We previously showed that PSD-95 recruited at NLGN1 clusters induced by βNRXN1 cross-linking was inhibited by a broad spectrum tyrosine kinase inhibitor (genistein), which led us to postulate that NLGN1 was phosphorylated (Giannone et al., 2013). Here, we used the same cluster assay in the presence of selective inhibitors of the tyrosine kinases FGFR, Trk, or JAK (Inglés-Prieto et al., 2015), that were found to phosphorylate NGLN1 by in vitro kinase assays (Letellier et al., 2018). Dissociated hippocampal neurons were co-transfected at DIV 3 with PSD-95-GFP and recombinant NLGN1 N-terminally tagged with the biotin acceptor peptide (bAP-NLGN1) (Howarth et al., 2005; Chamma et al., 2016), then neurons at DIV 7 were pre-treated with tyrosine kinase inhibitors or vehicle (DMSO) for 2 h, before incubation with a 2:1 mixture of βNRXN1-Fc and Cy5-conjugated anti-human Fc antibodies. This causes the specific clustering of NLGN1 and PSD-95 in a time range of 30–60 min (Mondin et al., 2011; Giannone et al., 2013). This short time window is chosen before the massive synaptogenesis period typically situated between DIV 7 and 14 in those cultures (Cottrell et al., 2000; Czöndör et al., 2012; Chanda et al., 2017), because we want to clearly distinguish de novo clusters of PSD-95 induced by βNRXN1 binding to NLGN1 from pre-existing synaptic PSD-95 clusters (Gerrow et al., 2006). Neurons were observed live and PSD-95-GFP enrichment (i.e., cluster versus shaft intensity level) was analyzed within βNRXN1-Fc clusters (Figure 1A). Treatments with either the FGFR inhibitor (PD166866) or the pan Trk inhibitor (GNF5837) significantly decreased the PSD-95-GFP recruitment to NLGN1 by about 30% compared to the DMSO control, while the JAK inhibitor did not have a significant effect (Figures 1B,C). Based on our previous finding that both FGFR1 and Trks are involved in the phosphorylation of endogenous NLGN1 in neurons (Letellier et al., 2018, 2020), these results indicate that NLGN1 tyrosine phosphorylation through these RTKs is implicated in the recruitment of PSD-95 to adhesive contacts formed between βNRXN1 and NLGN1.

Over-expressing recombinant NLGN1 in neurons is well known to increase the density of excitatory synapses (Levinson et al., 2005; Chih et al., 2006; Letellier et al., 2018), and we wondered if inhibiting FGFR or Trks could also impair this synaptogenic effect of NLGN1. To this aim, we transfected neurons at DIV 7 with bAP-NLGN1 or a control membrane marker (bAP-pDisplay), together with ER-resident biotin ligase BirA^ER that biotinylates the bAP tag for subsequent detection by a fluorescent streptavidin conjugate (SA-AF6457). Neurons were cultured for 48 h in medium containing PD166866, GNF5837, or DMSO vehicle (control), then endogenous PSD-95 was immunostained and the number of PSD-95 puncta per unit dendrite area was determined (Figure 2A). This normalization minimizes the effect of potential variations in dendrite width. As reported previously (Giannone et al., 2013), the density of PSD-95 puncta increased significantly (by 30%) upon NLGN1 OE, and this effect was entirely reversed by both FGFR and pan-Trk inhibitors (Figure 2B), indicating that the phosphorylation of NLGN1 by those RTKs is implicated in the formation of new synapses induced by NLGN1 over-expression. Because the NRXN-NLGN complex has also been implicated in dendrite development prior to synaptogenesis (Chen et al., 2010; Constance et al., 2018), we checked for a potential effect of NLGN1 OE and RTK inhibitors on dendrite length in our conditions. There was a small, but not significant, increase in average dendrite length in neurons expressing bAP-NLGN1 as compared to neurons expressing bAP-pDisplay (Supplementary Figure S1). Moreover, treatments with either FGFR or Trk inhibitors did not reduce dendrite length in neurons expressing bAP-NLGN1. Thus, the inhibitory effects of RTKs on NLGN1-dependent synaptogenesis are not due to prior effects of NLGN1 on dendrite development.

Given that the FGFR inhibitor has more modest effects on NLGN1 tyrosine phosphorylation than the pan Trk inhibitor (Letellier et al., 2018), we thereafter focused our analysis on the role of the Trk family in the regulation of NLGN1-dependent synapses. To differentiate between the various Trk members involved in NLGN1 tyrosine phosphorylation, we treated cortical neurons with ligands of TrkA, TrkB and TrkC (NGF, BDNF, and NT-3, respectively) for 20 min, followed by immunoprecipitation (IP) of endogenous NLGN1 and pTyr immunoblot (Figure 3A), as previously described (Letellier et al., 2018). We used cortical cultures instead of hippocampal cultures for those experiments because they allow us to obtain a larger number of neurons, as required for a high-yield biochemical isolation of NLGN1. Out of the three RTK ligands, BDNF was the only one to significantly increase the NLGN1 pTyr signal as compared to untreated control cultures (Figure 3B), suggesting that TrkB is a major regulator of NLGN1 tyrosine phosphorylation in neurons. Based on this result, we thus explored in more detail the function of TrkB in relation to NLGN1-dependent synapses.

Use of the pan-Trk pharmacological inhibitor did not allow us to discriminate which specific Trk family member was involved in the NLGN1-dependent increase in synapse formation (Letellier et al., 2018). We thus turned here to a knock-down and rescue approach, focusing on the role of TrkB in NLGN1 phosphorylation and NLGN1-dependent excitatory synapse assembly. To first check the validity of the approach, we transfected COS-7 cells with HA-NLGN1 and TrkB-RFP constructs, in the presence or not of shRNA against TrkB (sh-TrkB). Western blots performed on proteins isolated from those cell lysates showed a 70% decrease of TrkB-RFP expression by sh-TrkB, and the absence of effect on rescue TrkB-RFP (TrkB-res) (Supplementary Figures S2A,B). In parallel, the phosphotyrosine (pTyr) signal associated with NLGN1 was increased by the expression of TrkB compared to an empty vector (Supplementary Figures S2A,C). This increase was impaired by the co-expression of sh-TrkB, and reversed by the expression of rescue TrkB-RFP.

Then, hippocampal neurons were co-transfected with bAP-NLGN1 (or bAP-dDisplay), BirA^ER, and either scramble shRNA, sh-TrkB, or sh-TrkB + rescue TrkB-RFP, and neurons were stained at DIV 14 with an antibody to endogenous PSD-95 and with streptavidin-AF647 to label biotinylated proteins (Figure 4A). As shown above, the expression of bAP-NLGN1 increased the density of PSD-95 puncta by 30% as compared to neurons expressing bAP-pDisplay, both conditions in the presence of sh-scramble (Figures 4A,B). This effect of NLGN1 OE on PSD-95 was entirely blocked by the co-expression sh-TrkB, and reversed by the co-expression of rescue TrkB-RFP (Figures 4A,B). Thus, TrkB knock-down selectively impaired the synaptogenic effect of NLGN1. To prove that this effect was mediated by NLGN1 phosphorylation, we expressed a NLGN1 mutant bearing an alanine mutation of the unique tyrosine residue in the NLGN1 intracellular tail, i.e., NLGN1-Y782A (Giannone et al., 2013; Letellier et al., 2018). Because it is unable to bind gephyrin and instead binds well PSD-95, this NLGN1-Y782A mutant is thought to mimic phosphorylated NGLN1 (Giannone et al., 2013). As expected, the density of PSD-95 puncta in neurons expressing bAP-NLGN1-Y782A was as high as in neurons expressing bAP-NLGN1-WT, and not decreased by the co-expression of sh-TrkB (Supplementary Figures S3A,B). Thus, the increase in synapse formation mediated by NLGN1 involves the phosphorylation of the Y782 residue by TrkB.

In addition, we computed the surface density and projected area of dendritic spines by analyzing the same images taken from DIV14 neurons (Figure 5A), using a recently released program based on deep learning (Fernholz et al., 2024). Surprisingly, the linear density of spines was not altered by NLGN1 OE as compared to the pDisplay condition, and not affected either by the manipulation of TrkB levels (Figure 5B). In contrast, NLGN1 OE increased the projected area of spines (Figure 5C), an effect that may be mediated by the interaction between NLGN1 and the WAVE complex (Chen et al., 2014; Xing et al., 2018), that in turn regulates the assembly of a branched actin network in dendritic spines (Chazeau et al., 2014). Interestingly, this increase in spine volume caused by NLGN1 OE was reversed by the co-expression of sh-TrkB (Figure 5C). These data combined to the increase in PSD-95 puncta induced by NLGN1 OE suggest the formation of dendritic spines containing several post-synaptic density modules (Hruska et al., 2018).

GFP-βNRXN1 (Neupert et al., 2015) was a gift from M. Missler (Münster University). Plasmids for mouse HA-NLGN1 harboring both A and B splice inserts and mouse βNRXN1 without splice site 4 (SS4-) C-terminally fused to human Fc (Dean et al., 2003) were gifts from P. Scheiffele (Biozentrum, Basel). Mouse NLGN1 containing both A and B splice inserts and N-terminally fused to the biotin acceptor peptide (bAP-NLGN1), as well as endoplasmic reticulum (ER) resident biotin ligase (BirA^ER) (Howarth et al., 2005) were gifts from A. Ting (Stanford University, CA). PSD-95-GFP (Kuriu et al., 2006) was a gift from S. Okabe (Tokyo University, Japan). shRNA to TrkB and corresponding scramble shRNA (both containing a GFP reporter), wild type rat TrkB-RFP, and shRNA-resistant (rescue) TrkB-RFP (Jeanneteau et al., 2008) were gifts from F. Jeanneteau (Montpellier, France).

The protocol for the production and purification of the βNRXN1-Fc protein was previously described (Mondin et al., 2013). Briefly, the conditioned medium from a stable hygromycin-resistant HEK cell line producing βNRXN1-Fc was collected, and recombinant βNRXN1-Fc was purified on a protein G column (HiTrap Protein G HP, GE Healthcare) to a concentration of 1.0 mg.mL⁻¹ after elution with glycine in tubes containing Tris buffer. The protein was stored in small aliquots at −80°C and thawed right before use.

Hippocampal cultures were prepared from E18 Sprague–Dawley rat embryos obtained from gestant females purchased weekly (Janvier Labs, Saint-Berthevin, France), as described in published protocols with small modifications (Kaech and Banker, 2006; Mondin et al., 2013). Animals were handled and killed according to European ethical rules, with an authorization from the local Bordeaux Ethics committee (EC 50). Hippocampi were dissected out from whole brains under a binocular (Nikon) placed in a laminar flow hood, within a 60-mm Petri dish filled with 5 mL Hibernate medium (made of 500 mL Minimal Essential Medium - MEM, 1 g 3-(N-morpholino)propanesulfonic acid (MOPS) MOPS, 3.9 mL Glucose 45%, 5 mL Na Pyruvate, pH = 7.3). After dissection, hippocampi from 8 to 12 embryos were incubated in 5 mL 0.05% trypsin–EDTA, 10 mM HEPES at 37°C for 20 min, then cells were aspirated up and down for 20 times in a flame-polished Pasteur pipet pre-coated with horse serum. Cell concentration was determined using a Malassez cell counting chamber. Neurons were resuspended in MEM supplemented with 10% horse serum (MEM-HS) and plated at a density of 3,000 cells.cm⁻² on 18-mm coverslips previously coated for 2 h with 1 mg.mL⁻¹ polylysine in borate buffer (pH = 8.3). Three hours after plating, coverslips were flipped onto 60-mm dishes containing a glial cell layer in Neurobasal medium supplemented with 2 mM L-glutamine and 1x NeuroCult SM1 Neuronal supplement (STEMCELL Technologies) and cultured for 2 weeks at 37°C and 5% CO₂. Astrocytes were prepared from the same embryos, plated between 20,000 and 40,000 cells per 60-mm dish and cultured in MEM (Fisher Scientific) containing 4.5 g.L⁻¹ glucose, 2 mM L-glutamine, and 10% horse serum (Invitrogen) for up to 14 days before being used as feeder layers for neurons.

Neurons were transfected using a lipofection protocol (Effecten, Qiagen). The following combinations and concentrations of plasmids were used: (i) For the βNRXN1-Fc cluster assay in the presence of RTK inhibitors (Figure 1): 1 μg bAP-NLGN1: 1 μg PSD-95-GFP and 0.5 ug BirA^ER (transfection at DIV 3, observation at DIV 7); (ii) for the synapse induction assay in the presence of RTK inhibitors (Figure 2): 1 μg bAP-pDisplay or bAP-NLGN1: 0.5 μg BirA^ER (transfection at DIV 7, observation at DIV 9–10); for the synapse induction assay upon genetic manipulation of TrkB (Figures 4, 5): 1 μg bAP-pDisplay or bAP-NLGN1; 0.5 μg BirA^ER; 1 μg sh-TrkB (or sh-scramble), or 1 μg sh-TrkB +1 μg rescue-TrkB (transfection at DIV 7, observation at DIV 14–15). The culture medium contains free biotin in sufficient amounts so that BirA^ER is able to biotinylate the bAP tag on pDisplay or NLGN1.

DIV 7 neurons co-transfected with bAP-NLGN1 and PSD-95-GFP were pre-treated for 2 h with 0.5 μM tyrosine kinase inhibitors (FGFR inhibitor PD166866, Sigma-Aldrich PZ0114), pan-Trk inhibitor (GNF5837, Tocris 4559), pan-JAK inhibitor I (Calbiochem 420099) or vehicle (DMSO, concentration 1:2,000). Then neurons were incubated live for 30 min at 37°C with a mix of 10 μL βNRXN1-Fc (at 1 mg/mL) and 5 μL Alexa647-conjugated goat anti-human Fc (Jackson Immunoresearch 1 mg/mL) diluted in 100 μL of Tyrode solution (15 mM D-glucose, 108 mM NaCl, 5 mM KCl, 2 mM MgCl₂, 2 mM CaCl₂ and 25 mM HEPES, pH 7.4) supplemented with 1% immunoglobulin-free BSA (Sigma-Aldrich A7030). To save protein, coverslips were turned upside down onto the drop placed on a piece of Parafilm. After incubation, neurons were rinsed in warm Tyrode and mounted in an Inox Chamber (Life Imaging Services, Basel) under a TiE Nikon microscope thermostated at 37°C.

Cultures were fixed for 10 min in warm 4% paraformaldehyde-4% sucrose solution and permeabilized for 5 min with 0.1% Triton X-100 in PBS. Non-specific binding was blocked with PBS containing 1% Bovine Serum Albumin (BSA Sigma-Aldrich A7030). In Figures 2, 4, neurons expressing biotinylated bAP-pDisplay or bAP-NLGN1 were stained with streptavidin-AF647 (Thermo Fisher Scientific S21374, 2 mg/mL, 1:800) and endogenous PSD-95 was immunostained with a mouse monoclonal antibody (Thermo Fisher Scientific clone 7E3-1B8, 1:400), followed by Alexa568-conjugated goat anti-mouse antibody (Thermo Fisher Scientific A11031, 2 mg/mL 1:800). In Figures 4, 5, she GFP reporter of the shRNAs was not amplified. Coverslips were mounted in Mowiol (Calbiochem).

All stainings were visualized on an inverted epifluorescence microscope (Nikon Eclipse TiE) equipped with a 60x/1.40 NA oil immersion objective and filter sets for EGFP/Alexa488 (excitation: FF01-472/30; dichroic: FF-495Di02; emission: FF01-525/30); TRITC/Alexa568 (excitation: FF01-543/22; dichroic: FF-562Di02; emission: FF01-593/40); and Cy5/Alexa647 (excitation: FF02-628/40; Dichroic: FF-660Di02; Emission: FF01-692/40) (SemROCK). Full field images of 1,200 × 1,200 pixels (pixel size = 11 μm) were acquired with an sCMOS camera (Photometrics PRIME 95B), using the Metamorph^® software (Molecular Devices).

The detection of βNRXN1-Fc clusters and the quantification of the PSD-95-GFP enrichment level within those clusters (Figure 1) was carried out using a semi-automatic wavelet-based segmentation program written in Metamorph, as previously described (Racine et al., 2007; Czöndör et al., 2013; Mondin et al., 2013). Briefly, the user selects the appropriate wavelet planes and defines a threshold to correctly segment the βNRXN1-Fc clusters on the anti-Fc AF647 reference image. Puncta smaller than 10 square pixels are discarded. From this segmentation, a binary image is generated which is further used in Metamorph to automatically generate regions around those puncta. The regions are then transferred to the PSD-95-GFP images acquired in parallel, and the average intensity within those regions is measured. Finally, a threshold is applied on the overall image to highlight the dendrite contour and measure its area and intensity. The fluorescence intensity within puncta is divided by the intensity in the shaft, this ratio being defined as the enrichment factor. The same program was used to segment endogenous PSD-95 puncta for 2 dendritic segments per neuron and measure their number. This number is divided by the surface area of the dendritic segment to yield the surface density of PSD-95 puncta/μm² (Figures 2, 4). The length of the 2–4 longest dendrites was determined manually in Metamorph (Supplementary Figure S1). The analysis of dendritic spines was performed using DeepD3, a deep learning-based framework that segments dendrite shafts and spines in a fully automated fashion (Fernholz et al., 2024). This program was applied to images of shRNA GFP reporters in DIV14 neurons, giving a binary image of dendrite shafts, an image of dendritic spines as a localization probability map, and a full list of the detected dendritic spines as regions of interest, whose number and projected area was determined in FIJI. The cumulated length of the corresponding binarized dendritic segments was determined in Metamorph for each individual neuron. The number of detected spines was then divided by dendrite length.

COS-7 cells from the European Collection of Authenticated Cell Cultures (ECACC) purchased via (Sigma-Aldrich, Acc Nc 87021302) were cultured in DMEM (Eurobio) supplemented with 1% glutamax (Sigma-Aldrich, #3550–038), 1% sodium pyruvate (Sigma-Aldrich, #11360–070), 10% Fetal Bovine Serum (Eurobio). Cells were thawed from frozen vials at passage +4, and maintained up to passage 20. Cells were regularly tested negative for mycoplasma, using the MycoAlert detection kit (Lonza, #LT07-218). COS-7 cells were plated in 6-well plates at a density of 150,000 cells/well. After 5–6 h, cells were transfected using the X-tremeGENE HP DNA Transfection Reagent (Roche) with 1 µg HA-NLGN1 with or without RTKs (1 µg wild type TrkB-RFP, 4 µg shTrkB + 1 µg wild type TrkB-RFP, 4 µg shTrkB + rescue TrkB-RFP), Transfected cells were and left under a humidified 5% CO₂ atmosphere (37°C) for 2 days before being processed for NLGN1 immunoprecipitation (IP).

To obtain the high protein yields required to perform reproducible biochemistry experiments, we cultured cortical neurons. Whole cortices from E18 rat embryos were dissected out and cut into several thin pieces with a scalpel. Neurons were dissociated using the same procedure described above for hippocampal neurons, then plated at a density of 300,000–600,000 cells per well in 6-well plates coated with 1 mg.ml⁻¹ polylysine and cultured for 10 days. Ligands specific for TrkA, TrkB, and TrkC, i.e., respectively NGF (50 ng/mL; Peprotech 450–01), BDNF (50 ng/mL; Peprotech 450–02), or NT-3 (50 ng/mL; Peprotech 450–03), were applied for 20 min to the cultures prior to cell lysis. Cultures were harvested and solubilized proteins were immediately processed for IP.

COS-7 cells or cortical neurons were treated with 100 μM pervanadate for 15 min before cell lysis, to preserve phosphate groups on NLGN1. Whole-cell protein extracts were obtained by solubilizing cells in lysis buffer (50 mM HEPES, pH 7.2, 10 mM EDTA, 0.1% SDS, 1% NP-40, 0.5% deoxycholate - DOC, 2 mM Na-Vanadate, 35 μM PAO, 48 mM Na-Pyrophosphate, 100 mM NaF, 30 mM phenyl-phosphate, 50 μM NH₄-molybdate and 1 mM ZnCl₂) containing protease Inhibitor Cocktail Set III, EDTA-Free (Calbiochem). Lysates were clarified by centrifugation at 8,000 × g for 15 min. For NLGN1 immunoprecipitations, 500–1,000 μg of total protein (estimated by Direct Detect assay, Merck Millipore), were incubated overnight with 2 μg of, rabbit anti-NLGN1 (Synaptic systems 129013), then precipitated with 20 μL protein G beads (Dynabeads Protein G, Thermo Fisher Scientific) and washed 4 times with lysis buffer. At the end of the IP, 20 μL beads were resuspended in 20 μL of 2x loading buffer, and supernatants were processed for SDS-PAGE and Western blotting.

Proteins were loaded on 4–15% (Mini-PROTEAN TGX Precast Protein Gels, Bio-Rad) for separation (200 V, 400 mA, 40 min) and afterwards transferred to nitrocellulose membranes for semi-dry immunoblotting (7 min, BioRad). Membranes were rinsed in Tris-buffered saline Tween-20 (TBST; 28 mM Tris, 137 mM NaCl, 0.05% Tween-20, pH 7.4) and incubated with 5% non-fat dried milk for 45 min at room temperature. Membranes were incubated for 1 h at room temperature (or overnight at 4°C) in 0.5% non-fat dried milk in TBST containing the appropriate primary antibodies, as follows: mouse anti-pTyr (1:1,000, Cell Signaling 9411S and Merck 05–1050), rabbit anti-NLGN1 (1:1,000, Synaptic systems 129013), anti-GAPDH (1:10,000, GeneTex GTX627408), anti-GFP (1:5,000, Roche 11814460001), anti-HA (1:1,000, Ozyme BLE901513), anti-RFP (1:5,000, Abcam ab124754). After washing 3 times with TBST buffer, blots were incubated for 1 h at room temperature with horseradish peroxidase (HRP)–conjugated donkey anti-mouse or anti-rabbit secondary antibodies (Jackson Immunoresearch, 715–035-150 and 711–035-152, respectively, 1:5,000 dilution) for the SM or anti-mouse or anti-rabbit Easyblot (GeneTex, GTX221667-01 or GTX221666-01, respectively, 1:1,000 dilution) for the IP material. The latter was used to avoid the detection of primary antibodies from the IP. Target proteins were detected by chemiluminescence with Super signal West Dura or Super signal West Femto (Pierce) or Clarity Western ECL Substrate (Bio-Rad) on the ChemiDoc Touch system (Bio-Rad). Average intensity values were calculated using Image Lab 5.0 software (Bio-Rad). The ratio of phospho-NLGN1 divided by the total NLGN1 signal was normalized to the control samples.