Examining molecular determinants underlying heterogeneity of synaptic release probability using optical quantal imaging

Neurons communicate through neurotransmitter release at specialized synaptic regions known as active zones (AZs). Using transgenic biosensors to image postsynaptic glutamate receptor activation following single vesicle fusion events at Drosophila neuromuscular junctions, we analyzed release probability (Pr) maps for a defined connection with ~300 AZs between synaptic partners. Although Pr was very heterogeneous, it represented a stable and unique feature of each AZ. Pr heterogeneity was not abolished in mutants lacking Synaptotagmin 1, suggesting the AZ itself is likely to harbor a key determinant(s). Indeed, AZ Pr was strongly correlated with presynaptic Ca2+ channel density and Ca2+ influx at single release sites. In addition, Pr variability was reflected in the postsynaptic compartment, as high Pr AZs displayed a distinct pattern of glutamate receptor clustering. Developmental analysis suggests that high Pr sites emerge from earlier formed AZs, with a temporal maturation in transmission strength occurring over several days.

release rates (75% percentile of P r was 0.1, with a maximum P r of 0.73, Figure 1B). Indeed, the 136 release probability data did not fit a normal distribution (D'Agostino K 2 test (p<0.0001),  Wilk test (p<0.0001), Kolmogorov-Smirnov test (p<0.0001)). Beyond the heterogeneous P r 138 distribution, 9.7% of all release sites with apposed GluRIIA receptors displayed only 139 spontaneous fusion events, and another 14.6% of the AZ population was silent for both 140 spontaneous and evoked release during the recording period ( Figure 1B). In addition, the 141 majority of AZs rarely released a synaptic vesicle following an action potential, with a P r in the 142 range of 0.01 to 0.2. To functionally examine differences between high and low releasing sites, 143 we categorized all AZs with a release rate greater than 2 standard deviations above average as 144 "high P r ", and the remaining AZs that showed evoked release as "low P r ". Using these criteria, 145 65.8% of all AZs fell in the low P r category with an average release probability of 0.049 ± 0.004. 146 9.9% of AZs were classified as high P r sites, with an average release probability of 0.277 ± 0.015 147 ( Figure 1C), indicating high P r AZs displayed on average a 5.7-fold increased chance of vesicle 148 fusion following an action potential compared to low P r AZs. presence of GluRIIA-RFP allowed precise mapping of release sites between live and fixed tissue, 164 as well as correlation of high P r sites with SIM labeled BRP-positive AZs ( Figure 1D). Using an 165 automated detection algorithm in the Volocity 3D image analysis software, we were able to 166 identify all AZs labeled with BRP ( Figure 1D, right panel), and to resolve individual AZ clusters 167 that were not separated using conventional spinning disk microscopy where P r was determined. 168 The theoretical lateral resolution of the spinning disk confocal microscope for RFP labeled 169 structures is ~280 nm. Analysis of distances between different AZs by SIM indicated that 2.45 ± 170 0.4% (n = 9 NMJs from 9 animals) of all AZs were located close enough to each other (within 171 280 nm) such that they would not be resolvable in our live imaging. In contrast, 9.9% (n = 16 172 NMJs from 16 animals) of AZs were functionally classified as high P r sites. Therefore, the 173 9 majority of high releasing sites are not likely to be explained by release events occurring from 174 closely linked AZs. 175 To further analyze single versus closely spaced AZs, release maps were generated where 176 P r was color-coded to visualize the range of probabilities for all release sites. P r maps were then 177 compared to BRP-positive AZs identified by SIM. As shown in Figure 1D, the vast majority of 178 high P r sites were represented by a single BRP-positive AZ that was not further resolvable after 179 SIM imaging ( Figure 1D, red circles). These single BRP clusters at high P r sites were larger and 180 brighter than most other BRP positive puncta ( Figure 1E). The average total fluorescence of 181 single BRP puncta from high P r AZs (3.95 x 10 6 ± 2.67 x 10 5 , n = 24 AZs from 9 NMJs from 9 182 animals) was 1.7-fold greater than the fluorescence of randomly selected low P r BRP clusters 183 (2.33 x 10 6 ± 0.98 x 10 5 , n = 60 AZs from 9 NMJs from 9 animals, p<0.0001). To further 184 examine these large single BRP clusters and their release properties, larger clusters that could not 185 be resolved using conventional spinning disk microscopy were separately analyzed. All BRP 186 clusters larger than 280 nm were automatically detected and assigned their release probability 187 parameters measured during live imaging. We then determined whether these sites were 188 represented by single or multiple AZs using SIM microscopy. Clusters > 280 nm in diameter that 189 could be resolved to multiple BRP positive AZs after SIM imaging had a lower P r (0.10 ± 0.02, n 190 = 35 AZs from 5 NMJs from 5 animals) than those comprised of a single large BRP positive AZ 191 (0.19 ± 0.02, n = 42 AZs from 5 NMJs from 5 animals, Figure 1F). As such, high resolution SIM 192 microscopy confirms that most high P r sites correspond to single AZs with more intense BRP  Individual AZ P r is stable across imaging sessions 197 We next examined if the non-uniform distribution of P r across AZs was stable when no 198 plasticity changes were induced. If P r was highly dynamic at individual AZs over time, unique 199 local synaptic vesicle pools might be an important contributor to the distribution of variable 200 release properties. However, a more stable P r would argue for a specific factor(s) resident at 201 individual AZs. We were limited in our ability to examine P r continuously over time intervals 202 greater than 10-15 minutes due to bleaching of GCaMP6s from the high frequency capture rate. 203 Within this constraint, we conducted a 3-minute imaging session using 0.3 Hz stimulation to 204 generate an initial P r map, and then allowed the preparation to rest for 5 minutes without 205 stimulation or imaging. P r was then re-mapped in a final 3-minute imaging session using 0.3 Hz 206 stimulation. The activity level of individual AZs was very stable between the two sessions 207 (Figure 2A). This was especially evident for high P r sites, which sustained high levels of activity 208 during both imaging sessions. Plotting the release rate for all AZs revealed a strong correlation 209 for P r across the two imaging sessions (Pearson r = 0.77, R 2 =0.59, p<0.0001, n = 988 AZs from 210 8 NMJs from 7 animals, Figure 2B). These data suggest that release rate is a unique property of 211 each AZ and is stable over this time interval.

212
Heterogeneous release rates between AZs might be sensitive to the accumulation of 213 different vesicle populations with variable levels of fusogenicity. If so, a stronger stimulation 214 paradigm that is sufficient to drive vesicle cycling and intermixing would be expected to alter the 215 P r map. To test this, NMJ preparations were imaged during two low frequency 0.3 Hz 216 stimulation periods separated by a 5-minute 5 Hz stimulation session to induce robust synaptic 217 vesicle turnover and recycling ( Figure 2C). Release maps were not dramatically altered by 5 Hz 218 stimulation, with the overall correlation of P r similar to maps generated without stimulation 219 11 (Pearson r = 0.78, R 2 = 0.61, p<0.0001, n = 613 AZs from 6 NMJs from 6 animals, Figure 2D). heterogeneity, in addition to their established role in determining overall P r, then elimination of 230 Syt1 would likely disrupt this heterogeneity. We therefore expressed myrGCaMP6s with Mef2-231 GAL4 in syt1 null mutants. As observed electrophysiologically, quantal imaging in syt1 null 232 mutants revealed a dramatic reduction in evoked release, a shift from synchronous to highly 233 asynchronous fusion, and an increase in spontaneous release rates (Movie 2). To estimate AZ 234 heterogeneity in syt1 nulls, preparations were stimulated at 5 Hz and release events were 235 normalized to the number of stimuli ( Figure 3A). The average release rate per AZ per second in 236 syt1 nulls during 5 Hz stimulation was 0.03 ± 0.001 (n = 719 AZs from 7 NMJs from 6 animals, 237 Figure 3B). In contrast, spontaneous release rate per AZ in the absence of stimulation was 0.018 238 ± 0.001 per second in syt1 nulls (n = 719 AZs from 7 NMJs from 6 animals) compared to 0.011 239 ± 0.001 in controls (n = 559 AZs from 6 NMJs from 4 animals, p<0.0001, Figure 3B). All 240 visualized release events were mapped to specific AZs and representative P r heatmaps were 241 12 generated ( Figure 3A). Although release rate is dramatically reduced in syt1 nulls, AZs still 242 maintain the overall heterogeneity in P r distribution across AZs ( Figure 3C-E). Comparing the 243 distribution of AZ release rates for syt1 nulls and controls, release was proportionally decreased 244 across all AZs in syt1 ( Figure 3C). Frequency distribution analysis of AZs with normalized 245 release rates (from 0 to maximum release) confirmed that there was no significant change in the 246 heterogeneity of release between syt1 mutants and controls ( Figure 3D). Likewise, the 247 cumulative frequency distribution of normalized AZ P r was similar between syt1 mutants and 248 controls ( Figure 3E). Given that AZ release remains highly heterogeneous in the absence of Syt1, 249 these data suggest that variable distribution of key AZ components, rather than heterogeneity of 250 local synaptic vesicle proteins, is likely to control P r distribution across Drosophila NMJ AZs.   To gain confidence that the observed Cac-TdTomato intensity accurately reflects Cac 281 channel distribution, Cac channels transgenically tagged with GFP were also examined. dimmer and fully bleached within 7-10 minutes of imaging. Therefore, preparations were 292 stimulated at 1 Hz for shorter two minute imaging sessions to generate P r maps in myr-293 jRGECO1a expressing larvae ( Figure 4D). Using this approach, a strong correlation (Pearson r = 294 0.54, R 2 = 0.29, p<0.0001, n = 651 AZs from 7 NMJs from 7 animals) between AZ P r detected 295 by myr-jRGECO1a and Cac-GFP density was observed (representative experiment shown in 296 Figure 4E). Again, a weaker correlation was found between rates of spontaneous events and Cac-297 GFP density (Pearson r = 0.17, R 2 = 0.03, p<0.0001, n = 651 AZs from 6 NMJs from 6 animals).

298
Hence, P r for action-potential evoked fusion is strongly correlated with the local density of Cac 299 channels at individual AZs, regardless of which fluorophore is used to visualize Cac.

300
To determine the relative levels of Cac that defined low and high P r sites, the distribution 301 of Cac-GFP and BRP across the AZ population was examined using SIM microscopy. Similar to 302 the variable levels of BRP described earlier ( Figure 1D), variability in the distribution and mean  Cac-GFP fluorescence for these bright AZs (>2 standard deviations above average) was 2.1-fold 307 greater than that observed for the remaining sites (p<0.0001, Figure 4 -figure supplement 1C). 308 We next compared Cac-GFP fluorescence obtained for AZs that were functionally classified as 309 15 either low or high P r sites by quantal imaging using myr-jRGECO1a ( Figure 4F). The average 310 fluorescence of single Cac-GFP puncta from high P r AZs (normalized intensity = 0.6 ± 0.04, n = 311 38 AZs from 7 NMJs from 7 animals) was 2.09-fold greater than the average fluorescence of low 312 P r AZs (normalized intensity = 0.29 ± 0.01, n = 638 AZs from 7 NMJs from 7 animals, 313 p<0.0001). The P r of AZs classified on the basis of the levels of their Cac-GFP fluorescence was 314 also examined. The average P r for AZs displaying high Cac-GFP fluorescence (>2 standard 315 deviations above average) was 0.2 ± 0.016 (n = 7 NMJs from 7 animals) compared to 0.06 ±   indicating the abundance of Ca 2+ channels may not be the best proxy for AZ Ca 2+ entry. In 332 addition, a direct measure of Ca 2+ influx would be useful to bypass any potential unknown 333 effects on P r generated by expressing fluorescently tagged Cac. To generate an estimate of the 334 Ca 2+ influx that each AZ experiences independent of tagging Cac channels, Ca 2+ influx was  Figure 5A). However, stimulation at 10 Hz resulted in a robust increase in discrete punctated 347 fluorescence that remained confined to single AZs during stimulation ( Figure 5A).

348
To assay the ability of the sensor to detect local Ca 2+ influx, the stability of GCaMP BRP-349 visualized Ca 2+ signals during multiple rounds of 5-second 10 Hz stimulation was determined. confirmed that GCaMP BRP-short displayed a heterogeneous distribution of ΔF across AZs during 356 stimulation (n = 205 AZs from 6 NMJs from 3 animals, Figure 5C). 357 We next assayed if Ca 2+ influx detected by GCaMP BRP-short is correlated with Cac  However, there were some instances where specific AZs experienced a disproportionally low ΔF 367 of GCaMP BRP-short signal relative to their Cac-TdTomato intensity ( Figure 5D, arrows). This 368 observation suggests that Ca 2+ influx can be fine-tuned and regulated independently of Ca 2+ 369 channel abundance at certain AZs. As such, measuring both Ca 2+ channel density and Ca 2+ influx 370 is likely to provide a more accurate readout of how P r is controlled by local Ca 2+ concentrations 371 near the mouth of Ca 2+ channel clusters.

372
Using GCaMP BRP-short as a tool to estimate Ca 2+ influx at individual AZs, we analyzed 373 the correlation between GCaMP BRP-short ΔF induced by 10 Hz stimulation and release rate 374 visualized by myr-jRGECO1a during 1 Hz stimulation at single release sites ( Figure 6A, B). AZ 375 heatmaps for both P r and Ca 2+ influx fluorescence intensity were generated and compared across 376 the AZ population ( Figure 6A). AZs that experienced stronger Ca 2+ influx displayed the highest 377 18 P r during stimulation. Overall, there was a strong correlation between Ca 2+ influx and AZ P r 378 (Pearson r = 0.56, R 2 = 0.31, p<0.0001, n = 492 AZs from 6 NMJs from 6 animals, Figure 6B), 379 indicating the levels of Ca 2+ influx play a major role in determining whether a synaptic vesicle 380 undergoes fusion during an evoked response. In contrast, the frequency of spontaneous vesicle 381 fusion per AZ was only mildly correlated with the amount of Ca 2+ influx detected by GCaMP BRP-382 short (Pearson r = 0.23, R 2 = 0.07, n = 492 AZs from 6 NMJs from 6 animals, representative 383 experiment shown in Figure 6C), consistent with spontaneous release being largely independent 384 of extracellular Ca 2+ at this synapse. It is worth noting that although a strong correlation between 385 Ca 2+ influx and evoked P r was observed at most AZs, a minority population of AZs that 386 displayed robust Ca 2+ influx had very low P r ( Figure 6B). In summary, these data indicate that 387 Ca 2+ channel density and Ca 2+ influx are key factors regulating evoked release at individual AZs.

388
In addition, other factors can negatively influence P r at a small minority of AZs independent of 389 Ca 2+ influx. PSDs was observed ( Figure 7A). To determine if AZs that preferentially accumulate high levels 414 of GluRIIA correspond to high P r release sites, we mapped P r across the AZ population in 415 GluRIIA-RFP/GluRIIB-GFP expressing animals. Analysis of the P r map revealed a strong 416 positive correlation between GluRIIA-RFP and P r (Pearson r = 0.56, R 2 = 0.32, p<0.0001, n = 417 756 AZs from 8 NMJs from 4 animals, Figure 7B). In contrast, correlation with the levels of 418 GluRIIB-GFP was weaker (Pearson r = 0.32, R 2 = 0.1, p<0.0001, n = 756 AZs from 8 NMJs 419 from 4 animals, Figure 7C). A correlation between brighter GluRIIA PSD puncta and higher P r 420 sites was also observed in the analysis of syt1 mutants ( Figure 3A lines. After mapping P r in the larvae, preparations were fixed and stained with anti-BRP antisera.

426
As previously observed in animals lacking tagged glutamate receptors (Figure 1), a positive 427 correlation between AZ P r and BRP levels was observed (Pearson r = 0.44, R 2 = 0.2, p<0.0001, n 428 = 399 AZs from 6 NMJs from 4 animals, Figure 7D). In summary, these data indicate that 429 GluRIIA more strongly accumulates at PSDs apposing high P r AZs.

430
Beyond the preferential GluRIIA accumulation at high P r sites, we also observed a intensity of pixels along that line for each fluorophore was then analyzed. Average pixel 444 intensity revealed drastically distinct profiles for GluRIIB distribution between "bright" and 445 21 "dim" PSDs classified based on their GluRIIA intensity. GluRIIB was more evenly distributed 446 across the entire PSD at dim GluRIIA sites, but was segregated outward, forming a circular 447 donut-like ring around central GluRIIA puncta at bright GluRIIA sites ( Figure 7F). In addition, 448 presynaptic BRP intensity was more strongly correlated with postsynaptic GluRIIA levels 449 (Pearson r = 0.53, R 2 = 0.28, p<0.0001, n = 2496 AZs from 19 NMJs from 7 animals, Figure 7G)   Zito et al., 1999). Given the correlation between Ca 2+ channel 459 density, GluRIIA/GluRIIB segregation and high P r , we were interested in determining how AZs 460 acquire a specific P r during a larval developmental period that lasts 6-7 days. One model is that 461 certain AZs gain a higher P r status during development through a tagging or activity-dependent 462 mechanism that would lead to preferential accumulation of key AZ components compared to 463 their neighbors. Alternatively, high P r AZs might simply be more mature than their low P r 464 neighbors, having an earlier birthdate and a longer timeframe to accumulate AZ material. To 465 differentiate between these models for release heterogeneity, it would be desirable to follow P r 466 development from the embryonic through larval stages. However, this is not technically feasible instar larvae became more heterogeneous at the 3 rd instar stage (Figure 8 -figure supplement 2). 518 Indeed, histograms of normalized fluorescence intensity (relative intensity scaled from 0 to 1) 519 revealed that GluRIIA and GluRIIB were distributed relatively uniformly at 1 st instar larval 520 PSDs, with GluRIIA distribution becoming more skewed at later stages ( Figure 9A when it acquired the segregated GluRIIA/B pattern observed at high P r AZs was 3.20 ± 0.08 531 days (n = 41 AZs from 7 NMJs from 3 animals, Figure 9C). In a small subset of PSDs (5%), a 532 slightly faster accumulation of GluRIIA and the formation of GluRIIB peripheral rings was 533 observed, but never faster than 2 days (Figure 8 -figure supplemental 2). Given that the would fit with the 9.9% of high P r sites observed at the early 3 rd instar stage by GCaMP imaging.

537
The number of synapses present at the same NMJ from the 1 st instar through the early 3 rd instar 538 stage was quantified from live imaging experiments ( Figure 9D). AZ number doubled each day, 539 such that the average number of AZs found at the 1 st instar stage (day 1) represented 14.7 ± 1.4% 540 (n = 8 NMJs from 3 animals) of all AZs present by day 4 (3 days after initial imaging in 1 st 541 instars). This value is similar to the 10% population of high P r AZs observed during P r mapping.

542
Overall, these data support the hypothesis that AZ maturation is a key factor in regulating P r , 543 leading to increased accumulation of Ca 2+ channels and GluRIIA/GluRIIB segregation at high P r 544 sites compared to AZs that are newly formed (<2 days).

547
In the current study we used optical quantal imaging to examine the source of 548 heterogeneity in evoked P r across the AZ population at Drosophila NMJs. By combining quantal 549 imaging with SIM microscopy, we first confirmed that release heterogeneity was not caused by 550 summation of fusion events from multiple AZs. By monitoring release over 15 minute intervals, 551 we also observed that P r was a stable feature of each AZ. The Drosophila genome encodes a 552 single member of the N/P/Q-type Ca 2+ channel α1 subunit family (Cac) that is present at AZs Drosophila NMJ have a low P r . For the current study, the AZ pool was artificially segregated 577 into low and high release sites, with high releasing sites defined based on having a release rate 578 greater than two standard deviations above the mean. Given that birthdate is a key predictor of 579 glutamate receptor segregation, and by proxy P r , we expect the AZ pool to actually reflect a 580 continuum of P r values based on their developmental history. However, using the two standard 581 deviation criteria, 9.9% of AZs fell into the high P r category, with an average P r of 0.28 in  Beyond low and high P r sites, we found that 9.7% of the AZs analyzed displayed only 604 spontaneous release. We could detect no fusion events for either evoked or spontaneous release 605 28 for another 14.6% of AZs that were defined by a GluRIIA-positive PSD in live imaging. categories reflective of differences in AZ content, is unknown. For spontaneous-only sites, we 608 previously found that the ΔF/F avg quantal signal detected postsynaptically by GCaMP imaging 609 was similar to that observed at mixed mode AZs displaying both evoked and spontaneous events, 610 indicating that there is unlikely to be a dramatic difference in glutamate receptor density at these  Other factors we examined for regulating P r were differences in local synaptic vesicle 615 pools or synaptic vesicle protein content or state (for example, phosphorylation). P r was largely 616 unchanged with either 5-minute rest or 5 minute 5 Hz stimulation between imaging sessions.

617
Although P r may be more dynamic over longer intervals, the observation that developmental 618 maturation of glutamate receptor segregation occurs over ~ 3 days is consistent with P r being a 619 stable feature of the AZ over shorter time periods (hours to ~1 day). To further examine the role 620 of potential heterogeneity due to differences in synaptic vesicle proteins, we assayed confirmed that release was dramatically reduced and largely asynchronous in syt1 nulls, and that 625 mini frequency per AZ was increased. Although P r was reduced in syt1, P r distribution among 626 different AZs remained heterogeneous, suggesting that the AZ, rather than differential 627 distribution of Syt1, is critical. Although roles for other synaptic vesicle proteins in P r 628 29 heterogeneity cannot be excluded, the observation that prolonged 5 Hz stimulation, which would 629 be predicted to turn over the synaptic vesicle pool, does not change P r argues against this 630 hypothesis. Instead, these data support a model that differences in the abundance of presynaptic 631 Ca 2+ channels underlies heterogeneous AZ P r . 632 We considered several models for how AZs acquire this heterogeneous nature of P r 633 distribution. One possibility is that unique AZs gain high P r status through a mechanism that 634 would result in preferential accumulation of key AZ components compared to their neighbors.

635
Given that retrograde signaling from the muscle is known to drive synaptic development at ensures that the overall ratio of high to low P r sites remain relatively fixed at a low percentage, 662 depending on developmental stage and the rate of new AZ addition. 663 We did not test the correlation of P r with other AZ proteins besides Cac and BRP, but it allowing these AZs greater access to key components, and subsequently increasing their rate of 688 P r acquisition. In summary, our data indicate that Ca 2+ channel density and Ca 2+ influx at single 689 AZs is a key determinant for release heterogeneity, and that developmental AZ maturation is a 690 key factor in P r at the Drosophila NMJ.           imaging with myr-jRGECO1a. Student's t-test was used for statistical analysis (*** = p≤0.001).

1214
Error bars represent SEM.