesicle protein with very low surface fraction (≈2%) and (ii) high-collection efficiency optical readout, capable of single-molecule detection. These improvements allowed access to quantitative details of exocytosis and endocytosis that permitted us to estimate the time constant for endocytosis at nerve terminals in two ways (Figs. 2C and 5B). The first was a measure of the ensemble response to single AP stimulation across many boutons and an examination of the poststimulus decay. This decay was completely blocked (Fig. 2D) when vesicle reacidification was inhibited, indicating that the normal signal decay arises from endocytosis followed by vesicle reacidification. Our data show that the average endocytosis time constant for unitary responses is ≈15 s, in good agreement with previous estimates (4). The second approach examined the stochastic behavior of decay signals for individual vesicles and the endocytic dwell-time histogram for single-vesicle retrieval. This approach required successful identification of events corresponding to single vesicles. Analysis of the size distribution of exocytosis signals (Figs. 3 and 4) indicates that fluorescent signals corresponding to exocytosis appear quantized, and that each vesicle likely contains a single copy of vGpH. The fluorescence decay of individual vesicles reveals two distinct components, a stochastically determined dwell time of residence on the surface with a mean of ≈14 s, followed by an exponential decline we attribute to reacidification (Fig. 5).

The kinetics of endocytosis at nerve terminals has been a subject of considerable debate in the last 35 years. Central to the discussion has been the proposal that a fast mechanism of vesicle retrieval, kiss-and-run, might operate under certain conditions to enhance synaptic performance. The evidence for kiss-and-run at small nerve terminals has largely been based on either the observation of incomplete destaining of organic tracers during stimulation (8, 9, 21) or the differential access of fluorescence quenchers compared with loss of fluorescent tracers during repetitive cycles of exo endocytosis (10). Here, using methods that allow direct monitoring of fusion, we find that endocytic retrieval is dominated by a single mode of endocytosis whose average time constant is ≈14 s. Furthermore, the stochasticity indicates that endocytosis kinetics is likely dominated by a single rate-determining step. It is possible that previous evidence of fast endocytosis simply represented a selective sampling of this distribution due to technical constraints, ascribing different timescales of this stochastic process to different modes of endocytosis. It is important to note, for example, that for any Poisson process, although the mean retrieval time is 1/k, the most probable retrieval time is at short time intervals.

Recent experiments using siRNA directed against clathrin indicated that endocytic recovery in response to single AP requires clathrin (4) and, by extension, this suggests that the single mode we describe here depends on clathrin function. These measurements, however, required averaging over many trials and many boutons and included a significant correction for photobleaching. Previous measurements using pHluorin-based assays with VAMP-2 as the carrier protein after photobleaching the surface fraction reported the three modes of endocytosis (21); however, some of these data have been questioned as suffering from too low a signal-to-noise ratio to be conclusive (4), and we and others recently showed that surface VAMP-2 molecules exchange with vesicular VAMP during exoendocytosis (15, 22). This, together with photobleaching during data acquisition, complicates the analysis and interpretation of those results. Electrical capacitance measurements of fusion events from small synaptic vesicles were recently successfully recorded in a giant calyceal synapse, where flicker-type retrieval with times <1 s occurs in ≈20% of fusion events. It is unclear, however, whether these represent one bin of an exponential distribution of a longer timescale process. In our experiments, the fastest events (2.75-sec time bin) represented ≈17% of all events; however, they were simply part of a continuous exponential distribution with a mean retrieval timescale of 14 s. If a separate fast mode were present in our experiments, it would manifest as excess events in the first time bin, which were not observed. Our stochastic distribution is well described by a single mean lifetime of ≈14 s. Although it is possible that kiss-and-run might operate only under higher-frequency stimulation, endocytosis in this regime was recently shown to depend completely on dynamin 1 (3). Although it is premature to assume that within this single stochastic distribution all events are based on identical molecular mechanisms, direct tests of the molecular basis of endocytosis should be possible at much higher sensitivity than was previously feasible. However, the data on clathrin knockdowns, together with the high-sensitivity measurements presented here, as well as the analysis of endocytosis during high-frequency activity in the absence of dynamin 1, all indicate that kiss-and-run must occur only rarely at these small CNS synapses. Instead, we propose that the previous evidence for faster retrieval of synaptic vesicles was simply one end of an exponential distribution of retrieval times that are part of a single mode of endocytosis.

In summary, we described a quantitative characterization of endocytosis at the single-vesicle level indicating that a single mode of endocytosis controls synaptic vesicle retrieval. The stochastic nature of the coupling suggests that a single rate-limiting step dominates the retrieval time for endocytosis.

Materials and Methods

Cell Culture.

Hippocampal CA3–CA1 regions were dissected from Sprague–Dawley rat pups, dissociated, and plated onto polyornithine-coated glass coverslips, as described (23). Optical recordings were performed 14–21 days after plating, 8–14 days after transfection with vGpH, on coverslips using a laminar-flow perfusion and stimulation chamber (volume, ≈75 μl) on the stage of a custom-built laser-illuminated epifluorescence microscope. Images (512 × 512) were acquired every 170 msec using 100-msec integration with an Andor iXon+ (model no. DU-897E-BV) back-illuminated EMCCD camera using <1-μW/μm² 488-nm Ar⁺ ion acoustooptically shuttered laser whose beam was expanded to fill the field of view, a 40 × 1.3-N.A. Fluor Zeiss objective, and a 1.6 × tube lens using 515- to 560-nm emission and 510-nm dichroic filters. Action potentials were evoked by passing 1-ms current pulses, yielding fields of ≈10-V/cm via platinum-iridium electrodes. Cells were continuously perfused (1–1.5 ml/min, ≈25°C) in a saline solution containing 119 mM NaCl, 2.5 mM KCl, 2 mM CaCl₂, 2 mM MgCl₂, 25 mM Hepes (buffered to pH 7.4), 30 mM glucose, 10 μM 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX; Research Biochemicals), and 50 μM d,l-2-amino-5-phosphonovaleric acid (AP5; Research Biochemicals). When different [CaCl₂] was used, [MgCl₂] was increased to keep the divalent ion concentration constant. Bafilomycin A1 (Baf, Calbiochem) was used at 0.5 μM and applied for 30 s before the beginning of data acquisition. Unless otherwise noted, all chemicals were obtained from Sigma.

EGFP Studies.

EGFP (Biovision) was immobilized in acrylamide gel according to ref. 24. Images of several regions in the gel were acquired before and after bleaching with 100-W Hg Arc lamp illumination for 5 min. Average pixel intensities of diffraction limited spots were measured after 3 × 3 smoothing and field flattening based on images of a dilute fluorescein solution. Only puncta whose intensity was >2.5 times the standard deviation of the background signal in the absence of GFP were used for analysis.

Photobleaching Estimates.

Photobleaching rates were estimated by measuring the fluorescence decay of vGlut1-pHLuorin-expressing boutons in the presence of NH₄Cl. The bleaching rate for pHluorin in the alkaline state for the majority of the data was ≈0.01% per frame. In the acid-quenched state, the bleaching rate should be ≈25 times lower.

Image and Data Analysis.

Images were analyzed in ImageJ (http://rsb.info.nih.gov/ij) by using custom-written plugins (http://rsb.info.nih.gov/ij/plugins/time-series.html). All visible varicosities were selected for analysis by testing their responsiveness to multiple (three to four) rounds of test stimuli delivered as six AP trains at 33 Hz. Only boutons showing fluorescence changes of at least 1quantum in the 16–24 stimuli were included for further analysis. This effectively sets a lower bound on the release probability included in the data set of P ≈ 0.06. Average fluorescence intensities were obtained over a circular region of interest of 1.25-μm radius for each bouton that avoided fluorescence decay due to faster diffusive processes (4). ΔF_1AP was calculated as two point difference (ΔF_i) in intensity of the frame (n) on which the stimuli arrived and (n + 1) or (n + 2), whichever is greater. A histogram of ΔF_1AP values was fit to multiple Gaussians. For single-vesicle fluorescence decays, the t_1/2 time was identified as the first instant in time at which the adjacent average has decayed to half the initial amplitude and remains below the 50% intensity point for the remainder of the trace. The ensemble average curve in Fig. 2C was fit to the function

that describes the time course of two coupled first-order processes with time constants for endocytosis (τ_n) and reacidification [τ_r, set to 4 s, as determined (4, 16)] where amplitudes A₁ and A₂ are positive vlaues.

Fitting.

Levenberg–Marquardt χ² minimization for nonlinear least-square fitting was carried out in Origin (OriginLab). The areas and positions of each peak the quantal histograms were allowed to vary freely, but the width of all peaks was taken as a single-fitting parameter, which assumes that the noise in our system is dominated by a combination of systematic instrumentation noise (e.g., drifts, EMCCD clock noise, laser-illumination noise), and the noise associated with a baseline fluorescence derived from the population of acidic vesicles whose fluorescence is each ≈1/25th of an alkaline vesicle. We verified the robustness of the quantal description of the data by fitting the derived best-fit peak positions as a linear function of peak number (Fig. 1D); the slope of this fit gives the quantal size. Quantal-size estimates remained invariant over a wide range of bin sizes, as shown in Table 1. Poissonian error is assumed for calculating the reduced χ² value in each of our fittings.

Table 1.

Fitting errors as a function and χ² estimates for quantal analysis

Bin size, a.u.	Quantal size, a.u.	Error	Reduced χ2
4.75	52.0	2.0	0.73
9.75	50.5	2.0	0.97
12.5	49.5	0.9	1.14