Abstract
Although neurons release neurotransmitter before contact, the role for this release in synapse formation remains unclear. Cortical synapses do not require synaptic vesicle release for formation (Verhage et al., 2000; Sando et al., 2017; Sigler et al., 2017; Held et al., 2020), yet glutamate clearly regulates glutamate receptor trafficking (Roche et al., 2001; Nong et al., 2004) and induces spine formation (Engert and Bonhoeffer, 1999; Maletic-Savatic et al., 1999; Toni et al., 1999; Kwon and Sabatini, 2011; Oh et al., 2016). Using rat and murine culture systems to dissect molecular mechanisms, we found that glutamate rapidly decreases synapse density specifically in young cortical neurons in a local and calcium-dependent manner through decreasing N-methyl-d-aspartate receptor (NMDAR) transport and surface expression as well as cotransport with neuroligin (NL1). Adhesion between NL1 and neurexin 1 protects against this glutamate-induced synapse loss. Major histocompatibility I (MHCI) molecules are required for the effects of glutamate in causing synapse loss through negatively regulating NL1 levels in both sexes. Thus, like acetylcholine at the neuromuscular junction, glutamate acts as a dispersal signal for NMDARs and causes rapid synapse loss unless opposed by NL1-mediated trans-synaptic adhesion. Together, glutamate, MHCI, and NL1 mediate a novel form of homeostatic plasticity in young neurons that induces rapid changes in NMDARs to regulate when and where nascent glutamatergic synapses are formed.
Significance Statement
The role for neurotransmitter release in synaptogenesis in the central nervous system remains unclear. Here, we reconcile conflicting results in the field by showing that glutamate plays an important role in synapse formation by acting as a dispersal signal for N-methyl-d-aspartate receptors that is counteracted by trans-synaptic adhesion in intact tissue, similar to the role for neurotransmitter at the neuromuscular junction. We also describe a novel form of homeostatic plasticity in young neurons that allows them to respond to changes in activity through surprisingly rapid changes in synapse density. Finally, we show that this plasticity is modulated by immune proteins—major histocompatibility I molecules—through negative regulation of neuroligin levels, connecting two important synaptic signaling pathways for the first time.
Introduction
Synapse formation is an intricate process that is integral to the establishment of the complex circuitry underlying perception, cognition, and behavior. Impairment of this process is associated with neurodevelopmental and psychiatric disorders (Cameron and McAllister, 2018; Sudhof, 2021). During glutamatergic synapse formation, pre- and postsynaptic proteins accumulate within minutes of stabilized axo-dendritic contact (Waites et al., 2005; McAllister, 2007; Washbourne, 2015). Recruitment of transport packets containing glutamate receptors is integral to this process. N-Methyl-d-aspartate receptor (NMDAR) transport packets (NRTPs) rapidly deliver NMDARs to newly forming synapses following contact (Washbourne et al., 2002) via a neuroligin (NL1)-dependent mechanism (Barrow et al., 2009; Wu et al., 2019). NRTP transport along dendrites occurs rapidly and bidirectionally, interspersed frequently with pauses at sites where NMDARs cycle with the plasma membrane and where the initial complement of synapses selectively forms (Washbourne et al., 2004). Periodic delivery to the dendritic surface raises the intriguing possibility that NMDARs are capable of sensing glutamate before synaptic contacts are established. Yet, the consequences of NMDAR activation for receptor trafficking and synapse formation remain unknown.⇓
Key resources
During synaptogenesis, axonal growth cones release neurotransmitter before they contact other cells (Hume et al., 1983; Young and Poo, 1983; Matteoli et al., 1992; Sabo and McAllister, 2003). In cortical neurons, vesicles containing the glutamate transporter vGlut1 rapidly cycle along the axonal membrane and at the growth cone (Sabo and McAllister, 2003; Sabo et al., 2006). However, in contrast to the unambiguous role for activity in refinement of connections (Hooks and Chen, 2020), the role for neurotransmitter release in the initial formation of synapses is unclear. Decades of research indicate that synapse formation is independent of activity because synapses form in the absence of neurotransmitter release (Verhage et al., 2000; Varoqueaux et al., 2002). Yet, glutamate clearly induces dendritic filopodial extension and spine formation (Engert and Bonhoeffer, 1999; Maletic-Savatic et al., 1999; Toni et al., 1999). Moreover, glutamate uncaging induces the formation of dendritic spines with synapses in an NMDAR- and NL-1-dependent manner (Kwon and Sabatini, 2011; Kwon et al., 2012), and GABA uncaging induces both dendritic spines and clustering of inhibitory proteins (Oh et al., 2016).
This proposed positive role for glutamate in inducing synapse formation at cortical synapses is surprising. First, neurotransmitter plays the opposite role at the neuromuscular junction (NMJ). During NMJ formation, acetylcholine (ACh) causes internalization of ACh receptors (AChRs) and is opposed by an antidispersal signal (Misgeld et al., 2005). Moreover, glutamate activation of NMDARs at the NMJ during synaptogenesis causes synaptic pruning (Personius et al., 2016). Second, in the central nervous system (CNS), activation of NMDARs causes their internalization (Tovar and Westbrook, 2002; Nong et al., 2003; Lavezzari et al., 2004), and NMDARs in young neurons undergo substantial glutamate-induced internalization (Roche et al., 2001). Finally, our lab has shown that a global reduction in activity over days increases synapse density selectively in young cortical neurons in a manner requiring major histocompatibility I (MHCI) molecules (Glynn et al., 2011), a classical immune molecule family that negatively regulates synaptic density and function (Glynn et al., 2011; Elmer et al., 2013).
Here, using a cell culture approach essential for examining sufficiency and molecular mechanisms, we tested the hypotheses that glutamate at CNS synapses may act like ACh at the NMJ—dispersing NMDARs and causing synapse elimination unless opposed by trans-synaptic adhesion—and that MHCI mediates this process. We found that acute activation of NMDARs decreases the mobility and surface expression of NMDARs, the proportion of mobile NMDARs transported with NL1, and synapse density. The glutamate-induced decrease in NMDAR transport is local and dependent on NMDAR-mediated Ca2+ influx. Importantly, these effects of glutamate are prevented at sites of association of the NMDAR/NL1 complex with Nrxn1 in a hemisynapse assay. Finally, MHCI is required for glutamate-induced synapse loss, through a novel negative regulation of NL1 protein levels. Together these data show that glutamate causes NMDAR internalization and surprisingly rapid synapse loss in an MHCI-dependent manner unless opposed by NL1-neurexin-mediated trans-synaptic adhesion in young cortical neurons.
Materials and Methods
Resource availability
Lead contact
Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Dr. A. Kimberley McAllister (kmcallister{at}ucdavis.edu).
Materials availability
This study used two newly generated, unpublished plasmids. These can be made available upon request. Other key materials are listed in Table 1.
Data and code availability
This paper does not report the original code.
Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.
Experimental model and subject details
All studies were conducted with approved protocols from the University of California Davis Animal Care and Use Committee, in compliance with NIH guidelines for the care and use of experimental animals. Timed pregnant Sprague Dawley rats were purchased from Harlan. Mice were generated at UC Davis from C57BL/6, NL1−/− [knock-out (KO)], β2m−/− (KO), and H2-Kb−/−/H2-Db−/− [double KO (DKO)] lines. All mouse strains were kept on a pure C57BL/6 background, group housed, and on a 12:12 light/dark cycle. For all experiments, genotypes and treatment were blinded until postanalysis. For all culture experiments, male pups were used. For the β2m-KO biochemistry experiments, three littermate pairs [KO and wild-type (WT)] were male and two littermate pairs were female at the indicated age.
Reagents
N-Methyl-d-aspartic acid (NMDA), glutamate (l-glutamic acid), d(−)-2-amino-5 phosphopentanoic acid (APV), 6-cyano-7-nitroquinoxaline-2,3-dione (CNQX), and α-methyl-4-carboxyphenylglycine (MCPG), all from Sigma-Aldrich, were dissolved at the indicated concentrations in ACSF (rat) or Neurobasal medium (mouse) from Thermo Fisher Scientific, with the diluent acting as the control indicated in all figures. All drugs were added 10 min prior to fixation, or following a 10 min control baseline recording via a gravity-fed perfusion system.
Neuronal culture and transfections
Neurons from Postnatal Day (P) 0–2 rat or P1–2 mouse occipital cortex were cultured using established protocols (Glynn et al., 2011). Rat neurons were plated at a density of 25 K/18 mm coverslip onto either astrocyte monolayers or poly-l-lysine-coated coverslips inverted over astrocyte monolayers as previously described (Glynn et al., 2011). Mouse neurons were plated at a density of 45 K/18 mm coverslip in astrocyte-conditioned media (AsCM), but otherwise treated the same as rat neurons. AsCM was made by first culturing cortical astrocytes in MEM (Invitrogen), media containing 10% fetal calf serum (Gemini) and 1:500 MycoZap (Lonza) and replated after growth to 90% confluency in flasks to standardize astrocyte numbers and remove oligodendrocytes. Upon reconfluence, media were replaced with Neurobasal media supplemented with B27 plus (Thermo Fisher Scientific), glutamate (Thermo Fisher Scientific), sodium pyruvate (Thermo Fisher Scientific), and HEPES (Thermo Fisher Scientific) for 48 h, collected, and filter sterilized for storage and use. Rat neurons were transfected with Lipofectamine 2000 (Invitrogen) 18–24 h before imaging, while mouse neurons were transfected with the same reagent at 3 DIV for use at 8 DIV. Data was collected from at least two separate neuronal cultures for all experiments as indicated in the figure legends.
Plasmid constructs
HA-CD4 and HA-Nrx1βSS4−, lacking an insert in splice site number 4 (Ko et al., 2009), have been described previously (Ko et al., 2009; Takahashi et al., 2012). SEP-GluN1-1a cDNA was constructed by fusing the superecliptic pHluorin (enhanced mutant of pH-sensitive GFP; Kopec et al., 2006) to the N terminal of rat GluN1-1a. β2m shRNA hairpin and H2-Kb-mCherry plasmids were previously described (Glynn et al., 2011; Elmer et al., 2013). NL1-mCherry (NM_053868.2) was generated by subcloning the rat NL1 sequence into the pmCherry-N1 vector. H2-Kb-YFP was generated by subcloning the mouse H2-Kb sequence into the EYFP-N1 vector. NL1-EGFP was previously described (Fu et al., 2003).
Immunocytochemistry
Neurons were fixed with 4% paraformaldehyde and 4% sucrose in PBS at 4°C for 10 min, permeabilized with 0.25% Triton X-100 in PBS for 15 min, blocked with 10% BSA in PBS for 30 min, and incubated with primary and secondary antibodies in 3% BSA. Surface staining was performed on fixed cells without permeabilization. Coverslips were mounted in Fluoromount (Thermo Fisher Scientific). Primary antibodies used were as follows: vGlut1 (AB5905, 1:1,000; Chemicon), GluN2A (07-632, 1:400; Upstate Biotechnology), GluN2B (AB1557, 1:400; Chemicon), GFP (A11122, 1:1,000; Molecular Probes), HA (Clone 3F10, 5 µg/ml, Roche), mCherry (AB0040-200, 1:2,000; SICGEN), VAMP2 (104-202, 1:1,000; Synaptic Systems), and PSD-95 (clone K28/43, 1:750; Antibodies). Secondary antibodies used were Alexa Fluor-488, Alexa Fluor-405-, Alexa Fluor-647-, or Alexa Fluor-568-conjugated anti-rabbit, anti-mouse, and anti-goat (1:200–1:400; Invitrogen).
Imaging
Live imaging was conducted in an imaging perfusion chamber for 18 mm coverslips (QE-1; Warner) on an Eclipse TE300 Nikon inverted microscope using a 60× oil immersion objective (1.45 NA). Fluorophores were excited at their absorption maxima using a Lambda filter wheel for excitation and emission (Sutter Instrument) with double and triple bandpass filters (Chroma). Bleed-through from fluorophores into other channels was tested by using only one fluorophore and checking all other wavelength and filter combinations for detectable signal. Images were acquired sequentially with a CoolSNAP HQ CCD camera (Roper Scientific) and Simple PCI software (C-Imaging, Compix). Imaging was conducted with continual perfusion of ACSF as follows (in mM): 120 NaCl, 3 KCl, 2 CaCl2, 2 MgCl2, 30 glucose, and 20 HEPES and 0.2% sorbitol, pH 7.3, at 37°C from a gravity-fed perfusion system. Images were typically collected at 10 s intervals, as this provided the best compromise between high temporal resolution and minimizing neuronal toxicity and photobleaching. For imaging of immunocytochemistry, coverslips were viewed with an Olympus FluoView 2.1 laser scanning confocal system with a 60× PlanApo oil immersion objective (1.4 NA) on an IX70 inverted microscope. Images for each fluorophore were acquired sequentially with 2.5× digital zoom and 2× Kalman averaging. Rat cortical neuron images used for punctal density or colocalization analysis underwent one iteration of deconvolution with the appropriate point spread function for 488, 568, or 647 nm light before image analysis (SVI Huygens Essential).
Focal perfusion
Glutamate was perfused locally onto small sections of dendrite through a fire-polished micropipette (tip opening ∼1–2 μm; McAllister and Stevens, 2000). The micropipette was filled with ACSF containing glutamate (500 µM) and Alexa hydrazide-488 for visualization. Repetitive pressure injection at 2 psi was applied to the micropipette with an electrically gated valve (Picospritzer, General Valves) for five, 2 s pulses at 30 s intervals for 10 min. For quantification, a region of interest (ROI) was drawn around the size of the focal perfusion area and the mobility of DsRed-GluN1 puncta within the ROI were quantified and compared with nonperfused neighboring regions of the same dendrite using MetaMorph.
Electrophysiology
Whole-cell patch-clamp recordings were made from rat cortical neurons at 5–7 DIV (McAllister and Stevens, 2000). To isolate miniature excitatory postsynaptic currents (mEPSCs), recordings were made at −70 mV in the presence of 0.5 μM TTX. The extracellular solution consisted of the following (in mM): 110 NaCl, 3 KCl, 10 HEPES, 10 d-glucose, 10 glycine, 2 CaCl2, and 2 MgCl2, pH 7.3. The osmolarity was adjusted to 305 using sorbitol. The intracellular solution consisted of the following (in mM): 130 potassium gluconate, 10 KCl, 10 HEPES, 10 EGTA, 1 CaCl2, 5 ATP-Mg2+, and 5 GTP-Li2+, pH 7.3. Recordings were filtered at 2 kHz using an Axopatch 200B amplifier and digitized at 5 kHZ using Clampex 8 (Axon Instruments). Events were detected with MiniAnalysis software (Synaptosoft), at a threshold fourfold larger than the RMS noise and confirmed by visual inspection. Statistical analysis of average amplitude and frequency was performed using GraphPad Prism software.
HEK cell coculture assay
HEK 293 cells were transfected with Lipofectamine 2000 (Invitrogen) with HA-neurexin-1β or HA-CD4 plasmids. After 24 h, transfected cells were trypsinized, plated onto 6 DIV cortical neurons expressing SEP-GluN1-1a and cocultured for a further 24 h. Cocultured cells were then fixed and immunostained under nonpermeabilizing conditions using anti-GFP (A11122, 1:1,000; Molecular Probes) and anti-HA (5 µg/ml, Roche) antibodies. Confocal Z-stacks were acquired at 0.5 μm steps using a 60× objective on an inverted Olympus FluoView 2.1 laser scanning confocal system and processed with MetaMorph software (v7.5, Molecular Devices) using the maximum fluorescent intensity projection. For quantification, the contours of transfected HEK293 cells were chosen as ROIs, and the density of dendritic surface GluN1 puncta within the ROI was quantified using custom-written journals in MetaMorph.
Biochemical experiments
Hippocampi from sex-matched littermate pairs of P15 WT and β2m-KO animals were harvested and dounced in 0.3 M sucrose/HEPES buffer with protease and phosphatase inhibitors (A32961; Thermo Fisher Scientific). Samples were spun at 600 g for 10 min at 4°C, and the supernatant was collected and denatured in Laemmli buffer with β-mercaptoethanol. Samples were heated at 85°C for 15 min and stored at −20°C until they were run on SDS-PAGE gel for Western blot analysis. For surface labeling, neurons were treated with Neurobasal or 50 μM glutamate for 10 min at 37°C before removal of media and incubated with 1 mg/ml Sulfo-NHS-Biotin in ice-cold PBSCM for 20 min. Labeling was quenched with 20 mM glycine before lysing in RIPA buffer with protease and phosphatase inhibitors (A32961; Thermo Fisher Scientific). An aliquot for total protein was collected and denatured in Laemmli buffer with β-mercaptoethanol. Lysate was incubated with streptavidin agarose beads (20353 Pierce) overnight on a rotator, washed in RIPA, and denatured in Laemmli buffer with β-mercaptoethanol. Samples were heated at 85°C before running on SDS-PAGE gel. Gels were transferred to nitrocellulose membranes and incubated with primary and secondary antibodies in LI-COR intercept blocking buffer (LI-COR 927-60001). Secondary antibodies used were donkey anti-rabbit 680 (LI-COR 925-68073) and anti-mouse 800 (LI-COR 925-32212) for imaging on the LI-COR Odyssey system.
Data analysis
Quantification of immunocytochemistry and colocalization during live imaging was performed using the raw images in MetaMorph software (Washbourne et al., 2002, 2004; Barrow et al., 2009). Pre- and postsynaptic protein colocalization and density were quantified as described previously using custom-written journals in MetaMorph software (v7.5, Molecular Devices; Glynn et al., 2011). Intensities were measured within ∼1 µm diameter circles around the center of manually defined puncta in each channel. Quantification of transport velocities was determined by measuring the time from the start to the end of a unidirectional motion to give mean velocity. Puncta were considered mobile when they performed a unidirectional movement of >2 µm across at least three successive images (Washbourne et al., 2004; Barrow et al., 2009). Images were processed postquantification using Adobe Photoshop for presentation purposes.
Experimental design and statistical analysis
For analysis related to dendritic segments (synapse density, live imaging, surface GluN1), unless noted as individual dendrites, 2–4 segments were analyzed for each cell and averaged per cell. These cell averages reflect the N used in the experiments and for statistical analysis. Each experiment was repeated biologically 2–4 times. For neuronal experiments, only male pups were used in this paper.
For any experiments with a red dotted line indicated in the figure, conditions were normalized to the control condition within the biological repeat (i.e., individual neuronal cultures) for comparison across experiments. In cases in which multiple genotypes were used, all conditions were normalized to the WT control. For each experiment, the statistical test was selected based on the number of groups being compared, the normality of the distribution, and the equality of the standard deviations. This resulted in Student's t tests and ANOVAs as noted in the results. Specifically, ANOVAs were used when more than two groups were experimentally manipulated and analyzed from the same pool, while t tests were used when two groups were directly compared within condition (e.g., individual dendrite pre- and posttreatment or same animal different wells with alternate drug treatment). Statistics were run using GraphPad Prism (V9) software. All data are expressed as mean ± SEM. Sample sizes are included within the text for each experiment.
Results
Acute neuronal activity alters glutamatergic synapse density within minutes
At very early stages of CNS development, even prior to synapse formation, dendrites are speckled with highly mobile ionotropic glutamate receptor transport packets that can detect, and be activated by, ambient glutamate as they rapidly cycle along the dendritic shaft (Washbourne et al., 2002, 2004). Given that their recruitment to nascent synapses occurs within ∼8 min following initial axonal contact (Washbourne et al., 2002) and activity regulation as short as 5 min significantly alters the phosphorylated proteome in synapses (Desch et al., 2022), it is conceivable that neurotransmitter release during cortical development could not only regulate cortical connectivity and the number of synapses formed but could also do so in a timeframe of minutes rather than hours or days as previously reported (Rao et al., 1998; Friedman et al., 2000). To test this hypothesis, neurotransmitter inhibitors and agonists were applied to 6–7 DIV rat cortical cultures 10 min prior to fixing and immunostaining with antibodies against the presynaptic vesicle protein vGlut1 and the postsynaptic NMDAR subunits GluN2A and GluN2B to label synapses. The inhibitors used were TTX (500 nM) to block action potentials, CNQX (10 µM) to block α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (AMPARs), and APV (50 µM) to block NMDARs, as well as a receptor antagonist cocktail (ACM) comprising APV, CNQX, and MCPG (500 µM) to also block metabotropic glutamate receptors. We found that, after just 10 min of APV treatment, there was a significant 25% increase in the density of synapses determined by quantification of overlap between vGlut1 and GluN2A/GluN2B (Fig. 1A,B), whereas no changes were observed following TTX, CNQX, or ACM treatment (Fig. 1B). Conversely, stimulation of NMDARs for just 10 min with either NMDA (50 µM), or NMDA/glycine (N/Gly; 1 mM/10 µM) or glutamate (glut; 50 µM) all elicited a significant decrease in glutamatergic synapse density by 23, 30, and 35%, respectively (Fig. 1B) [control n = 20, TTX = 13, CNQX = 8, APV = 24, ACM = 8, NMDA = 22, N/Gly = 14, glut = 9 cells; F(7,186) = 21.20, p < 0.0001, ANOVA; control vs glut p < 0.0001, control vs ACM p = 0.1173, control vs NMDA p = 0.0249, control vs N/Gly p = 0.0058, control vs APV p = 0.2799]. Spontaneous activity is developmentally regulated in cortical sensory areas (Okawa et al., 2014), so we next determined if activity regulates synapse density in a similar way at later ages. Neither stimulating with glutamate or activators of NMDARs nor blocking with ACM or APV elicited changes in synapse density in older cultures [14 DIV; Fig. 1C,D; control n = 9, TTX = 8, APV = 9, ACM = 12, N/Gly = 10, glut = 8 cells; F(6,56) = 0.5383, p = 0.7463, ANOVA]. Thus, activation of NMDARs rapidly and negatively alters glutamatergic synapse density selectively during the period of initial establishment of cortical synapses.
Activity rapidly regulates cortical synapse density in young neurons. A, Representative confocal images of dendrites show the density of synapses, determined by colocalization between postsynaptic GluN2A and GluN2B puncta (“green”) and presynaptic vGlut1 puncta (“red”), from cortical neurons at 6–7 DIV incubated with ACSF (control; left panels), APV (middle panels), or glutamate (glut: right panels) for 10 min before fixation. B, Glutamatergic synapse density was significantly increased following blockade of NMDARs with APV for 10 min and decreased following activation of NMDARs with NMDA, NMDA plus glycine (N/Gly), or glut. In contrast, a cocktail of APV, CNQX, and MCPG (ACM) or TTX or CNQX treatment alone did not change synapse density compared with control. **p < 0.01, ANOVA. C, Representative confocal images of dendrites from 14 DIV cortical neurons following 10 min treatment with ACSF (control), APV, or glutamate. D, Glutamatergic synapse density was unaltered following acute blockade or activation of NMDARs for 10 min in older cultures. E, Representative traces of mEPSCs from 7 DIV cortical neurons incubated for 10 min with ACSF (control), APV, or glutamate. F, Acute blockade of NMDARs with APV significantly increased mEPSC frequency, whereas acute activation of NMDARs with glutamate significantly decreased it. *p < 0.05, **p < 0.01, t test. Data are plotted as mean ± SEM, normalized to control cells from the same culture/experiment (“red dotted line”). Scale bar, 5 µm.
Since excess glutamate can be excitotoxic to neurons, we next examined whether acute application of glutamate-induced cell death under our experimental conditions. Six days in vitro cortical neurons, expressing GluN1-DsRed, were treated with glutamate (50 µM) for 10 min and then loaded with the cell death fluorescent marker SYTOX Green (0.5 µM). This DNA-binding dye stains the nuclei of cells with a compromised plasma membrane and therefore represents an ideal indicator of cell death (Evans and Cousin, 2007). Following a 10 min incubation of cells with glutamate, no detectable SYTOX Green fluorescence was observed. Neuronal health was confirmed by examining the neurons under bright-field illumination. Neurons were round and translucent in appearance, characteristic of healthy viable cells (“data not shown”).
Next, to determine if manipulation of NMDAR activation alters the number of functional synapses, in addition to its effects on synapse density, whole-cell patch-clamp recording of spontaneous mEPSCs was performed. Following blockade of NMDARs for 10 min with APV, mEPSC frequency increased by 55% on average, compared with control neurons incubated with ACSF (Fig. 1E,F; control n = 13, APV = 16 cells; t = 2.18, DF = 27, p = 0.0381, t test unpaired). Conversely, and with a similar bidirectional result to synapse density measured with ICC, glutamate stimulation induced a significant reduction in mEPSC frequency by 46% (Fig. 1E,F; control n = 14, glut = 13 cells; t = 2.38, DF = 25, p = 0.0253, t test unpaired). There was no change in mEPSC amplitude under any condition (“not shown”). Taken together, these results provide evidence that during the initial stages of synapse formation, glutamate release can rapidly, within minutes, alter the density of synapses on cortical neurons.
Glutamate rapidly alters NMDAR transport and surface expression
Because synapse formation requires transport of NMDARs in NRTPs to sites of contact (Washbourne et al., 2002), it is possible that the rapid glutamate-induced decrease in synapse density could be due to alterations in NRTP transport (Fig. 2A–H). To test this hypothesis, young cortical neurons at 4 DIV were transfected with GluN1-DsRed and time-lapse imaged 24 h later, 10 min before, and 10 min during ACSF, ACM, or glutamate treatment (Fig. 2A–C). Trafficking of tagged NMDARs mimics that of endogenous protein since overexpression of tagged GluN1 constructs in young cortical neurons does not change the density or intensity of NMDAR puncta in transfected compared with neighboring untransfected neurons, and NRTP transport is not changed by the location (N or C terminal) or the type of tag (EGFP or DsRed; Washbourne et al., 2002). The mobile fraction of NRTPs, calculated as the percent of total GluN1-DsRed puncta that moved >2 µm during the imaging period, was substantially altered following 10 min incubation with either glutamate receptor antagonists or agonists. Acute incubation with APV for 10 min increased NRTP mobility by roughly 29% compared with that of control ACSF-treated neurons (Fig. 2G), while ACM elicited a significant increase in the number of mobile NRTPs ∼64% (Fig. 2B,G), indicating a role for NMDAR activation and possibly also metabotropic glutamate receptors in regulating NRTP transport. Conversely, acute glutamate exposure significantly decreased the proportion of mobile NRTPs to <50% of control (Fig. 2C,G), as did treatment with NMDA and N/Gly, which reduced mobility by 45 and 61%, respectively [Fig. 2G; control n = 8, CNQX = 10, APV = 9, ACM = 8, NMDA = 5, N/Gly = 7, glut = 9 cells; F(7,56) = 32.55, p < 0.0001, ANOVA: control vs CNQX p = 0.8417, control vs APV p = 0.0393, control vs ACM p < 0.0001, control vs NMDA p = 0.0037, control vs N/Gly p < 0.0001, control vs glut p = 0.0007]. In fact, the effects of glutamate were so striking that cessation of NRTPs occurred almost instantaneously following the addition of glutamate (Fig. 2C), suggesting that mobile NRTPs can sense and fully respond to ambient glutamate almost immediately (Washbourne et al., 2004).
NMDAR activation rapidly regulates NRTP transport. A–C, Time-lapse images of dendrites from GluN1-DsRed-expressing 5 DIV cortical neurons before and during acute (10 min) treatment (indicated by the red line) with (A) ACSF (control), (B) a cocktail of glutamate receptor inhibitors (ACM), or (C) glutamate. Arrowheads indicate mobile (“open arrowheads”) and stable (“closed arrowheads”) NRTPs moving throughout the dendrite. Time is in seconds. D–F, Representative confocal images of 5 DIV cortical neurons expressing SEP-GluN1-1a, fixed and stained with antibodies against GFP in nonpermeabilizing conditions, following 10 min treatment with (D) ACSF (control), (E) ACM, or (F) glutamate. Boxed dendritic regions are magnified below each respective image. G, Blockade of NMDARs following 10 min incubation with APV or ACM significantly increased NRTP mobility, whereas acute activation of NMDARs with NMDA, N/Gly, and glutamate significantly decreased it. All values are shown normalized to NRTP mobility in the same dendritic segments prior to drug treatment (“red dotted line”). H, Acute activation of NMDARs with APV and ACM significantly increased sNMDAR expression, while N/Gly and glutamate significantly decreased it. Values are shown normalized to control (“red dotted line”). Data are plotted as mean ± SEM, normalized as indicated for each panel. *p < 0.05, **p < 0.01, *p < 0.001, ANOVA. Scale bar, 5 µm.
Another measure of receptor trafficking is their surface expression. We therefore tested whether glutamate treatment and NMDAR activation altered the number of NMDARs expressed on the surface of the plasma membrane. To reliably detect surface NMDARs (sNMDARs), an N-terminal-tagged GluN1 SEP construct (SEP-GluN1-1a) was used so that we could detect the SEP tag on the surface of the cell using antibodies against GFP under nonpermeabilized staining conditions (Ashby et al., 2004). Eight days in vitro cortical neurons that had been transfected with SEP-GluN1-1a at 5 DIV were acutely exposed to activators or inhibitors of NMDARs for 10 min and then immediately fixed and stained, under nonpermeabilizing conditions, with antibodies to GFP (Fig. 2D–F). Consistent with the bidirectional changes observed for synapse density and NRTP mobility, blockade of NMDARs with APV or with the ACM significantly increased the proportion of total NMDARs on the dendritic surface by 28 and 37%, respectively, while glutamate and N/Gly treatment significantly decreased it by 55 and 34%, respectively [Fig. 2H; control n = 17, APV = 16, ACM = 14, NMDA = 5, N/Gly = 15, glut = 11 cells; F(5,72) = 19.59, p < 0.0001, ANOVA: control vs APV p = 0.0372, control vs ACM p = 0.0037, control vs NMDA p = 0.3889, control vs N/Gly p = 0.0082, control vs glut p < 0.0001]. Taken together, these results strongly suggest that activity may rapidly regulate synapse density in young neurons via rapid changes in NMDAR mobility and surface expression.
Glutamate-induced alterations in NRTP transport require Ca2+ influx through NMDARs
NMDARs are cation channels, permeable to both sodium (Na+) and Ca2+ ions. NMDAR-mediated influx of Ca2+ initiates a second messenger cascade responsible for regulating a diverse array of neuronal functions during development including neurite outgrowth and long-term potentiation and long-term depression (Cummings et al., 1996; Connor et al., 1999; Zucker, 1999; Spitzer, 2002). We therefore determined whether alterations in NRTP mobility following glutamate treatment were dependent on Ca2+ influx following NMDAR activation. First, we imaged NMDAR-mediated Ca2+ influx elicited by glutamate using the high-affinity cell-permeant Ca2+ indicator Fluo-4 AM in cortical neurons expressing GluN1-DsRed to simultaneously monitor NRTP transport and changes in intracellular Ca2+. Four days in vitro cortical neurons were transfected with GluN1-DsRed and then loaded with Fluo-4 24 h later. Time-lapse imaging was then performed 10 min prior to and during acute application of the ACM inhibitor (Fig. 3A) or glutamate (Fig. 3B). Upon addition of ACM to prevent NMDAR activation and Ca2+ influx, NRTP mobility increased, as expected, while no changes in Fluo-4 fluorescence were observed (Fig. 3A,C). In contrast, glutamate elicited a sustained intracellular Ca2+ rise that remained elevated throughout the 10 min imaging period. Simultaneous monitoring of NRTP mobility revealed that this Ca2+ elevation occurred immediately prior to cessation of mobility (Fig. 3B,D). To further confirm that Ca2+ influx was responsible for inhibition of NRTP mobility following glutamate stimulation, neurons were preincubated with BAPTA-AM, a Ca2+ chelator. For this experiment, all bars show the NRTP mobility normalized to the “pretreatment” mobility as indicated by the red dotted line. In the presence of BAPTA, NRTP mobility remained unchanged following acute glutamate exposure compared with the 67% reduction in mobility observed with glutamate (Fig. 3E; glut n = 4, glut + BAPTA = 5, glut + APV = 5, APV before glut = 7, glut + CNQX = 6 dendrites; glut + BAPTA: t = 2.162, DF = 11.01, p < 0.0001, ANOVA, glut vs BAPTA + glut p = 0.0023, glut vs APV + glut p = 0.0036, glut vs APV before glut p < 0.0001, APV before glut vs glut + CNQX p = 0.0017). In addition to Ca2+ influx, the ability of glutamate to alter NRTP transport also depended on its ability to bind to the receptor, since incubation with glutamate in the presence of APV to block NMDARs fully prevented glutamate-induced inhibition of NRTP transport. CNQX had no significant effect compared with glutamate treatment alone, suggesting that influx through AMPARs was not required (Fig. 3E; glut vs glut + CNQX p = 0.2610). Thus, glutamate requires Ca2+ influx through NMDARs to rapidly decrease NRTP transport.
The effects of glutamate on NRTP mobility are NMDAR-dependent and require Ca2+ influx. A, B, Representative time-lapse images of 5 DIV cortical neurons expressing GluN1-DsRed (“left column”) and loaded with Fluo-4 AM (“right column”), to simultaneously monitor NRTP mobility and intracellular Ca2+ in the same cell, before and during 10 min ACM (A) or glutamate (B) treatment indicated by the red line. The magnitude of the changes in Fluo-4 signal is shown in pseudocolor. Scale bar, 5 µm. C, D, The increase in NRTP mobility (“black bars”) following acute exposure to ACM (C) occurred in the absence of an increase in Ca2+ (F/F0, “red trace”), whereas the inhibition of NRTP mobility following acute treatment with glutamate (D) occurred immediately following influx of Ca2+ through NMDARs. Note the timescale is extended postglutamate in D to monitor whether the glutamate effect washed out. E, BAPTA-AM and APV treatment prevented the glutamate-induced reduction in NRTP mobility, whereas CNQX was without effect. In the APV before glut condition, APV was added 2 min prior to glutamate addition. Data are plotted as mean ± SEM, normalized to GluN1 mobility in the same dendritic segments prior to drug treatment (“red dotted line”). **p < 0.01, ****p < 0.0001, ANOVA. Unless indicated, drug concentrations are as in Figure 1.
Glutamate acts locally to alter NRTP transport
We next tested whether local changes in glutamate, as may occur when axons approach dendrites during synapse formation, also alter NRTP transport. Using focal perfusion, pipettes, pulled to a tip diameter of 1–2 µm, were positioned ∼10 µm from a dendritic segment expressing GluN1-DsRed. A suction pipette was positioned upstream of the application pipette, enabling a small, confined stream of glutamate or ACSF to be locally applied to a limited section of dendrite while avoiding other dendrites or the cell soma (Fig. 4A–C). ACSF (Fig. 4A), glutamate (500 µM; Fig. 4B), or glutamate in the presence of APV (Fig. 4C) was applied focally by pressure application (2 psi) for five 2 s pulses at 30 s intervals for 10 min. For each experiment, neighboring, nonfocally perfused regions of the same dendrite were used for comparison; please note that data were normalized to pretreatment NRTP mobility (Fig. 4D).
Glutamate acts locally to alter NMDAR mobility in young cortical neurons. A–C, Representative images showing a dendritic segment expressing GluN1-DsRed focally perfused (“gray box”) with control ACSF (A), glutamate (B), or glutamate with APV (C) or neighboring nonperfused dendrites (“orange box”). Five puffs of these solutions (2 s duration at 30 s intervals) were perfused onto limited sections of dendrites for 10 min (pipette tip) using a Picospritzer. The perfusion area is indicated by the Alexa-488 plume. Boxed regions at higher magnification below each image show changes in NRTP mobility before and during 10 min of puffing, indicated by the red line. Arrowheads indicate mobile NRTPs (“open arrowheads”) and stable NMDAR clusters (“closed arrowheads”) throughout the dendrite. Time in seconds. Scale bar, 5 µm. D, Focal perfusion of glutamate locally decreased NRTP mobility (“black bars”) compared with neighboring, nonperfused dendritic segments (“open bars”). Perfusion of ACSF alone (control) had no effect on NRTP mobility, but perfusion of APV prevented local inhibition of NRTP transport during focal perfusion of glutamate. Data are plotted as mean ± SEM, normalized to GluN1 mobility within dendrite before drug treatment (“control black bar”; “red dotted line”). ***p < 0.001, ANOVA. E, Focal perfusion of glutamate does not attract NMDARs toward or disperse NMDARs away from the focal stream in neighboring, nonperfused dendritic segments. Data are plotted as mean ± SEM, normalized to control mobility prior to focal perfusion (“red dotted line”).
Focal perfusion of ACSF did not itself alter NRTP trafficking since NMDARs moved freely, exhibiting normal mobility and bidirectional transport behavior throughout the focally perfused segment of dendrites (Fig. 4A). However, perfusion of glutamate onto limited sections of dendrites induced local inhibition of NRTP mobility by 35% specifically within the perfused region, compared with neighboring, nonfocally perfused regions (Fig. 4B,D). APV (500 µM), when perfused together with glutamate, prevented the inhibitory effect of glutamate on NRTP mobility, presumably by blocking NMDAR-mediated Ca2+ influx; NRTPs continued to travel bidirectionally throughout the dendrite in this condition [Fig. 4C,D; control (ACSF) n = 5, glut = 6, glut + APV = 5 dendrites; F(6,51) = 32.55, p = 0.0079, ANOVA]. Finally, NMDARs in the neighboring nonperfused regions were not attracted to glutamate since there were no changes in directionality toward or away from the perfused region (Fig. 4E; n = 6 dendrites; t = 0.4476, DF = 10, p = 0.6640, t test Welch's correction). Thus, activity can act locally to rapidly alter NRTP dynamics prior to synaptogenesis.
Glutamate inhibits trafficking of NL1 and its cotransport with NRTPs
The cotransport of NRTPs with NL1 is critical for the rapid recruitment of NMDARs to nascent synapses as they form during development (Barrow et al., 2009; Budreck et al., 2013). Given that glutamate inhibits NRTP mobility, we hypothesized that glutamate would also decrease the cotransport of NRTPs with NL1 and potentially even their colocalization. To test this idea, 4–5 DIV cortical neurons were cotransfected with GluN1-DsRed and eGFP-NL1 and time-lapse imaged 24 h later to visualize the dynamics of NRTPs and NL1 puncta simultaneously. Trafficking of tagged NL1s likely mimics that of endogenous protein since eGFP-NL1 can bind β-NRX and can induce the formation of presynaptic terminals onto transfected nonneuronal cells (Fu et al., 2003) and its dendritic localization mimics the distribution of endogenous NL1 (Barrow et al., 2009). NRTP and NL1 puncta exhibited dynamic behavior, rapidly moving bidirectionally throughout dendritic compartments (Fig. 5A,B), as previously reported (Washbourne et al., 2002, 2004; Barrow et al., 2009). In addition to rapid inhibition of NRTP mobility, acute application of glutamate (50 µM) dramatically and rapidly decreased NL1 mobility [Fig. 5C; ACSF n = 6, ACM = 15, N/Gly = 9, glut = 9 dendrites; F(3,35) = 14.00, p < 0.0001, ANOVA: control vs ACM p = 0.6229, control vs N/Gly p = 0.0825, control vs glut p = 0.0009] and caused an 87% reduction in NRTPs cotransported with NL1 [Fig. 5D; F(3,35) = 11.20, ANOVA: control vs ACM p = 0.6876, control vs N/Gly p = 0.0271, control vs glut p = 0.0086] without significantly altering overall NL1/GluN1 colocalization [Fig. 5E; F(3,35) = 4.761, p < 0.0001, ANOVA: control vs ACM p = 0.9858, control vs N/Gly p = 0.0770, control vs glut p = 0.1108]. Similarly, 10 min treatment with N/Gly significantly reduced the cotransport of NRTPs with NL1 by 74% (Fig. 5D; stats reported above). Thus, the inhibitory effect of glutamate is not restricted to NRTPs, but also rapidly alters the transport of other synaptic proteins involved in synaptogenesis. The loss in mobility between colocalized NL1 and NRTPs could explain the rapid loss in synapse density observed following acute glutamate treatment. In contrast, bath application of the ACM failed to have any effect on NL1 mobility or its colocalization or cotransport with NRTPs, suggesting that the population of NRTPs that exhibit a rapid increased mobility following ACM treatment (Fig. 2G) are likely not associated with NL1 (Fig. 5C–E). These results demonstrate that rapid alterations in glutamate release can specifically disrupt the cotransport of NRTPs and NL1, thereby preventing their recruitment or stabilization at nascent synapses.
Glutamate inhibits the transport of NL1 and its cotransport with NMDARs. A, B, Time-lapse images of 5 DIV cortical neurons simultaneously expressing GluN1-DsRed (“left column”) and GFP-NL1 (“central column”) to visualize colocalization (“overlay,” “right column”) and cotransport, before and during (“red line”) ACSF (A) or glutamate (B) treatment. Open and closed arrowheads indicate mobile and stationary clusters, respectively. Scale bar, 5 µm. C, Acute glutamate treatment, but not ACM or NMDA/glycine significantly decreased NL1 mobility. D, Acute glutamate and NMDA/glycine treatment, but not ACM, significantly decreased NL1/GluN1 comobility. E, NL1 colocalization with GluN1 was not significantly altered by any of the treatments. The same samples were analyzed for graphs in C–E. Data are plotted as mean ± SEM. Data were normalized to measures in the same dendritic segment before treatment (“red dotted line”). *p < 0.05, **p < 0.01, ***p < 0.001, ANOVA.
Decreased NL1 is necessary for glutamate-induced synapse loss
To determine if acute glutamate-induced synapse loss requires changes in NL1 levels, we switched our assay to mouse neurons so we could use a genetic deletion of NL1 (Varoqueaux et al., 2006; Blundell et al., 2010). In our hands, mouse cultures have lower synaptic density compared with the rat cultures used in earlier figures. First, we confirmed that 8 DIV occipital WT mouse neurons show the glutamate-induced reduction in excitatory synapses (Fig. 6A,B; control n = 34 cells, glut = 23 cells; t = 3.181, DF = 53.22, p = 0.0024, t test Welch's correction). Second, cultured WT neurons were treated with 10 min of glutamate at 8 DIV and surface proteins were labeled using cell-impermeable biotin before fixation. Glutamate treatment caused a 41% reduction in sNL1 (Fig. 6C,D; N = 4; t = 3.376, DF = 3, p = 0.0432, paired t test). Next, we tested whether acute glutamate treatment requires this decrease in NL1 to cause synapse loss, since NL1 is well-known to stabilize synapses at older ages (Sudhof, 2018). If so, then preventing the glutamate-induced decrease in NL1 should prevent synapse loss. Cortical neurons were transfected with a construct to overexpress NL1 (NL1-mCherry) at 3 DIV, treated with or without glutamate for 10 min at 8 DIV, and synapses were quantified as above. Consistent with our hypothesis, preventing the decrease in NL1 by overexpression of NL1-mCherry rescued the glutamate-induced synapse loss (Fig. 6E,F; control n = 26, glut = 23 cells; t = 0.9304, DF = 45.10, p = 0.3571, t test Welch's correction), suggesting that excess NL1 stabilizes a higher proportion of excitatory synapses to offset the effects of glutamate either through increased adhesion or possibly closing a developmental window where glutamate can reduce synapse numbers. Regardless of mechanism, decreases in NL1 levels are clearly necessary for the effects of glutamate in causing synapse loss.
NL1/Nrxn binding prevents glutamate-induced internalization of sNMDARs and synapse loss. A, Representative confocal images of WT occipital cortical mouse neurons at 8 DIV. Neurons were fixed and stained with antibodies to identify glutamatergic synapses defined as colocalized puncta of postsynaptic PSD-95 (“green”) and presynaptic VAMP2 (“red”). Scale bar, 5 µm. B, Treatment with 10 min of glutamate significantly reduced synapse density in WT murine neurons. Data are plotted as mean ± SEM, normalized to control cells from the same culture/experiment (“red dotted line”). **p < 0.01, t test. C, Eight DIV WT neurons were incubated with glutamate and then labeled with cell-impermeable biotin. Biotinylated proteins were collected on streptavidin agarose and run for Western blot. Surface NL1 (sNL1) was significantly reduced following glutamate treatment when compared with the vehicle-treated (control) condition. Data are plotted as mean ± SEM, normalized to the control condition (“red dotted line”) *p < 0.05, paired t test. D, Western blot showing similar total levels of NL1 compared with reduced levels of sNL1, with actin as both a loading and cytosolic control. E, Representative confocal images of occipital WT cortical mouse neurons transfected with NL1-mCherry that were incubated with glutamate or control for 10 min at 8 DIV and then fixed and stained with antibodies against VAMP2 and PSD-95. Synapses were defined as colocalized postsynaptic PSD95 (“green”) and presynaptic VAMP2 puncta (“red”). F, Glutamatergic synapse density was not altered following 10 min glutamate treatment of occipital WT mouse cultures. Data are plotted as mean ± SEM, normalized to the values from control cells from the same culture/experiment (“red dotted line”). G, Density of sGluN1 puncta was calculated for dendritic segments contacting HA-positive HEK cells expressing HA-CD4 (“open bar”) or HA-Nrxn-1β (“black bar”) normalized to the HA-CD4 condition (“red dotted line”). HA-Nrxn-1β-contacting dendrites showed a significantly increased density of sGluN1 puncta. *p < 0.05, t test. H, Representative maximum projection images of HEK cells expressing HA-CD4 (control) or HA-Nrxn-1β cocultured with cortical neurons at 6 DIV expressing SEP-GluN1-1a, fixed and immunostained for GFP (“green”) and HA (“red”) following 10 min treatment with ACM or glutamate. Scale bar, 5 µm. I, Glutamate and NMDA/glycine, but not ACSF or ACM, treatment caused a loss of surface GluN1 in sections of dendrites that were not in contact with HA-Nrxn-1β expressing HEK cells (“open bars”). This glutamate-induced loss is prevented at contacts with HEK cells expressing HA-Nrxn-1β (“black bars”). Density is normalized to sGluN1 density in noncontacting dendrites from vehicle-treated cells (control; “red dotted line”). *p < 0.05, ***p < 0.001, ANOVA. J, Representative confocal images of occipital NL1-KO cortical mouse neurons treated with glutamate and stained for PSD95 (“green”) and VAMP2 puncta (“red”). Scale bar, 5 µm. K, Glutamatergic synapse density in NL1-KO cultures was significantly decreased following 10 min of glutamate treatment. Data are plotted as mean ± SEM, normalized to the control cells treated with vehicle from the same culture/experiment (“red dotted line”). *p < 0.05, t test.
Nxrn/NL1 adhesion counteracts the inhibitory effects of glutamate
So far, our results suggest that NL1 opposes the effects of glutamate in causing synapse loss and NMDAR internalization. This model is similar to the role for ACh at the NMJ where ACh destabilizes postsynaptic sites through the dispersal of AChRs (Misgeld et al., 2005). This effect is counteracted by an antidispersal signal in the form of agrin, which prevents neurotransmitter-induced AChR loss (Misgeld et al., 2005). Since NL1 mediates the trafficking and recruitment of NMDARs to nascent synapses (Fu et al., 2003; Barrow et al., 2009), and since it requires binding to its presynaptic ligand Nrxn1 in order to initiate synapse formation (Scheiffele et al., 2000; Chih et al., 2005; Tsetsenis et al., 2014; L. Y. Chen et al., 2017), we reasoned that NL1 and its interaction with Nrxn1 may act as a similar antidispersal signal to protect NMDARs against the effects of glutamate.
To test this hypothesis, we utilized a hemisynapse assay (Biederer and Scheiffele, 2007). Cortical neurons at 6 DIV were cocultured with HEK293 cells expressing HA-Nrxn-1β in the presence or absence of glutamate or N/Gly to activate, or ACM to block, NMDARs, respectively. In this classic hemisynapse assay, postsynaptic specializations are induced to cluster along dendrites at sites of contact with cocultured HEK cells expressing the synaptogenic molecule Nrxn1 (Graf et al., 2004; Biederer and Scheiffele, 2007). We used the splice variant of Nrxn1β lacking an insert at Splice Site 4 that potently binds with postsynaptic NL1 (Ko et al., 2009). HA-Nrxn-1β-induced clustering of postsynaptic GluN1 was assessed using antibodies to GFP to detect surface-expressed SEP-GluN1-1a and to HA to detect CD4- or Nrxn1-expressing HEK cells. As expected (Graf et al., 2004; Biederer and Scheiffele, 2007), GluN1 was induced to cluster selectively at sites of contact with Nrxn1-expressing HEK cells, compared with HEK cells expressing HA-CD4 control (Fig. 6G,H; HA-CD4 n = 12, HA-Nrxn-1β n = 6 dendrites; t = 2.709, DF = 6.452, p = 0.0327, t test Welch's correction). Also, as expected from data presented above, acute incubation with glutamate caused a 31% loss in surface GluN1 puncta in dendritic segments that were not in contact with HEK cells. Importantly, this glutamate-induced loss of GluN1 was prevented at sites of contact with Nrxn1-expressing HEK cells. Similarly, the 38% loss of surface GluN1 induced by N/Gly was also prevented in Nrxn1-contacting dendrites compared with the noncontacting dendrites [Fig. 6I; control n = 16, ACM = 23, N/Gly = 14, glut = 23 cells; F(3,57.77) = 3.876, p = 0.0136, ANOVA: (uncontacted) control vs ACM p = 0.4715, control vs N/Gly p = 0.0087, control vs glut p = 0.028); F(3,49.53) = 1.772, p = 0.1646, ANOVA: (contacted) control vs ACM p = 0.5613, control vs N/Gly p = 0.0642, control vs glut p = 0.9411]. Thus, binding of NL1 to Nrxn1 stabilizes the complex and protects NMDARs from dispersal by glutamate, similar to the role for agrin as an antidispersal signal that protects AChRs from ACh-induced dispersal at the NMJ (Misgeld et al., 2005).
Finally, if our hypothesis is true, then glutamate should still cause synapse loss in neurons lacking NL1 because of the lack of this antidispersal signal. Indeed, 10 min of glutamate treatment induced a significant 25% reduction in excitatory synapses in 8 DIV NL1-KO neurons compared with control NL1-KO neurons (Fig. 6J,K; NL1-KO control n = 42, NL1-KO glut = 27 cells; t = 2.269, DF = 65.59, p = 0.0266, t test Welch's correction). Thus, Nrxn-NL1-induced NMDAR puncta are resistant to glutamate-induced loss, suggesting that an interaction across the synaptic cleft protects and stabilizes developing synapses from the effect of glutamate release. Together, these results indicate that NL1 is a critical component of the antidispersal signal that opposes glutamate-induced synapse loss in cortical neurons during the early stages of synapse formation.
MHCI is necessary for glutamate-induced synapse loss through negatively regulating NL1
To further dissect the molecular mechanisms that mediate the effects of glutamate on synapse density, we hypothesized that other molecules present at the synapse that cause synapse loss in an activity-dependent manner may play a role. MHCI molecules are optimal candidates since they are activity regulated, expressed in neurons at synapses (Corriveau et al., 1998; Huh et al., 2000; Needleman et al., 2010), and negatively regulate synapse density selectively during the period of the initial establishment of connections when glutamate rapidly alters synapse number (Needleman et al., 2010; Glynn et al., 2011). Importantly, little is known about how MHCI causes synapse loss (Elmer and McAllister, 2012; Cameron and McAllister, 2018).
If glutamate requires MHCI to cause synapse loss, then we would expect that acute glutamate treatment would increase MHCI levels in neurons and especially surface MHCI (sMHCI) which is required for synapse loss (Glynn et al., 2011). WT occipital 8 DIV cultures were treated with 50 μM glutamate and sMHCI levels were quantified. While total MHCI levels are clearly regulated by activity over longer timeframes of days (Glynn et al., 2011), acute effects of altering neural activity on MHCI levels are unknown. Although 10 min of glutamate did not significantly alter sMHCI density (Fig. 7A,B; control n = 21, glut n = 27 cells; t = 0.2501, DF = 40.41, p = 0.8037, t test Welch's correction), it did significantly increase the intensity of sMHCI puncta (Fig. 7C,D; control n = 21 cells, 922 puncta, glut n = 27 cells, 1,158 puncta; average intensity: t = 3.009, DF = 30.65, p = 0.0052, t test Welch's correction; cumulative probability intensity histogram: Kolmogorov–Smirnov test, D = 0.1263, p < 0.0001). Thus, MHCI levels are increased by acute glutamate treatment.
An increase in sMHCI is required for glutamate-induced synapse loss. A, Representative confocal images of nonpermeabilized WT occipital cortical mouse neurons at 8 DIV. Neurons were fixed and stained with antibodies to identify sMHCI (“magenta”) in the absence of permeabilization as defined as no MAP2 (“blue”) colocalization. Scale bar, 5 µm. B, There was no significant change in sMHCI punctal density in response to an acute 10 min treatment with glutamate. Data are plotted as mean ± SEM, normalized to control cells from the same culture/experiment (“red dotted line”). t test. C, There was a significant increase in the average punctal intensity of sMHCI in response to acute glutamate treatment. Data are plotted as mean ± SEM, normalized to control cells from the same culture/experiment (“red dotted line”). t test. D, sMHCI intensity shifts to more intense puncta in response to glutamate treatment in WT 8 DIV occipital neurons. ****p < 0.0001, Kolmogorov–Smirnov test. Data are plotted as a cumulative probability plot for control (“red”) and glutamate (“black”), with intensity normalized to control cells from the same culture/experiment. Specifically, intensity for sMHCI was averaged for each neuron in each of the three independent cultures. From that average, all intensity was normalized within culture, thus reflecting a control average of “1.” E, Representative confocal images of MHCI-DKO occipital cortical mouse neurons at 8 DIV that were incubated with glutamate for 10 min and then fixed and stained with antibodies to postsynaptic PSD95 (“green”) and presynaptic VAMP2 puncta (“red”); glutamatergic synapses are defined as colocalized puncta. Scale bar, 5 µm. F, Glutamatergic synapses were not significantly altered by glutamate treatment in the absence of H2-Kb and H2-Db. G, Representative confocal images of β2m-KO occipital cortical mouse neurons at 8 DIV that were incubated with glutamate for 10 min and then fixed and stained with antibodies to postsynaptic PSD95 (“green”) and presynaptic VAMP2 puncta (“red”); glutamatergic synapses are defined as colocalized puncta. Scale bar, 5 µm. H, Glutamatergic synapses were not significantly altered by glutamate treatment in the absence of sMHCI via genetic deletion of β2m.
Our lab has previously demonstrated that elevations in MHCI cause a loss of NMDARs and glutamatergic synapses selectively during the same early developmental period that glutamate rapidly alters synapses (Glynn et al., 2011; Elmer et al., 2013). To test whether increases in MHCI are necessary for glutamate-induced synapse loss, 8 DIV neurons from mice that lack the two major classical forms of MHCI, H2-Kb and H2-Db (MHCI-DKO), or lacking sMHCI (β2m-KO; Needleman et al., 2010; Glynn et al., 2011) were treated with glutamate or control media for 10 min before fixing and staining for PSD-95 and VAMP2 (Fig. 7E–H). Glutamate failed to reduce synapse density in neurons that lacked H2-Kb and H2-Db (Fig. 7E,F; MHCI-DKO control n = 29, MHCI-DKO glut = 28 cells; t = 0.4308, DF = 55.05, p = 0.6683, t test Welch's correction) or sMHCI (Fig. 7G,H; β2m-KO control n = 20, β2m-KO glut = 20 cells; t = 0.4261, DF = 36.59, p = 0.6725, t test Welch's correction), showing the necessity of the two major forms of classical MHCI in C57BL/6 mice and all sMHCI for glutamate-induced synapse loss in early cortical cultures.
Because both MHCI and NL1 are required for glutamate-induced synapse loss, we next asked if MHCI acts upstream of NL1 to mediate glutamate-induced changes in synapse density. First, we determined if altering MHCI levels negatively regulates NL1 using cultured cortical neurons and a lentiviral strategy to overexpress MHCI H2-Kb-YFP from 3 to 8 DIV, with GFP or NL1 shRNA as a negative and positive control, respectively. When measured via Western blot and normalized to actin as a loading control, H2-Kb-YFP significantly decreased NL1 protein levels by 34% compared with GFP expression alone [Fig. 8A,B; n = 4 cultures; F(2,3,6) = 27.48, p = 0.001, ANOVA, repeated measures: H2-Kb-YFP vs GFP p = 0.0229; H2-Kb-YFP vs NL1 shRNA p = 0.0238, GFP vs NL1 shRNA p = 0.0008]. Using the β2m-KO mouse in which most forms of classical MHCI are not expressed on the surface of cells (Needleman et al., 2010), we harvested and lysed sex-matched littermate hippocampi for Western blot analysis (Fig. 8C). Removal of sMHCI significantly increased NL1 by 162% compared with WT controls (Fig. 8D; n = 5 littermate pair hippocampi; t = 4.141, DF = 4, p = 0.0144, paired t test), indicating that NL1 is bidirectionally modulated by MHCI.
MHCI mediates glutamate-induced synapse loss by negatively regulating NL1 levels. A, Western blot of samples from lysed 8 DIV WT cortical neurons that were cultured and infected with lentivirus-expressing GFP, H2-Kb-YFP, or NL1 shRNA at 3 DIV. B, Protein levels normalized to actin from Western blots as in (A) indicate that overexpression of H2-Kb-YFP significantly reduced NL1 total protein levels, as did NL1 shRNA. **p < 0.01, ANOVA. C, Western blot of samples from lysed hippocampi from sex-matched littermate pairs of WT and β2m-KO mice. D, Protein levels normalized to actin from Western blots as in (C) indicate that loss of sMHCI expression in the β2m-KO mice significantly increased NL1 total protein. *p < 0.05, t test. E, Representative confocal images of WT occipital cortical mouse neurons transfected with H2-Kb-YFP that were incubated with glutamate for 10 min at 8 DIV and then fixed and stained with antibodies to postsynaptic PSD95 (“green”) and presynaptic VAMP2 puncta (“red”); glutamatergic synapses are defined as colocalized puncta. Scale bar, 5 µm. F, Glutamatergic synapses were significantly altered by glutamate treatment in the presence of excess H2-Kb-YFP. **p < 0.01, t test. G, WT neurons were transfected either with mCherry or H2-Kb-mCherry for direct comparison of synaptic changes in response to glutamate. Neurons were incubated with glutamate for 10 min at 8 DIV and then fixed and stained with antibodies to postsynaptic PSD95 (“green”) and presynaptic VAMP2 puncta (“red”); glutamatergic synapses are defined as colocalized puncta. Glutamatergic synapses were significantly decreased by glutamate treatment in the presence of either mCherry or H2-Kb-mCherry, while overexpression of H2-Kb-mCherry alone was sufficient to decrease synapse density to an intermediate level. *p < 0.05, **p < 0.01, ***p < 0.001, Two-way ANOVA. Scale bar, 5 µm. H, Representative confocal images of WT occipital cortical mouse neurons transfected with H2-Kb-YFP and NL1-mCherry that were incubated with glutamate for 10 min at 8 DIV and then fixed and stained with antibodies to postsynaptic PSD95 (“green”) and presynaptic VAMP2 puncta (“red”); glutamatergic synapses are defined as colocalized puncta. Scale bar, 5 µm. I, Glutamatergic synapses were not altered by glutamate treatment in the presence of excess H2-Kb-YFP and excess NL1-mCherry.
Given that NL1 protein is negatively modulated by MHCI (Fig. 8A–D) and NL1/Nrxn binding stabilizes synapses (Fig. 6F,G), glutamate would be expected to cause synapse loss in neurons with elevated MHCI levels, phenocopying the effects of glutamate in NL1-KO neurons (Fig. 6I). As predicted, there was a significant reduction in synapse density in WT occipital cortical mouse neurons that overexpressed H2-Kb-YFP after 10 min 50 μM glutamate treatment (Fig. 8E,F; H2-Kb-YFP control n = 23, H2-Kb-YFP glut = 25 cells; t = 2.927, DF = 50.66 p = 0.0051, t test Welch's correction), reinforcing the necessity of NL1 in preventing glutamate-induced synapse loss. Next, an additional set of experiments was performed to compare the magnitude of glutamate-induced synapse loss in control and MHCI-overexpressing neurons. As we previously reported (Glynn et al., 2011; Elmer et al., 2013), overexpression of H2-Kb-mCherry decreased synapse density compared with mCherry alone (Fig. 8G). Importantly, treatment with glutamate decreased synapse density in both mCherry- and H2-Kb-mCherry-overexpressing neurons to a similar level, indicating that the effects of MHCI overexpression and glutamate on synapse loss are not additive, which is consistent with the interpretation that they act in an overlapping pathway to control synapse density [Fig. 8G; two-way ANOVA: treatment F(1,62) = 35.59 p < 0.0001, transfection F(1,62) = 3.385 p = 0.0706, treatment × transfection interaction F(1,62) = 5.06, p = 0.028, individual comparisons: mCherry con vs mCherry glut p < 0.0001, mCherry con vs H2-Kb-mCherry con p = 0.0181, mCherry con vs H2-Kb-mCherry p < 0.0001, mCherry glut vs H2-Kb-mCherry con p = 0.0309, H2-Kb-mCherry con vs H2-Kb-mCherry glut p = 0.0497, mCherry glut vs H2-Kb-mCherry glut p = 0.9925].
Finally, we investigated whether MHCI requires changes in NL1 levels to mediate the effects of glutamate. Three days in vitro cortical neurons were transfected with both H2-Kb-YFP and NL1-mCherry before treatment with 10 min of glutamate or vehicle at 8 DIV and staining for PSD-95 and VAMP2 (Fig. 8H). As predicted by our model, overexpression of NL1-mCherry together with H2-Kb-YFP phenocopied NL1-mCherry overexpression alone (Fig. 6C,D) with no change in synapse density in response to glutamate (Fig. 8I; H2-Kb-YFP/NL1-mCherry control n = 27, H2-Kb-YFP glut = 27 cells; t = 0.5726, DF = 47.57, p = 0.5695, t test Welch's correction), indicating that MHCI overexpression affects glutamate-induced synapse loss through modulation of NL1 and suggesting that excess NL1 can stabilize a higher proportion of excitatory synapses even in the presence of glutamate and elevated MHCI.
Discussion
Here, we describe a novel form of homeostatic plasticity that exists in young cortical neurons to rapidly alter glutamatergic synapse density in response to changes in activity. Acute elevations in glutamate (10 min) act as a dispersal signal for NMDARs and NL1 that reduces trafficking and causes loss of synapses, selectively during a period when connections are initially forming. This dispersal signal is opposed by trans-synaptic adhesion through binding of NL1 to Nrxn1, which stabilizes synapses. The effect of glutamate in causing synapse loss requires elevations in MHCI, which negatively regulates NL1 levels. In the presence of excess NL1 (NL1 overexpression or MHCI-DKO), or at sites of NL1 binding to presynaptic Nrxn1 in the hemisynapse assay, glutamate does not cause synapse loss. In the absence (NL1-KO) or reduced presence of NL1 (MHCI overexpression), or at dendritic sites without binding of NL1 to Nrxn1, glutamate disperses NMDARs and causes glutamatergic synapse loss. Together, this glutamate–MHCI-NL1 signaling pathway represents a mechanism that could act to tightly regulate the number and location of the initial complement of synapses formed along developing dendrites.
Although homeostatic plasticity is essential for globally altering and resetting synaptic strength to counter the destabilizing effects of Hebbian plasticity in mature networks (G. Turrigiano, 2012; L. Chen et al., 2022), the effects of acute global and local changes in neural activity on the initial establishment of neural networks have been unclear. Because synapses form in the absence of neurotransmitter release (Verhage et al., 2000; Varoqueaux et al., 2002) and because synapse density is not altered by activity blockers in mature hippocampal neurons (Ko et al., 2011; G. G. Turrigiano, 2017; Andreae and Burrone, 2018), it has been assumed that neurotransmitter does not play a role in the initial formation of connections (Sudhof, 2021). In contrast, our results show that NMDAR activation in cortical networks regulates glutamatergic synapse density on a surprisingly short timescale of minutes, but only in networks that are initially establishing connections. Although our application of glutamate is unlikely to precisely mimic the concentration and especially the kinetics of glutamate in young intact tissue in our neuronal cultures, the concentration of glutamate used here is well below the concentration estimated at the peak of synaptic release (>1 mM; Clements et al., 1992; Dzubay and Jahr, 1999) and is more comparable but still below the estimated concentration of glutamate in the extrasynaptic space following vesicular release (160–190 µM; Dzubay and Jahr, 1999). The effects of glutamate were robust, leading to both a 35% decrease in glutamatergic synapse density and a 46% decrease in mEPSC frequency after only 10 min of glutamate treatment. Moreover, the glutamate-induced synapse loss occurred in both rat and mouse cultures, assessed by multiple combinations of colocalized proteins to define synapses. These changes occur on both global and local levels and are consistent with reports of high levels of spontaneous neurotransmitter release in developing tissues (Matteoli et al., 1992; Sabo and McAllister, 2003; Andreae and Burrone, 2018) as well as a lack of neurotransmitter transporters in young tissues (Thomas et al., 2011), which could allow for greater levels of ambient neurotransmitter. The rapid and selective decrease in NRTP transport specifically at sites of glutamate perfusion suggests that the local release of glutamate by axonal growth cones (Matteoli et al., 1992; Sabo and McAllister, 2003; Sabo et al., 2006) could also regulate synapse formation. This new form of homeostatic plasticity resulting in rapid changes in synapse density during the initial formation of neural networks complements the emerging idea that networks tap into distinct forms of homeostatic plasticity at different stages of development throughout the lifespan (Wierenga et al., 2006; Wen and Turrigiano, 2021). The rapid activity-induced changes in synapse density shown here may be necessary for neural networks to respond dynamically to rapidly changing input number as connections initially form, while more established networks rely on altering synaptic weights or types to maintain their network stability over longer timescales (G. Turrigiano, 2011).
The central role for NMDARs in glutamate-induced synapse loss in young neural networks makes sense given that NMDARs are often the first glutamate receptors recruited to nascent synapses (Washbourne et al., 2002). In addition, the rapid trafficking of NRTPs within dendrites and their cycling with the membrane at pause sites before and during synapse formation (Tovar and Westbrook, 2002; Washbourne et al., 2004) indicate that they are well positioned to respond to glutamate. Indeed, both constitutive and agonist-induced internalization of NMDARs have been well characterized in heterologous cells and in young neural networks (Carroll et al., 1999; Roche et al., 2001; Vissel et al., 2001; Nong et al., 2003, 2004; Sans et al., 2003). Our results extend this work by showing that this glutamate-induced internalization of NMDARs occurs in conjunction with synapse loss and changes in surface levels and cotransport of NL1, a synaptic adhesion molecule that regulates NMDAR accumulation at synapses (Chih et al., 2005; Chubykin et al., 2007; Barrow et al., 2009) and that is regulated by activity (Schapitz et al., 2010; Peixoto et al., 2012; Suzuki et al., 2012; Bemben et al., 2014). Given the likely central role of GluN2B in this process due to its abundance in the young cortex and its rapid glutamate-induced internalization in young neurons (Roche et al., 2001; Nong et al., 2004), it is possible that either the developmental subunit switch to GluN2A-containing NMDARs (Barria and Malinow, 2002) or the recruitment of synaptic scaffolding molecules such as PSD-95 (Lavezzari et al., 2004) may shift this novel form of homeostatic plasticity from altering synapse density in new networks to selectively altering synapse strength in more mature networks.
The role for glutamate in acting as a dispersal signal for NMDARs and a negative regulator of synapses in young networks is similar to the well-described role for ACh at the NMJ (Misgeld et al., 2005; Rodriguez Cruz et al., 2020). At developing NMJs, ACh causes dispersal and internalization of AChRs unless an antidispersal signal, agrin, is present (Misgeld et al., 2005). Here, we show that glutamate causes internalization of NMDARs and synapse loss unless the receptors and synapses are stabilized by NL1. Similar to the role for glutamate activation of NMDARs at the NMJ causing synapse pruning during the initial stages of early synaptogenesis (Personius et al., 2016), we found that glutamate causes internalization of NMDARs and glutamatergic synapse loss in young cortical cultures. Thus, the role for neurotransmitter in the formation of such disparate structures as the NMJ and cortical excitatory synapses appears to be conserved. Finally, it is likely that additional synaptic adhesion molecules play a similar role as NL1/Nrxn1 in opposing neurotransmitter-induced receptor internalization and synapse loss (Ko et al., 2011; Sudhof, 2018, 2021); future studies are needed to address this issue. If so, then this antidispersal function could contribute to the matching of presynaptic neurotransmitter with the accumulation of the correct type of neurotransmitter receptors at the synapse.
At first glance, our discovery that glutamate acts to cause synapse loss appears to be contradictory to evidence that neurotransmitters promote synapse formation. Many studies have shown that activity can induce dendritic spine formation in an NMDAR-dependent manner (Engert and Bonhoeffer, 1999; Maletic-Savatic et al., 1999; Toni et al., 1999; Jourdain et al., 2003; Nagerl et al., 2004) and the Sabatini lab critically showed that focal uncaging of glutamate rapidly induces the formation of new dendritic spines with functional synapses selectively in young cortical slices in an NL1-dependent manner (Kwon and Sabatini, 2011; Kwon et al., 2012). Because these uncaging experiments were performed in an intact tissue, which is tightly packed with neuronal processes, it is likely that any new dendritic spines stimulated to extend by local glutamate (Dailey and Smith, 1996; Wong and Wong, 2001; Portera-Cailliau et al., 2003; Cruz-Martin et al., 2012) will rapidly come into contact with presynaptic axons and thereby facilitate binding of NL1 to Nrxn1 to stabilize the nascent synapses (Sudhof, 2018, 2021). In contrast, the low-density culture system used in our study limits the density of neighboring axons, revealing that glutamate causes NMDAR internalization and synapse loss in the absence of presynaptic contact. Mimicking presynaptic contacts with Nrxn1-expressing HEK cells prevents glutamate-induced synapse loss.
In addition to reconciling conflicting literature about the role for glutamate in synaptogenesis, we have revealed a novel molecular mechanism—neuronal MHCI signaling—for glutamate-induced synapse loss. MHCI negatively regulates synapse density, selectively during the initial establishment of connections when glutamate rapidly alters synapse density (Glynn et al., 2011). MHCI levels are positively regulated by activity (Corriveau et al., 1998; Glynn et al., 2011; Ribic et al., 2011; Lv et al., 2015), and they contribute to synaptic scaling-like responses (Goddard et al., 2007; Glynn et al., 2011; Elmer and McAllister, 2012; McAllister, 2014). In young cortical cultures, we previously showed that blocking activity with TTX for 2 d decreases sMHCI and increases synapse density in an MHCI-dependent manner (Glynn et al., 2011). Here, we show that sMHCI levels are rapidly increased in pre-existing sites by 10 min glutamate, and that increase is required to decrease NL1 levels and cause synapse loss. This is the first demonstration that MHCI negatively regulates NL1 levels to control synapse density. Future studies will be needed to determine how MHCI alters NL1 and whether that interaction also plays a role in homeostatic plasticity of synaptic strength at later ages (Goddard et al., 2007; Glynn et al., 2011; McAllister, 2014; Adelson et al., 2016) and in synapse loss downstream of peripheral immune activation (Elmer et al., 2013; McAllister, 2014, 2017; Estes and McAllister, 2015, 2016) and during neurodegeneration (Zalocusky et al., 2021; Kim et al., 2023).
Footnotes
This work was supported by NIH R01-EY13584 (A.K.M.) and R01-NS060125 (A.K.M.) and by philanthropic support from H. Britton Sanderford, Jr. We thank Ann-Marie Craig for providing HA-CD4, Thomas Sudhof for providing HA-Nrx1βSS4−, and Dr. R. Malinow for providing SEP-GluN2A. We thank Drs. Samantha Spangler and Leigh Needleman for generating rat neuronal cultures and Faten El-Sabeawy for contributing valuable technical support.
↵*G.L.S. and S.L.B. contributed equally to this work.
The authors declare no competing financial interests.
- Correspondence should be addressed to A. Kimberley McAllister at mcallik{at}wfu.edu.
