Dissociated hippocampal neuron cultures were prepared from E18 rat embryos and plated with a density of 50k per well on poly-lysine coated glass coverslips in a 12-well plate. All procedures conformed to the guidelines established by the Institutional Animal Care and Use Committees at the University of Science and Technology of China (USTC) and the Chinese Academy of Sciences (CAS). The activity blockade was performed by the addition of tetrodotoxin (TTX, 0.5 μM) or 2,3-dihydroxy-6-nitro-7-sulfamoylbenzo(f)-quinoxaline (NBQX, 10 μM) at DIV10. To make sure the blockade was effective through the long period, we added another dose at DIV14. Whenever the cultures were fed, TTX or NBQX was included in the fresh medium.

Cells were fixed with 4% paraformaldehyde (PFA) and 4% sucrose in phosphate-buffered saline (PBS) (pH 7.4) for 10 min at room temperature (RT), followed by washing with 50 mM glycine in PBS. Cells were then permeabilized and blocked using 3% bovine serum albumin (BSA) or 5% donkey serum in PBS/Gly with 0.3% Triton X-100, followed by incubation with primary antibodies (3 h RT) and then secondary antibodies (1 h RT).

The following primary antibodies were used in this study: rabbit anti-RIM1/2 (1:500, Synaptic Systems 140213), mouse anti-PSD-95 (1:200, NeuroMab 75-028), and rabbit anti-GluA1 (1:500, Merck Millipore AB1504). For co-staining of GluA1 and RIM1/2, as both antibodies were from rabbits, staining was performed separately and the first primary antibody (GluA1) was blocked and converted with a goat anti-rabbit Fab fragment (JacksonImmuno 111-007-003) for 2 h at RT, and then recognized by secondary antibodies. The secondary antibodies were Alexa-647 conjugated donkey anti-Goat (JacksonImmuno 705-605-147), Alexa-647 conjugated donkey anti-rabbit (JacksonImmuno 711-605-152), and unconjugated donkey anti-mouse antibody (JacksonImmuno 715-005-151) or donkey anti-rabbit antibody (JacksonImmuno 711-005-152) labeled with Cy^TM3B (Mono NHS Ester, GE Healthcare 16889934).

Imaging was carried out on a Nikon ECLIPSE Ti2 inverted microscope equipped with a perfect focusing system and a 100 × /1.49 TIRF oil-immersion objective controlled using NIS-Elements AR 4.30.02 software. The typical incident power was ~40 mW (measured through the objective). Samples were imaged in a freshly made STORM imaging buffer containing: 50 mM Tris, 10 mM NaCl, 20% glucose, 56 μg/ml glucose oxidase (Sigma), 18 μg/ml catalase (Sigma), and 100 mM cysteamine (Sigma). To reduce background fluorescence while maximizing the depth of view, we adjusted the incident angle of the excitation beam to near but less than the critical angle and to achieve oblique illumination of the sample. Emission was collected using a CMOS camera (ORCA-Flash4.0, Hamamatsu) at a frame rate of 50 Hz and stored as images with a pixel size of 160 nm (with an 0.4× lens in the emission path). Z positions were determined by the ellipticity of the single peaks generated by a cylindrical lens (focal length 100 mm) with a precision of 40–50 nm. Total 50k images were collected for each channel. TetraSpeck beads (100 nm; Invitrogen) deposited on a coverslip were localized for generating the calibration curves. In our system, the average deviation of the bead localizations after t correction was 10–15 nm in x/y and 40–50 nm in z. Localization detection, calibration, and drift correction were performed using the NIS-Elements AR analysis 4.40.00 software. Localization coordinates were then rendered into images (pixel size of 5 nm) using a two-dimensional Gaussian kernel (σ = 20 nm) with custom routines in MATLAB (Mathworks).

Whole-cell patch clamp was carried out with patch-clamp amplifiers (MultiClamp 700B, Axon Instruments) at RT. The data were acquired and analyzed using custom Igor Pro (WaveMetrics) programs. Intracellular pipette solution (pH 7.3) contained 136.5 mM potassium gluconate, 17.5 mM KCl, 9 mM NaCl, 1 mM MgCl₂, 10 mM HEPES, 0.2 mM EGTA, 2 mM ATP-Mg, and 0.3 mM GTP-Na. External bath solution (pH 7.3) contained 150 mM NaCl, 3 mM KCl, 3 mM CaCl₂, 2 mM MgCl₂, 10 mM HEPES, 5 mM glucose, 1 μM TTX, and 20 μM bicuculline.

Detailed analysis on synaptic clusters was formed using custom routines in MATLAB as described previously (Tang et al., 2016; Chen et al., 2020). A synaptic cluster was identified in a 2D scatter plot of all localizations. By rotating a 3D scatterplot of localizations of a selected potential synaptic cluster pair, we evaluated the data quality and selected only those with clear edges (e.g., no nearby third cluster which may indicate more than two synaptic clusters in close proximity) for further analysis. To define a synaptic cluster on the random background, the nearest neighbor distances (NNDs) between localizations were calculated, and the mean + 2 × SD of NND was used as a cutoff to divide the localizations into subclusters. All localizations outside the primary subclusters were considered to be a background and were not used in further analysis. The border of the synaptic cluster was defined using the alpha-shape of the set of 3D localizations with α = 150 nm.

Nanoclusters were detected based on local density which was defined as the number of molecules within a radius of 2.5 times the standard median nearest neighbor distance (MdNND) for the calculation of the density of the synaptic cluster (Chen et al., 2020). The threshold of local density for nanocluster detection was defined as mean (LD0) + 4 × SD (LD0), where LD0 is the local density of a randomized cluster with the same overall density as the synaptic cluster. All localizations with a local density over this threshold were considered within nanoclusters. These localizations were then divided into subclusters with a “top-down” divisive strategy with a minimal peak-to-peak distance of 80 nm. Finally, to be counted as nanoclusters, those subclusters had to meet a set of criteria, including the number of localizations ≥8, which was derived empirically based on tests on our measured and simulated synapses to reduce the false positives arising from repeated localizations of the same molecule.

Data analysis was performed using the Origin software. Data are reported as average ± SEM values, and statistical significance was evaluated using one-way ANOVA, Kolmogorov–Smirnov test for cumulative curves, and z-test for proportions. Asterisks above brackets in data bar graphs indicate the level of statistical significance (^*p < 0.05; ^**p < 0.01; and ^***p < 0.001). Detailed results of statistical analysis are listed in Supplementary Table 1.

Previous ultrastructural and fluorescence imaging studies have shown a strong correlation between the sizes of membrane compartments and the amount of synaptic proteins on presynaptic and postsynaptic sides. In this study, we reexamined this with STORM super-resolution microscopy on presynaptic RIM1/2, the key AZ protein for action potential-dependent vesicle fusion, and postsynaptic AMPAR subunit GluA1 and scaffold PSD-95. We performed the staining and imaging as pairs of RIM1/2-GluA1 and RIM1/2-PSD-95, at four developmental stages in cultured hippocampal neurons: 6–8 days in vitro (marked as DIV7 in the following text) when synaptic contacts are newly formed, DIV10 (±1), DIV14 (±1), and DIV18 (±1) when synapses are thought mature.

At DIV7, synaptic contacts could be identified as sites of colocalization between RIM1/2 and GluA1 (or PSD-95), while there were many protein clusters without its counter partner in the vicinity (Figure 1A). Synaptic clusters were filtered, and the borders were defined with an alpha-shape as previously described (Tang et al., 2016; Chen et al., 2020). When the cultures grew older, the volume of synaptic clusters expanded gradually (Figures 1B–G), with a 4–5-fold increase in cluster sizes of all three proteins from DIV7 to DIV18 (p < 0.001 for all groups, one-way ANOVA; for details, refer to Supplementary Table 1), similar to previous observations (Chanda et al., 2017). More importantly, we found that this expansion in synaptic size was proportional between the presynaptic and postsynaptic compartments (Supplementary Figure 1). Actually, even at the same developmental state, the correlations between the volumes were strong for each protein (Figures 1H,I; Supplementary Figure 1). Moreover, a similar strong correlation was also found in the localization numbers of synaptic proteins which represent the amount of synaptic proteins (Supplementary Figure 2). These results suggest a synchronous growth of AZ and PSD during synapse maturation, and no obvious evidence indicating the remodeling in one compartment preceded that in the other.

We then set out to quantify the subsynaptic distribution of these proteins at different developmental stages. To visualize the protein pattern within the synapse, local density for each localization was calculated and color-coded in the distribution map (Figures 2A–D). Comparing with that in mature synapses, GluA1 at DIV7-10 distributed more homogeneously within the synapse boundaries. To quantify the heterogeneity within the synaptic cluster, we employed a normalized autocorrelation function (g_a) (Veatch et al., 2012; Tang et al., 2016). The autocorrelation showed a significant heterogeneity over a range of 0–100 nm for GluA1, RIM1/2, and PSD-95 at DIV14-18, while the amplitudes were significantly smaller and the range over which the heterogeneity was above the chance level was narrower for more immature synapses (Figures 2E–G). These differences were further confirmed by the properties of high-density nanoclusters identified with an automated algorithm based on local density (Chen et al., 2020). Immature synapses had a smaller number of nanoclusters, lower localization density within nanoclusters, and smaller nanocluster volume (Figures 2H–J). The gradually increased heterogeneity in these proteins during the maturation of synapses suggests that the well-organized subsynaptic distribution of both presynaptic transmitter release and postsynaptic receptors is a hallmark for mature synapses and may be essential for their functions.

To compare the organizations on presynaptic and postsynaptic compartments, we averaged the g_a for radius from 0 to 50 nm (ga¯) and used it as a simplified index of subsynaptic heterogeneity. We found a significant correlation between ga¯ of presynaptic RIM1/2 and postsynaptic GluA1 or PSD-95 (p < 0.001 for both pairs, Figures 3A,B). This result suggests that the subsynaptic organizations of presynaptic and postsynaptic components are evolved synchronously during development, similar to the size of both compartments. When we examined the ga¯ correlations in each developmental stage, we found that the correlation was only significant in mature synapses (Figures 3C,D). Since more mature synapses had a larger volume, we wonder whether stronger heterogeneity is an intrinsic property of larger synapses. We, therefore, plotted ga¯ of RIM1/2 against the cluster volume for each synapse and found that there was in fact a positive correlation (Figure 3E) for both mature and immature synapses. However, synapses of DIV18 showed a general larger ga¯ than those of DIV7. To eliminate the effect of cluster volume on ga¯, we picked up only those large synapses with a cluster volume of 1–4 × 10⁷ nm³ and compared the subsynaptic heterogeneity. The mature synapses showed a significantly larger ga¯ than the immature for both RIM1/2 and GluA1, but the heterogeneity of PSD-95 was similar (Figure 3F). These results suggest the subsynaptic heterogeneity of RIM1/2 or GluA1 is not an intrinsic property of the protein clusters, and there may be other active processes underlying the formation and evolvement of subsynaptic organizations during synaptic maturation.

We previously showed that the presynaptic RIM1/2 nanoscale organization represents the preferential sites for transmitter release, and its alignment with postsynaptic AMPAR densities could efficiently modulate the synaptic strength (Tang et al., 2016). Therefore, it would be crucial to examine the nanoscale alignment between nanoclusters across the cleft during development.

Representative examples showed that synapses at different developmental stages all have a certain degree of alignment between the nanoclusters of RIM1/2 and GluA1 (Figures 4A–D). To further quantify this, we performed the enrichment analysis by calculating the averaged local density of GluA1 along with different distances from the projected center of defined RIM1/2 nanoclusters (Figure 4E). All groups showed a trend of elevated GluA1 density at distances close to the RIM1/2 nanocluster centers, but more mature synapses exhibited a more significant enrichment. To simplify the comparison, we averaged the normalized GluA1 density within the distance of 50 nm to define an enrichment index (EI) and compared the measured indices with that of simulated synapses with the positioning of GluA1 nanoclusters randomized within the GluA1 cluster. Synapses in cultures of DIV14 and 18 showed an EI significantly above the randomized simulations (Figure 4F; p < 0.05, one-way ANOVA using the Tukey's multiple comparison test, for details, refer to Supplementary Table 1), while the EIs of immature synapses at DIV7 and 10 were not significantly different from the chance level.

By performing multiple simulations on the same synapse, we could get a threshold of EI (usually around 1.25–1.3) above which the tested RIM1/2 nanocluster was defined as a nanocluster with significant GluA1 enrichment, or simply “enriched nanocluster,” with a 95% CI (Tang et al., 2016). There were 33.3% of RIM1/2 nanoclusters above this threshold for synapses at DIV7, and the percentage increased gradually with development to 51.9% at DIV18 (Figure 4G). Similar results were found for the alignment between RIM1/2 and PSD-95 (Figures 4H,I), but one minor difference is that there were slightly more enriched nanoclusters at DIV10 compared with the GluA1-RIM1/2 pair (Figure 4J; 47.7% RIM1/2 and 48.1% PSD-95, vs. 10.0% RIM1/2, and 33.3% GluA1), suggesting the alignment between RIM1/2 and PSD-95 may form prior to that between RIM1/2 and GluA1.

Previous studies found that the functional correlation between presynaptic and postsynaptic compartments requires neuronal activity to develop (Kay et al., 2011). We then sought to determine whether the correlation between nano-organizations was also dependent on synaptic activity. The hippocampal cultures were incubated in TTX (0.5 μM) or NBQX (10 μM) to block action potentials or AMPA receptors, respectively, at DIV10-18 during which the transsynaptic alignment and organizations show most dramatic changes (Figures 2, 4). This blockade lasted much longer than the treatment commonly used for homeostatic plasticity studies, and therefore, more processes may have taken place. In fact, though the frequency of miniature EPSCs showed a similar trend of increase after TTX incubation, the effects of both treatments were much smaller and insignificant than that of the previous 1–2 days' treatment (Han and Stevens, 2009) (Supplementary Figure 4). We found that AMPAR blockade significantly increased the synaptic cluster volume of both RIM1/2 and PSD-95 at DIV18 (Figures 5A,B,E,F) as well as the nanocluster number of PSD-95 (Figures 5C,G). Synapses in cultures treated with TTX showed a similar trend but the differences did not reach a significant level. These changes and trends were largely consistent with previous observations under similar treatment (MacGillavry et al., 2013; Glebov et al., 2017; Venkatesan et al., 2020).

However, the normalized density within nanoclusters showed no change with the treatment (Supplementary Table 1). Consistently, the autocorrelation profiles of either protein were largely overlapped (Figures 5D,H), except that PSD-95 showed slightly larger values of heterogeneity. To test whether this results from the larger cluster volume (Figures 3E, 5I), we similarly plotted ga¯ of PSD-95 against the cluster volume for each synapse and found that ga¯ of PSD-95 was, in fact, positively correlated with the cluster volume (Figure 5J). To rule out the effect of cluster volume, we selected only the synaptic clusters with a volume of (1–3) × 10⁷ nm³ and found that ga¯ of either PSD-95 or RIM1/2 showed no significant difference between control and AMPAR blockade (Figure 5K), suggesting that the general heterogeneity of synaptic proteins was largely unchanged by activity blockade.

We further examined whether the formation of trans-synaptic nano-alignment depends on synaptic activity. We found that neither TTX nor NBQX treatment had a significant effect on the averaged enrichment between RIM1/2 and PSD-95 (Figures 6A–C). The lack of changes in general heterogeneity of protein organizations and the averaged nanoscale trans-synaptic enrichments is well-consistent with the electrophysiological result that there was no significant change in either the frequency or amplitude of mEPSCs (Supplementary Figure 4). However, the percentage of nanoclusters (NCs) that were enriched with their counter partners showed a reducing trend on activity blockade, with the difference significant for the proportion of enriched PSD-95 nanoclusters between NBQX and control groups (Figure 6D; p = 0.025, z-test). Therefore, in spite of the similar average overall enrichment between RIM1/2 and PSD-95, there might be some nanoscale reorganizations that concentrate the proteins within a subset of nanocolumns, especially for the RIM1/2-to-PSD-95 enrichment.