The following constructs were used for cell-based assays and have previously been described (Tanabe et al., 2017): extracellularly HA-tagged mouse IgSF21 (IgSF21-HA), mouse Nrxn2α LNS1 to LNS3 domains subcloned into pDisplay (aa 29-680; pDisplay-Nrxn2α LNS1-3), soluble Fc-tagged mouse Nrxn2α ectodomain (Nrxn2α-Fc), and extracellularly HA-tagged human CD4 (HA-CD4). Novel constructs were generated for IgSF21 and Nrxn2α mutants by site-directed mutagenesis using IgSF21-HA and pDisplay-Nrxn2α LNS1-3 as templates. The IgSF21-ires-GFP construct was made by subcloning untagged rat IgSF21 into the pCAG-IRES-GFP vector (Lee A. K. et al., 2023). pNICE-NLGN2(+A) (Addgene: Plasmid #15259) was a gift from Peter Scheiffele. All constructs were verified by DNA sequencing.

The following chemicals were used to treat rat hippocampal neurons at the indicated concentrations: cytochalasin D (MedChemExpress, Monmouth Junction, NJ, USA; HY-N6682; 2 μM), nocodazole (MedChemExpress, Monmouth Junction, NJ, USA; HY-13520; 3.3 μM), PKI 14-22 (MedChemExpress, Monmouth Junction, NJ, USA; HY-106377A; 0.1 μM), bisindolylmaleimide I (Sigma-Aldrich; #203290; 4 μM), KN-93 (MedChemExpress, Monmouth Junction, NJ, USA; HY-15465; 5 μM), CaMKII Inhibitor XII (Sigma-Aldrich; #208923; 10 μM), PP2 (Sigma-Aldrich; #529573; 10 μM), SP600125 (MedChemExpress, Monmouth Junction, NJ, USA; HY-12041; 25 μM), and PD98059 (Selleck; #S1177; 50 μM). All compounds were dissolved in dimethyl sulfoxide (DMSO) and added to neurons to reach the indicated concentration and a final concentration of 0.2% DMSO.

All animal experiments were carried out in accordance with Canadian Council on Animal Care guidelines and approved by the Institut de Recherches Cliniques de Montréal (IRCM) Animal Care Committee, the Animal Care Committee of the University of Manitoba and the University of Leeds Animal Welfare and Ethical Review Body. The Nrxn2α knockout (KO) mouse line was made in a previous study by crossing B6;129-Nrxn3^TM1Sud/Nrxn1^TM1Sud/Nrxn2^TM1Sud/J mice (JAX #006377, The Jackson Laboratory, Bar Harbor, ME, USA) with the C57BL/6NCrl strain (Charles River, Margate, UK) (Dachtler et al., 2014). Genotyping by PCR used the following primers: Nrxn2α wild-type (WT) 5′ forward primer 5′-TGAGATGGAGAGCCAGACACCG-3′, Nrxn2α WT 3′ reverse primer 5′-ACAGTGCCATGGACTCAGTAGCC-3′ and Nrxn2α KO mutant 3′ reverse primer 5′-TGCATCGCATTGTCTGAGTAGGTGTC-3′. Expected PCR product sizes are 202 bp (WT) and 315 bp (KO). Mice were group-housed (two to five per cage) and maintained on a 12-h light/dark cycle.

The amino acid sequences coding for mouse IgSF21 (a.a. 25–132; UniProt ID: Q7TNR6) and mouse Nrxn2α (a.a. 29–206; Uniprot ID: E9Q7 × 7) were submitted to the AlphaFold2 server ColabFold (Mirdita et al., 2022). The number of relaxed models was set to five and the template mode was set to pdb100. Among the resulting protein complex predictions, we selected the relaxed model with the highest local structure confidence score given as pLDDT (predicted local-distance difference test). Further structural analysis and representation were performed with Pymol version 2.5.5 (PyMOL). Electrostatic potentials were mapped with the Adaptive Poisson-Boltzmann Solver (APBS) Electrostatics Plugin from Pymol (Dolinsky et al., 2004).

Cultures and transfection of rat or mouse hippocampal neurons were performed essentially as described previously (Kaech and Banker, 2006; Tanabe et al., 2017; Lee A. K. et al., 2023). Briefly, hippocampal neurons from E18 rat embryos were cultured on poly-L-lysine-coated glass coverslips in neurobasal medium (Gibco; #21103-049) supplemented with NeuroCult SM1 (StemCell; #05711) and GlutaMaX (Gibco; #35050061). For Nrxn2α KO mouse neuron cultures, hippocampi from E16 mice embryo littermates derived from the crossing of Nrxn2α heterozygous mice were extracted and stored individually in tubes containing Hibernate E (BrainBits, HE500) for around 4 h until genotyping results were obtained. Meanwhile, genotyping of the mice was carried out as described above to identify WT and Nrxn2α homozygous KO littermates. Hippocampi from littermates of the same genotype were pooled, incubated in 0.25% trypsin-EDTA and dissociated, then the cells were plated onto poly-L-lysine-coated coverslips. For mouse neuron cultures, 25 mg/ml insulin (Invitrogen; #I5506) was added to the supplemented neurobasal medium. For low density cultures, the neurons were cultured on glass coverslips inverted over a feeder layer of rat astrocytes.

Neuronal transfections and immunocytochemistry were performed essentially as described previously (Tanabe et al., 2017). Neuronal transfections were performed at 8 days in vitro (DIV) using a ProFection Mammalian Transfection System kit (Promega; #E1200) kit according to the manufacturer’s instructions. For immunocytochemistry, neurons were fixed in prewarmed 4% (v/v) formaldehyde/4% (w/v) sucrose for 12 min, permeabilized in 0.2% (v/v) Triton X-100 in phosphate-buffered saline (PBS) and blocked with 5% (v/v) normal donkey serum (NDS) and 3% (w/v) bovine serum albumin (BSA) in PBS for 1 h at room temperature. Neurons were incubated overnight at 4°C with the appropriate primary antibodies: anti-vesicular GABA transporter (VGAT) (1:500; rabbit; Synaptic Systems, Goettingen, Germany, #131003), anti-vesicular glutamate transporter 1 (VGLUT1) (1:500; guinea pig; Synaptic Systems, Goettingen, Germany; #135 304), anti-GFP (1:2,000; chicken; Abcam, Cambridge, UK; #ab13970), anti-HA (1:1,000; rat; Roche; #ROAHAHA), anti-acetylated Tubulin (1:1,000; mouse; Sigma-Aldrich; #T6793) and anti-MAP2 (1:2,000; rabbit; Abcam, Cambridge, UK; #5392). Cells were then incubated with highly cross-adsorbed Alexa dye–conjugated secondary antibodies generated in donkey toward the appropriate species (1:500; Jackson ImmunoResearch). To assess the effect of JNK inhibition on the maintenance of native synapses, neurons were treated at 14 DIV with SP600125 or DMSO as a vehicle control for 24 h.

Cocultures of hippocampal neurons with HEK293T cells were set up as previously described (Tanabe et al., 2017; Lee H. et al., 2023). For neuron-HEK293T coculture assays, appropriate extracellularly HA-tagged plasmids were transfected in HEK293T cells using TransIT-LT1 (Mirus Bio. LLC; catalog number: MIR2305) and cultured in DMEM containing 10% (v/v) FBS. 24 h after transfection, cells were harvested by trypsinization and seeded on the neuron cultures. After 24 h of coculture, cells were fixed in prewarmed 4% (v/v) formaldehyde and 4% (w/v) sucrose in PBS for 12 min and blocked with 5% (v/v) NDS and 3% (w/v) BSA in PBS for 1 h at room temperature. Cells were incubated with anti-HA (1:1,000; rat; Roche; #ROAHAHA) overnight at 4°C to label transfected surface-expressed HA-tagged synaptic organizers. Cells were then permeabilized in 0.2% (v/v) Triton X-100 in PBS and incubated with anti-VGAT (1:500; rabbit; Synaptic Systems, Goettingen, Germany, #131003) or anti-VGLUT1 (1:500; guinea pig; Synaptic Systems, Goettingen, Germany; #135 304) overnight at 4°C. Cells were then incubated with highly cross-adsorbed Alexa dye–conjugated secondary antibodies generated in donkey toward the appropriate species (1:500; Jackson ImmunoResearch). For the pharmacological treatments, indicated compounds were added to primary hippocampal neuron cultures (14 DIV) and then 1 h later, HEK293T cells were seeded onto the neuron cultures. To investigate the effect of JNK inhibition on the maintenance of induced synapses, the transfected HEK293T cells were first cocultured with neurons for 17 h, and then the cocultures were treated with SP600125 or DMSO for another 7 h.

To produce soluble Fc-tagged proteins, plasmids encoding mouse IgSF21-Fc, Nrxn2α-Fc or Fc alone were transfected into HEK293T cells using TransIT-PRO (Mirus Bio LLC; #MIR5740) and cultured in Dulbecco’s modified Eagle’s medium (DMEM) (Gibco; #11965118) containing 10% (v/v) fetal bovine serum (FBS) (Wisent; #080-150). 24 h after transfection, the medium was replaced with serum-free AIM V synthetic medium (Gibco; #12055083). The conditioned medium was collected after 5 days. Purified human IgSF21-His (R&D Systems; #10330-S2) was purchased from R&D Systems. Protein binding assays were performed as described previously (Tanabe et al., 2017; Lee H. et al., 2023). Briefly, to assess protein binding on the surface of COS7 cells, the cells were transfected using TransIT-LT1 (Mirus Bio. LLC; #MIR2305) with plasmids encoding the target proteins fused to an extracellular HA-tag and then cultured in DMEM containing 10% (v/v) FBS. 24 h after transfection, the transfected COS7 cells were washed with extracellular solution (ECS) containing 168 mM NaCl, 2.4 mM KCl, 20 mM HEPES, pH 7.4, 10 mM D-glucose, 2 mM CaCl₂, 1.3 mM MgCl₂, and 100 μg/ml bovine serum albumin (BSA; Sigma, #A9647); the cells were then incubated for 1 h at 4°C with the appropriate Fc- or His-fusion proteins. IgSF21-His proteins diluted in ECS/BSA to 70 nM or Nrxn2α-Fc containing HEK293T conditioned medium were applied to COS7 cells. Cells were washed with ECS and subsequently fixed in prewarmed 4% (v/v) formaldehyde/4% (w/v) sucrose for 12 min and blocked in 5% (v/v) NDS and 3% (w/v) BSA in PBS for 1 h at room temperature. COS7 cells were incubated with anti-HA (1:1,000; rat; Roche; #ROAHAHA) and anti-6xHis (1:1,000; rabbit; Abcam, Cambridge, UK, catalog number: ab213204) without permeabilization overnight at 4°C. Cells were then incubated with highly cross-adsorbed Alexa dye–conjugated secondary antibodies generated in donkey toward the appropriate species (1:500; Jackson ImmunoResearch).

Images were acquired on a Leica DM6 fluorescence microscope with 40 × 1.25 or 60 × 1.40 numerical aperture oil objectives and a Hamamatsu C11440 ORCA-Flash 4.0 camera using LasX software (Leica). Images were acquired as 16-bit grayscale. For quantification, sets of cells were stained simultaneously and imaged with identical settings.

To quantify binding levels and cell surface expression levels, we measured the average intensity of each channel within the delineated COS7 or HEK293T cell area subtracted by the average intensity of the off-cell background. For in situ binding assays, the average intensity of bound soluble Fc- or His-proteins was normalized using the average surface intensity of the HA-tagged protein signal. COS7 cells expressing similar levels of HA-tagged proteins were selected to quantify bound soluble Fc- or His-proteins. Analyses were performed using Volocity 6.0, Excel for Microsoft 365 (Microsoft), and Prism 9 (GraphPad Software, Inc.). For the artificial synapse formation assays, HEK293T cells displaying similar surface levels of HA-tagged proteins were imaged without considering the other fluorescence channels. To assess GABAergic or glutamatergic presynaptic differentiation, the fluorescence channel corresponding to VGAT or VGLUT1 was thresholded, and the total thresholded intensity within regions positive for surface HA was measured. Analysis was performed using Metamorph 7.8 (Molecular Devices), Excel 2003, and Prism 9. For analysis in low-density neuron experiments, neurons were chosen randomly based on healthy morphology and, for transfected neurons, expression level. Images for each synaptic marker were thresholded to extract the clusters. To quantify the number of puncta, a segment of secondary dendrite per neuron or whole soma were chosen randomly based on healthy morphology and expression level, and the number of clusters per dendrite length was measured. For immunoblot quantification, band signal intensity was measured using ImageJ software and normalized to the β-actin signal or total protein (for phosphorylated targets) intensities as indicated.

Surface biotinylation assays were performed according to the kit’s instructions (Thermo Scientific, Waltham, MA, USA; #A44390). Briefly, rat hippocampal neurons were washed once with PBS and then incubated with 0.25 mg/ml EZ-Link Sulfo-NHS-SS-Biotin in PBS at room temperature for 10 min. The remaining active biotin was quenched by washing three times with ice-cold TBS. After homogenization in lysis buffer supplemented with a cocktail of protease inhibitors (Roche; #CO-RO), the biotinylated proteins were isolated using NeutrAvidin-conjugated agarose beads, resolved by SDS-PAGE, and immunoblotted with the corresponding antibodies. For pharmacological inhibition of the JNK pathway, neurons were treated at 14 DIV with SP600125 or DMSO as vehicle and surface biotinylation assays were performed 24 h post-treatment.

Synaptic fractionation was carried out essentially as previously described (Carlin et al., 1980). All steps were performed on ice or at 4°C. Briefly, forebrains from 1 to 2-month-old rats or cortices from 2-month-old Nrxn2α KO and WT littermate mice were homogenized using a Douncer in homogenization buffer [0.32 M sucrose, 4 mM HEPES pH 7.4, 1 mM MgCl₂, supplemented with a cocktail of protease (Roche; #CO-RO) and phosphatase inhibitors (Roche; #PHOSS-RO) (homogenate fraction)]. The homogenate was centrifuged at 1,500 × g for 15 min to collect the supernatant (postnuclear fraction). The supernatant was further centrifuged at 18,000 × g for 20 min and the resulting supernatant (cytosol fraction) and pellet (crude synaptosome fraction) were collected. The pellet was resuspended in homogenization buffer and loaded on a discontinuous sucrose gradient (0.85 M/1.0 M/1.2 M) and centrifuged at 78,000 × g for 2 h. The material at the 1.0 M/1.2 M interface was carefully collected (synaptosome fraction). Triton X-100 was added to 0.5% (v/v) and the synaptosome samples were incubated at 4°C on a rotating platform for 20 min before being centrifuged at 32,000 × g for 2 h. The supernatant was collected (Triton-soluble synaptosome fraction) and the pellet was resuspended in homogenization buffer, loaded on a discontinuous sucrose gradient (1.0 M/1.5 M/2.0 M) and centrifuged at 170,000 × g for 2 h. The material at the 1.5 M/2.0 M interface was carefully collected. Triton X-100 was added to 0.5% (v/v) and incubated at 4°C on a rotating platform for 10 min before being centrifuged at 100,000 × g for 20 min. The resulting pellet was resuspended in homogenization buffer (Triton-insoluble synaptosome fraction). Proteins were resolved by SDS-PAGE and immunoblotted with appropriate antibodies.

Surface plasma resonance (SPR) analyses were performed using a Biacore T200 instrument (GE Healthcare, Chicago, IL, USA). Recombinant human His-tagged IgSF21 protein (R&D Systems; #10330-S2) was immobilized on carboxymethylated dextran CM5 sensor chips (Cytiva) using an amine-coupling strategy. Briefly, the sensor chip surface was activated with a 1:1 mixture of N-hydroxysuccinimide and 3-(N,N-dimethylamino)-propyl-N-ethylcarbodiimide. IgSF21 protein solution (solubilized in acetate buffer, pH 5.5) was injected at a flow rate of 20 μl/min in PBS running buffer (PBS 1X, pH 7.4, 0.025% [v/v] Tween-20) to reach a level of immobilization of 500 relative units (RU) on the CM5 sensor chip. After the immobilization step, surfaces (protein and reference) were deactivated by the injection of an ethanolamine solution to prevent further chemical reaction. Binding kinetics of recombinant human Fc-tagged NRXN2α (analyte) over the IgSF21 (target) sensor chip was evaluated in PBS running buffer (PBS 1X, pH 7.4, 0.025% [v/v] Tween-20), using a concentration series ranging from 0 to 200 nM analyte. All tests were performed at 25°C using a flow rate of 20 μl/min with a 180 sec association step (analyte) and a 240 sec dissociation step (PBS running buffer). Sensor chip surfaces were regenerated by injecting 15 μl of a 10 mM Glycine pH 3.0 solution at a flow rate of 30 μl/min. Binding sensorgrams were obtained by subtracting the signal of the reference flow cell (no protein target coupled to the sensor) from the signal obtained from the ligand flow cell (protein target coupled to the sensor). Data analysis was performed using the BIA Evaluation Software (GE Healthcare, Chicago, IL, USA). The data were fitted to a one-site Langmuir binding model using kinetic analysis and a global or local fit.

Crude synaptosomal fractions were prepared essentially as described (Siddiqui et al., 2013) with some modifications. Briefly, following anesthesia of mice using isoflurane, brains were rapidly removed, rinsed with ice-cold PBS, and homogenized in homogenization buffer (320 mM sucrose, 4 mM HEPES-NaOH, pH 7.3, and protease inhibitors) in a glass Teflon homogenizer (nine strokes, 900 rpm). The homogenate was centrifuged for 10 min at 1,000 g to remove nuclei. The resulting pellet P1 was discarded. The supernatant S1 was centrifuged for 15 min at 12,000 g. The resulting pellet was resuspended in homogenization buffer and further centrifuged for 15 min at 11,000 g. The crude synaptosomal fraction was resuspended in lysis buffer (Complexiolyte-47 or -48 buffer, Logopharm) containing protease inhibitors, incubated for 1 h at 4°C and centrifuged at 14,000 g for 20 min to separate the insoluble fraction. The solubilization efficiency of synaptosomal lysates was approximately 95%. The supernatant (1.5 mg/ml protein concentration) was incubated with anti-IgSF21 (Proteintech, Chicago, IL, USA; #21465-1-AP) antibodies that were pre-conjugated to Protein G Sepharose beads for at least 12 h at 4°C. After overnight incubation of lysates with antibody-conjugated beads at 4°C, the beads were washed 4 times with Complexiolyte-47 or -48 dilution buffer and boiled in 2X Laemmli buffer (Bio-Rad, Hercules, CA, USA). To avoid detection of multiple bands due to numerous heparan sulfate modifications of Nrxns, lysates of crude synaptosomal fractions or immunoprecipitates bound to protein-G Sepharose (GE Healthcare, Chicago, IL, USA) were further incubated with a cocktail of 1 U/ml each of Heparinases I, II, III for 2 h at 37°C prior to boiling in 2X Laemmli buffer.

For pull-down assays with highly purified proteins, 3 μg of human Nrxn2α-Fc (R&D Systems; #6636-NX), Nrxn1β-Fc (R&D Systems; #5268-NX), or Fc protein alone were incubated with 2.5 μg human IgSF21-His (R&D Systems; #10330-S2) in 500 μl of binding buffer (10 mM HEPES pH 7.4, 150 mM NaCl, 2 mM CaCl₂, 1 mM MgCl₂ and 0.1% Tween-20) for 1 h at room temperature. Then, 50 μl of Protein-G magnetic bead slurry (Dynabeads; Invitrogen; #10004D) was added and the samples were incubated for 1 h at 4°C on a rotating platform. Protein complexes pulled down by the Protein-G beads were purified by magnetic separation (DynaMag-2 Magnet; Invitrogen; #12321D) by washing the Protein-G beads once with binding buffer followed by four washes with PBS. Samples containing protein complexes attached to the beads were boiled in 2x Laemmli buffer containing β-mercaptoethanol, resolved by SDS-PAGE, and immunoblotted with the corresponding antibodies.

For pull-down assays coupled to mass spectrometry analysis, crude synaptosome fractions were prepared as previously described (Condomitti et al., 2018). Briefly, brains from 1-month old mice were homogenized using a Douncer with homogenization buffer [0.32 M sucrose, 4 mM HEPES pH 7.4, 1 mM MgCl₂, supplemented with a cocktail of protease inhibitors (Roche; #CO-RO)]. The homogenate was centrifuged at 1,500 × g for 15 min and the supernatant was further centrifuged at 18,000 × g for 20 min. The pellet (crude synaptosome fraction) was resuspended in homogenization buffer containing 1.5% CHAPS detergent and incubated at 4°C for 2 h on a rotating platform before being centrifuged at 20,000 × g for 30 min. Protein-G magnetic beads (Dynabeads; Invitrogen; #10004D) were coated with IgSF21-Fc or control Fc proteins. Coated beads were added to the extracted crude synaptosome fractions, and the samples were incubated at 4°C overnight on a rotating platform. Bound proteins were purified by magnetic separation (DynaMag-2 Magnet; Invitrogen; #12321D) and washed three times with homogenization buffer containing 1.5% CHAPS detergent and five times with 50 mM ammonium bicarbonate (AmBic).

Protein samples were run on 8 to 12% polyacrylamide gels that were then wet-transferred onto 0.2 μm PVDF membranes (Bio-Rad, Hercules, CA, USA; #1620177). Membranes were blocked in 5% BSA diluted in tris-buffered saline with Tween-20 (TBST) (0.1% Tween-20) or 5% skim milk in TBST (0.001% Tween-20) and incubated with one of the following primary antibodies: anti-IgSF21 (1:1,000 to 1:2,000; rabbit; Proteintech, Chicago, IL, USA; #21465-1-AP), anti-PSD-95 family (1:2,000; mouse IgG2a; clone 6G6-1C9; Thermo Scientific, Waltham, MA, USA; MA1-045), anti-synaptophysin (1:10,000; mouse IgG1; clone SY38; Millipore, Kankakee, IL, USA; MAB5258), anti-Nrxn1/2/3 (1:1,000; rabbit; Millipore, Kankakee, IL, USA; #ABN161-I), anti-Nrxn1/2/3 (1:1,000; rabbit; Synaptic Systems, Goettingen, Germany; #175 003), anti-His (1:1,000; rabbit; Abcam, Cambridge, UK; #ab9108), anti-CaMKIIα [1:2,000; rabbit; Cell Signaling Technology (CST), Danvers, MA, USA; #11945], anti-CaMKIIδ (1:1,000; rabbit; LSBio, Seattle, WA, USA; #LS-C329304), anti-Src (1:1,000; mouse; Millipore, Kankakee, IL, USA; #05-184-I), anti-Fyn (1:1,000; mouse; Santa Cruz Biotechnology, Dallas, TX, USA; sc-434); anti-JNK (1:1,000; rabbit; CST; #9252), anti-phosphorylated JNK (1:1,000; rabbit; CST; #4668) and anti-β-actin (1:2,000; rabbit; Abcam, Cambridge, UK; #ab8227). Membranes were washed with TBST and incubated with horseradish peroxidase-conjugated secondary antibodies recognizing the appropriate species (1:10,000; raised in donkey, Jackson ImmunoResearch or raised in goat, Southern Biotech) or a conformation-specific mouse anti-rabbit IgG (L27A9 mAb, CST). This conformation-specific secondary antibody that does not bind to reduced and denatured IgGs was used to prevent masking of the various Nrxn isoforms by the IgG heavy and light chains of the antibody used in the co-immunoprecipitation experiments (the antibodies used in immunoprecipitation and immunoblotting are from the same host species). Signals were developed using Clarity Western ECL Substrate (Bio-Rad, Hercules, CA, USA; #1705061) or Clarity Max Western ECL Substrate (Bio-Rad, Hercules, CA, USA; #1705060) and captured by Image Lab software (Bio-Rad, Hercules, CA, USA).

The on-bead pulled down proteins were first incubated in 4 M urea with 50 mM AmBic for 10 min and then diluted in 2 M Urea with 50 mM AmBic for a tryptic digestion performed overnight at 37°C with agitation using 0.3 μg of Sequencing Grade Modified Trypsin (Promega; #V5111). The samples were then reduced with 13 mM dithiothreitol at 37°C for 30 min and, after cooling for 10 min, alkylated with 23 mM chloroacetamide at room temperature for 20 min in the dark. The supernatants were then acidified with trifluoroacetic acid for desalting and removal of residual detergents using an Oasis MCX 96-well Elution Plate (Waters; #186001830BA) following the manufacturer’s instructions. After elution in 90% methanol with 10% ammonium hydroxide (v/v), samples were dried with a Speed-vac, reconstituted under agitation for 15 min in 12 μL of 2% acetonitrile and 1% formic acid and loaded onto a 75 μm i.d. × 150 mm Self-Pack C18 column installed in an Easy-nLC II system (Proxeon Biosystems, Odense, Denmark). Peptides were eluted with a two-slope gradient at a flowrate of 280 nl/min. The buffers used for chromatography were 0.2% formic acid in water (solvent A) and 0.2% formic acid in acetonitrile (solvent B). Solvent B was first increased from 2 to 37% over 90 min and then from 37 to 80% over 10 min. The HPLC system was coupled to an Orbitrap Fusion mass spectrometer (Thermo Scientific, Waltham, MA, USA) through a Nanospray Flex Ion Source. Nanospray and S-lens voltages were set to 1.3-1.7 kV and 60 V, respectively. The capillary temperature was set to 250°C. Full scan MS survey spectra (m/z 360–1,560) in profile mode were acquired in the Orbitrap with a resolution of 120,000 with a target value of 3 × 10⁵. The 25 most intense peptide ions were fragmented in an HCD collision cell and analyzed in the linear ion trap with a target value of 2 × 10⁴ and a normalized collision energy at 29 V. Target ions selected for fragmentation were dynamically excluded for 15 sec after two MS2 events.

Liquid chromatography-tandem mass spectrometry (LC-MS/MS) raw files were analyzed with the Mascot search engine using the Mus musculus RefSeq database (version 20151112) supplemented with “common contaminants” from the Max Planck Institute¹, the global proteome machine (GPM)² and decoy sequences. The search parameters were set with trypsin specificity (two missed cleavage sites allowed), and variable modifications involving oxidation (M), deamidation (NQ) and carbamidomethyl (C) were set as fixed modifications. The mass tolerances for precursor and fragment ions were set to 10 ppm and 0.6 Da, respectively, and peptide charges of +2,+3,+4 were considered. Search results were individually processed by PeptideProphet (Keller et al., 2002), and peptides were assembled into proteins using parsimony rules as previously described (Nesvizhskii et al., 2003) by using the trans-proteomic pipeline (TPP). TPP settings were the following: -p 0.05 -x20 -PPM –d “DECOY,” iProphet option pPRIME and PeptideProphet pP. Only proteins having at least one unique peptide and an iProphet probability ≥ 0.9 were considered.

To estimate the interaction statistics, proteins having iProphet protein probability ≥ 0.9 and unique peptides ≥ 1 were submitted to SAINTexpress (version 3.6.1) (Teo et al., 2014) without control or bait compression. Potential hit proteins identified by the pulldown using IgSF21-Fc as bait were compared against pulldowns using the Fc control antibody, and interactions displaying a Bayesian false discovery rate (BFDR) ≤ 0.03 were considered as statistically significant.

Agostino’s K-squared test was used to assess normal distributions, and Bartlett’s test was used to check whether standard deviations (SDs) were significantly different across groups. To compare two groups showing normal distribution with equal variance and unequal variance, unpaired Student’s t-tests and Welch’s t-tests were used, respectively. To compare two groups without normal distribution, Mann–Whitney tests (non-parametric t-tests) were used. For statistical tests comparing to a hypothetical constant value, calculation of the 95% confidence interval (CI) of the mean was used. To compare three or more groups with normal distribution, Welch’s ANOVA with Dunnett’s T3 post hoc analysis was used. For comparisons between three or more groups without normal distribution, Kruskal–Wallis tests with Dunn’s post hoc analysis were used. To assess statistical differences in cumulative distribution of two groups, Kolmogorov–Smirnov tests were used. Statistical significance was examined with appropriate tests as indicated in the figure legends. All data are reported as the mean ± SEM from at least three independent experiments, and statistical significance was defined as p < 0.05.

Through in situ binding assays using cell lines, we have previously shown that IgSF21 binds to Nrxn2α, but not to other Nrxn isoforms, e.g., Nrxn1α, Nrxn3α, or the β-Nrxns (Tanabe et al., 2017). To confirm the binding specificity of IgSF21 to Nrxn2α in a more physiologically relevant context, we used IgSF21-Fc protein as bait in pull-down assays using mouse forebrain synaptosome samples (Figure 1A). Proteins pulled down by IgSF21-Fc were resolved by LS-MS/MS, and we unambiguously identified Nrxn2α as the top binding partner for IgSF21 (Figure 1A). These data indicate that IgSF21 has the ability to bind Nrxn2α in the brain.

Next, to address whether endogenous IgSF21 and Nrxn2α form protein complexes in the brain, we performed co-immunoprecipitation (co-IP) assays on mouse brain synaptosomes using an IgSF21 antibody for immunoprecipitation and a pan-Nrxn antibody for immunoblotting (IB) (Figure 1B). A band with an apparent molecular weight of about 200 kDa was detected in the samples immunoprecipitated by the IgSF21 antibody, but not those immunoprecipitated by rabbit IgG (Figure 1B). Together with the specificity of IgSF21 for Nrxn2α that we identified using cell-based studies (Tanabe et al., 2017), the results of the MS-coupled pull-down assay (Figure 1A) and the co-IP experiment (Figure 1B) support that native IgSF21-Nrxn2α complexes are present in mouse brain synapses.

To test whether IgSF21 and Nrxn2α engage in direct protein interaction, we performed pulldown assays using highly purified recombinant proteins. Nrxn2α-Fc, but not two negative control Fc proteins (Nrxn1β-Fc and Fc alone), pulled down IgSF21-His, revealing a direct protein interaction (Figure 1C). To gain insight into the affinity and dynamics of the IgSF21-Nrxn2α complex, we proceeded to test the binding of purified IgSF21 immobilized on a biosensor chip to soluble Nrxn2α ectodomain (0–200 nM) using surface plasma resonance (SPR). The results fit a model of bimolecular (1:1) association between an IgSF21 monomer and an Nrxn2α monomer (Figure 1D). Soluble Nrxn2α bound immobilized IgSF21 with low nanomolar concentration (K_D = 5.7 nM). Thus, Nrxn2α forms a direct, high affinity interaction with IgSF21.

IgSF21 has been shown to induce GABAergic, but not glutamatergic, presynaptic differentiation in artificial synapse formation assays. To test whether endogenous presynaptic Nrxn2α is the functional receptor for IgSF21 and is indispensable for the latter’s synaptogenic activity, we performed artificial synapse formation assays using primary hippocampal neuron cultures derived from Nrxn2α KO and WT mouse embryos (Figure 1E). IgSF21 expressed on the surface of HEK293T cells was only able to induce the accumulation of VGAT, an inhibitory presynaptic vesicle protein, on contacting axons of WT neurons and not on those of Nrxn2α KO neurons (Figure 1G and Supplementary Figure 1A). These results indicate that Nrxn2α is indispensable for the synaptogenic activity of IgSF21 and further suggest that neither other Nrxns nor any other cell surface molecules expressed on Nrxn2α KO neurons can compensate for the loss of Nrxn2α. In contrast, Nlgn2-expressing HEK293T cells induced VGAT accumulation in both WT neurons and Nrxn2α KO neurons at an equivalent level (Figures 1F, H and Supplementary Figure 1B). This is consistent with the evidence that Nrxn2α is expressed at lower abundance compared to Nrxn1α and Nrxn3α and that Nlgn2 can interact with α- and β-Nrxn1/2/3 to induce presynaptic differentiation (Treutlein et al., 2014; Schreiner et al., 2015).

In our model, IgSF21 acts as a postsynaptic organizer interacting in a trans manner with presynaptic Nrxn2α. To verify that IgSF21 is indeed postsynaptically expressed, we assessed its expression in subcellular fractions from rat forebrains probing the Triton-insoluble synaptosome fraction, which corresponds to the postsynaptic fraction containing PSD-95, and the Triton-soluble fraction, which contains presynaptic components including synaptophysin (Figure 1I). While there are two alternative splicing isoforms of IgSF21, a longer one (IgSF21 L) and a shorter one (IgSF21 S) (Tanabe et al., 2017), both isoforms of IgSF21 were enriched in the Triton-insoluble synaptosome fraction but absent from the Triton-soluble fraction (Figure 1I). Taken together, our results indicate that endogenous presynaptic Nrxn2α acts as a functional receptor for postsynaptic IgSF21 to mediate IgSF21-induced inhibitory presynaptic differentiation through a direct high-affinity protein interaction.

So far, the tertiary structure of the IgSF21 protein or the IgSF21-Nrxn2α protein complex has not been determined. The AlphaFold Protein Structure Database (Identifier: AF-Q7TNR6-F1) and our previous study (Tanabe et al., 2017) have suggested that IgSF21 L has three immunoglobulin-like domains (Ig1-3), whereas IgSF21 S possesses only Ig1 and Ig3. According to our previous mutagenesis analysis (Tanabe et al., 2017), the IgSF21 Ig1 domain, which is common between IgSF21 L and IgSF21 S, is necessary and sufficient for Nrxn2α binding. In addition, the first laminin, neurexin and sex hormone-binding globulin-like domain (LNS1) of Nrxn2α is necessary and sufficient for IgSF21 binding. Because many Ig, Ig-like, and LNS domains exist in the Protein Databank (PDB), we chose to use AlphaFold2 (Jumper et al., 2021) to predict the tertiary structure of the IgSF21 Ig1-Nrxn2α LNS1 binding interface (Figure 2A). Five ranks were chosen for AlphaFold2 modeling of the binding interface (Ig1/LNS1) to determine the best prediction (Supplementary Figures 2A–C). Predicted local-distance difference test (pLDDT) scores were calculated for each residue of the polypeptide chains. In addition, the predicted alignment errors (PAEs) were measured between residues in the model. Rank 1 was chosen for further analysis because it contained the best chain (averaged) pLDDT and PAE scores from the 5 ranks. Our prediction modeled a binding interface in which the β-sheets (R⁵³ to D⁷⁶) of the IgSF21 Ig1 domain faced two helical protrusions (L¹⁶¹ to Y¹⁷¹) found within the loop between the 11 and 12th β-sheets (the β11-β12 loop, composed of residues from G¹⁵³ to F¹⁷⁵) of the Nrxn2α LNS1 domain (Figures 2B, C). A 4 Å cut-off value was used to assess which intermolecular forces (IMFs) could maintain the binding interface (Figures 2B, C) between IgSF21 Ig1 and Nrxn2α LNS1 in the AlphaFold2 model. A combination of many hydrogen bonds and a few hydrophobic interactions were predicted to mediate the association between IgSF21 Ig1 and Nrxn2α LNS1 (Figure 2D). For instance, IgSF21 Ig1 was predicted to utilize side chains (R⁵⁵, E⁵⁶, Y⁶⁰, K⁷¹, D⁷⁶, H¹¹⁷) to pack against Nrxn2α LNS1 through hydrogen bonding. On the other hand, Nrxn2α LNS1 was predicted to use side chains (R¹⁶⁰, L¹⁶¹, S¹⁶², S¹⁶⁷, Y¹⁷¹) and elements of the polypeptide backbone (P¹⁵⁷, V¹⁵⁹, L¹⁶¹, S¹⁶², L¹⁶⁴) to pack against IgSF21 Ig1 via hydrogen bonding. Hydrophobic interactions between Ig1 (V⁵⁸, I⁶⁸) and LNS1 (P⁹⁸, L¹⁶¹) were also possible from our predicted binding interface as listed in Figure 2D. Our binding interface model predicted more residue contacts through hydrogen bonding than hydrophobic clustering. Therefore, we concluded that perturbing the hydrogen bonding network of the interface would be most likely to destabilize the IgSF21 Ig1-Nrxn2α LNS1 complex.

Next, we performed cell-based validation of the IgSF21 Ig1:Nrxn2α LNS1 model predicted by AlphaFold2 (Figures 3, 4) using site-directed mutagenesis to target key residues at the interface between the two domains and disrupt the binding of the two proteins (Figure 2D and Supplementary Figure 2D). Given the exploratory nature of the IgSF21 Ig1:Nrxn2α LNS1 model, we leveraged single mutations that could have a broad impact on the interface such as charge reversal (IgSF21 Ig1 R⁵⁵D, E⁵⁶R, and D⁷⁶K) or charge removal (IgSF21 Ig1 D⁷⁶N) as well as a triple mutant (Nrxn2α LNS1 R¹⁶⁰A/L¹⁶¹A/S¹⁶²A), which would be expected to impact both electrostatic interactions and hydrogen bonds. Because charge reversal mutagenesis has the potential to disrupt overall protein structure, we also implemented conservative replacements (e.g., IgSF21 R⁵⁵K, E⁵⁶D, and D⁷⁶E) which changed the shape and/or size of the side chain but maintained the charged state. To additionally engineer an impactful mutant that would be expected to be well-tolerated by the protein fold, we introduced an N-linked glycosylation site on the surface of IgSF21 Ig1 (consensus site N-X-S/T, where X is any residue except proline) in a double mutant, IgSF21 E⁵⁶N/V⁵⁸T. As negative controls, we selected two residues distant from the predicted binding interface (K⁴⁵ and R¹⁰²) for charge reversal mutagenesis (IgSF21 K⁴⁵D and R¹⁰²D) and conservative replacement (IgSF21 K⁴⁵R and R¹⁰²K).

We first tested whether and how membrane-bound HA-tagged IgSF21 WT and mutants bound soluble Nrxn2α-Fc ectodomain in in situ cell surface binding assays. All of the IgSF21 molecules harboring mutations at the predicted binding interface failed to interact with Nrxn2α (Figures 3A–C). Importantly, surface expression of these IgSF21 mutants was similar to that of WT, suggesting that the lack of binding of Nrxn2α on the cell surface was unlikely to be due to a lack of surface expression of the IgSF21 mutants. Intriguingly, the very conservative mutations in IgSF21 Ig1 (R⁵⁵K, E⁵⁶D, and D⁷⁶E) were as effective at disrupting binding to Nrxn2α-Fc as the more drastic mutations (R⁵⁵D, E⁵⁶R, D⁷⁶K, and D⁷⁶N), indicating the very precise nature of the packing interface. In contrast, conservative replacement of the IgSF21 residues distant to the binding site (i.e., K⁴⁵R and R¹⁰²K) supported the predicted binding model because these replacements did not significantly affect Nrxn2α-Fc binding (Figures 3A–C). On the other hand, charge reversal substitution of these distant IgSF21 Ig1 residues (K⁴⁵D and R¹⁰²D) did diminish Nrxn2α-Fc binding; however, they also significantly decreased cell surface expression, suggesting that they additionally disrupted protein folding and/or protein trafficking. Our in-depth analysis of the AlphaFold2 model indicated that IgSF21 Ig1 K⁴⁵D and R¹⁰²D could indeed introduce unfavorable steric clashes with nearby residues (data not shown) which in turn could negatively impact their trafficking to the cell surface (Figures 3A, C). In a complementary set of binding assays, we tested whether membrane-bound mutated Nrxn2α harboring three alanine substitutions at the predicted binding interface (Nrxn2α LNS1 R¹⁶⁰A/L¹⁶¹A/S¹⁶²A) could bind to soluble IgSF21-His protein. For this test, we introduced mutations into extracellularly HA-tagged Nrxn2α encoding just the LNS1 to LNS3 domains (Nrxn2α LNS1-3) which our previous studies demonstrated are sufficient to bind to soluble IgSF21 ectodomain (Tanabe et al., 2017). Indeed, cells expressing Nrxn2α LNS1-3 WT showed significant binding of soluble IgSF21-His, while those expressing Nrxn2α LNS1-3 R¹⁶⁰A/L¹⁶¹A/S¹⁶²A construct failed to bind IgSF21-His as did Nrxn1β, whose only LNS domain is one equivalent to LNS6 and thus lacks the binding site for IgSF21-His (Figures 3D–F).

Given that IgSF21 induces presynaptic differentiation through Nrxn2α (Figures 1G, 1I), we next tested whether IgSF21 mutants that failed to bind to Nrxn2α also had impaired synaptogenic activity in artificial synapse formation assays (Figure 4). In these assays, HEK293T cells expressing either IgSF21 WT or a mutant from our panel of conservative mutants were cocultured with rat primary cultured hippocampal neurons. As expected, IgSF21 Ig1 forms introducing a glycosylation site (E⁵⁶N/V⁵⁸T) or carrying conservative mutations (R⁵⁵K, E⁵⁶D, and D⁷⁶E) at the predicted binding interface failed to induce VGAT accumulation, while those with conservative substitutions at distant residues (K⁴⁵R and R¹⁰²K) induced VGAT accumulation at an equivalent level to the WT construct (Figure 4A–C). Altogether, our results show that residues predicted to form the interface between IgSF21 Ig1 and Nrxn2α LNS1 are crucial for the interaction between these two proteins and consequently for the ability of IgSF21 to induce GABAergic presynaptic differentiation.

Previous in vivo KO mouse studies have suggested that IgSF21 may preferentially regulate dendritic synaptic inhibition, whereas Nlgn2 predominantly governs somatic inhibition (Poulopoulos et al., 2009; Tanabe et al., 2017). To test whether such subcellular-compartment specificity is also maintained in an in vitro gain of function paradigm, we overexpressed IgSF21 and Nlgn2 in primary rat cultured hippocampal neurons and assessed pre- and postsynaptic marker immunoreactivity. IgSF21-IRES-GFP or IRES-GFP (a negative control) were transfected at 7 DIV, and immunostaining for VGAT and VGLUT1, a presynaptic marker of excitatory synapses was performed at 21 DIV. IgSF21 overexpression led to a significant increase in VGAT immunoreactivity on dendrites but not on somatic regions (Figures 5A, C, E, F). Furthermore, consistent with previous artificial synapse formation assays (Tanabe et al., 2017), IgSF21 overexpression had no effect on VGLUT1 immunoreactivity on dendrites (Figures 5B, D). In contrast, Nlgn2 overexpression led to a significant increase in VGAT immunoreactivity on both dendritic and somatic regions (Figures 5G–J). These results demonstrate that primary cultured neurons maintain molecular mechanisms that govern the subcellular-compartmental specificity of inhibitory presynaptic organization.

To test whether the specific effect of IgSF21 on dendritic, but not somatic, VGAT accumulation is due to selective dendritic expression and/or localization of IgSF21, we immunostained for IgSF21 in the transfected neurons. Interestingly, IgSF21 immunoreactivity was observed in both somatic and dendritic regions of neurons transfected with IgSF21-IRES-GFP (Figure 5K). In the neurons transfected with IRES-GFP, IgSF21 immunoreactivity was hardly detected, presumably because the IgSF21 antibody has insufficient sensitivity to detect endogenous IgSF21 (Figure 5K). These results suggest that preferential regulation of dendritic inhibition by IgSF21 is unlikely to be due to selective trafficking of IgSF21 exclusively to dendrites.

The distinct phenotypes in the above overexpression experiments between the two Nrxn ligands, IgSF21 and Nlgn2, could result from interneuron-type-dependent intracellular signaling mechanisms at GABAergic presynaptic terminals. Indeed, a previous study showed that the presynaptic phenotypes resulting from pan-Nrxn deletion were interneuron-type-specific (Chen et al., 2017). To test whether IgSF21- and Nlgn2-induced presynaptic differentiation rely on distinct intracellular signaling mechanisms, we performed artificial synapse formation assays while targeting kinase-mediated signaling pathways pharmacologically (Figure 6). Dosage of the kinase inhibitors was based on those described in Jiang et al. (2021). We found that treatment with SP600125, an inhibitor of c-jun N-terminal kinases (JNK), significantly suppressed VGAT accumulation induced by both IgSF21 and Nlgn2 (Figures 6A–D). However, treatment with KN93 or XII, two independent CaMKII inhibitors, or with PP2, an Src kinase inhibitor, did not impact the synaptogenic activity of IgSF21, while significantly blocking that of Nlgn2 (Figures 6A–D). In contrast, inhibitors of protein kinase A (PKI), protein kinase C (BIS), and MAPK/ERK kinases (PD98059) had no statistically significant effect on either IgSF21- or Nlgn2-induced presynaptic differentiation (Figures 6A–D). These results suggest that JNK signaling is widely required for inhibitory presynaptic organization, but that only a subset of synaptic organizers leverage CaMKII and Src kinases in order to regulate inhibitory presynaptic organization.

Given that JNK, CaMKII, and Src kinases are important signaling molecules involved in the regulation of the actin and microtubule cytoskeleton (Thomas et al., 1995; Shen et al., 1998; Chang et al., 2003; McVicker et al., 2015; Benoit et al., 2021), we next investigated whether the synaptogenic activity of IgSF21 and Nlgn2 was altered by the disruption of actin and microtubule cytoskeletal dynamics. We performed artificial synapse formation assays in the presence or absence of nocodazole and cytochalasin D (Cyto D) to interfere with the polymerization and stability of microtubules and actin filaments, respectively. Both compounds almost fully suppressed presynaptic differentiation induced by both IgSF21 and Nlgn2 (Figures 6E–H). Importantly, the effect of these compounds on surface expression of exogenous IgSF21 and Nlgn2 on HEK293T cells was not responsible for the loss of their synaptogenic activity (Supplementary Figures 1C, D). These results thus suggest that both synaptic organizers require intact microtubules and microfilaments to produce VGAT-positive puncta during presynaptic differentiation despite their distinct dependencies on different kinases.

The above pharmacological results suggest that JNK, CaMKII and Src kinases act as presynaptic signaling molecules. Therefore, we next tested whether these kinases are indeed expressed at presynaptic sites (Figure 7) using Triton-soluble and -insoluble synaptosome fractions from rat forebrains, which correspond to pre- and postsynaptic fractions, respectively (Figure 1I). Both isoforms of JNK kinases, p54 and p46, were enriched in the Triton-soluble fraction, confirming the presynaptic localization of JNK kinases as previously reported (Biggi et al., 2017). Src and Fyn kinases were predominantly present in the Triton-insoluble fraction, but also in the Triton-soluble fraction, albeit with a lower enrichment. Interestingly, of the two CaMKII isoforms that we tested, CaMKIIδ was detected almost exclusively in the Triton-soluble fraction, suggesting presynaptic localization, whereas CaMKIIα was detected exclusively in the Triton-soluble fraction consistent with postsynaptic enrichment. These results indicate that presynaptic JNK regulates both IgSF21- and Nlgn2-induced inhibitory presynaptic differentiation, while presynaptic CaMKIIδ, but not postsynaptic CaMKIIα, is involved in Nlgn2-induced inhibitory presynaptic differentiation.

Having shown that JNK signaling is required for de novo formation of inhibitory presynaptic terminals induced by IgSF21 and Nlgn2 (Figures 6A–D), we next tested whether JNK signaling is also important for the maintenance of already formed inhibitory synapses. To do so, we first performed artificial synapse formation assays in which HEK293T cells expressing IgSF21 or Nlgn2 were cocultured with hippocampal neurons for 17 h, which is sufficient to induce presynaptic differentiation (Sudhof, 2018). Afterward, we treated the cultures with the JNK inhibitor SP600125 for another 7 h. We found that JNK inhibition had no significant effect on VGAT signals induced by either IgSF21 or Nlgn2 once these had been established (Figures 8A–D). Next, we investigated whether and how JNK signaling is involved in the maintenance of native formed synapses by treating primary rat hippocampal neurons at 14 DIV with SP600125 for 24 h. Like the results from the artificial synapse formation assays, JNK inhibition had no effect on the density and size VGAT puncta; likewise, it also had no effect on VGLUT1 puncta (Figures 8E–I). Given a previous study showing that JNK signaling is required for excitatory and inhibitory presynaptic differentiation induced by other Nrxn ligands such as Nlgn1 and LRRTM2 (Jiang et al., 2021), these results suggest that JNK signaling is involved in the differentiation, but not the maintenance, of presynaptic terminals of inhibitory synapses, and likely excitatory synapses as well.

One explanation for the wide effects of JNK signaling on synaptogenic activity of multiple Nrxn ligands (Figures 6A–C; Jiang et al., 2021) could be a reduction of cell surface expression of endogenous Nrxns following JNK inhibition. We thus used rat hippocampal neuron cultures to perform cell surface biotinylation assays after application of SP600125 for 24 h. To detect endogenous α-Nrxns, we used two different antibodies that recognize Nrxn1/2/3 isoforms. We found that JNK inhibition did not affect total or surface expression levels of α-Nrxns (Figures 9A–C), suggesting that the inhibition of synaptogenic activity of Nrxn ligands by the JNK inhibitor is not due to a reduction in cell surface Nrxn expression but, rather, due to loss of a functional linkage between Nrxns and JNK signaling.

Next, to test whether Nrxns are functionally linked with JNK signaling, we measured the level of active phosphorylated JNK in cortical synaptosomes prepared from Nrxn2α KO mice and WT littermates. Given that previous studies showed sex-dependent phenotypes in Nrxn2α KO mice (Born et al., 2015; Dachtler et al., 2015), we prepared synaptosome samples from males and females separately. Interestingly, synaptosomes from Nrxn2α KO females displayed hyperactivation of JNK signaling (Figures 9D–F), whereas synaptosomes from Nrxn2α KO males showed hypoactivation of JNK signaling (Figures 9D, G, H). These data suggest a sex-specific functional linkage between the JNK signaling pathway and Nrxn2α. We further investigated synaptic expression level of IgSF21 in Nrxn2α KO and WT synaptosomes and found no significant difference between Nrxn2α KO and WT mice (Supplementary Figure 3). This data suggest that altered JNK activity by Nrxn2α KO may not be related to synaptic expression level of IgSF21.