Single synapse glutamate imaging reveals multiple levels of release mode regulation in mammalian synapses.

Mammalian central synapses exhibit vast heterogeneity in signaling strength. To understand the extent of this diversity, how it is achieved, and its functional implications, characterization of a large number of individual synapses is required. Using glutamate imaging, we characterized the evoked release probability and spontaneous release frequency of over 24,000 individual synapses. We found striking variability and no correlation between action potential-evoked and spontaneous synaptic release strength, suggesting distinct regulatory mechanisms. Subpixel localization of individual evoked and spontaneous release events reveals tight spatial regulation of evoked release and enhanced spontaneous release outside of evoked release region. Using on-stage post hoc immune-labeling of vesicle-associated proteins, Ca2+-sensing proteins, and soluble presynaptic proteins we were able to show that distinct molecular ensembles are associated with evoked and spontaneous modes of synaptic release.

Introduction

More than half a century ago, seminal work at the neuromuscular junction identified action potential (AP)-evoked and AP-independent spontaneous neurotransmission as two discrete modes of information transmission at chemical synapses (Fatt and Katz 1952). Upon stimulation of presynaptic terminals, the size of postsynaptic currents tended to occur at integer multiples of a quantal response. This quantal response corresponded with the amplitude of spontaneous “mini” postsynaptic potentials that occur in the absence of depolarization. Subsequent work identified synaptic vesicles (SVs) as the means of packaging and releasing these quanta of neurotransmitter during synaptic transmission. In the following decades a simplified model of the synaptic transmission was established in which, upon AP-induced calcium influx, synaptic vesicles fuse with a variable release probability at a variable number of discrete release sites within a presynaptic bouton. These two sources of variability have been suggested to be the key contributors to heterogeneity in AP-evoked postsynaptic potentials among synapses (Dobrunz and Stevens 1997).

Although there is consensus on many fundamental aspects of evoked synaptic transmission, the mechanisms and physiological relevance of spontaneous synaptic release remain contentious. Initially spontaneous release was believed to share the same presynaptic machinery as evoked release and merely be stochastic noise inherent in all synapses. Likewise, an abundance of work suggests that evoked and spontaneous release originate from the same pool of synaptic vesicles (Groemer and Klingauf 2007; Hua et al., 2010; Wilhelm et al., 2010). However, other work, employing similar methods has reached the opposite conclusion (Sara et al., 2005; Fredj and Burrone 2009). There is an emerging view that these two forms of release do result from differential presynaptic regulation (Groffen et al., 2010; Ramirez et al., 2012), can even target different postsynaptic receptors (Reese and Kavalali 2016), and play essential roles in homeostatic plasticity (Sutton and Schuman 2005). One key hindrance, preventing resolution of these inconsistencies, is the lack of adequate methods to directly study discrete release events at the synapse.

Our current mechanistic understanding of synaptic transmission is based primarily on decades of electrophysiological work at a limited number of large model synapses (Llinas et al., 1994; Geiger and Jonas 2000; von Gersdorff and Borst 2002). Interrogation of synaptic release from smaller “simple synapses” that constitute the majority of synapses in the central nervous system remains challenging (Malagon et al., 2016). Various fluorescent reporters allow for spatiotemporal discrimination of synaptic activity with optical microscopy. FM dyes and pH-sensitive GFPs (pHluorins) have been used extensively to interrogate the synapse (Harata et al., 2001; Waters and Smith 2002; Leitz and Kavalali 2014). However, robust single vesicle response has been difficult to resolve above noise and variable probe copy number per vesicle convolutes quantal analysis. Calcium reporters that report on N-methyl-D-aspartate receptor (NMDAR)-mediated [Ca2+] accumulation in individual spines have been shown to reliably report on single release events. However, the diffusion and accumulation of calcium limits accurate localization of release in space and time. Furthermore, both pre- and postsynaptic processes regulate the measured response amplitude (Melom et al., 2013; Peled et al., 2014). Recently, the development of fluorescent probes that directly report on glutamate concentration in the synaptic cleft (Okubo et al., 2010; Marvin et al., 2013) have shown great promise in providing the resolution required to “see” synaptic release.

In this work, we present a method for detecting and quantifying individual exocytotic events. We do this in a large field of view covering hundreds of synapses from dissociated hippocampal cultured neurons. We show that intensity-based glutamate-sensing fluorescent reporter (iGluSnFR) (Marvin et al., 2013) provides sufficient sensitivity to resolve quantal vesicle release. This resolution allows us to not only accurately characterize evoked release properties but also, importantly, reliably detect spontaneous events. We characterize evoked release probability as the fraction of stimuli for which response is detected and spontaneous fusion rates as the number of detected events per given time in the absence of stimuli. We measure these two properties at thousands of individual synapses and show that there is substantial variability among them but no correlation between them. We show that spontaneous release tends to occur over a larger area of the synapse. Utilizing on-stage post hoc immunostaining to measure the local abundance of synaptic constituents, we show that the inherent variability in synaptic properties and molecular composition can be used to predict the roles of synaptic proteins. Our results shed new light on the relationship between evoked and spontaneous release. Rather than the existence of master regulators of spontaneous release, this work suggests that a combination of different factors is preferentially associated with each release mode.

Results

Glutamate imaging provides a robust readout of discrete synaptic transmission events

To measure synaptic transmission at individual central synapses, we prepared primary hippocampal cultures from newborn (P0) rat brains and infected the cells with an adenovirus-associated construct to express iGluSnFR (Marvin et al., 2013). iGluSnFR uniformly distributed on the extracellular surface of cell body, dendritic arbor, and axonal projections of the neurons where it reports on extracellular glutamate concentration (Figures 1A and S1A, Videos S1 and S2). Upon low-frequency field stimulation, we observed small increases in fluorescence above background (Figure 1B, top panels), which were localized after high-pass filtering (magenta spots in Figure 1B). In addition to robust quantification of evoked release (Figure 1C, top trace, Figure S2), we were also able to resolve spontaneous glutamate release in the presence of 500 nM tetrodotoxin (TTX) at the dendritic spines (Figure 1C, bottom trace) and both release modes at axonal terminals (Figure S1C).

Figure 1

Glutamate imaging provides a robust signal of discrete synaptic transmission events

(A) When expressed in hippocampal neurons, iGluSnFR localizes to the outer leaflet of the plasma membrane.

(B) During synaptic release small increases in fluorescence above the relatively high background were observed. After subtracting pre-stimulation from the post-stimulation image, areas of activity became more apparent. For precise localizations of active regions of interests (ROIs), the maximum projection of the first derivative over time of an image stack was obtained after high-pass image filtering (magenta spots in the bottom right panel).

(C) Examples of single-synapse fluorescence traces sampled from raw images in response to electrical stimulation (top) and in the presence of TTX (bottom) to capture spontaneous activities. Scale bars, 10 μm in (A and B).

Quantal resolution and presence of multi-quantal release

To assess the quantal resolution of the sensor, we performed 5-min-long AP-evoked measurements of glutamate release during a train of electrical field stimuli at 0.5 Hz in the presence of 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX) and 2-amino-5-phosphonovaleric acid (AP5), to block recurrent network activity. We then built histograms of stimulus-evoked fluorescence amplitudes from individual synapses. The resulting histograms often showed multiple peaks and fit well to a sum of Gaussian distributions (Akaike, 1973, McLachlan and Peel, 2000). The histograms shown in Figures 2A and 2B both have prominent peaks around zero, indicating frequent release failures. The second peak is centered about the unitary quantal amplitude. In some cases, clearly resolved peaks at integer multiples of the quantal amplitude were observed (Figures 2B and S3), providing an evidence of multiquantal release. These data clearly illustrate the quantal resolution of iGluSnFR, giving us the confidence to use this sensor to reliably identify all release events at small central synapse.

Figure 2

Resolution of unitary release quanta and presence of multivesicular release at individual boutons

(A) The histogram of individual stimulus-evoked fluorescence change amplitudes measured during 0.5-Hz stimulus trains fit well to a sum of Gaussian distributions (black dashed line). The prominent distribution centered at zero represents release failures, and the first non-zero peak represents the unitary quantal amplitude at that synapse.

(B) In approximately 10% boutons, multi-quantal release at individual boutons is indicated by additional distributions centered near integer multiples of the unitary quantal amplitude.

(C) Fluorescence changes during field stimulation-evoked release and AP-independent spontaneous glutamate release show similar rise time, amplitudes, and recovery.

(D) The spontaneous amplitude (acquired during the first minute)/evoking amplitude ratio decreases with increasing evoked release probability, indicating the increased likelihood of multi-quantal release. Error bars represent SEM (n > 1,000 synapses).

To compare the relative quantal amplitudes of evoked and spontaneous glutamate release at the same subset of synapses, we followed measurements during a 20 AP, 0.5-Hz train stimuli with a 5-min incubation in 500 nM TTX to block AP generation and then a 5-min measurement to sample spontaneous glutamate release. The amplitude and time course of evoked and spontaneous fluorescence changes in the same synapse were similar, with amplitudes in the range of 0.2–0.6 relative change in fluorescence (ΔF/Fo), rise time within a single frame (50 ms), and decay constants on the order of ∼100 ms (Figure 2C).

We then compared the spontaneous release amplitude of events occurring within the first minute of the measurement (unbleached unitary quantal amplitude) with the averaged amplitudes of evoked release over a range of probabilities. We see that the ratio of spontaneous to evoked amplitudes starts at one for synapses with low release probability, likely uniquantal evoked release, and decreases with increasing evoked release probability (Figure 2D). This, together with Figure 2B, clearly indicates that even for synapses with moderate evoked release probability, multi-quantal release can occur in dissociated hippocampal culture.

Differences in the distribution of evoked and spontaneous fusion sites

During the measurements of evoked and spontaneous release described above, the maximum intensity projection of the difference image was used to localize active synapses. These often occurred at morphologically distinct regions identifiable as axonal boutons or dendritic spines (Figure 3A). When reviewing images of individual events occurring at an identified synapse, there was strong overlap but often an apparent jitter in the location of release. To characterize whether there are differences in spatial distribution between release modes, we selected synapses with similar numbers of measured evoked cand spontaneous responses and performed subpixel localization of the centroid of release by least-square fitting of a Gaussian distribution to the initial release signal of each event (Figures 3B and 3C) (Ovesny et al., 2014). Evoked release tended to occur within a tight spatial distribution, likely indicating the bounds of the active zone, whereas spontaneous release was more diffuse likely occurring at both the active and periactive zones. Farthest distance between each type of release was computed from 182 synapses. The cumulative distribution of farthest distances shows significantly tighter spatial regulation of evoked release sites when compared with spontaneous release sites (Figure 3D). Although the measurements of spontaneous release do occur over a longer window than those of evoked, there is no dependence of the distance between observed release events and the time between these events (Figure S4). Further analysis of the areas covered by spontaneous and evoked release events (Figure 3E) showed much higher spontaneous release densities outside of evoked release areas (Figure 3F). These data are in complete agreement with a previous study where a significantly smaller evoked fusion area compared with spontaneous fusion sites was observed in pHluorin-based measurements in individual hippocampal boutons (Tang et al., 2016).

Figure 3

Subpixel localization of evoked and spontaneous release events reveals distinct spatial regulation

(A) To assess the spatial organization of AP-evoked and spontaneous release sites in the same bouton, synapses with at least 10 evoked responses and 5 spontaneous release events were selected.

(B) Sub-pixel coordinates of the release sites for both evoked (red) and spontaneous release events (cyan) were estimated from the centroid of the initial change in fluorescence.

(C) Release site coordinates for evoked (red) and spontaneous (cyan) release site localizations shown as normalized symmetric 2D Gaussian functions with a standard deviation equal to their localization uncertainty. The mean localization uncertainty for fits of evoked and spontaneous release events were 86.9 and 89.5 nm, respectively.

(D) Farthest distances between release sites were computed for evoked and spontaneous release in 182 synapses. The cumulative distribution of farthest distances shows significantly tighter spatial regulation of evoked release sites (Kolmogorov-Smirnov test, p value = 0.008).

(E) The surface areas covered by evoked and spontaneous release events were estimated by the convex hulls around their sub-pixel coordinates.

(F) The density of detected spontaneous release is on average greater inside spontaneous release areas than in evoked release areas. Scale bars: 10 μm in (A) and 1 μm in (B).

Intersynaptic variability in strength of evoked and spontaneous release

From the measurements described above, we could measure the evoked (Figures 4A and 4B) and spontaneous (Figures 4C and 4D) glutamate release events at over 400 individual synapses in a single field of view. By performing these measurements at more than 60 cells, we could characterize the evoked release probability and spontaneous fusion frequency at over 24,000 individual synapses. We obtained an averaged release probability of 0.25 (Figure 4E) and spontaneous frequency of 0.011 Hz (Figure 4F). Excluding synapses that exhibited only spontaneous activity increased the averaged release probability to 0.37, which agrees with the recent Ca2+ imaging measurements in hippocampal cultures (Reese and Kavalali 2016). However, as in our measurements evoked probabilities smaller than 0.05 are rounded to zero, it is likely that the true estimate of mean release probability lies in between the two reported values.

Figure 4

Imaging evoked and spontaneous synaptic release reveals intersynaptic variability of release properties

(A and B) The location of release sites and corresponding evoked release probability, designated by color, of hundreds of active synapses in a single field of view (A) can be computed from release events detected during a train of 20 stimuli applied at 0.5 Hz (B).

(C and D) Spontaneous release frequency is measured in the same sample (C) for 5 min following an initial 5-min incubation in 500 nm TTX (D). Each dot in (B and D) represents one release event.

(E) A histogram of evoked release probability from over ~24,000 individual synapses reveals a mean release probability of 0.25 and a large population (31%) that shows no evoked release.

(F) The histogram of the spontaneous frequency shows a mean of 0.011 Hz.

(G) The evoked release probability and spontaneous release frequency from all measured synapses are plotted in a density-scatter plot. The color represents the number of synapses for which a given pair of values was measured. No correlation was measured between evoked probability and spontaneous frequency (white line, R2 = 0.014). Scale bar: 10 μm in (A and C).

Similar to previous observations at the Drosophila neuromuscular junction (Melom et al., 2013) and hippocampal synapses (Reese and Kavalali 2016), the distribution of the number of spontaneous events observed within a certain time (5 min) was non-Poisson (Figure S5), implying that the differences measured among synapses is not the result of the same stochastic process. When we plotted the two functional properties of synapses against each other, we observed a remarkable variability. The majority of synapses exhibited both evoked and spontaneous neurotransmission (Figure 4G). No significant correlation (dashed white line in Figure 4G, R2 = 0.014) was observed between the two forms of release, suggesting that discrete mechanisms are employed.

Release properties did seem to be correlated for neighboring hippocampal boutons. We observed large spatial correlation of evoked release probability between synapses with a length constant of 6.7 μm (Figure S6A). For spontaneous frequencies, we observed much smaller correlation between synapses with a shorter length constant of 2.7 μm (Figure S6B). The functional correlation between neighboring synapses has been observed by numerous other studies in both primary cultures (Murthy et al., 1997; Branco et al., 2008; Taschenberger et al., 2016), and tissue slices (Koester and Johnston 2005; Staras et al., 2010). This is likely due to the high probability of adjacent pre-synapses originating from the same axon where inter-synaptic sharing of second messengers, proteins, and SVs have been shown to occur (Darcy et al., 2006; Staras et al., 2010).

Molecular variability among central synapses

To test whether the observed functional variability may be the result of molecular heterogeneity at the synapse, we first characterized the relative expression of a variety of presynaptic proteins including vesicle-associated proteins and Ca2+-sensing and presynaptic soluble proteins at hippocampal synapses. We performed immunolabeling of a given protein of interest and the active zone protein Bassoon (Figures 5A and S7). The punctate Bassoon labeling was used to identify synapses, provide a measure of relative active zone size (Tang et al., 2016), and to identify regions from which the intensity of the other labeled proteins were sampled. The variability of expression relative to Bassoon differed extensively across tested proteins (Figure 5B). To compare the degree to which expression heterogeneity differed among proteins, we computed the Pearson correlation coefficient of the intensity of labeling relative to that of Bassoon. Canonical SV proteins like vesicle-associated membrane protein (VAMP) 2 displayed a relatively tight correlation with Bassoon expression (R2 = 0.69), whereas some targets, such as VAMP7, were much more heterogeneously expressed across hippocampal boutons (R2 = 0.16). These measurements revealed varying levels of heterogeneity of individual synaptic proteins (Figures 5C–5E) that when taken together compound into vast heterogeneity in molecular composition at presynaptic boutons.

Figure 5

Vast variability in relative molecular expression across central synapses

(A) Representative images of co-labeling for Bassoon and different presynaptic proteins; scale bar, 10 μm.

(B) Flat-field corrected intensities of tested proteins against flat-field corrected Bassoon intensities shows that the expression profile of different presynaptic proteins varies extensively among central synapses (dashed lines represent linear fitting; R2 for VAMP2, Munc18-1, and VMAP7 are 0.69, 0.42, and 0.16, respectively).

(C–E) Analysis of correlation coefficient between Bassoon and vesicle-associated (C), Ca2+-sensing (D), and presynaptic soluble proteins (E) show the greatest heterogeneity (smallest correlation coefficient) for non-canonical SV proteins (box: first and third quartile, line: median, whiskers: the minimum and maximum values [n = 4–8 cells per protein of interest]).

Relative abundance of synaptic proteins correlates with functional properties

To directly compare the functional properties and molecular composition of synapses, we performed on-stage immuno-labeling following glutamate release imaging. For each measurement, we obtained the evoked probability, the spontaneous release frequency, and the abundance of given protein relative to Bassoon at individual synapses (Figure 6A, compare with Figures 4A and 4C). In line with a previous report (Holderith et al., 2012), we observed a positive correlation between evoked release probability and active-zone size (Bassoon intensity) (Figure 6B). We also show that spontaneous release rate scales with synapse size (Figure 6C). Consistent with the observed intersynaptic spatial correlation of release properties (Figures S6A and S6B), we observed that the amount of Bassoon labeling was also correlated among neighboring synapses with a similar length constant of 5 μm (Figure S6C).

Figure 6

Functional properties of synapse correlate with synapse size and molecular composition

(A) Active synapses (white circles) superimposed on the Bassoon-labeled image (see Figures 4A and 4C, scale bar, 10 μm).

(B) The averaged evoked probabilities of synapses show an increase as the synapse size (estimated by the abundance of Bassoon) increases.

(C) Averaged spontaneous frequencies of hippocampal boutons also scale with their size. Error bars in (B and C) represent SEM (n is on average 1,731 synapses per point).

(D) “Spontaneous” (gray) and “evoked” (blue) pools marked on the scatterplot of the total dataset.

(E) Paired comparison between the relative abundance of each protein in spontaneous synapses and evoked synapses of the same cell (each two black circles connected by a dashed line represent one cell. Error bars represent SEM [n > 100 synapses per black circle]).

(F–H) Relative enrichments of vesicle-associated proteins (F) and Ca2+-sensing (G) and presynaptic soluble proteins (H) in spontaneous synapses (box: first and third quartile, line: median, whiskers: the minimum and maximum values [n = 4–8 cells per protein of interest]).

To check for the correlation between release modes and molecular properties of synapses, we divided the synaptic measurements into three equal-sized pools (Figure 6D): (1) a “spontaneous release” pool with no or relatively low evoked release (marked in gray), (2) a “mixed” pool with similar relative of evoked and spontaneous activity and (3) an “evoked release” pool with no or relatively low spontaneous release (marked in blue). To mitigate variation in labeling efficiency between cells, we only compared the relative abundance of each protein within the same experiment (Figures 6E and S8). We were then able to distill these values into the relative enrichment of a protein in spontaneous synapses by dividing their relative abundance in “spontaneous” pool to that of the “evoked release” pool (Figures 6F–6H).

In case of vesicle-associated proteins, we observed more than 1.5-fold enrichment (did not reach statistical significance) of non-canonical N-ethylmaleimide-sensitive factor attachment receptors (SNAREs) proteins, namely, Vti1a and VAMP7, in the “spontaneous” pool synapses, whereas VAMP2, the classical vesicular SNARE, displayed a significant relative enrichment in the “evoking” pool of synapses. Synapsin-1 also showed significant enrichment in the “evoking” synapses (Figure 6F). Among tested Ca2+ sensors we observed significant enrichment of Doc2b in the “spontaneous” pool and Synaptotagmin 1 in the “evoked” pool (Figure 6G). Interestingly, all tested presynaptic soluble proteins also showed significantly higher expression in the “spontaneous” pool synapses (Figure 6H).

Discussion

Here we used genetically encoded glutamate sensor to image release at thousands of synapses in neuronal cultures. This enabled us to characterize the variability, interdependency, spatial localization, and molecular regulation of AP-evoked and spontaneous synaptic vesicle exocytosis.

Although still most often used to measure glutamate release during stimulus trains and bursts of activity (Namiki et al., 2007; Marvin et al., 2013) some glutamate imaging work has focused at discrete events at single synapses (Taschenberger et al., 2016). Recently, Sakamoto and colleagues used an enhanced E (glutamate) optical sensor (EOS) hybrid glutamate sensor to investigate activity at single synapses (Sakamoto et al., 2018). Using fluctuation analysis, they were able to fit a binomial model of glutamate release at the synapse to determine quantal parameters. With iGluSnFR, we show that the signal amplitude from evoked release synapses with low evoked probability is equivalent to that released during spontaneous fusion. We obtained the same results with a new variant of iGluSnFR (Figure S9), which became available during the time course of our study (Marvin et al., 2018; Jensen et al., 2019). With both sensors, we also clearly show that increasing the evoked release probability results in an increase in amplitude compared with spontaneous “mini” amplitude indicating that multi-quantal release occurs in small synapses, even at relatively low evoked probabilities (Figures 2D and S9B).

It should be noted that our interpretation of these data is based on the assumption that release events are happening at individual synapses. It has been shown that multi-synapse aggregates do exist in culture (Kavalali et al., 1999; Welzel et al., 2010). However, the coefficient of variation (CV) for our estimates of active zone size, based on Bassoon labeling (Figure S10, CV = 0.5), are in close agreement with the single active zone area distribution measured by electron microscopy (EM) (CV = 0.56 and 0.7 in brain slices and culture, respectively, Schikorski and Stevens 1997), indicating that the vast majority of sites from which we measure activity are single synapses.

Typically, our measurements provide coverage of most of the dendritic arbor of a cultured neuron (219 μm × 219 μm field of view), capturing activity of hundreds of synapses in a single experiment. This enhanced throughput allowed us to characterize the evoked release probability and spontaneous fusion frequency in more than 24,000 synapses. Early work using styryl dyes reported tight correlation of evoked and spontaneous activity at individual synapses (Prange and Murphy 1999, r = 0.59; p < 0.0001). This work concluded that both forms of release were regulated in parallel, originating from the same pool of readily releasable vesicles and utilizing the common release machinery. However, more recent studies at the Drosophila neuromuscular junction offered conflicting reports. One group reported on a negative correlation between these properties (Peled et al., 2014) suggesting that highly active synapses preferred one mode of transmission, whereas the others concluded that these properties were poorly correlated (Melom et al., 2013). Work that utilized postsynaptic calcium measurements in hippocampal cultures also indicated that these properties are poorly correlated (Reese and Kavalali 2016, R2 = 0.012). Our results, confirm that there is no positive or negative correlation (R2 = 0.014) between evoked and spontaneous release probability in hippocampal boutons. This would indicate that, for the most part, these forms of release are independently regulated.

That is not to say that evoked and spontaneous release are completely free from co-regulation. It is well established that synaptic structure is a major factor regulating synaptic release. Evoked and spontaneous release efficiency is known to scale with vesicle pool size (Murthy et al., 1997). Evoked release has also been shown to scale with active zone area (Holderith et al., 2012). Indeed, our results indicate that both evoked (Figure 6B) and spontaneous release (Figure 6C) scale with active zone size in hippocampal boutons. The existence of scaling with active zone size for both forms of release, yet lack of correlation between the properties, provides additional support of the presence of other stronger mechanisms of differential regulation.

The ultrastructural organization within a synapse also regulates exocytosis. Tight spatial coupling of primed vesicles to voltage-gated calcium channels is essential for synchronous evoked release (Eggermann et al., 2011). Many synapses employ specialized structures such as ribbons or T-bars to facilitate vesicle clustering and coupling to calcium channels to further increase evoked release fidelity (Ackermann et al., 2015). The spatial regulation of spontaneous release, however, remains poorly understood. Work in the large ribbon synapses of retinal bipolar (Zenisek 2008) and photoreceptors cells (Cork et al., 2016) has shown differences in spontaneous fusion rates across the synapse. The ribbon is the predominant site of evoked release, whereas higher rates of spontaneous release occurred at non-ribbon sites. Based on the subpixel localization of evoked and spontaneous glutamate release, we also observed significant differences in spatial distribution of evoked and spontaneous release sites. We show that spontaneous release occurs over a larger area of hippocampal boutons than evoked release (Figures 3B–3F) and with higher frequency outside the bounds of evoked release. Super-resolution imaging has shown tight clustering of active zone proteins at discrete release sites of hippocampal boutons (Sakamoto et al., 2018). Similarly, tight sequestration and peripheral depletion of factors responsible for spontaneous release clamping (Schupp et al., 2016), could leave peripheral vesicles more susceptible to spontaneous exocytosis. Together with the previously shown transsynaptic alignment of postsynaptic receptors at the fusion sites (Tang et al., 2016), this spatial regulation of spontaneous and evoked release sites could explain how these release modes preferentially activate non-overlapping populations of postsynaptic receptors (Atasoy et al., 2008; Sara et al., 2011).

It should be noted, that the subset of synapses selected for subpixel localization possessed above average release properties to provide sufficient event localizations for comparison of evoked and release distribution. It is likely that release properties and release localization distribution are interdependent and there may be combinations of evoked and spontaneous release properties where the observations presented in Figure 3 do not hold. Future investigation using this same experimental framework may provide insight into these dependencies as well as the dynamics of release site localization during synaptic maturation or plasticity.

Several decades of work, mostly reliant on genetic manipulation, has provided a solid foundation for the molecular regulation of synaptic release. This past work has identified a number of factors that are essential for SV exocytosis and others that serve more modulatory roles. We show that a number of vesicle-associated, calcium-sensing, and soluble presynaptic proteins are present at dramatically different levels across synapses. For some of the tested proteins, we noticed that in addition to the established presynaptic localization, apparent dendritic staining was observed (Figures 5 and S7). Nevertheless, by performing post hoc immunolabeling of synapses that show more relative spontaneous activity, more relative evoked release, or equal contributions to both forms of release, we were able to show that for a number of these proteins their abundance relative to Bassoon does correlate with presynaptic function. Although we show that iGluSnFR is able to reliably report on the spontaneous release events, the window over which sufficient signal to noise ratio (SNR) is achieved is limited. This, in turn, limits the resolution with which release frequency can be measured. With a median number of observed spontaneous events of approximately 3 per 5-min window the confidence of spontaneous release frequency and thus molecular and spatial correlates for a single synapse may be low. However, we show that by pooling thousands of synapses by their evoked and spontaneous release properties significant molecular correlations become apparent.

VAMP2 and synaptotagmin 1, whose absence results in severe decrease in AP-evoked SV fusion (Geppert et al., 1994; Schoch et al., 2001; Chapman 2008) displayed significantly greater relative abundance in synapses that tend to have relatively stronger evoked release. Similarly, Synapsin 1 was found to be expressed to a greater extent at synapses with relative higher evoked probabilities. Synapsin 1 plays an important role in evoked neurotransmission by tightening the vesicular release to physiological stimuli (Chi et al., 2003); however, spontaneous release has been reported to stay unchanged after its loss (Chiappalone et al., 2009).

Doc2b, on the other hand, was found to be enriched in synapses with higher relative spontaneous fusion. Doc2b has been proposed to be a calcium sensor for spontaneous release (Groffen et al., 2010) and a calcium-independent enhancer of spontaneous release (Pang et al., 2011). In addition, two vesicle-associated membrane proteins, Vti1a and VAMP7, showed higher expression in synapses with higher relative spontaneous release but did not show significant enrichment. In addition to their roles in fusion of intracellular organelles such as late endosomes (Medigeshi and Schu 2003; Ganley et al., 2008) and lysosomes (Pryor et al., 2004), these proteins are proposed to define a “spontaneous pool” of SVs (Hua et al., 2011; Ramirez et al., 2012; Bal et al., 2013). The abundance of Vti1a was shown to correlate with spontaneous release activity, and its loss selectively reduced spontaneous neurotransmission (Ramirez et al., 2012). Vamp7, however, is proposed to regulate asynchronous release as well (Scheuber et al., 2006), which might explain its lower enrichment when compared with Vti1a at synapses with higher relative spontaneous activity.

Surprisingly the three soluble synaptic proteins we tested, Munc13, Munc18, and Complexin, were found to be significantly enriched in synapses with stronger relative spontaneous release. Munc18 and Munc13 cooperate to prime the ternary SNARE complex (Lai et al., 2017). In Munc18-1-deficient mice and Munc13-1/2 double knockout mice both evoked and spontaneous synaptic neurotransmitter release are completely abolished (Verhage et al., 2000; Varoqueaux et al., 2002). Our results indicate that excess abundance of these factors, relative to Bassoon, leads to enhanced spontaneous release. Munc13 has been shown to tightly cluster around sites of evoked release (Sakamoto et al., 2018). It may be the case that when local abundance of a protein exceeds that of other active zone proteins (i.e., Bassoon), the capacity to effectively sequester it within evoked release sites is diminished. Then excess Munc13 and Munc18 are free to dock and prime vesicles outside of the organized active zone, where we show more spontaneous release occur. This provides insight into why phorbol esters, that potentiate all release through Munc13 (Rhee et al., 2002), have been shown to specifically enhance spontaneous release outside of organized release sites (Zenisek 2008).

Despite a history of conflicting reports, Complexin is now believed to interact with the trans-SNARE complex to regulate the energetic requirements for SV fusion. Whether this interaction clamps or triggers SV fusion seems to depends on the isoform of protein, the synapse type, as well as the model system used in the study (Neher 2010; Trimbuch and Rosenmund 2016). Our results showing an enrichment of Complexin in synapses with higher relative spontaneous activity suggests a facilitatory role for Complexin, in line with numerous other works (Xue et al., 2008; Strenzke et al., 2009; Xue et al., 2010; Chang et al., 2015). Most recently, fluorescence resonance energy transfer assay using mammalian proteins has shown that although it is not essential to form the primed state of the trans-SNARE complex, Complexin, together with Munc13 and Munc18, protects it from disassembly (Prinslow et al., 2019). It would follow that synapses with greater abundance of these factors would have a greater number of primed vesicles available for release.

Taken together, our results show that the regulation of evoked and spontaneous release occurs at many different levels that, when combined, result in vast functional variability across synapses. It is evident that further resolution of the coordinated and divergent mechanisms will require greater resolution of key molecules within the ultrastructure of the synapse. The combination of functional optical assays, such as the glutamate release imaging used in this work, together with multicolor super-resolution techniques, will allow us to better understand how molecular ultrastructure governs the strength of different release modes at the synapse.

Limitations of the study

As with most works that utilize fluorescent indicators, particularly genetically encoded indicators, the characterization of synaptic release in this work is limited by the finite photon budget of iGluSnFR. Care was taken to find an optimal balance between spatial, temporal, and signal resolution in the detection of synaptic release. Indeed, sufficient SNR was achieved to resolve unitary release events. Measurements of evoked release allowed us to assign synaptic release probabilities in 5% intervals. However, the window over which sufficient SNR was maintained while measuring spontaneous release was limited to 5 min. During this time the vast majority of the synapses exhibiting fewer than five spontaneous release events, resulting in courser discretization of spontaneous release frequency. Several groups are working to further optimize glutamate sensors to increase their brightness and modulate their affinity. These advancements together with advanced imaging methods, such as targeted illumination, may increase our photon budget and thus the amount of information about glutamate release one is able to gather from a single synapse.

Resource availability

Lead contact

Further information and requests for resources and reagent should be directed to the Lead Contact Andrew Woehler (andrew.woehler@mdc-berlin.de).

Material availability

This study did not generate any new or unique reagents.

Data and code availability

The datasets and code supporting the current study have not been deposited in a public repository due to their size but are available from the corresponding author upon request.

Methods

All methods can be found in the accompanying Transparent methods supplemental file.