Neurexin-3 subsynaptic densities are spatially distinct from Neurexin-1 and essential for excitatory synapse nanoscale organization in the hippocampus

Abstract

Proteins critical for synaptic transmission are non-uniformly distributed and assembled into regions of high density called subsynaptic densities (SSDs) that transsynaptically align in nanocolumns. Neurexin-1 and neurexin-3 are essential presynaptic adhesion molecules that non-redundantly control NMDAR- and AMPAR-mediated synaptic transmission, respectively, via transsynaptic interactions with distinct postsynaptic ligands. Despite their functional relevance, fundamental questions regarding the nanoscale properties of individual neurexins, their influence on the subsynaptic organization of excitatory synapses and the mechanisms controlling how individual neurexins engage in precise transsynaptic interactions are unknown. Using Double Helix 3D dSTORM and neurexin mouse models, we identify neurexin-3 as a critical presynaptic adhesion molecule that regulates excitatory synapse nano-organization in hippocampus. Furthermore, endogenous neurexin-1 and neurexin-3 form discrete and non-overlapping SSDs that are enriched opposite their postsynaptic ligands. Thus, the nanoscale organization of neurexin-1 and neurexin-3 may explain how individual neurexins signal in parallel to govern different synaptic properties.

Fig. 1. Conditional deletion of Nrxn3 impairs nano-organization of GluA1 and RIM1 at excitatory synapses in the hippocampus.

a Representative 3D dSTORM images of presynaptic active zone maker RIM1 (cyan) and postsynaptic AMPA receptor subunit GluA1 (red) in Control and Nrxn3 knockout neurons (left). The schema used to quantify 3D dSTORM data (right). Nrxn3 KO reduces GluA1 compartment volume p = 0.0148 (b) and number of SSDs per synapse p = 0.0045 n = 122 control and 149 Nrxn3 KO synapses (c) but not SSD volume p = 0.8995 n = 115 and n = 135 synapses (d). Nrxn3 KO reduces the relative density of GluA1 in SSDs p = 0.0098 n = 103 synapses and n = 123 synapses (e). Significance: Mann–Whitney test (two-tailed). f–i Same quantification as (b–e) except for presynaptic RIM1. RIM1 compartment volume, p < 0.0001; SSDs per synapse, p = 0.5016 n = 191 control and n = 185 Nrxn3 KO synapses; SSD volume p < 0.0001 n = 163 control and n = 150 Nrxn3 KO synapses; relative density in SSDs p = 0.7732 n = 169 control and n = 160 Nrxn3 KO synapses. Significance: Mann–Whitney test (two-tailed). j Nrxn3 KO reduces transsynaptic enrichment of GluA1 relative to RIM1 SSDs, p = 0.0150 n = 87 control and n = 97 Nrxn3 KO SSDs and a reduction in the GluA1 transsynaptic enrichment index at radii ≤60 nm, p = 0.0471 n = 87 control and n = 97 Nrxn3 KO SSDs. Significance: 2-way repeated measures ANOVA main effect of Nrxn3 KO and Student’s t test (two-tailed) respectively. k, l Nrxn3 KO reduces GluA1 to RIM1 cross-correlation over 10–100 nm radius, p = 0.0001 n = 174 control and n = 188 Nrxn3 KO synapses; (k) and average cross-correlation at radii ≤50 nm, p = 0.0307 (l). Significance: 2-way repeated measures ANOVA main effect of radius x Nrxn3 KO; Mann–Whitney test (two-tailed). Data from three independent experiments. Number of synapses indicated on the graph unless stated in the legend. Bar graphs and line graphs: average ± SEM. Violin plots: median ± upper and lower quartiles. *p < 0.05; **p < 0.01. Source data are provided as a Source Data file.

Fig. 2. Ablation of Nrxn3 impairs the nano-organization of PSD-95 in hippocampus.

a Reconstructed 3D dSTORM images of presynaptic RIM1 and postsynaptic PSD-95 in control (top) and Nrxn3 KO hippocampal neurons (bottom). Nrxn3 KO reduces PSD-95 compartment volume p < 0.0001 (b) and SSDs per synapse p = 0.0005 n = 149 control and n = 164 Nrxn3 KO synapses (c) without changing SSD volume p = 0.8609 n = 123 control and n = 136 Nrxn3 KO synapses (d). Significance: Mann–Whitney test (two-tailed). e–g Same as in (b–d) but for RIM1. RIM1 compartment volume, p < 0.0001; SSDs per synapse, p = 0.1097 n = 143 control and n = 148 Nrxn3 KO synapses; SSD volume, p = 0.0017 n = 115 control and 126 Nrxn3 KO synapses. Significance: Mann–Whitney test (two-tailed). h Reconstructed 3D dSTORM images of presynaptic RIM1 and postsynaptic surface GluA1 in control (top) and Nrxn3 KO cortical neurons (bottom). Nrxn3 KO does not alter nanoscale GluA1 properties including compartment volume p = 0.4802 (i), SSDs per synapse p = 0.6117 n = 111 control and n = 97 Nrxn3 KO synapses (j), or SSD volume p = 0.8620 n = 85 control and n = 72 Nrxn3 KO synapses (k). Significance: Mann–Whitney test (two-tailed). Nrxn3 KO results in no changes in nanoscale RIM1 properties including compartment volume p = 0.5918 (l), SSDs per synapse p = 0.5806 n = 122 control and n = 115 Nrxn3 KO synapses (m), or SSD volume p = 0.3018 n = 100 control and n = 96 Nrxn3 KO synapses (n). Significance: Mann–Whitney test (two-tailed). Data from three independent experiments. Number of synapses indicated on the graph unless stated in the legend. Bar graphs and line graphs: average ± SEM. Violin plots: median ± upper and lower quartiles. **p < 0.01; ***p < 0.001; ****p < 0.0001. Source data are provided as a Source Data file.

Fig. 3. Computational modeling of Nrxn3 conditional knockout hippocampal synapses shows deficits in AMPAR transmission.

a Schematic of pyramidal neuron and synapse. b Diagram of modeled synapse. c Schematic of AMPAR kinetics modeled in simulation. d Diagram of synapse indicating regions of AMPARs included in the simulation. e Isolation of the AMPAR current from three regions of the modeled synapse from left to right: Ring 1 which corresponds to GluA1 SSDs, Ring 2 the peri-SSD region, and Ring 3 which is peripheral to SSDs. f Computational simulation of potential effects on EPSC in Nrxn3 KO hippocampal neurons. From left to right, there are cartoons (above) and simulated EPSC traces (below) from WT (black), decreased GluA1 SSDs (red), altered SSD and compartment volumes (red), reduced density of GluA1 (red), and the cumulative effects after Nrxn3 KO (red). The gray traces in the left panel are 160 runs with release sites randomly distributed through the active zone, the black trace is the mean value. For parameters used, see Table 1. Schematic drawing in (b) is adapted from Han et al., 2022 (17). Source data are provided as a Source Data file.

Fig. 4. Homozygous V5-Nrxn3 knock-in mice show no deficits in survival, synaptic protein expression, or synaptic function and are suitable for Nrxn3 localization.

a Schematic of the V5-tag knock-in site. b Sanger Sequencing validation of F1 V5-Nrxn3  mice. c Survival rates of offspring from heterozygous V5-Nrxn3 mice (expected Mendelian ratio = red outline) p = 0.2999 χ² test, n = 71 mice. d–f Immunoprecipitation of V5-Nrxn3 from whole brain lysate (d). Representative Western blots of excitatory synaptic proteins from wildtype littermate (left) and V5-Nrxn3 (right) mice (e). Western blot quantification of synaptic proteins in V5-Nrxn3 mice (f) α-Nrxns, p = 0.5456; Homer1, p = 0.9076; PSD-95, p = 0.3289; GluA1, p = 0.4543; GluA2, p = 0.5144; or RIM1/2, p = 0.5853. One sample t test (two-tailed), n = 3 independent animals for each condition. M: molecular weight ladder. g–j Functional analysis of V5-Nrxn3 mice. Recording schematic (g). Representative mEPSCs from V5-Nrxn3 and wild-type littermate controls (h). mEPSC frequency (i); (p = 0.7493). mEPSC amplitude (j); (p = 0.1275). Wildtype: 24 cells from 4 independent animals; V5-Nrxn3: 25 cells from 4 independent animals. Significance: unpaired t test (two-tailed). k Representative confocal images (left) of surface anti-V5 labeling of wildtype and V5-Nrxn3 neurons. Quantification (right) of surface expression of V5-Nrxn3; n = 15 Wild-type and n = 14 V5-Nrxn3 cells from 3 independent experiments unpaired t-test (two-tailed). l Representative image (left) and summary graphs of V5-Nrxn3 co-localization with vGluT1 (right). Number of neurons and independent experiments are indicated in the figure. Bar graphs indicate mean ± SEM. ****p < 0.0001. Source data are provided as a Source Data file.

Fig. 5. Endogenous neurexin-3 is present at the majority of excitatory synapses and localizes near the synaptic nanocolumn.

a Representative 3D dSTORM of endogenous V5-Nrxn3 (cyan) and Homer1 (red). b V5-Nrxn3 Homer1 synaptic abundance (left) and number of SSDs (right). Distribution of Nrxn3 localization distance (c) and SSD distance (d) to the centroid of Homer1. Mean and 95% confidence interval of the Gaussian fitted distribution indicated on the graph. e Representative 3D dSTORM of V5-Nrxn3 (cyan) and PSD-95 (red). f Same as (b) for PSD-95+ synapses. g Distributions of synaptic of V5-Nrxn3 (left) and PSD-95 (right). Blue spheres identify V5-Nrxn3 SSDs opposite a PSD-95 of varying density (heatmap). Arrow identifies V5-Nrxn3 SSD that opposes a region of high PSD-95 density. h Enrichment analysis of PSD-95 density opposite Nrxn3 SSDs. n = 154 synapses. i Nearest neighbor distance from Nrxn3 SSDs to the closest PSD-95 SSD is left-shifted compared to randomized data. Violin plot (inset) of nearest neighbor distances, p = 0.0004; Mann–Whitney test (two-tailed). n = 303 experimental and randomized SSDs. j Same as in H, except for nearest neighbor distances of PSD-95 SSDs to nearest Nrxn3 SSD compared to randomized data. Violin plot (inset) of nearest neighbor distances, p < 0.0001; Mann–Whitney test (two-tailed). n = 465 experimental and randomized SSDs. Data from three independent experiments. Number of synapses indicated on the graph unless stated in the legend. Bar graphs and line graphs: average ± SEM. ***p < 0.001; ****p < 0.0001. Source data are provided as a Source Data file.

Fig. 6. Endogenous neurexin-1 organizes near the periphery of synapses and synaptic nanocolumns.

a Representative 3D dSTORM field view of surface HA-Nrxn1 (cyan) and Homer1 (red). b Nrxn1 Homer1 synaptic abundance (left) and number of SSDs (right). Distribution of Nrxn1 localization distance (c) and SSD distance (d) to the centroid of Homer1. Mean and 95% confidence interval of the Poisson fitted distribution are indicated on the graph. (e) Representative 3D dSTORM of Nrxn1 (cyan) and PSD-95 (red). f Same as (b) for PSD-95+ synapses. g Distributions of synaptic Nrxn1 and PSD-95. Nrxn1 SSDs (green spheres) opposing relative PSD-95 density (heatmap). h Average PSD-95 enrichment opposite Nrxn1 SSDs at increasing distances from the translated Nrxn1 SSD center. n = 111 synapses. i Nearest neighbor distances of Nrxn1 SSDs to PSD-95 SSDs are unchanged from randomized data. Violin plot (inset) of nearest neighbor distance, p = 0.0807; Mann–Whitney test (two-tailed). N = 225 experimental and randomized SSDs. j Same as in (h) but for nearest neighbor distances of PSD-95 SSDs to Nrxn1 SSD is modestly left-shifted relative to a randomized SSD location. Violin plot (inset) of median nearest neighbor distance is random, p = 0.0208; Mann–Whitney test (two-tailed). n = 330 experimental and randomized SSDs. Data from three independent experiments. Number of synapses indicated on the graph unless stated in the legend. Bar graphs and line graphs: average ± SEM. *p < 0.05. Source data are provided as a Source Data file.

Fig. 7. Endogenous LRRTM2 and GluD1 assemble into SSDs near sites of neurexin-1 and neurexin-3 enrichment.

a Representative 3D dSTORM reconstruction of LRRTM2 (cyan) and Homer1 (red). b LRRTM2 Homer1 synaptic abundance n = 232 synapses from 3 independent experiments (left) and number of SSDs n = 151 synapses from 3 independent experiments (right). LRRTM2 localization distance (c) and SSD distance (d) to Homer1 centroid. The mean and 95% confidence interval of Poisson fitted distribution are indicated on the graph. e Representative 3D dSTORM reconstruction of LRRTM2 (cyan) and Nrxn1 (left) or Nrxn3 (right; red) at Homer1+ synapse (green) in an enface view. f Nrxn1 is farther from LRRTM2 SSDs than Nrxn3, p = 0.0135; Mann–Whitney test (two-tailed). g Representative 3D dSTORM reconstruction of GluD1 (cyan) and Homer1 (red) in an enface view. h GluD1 Homer1 synaptic abundance n = 196 synapses from 3 independent experiments (left) and number of SSDs n = 134 synapses from 3 independent experiments (right). i–j GluD1 localization distance (g) and GluD1 SSD distance (h) from the Homer1 centroid. The mean and 95% confidence interval of the Poisson fitted distribution indicated on the graph. k Representative 3D dSTORM reconstruction of GluD1 (cyan) and Nrxn1 (left) or Nrxn3 (right; red) at Homer1+ synapse (green) in an enface view. l Nrxn1 is closer to GluD1 SSDs than Nrxn1, p = 0.0314; Mann–Whitney test (two-tailed). Number of synapses and independent experiments are indicated in the figure. Bar graphs and line graphs: average ± SEM. Violin plots: median ± upper and lower quartiles. *p < 0.05. Source data are provided as a Source Data file.

Fig. 8. Nrxn1 and Nrxn3 form discrete SSDs at Homer1+ hippocampal synapses.

a Breeding strategy to create HA-Nrxn1^KI/KI::V5-Nrxn3^KI/KI mice. Representative 3D dSTORM field view of Nrxn1 (b); (cyan) or Nrxn3 (c); (cyan) and RIM1 (red) with a widefield Homer1 overlay (green). The nearest neighbor distance is shorter from RIM1 SSDs to Nrxn3 SSDs than to Nrxn1 SSDs, p < 0.0001; (d) and from Nrxn3 SSDs to RIM1 SSDs, p < 0.0001; (e). Significance: Mann–Whitney test (two-tailed). f Representative 3D dSTORM field view of Nrxn3 (red) and Nrxn1 (cyan) with a wide field Homer1 overlay (green). g Stacked bar graph of excitatory synapses with ≥5 Nrxn1 and/or Nrxn3 localizations. h Schematic of the overlap mask method. SSD overlap is determined by quantifying the volume overlap of Nrxn1 and Nrxn3 SSDs masks. Stacked bar graph of the SSD volume overlap of Nrxn3 with Nrxn1 (I) and Nrxn1 with Nrxn3 (J). n = 212 synapses. (k) Histogram of the percent overlap of Nrxn1 and Nrxn3 SSDs using the overlapping mask method. n = 212 synapses. l Representative scatter plots depicting Nrxn1 SSDs (green) and Nrxn3 SSDs (blue) showing minimal (<10%, left), moderate (10–50% middle), and high (>50%, right) overlap of SSDs. Data from three independent experiments. Number of synapses indicated on the graph unless stated in the legend. Violin plots: median ± upper and lower quartiles. ****p < 0.0001. Source data are provided as a Source Data file.

Fig. 9. Summary model: Nrxn3 controls excitatory synapse nano-organization in hippocampus and localizes discretely from Nrxn1.

We find that Nrxn3 controls excitatory synapse nano-organization in hippocampus and that Nrxn1 and Nrxn3 localize in discrete, non-overlapping SSDs that are preferentially opposite GluD1 and LRRTM2 SSDs respectively. Nrxn3 deletion results in decreased AMPAR SSDs per synapse as well as a decrease in the relative number of AMPARs in SSDs.