Abstract
α-Neurexins are essential and highly expressed presynaptic cell-adhesion molecules that are frequently linked to neuropsychiatric and neurodevelopmental disorders. Despite their importance, how the elaborate extracellular sequences of α-neurexins contribute to synapse function is poorly understood. We recently characterized the presynaptic gain-of-function phenotype caused by a missense mutation in an evolutionarily conserved extracellular sequence of neurexin-3α (A687T) that we identified in a patient diagnosed with profound intellectual disability and epilepsy. The striking A687T gain-of-function mutation on neurexin-3α prompted us to systematically test using mutants whether the presynaptic gain-of-function phenotype is a consequence of the addition of side-chain bulk (i.e., A687V) or polar/hydrophilic properties (i.e., A687S). We used multidisciplinary approaches in mixed-sex primary hippocampal cultures to assess the impact of the neurexin-3αA687 residue on synapse morphology, function and ligand binding. Unexpectedly, neither A687V nor A687S recapitulated the neurexin-3α A687T phenotype. Instead, distinct from A687T, molecular replacement with A687S significantly enhanced postsynaptic properties exclusively at excitatory synapses and selectively increased binding to neuroligin-1 and neuroligin-3 without changing binding to neuroligin-2 or LRRTM2. Importantly, we provide the first experimental evidence supporting the notion that the position A687 of neurexin-3α and the N-terminal sequences of neuroligins may contribute to the stability of α-neurexin–neuroligin-1 trans-synaptic interactions and that these interactions may specifically regulate the postsynaptic strength of excitatory synapses.
- alpha neurexins
- excitatory synaptic transmission
- LRRTM
- neurexins
- neuroligins
- neurosciencepsychiatric disease
Significance Statement
Although neurexins were discovered over 30 years ago, our understanding of how the complex extracellular sequences unique to α-neurexins participate in synapse function remains incomplete. We leveraged a previously studied human missense mutation, located in a conserved extracellular region of neurexin-3α and linked to profound intellectual disability and epilepsy, to systematically assess the tolerance of neurexin-3α function to mutations within this region. Using molecular replacement, we assessed how single amino acid substitutions in this extracellular region alter synapse morphology, presynaptic calcium dynamics, and synaptic transmission. We reveal that multiple neurexin ligands unexpectedly use this region to modulate trans-synaptic binding and that different amino acid substitutions in place of the disease mutation result in dramatically different changes to synaptic transmission.
Introduction
Neurexins (Nrxns) are evolutionarily conserved and essential presynaptic cell-adhesion molecules that play critical roles in synapse function (Südhof, 2017). However, despite 30 years of intense scrutiny, an understanding of the precise synaptic functions that individual Nrxns control remains incomplete (Sudhof, 2017). Three Nrxn genes (Nrxn1-3) produce long alpha (α-Nrxn), short beta (β-Nrxn), and an even shorter gamma (γ-Nrxn, Nrxn1 only) isoforms (Sterky et al., 2017; Gomez et al., 2021). While α- and β-Nrxns share identical transmembrane and intracellular sequences, their extracellular sequences differ in length and complexity (Reissner et al., 2013; Sudhof, 2017). α-Nrxn extracellular sequences contain six laminin-neurexin-sex hormone (LNS1–6) domains interspersed with three EGF-like repeats (EGFA–C) and can be alternatively spliced at up to six conserved splice sites (SS1–6; Schreiner et al., 2014; Gomez et al., 2021). In contrast, β-Nrxn extracellular sequences are less elaborate and only possess LNS6 and two splice sites (SS4 and SS5). Of the six splice sites, SS4, located in LNS6, has received the most attention. Exclusion (SS4−) or inclusion (SS4+) of the alternative SS4 exon influences the binding of Nrxns to almost all postsynaptic binding partners (Gomez et al., 2021).
β-Nrxns have received a disproportionate amount of scientific scrutiny even though α-Nrxns are more abundantly expressed in almost all brain regions and cell types studied and mutations in the extracellular sequences of α-Nrxns are more commonly linked to disease (Restrepo et al., 2019; Uchigashima et al., 2019). Nrxns engage in trans-synaptic interactions with a growing number of postsynaptic ligands (Gomez et al., 2021; Kim et al., 2021). At excitatory synapses, presynaptic Nrxns interact with the postsynaptic ligands neuroligin-1 and neuroligin-3 (NL1 and NL3) and leucine-rich repeat transmembrane proteins 1–4 (LRRTM1–4), which compete for binding to the same surface in the LNS6 domain (Ko et al., 2009; Siddiqui et al., 2010; Chen et al., 2011; Paatero et al., 2016). Although untested, crystal structures of the Nrxn1α suggest that α-Nrxns may function as extracellular scaffolds to simultaneously interact with multiple extracellular partners or utilize individual domains to cooperatively stabilize ligand binding (Chen et al., 2011; Miller et al., 2011; Reissner et al., 2013; Liu et al., 2018). However, the considerable length (∼1,470 aa) and complex domain organization of α-Nrxns have made studying how α-Nrxn extracellular sequences contribute to synapse function difficult.
We previously identified a Nrxn3α compound heterozygous patient diagnosed with epilepsy and profound intellectual disability who possesses a paternal truncation mutation and a maternal rare missense mutation (A683T; NP_001317124.1). The missense mutation is located in a highly conserved interface between the EGF-B-like domain and the LNS4 domain (Fig. 1A). Interestingly, other missense mutations have been identified in this region including a Nrxn3α mutation linked to autism spectrum disorder (ASD) and intellectual disability (T678I; Fig. 1A, yellow; Landrum et al., 2018) and a Nrxn1α mutation [c.2148 G>A (NC_000002.10); E715K (NM_004801.2)] located immediately upstream of the analogous A683T mutation in Nrxn3α (Fig. 1A, magenta; Yan et al., 2008), suggesting that this region may be important for synapse function. The A683T orthologous missense mutation in mouse Nrxn3α, Nrxn3αA687T (A687T), produced a presynaptic gain-of-function exclusively at excitatory synapses due to selectively enhanced interactions with LRRTM2 (Restrepo et al., 2019). Importantly, Nrxn3αA687T SS4−, but not SS4+, was required to manifest the gain-of-function phenotype. To gain greater mechanistic insight, we systematically characterized the amino acid side-chain properties of Nrxn3αA687 that are required to regulate synapse function.
A687T and A687S substitutions increase Nrxn3α-mediated postsynaptic recruitment of PSD-95 and gephyrin, forming more heterologous synapses in vitro. A, Diagram indicating the location of the A687 substitutions in the EGF-B-LNS4 linker of Nrxn3α. A mutation in this linker in Nrxn1α are highlighted in magenta and imposed onto the Nrxn3α primary amino acid sequence. The conserved alanine (bold) is indicated. The human A687 (purple), the patient mutation (A687T, green), the substitution to mimic side-chain bulk (A687V, orange), and the substitution to mimic polar/hydrophilic side-chain properties (A687S, blue) are indicated. A Nrxn3α mutation (T678I) linked with intellectual disability is highlighted in yellow and a mutation in human Nrxn1α (E715K) associated with ASD is highlighted in magenta. B, Diagrams depicting the different A687 amino acid side-chain substitutions in Nrxn3α. C, D, Schematic depicting the experimental workflow of the heterologous synapse formation assay (C) and a representation of the recruitment of postsynaptic specializations to HEK cells expressing Nrxn3α variants (D). E, Representative images of immunolabeled HEK293T expressing GFP only, WT, or A687 substitutions and DIV14 primary hippocampal neuron cocultures. F, G, Summary graphs of the intensities of the excitatory postsynaptic scaffold, PSD-95 (F), and of the inhibitory postsynaptic scaffold, gephyrin (G), reveal that A687S selectively increased the amount of PSD-95 recruited without changes in gephyrin recruitment. Data represented are mean ± SEM; mean values are quantified from the average of the number of cells in all experiments. Numbers in bar graphs describe cells from each individual experiment. PSD-95 intensity: GFP: 0.1723 ± 0.033 A.U.; WT: 0.7925 ± 0.095 A.U.; p = 0.0007; A687T: 0.64 ± 0.090 A.U.; p = 0.8700; A687V: 1.354 ± 0.07607 A.U.; p = 0.7423; A687S: 1.35 ± 0.209 A.U.; p = 0.0039; one-way ANOVA, multiple comparisons relative to WT. Gephyrin intensity: GFP: 0.64 ± 0.012 A.U.; WT: 1.178 ± 0.2188 A.U.; p = 0.0003; A687T: 1.04 ± 0.1585 A.U.; p = 0.9859; A687V: 0.9086 ± 0.1585 A.U.; p = 0.8349; A687S: 1.28 ± 0.2692 A.U.; p = 0.9651; one-way ANOVA, multiple comparisons relative to WT. **p < 0.01. Number of cells analyzed from the number of independent experiments indicated in each bar graph.
Here, we generated two “pseudo-mimetic” A687T mutant Nrxn3α: (1) Nrxn3αA687V (A687V, gold) to recapitulate the added side-chain bulk of the mutant threonine residue and (2) Nrxn3αA687S (A687S, blue) to investigate the impact of altered polar-mediated interactions (Fig. 1A,B). The A687V substitution did not dramatically alter synapse morphology, function, or binding to postsynaptic ligands. Surprisingly, while A687S also led to a gain-of-function phenotype at excitatory synapses, distinct from A687T, A687S enhanced postsynaptic properties and modulated binding in trans with a different population of postsynaptic ligands.
Materials and Methods
Mouse generation and husbandry
All mouse work was approved by the University of Colorado Anschutz Medical Campus animal use committees. For all experiments, wild-type mice on a C57BL/6 background from both sexes were utilized. Mice were housed in the Anschutz Medical Campus animal care facility and were regulated on a 12 h light/dark cycle, at 21–23°C, 35% humidity. Mice were housed in ventilated cages with same-sex littermates in groups of two–five with food and water intake ad libitum. All procedures were conducted in accordance with guidelines approved by the Administrative Panel on Laboratory Animal Care at University of Colorado, Anschutz School of Medicine, accredited by Association for Assessment and Accreditation of Laboratory Animal Care International (AAALAC; 00235).
Analysis of hippocampal neuronal cultures
Hippocampi from postnatal day (P) 0 C57BL/6J mice were cultured on Matrigel-coated glass coverslips (Carolina Biological, 633009) as previously described (Aoto et al., 2013). Briefly, hippocampi were isolated in Hanks balanced salt solution (HBSS; Sigma-Aldrich, catalog #H2387), digested in 10 U/ml papain (Worthington Biochemical, catalog #LS003126) in HBSS at 37°C for 20 min. Digested hippocampi were washed with HBSS and triturated in plating media (MEM) with 10% FBS (Sigma-Aldrich), 0.5% glucose (Invitrogen), 2 mM ʟ-glutamine (GeminiBio, catalog #400-106-100), 0.02% NaHCO3 (Sigma-Aldrich, catalog #S5761), and 0.1 mg/ml transferrin (GeminiBio, catalog #800-130P-100) before being plated. Twenty-four hours after plating, 70% of the plating media was exchanged with neuronal growth media (MEM) with 2 mM ʟ-glutamine, 2.38 mM NaHCO3, 0.1 mg/ml transferrin, 0.4% glucose, 5% FBS, and 1:50 Gem21 (GeminiBio, catalog #400-160-010). On day in vitro 4 (DIV4), 70% of the growth media was replaced with growth media supplemented with 4 μM cytosine β-D-arabinofuranoside (AraC; Sigma, catalog #C1768) to arrest glial cell growth. Lentivirus transduction was performed on DIV3–4, infected with lentiviruses at DIV4, and analyzed at DIV13–15.
Virus production
Lentiviruses were produced in HEK293T cells as previously reported (Aoto et al., 2013). Briefly, HEK293T cells were transiently transfected using the calcium phosphate transfection method with pMDL gag/pol, pVSV-g, pRSV-REV, and a transfer vector encoding shRNA against Nrxn3α and/or Nrxn3αA687 molecular replacement cDNAs. Lentiviruses were constructed utilizing the human synapsin prompter to induce expression. shRNA-resistant cDNAs were generated as rescue constructs packaged in separate lentiviral envelopes.
Plasmids
The A687 mutations were introduced into mouse Nrxn3α cDNA that contained splice inserts at SS1 and SS3 and lacked splice inserts at SS2, SS4, and SS5 (Nrxn3α SS4−) as previously reported (Aoto et al., 2013), by Gibson Assembly into FSW: shRNA F: TGCACTGGGCTGGTGATTGATTCAAGAGATCAATCACCAGCCC AGTGCTTTTTTC R: TCGAGAAAAAAGCACTGGGCTGGTGATTGATCTCTTGAATCAATCACCAGCCCAG TGCA; Nrxn3α: F: TTTAAACTTAAGCTTGGTACCGAGCTCGGATCCGCCACCATGAGCTTTACCCTCCACTC R: GAGCGGCCGCCACTGTGCTGGATATCTGCAGAATTCTTACACATAATACTCCTTGTCCTT; A687T; F: ACCTGCGAAAGAGAGACATCTATCCTGAGCT R: AGCTCAGGATAGATGTCTCTCTTTCGCAGGT; A687V F: GGGGAGAACCTGCGAAAGAGAGGTGTCTATCCTGAGCTATGACGGCAGCA R: TGCCGTC ATAGCTCAGGATAGACACCTCTCTTTCGCAGGTTCTCCCCCAG; A687S F: GGGGAGAACCTGCGAAAG AGAGAGCTCTATCCTGAGCTATGACGGCAGCA R: TGCCGTCATAGCTCAGGATAGAGCTCTCTCTTTC GCAGGTTCTCCCCCAG. These A687 cDNAs were subsequently subcloned via Gibson Assembly into pcDNA3.1(+) (Invitrogen) digested with EcoRI and BamHI using the following primers: F: TTTAAACTTAAGCTTGGTACCGAGCTCGGATCCGCCACCATGAGCTTTACCCTCCACTC, R: GAGCGGCCGCCACTGTGCTGGATATCTGCAGAATTCTTACACATAATACTCCTTGTCCTT. Mouse NL cDNAs all contained splice-site A and were inserted into pcDNA3.1(+) (Invitrogen). The NL1 cDNA lacked splice-site B (NL1 AΔB). pCMV-IgC LRRTM2 was previously described (Ko et al., 2009).
Heterologous synapse formation assay
Mouse hippocampal neurons were isolated and cultured at P0. On DIV10, HEK293T cells are transfected with GFP alone or GFP with pcDNA3.1 Nrxn3α-A687 SS4−. On DIV11, control or neurexin-expressing HEK293T cells are cocultured with the hippocampal culture. On DIV14 cocultures are fixed in 4% PFA with 4% sucrose, washed 3× 5′ with PBS, permeabilized with PBS + 0.2% Triton X-100, and fixed for 1 h in blocking buffer (2% normal goat serum + 0.2% Triton X-100 in PBS). Cocultures were incubated with primary antibodies to detect for postsynaptic scaffold proteins PSD-95 (1:1,000/Millipore, catalog #CP35) and gephyrin (1:500/Synaptic Systems, catalog #147111). Cocultures were then incubated in conjugated secondary antibodies CF568 (1:500/Biotium, catalog #20802) and Alexa Fluor 647 (Jackson ImmunoResearch, catalog #715-175-151), and mounted onto glass slides using Fluoromount-G (SouthernBiotech, catalog #0100-01). Slides were imaged utilizing spinning disk confocal microscopy at room temperature using the Zeiss Axio Observer microscope on a 63× oil-immersion objective utilizing 488, 594, and 647 laser excitation and a CSU-XI spinning disk confocal scan head (Yokogawa) paired with an Evolve 512 EM-CCD camera manipulated by SlideBook software. Numerous coverslips per condition were used to acquire ∼10–15 cells, per imaging session. The laser power and photomultiplier were unchanged per imaging session. Acquired images underwent threshold by intensity to omit background noise and analyzed using a MATLAB toolbox previously reported (Aoto et al., 2013).
RT-qPCR of Nrxn3α mRNA in primary hippocampal cultures
β-Actin was used as an internal control to quantitatively measure Nrxn3α mRNA abundance. mRNA was isolated from DIV14 primary hippocampal culture utilizing Quick-RNA Microprep mRNA isolation kit (Zymo Research, catalog #R1050). qScript xLT 1-Step RT-qPCR, Low Rox ToughMix Kit (Quantabio, catalog #95133-100) was used to induce reverse transcription. Mixture was added to Clear TempPlate 384-well Full-Skirt PCR, Single Notch polypropylene plates (USA Scientific, catalog #1,438–4,700). Plate was run and analyzed on CFX384 Real-Time System RT-PCR instrument (Bio-Rad). β-actin: F: GACTCATCGTACTCCTGCTTG, R: GATTACTGCTCTGGCTCCTAG, Probe: CTGGCCTCACTGTCCACCTTCC. Nrxn3α: F: TCCAGTTCAAGACCACTTCAG, R: GTCACTGTTGCCTTTGATCAC, Probe: TGGGACCATTGCCGA GATCAAACA.
Immunocytochemistry
Live neuron surface immunolabeling of HA-tagged Nrxn3α WT or A687, GluA1 subunit of α-amino-3-hydroxy-5-methyl-4-isoxazole propionic acid receptor (AMPAR), and γ2 subunit of γ-aminobutyric acid receptor (GABAAR) was adapted from previously reported protocols (Aoto et al., 2013). Briefly, live primary hippocampal cultures were prewashed with room temperature HEPES-based aCSF (Hb-aCSF) and immunolabeled with either mouse anti-HA (1:250; BioLegend, catalog #901523), anti-rabbit GluA1 (1:250; Millipore, catalog #MAB2263), or anti-guinea pig γ2 (1:250; Synaptic Systems, catalog #224004) diluted in Hb-aCSF for 20 min at 37°C. Live cells were washed 2 × 2′ with Hb-aCSF at 4°C before fixation with 4% PFA + 4% sucrose. Fixed cells were then incubated for 1 h in either donkey anti-mouse (1:500/Biotium, catalog #20802), donkey anti-rabbit (20795), or donkey anti-guinea pig (20838) CF568 secondary antibodies diluted in blocking buffer (2% normal goat serum + 0.2% Triton X-100 in PBS).
Fixed neuron immunolabeling was conducted as previously reported (Aoto et al., 2013). Briefly, DIV14 primary hippocampal cultures are fixed with 4% PFA + 4% sucrose, washed 3× 5′ with PBS, permeabilized with PBS + 0.2% Triton X-100, and fixed for 1 h in blocking buffer (2% normal goat serum + 0.2% Triton X-100 in PBS). Cocultures were incubated with primary antibodies to detect for presynaptic protein markers, vGlut1 (1:1,000/Synaptic Systems, catalog #135304) and vGAT (1:500/Synaptic Systems, catalog #131002) or postsynaptic scaffold proteins PSD-95 (1:1,000/Millipore) and gephyrin (1:200/Synaptic Systems). Fixed neurons were then incubated in conjugated secondary antibodies CF568 (1:200/Biotium) and Alexa Fluor 647 (1:200/Jackson ImmunoResearch, catalog #127706) and mounted onto glass slides using Fluoromount-G. Slides were imaged utilizing the same microscope and conditions mentioned above. Numerous coverslips per condition were used to acquire ∼15–25 cells, per imaging session. Acquired images were thresholded by intensity to omit background noise and analyzed using a MATLAB toolbox previously reported (Aoto et al., 2013).
Primary hippocampal culture electrophysiology
Primary hippocampal neurons were coinfected with a lentivirus encoding eGFP and a control or Nrxn3α shRNA and a separate lentivirus encoding Nrxn3αA687 variants. GFP-positive pyramidal neurons were identified by the presence of dendritic spines and subjected to whole-cell voltage-clamp electrophysiology. Glass pipettes were pulled to 3–5 mΩ, and series resistance was continuously monitored during recording. Room temperature aCSF (in mM: 126 NaCl, 2.5 KCl, 1 NaH2PO4, 26.2 NaHCO3, 2.5 CaCl2, 1.3 MgSO4•7H2O, 11 D-glucose, ∼290 mOsm) was constantly superfused over cultures. The internal solution included the following (in mM): 115 Cs-methanesulfonate, 15 CsCl, 8 NaCl, 0.3 EGTA, 10 HEPES, 4 Mg-ATP, 0.3 Na-GTP, 10 TEA-Cl, 10 Na2-phospocreatine, 1 MgCl2, ∼290 mOsm. Pyramidal cells were held at −70 mV to record mEPSCs (0.5 µm tetrodotoxin and 100 µm picrotoxin) and mIPSCs (0.5 µm tetrodotoxin, 50 µm D-AP5, and 0.5 µm NBQX). Miniature events were picked by hand and analyzed in Molecular Devices Clampfit 10 employing template matching at a threshold of 5 pA. The electrophysiological properties were measured from all five conditions in a single experiment.
Calcium imaging of primary hippocampal cultures
Primary hippocampal neurons were infected with lentivirus encoding mCherry alone or Nrxn3α shRNA that also expressed mCherry along with Nrxn3α molecular replacement lentivirus and a lentivirus encoding GCaMP5G. Neurons were constantly superfused with room temperature aCSF (see above, Primary hippocampal culture electrophysiology), and evoked calcium signals were acquired after one 50 Hz electrical stimulus using a 40× water dipping objective (Olympus) and a Hamamatsu ORCA-Flash 4.0 V3 digital camera. The recorded GCaMP5G signal was processed using SynActJ (Synaptic Activity in ImageJ; Schmied et al., 2021). Briefly, SynActJ automatically determines segmentation parameters to identify GCaMP5G-expressing presynaptic terminals. These segmentation parameters were then validated by a blinded experimenter before automatic analysis of the change in fluorescent signal before and after the 50 Hz stimulus. The fluorescent signal was averaged from all automatically detected presynaptic GCaMP5G-positive boutons in each video. To determine Ca2+ transient kinetics, the average fluorescence intensity was averaged in each frame of a 100-frame video, which was then averaged between experiments. We confirmed that the automatically detected boutons terminated on mCherry-positive dendrites. The automatic thresholding differed between experiments but stayed consistent within experiments.
HEK cell aggregation assay
Aggregation was conducted as previously reported (Restrepo et al., 2020). Briefly one population of HEK293T cells were transfected with either GFP alone or together with LRRTM2, neuroligin-1, neuroligin-2, neuroligin-3, or neuroligin-4. Another population of HEK293T cells are transfected with either mCherry alone or together with Nrxn3αA687 SS4− mutants. Both populations are mixed in a 1:1 ratio and incubated for 30–60 min in DMEM containing the following (in mM): 50 HEPES-NaOH, pH7.4, 10% FBS, 10 MgCl2, and 10 CaCl2. Incubated HEK293T cells were imaged by epifluorescence microscopy utilizing a wide-field Zeiss Axiovert 200M with an Olympus 10× air objective (NA, 0.55; WD, 26 mm). Aggregation index was analyzed utilizing Fiji (ImageJ) scripts quantifying the percentage overlap of red (mCherry) on green (GFP) puncta.
Production of recombinant Ig-fusion proteins in HEK293T cells
For recombinant protein expression, HEK293T cells were cultured to ∼70–80% confluence in 150 mm cell culture dishes and then transfected with 30 μg of cDNA corresponding to the Ig protein(s) of interest using PEI. Medium containing soluble Ig-fusion proteins was harvested 4 d after transfection and cleared by centrifugation at 2,000 rpm. The supernatant composition was then adjusted by adding 20 mM HEPES-NaOH, pH 7.4, 1 mM EDTA, and a protease inhibitor mixture (Thermo Scientific) and then incubated for 2 h with protein A-Sepharose (GE HealthCare), which binds the Fc domain of human IgG Fc. Unbound proteins were removed by washing beads with PBS, and bead-bound proteins were eluted with 0.1 M glycine, pH 2.2.
Cell-surface binding assays
Ig-fusion proteins of Nlgn1, Nlgn1.2, Nlgn2.1, LRRTM2, Nrxn3 WT, Nrxn3 A687S, and IgC alone (Control) were produced in HEK293T cells, as previously described (Han et al., 2020). Soluble Ig-fused proteins were purified using protein A-Sepharose beads (GE HealthCare). Bound proteins were then eluted with 0.1 M glycine, pH 2.2, and immediately neutralized with 1 M Tris-HCl, pH 8.0. Transfected HEK293T cells expressing the indicated plasmids were incubated with 10 μg/ml Ig-fused proteins for 2 h at 37°C. After 2 h, cells were washed twice with PBS, fixed with 3.7% formaldehyde for 10 min at 4°C, and blocked with 3% horse serum/0.1% bovine serum albumin (BSA; crystalline grade) in PBS for 15 min at room temperature. Surface-expressed protein was then detected by staining with mouse anti-HA antibody at room temperature. After 90 min, cells were washed twice with PBS and incubated with Cy3-conjugated anti-human IgG and FITC-conjugated anti-mouse secondary antibodies for 1 h at room temperature. Images were acquired using a confocal microscope (LSM800; Carl Zeiss). Commercially available antibodies, mouse monoclonal anti-HA antibodies (clone 16B12; BioLegend; catalog #901501; RRID: AB_2565006), and Cy3-donkey anti-human IgG antibodies (Jackson ImmunoResearch; catalog #709-165-149; RRID: AB_2340535) were used.
Molecular modeling
All molecular modeling was performed using ChimeraX (Meng et al., 2023) using the following PDB identifiers: 3QCW, 3BIW, and 5Z8Y. To predict the impact of A687T, A687V, and A687S, we identified the corresponding residue in the Nrxn1α structure (A723) and used the function swapaa to introduce a substitution at A723. We used a Dunbrack backbone-dependent rotamer library, curated from experimentally solved protein structures, to identify the side-chain rotamer with the highest probability of occurrence. Clashes and contacts were determined in ChimeraX using the following default criteria: clashes, Van der Waals radii (VDW) overlap cutoff of 0.6 Å and an H-bond allowance of 0.4 Å; contacts, VDW overlap cutoff of −0.4 Å and an H-bond allowance of 0.0. Hydrogen bonds were predicted using the following criteria based on a survey of small molecule crystal structures (Mills and Dean, 1996): distance tolerance of 0.4 Å and an angle tolerance of 20°. Distances between side chains were calculated by selecting side-chain atoms and using the distances function.
Quantification and statistical analysis
Statistical analysis for all experiments was performed in GraphPad Prism 7 and GraphPad Prism 9. Data distribution was assessed for normality using the Shapiro–Wilk test. Further statistical analysis included Student's unpaired t test and one-way ANOVAs. Multiple comparisons were also used where applicable. Cumulative properties statistics were analyzed by Kolmogorov–Smirnov tests. Every bar graph represented are the means from at least three independent experiments. Reported data values are presented as mean ± SEM.
Results
Nrxn3αA687S enhances the recruitment of postsynaptic specializations in the heterologous synapse formation assay
To build on our previous work, which characterized a human mutation in Nrxn3αA687T (A687T), we sought to interrogate how the mutant threonine residue alters Nrxn3α function. We generated two Nrxn3α variants to recapitulate the A687T-mediated phenotypes and test the impact of the bulk (A687V) versus the polar and hydrophilic properties (A687S) of the threonine side-chain (Fig. 1B). We first asked whether substitutions at position A687 alter the ability of Nrxn3α to participate in trans interactions to recruit excitatory and inhibitory postsynaptic specializations. Although not inherently synaptogenic in vivo, when coated on microspheres or expressed on non-neuronal cells, α- and β-Nrxns potently induce the formation of postsynaptic specializations with cocultured neurons through trans interactions with Nrxn ligands expressed on contacting dendrites (Graf et al., 2004; Biederer and Scheiffele, 2007; Gokce and Sudhof, 2013). Our previous work demonstrated that Nrxn3 SS4−, but not SS4+, was functionally relevant in primary hippocampal neurons; thus, we used the Nrxn3 SS4− splice variant in all experiments (Aoto et al., 2013; Restrepo et al., 2019).
We transfected HEK293T cells with GFP alone or cotransfected cells with GFP and Nrxn3αWT (WT), A687T, A687S, or A687V. We cocultured the transfected HEK293T cells with primary hippocampal neurons on DIV11 (Fig. 1C). After 3 d of coculture (DIV14), we performed indirect immunofluorescence labeling with antibodies against PSD-95 and gephyrin, which are postsynaptic scaffold proteins that demarcate excitatory and inhibitory synapses, respectively (Fig. 1D). HEK293T cells expressing GFP alone did not recruit postsynaptic scaffolds. Consistent with previous studies that utilized Nrxn3 in this assay, we found that all Nrxn3αA687 variants recruited excitatory and inhibitory postsynaptic scaffolds to HEK cells (Gokce and Sudhof, 2013). The robust recruitment of PSD-95 and gephyrin by Nrxn3α onto HEK cells prevented the measurement of individual scaffolding molecule clusters. Instead, we quantified the fluorescence integrated intensity of PSD-95 or gephyrin as an indirect measurement of the abundance of each recruited protein (Fig. 1E–G). Unexpectedly, although the expression of the A687T mutation in neurons produced a robust gain-of-function phenotype (Restrepo et al., 2019), when expressed in HEK cells, the activity of Nrxn3α A687T SS4− in this assay was identical to Nrxn3α WT (Fig. 1E–G). The substitution of alanine to valine did not alter gephyrin or PSD-95 recruitment and was nearly identical to WT (Fig. 1E–G). However, distinct from the A687T mutation, the A687S mutation significantly increased the integrated intensity of PSD-95 but not gephyrin (Fig. 1F,G). Taken together, A687S was the only variant tested that was capable of enhancing the formation of excitatory hemisynapses, which suggests that the introduction of a serine residue at A687 may either enable stronger interactions with LRRTM2 than A687T or enhance Nrxn3α's engagement in trans interactions with a different population of postsynaptic adhesion molecules. While the heterologous synapse formation assay indicates that Nrxn3αA687S engages in more robust trans-synaptic interactions than WT or A687T, this assay does not provide insight into the functional relevance of the amino acid side chain. Thus, we next aimed to build upon our findings to assess whether the morphological and functional phenotypes reported for the A687T mutation in primary hippocampal neurons are phenocopied by the A687S substitution.
Molecular replacement with A687 substitutions does not alter Nrxn3α surface expression in hippocampal neurons
To assess if A687V or A687S impacts the morphology and function of hippocampal excitatory synapses like A687T, we employed a previously used molecular replacement approach which consists of simultaneous lentivirus-mediated short hairpin (shRNA) knockdown of endogenous Nrxn3α and replacement with shRNA-resistant WT or Nrxn3αA687 variants via separate lentiviral vectors (Fig. 2A; Restrepo et al., 2019). Lentivirus-mediated transduction reliably results in the infection of >95% of all neurons, which is essential to interrogate the role of presynaptic molecules in vitro (Aoto et al., 2013). Consistent with our previous work, infection with a dual-promoter lentivirus encoding Nrxn3α shRNA and enhanced GFP (eGFP) reduced endogenous Nrxn3α mRNA in primary neurons by ∼75% relative to control infection (Restrepo et al., 2019). Molecular replacement with shRNA-resistant WT or Nrxn3αA687 variants resulted in an approximately eightfold increase of transcript abundance relative to mock-infected neurons (Fig. 2B). We next asked if the A687 substitutions altered the surface abundance of Nrxn3α. We performed molecular replacement in dissociated hippocampal neurons on DIV3 with HA-tagged Nrxn3α WT or Nrxn3αA687 variants followed by live anti-HA surface labeling on DIV14 (Fig. 2D). Relative to WT, substitutions at A687 did not alter the surface expression of Nrxn3α (Fig. 2E). Together, we observed no differences in Nrxn3α mRNA levels or surface expression following molecular replacement with A687, A687T, A687V, or A687S. Thus, we next wanted to interrogate the influence of these substitutions on synaptic morphology and function.
The A687 substitutions do not alter Nrxn3α surface expression in hippocampal cultured neurons. A, Schematic illustrating the molecular replacement approach (top) and representative image of lentivirus infected primary hippocampal cultures (bottom). B, qRT-PCR of Nrxn3α mRNA after shRNA-mediated knockdown of endogenous Nrxn3α after lentiviral infection (left), normalized mRNA levels after molecular replacement approach with A687 substitutions (right). Nrxn3α relative to shRNA knockdown: WT: 9.149 ± 0.6306, n = 3, p = <0.0001; A687T: 10.44 ± 1.908, n = 3, p = <0.0001; A687V: 10.99 ± 2.590, n = 4, p = <0.0001; A687S: 9.793 ± 0.4335, n = 3, p = <0.0001; one-way ANOVA, multiple comparisons. C, Diagram highlighting the locations of the A687 substitutions and the corresponding codons, and the relative location of the HA-epitope tag. D, Representative images of DIV14 hippocampal neurons live surfaced immunolabeled HA-tagged isoforms of Mock, WT, and A687 substitutions. E, Quantification of the surface levels of HA-Nrxn3α WT and A687 variants following molecular replacement. Intensity, GFP: 0.0 ± 0.0, n = 44/3, p = <0.0001; WT: 60.41 ± 3.187 A.U., n = 58/3; A687T: 68.75 ± 3.395 A.U., n = 65/3, p = 0.3341; A687V: 63.82 ± 3.281 A.U., n = 79/3, p = 0.9301; A687S: 57.46 ± 2.859 A.U., n = 73/3, p = 0.9608; density per μm2, GFP: 0.000 ± 0.000, p = <0.0001; WT: 4.750 ± 0.3142; A687T: 4.659 ± 0.2757, p = 0.9989; A687V: 4.994 ± 0.2100, p = 0.9475; A687S: 4.314 ± 0.20181, p = 0.6607; GFP: 0.000 ± 0.000 μm2, p = <0.0001; WT: 1.124 ± 0.1114 μm2; A687T: 1.425 ± 0.1224 μm2, p = 0.2228; A687V: 1.285 ± 0.0932 μm2, p = 0.7677; A687S: 1.301 ± 0.0884 μm2, p = 0.7127; one-way ANOVA, multiple comparisons. Number of independent experiments identified in bar graphs in B. Number of cells analyzed from the number of independent experiments indicated in each bar graph in E.
Molecular replacement with Nrxn3αA687S selectively increases postsynaptic PSD-95 puncta size and AMPAR surface expression in primary hippocampal neurons
In the heterologous synapse assays, the HEK293T cells that expressed A687S, but not A687T, recruited excitatory postsynaptic scaffolding molecules more robustly than WT (Fig. 1E,F). We next asked if A687S, when expressed in neurons, exerts a similar or distinct effect as A687T on excitatory synapse morphology in dissociated hippocampal cultures. We previously found that molecular replacement with A687T selectively increased the puncta size of the presynaptic excitatory synaptic vesicle marker vGluT1 without altering the morphological properties of postsynaptic PSD-95 or the morphological properties at inhibitory synapses (Restrepo et al., 2019). To assess the impact of A687S on excitatory synapse morphology, we infected primary cultures with the following controls: (1) a lentivirus encoding GFP and (2) a dual-promoter lentivirus encoding Nrxn3α shRNA and eGFP. Using this dual-promoter lentivirus, we performed molecular replacement to knock down endogenous Nrxn3α and coinfected neurons with lentivirus encoding Nrxn3α WT or A687 variants on DIV3. We performed immunocytochemistry on DIV14 with antibodies for vGluT1 and PSD-95. We selectively imaged pyramidal neurons, which we identified by the presence of dendritic spines. The knockdown of Nrxn3α did not alter synapse morphology, which is consistent with previous reports (Aoto et al., 2015; Restrepo et al., 2019). Relative to control neurons, molecular replacement with WT or A687V did not alter the cluster density, intensity, or size of presynaptic vGluT1 or postsynaptic PSD-95 clusters (Fig. 3A–C). Consistent with our previous data, the A687T mutation selectively increased the size of vGluT1 puncta by ∼30% relative to GFP without altering vGluT1 puncta intensity or density or the properties of PSD-95 clusters (Fig. 3A,B; Restrepo et al., 2019). In contrast, we observed no presynaptic morphological changes following molecular replacement with A687S. Instead, A687S significantly increased the density of PSD-95 clusters by ∼28% and the size of PSD-95 clusters by ∼38% without altering cluster intensity (Fig. 3A,C). While molecular replacement with both A687T and A687S led to enhanced morphological properties of excitatory synapses; A687T specifically enhanced presynaptic vGluT1 morphology, whereas molecular replacement with A687S selectively increased the properties of postsynaptic PSD-95 clusters. These findings indicate that the A687T and the A687S substitutions may differentially alter the synaptic functions encoded by Nrxn3α at excitatory synapses.
Molecular replacement with Nrxn3αA687S selectively increases the size and numbers of PSD-95 puncta and surface AMPAR puncta in vitro. A, Representative images of DIV14 primary hippocampal neurons immunolabeled for vGluT1 and PSD-95. B, C, Summary graphs of presynaptic vGluT1 (B) and postsynaptic PSD-95 (C) intensity (left), density (middle), and size (right) reveals that A687T selectively increases the size of vGluT1, and A687S significantly increases PSD-95 size and density. vGlut1 intensity: GFP: 89.47 ± 2.079 A.U. n = 102/4; shRNA: 90.54 ± 2.656 A.U., n = 105/4, p = 0.9998; WT: 97.88 ± 3.294 A.U., n = 74/4, p = 0.3276; A687T: 95.82 ± 2.419 A.U., n = 106/4, p = 0.6507; A687V: 95.21 ± 2.719 A.U., n = 105/4, p = 0.6507; A687S: 86.62 ± 2.814 A.U., n = 96/4, p = 0.9773; vGlut1 density per 10 μm: GFP: 4.028 ± 0.1683; shRNA: 4.314 ± 0.1910, p = 0.8744; WT: 3.818 ± 0.1780, p = 0.9760; A687T: 4.355 ± 0.1563, p = 0.8833; A687V: 4.080 ± 0.1987, p > 0.9999; A687S: 4.512 ± 0.2110, p = 0.4335; vGlut1 size: GFP: 0.8980 ± 0.06289 μm2; shRNA: 1.193 ± 0.07321 μm2, p = 0.2745; WT: 1.194 ± 0.08530 μm2, p = 0.3747; A687T: 2.065 ± 0.1666 μm2; A687V: 1.170 ± 0.05989 μm2, p = 0.3650; A687S: 1.277 ± 0.1005 μm2, p = 0.0825; PSD-95 intensity: GFP: 130.5 ± 6.2224 A.U.; shRNA: 141.6 ± 6.395 A.U., p = 0.8390; WT: 128 ± 7.289 A.U., p = 0.9999; A687T: 138.4 ± 6.524 A.U., p = 0.9562; A687V: 132.9 ± 6.487, p = 0.9998; A687S: 134.5 ± 7.378 A.U., n = 96/4, p = 0.9982; PSD-95 density per 10 μm: GFP: 3.771 ± 0.2545; shRNA: 3.351 ± 0.2175, p = 0.7890; WT: 2.817 ± 0.1914, p = 0.0764; A687T: 4.336 ± 0.0204, p = 0.4937; A687V: 3.847 ± 0.2459, p = 0.9999; A687S: 4.833 ± 0.2584, p = 0.0100; PSD-95 size: GFP: 0.6241 ± 0.03119 μm2; shRNA: 0.6087 ± 0.03365 μm2, p = 0.9996; WT: 0.5588 ± 0.03339 μm2, p = 0.8485; A687T: 0.7021 ± 0.03544 μm2, p > 0.6371; A687V: 0.6185 ± 0.02780 μm2, p > 0.9999; A687S: 0.8623 ± 0.05375 μm2, p < 0.0001; one-way ANOVA, multiple comparisons. D, Representative images of DIV14 primary hippocampal neurons live surface labeled for GluA1. E, Summary graphs of surface GluA1 intensity (top left), density (top right), and size (bottom left) reveals that A687S significantly increases the size and density of surface localized GluA1. There was no change in AMPAR colocalization PSD-95 clusters (bottom right). Intensity, GFP: 64.65 ± 2.346 A.U., n = 106/3; shRNA: 65.01 ± 2.377 A.U., n = 110/3, p = 0.9998; WT: 62.90 ± 1.618 A.U., n = 116/3, p = 0.9659; A687T: 60.92 ± 1.745 A.U., n = 125/3, p = 0.35769; A687V: 61.20 ± 2.146 A.U., n = 112/3, p = 0.6677; A687S: 58.71 ± 2.230 A.U., n = 122/3, p = 0.1658; density per 10 μm, GFP: 5.811 ± 0.1643; shRNA: 6.165 ± 0.1444, p = 0.3394; WT: 5.808 ± 0.1670, p > 0.9999; A687T: 5.669 ± 0.1281, p = 0.9518; A687V: 6.059 ± 0.1661, p = 0.7164; A687S: 6.468 ± 0.1792, p = 0.0169; size, GFP: 1.047 ± 0.07031 μm2; shRNA: 0.8941 ± 0.06724 μm2, p = 0.2191; WT: 0.9147 ± 0.07112 μm2, p = 0.3306; A687T: 1.012 ± 0.05476 μm2, p = 0.9899; A687V: 0.9884 ± 0.06772 μm2, p = 0.8995; A687S: 1.568 ± 0.1120 μm2, p < 0.0001; one-way ANOVA, multiple comparisons. Data represented are mean ± SEM; mean values are quantified from the average of the number of dendrites in all experiments. Numbers in bar graphs represents the number of dendrites analyzed from the number of independent experiments. **p < 0.01; ****p < 0.0001.
Nrxn3 SS4− nonredundantly controls AMPAR trafficking at excitatory synapses in the hippocampus (Aoto et al., 2013, 2015; Dai et al., 2019). We next performed live surface staining of GluA1 as previously described (Lloyd et al., 2023) and then colabeled the cultures for the PSD-95 (Fig. 3D). Relative to neurons transduced with GFP, molecular replacement with A687S led to a modest but significant ∼11% increase in GluA1 cluster density and a striking ∼60% increase in GluA1 size (Fig. 3E). In contrast, neither A687T nor A687V altered GluA1 cluster properties when compared with GFP control neurons (Fig. 3E). These enlarged GluA1 clusters exhibited an identical degree of colocalization with PSD-95 clusters when compared with the colocalization of GluA1 with PSD-95 in control neurons (Fig. 3D,E). Although we observed a modest A687S-dependent increase in the density of PSD-95 and GluA1 clusters, this density increase was not matched by a concomitant increase in the density of presynaptic vGluT1 relative to neurons infected with GFP only (Fig. 3B–E). Thus, A687S may produce a modest number of postsynaptic structures that are not opposed to functional presynaptic terminals. However, the primary morphological effect of the A687S substitution is an increase in the size of excitatory postsynapses. The A687S-dependent increase in PSD-95 and GluA1 morphological properties prompted us to next investigate if these morphological changes impact the function of excitatory synapses.
Molecular replacement with Nrxn3αA687S selectively increases excitatory postsynaptic strength
The disparate morphological phenotypes at excitatory synapses induced by molecular replacement with A687T and A687S were unexpected given that threonine and serine possess polar, hydrophilic side chains. We next asked if there was a corresponding functional consequence of molecular replacement with A687S on excitatory synaptic transmission. We performed molecular replacement on DIV3 with Nrxn3αWT or Nrxn3αA687 SS4− variants. On DIV14-15, we performed whole-cell patch-clamp electrophysiology from eGFP-positive spiny pyramidal neurons. Nrxn3αA687 substitutions did not alter neuronal intrinsic membrane properties (Fig. 4A). We next monitored the impact of Nrxn3αA687 substitutions on the frequency and amplitude of miniature excitatory postsynaptic currents (mEPSCs). Changes in mEPSC frequency are commonly correlated with changes in synapse density and/or presynaptic release probability, while changes in mEPSC amplitude are commonly thought to reflect changes in postsynaptic strength. Consistent with prior studies, knockdown of endogenous Nrxn3α did not alter mEPSC properties. Further, molecular replacement with WT or A687V did not alter mEPSC parameters relative to control conditions. However, A687T and A687S produced strikingly different gain-of-function synaptic phenotypes. Consistent with our previous findings, molecular replacement with A687T induced a ∼57% increase in mEPSC frequency without altering mEPSC amplitude (Restrepo et al., 2019). In contrast, molecular replacement with A687S significantly increased mEPSC amplitude by 30% without altering mEPSC frequency (Fig. 4B,E,F). Thus, consistent with the distinct excitatory synapse morphology phenotypes (Fig. 3), molecular replacement with A687T increased presynaptic strength while molecular replacement with A687S increased postsynaptic strength.
Molecular replacement with Nrxn3αA687S selectively increases mEPSC amplitude and evoked AMPAR-mediated transmission at excitatory synapses. A, Summary graphs of the passive membrane properties of neurons infected with lentiviruses encoding only GFP or Nrxn3α shRNA and following molecular replacement with Nrxn3α WT or A687. There were no changes in membrane capacitance (Cm) and membrane resistance (Rm) between conditions. Cm: GFP Control: 173.3 ± 9,875 pF, n = 18; shRNA: 164.7 ± 11.01 pF, n = 21, p = 0.9797; WT: 180.9 ± 6.328 pF, n = 16, p = 0.9912; A687T: 167.3 ± 6.328 pF, n = 20, p = 0.9962; A687V: 176.0 ± 7.296 pF, n = 17, p > 0.9999; A687S: 178.2 ± 7.338 pF, n = 17, p = 0.9989; Rm: GFP Control: 190.1 ± 14.23 mΩ; shRNA: 242.1 ± 21.43 mΩ, p = 0.1132; WT: 224.8 ± 11.46 mΩ; A687T: 237.5 ± 10.53 mΩ, p = 0.5386; A687V: 221.6 ± 8.653 mΩ, p = 0.5929; A687S: 207.8 ± 15.64 mΩ, p = 0.9564; one-way ANOVA, multiple comparisons versus GFP. B, Representative traces from DIV14 to DIV15 primary hippocampal neurons infected with the lentiviruses indicated in the panel. C, D, Summary graphs of the cumulative distribution of mEPSC frequency (C) and average frequency per condition (D). E, F, Summary graphs of the cumulative distribution of mEPSC amplitude (E) and average amplitude per condition (F). The A687S substitution selectively increased mEPSC amplitude without altering mEPSC frequency. Frequency: GFP Control: 1.293 ± 0.1362 Hz; shRNA: 1.289 ± 0.1278 Hz, p > 0.9999; WT: 1.180 ± 0.1577 Hz, p = 0.0853; A687T: 1.984 ± 0.1984 Hz, p = 0.0105; A687V: 1.292 ± 0.1279 Hz, p > 0.99999; A687S:1.089 ± 0.1905 Hz, p = 0.8429; amplitude: GFP Control: 12.03 ± 0.5846 pA; shRNA: 12.98 ± 0.4583 pA, p = 0.6454; WT: 11.73 ± 0.4320 pA, p = 0.9963; A687T: 11.90 ± 0.5683 pA, p = 0.9998; A687V: 12.64 ± 0.6503 pA, p = 0.9228; A687S: 15.47 ± 0.7739 pA, p = 0.0004; one-way ANOVA, multiple comparisons. G, Representative traces (left) and summary graph (right) of evoked AMPAR EPSC amplitudes reveal A687S significantly increased AMPAR-mediated synaptic transmission. GFP: 732 ± 90 pA; shRNA: 805 ± 110 pA, p = 0.98; WT: 714 ± 100 pA, p = 0.999; A687S: 1323 ± 160 pA, p = 0.008; one-way ANOVA multiple comparisons to GFP. H, Representative traces (left) and summary graph (right) of evoked NMDAR EPSC amplitudes. GFP: 828 ± 222 pA, p = 0.931; shRNA: 649 ± 195 pA, p = 0.969; WT: 699 ± 228 pA; A687S: 655 ± 168 pA, p = 0.924; one-way ANOVA multiple comparisons to GFP. Data represented are mean ± SEM; mean values are quantified from the average of the number of cells in all experiments. Numbers in bar graphs indicate the number of cells analyzed from the number of independent experiments.
The increase in mEPSC amplitude following molecular replacement with A687S was unexpected given the robust presynaptic gain-of-function observed for A687T. Although mEPSC amplitudes are dominated by AMPAR-mediated currents, we next asked whether the postsynaptic gain-of-function observed with A687S reflected by changes in AMPAR- and/or NMDAR-mediated synaptic strength. To address this question, we pharmacologically blocked NMDARs and monitored electrically evoked AMPAR-mediated EPSCs. In agreement with our immunocytochemistry indicating more surface GluA1, molecular replacement with A687S increased AMPAR-EPSC amplitudes by 84%. In contrast, A687S did not elicit changes in NMDAR-mediated EPSCs (Fig. 4G,H). Taken together, in striking contrast to the A687T-dependent presynaptic gain-of-function, A687S-dependent postsynaptic gain-of-function selectively impacts postsynaptic AMPAR-mediated synaptic transmission independent of changes to NMDAR function or presynaptic release.
Molecular replacement with Nrxn3αA687S selectively increases synapse density, independent to changes in GABAAR surface expression in vitro
In primary hippocampal cultures, unlike in vivo, the genetic manipulation of Nrxn3α or Nrxn3α/β has not resulted in functional or morphological phenotypes at inhibitory synapses (Aoto et al., 2013, 2015; Restrepo et al., 2019; Boxer et al., 2021). However, in the heterologous synapse formation assay, all Nrxn3α variants tested exhibited similar levels of synaptogenic activity of inhibitory synapses in this reduced system, suggesting that Nrxn3α can engage in interactions with ligands found at inhibitory synapses. We performed molecular replacement in dissociated hippocampal neurons on DIV3 and fixed and immunolabeled cells on DIV14 to visualize vGAT and gephyrin, which are pre- and postsynaptic markers of inhibitory synapses, respectively. Consistent with our previous study, knockdown of endogenous Nrxn3α or molecular replacement with WT or A687T did not alter vGAT or gephyrin intensity, density, or size (Fig. 5A–C). Importantly, molecular replacement with A687S led to an unexpected, and selective, ∼28% increase in density of vGAT and a ∼14% increase in the density of gephyrin clusters (Fig. 5A–C). The unexpected increase in the density of vGAT and gephyrin prompted us to next test whether these synapses are functional and contain surface GABAA receptors. We live surfaced labeled neurons for the γ2 subunit of GABAARs, then fixed, and stained for gephyrin. Relative to GFP and shRNA-infected control neurons, molecular replacement with WT, A687T, or A687V did not alter the intensity, density, or size of GABAAγ2 positive clusters (Fig. 5D,E). However, A687S reduced the density of surface GABAAγ2 clusters by 15% and decreased the percent colocalization of surface GABAAγ2 with gephyrin by ∼14% (Fig. 5E). Together, these data indicate that although A687S increased the density of vGAT and gephyrin clusters by 28 and 14%, respectively, at least some of these nascent synapses may be nonfunctional, ectopic synapses.
Molecular replacement with Nrxn3αA687S selectively increases synapse density, independent of changes in GABAAR surface expression in vitro. A, Representative images of DIV14 primary hippocampal neurons immunostained for VGAT and gephyrin. B, C, Summary graphs of presynaptic vGAT (B) and postsynaptic gephyrin (C) intensity (left), density (middle), and size (right) reveals that A687S significantly increases VGAT and gephyrin density. vGAT intensity: GFP: 78.62 ± 2.072 A.U., n = 111/3; shRNA: 93.66 ± 2.11 A.U., n = 96/3, p < 0.00001; WT: 87.11 ± 1.90 A.U., n = 111/3, p = 0.0072; A687T: 85.76 ± 2.513 A.U., n = 104/3, p = 0.2925; A687S: 78.62 ± 1.997 A.U., n = 109/3, p > 0.9999; 83.66 ± 2.453 A.U., n = 104/3, p = 0.5568; vGAT density per 10 μm: GFP: 2.223 ± 0.095; shRNA:1.969 ± 0.093, p = 0.4868; WT: 2.418 ± 0.109, p = 0.7216; A687T: 1.848 ± 0.084, p = 0.0831; A687S: 2.857 ± 0.105, p < 0.00001; A687V: 2.268 ± 0.105; vGAT size: GFP: 1.717 ± 0.067 μm2; shRNA: 2.281 ± 0.108 μm2, p < 0.0001; WT: 1.827 ± 0.088 μm2, p = 0.9347; A687T: 1.8143 ± 0.087, p = 0.8975; A687V: 1.811 ± 0.086 μm2, p = 0.9661; A687S: 1.615 ± 0.068 μm2, p = 0.9547; one-way ANOVA, multiple comparisons. Gephyrin intensity: GFP: 95.54 ± 2.889 A.U.; shRNA: 95.93 ± 1.813 A.U., p > 0.9999; WT: 104.7 ± 2.458 A.U., p = 0.0613; A687T: 99.68 ± 2.273 A.U., p = 0.8167; A687V: 99.67 ± 2.338 A.U., p = 0.8044; A687S: 95.13 ± 2.087 A.U., p > 0.9999; gephyrin density per 10 μm: GFP: 3.565 ± 0.135; shRNA: 3.277 ± 0.133, p = 0.6404; WT: 3.369 ± 0.128, p = 0.8875; A687T: 3.130 ± 0.121, p = 0.1687; A687V: 3.1116 ± 0.130, p = 0.1276; A687S: 4.106 ± 0.132, p = 0.0356; gephyrin size: GFP: 1.258 ± 0.068 μm2; shRNA: 1.346 ± 0.097 μm2, p = 0.9378; WT: 0.9884 ± 0.054 μm2, p = 0.0387; A687T: 1.091 ± 0.059 μm2, p = 0.4671; A687V: 0.9859 ± 0.054 μm2, p = 0.034; A687S: 1.121 ± 0.061 μm2, p = 0.6742; one-way ANOVA, multiple comparisons. D, Representative images of DIV14 primary hippocampal neurons live surface labeled for the GABAAR subunit, GABAγ2. E, Summary graphs of surface GABAA intensity (top left), density (top right), and size (bottom left) reveals that A687S significantly decreased the density of surface localized GABAA. Intensity: GFP: 60.65 ± 1.321 A.U., n = 176/5; shRNA: 56.51 ± 1.425 A.U., n = 196/5, p = 0.1426; WT: 55.53 ± 1.335 A.U., n = 186/5, p = 0.0474; A687T: 53.62 ± 1.516 A.U., n = 174/5, p = 0.0030; A687S: 54.24 ± 1.364 A.U., n = 191/5, p = 0.0067; A687V: 54.64 ± 1.502A.U., n = 188/5, p = 0.0132; density per 10 μm2: GFP: 5.344 ± 0.123; shRNA: 5.204 ± 0.125, p = 0.9589; WT: 5.060 ± 0.125, p = 0.5364; A687T: 4.952 ± 0.108, p = 0.1949; A687S: 4.744 ± 0.011, p = 0.044; A687V: 4.894 ± 0.108, p = 0.0767; size: GFP: 1.157 ± 0.050 μm2; shRNA: 1.236 ± 0.053 μm2, p = 0.9069; WT: 1.282 ± 0.058 μm2, p = 0.5907; A687T: 1.363 ± 0.061 μm2, p = 0.0964; A687V: 1.157 ± 0.044 μm2, p > 0.9999; A687S: 1.280 ± 0.057 μm2, p = 0.6000; one-way ANOVA, multiple comparisons. A687S induced a decrease in GABAAR colocalization with gephyrin clusters (bottom right). GFP: 0.8083 ± 0.016 n = 91/3; shRNA: 0.7601 ± 0.016, n = 114/3 p = 0.2149; WT: 0.8199 ± 0.014, n = 109/3 p = 0.9947; A687T: 0.8121 ± 0.014, n = 102/3, p > 0.9999; A687V: 0.8430 ± 0.012, n = 110/3, p = 0.5925; A687S: 0.7301 ± 0.016, n = 110/3, p = 0.0042; one-way ANOVA, multiple comparisons. Data represented are mean ± SEM; mean values are quantified from the average of the number of dendrites in all experiments. Numbers in bar graphs describe number of dendrites from the number of independent experiments.
Molecular replacement with Nrxn3αA687S does not alter inhibitory transmission in vitro
The A678S substitution unexpectedly increased inhibitory synapse density without a concomitant increase in the density of surface GABAARs. We tested whether inhibitory synaptic transmission was altered by molecular replacement with Nrxn3αA687S. After 10–11 d of molecular replacement, intrinsic membrane properties were unaltered independent of experimental condition (Fig. 6A). Miniature inhibitory postsynaptic current (mIPSC) frequency and amplitude were unchanged across all conditions, consistent with the notion that Nrxn3α does not have a functional role at inhibitory synapses in mixed primary cultures (Fig. 6B–F; Aoto et al., 2013, 2015; Restrepo et al., 2019). Thus, despite a modest increase in the density of vGAT and gephyrin clusters after the replacement with A687S, mIPSC frequency and amplitude are unchanged, suggesting that a proportion of inhibitory synapses, induced by Nrxn3αA687S are likely nonfunctional. Taken together, the molecular replacement with A687T and A687S shifts the balance of excitation to inhibition toward excitation, and this shift occurs via different mechanisms; A687T increased presynaptic release of glutamate, whereas A687S increased postsynaptic strength (Restrepo et al., 2019).
Molecular replacement with Nrxn3αA687S does not alter inhibitory synaptic transmission in vitro. A, Summary graph of passive membrane properties of neurons infected with lentivirus encoding GFP alone, Nrxn3α shRNA along or following molecular replacement with Nrxn3α WT or A687. There were no changes in membrane capacitance (Cm) and membrane resistance (Rm) between conditions. Cm: GFP Control: 167.7 ± 8.618 pF, n = 16/4; shRNA: 158.2 ± 7.560 pF, n = 20/4, p = 0.9616; WT: 155.3 ± 6.069 pF, n = 19/4, p = 0.8909; A687T: 156.2 ± 6.734 pF, n = 20/4, p = 0.9169; A687V: 161.7 ± 8.118, pF, n = 22/4, p = 0.9946; A687S: 150.1 ± 8.610 pF, n = 20/4, p = 0.6396; one-way ANOVA, multiple comparisons; Rm: GFP Control: 207.2 ± 26.83 mΩ; shRNA: 270.5 ± 21.20 mΩ, p = 0.3276; WT: 262 ± 21.15 mΩ, p = 0.4797; A687T: 253.4 ± 17.66 mΩ, p = 0.6730; A687V: 272.1 ± 1731 mΩ, p = 0.2546; A687S: 269.5 ± 20.20 mΩ, p = 0.3196; one-way ANOVA, multiple comparisons. B, Representative traces from DIV14 to DIV15 primary hippocampal neurons infected with lentivirus indicated above each trace. C, D, Summary graphs of the cumulative distribution of mIPSC frequency (C) and average frequency per condition. Frequency: GFP Control: 2.859 ± 0.3755 Hz; shRNA: 2.04 ± 0.2907 Hz, p = 0.4166; WT: 1.892 ± 0.2414 Hz, p = 0.2462; A687T: 2.348 ± 0.2787 Hz, p = 0.8475; A687V: 2.674 ± 0,2767 Hz, p > 0.9990; A687S: 2.283 ± 0.3062 Hz, p = 0.7700; one-way ANOVA, multiple comparisons; amplitude: GFP Control: 15.87 ± 0.7440 pA; shRNA: 12.81 ± 0.6756 pA, p = 0.0297; WT: 14.14 ± 0.8320 pA, p = 0.5231; A687T: 13.51 ± 0.6736 pA, p = 0.1738; A687V: 15.33 ± 0.5504 pA, p = 0.9940; A687S: 14.51 ± 0.5440 pA, p = 0.7472; one-way ANOVA, multiple comparisons (D). E, F, Summary graphs of the cumulative distribution of mIPSC amplitude (E) and average amplitude per condition (F). Data represented are mean ± SEM; mean values are quantified from the average of the number of cells in all experiments. Statistical significance was determined by one-way ANOVA, multiple comparisons (to GFP) **p < 0.01. The number of cells monitors from independent experiments indicated in the bar graphs in A.
Nrxn3αA687 substitutions do not alter evoked presynaptic calcium dynamics
The constitutive ablation of all α-Nrxns impairs spontaneous and evoked release of excitatory and inhibitory neurotransmitters through the decoupling of presynaptic Ca2+ channels (Missler et al., 2003; Tong et al., 2017; Brockhaus et al., 2018). Molecular replacement with A687T significantly increased the size of the readily releasable pool and increased presynaptic release probability (Restrepo et al., 2019); however, presynaptic Ca2+ dynamics were not previously assessed. We asked whether changes in presynaptic calcium dynamics might contribute to the synaptic gain of function phenotypes observed after molecular replacement with A687T (Restrepo et al., 2019) and A687S (this study). We infected primary neurons with a dual-promoter lentivirus encoding both Nrxn3α shRNA and mCherry, to knockdown endogenous Nrxn3α and fluorescently identify neuron morphology, respectively, and coinfected with lentiviruses encoding WT or A687 variants. We also infected the neurons with lentivirus encoding the Ca2+ indicator GCaMP5G fused to the synaptic vesicle protein synaptobrevin-2 (Syb2), which effectively targets GCaMP5G to presynaptic terminals (Anderson et al., 2015; Fig. 7A). Extracellularly evoked action potentials (1 × 50 Hz stimulus) reliably elicited Ca2+ transients that we measured as GCaMP5G fluorescence (Fig. 7B). The manipulation of Nrxn3α neither altered the peak GCaMP5G fluorescence nor the decay kinetics of Ca2+ transients (Fig. 7C–F). Due to the redundant regulation of presynaptic Ca2+ channel function by α-Nrxns, it is perhaps not surprising that the knockdown of Nrxn3α alone did not alter presynaptic calcium dynamics, as measured by GCaMP (Missler et al., 2003). Together, these data indicate that Ca2+ dynamics per se does not contribute to the synaptic phenotypes induced by molecular replacement with A687T and A687S (Restrepo et al., 2019).
Molecular replacement with Nrxn3αA687 does not alter peak presynaptic calcium influx or clearance after electrical stimulation in vitro. A, Schematic depicting molecular replacement using primary hippocampal neurons infected with lentivirus encoding mCherry alone or Nrxn3α shRNA and mCherry, lentivirus for molecular replacement with Nrxn3α WT and A687 and GCaMP5G lentivirus. B, Representative images of DIV14 primary hippocampal neurons with GCaMP activation after evoked electrical stimulus of 50 Hz. C, Representative traces of evoked calcium influx after control, Nrxn3α knockdown, and WT or A687 molecular replacement in primary hippocampal neurons. D, Summary graph of peak calcium influx (ΔF / F0) revealed no alterations in peak calcium influx in any conditions. E, Representative traces of evoked calcium influx normalized to the peak after control, Nrxn3α knockdown, and WT or A687 molecular replacement in primary hippocampal neurons. F, Summary graph of the decay time constant representing the calcium clearance from presynaptic terminals (ΔF / F0) revealed no alterations in peak calcium influx in any conditions. Data represented are mean ± SEM; mean values are quantified from the average of the number of cells in all experiments. Peak amplitude+: mCherry Control: 1.696 ± 0.077 ΔF / F0, n = 53/5; shRNA: 1.678 ± 0.111 ΔF / F0, n = 42/5, p > 0.9999; WT: 1.892 ± 0.079 ΔF / F0, n = 50/5, p > 0.9999; A687T: 1.787 ± 0.861 ΔF / F0, n = 50/5, p = 0.9786; A687V: 1.763 ± 0.092 ΔF / F0, n = 61/5, p = 0.9936; A687S: 1.843 ± 0.076 ΔF / F0, n = 64/5, p = 0.8133; one-way ANOVA, multiple comparisons versus GFP; exponential decay: mCherry Control: 14.20 ± 3.046 τ, n = 4; shRNA: 13.04 ± 2.532 τ, n = 4, p = 0.9976; WT: 16.93 ± 1.937 τ, n = 4, p = 0.4979; A687T: 13.51 ± 1.434, n = 4, p = 0.9968; A687V: 15.30 ± 1.463, n = 4, p = 0.9858; A687S: 18.54 ± 2.773 n = 4, p = 0.6122; one-way ANOVA, multiple comparisons (to GFP). Numbers in bar graphs represent the average fluorescence change monitored from a single video recording from four independent experiments (C, D). The data points in F represent the Ca2+ exponential decay time averaged from 100 frames per video and subsequently averaged across independent experiments.
Nrxn3αA687S selectively enhances Nrxn3α aggregation with NL1 and NL3 but not with NL2 or LRRTM2
Nrxn3αA687T and Nrxn3αA687S differ by a single methyl group yet molecular replacement with A687T and A687S resulted in distinct morphological and functional phenotypes at excitatory synapses (Figs. 3, 4). Two classes of postsynaptic Nrxn ligands important for synapse maintenance and plasticity are NL family proteins and LRRTM family proteins. While NLs and LRRTMs bind to LNS6 of Nrxns (Siddiqui et al., 2010; Sudhof, 2017; Kim et al., 2021; Connor and Siddiqui, 2023), they differ in their synapse localization: LRRTMs and NL1 are exclusively localized to excitatory synapses in neurons, NL2 is localized to inhibitory synapses, and NL3 is localized to both types of synapses (Südhof, 2017). Relevant to the impact of A687T (Restrepo et al., 2019) and A687S (this study) on distinct aspects of excitatory morphology and function, LRRTM1 and LRRTM2 redundantly control excitatory synapse formation during development and regulate AMPAR-mediated synaptic transmission (Soler-Llavina et al., 2013). In particular, LRRTM2 interacts with Nrxn3 to control postsynaptic AMPAR strength (Aoto et al., 2013). α-Nrxns can only bind NL1 isoforms that lack splice-site B (NL1ΔB; Boucard et al., 2005). While the genetic ablation of NL1 indicates that NL1 is dispensable for synapse formation but regulates NMDAR-mediated synaptic transmission (Wu et al., 2019), NL1ΔB can selectively increase the size of excitatory synapses (Boucard et al., 2005). Although present at excitatory and inhibitory synapses, the genetic ablation of NL3 does not alter synaptic transmission at either type of synapses (Chanda et al., 2017). At the nanoscale level, LRRTM2 and NL1 exhibit distinct subsynaptic distributions—LRRTM2 is enriched near the center of the synapse, while NL1 is more diffuse but enriched more peripherally (Chamma et al., 2016a, b; Nozawa et al., 2022; Lloyd et al., 2023). Thus, we hypothesized that the distinct excitatory synaptic phenotypes observed following molecular replacement with A687T and A687S are driven by selective interactions with different postsynaptic ligands. To test this, we performed transcellular aggregation assay, which measures the ability of Nrxn-ligand pairs to mediate stable interactions in trans between two populations of heterologous cells (Restrepo et al., 2020). Briefly, we cotransfected one population of HEK293T cells with mCherry and Nrxn3αWT or Nrxn3αA687 variants and cotransfected a second population of HEK293T cells with GFP and LRRTM2 or the indicated NL paralog. For these experiments, all NL constructs contained the splice-site A insert, and we excluded the NL1-specific splice-site B insert. These two HEK293T cell populations were incubated together in suspension, and cellular aggregation was quantified by mCherry and GFP colocalization.
We observed no cellular aggregation of HEK cells that expressed either mCherry or GFP expressed alone, which indicates that cellular aggregation is dependent on trans interactions between cells expressing Nrxn3α and a cognate ligand (Fig. 8A–E). As previously reported, relative to wild-type, the A687T mutant significantly increased the degree of aggregation with LRRTM2 by 20% without altering binding with any NLs (Restrepo et al., 2019). Consistent with our findings that A687T and A687S promote distinct morphological and functional phenotypes at excitatory synapses, the A687S mutant, relative to Nrxn3αWT, increased Nrxn3α aggregation with NL1 by 48% and with NL3 by 35% without altering binding to LRRTM2 or NL2 (Fig. 8A–E). The enhanced binding activity of Nrxn3αA687S to NL1 and NL3 provides insight into a possible mechanism that underlies the different morphological and functional gain-of-function phenotypes observed after molecular replacement (Figs. 3, 4). The increased Nrxn3αA687S–NL3 interaction may underlie the modest increase in the number of nonfunctional inhibitory synapses (Fig. 5B,C) and, consistent with previous work, Nrxn–NL3 interactions are most likely dispensable for synapse maturation or function (Chanda et al., 2017; Fig. 5E). Unexpectedly, we observed a modest (∼18%), yet statistically significant reduction in Nrxn3αA687V aggregation with NL2 relative to Nrxn3αWT, and this modest change in aggregation is not functionally or morphologically relevant at inhibitory synapses (Figs. 5, 6, 8D). Together, these results suggest that the divergent morphological and functional phenotypes induced by molecular replacement with Nrxn3αA687T and Nrxn3αA687S may be driven by a differential increase in binding interactions with LRRTM2 and NL1.
The A687S substitution selectively enhances Nrxn3α aggregation with the postsynaptic ligands, NL1 and NL3. A, Representative images of the heterologous cell aggregation assay. Inset images are magnified cell aggregates. B–E, Summary graphs of quantified heterologous cell aggregation assays normalized to WT aggregation. One sample of HEK293T cells are transfected with WT or A687 Nrxn3α and mCherry and coincubated with a sample of HEK293T cells transfected with Nrxn ligands. B, LRRTM2; WT: 1.00 ± 0.0, n = 5, A687T: 1.202 ± 0.038, n = 5, p = 0.0007; A687V: 0.9995 ± 0.073, n = 5, p > 0.9999; A687S: 1.126 ± 0.021, n = 5, p = 0.1043. C, NL1; WT: 1.00 ± 0.0, n = 5, A687T: 0.8051 ± 0.169, n = 5, p = 0.8688; A687V: 1.043 ± 0.1785: 0.8227, n = 5, p > 0.9999; A687S: 1.481 ± 0.0158, n = 5, p = 0.0158. D, NL2; WT: 1.00 ± 0.0, n = 5, A687T: 0.9104 ± 0.05682, n = 5, p = 0.5490; A687V: 0.8227 ± 0.04922, n = 5, p = 0.0063; A687S: 0.9959 ± 0.05292, n = 5, p > 0.9999. E, NL3; NL3: WT: 1.00 ± 0.0, n = 5, A687T: 0.9208 ± 0.169, n = 4, p = 0.9988; A687V: 0.9466 ± 0.1392, n = 5, p > 0.9999; A687S: 1.361 ± 0.104, n = 5, p = 0.0231. The index of aggregation was calculated by quantifying the number of colocalized cells (cells with red + green overlap) divided by the total number of both cell samples. Data represented are mean ± SEM; mean values are quantified from the average experiments analyzed. Numbers in bar graphs describe individual experiments. Statistical significance was determined by one-way ANOVA, multiple comparisons. *p < 0.05; ***p < 0.001; ****p < 0.0001. Number in bars represents the number of independent experiments. Scale bars: 200 μm, inset: 50 μm.
The unique N-terminal sequences of neuroligin-1 and neuroligin-2 differentially modulate cellular aggregation with Nrxn3αA687S
The extracellular sequences of NLs consist of short N-terminal sequences followed by a large cholinesterase domain and a juxtamembrane stalk region (Leone et al., 2010) that share high extracellular sequence homology (NL1 vs NL2: ∼71%, NL1 vs NL3: ∼75%; NL2 vs NL3: ∼71%) and bind identical surface residues in LNS6 of Nrxns. Thus, the selective A687S-mediated increase in Nrxn3α binding to NL1 and NL3 but not to NL2 in the cellular aggregation assay was unexpected. Given the sequence conservation of NLs, how does the A687S substitution specifically enhance binding to NL1 but not NL2? Nrxn1α–NL1 binding occurs at LNS6 of Nrxns (Reissner et al., 2008; Bemben et al., 2015); however, there is a possibility that the presence of splice-site A (A-site) of NLs may contact LNS4 to modulate interactions between α-Nrxns and NL1 (Chen et al., 2011). Curiously, these α-Nrxn-specific sequences, particularly the LNS3–EGF-B–LNS4 module, might suppress α-Nrxn binding to NLs and LRRTMs (Boucard et al., 2005; Kang et al., 2008; de Wit et al., 2009). The disease-relevant A687T mutation and the subsequent substitutions studied here reside at the interface between EGF-B and LNS4 and may make LNS4 more accessible to NL A-site interactions (Fig. 1A). While negative modulation of ligand binding imposed by the LNS3–EGF-B–LNS4 module might be disrupted by the A687T and A687S mutations, it does not completely explain how the A687S substitution increased cellular aggregation with NL1 but not NL2 because both expression constructs included the A-site exon. Although the cholinesterase-like domains of NLs exhibit significant sequence homology, the N-terminal sequences exhibit significant variability (Fig. 9A). NL1 possesses a shorter sequence preceding the cholinesterase-like domain than NL2 (NL1: 8 residues vs NL2: 28 residues), and these sequences only share ∼29% sequence homology. Additionally, the first 14 amino acids of the cholinesterase-like domain exhibit only ∼36% homology (Fig. 9B; Leone et al., 2010). Given the difference in N-terminal sequence homology between NL1 and NL2, we tested whether these highly variable sequences contribute to the differential effect of the A687S substitution in the aggregation assay.
Domain swapping reveals the sequences of NL1 are necessary to enhance trans interactions with Nrxn3αA687S. A, Diagram of the domain organization (left) and homology (right) of neuroligin-1 and neuroligin-2. The N-termini of NL1 and NL2 only share 27.3% homology. B, Schematic of the NL1/NL2 domain swap. The sequences immediately downstream of the signal peptide for NL1 (aa: 46–85) and NL2 (aa: 15-75) exhibit minimal homology (top). The NL2.1 chimera possesses amino acids 1–75 of NL2 fused in-frame to NL1 and NL1.2 possesses amino acids 1–85 of NL1 fused in-frame to NL2 (bottom). C, D, Representative images. Inset images are magnified cell aggregates (scale bar: 200 μm, inset: 50 μm; top) and summary graphs (bottom) of cellular aggregation between Nrxn3α WT and A687S with NL1 and NL2.1 (C). WT-NL1: 1.00; WT-NL2.1: 0.902 ± 0.085, p = 0.95; A687S-NL1: 1.45 ± 0.09, p < 0.0001; A687S-NL2.1: 1.07 ± 0.1, p = 0.991; one-way ANOVA multiple comparisons to WT-NL1. Cellular aggregation of Nrxn3α WT and A687S with NL2 and NL1.2 N = 13 independent experiments (D). WT-NL2: 1.00; WT-NL1.2: 1.27 ± 0.04, p = 0.65; A687S-NL2: 1.385 ± 0.15, p = 0.171; A687S-NL1.2: 1.79 ± 0.25, p < 0.0001. One-way ANOVA multiple comparisons to WT-NL2. N = 8 independent experiments. E, Coomassie blue-stained Fc proteins, expressed in HEK293T cells and purified by affinity chromatography, for in cell-surface–binding assays. F, Representative images showing HEK293T cells expressing N-terminally HA-tagged Nlgn1, Nlgn2, Nlgn1.2, Nlgn2.1, and LRRTM2 incubated with purified Ig-fused Nrxn3 WT (Ig-Nrxn3 WT), Nrxn3 A687S (Ig-Nrxn3 A687S), or negative control (IgC), analyzed by immunofluorescence imaging for Ig-fusion proteins (red) and HA (blue; left). Quantification of cell surface binding. Data are means ± SEM (n = 13–21 cells/group; *p < 0.05; Mann–Whitney U test). Scale bar, 10 μm (applies to all images; right). G, Representative images showing HEK293T cells expressing N-terminally HA-tagged Nrxn3 WT or Nrxn3 A687S incubated with purified Ig-Nlgn1, Ig-Nlgn1.2, Ig-Nlgn2.1, Ig-LRRTM2, or negative control (IgC), analyzed by immunofluorescence imaging for Ig-fusion proteins (red) and HA (blue; left) and quantification of cell surface binding. Data are means ± SEM (n = 15–20 cells/group). Scale bar, 10 μm (applies to all images; right). Statistical significance was determined by one-way ANOVA, multiple comparisons (Nrxn3α WT to Nrxn3αA687S). *p < 0.05; ***p < 0.001. Each data point in bars represents the average from a single experiment.
We generated NL1 and NL2 chimeras by swapping amino acids 46–67 of NL1 with amino acids 15–56 of NL2 (numbering includes the signal peptide), which resulted in the N-terminus of NL1 fused in-frame with NL2 (NL1.2) and N-terminal sequences of NL2 fused in-frame with NL1 (NL2.1). Each swapped sequence harbors its endogenous signal peptide and the variable N-terminal region (Fig. 9B). We examined whether the NL2.1 chimera was sufficient to abrogate the enhanced cellular aggregation of HEK cells expressing NL1 with HEK cells expressing Nrxn3αA687S (Fig. 9B). Consistent with our initial findings, A687S enhanced cell aggregation with wild-type (WT) NL1; however, the inclusion of the first 56 residues of NL2 in the NL2.1 chimera completely prevented this effect (Fig. 9C). In fact, the aggregation index for Nrxn3αA687S with NL2.1 was no different from the aggregation index for Nrxn3αWT with NL1, which indicates that the longer N-terminal sequences of NL2 are sufficient to block the effect of A687S. Conversely, the inclusion of the shorter N-terminal sequences (20 amino acids) of NL1 onto NL2 recapitulated the Nrxn3αA687S-NL1 phenotype by significantly increasing the aggregation index of NL1.2 with Nrxn3αA687S, but not with WT Nrxn3α (Fig. 9D). Taken together, these data suggest that the N-terminal sequences of NL1 and NL2 may modulate interactions in trans with Nrxn3αA687S.
The cell surface binding assay identifies differential ligand binding with Nrxn3αA687S
As an additional approach to the cellular aggregation assay, we utilized cell-surface–binding assays to assess whether A687S influences Nrxn3α binding to NL1, NL2, and LRRTM2. We measured the binding of recombinant NL1-Fc, NL2-Fc or LRRTM2-Fc, and HEK293T cells expressing Nrxn3WT or Nrxn3A687S and observed a modest but nonsignificant increase in NL1-Fc binding to A687S and an unexpected significant (17%) increase in binding between LRRTM2-Fc and A687S (Fig. 9E,F). To corroborate these observations, we performed reverse-orientation cell-surface–binding assays, employing recombinant Nrxn3αWT-Fc or Nrxn3αA687S-Fc to HEK293T cells expressing HA-tagged NL1, NL2, NL1.2, NL2.1, or LRRTM2. We found that Nrxn3αWT-Fc, but not Fc alone (negative control), bound to cells expressing NL1, NL2, or LRRTM2 (Fig. 9G). Again, semiquantitative analyses revealed that Nrxn3αA687S-Fc bound more robustly (18% increase) to cells expressing LRRTM2 than Nrxn3αWT-Fc, whereas Nrxn3αA687S-Fc exhibited comparable binding activity to cells expressing NL variants or LRRTM2 (Fig. 9G). These results suggest that in this cell-surface–binding assay, the A687S substitution in Nrxn3α specifically enhances LRRTM2-binding and not NL-binding. While the cellular aggregation assay monitors stable interactions formed engaged in trans by cells in suspension, the cell-surface–binding assay may be more sensitive to transient interactions that occur in cis and trans. Thus, both assays provide insight into the impact of the A687S substitution on ligand interactions. The stable Nrxn–NL1ΔB interactions may selectively increases synapse size (Boucard et al., 2005) while Nrxn3–LRRTM2 interactions stabilize synaptic AMPARs (Aoto et al., 2013, 2015).
Molecular modeling of A687 substitutions in the Nrxn1α crystal structure
NLs and LRRTMs canonically bind the “hypervariable” surface of the LNS6 domain present in all Nrxn isoforms. Curiously, A687 is located hundreds of amino acids upstream of LNS6 and buried at the interface of EGF-B and LNS4, yet A687T and A687S mutations modulate trans-synaptic interactions, synapse morphology, and function (Arac et al., 2007; Chen et al., 2011; Miller et al., 2011; Restrepo et al., 2019). The “hypervariable” surfaces of Nrxn LNS domains are sites of interaction with most postsynaptic ligands. LNS2, LNS4, and LNS6 have been experimentally determined to harbor a Ca2+-binding site; however, based on sequence homology, it is predicted that all LNS domains have the capacity to bind Ca2+ (Chen et al., 2011; Miller et al., 2011). Importantly, the LNS4 domain is only LNS domain where access to the hypervariable surface and the Ca2+-binding site is blocked. A loop from neighboring LNS3 (β4-β5) forms direct contacts with the hypervariable surface of LNS4 (Chen et al., 2011). The A687 mutation is buried in the interface between EGFB and LNS4; however, the mutation is part of a larger module containing LNS3–EGF-B–LNS4 that forms a tight “jelly roll–EGF–jelly roll” structure reminiscent of reelin repeats found in reelin and adopts a horseshoe arrangement (Fig. 10B). Molecular modeling of Nrxn1α and NL1 suggest that the A-site of NL1 may interact with LNS3 and LNS4 of Nrxn1α, suggesting that the α-Nrxn-specific LNS3–EGF-B–LNS4 module may contribute to trans-synaptic interactions (Chen et al., 2011).
Molecular modeling of Nrxn3αA687 substitutions in the Nrxn1α structure. A, The percent sequence identity (78.8%) of the LNS3–EGF-B–LNS4 reelin-like repeat module shared by bovine Nrxn1α and mouse Nrxn3α. B, Sequence alignment of EGF-B–LNS4 from bovine Nrxn1α and mouse Nrxn3α. These sequences share 80.4% sequence identity and have an 89.6% sequence similarity. The EGF-B repeat and LNS4 domain are indicated. Bovine Nrxn1α A723 and mouse Nrxn3α A687 are conserved and highlighted in orange. The conserved tyrosine Y714 (bovine Nrxn1α) and Y678 (mouse Nrxn3α) is indicated by a yellow box. The serine residue that is predicted to form hydrogen bonds with A687T and A687S in bovine Nrxn1α (S878) is conserved in mouse Nrxn3α (S843; indicated by a green box). C, Ribbon diagram of the Nrxn1 LNS3–EGF-B–LNS4 reelin repeat module. A723, which is analogous to the A687 site, is highlighted in red (dashed box). Green, LNS domains; yellow, EGF-B. D, Molecular modeling of A687 substitutions identifies predicted intramolecular interactions following substitutions with threonine, valine, and serine. The threonine substitution was predicted to induce a putative hydrogen bond (3.4 Å; dotted red line) with neighboring serine 878. The valine substitution was predicted to induce two putative steric clashes with the alpha carbon and carboxyl carbon backbone of tyrosine 714 (3.1 Å and 2.9 Å blue dotted line). The serine side chain was predicted to form a hydrogen bond with serine 878 (3.2 Å).
We utilized molecular modeling to predict if the A687T and A687S substitutions might have the capacity to alter intramolecular interactions within the LNS3–EGF-B–LNS4 reelin repeat. The extracellular sequences of Nrxn3α has not been crystalized; however, structures of bovine Nrxn1α were solved (Chen et al., 2011; Miller et al., 2011). The LNS3–EGF-B–LNS4 modules of bovine Nrxn1α and mouse Nrxn3α share high primary sequence homology (78.8%) and sequence identity (90.1%; Fig. 10A). Further, the sequences of EGF-B and LNS4 that surround A723 of Nrxn1α and A687 of Nrxn3α share sequency identity (80.4%) and similarity (89.6%); thus, we used a published crystal structure of bovine Nrxn1α (PDB: 3QCW) to model the impact of single amino acid substitutions at position Nrxn3αA687 at the corresponding site of Nrxn1α [A723; Fig. 10B (Nrxn1αA723 and Nrxn3αA687 highlighted in orange) and 10C; Chen et al., 2011]. We mutated this site in the Nrxn1α structure using a Dunbrack backbone-dependent rotamer library, curated from experimentally solved protein structures, and identified the side-chain rotamer with the highest probability of occurrence (Fig. 10D; Meng et al., 2023). First, the WT alanine residue did not engage in intramolecular hydrogen bonds nor introduce steric clashes with neighboring amino acids (Fig. 10D, summarized in Table 1). We next modeled a threonine at the Nrxn1αA723 site and observed the formation of a predicted hydrogen bond with serine 878 (Fig. 10D, summarized in Table 1). Similarly, the formation of a hydrogen bond with serine 878 was also predicted following the substitution of serine at Nrxn1αA723 (Fig. 10D, summarized in Table 1). Neither the threonine nor serine substitutions were predicted to introduce intramolecular steric clashes. Importantly, S878, which forms the predicted hydrogen bond with the substituted threonine and serine residues in Nrxn1α, is conserved in Nrxn3α (Fig. 10B, green box). In contrast, when we modeled valine at Nrxn1αA723, no hydrogen bonds were predicted, but we observed two predicted steric clashes with the backbone of a neighboring tyrosine residue (Y714; Fig. 10D, summarized in Table 1). Similar to serine 878, Y714 is conserved between bovine Nrxn1α and mouse Nrxn3α (Fig. 10B, yellow box). The computationally predicted differences in the number of predicted hydrogen bonds and steric clashes based on molecular modeling might explain why A687T and A687S drive synaptic phenotypes while A687V does not.
Summary table of the hydrogen bonds and steric clashes predicted from molecular modeling
Discussion
The function of the elaborate extracellular sequences of α-Nrxns have been largely ignored, perhaps due to its size and complexity. Our previous studies revealed that the patient A687T mutation promoted a presynaptic gain-of-function phenotype that is driven by the selective dysregulation of trans-synaptic interactions with LRRTM2 (Restrepo et al., 2019). In the current study, we initially aimed to define the specific side-chain properties of the A687T mutation that are responsible for the gain-of-function phenotypes. We hypothesized that the addition of a polar moiety in this region leads to synaptic dysfunction through the formation of potential ectopic polar side-chain–mediated interactions. We report that A687T and A687S, when expressed in neurons, unexpectedly yielded distinct morphological and functional phenotypes (summarized in Table 2). Consistent with our previous work, molecular replacement with A687T produced a presynaptic gain-of-function including increased vGluT1 cluster size and enhanced mEPSC frequency (Restrepo et al., 2019; Figs. 4, 5). In contrast, A687S replacement did not elicit any marked presynaptic change. We instead observed a striking postsynaptic gain-of-function represented by increased PSD-95 cluster size, greater surface expression of GluA1 (Fig. 3), and selectively enhanced strength of postsynaptic AMPARs (Fig. 4).
Summary of results
Mechanistically, cell aggregation assays revealed that A687S increased trans interactions with NL1ΔB and NL3, with no effect on LRRTM2 binding (Fig. 8). As a secondary approach, the cell-surface–binding assay indicated that Nrxn3αA687S increased binding with LRRTM2 and not NLs (Fig. 9E–G). While the surface binding assay results are seemingly incongruent with the cellular aggregation data, it may be inappropriate to directly compare the two results. The readout in both assays relies on the direct interaction between pairs of proteins in a non-neuronal system; however, it is important to consider the fundamental differences between each assay. First, the cellular aggregation assay relies on both receptor–ligand pair to be expressed and processed in cells and tethered to the cell surface, whereas in the surface binding assay, only one protein is embedded in the membrane while the other is a recombinant soluble protein. Second, cellular aggregation can only occur in the trans configuration. In contrast, the binding of soluble recombinant protein to its target can occur in both cis and trans configurations. Third, the interactions mediated in trans between two cells likely require stable interactions as the mixed populations of cells are rotated in suspension for 30–45 min. In the cell-surface–binding assay, soluble protein is added in saturation to adherent cells for multiple hours and then fixed, which might capture a transient population of interactions. Fourth, the IgG Fc region dimerizes the soluble protein, which may artificially increase binding. While this is reasonable for NLs, which endogenously form constitutive dimers, the Fc-dependent dimerization might have unintended binding consequences for monomeric Nrxns and LRRTMs. In contrast, Nrxn3α and LRRTM2 remain in a monomeric state in the cellular aggregation assay. Despite these differences, both assays provide potential mechanistic insight into the morphological and functional phenotypes observed following A687S molecular replacement. For example, the stable interactions between Nrxn3αA687S and NL1ΔB may underscore the increased size of postsynaptic puncta (Fig. 3B,E) while the A687S-dependent increase in AMPAR-mediated synaptic transmission may be due to enhanced trans-synaptic interactions between Nrxn3αA687S and LRRTM2 (Fig. 4).
The A687 side chain is buried in the interface between EGF-B and LNS4 and participates in the larger reelin repeat module comprising LNS3–EGF-B–LNS4. How can mutations in this buried residue alter trans-synaptic interactions, synapse morphology, and synaptic transmission? Functionally, the reelin repeat has been proposed to function as a suppressor of trans-synaptic interactions. Deletion of the reelin repeat increased interactions with NL2, and, relative to β-Nrxns, the presence of α-Nrxn-specific sequences reduces binding affinity to NL1 and LRRTM2 (Boucard et al., 2005; Kang et al., 2008; de Wit et al., 2009). Thus, the A687T and A687S mutations may relieve the suppressive influence that LNS3–EGF-B–LNS4 exerts on binding to trans-synaptic ligands. When we modeled the equivalent Nrxn3α A687T and A687S mutations in the Nrxn1α crystal structure, we found that neither mutation was predicted to produce steric clashes; however, they were each predicted to engage in a hydrogen bond with S878. In contrast, A687V mutation, which does not produce morphological or functional phenotypes, is not predicted to engage in hydrogen bonding and is critically predicted to produce steric clashes with the backbone of a neighboring residue (Y714). Perhaps the predicted hydrogen bond formed by the hydroxyl moiety of threonine and serine might differentially disrupt the suppressive activity of the reelin repeat to promote contacts with NL1 and LRRTM2.
Previous molecular modeling of Nrxn1α–NL1 suggests that splice-site A of NL1 might contact LNS3 and LNS4 of Nrxn1α (Chen et al., 2011). While the A687S mutation might reduce the suppressor activity of the LNS3–EGF-B–LNS4 module to facilitate interactions with NL1 splice-site A, it was intriguing that we did not observe a similar increase in cellular aggregation with NL2 (Fig. 8). All NL expression constructs used in this study included the splice-site A exons; therefore, other sequences unique to NL1 and NL2 might differentially regulate their ability to interact with Nrxn3αA687S. The N-terminal chimeras revealed that swapping the longer 42 aa variable sequences of NL2 onto NL1 eliminated the enhanced cellular aggregation between Nrxn3αA687S and NL1. The reciprocal swap of the shorter 22 aa sequences of NL1 onto NL2 resulted in enhanced Nrxn3αA687S–NL2 cellular aggregation. Future experiments are warranted to characterize how the LNS3–EGF-B–LNS4 reelin repeat mediates its proposed suppressive activity on NL1 and LRRTM2 binding. Perhaps it is conveyed by the b β4–β5 loop of LNS3, which blocks the Ca2+-binding site and hypervariable surface of LNS4.
The nanoscopic architecture of excitatory synapses is thought to be critical for synaptic function. Proteins critical for presynaptic release and detection of neurotransmitter assemble into nanoclusters that often align across the synapse to form trans-synaptic nanocolumns (MacGillavry et al., 2013; Tang et al., 2016). Similarly, we demonstrated that Nrxn1 and Nrxn3 each form homogenous nanoclusters that are spatially discrete and nonoverlapping at excitatory synapses (Lloyd et al., 2023). We demonstrated that Nrxn3 nanoclusters are centrally enriched at synapses, are in close proximity to the trans-synaptic nanocolumn, and most likely align with postsynaptic LRRTM2. One appealing possibility is that the A687T and A687S mutations alter the nanoscale architecture of Nrxn3α nanoclusters and redistribute Nrxn3α nanoclusters within excitatory synapses to align with different postsynaptic ligands. Alternatively, the nanoscopic organization of Nrxn3α within an excitatory synapse might be established independent of A687 mutations; however, postsynaptic ligands might undergo a nanoscale redistribution. Although molecular replacement is an excellent tool to develop testable hypotheses, this method inherently relies on the overexpression of the replaced Nrxn and restricts the splice identity of Nrxn to a single isoform. The sensitivity of super-resolution imaging will be required for the sophisticated characterization of the impact of the A687T and A687S mutations of endogenous Nrxn3α. Representing an additional technical limitation is the lack of specific antibodies for α- and β-Nrxns (Lloyd et al., 2023) and NL1 (Zhang et al., 2015; Nozawa et al., 2022), which has prevented the imaging of endogenous α-Nrxns and NL1 using diffraction-limited light microscopy and super-resolution imaging modalities.
While the molecular modeling suggests that the A687T and A687S substitutions may engage in putative hydrogen bonds (Fig. 10D), a limitation of this approach is that these mutations were modeled in the crystal structure of bovine Nrxn1α due to the absence of a crystal structure for mouse Nrxn3α. Although the amino acids that comprise the EGF-B–LNS4 region of bovine Nrxn1α and mouse Nrxn3α share ∼80% sequence identity and 90% amino acid similarity, the amino acids that are not conserved may result in unpredicted differences in the structural organization of the EGF-B–LNS4 region of mouse Nrxn3α that is not identical to bovine Nrxn1α. Future experiments using as-yet unavailable crystal structures of Nrxn3α will be necessary to validate the molecular modeling results. Given the growing appreciation that individual Nrxns control distinct aspects of synaptic transmission (Sudhof, 2017), the crystal structures of Nrxn2α and Nrxn3α would provide tremendous predictive insight into how potential differences in protein structure may contribute to individual Nrxn function and how the expanding number of missense mutations in individual α-Nrxns may alter protein function.
Altogether, our study (1) identifies the tolerance of A687 to further elucidate this largely unexplored extracellular sequence of Nrxn3α, (2) provided additional evidence for the role of the interface region between EGF-B and LNS4 in stabilizing interactions with prototypical neurexin ligands, (3) provides evidence about the potential importance of the variable N-terminal sequences of NL1 and NL2 that modulate binding with α-Nrxns, and (4) provides new insight into how distinct Nrxn3α–LRRTM2 and Nrxn3α–NL1 interactions may contribute to excitatory synapse function. While primary hippocampal cultures are an excellent platform to develop and test hypotheses due to the relative ease of testing how molecular manipulations alter synapse morphology and function, it will be important in future experiments to assess how the endogenous expression of disease-relevant Nrxn3α mutants in vivo alter circuit function and animal behavior.
Footnotes
This work was supported by grants from the National Institute of Mental Health (R01MH116901 and R21MH129620 to J.A.), National Creative Research Initiative Program of the Ministry of Science and ICT (2022R1A3B1077206 to J.K.), and an HHMI Gilliam Fellowship for Advanced Study (GT15852 to E.G.S.) at an NIGMS T32 (3T32GM007635 to E.G.S.). We thank the affected individual and their family for generously sharing genome sequencing data and members of the Aoto laboratory for helpful discussions. We thank Dr. Susana Restrepo Upegui for generating the HA-tagged A687V and A687S cDNAs and Thomas Südhof (Stanford University) for the NL2 and LRRTM2 plasmids.
The authors declare no competing financial interests.
- Correspondence should be addressed to Jason Aoto at jason.aoto{at}cuanschutz.edu.
