Presynaptic NMDA receptors – dynamics and distribution in developing axons in vitro and in vivo

Presynaptic N-methyl-D-aspartate (NMDA) receptors (NMDARs) have been suggested to play roles in cortical synapse development (Corlew et al., 2007). They have also been linked to pathogenesis of developmental disorders, including epilepsy (Yang et al., 2007; Yang et al., 2006) and fetal alcohol spectrum disorders (Valenzuela et al., 2008). However, the existence of presynaptically localized NMDARs at cortical synapses has remained controversial for decades.

During cortical development, activation of NMDARs in presynaptic neurons facilitates neurotransmitter release and is required for spike-timing-dependent long-term depression (tLTD) (Berretta and Jones, 1996; Brasier and Feldman, 2008; Buchanan et al., 2012; Corlew et al., 2007; Rodríguez-Moreno and Paulsen, 2008; Sjöström et al., 2003). Although dendritic NMDARs might contribute to presynaptic plasticity (Christie and Jahr, 2008; Christie and Jahr, 2009), axonal NMDARs are necessary for tLTD (Rodríguez-Moreno et al., 2011). In addition, Ca²⁺ transients are observed in presynaptic boutons upon focal uncaging of NMDA or glutamate (Buchanan et al., 2012; McGuinness et al., 2010). However, it is not clear whether the responsible axonal receptors are distributed throughout the axon or whether they are enriched at nerve terminals (Buchanan et al., 2012). If they localize to presynaptic terminals, it is unknown which presynaptic structures contain NMDARs. Understanding where axonal NMDARs are located is important because axonal NMDARs might (1) utilize distinct signaling mechanisms, (2) be activated by different sources and levels of glutamate (e.g. synapse-autonomous glutamate release versus spill-over or ambient glutamate, or spontaneous versus evoked release), or (3) serve distinct functions (e.g. control of plasticity at the level of individual synapses, axon branches or entire arbors) based on where they reside.

NMDARs in presynaptic neurons regulate neurotransmitter release before onset of the critical period for receptive field plasticity (Corlew et al., 2007). Furthermore, axonal NMDAR expression is highest during the postnatal first two weeks, a period of intense synapse formation, but is dramatically decreased later in development (Corlew et al., 2007; Ehlers et al., 1998; Herkert et al., 1998; Song et al., 2009; Wang et al., 2011). Therefore, it has been hypothesized that presynaptic NMDARs might facilitate synapse formation (Corlew et al., 2008). If so, it would be expected that NMDARs are expressed at nascent presynaptic terminals, but it is not known whether this is the case. In addition, the specific functions of presynaptic NMDARs is defined by when and where they are present: NMDARs can participate in early steps in synaptogenesis only if they are present during those steps, whereas late arrival to presynaptic terminals would support a role in presynaptic maturation or plasticity but not initial synaptogenesis.

To address these issues, we defined the spatio-temporal dynamics of axonal NMDARs during cortical synapse formation using a combination of live time-lapse confocal microscopy of NMDAR subunits, immuno-isolation of synaptic vesicles and ultrasynaptic fractionation of presynaptic terminals. NR1 (also known as GRIN1) appeared in discrete puncta throughout the axons of many cortical neurons during periods of intense synapse formation. Some NR1 puncta were mobile, whereas others localized to presynaptic terminals. NMDARs were recruited to terminals within 24 h, and contact with neuroligin-expressing cells was sufficient to induce this recruitment. The distribution and trafficking of axonal NMDARs was insensitive to acute changes in activity. Within presynaptic terminals in vivo, NR1 associated with the presynaptic active zone membrane and, surprisingly, both glutamatergic and GABAergic synaptic vesicles. These data provide strong evidence that NMDARs are integral to developing presynaptic terminals and are in the right place at the right time to locally regulate early steps in presynaptic development.

Because the existence and localization of presynaptic NMDARs in cortical neurons has been controversial, we first determined where NMDARs are found in the axons of developing neocortical neurons. To do so, we transfected cultured visual cortical neurons at 5–6 days in vitro (DIV) with GFP-tagged rat NR1 isoform 1a. NR1 is an obligate subunit of functional NMDARs (Granger et al., 2011). Previous studies have shown that transfecting neurons with NR1 does not lead to the overexpression of NR1 or increase the number of functional NMDARs, and that GFP-tagged NR1 accurately reflects the localization of endogenous NR1 in dendrites (Bresler et al., 2004; Luo et al., 2002; Prybylowski et al., 2002; Washbourne et al., 2002). Similarly, there was no increase in NR1 immunofluorescence in dendrites of our neurons when transfected with NR1–GFP as compared to neighboring neurons within the same culture that were not transfected (mean intensity for transfected, 80.2% of control untransfected; standard deviation of intensity, 117.5% of untransfected; maximum intensity, 92.5% of untransfected). To identify axons, neurons were co-transfected with synaptophysin–mRFP (or synaptophysin–mCherry). Transfected axons were initially identified based on a combination of criteria: (1) expression of synaptophysin–mRFP in a punctate pattern typical of axons, (2) non-tapering morphology, and (3) length of the processes, because axons typically extend farther than dendrites at this stage of development. Axons were then imaged live to determine the presence of NR1–GFP. Axonal identity was ultimately confirmed through retrospective double-immunofluorescence with antibodies against MAP2 and tau-1: MAP2 is highly enriched in dendrites and nearly absent from axons, whereas tau-1 is typically enriched in axons (Bernhardt and Matus, 1984; Binder et al., 1985; Kosik and Finch, 1987). Given that the absence of MAP2 was more reliable than the enrichment in tau for identifying axons in our developing cortical neurons, most of the experiments described below used the absence of MAP2 to identify axons.

Live confocal imaging of NR1–GFP indicated that many axons expressed NR1–GFP in discrete puncta (Fig. 1A,B). Axonal NR1–GFP puncta typically appeared smaller, dimmer and less dense than dendritic NR1–GFP in the same neuron (Fig. 1C). Axonal NR1 puncta appeared at a mean density of 0.04 puncta/µm and as high as 0.06 puncta/µm. In addition, some axons had spots of NR1 enrichment that appeared brighter than the background in the rest of the axon, but were less bright and more diffuse than typical NR1 puncta and were consequently below the minimum threshold intensity for puncta detection. The density of enriched-spots plus puncta was 0.1/µm.

Endogenous NR1 also was found in puncta in axons (Fig. 1D). In order to clearly distinguish axonal NR1 from NMDARs in dendrites and glia, endogenous NR1 was examined in neurons grown without contact with glia and regions of axon that did not contact dendrites. Therefore, the punctate distribution of NR1 does not depend on contact with dendrites or glia. NR1 puncta might correspond to clusters of NR1 in the axonal plasma membrane, vesicles transporting NR1 along axons, or both.

To determine the extent to which NR1 is found at presynaptic terminals, we examined NR1–GFP and synaptophysin–mRFP in the same axons (Fig. 1B; Fig. 2A). NR1 and synaptophysin puncta were identified in each channel independently, based on image statistics (Fig. 2B). Synaptophysin colocalized with 50.2±10.3% of NR1 puncta (mean±s.e.m.; n = 77). Conversely, NR1 colocalized with 51.1±8.2% of synaptophysin puncta (n = 78). Of the colocalized puncta, 91.9% of NR1 puncta overlapped with synaptophysin and 8.1% were adjacent to synaptophysin. Endogenous NR1 also colocalized with synaptophysin within axons (Fig. 2C).

We then determined whether the NR1 and synaptophysin signals were correlated. Cross-covariograms were made for NR1 and synaptophysin signals in each transfected axon, and the peaks of the cross-covariograms, which correspond to the degree of correlation between the two signals, were quantified (Fig. 2D). The degree of correlation varied (range, 0.1079–0.7651; mean, 0.4786; n = 11 axons), with a significant correlation occurring in 54.6% of axons (peak of covariogram>0.5, centered at 0 displacement). This indicates that in a population of axons, NR1 and synaptophysin signals are not only colocalized but also that their intensities co-vary.

To verify that sites of NR1 and synaptophysin colocalization correspond to presynaptic terminals, we imaged axo-dendritic contacts formed between (1) axons that expressed NR1–GFP and synaptophysin–mRFP, and (2) MAP2-labeled dendrites and somas of neurons that were not transfected (Fig. 2E). This ensured that the observed NR1 was restricted to the axon. At least one NR1 punctum was found in 84.9±6.5% of axo-dendritic contacts (n = 63 contacts from 20 images). NR1-positive axo-dendritic contact sites also contained synaptophysin. These data strongly support the hypothesis that NR1 is found at presynaptic terminals of cortical neurons.

Next, we verified that NMDARs localize to presynaptic terminals and determined where they reside within presynaptic terminals by using ultrastructural fractionation of synaptosomes (Fig. 3A), as previously described (Feligioni et al., 2006; Phillips et al., 2001; Pinheiro et al., 2005; Pinheiro et al., 2003). First, we verified the purity of the fractionation by immunoblotting for synaptophysin and the postsynaptic proteins PSD95 (also known as DLG4) and SHANKs (Fig. 3B). The majority of synaptophysin appeared in the non-synaptic synaptosomal protein (NSSP) fraction, which contains all synaptic cytosolic and membrane proteins that are not directly incorporated in or strongly associated with the active zone or postsynaptic density, including most synaptic vesicles. The small amount of synaptophysin in the active zone fraction is thought to correspond to docked synaptic vesicles (Feligioni et al., 2006). By contrast, PSD95 and SHANK proteins were exclusively found in the postsynaptic density (PSD) fraction. PSD95 and SHANK proteins were absent from the active zone (AZ) and NSSP fractions, regardless of the exposure times.

As expected, NR1 was abundant in the PSD fraction of both developing (P12) and adult synapses. However, in contrast to PSD95, NR1 was also observed in the AZ fraction and, to a lesser extent, in the NSSP fraction in developing synapses (Fig. 3B). Although a small amount of NR1 appeared in the AZ fraction in adults, the levels in adult appeared to be substantially reduced when compared to developing synapses. Similar to NR1, NR2B (also known as GRIN2B) was also observed in the AZ fraction of developing synapses, but NR2B was absent from the AZ fraction of mature synapses, even at long exposures. During development, synaptic NR2 subunits undergo a switch from predominantly NR2B-containing to NR2A-containing receptors. Here, NR2A levels were higher in PSDs of adults when compared to developing synapses, and NR2A was absent from the active zone fraction at both ages, even at the longest exposures possible. Taken together, these data indicate that NR2B-containing NMDARs are an integral component of the active zones of developing synapses and that association of NMDARs with the active zone is developmentally regulated.

Previous work has shown that α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid (AMPA) receptors (AMPARs) are found in small synaptic vesicles (Schenk et al., 2003); therefore, we next asked whether NMDARs associate with synaptic vesicles. To test this hypothesis, synaptic vesicles were immunoisolated from lysed synaptosomes using antibodies against glutamate and γ-aminobutyric acid (GABA) vesicular transporters. Remarkably, NR1 was associated with VGLUT1-, VGLUT2- and VGAT-positive synaptic vesicles from P6 brains (Fig. 4A,C,D), indicating that NMDARs are present in both excitatory and inhibitory presynaptic terminals during postnatal development (P6). This localization was developmentally regulated because NR1 was not observed in synaptic vesicles from adult brain (Fig. 4B–D), as previously reported (Schenk et al., 2003). The appearance of NR1 in both excitatory and inhibitory synaptic vesicles was specific because VGAT did not co-immunoisolate with VGLUTs and vice versa (Fig. 4E,F). Importantly, PSD95 was not observed in the synaptic vesicle immunoisolates or even in the starting material for the immunoisolation, the LS1 fraction (Fig. 4G). In addition, traces of NR1 were also evident in synaptic vesicles purified by ultra-pure controlled-pore glass chromatography (Fig. 4G). This synaptic vesicle location suggests that presynaptic NMDARs can be taken up from the presynaptic membrane by endocytosis and/or delivered to the presynaptic membrane upon exocytosis.

We next asked how and when NMDARs are delivered to presynaptic terminals. To address this, NR1–GFP-expressing axons were subjected to live, time-lapse imaging (every 30 s for 25 frames). From this analysis, it was clear that a population of NR1 puncta was mobile (Fig. 5A–C; supplementary material Movie 1). NR1 puncta moved in 54.2% of axons (13 of 24). In these axons, 24.5±6.6% of puncta were mobile (mean±s.e.m.; n = 139 puncta in 13 axons), moving anterogradely, retrogradely or both. Interestingly, in one axon, an NR1 punctum spontaneously formed from a more diffuse pool of NR1, suggesting that at least some NR1 puncta form from NR1 that is already in the axon (Fig. 5D).

The individual movements of 21 puncta were then tracked and quantified. The majority (67.7%) of puncta moved uni-directionally and at a fairly consistent velocity (Fig. 5A,B). A smaller subset of puncta (14.3%) displayed highly saltatory movement, sometimes pausing for several frames and/or switching direction (Fig. 5B). On average, mobile NR1 puncta spent half of their time moving (53.0±9.7%), with two distinct populations of puncta that moved either >80% or <35% of the time (Fig. 5C). NR1 puncta moved continuously for an average of 112.7±27.1 s (Fig. 5C; n = 32 continuous movements). The instantaneous velocities of mobile NR1 puncta were variable, with a maximum velocity of 0.59 µm/s and a mean instantaneous velocity of 0.23±0.01 µm/s (Fig. 5C).

During synapse formation, many synaptic vesicle and several active zone proteins are prepackaged together into Golgi-derived transport organelles, known as synaptic vesicle protein transport vesicles (STVs), for delivery to developing presynaptic terminals. This raises the question of whether NMDARs are delivered to presynaptic terminals in STVs together with synaptic vesicle proteins. To test this, we asked whether NR1–GFP and synaptophysin–mRFP move together along axons (Fig. 5E). NR1 puncta and synaptophysin puncta could be observed moving together; however, the majority of mobile NR1 puncta movements were independent of synaptophysin puncta, with only 25.9±8.9% of NR1 puncta moving with synaptophysin at some point during imaging (n = 53 NR1 puncta in seven axons; Fig. 5F). Similarly, the majority of mobile synaptophysin puncta moved independently of NR1, with only 30.7±8.0% of synaptophysin puncta moving with NR1 at some point (n = 45 synaptophysin puncta; Fig. 5F). Of the puncta that exhibited co-movement of NR1 and synaptophysin, 41.7% showed only partial co-movement, and most consisted of only short movements of no more than 10 µm (Fig. 5F). In addition, the characteristics of NR1 movement were generally distinct from the typical characteristics of STVs: NR1 puncta tended to move for extended periods at a relatively consistent velocity and direction (Fig. 5A,B). In contrast, synaptophysin puncta typically have highly discontinuous movements, frequently pausing and reversing directions (Bury and Sabo, 2011; Sabo et al., 2006). Taken together, these data suggest that NR1 and most synaptic vesicle proteins are delivered independently to developing presynaptic terminals. It is worth noting that these data also argue against a significant contribution of transport intermediates to the colocalization observed above in Fig. 2, further supporting the presynaptic identity of colocalized NR1 and synaptophysin.

Within dendrites, NMDA receptor localization and trafficking can be acutely regulated by activity (Barria and Malinow, 2002; Grosshans et al., 2002; Heynen et al., 2000); therefore, we next asked whether axonal NR1 trafficking is regulated by activity (Fig. 6). Upon acute (10–15 min) treatment of neurons with NMDA (15–100 µM) to activate NMDARs, the density of NR1 puncta was unchanged (Fig. 6A; mean±s.e.m. percentage change in density 13.1±6.7, P = 0.0667, repeated measures ANOVA). Similarly, exposure to TTX (1 µM) to inhibit action potential driven activity had no effect on NR1 density (mean percentage change in density, 9.0±9.2; P = 0.4285). In addition, the percentage of mobile NR1 puncta was unchanged by either NMDA or TTX (Fig. 6B; NMDA mean percentage change 44.2±25.1, P = 0.2706, n = 18 axons; TTX, mean percentage change, 46.9±31.8, P = 0.4425, n = 12 axons; repeated measures ANOVA). The appearance and movements of NR1–GFP puncta were also unaffected by uncaging of glutamate near transfected axons (Fig. 6C; P = 0.218 by repeated measures ANOVA, n = 5 axons). Glutamate uncaging was effective at activating NMDARs because focal uncaging near dendrites induced formation of new dendritic spines (data not shown).

NMDARs regulate presynaptic development (Sceniak et al., 2012), but it is not known whether presynaptic NMDARs are in the right place at the right time to mediate NMDAR-dependent presynaptic development. If presynaptic NMDARs are involved in synaptogenesis, they would be expected to (1) arrive at nascent synapses fairly early and (2) assemble at presynaptic terminals in response to synaptogenic signals. One well-characterized synaptogenic signal is trans-synaptic adhesion mediated through neuroligin. Neuroligin adhesion induces formation of fully functional presynaptic terminals (Scheiffele et al., 2000), and synaptic vesicle and active zone proteins are stably recruited to these new boutons within the first day after contact (Bury and Sabo, 2014; Lee et al., 2010). In addition, NMDAR activation regulates development of presynaptic terminals that are induced by neuroligin adhesion (Sceniak et al., 2012).

Therefore, we determined whether neuroligin adhesion recruits NMDARs to sites of new presynaptic terminal assembly. To do so, HEK-293 cells were transfected with neuroligin-1 then placed in contact with NR1–GFP-transfected axons to induce de novo presynaptic terminal formation in the absence of dendrites. Within 24 h after contact was induced, NR1 appeared in discrete puncta at sites of adhesion with neuroligin-expressing HEK-293 cells (Fig. 7A,B). The average percentage of axon length covered by NR1 puncta was 17.3±3.4% (mean±s.e.m.) (Fig. 7C). This was significantly higher than the average percentage of axon length covered by NR1 puncta at non-contact regions of the axon (8.4±1.7%; P = 0.0273, n = 12 contacts). In addition, the average density of NR1 puncta at neuroligin contacts was approximately double that at non-contact regions of the axon (Fig. 7D; P = 0.0346). When imaged live, NR1 was recruited to neuroligin contacts, even when other NR1 puncta were mobile within the same axon (Fig. 7E). These data indicate that synaptogenic adhesion is sufficient to induce stable recruitment of NR1 to nascent presynaptic terminals as they form. Interestingly, some NR1 passed by contact sites without being recruited, suggesting that there is a population of mobile NR1 that is unresponsive to the neuroligin–neurexin synaptogenic cue.

NMDARs are present in axonal growth cones of immature hippocampal neurons (Ehlers et al., 1998; Herkert et al., 1998; Wang et al., 2011), but their dynamics within growth cones have not yet been reported. Consistent with these reports, we observed bright NR1–GFP puncta in axonal growth cones of developing cortical neurons. NR1 puncta were present in extending (n = 7), retracting (n = 4) and stationary (n = 6) growth cones (Fig. 8; supplementary material Movie 2). NR1 fluorescence also appeared bright in nascent growth cones formed from existing axons, even as they emerged from the axon (n = 7). Interestingly, in 71.4% of examples, NR1 clearly appeared at the site of growth cone initiation prior to emergence of a distinct growth cone.

Synaptic vesicle proteins can enter and leave growth cone filopodia and lamellopodia (Sabo and McAllister, 2003), raising the question of whether NMDARs are also dynamic within these regions of axonal growth cones. We found that NR1 fluorescence was often very bright in the central domain of the growth cone and in the region of the axon immediately proximal to the growth cone (Fig. 8). NR1 puncta was highly dynamic in both the growth cone and adjacent axon. NR1 puncta also appeared and disappeared from growth cone filopodia from one frame to the next. Of the imaged growth cones, 80.0% (n = 15) had NR1 puncta that were transiently present in lamellipodia or filopodia. Interestingly, NR1 puncta also rapidly entered and exited axonal shaft filopodia (supplementary material Movie 2). Therefore, presynaptic NMDARs are in a position to contribute to synaptogenesis even during or before initial axo-dendritic contact.

NMDARs play key roles in cortical development, and it is likely that presynaptic NMDARs are responsible for some of these effects. For example, it has been suggested that presynaptic NMDARs could contribute to establishment of receptive field properties in the visual system, such as retinotopy, ocular dominance and orientation selectivity (Corlew et al., 2007). It has also been suggested that presynaptic NMDARs mediate some acute effects of brain-derived neurotrophic factor (BDNF), a strong regulator of cortical development (Madara and Levine, 2008). To understand when, where and how axonal and presynaptic NMDARs contribute to cortical development, it is necessary to define the prevalence, localization and dynamics of axonal and presynaptic NMDARs. Here, we used fluorescence live confocal imaging combined with immunoisolation from synaptosomes and ultrasynaptic fractionation to show for the first time that NMDARs are present in the active zones and synaptic vesicles of presynaptic terminals in developing cortical neurons in vitro and in vivo. NMDARs are found at presynaptic terminals from their earliest stages, and their expression within terminals is developmentally regulated.

Although NMDARs are present as puncta in both axons and dendrites, the morphologies of axonal and dendritic NMDAR puncta are distinct. First, NR1 puncta occur at a lower density in axons than in dendrites [axons, ∼0.04 puncta/µm, or 0.1 puncta/µm if hot-spots were also included; dendrites: ∼0.3–0.4 puncta/µm at 5–8 DIV (Washbourne et al., 2002)]. Second, we found that axonal NR1 puncta generally appear to be smaller and dimmer than dendritic NR1 puncta in the same neuron. Because axons express a lower density and intensity of NR1 than dendrites, identification of NR1 in the axons would have been difficult without an independent label to initially identify axons (e.g. synaptophysin–mRFP). This might explain why axonal NR1 was missed in previous studies (Bresler et al., 2004; Washbourne et al., 2002).

Trafficking of NMDAR puncta in axons and dendrites was also distinct. In axons, approximately one quarter of the NR1–GFP puncta were mobile, moving at a mean velocity of 0.23 µm/s. In dendrites, a previous study found that a third of dendritic NR1–GFP clusters are mobile, moving at a mean velocity of 0.07 µm/s (Washbourne et al., 2002), although another study failed to detect mobile dendritic NR1–GFP puncta (Bresler et al., 2004). Most NR2A and NR2B clusters are mobile within developing dendrites (Yin et al., 2012; Yin et al., 2011). However, dendritic NR2A and NR2B move anterogradely at ∼0.7–0.8 µm/s (Yin et al., 2012; Yin et al., 2011). Similar velocities (∼0.7 µm/s) have been reported for KIF17 (Guillaud et al., 2003; Wong-Riley and Besharse, 2012), a motor that transports dendritic NMDARs (Guillaud et al., 2003; Guillaud et al., 2008; Setou et al., 2000; Song et al., 2009; Yin et al., 2012; Yin et al., 2011).

It is unclear which anterograde motors transport axonal NMDARs. NR2 subunits target NR1 to postsynaptic membranes (Barria and Malinow, 2002). In dendrites, NR2B-containing NMDARs are transported by KIF17 (Guillaud et al., 2003; Guillaud et al., 2008; Setou et al., 2000; Yin et al., 2012; Yin et al., 2011). KIF17 does not enter axons (Guillaud et al., 2003; Song et al., 2009), suggesting that KIF17 is not responsible for NMDAR transport in axons. Because it is likely that NMDARs are transported by distinct motors in axons and dendrites, identification of the responsible motors in future experiments could allow us to selectively reduce expression of either presynaptic or postsynaptic NMDARs in order to understand their distinct functions.

We also observed dynamic NR1 in nascent axonal growth cones. Surprisingly, NR1 often appeared at sites of growth cone induction even before growth cones appeared. NR1 might be a component of vesicles that supply membrane for growth. Alternatively, NMDARs might regulate growth cone formation or behavior. Consistent with the latter hypothesis, growth cone behavior is strongly regulated by neurotransmitters and intracellular Ca²⁺ (Gomez and Spitzer, 2000; Henley et al., 2004; Ruediger and Bolz, 2007; van Kesteren and Spencer, 2003). In addition to functional NMDARs (Wang et al., 2011), growth cones also express functional AMPARs that are internalized or exposed to the surface upon activity (Schenk et al., 2003). In excitatory neurons, vesicles within axonal growth cones and their filopodia are thought to be capable of glutamate release (Sabo and McAllister, 2003), raising the intriguing possibility that both NMDA and AMPA receptors in growth cones might act as autoreceptors.

Elegant work has provided evidence that NMDARs regulate presynaptic plasticity in the visual and somatosensory neocortex, entorhinal cortex, hippocampus, amygdala, cerebellum and spinal cord (Bardoni et al., 2004; Bender et al., 2006; Berretta and Jones, 1996; Brasier and Feldman, 2008; Buchanan et al., 2012; Casado et al., 2000; Casado et al., 2002; Corlew et al., 2007; Duguid and Smart, 2004; Fiszman et al., 2005; Glitsch and Marty, 1999; Humeau et al., 2003; Jourdain et al., 2007; Li and Han, 2007; Li et al., 2008; Liu et al., 1997; Mameli et al., 2005; Rodríguez-Moreno et al., 2011; Rodríguez-Moreno and Paulsen, 2008; Sjöström et al., 2003; Yang et al., 2006). At least in certain developing neurons, the responsible receptors are found in the presynaptic rather than in the postsynaptic neuron. Recently, several studies have provided convincing functional evidence that the relevant receptors are localized to axons (Buchanan et al., 2012; McGuinness et al., 2010; Rodríguez-Moreno et al., 2011). However, it remained unclear whether these receptors were distributed throughout the axon, concentrated at presynaptic terminals or located in microdomains near synapses. Although electron microscopy studies support the existence of presynaptic NMDARs in the cortex and hippocampus, these studies have not been definitive because there is inconsistency in whether they are found at excitatory presynaptic terminals, in the timing and abundance of their expression, and in their precise localization within terminals (Aoki et al., 2003; Charton et al., 1999; Corlew et al., 2007; DeBiasi et al., 1996; Fujisawa and Aoki, 2003; Jourdain et al., 2007; McGuinness et al., 2010; Siegel et al., 1994).

Here, we have shown that NR1, the obligate subunit of NMDARs, is clustered at presynaptic terminals in vitro and in vivo. This clustering occurs early in synapse development. Surprisingly, presynaptic NMDARs could be found in synaptic vesicles of both excitatory and inhibitory synapses. The presence of NR1 in association with synaptic vesicles suggests that, in analogy to presynaptic AMPARs, NMDARs can be dynamically internalized by endocytosis and exposed to the presynaptic membrane by exocytosis in response to electrical activity and activation of signal transduction cascades. In addition, the localization of presynaptic NMDARs is developmentally regulated given that NMDARs are not found in synaptic vesicles in mature neurons in vivo, consistent with previous studies of presynaptic NMDAR function in cortical slices (Corlew et al., 2007; Wang et al., 2011).

It has recently been shown that NMDARs regulate presynaptic development (Sceniak et al., 2012). Future experiments will be necessary to determine whether presynaptic NMDARs are responsible for this NMDAR-dependent regulation; however, the early appearance of NMDARs at presynaptic terminals is consistent with a role for presynaptic NMDARs. This hypothesis is compelling given the prevalence of postsynaptically silent synapses during circuit development (Groc et al., 2006; Isaac, 2003). How might presynaptic NMDARs regulate presynaptic terminal development? One possibility is that NMDARs regulate presynaptic development by locally increasing Ca²⁺ influx. Ca²⁺ influx regulates the transport of synaptic vesicle proteins along axons during synaptogenesis and increases the anchoring of synaptic vesicle precursors at sites of synapse formation (Sabo et al., 2006), but the source of this Ca²⁺ signal has not been identified.

Here, we demonstrated that NR1 is recruited to sites of presynaptic terminal formation by trans-synaptic adhesion through postsynaptic neuroligin. Given that neuroligin exerts its synapse-organizing effects through its interaction with neurexin on the presynaptic side (Dean et al., 2003; Ichtchenko et al., 1995; Levinson et al., 2005; Nguyen and Südhof, 1997; Scheiffele et al., 2000; Varoqueaux et al., 2006), a link must exist between neurexin adhesion (or clustering) and recruitment of NMDARs to presynaptic terminals. In dendrites, transport of NR2B-containing NMDARs occurs through an interaction with a scaffolding complex that contains MALS/Velis/mLin7, CASK/mLin2 and Mint1/X11/mLin10 (Setou et al., 2000). Neurexin can also interact with CASK (Butz et al., 1998; Hata et al., 1996; Mukherjee et al., 2008; Zhang et al., 2001), providing a possible mechanism to link NMDARs to neurexin. This could represent a general mechanism for accumulation of NMDARs at presynaptic terminals because several additional trans-synaptic adhesion proteins, including SynCAMs and APP, can also interact with this complex (Bury and Sabo, 2010).

Our results demonstrate that presynaptic NMDARs are in the right place at the right time to regulate presynaptic development. Previous studies have shown that axonal NMDARs acutely regulate presynaptic transmitter release during postnatal development (Corlew et al., 2008; Pinheiro and Mulle, 2008). Taken together, these data support a new model of postnatal synapse development in which NMDARs occupy a privileged position within presynaptic active zones, where they ultimately act as auto-receptors to promote the development and plasticity of individual presynaptic boutons. They also suggest that prolonged disruption of trafficking, localization or function of presynaptic NMDARs is likely to result in abnormal synapse development. Consistent with this, presynaptic NMDARs have been suggested to play roles in epilepsy (Yang et al., 2007; Yang et al., 2006) and fetal alcohol spectrum disorders (Valenzuela et al., 2008). Mutations in NMDAR subunits are associated with autism, intellectual disability and epilepsy (Endele et al., 2010; Hamdan et al., 2011; O'Roak et al., 2011; O'Roak et al., 2012a; O'Roak et al., 2012b; Tarabeux et al., 2010; Yoo et al., 2012), and it will be interesting to determine whether presynaptic NMDARs are affected by these mutations and how presynaptic NMDARs might contribute to pathogenesis of these diseases.

Neurons were dissociated from visual cortices of 0–3-day-old Sprague-Dawley rats, cultured on glial monolayers and transfected, essentially as previously described (Berry et al., 2012; Bury and Sabo, 2011; Bury and Sabo, 2014; Sceniak et al., 2012). For most experiments, neurons were plated on a monolayer of cortical astrocytes in Neurobasal-A medium with glutamax and B27 supplement (Invitrogen). For NR1 immunofluorescence, neurons were plated directly on poly-L-lysine-coated coverslips, in the absence of glia (because glia can express NR1). Astrocytes were grown on collagen (BD Biosciences) and poly-D-lysine (Sigma)-coated coverslips in minimum essential medium (Invitrogen) with glutamax, 10% fetal calf serum (Sigma), glucose (Sigma), N2 supplement (Invitrogen), and penicillin-streptomycin. At 2–4 DIV, neuron cultures were treated with anti-mitotic (5′fluoro-2′dioxy-uridine/uridine; Sigma). Neurons were transfected with NR1–GFP and synaptophysin–mRFP, synaptophysin–mCherry or pTagRFP at 5–10 DIV using Lipofectamine 2000 (Invitrogen). In some experiments, neurons were transfected with synaptophysin-GFP and immunolabled for NR1. All animal experiments were conducted with an approved protocol from the Case Western Reserve University Institutional Animal Care and Use Committee, in compliance with the National Institutes of Health guidelines for the care and use of experimental animals.

HEK-293 cells were maintained in Dulbecco's modified Eagle's medium (DMEM) containing 10% fetal calf serum and penicillin-streptomycin. For presynaptic terminal induction, HEK-293 cells were transfected with HA–neuroligin-1 and hcRed or dsRed. At 20–30 h after transfection, cells were dissociated, washed and added to neurons (6–12 DIV). After 24–48 h, cells were fixed and immunolabeled or imaged live.

With the exception of NR1 labeling, cells were fixed with 4% paraformaldehyde in 0.1 M phosphate-buffered saline (PBS) containing 4% sucrose and permeabilized with 0.2% Triton X-100 in PBS. For NR1, neurons were fixed with −20°C methanol for 10 min then permeabilized with 0.025% Triton X-100 for 2 min. Primary antibodies were chicken anti-MAP2 (Thermo Scientific), anti-tau-1 (Synaptic Systems), rabbit anti-GFP (Invitrogen) and mouse anti-NR1 (BD Biosciences) antibodies. Secondary antibodies were an anti-chicken-IgG antibody conjugated to Alexa Fluor 647 and anti-mouse- or rabbit-IgG antibody conjugated to Alexa Fluor 555 or 488.

Imaging was performed on a Nikon C1 Plus confocal system on a Nikon Ti-E inverted microscope with 488 nm Ar, 543 nm HeNe and 633 nm HeNe lasers and 40× 1.0 NA and 0.95 NA objectives. For immunocytochemistry, 3D image stacks were collected using EasyC1 software (Nikon) with 2×Kalman averaging, a pixel size of 90–115 nm and 16-bit pixel depth. Each channel was imaged separately to avoid bleed-through. Imaging parameters were optimized to maintain fluorescence within the linear range and maximize intensity resolution. For live imaging, neurons were imaged with continual perfusion with artificial cerebrospinal fluid (ACSF, in mM: 120 NaCl, 3 KCl, 2 CaCl₂, 2 MgCl₂, 30 D-glucose, 20 HEPES, and 0.2% sorbitol, pH 7.3) or in culture medium. Imaging in culture medium was performed in a custom-built chamber maintained at 35–37°C. In most live imaging experiments, neurons were fixed after imaging for retrospective immunocytochemistry with MAP2 to confirm axonal identity. For treatment with TTX (1 µM) or NMDA (15–100 µM), imaging was performed in the same axons before and during treatment (beginning 0–10 min after start of treatment). NMDA treatment and glutamate uncaging were performed in the presence of glycine (20 µM) and nominal to low Mg²⁺ (0–0.5 mM). For uncaging, axons were imaged to establish baseline NR1–GFP movement then stimulated by uncaging RuBi-glutamate (30 µM, Tocris) with a 488-nm laser (Fino et al., 2009; Salierno et al., 2010). After stimulation, RuBi-glutamate was removed by washing with ACSF for 2 min, followed by additional imaging. Images were obtained every 1–15 s for 5 min. The effectiveness of glutamate uncaging was confirmed with the Ca²⁺ indicator fluo-4 (Invitrogen).

Synaptosomes were purified from P2 fractions from the pooled cortex of three adult mice (3–6 months old) and of six P12 mice. The tissue was homogenized in 10 ml of ice-cold HB buffer (0.32 M sucrose, 1 mM EDTA, 10 mM Tris-HCl, pH 7.4) containing protease inhibitors, using a glass-Teflon homogenizer. The resultant homogenate was centrifuged at 1000 g for 5 min at 4°C in order to remove nuclei and debris, and the supernatant was further centrifuged at 18,900 g for 10 min at 4°C to obtain the P2 fraction. The P2 pellet was resuspended in 2 ml of HB buffer and gently stratified on a discontinuous Percoll gradient (3–10%–23%). After a 10 min centrifugation at 18,900 g at 4°C, the synaptosomal fraction from the 10% and 23% Percoll interface was collected, washed in Krebs buffer (140 mM NaCl, 5 mM KCl, 5 mM NaHCO3, 1.3 mM MgSO4, 1 mM phosphate buffer pH 7.4, 10 mM Tris/Hepes pH 7.4) to eliminate Percoll, and used as the starting material for subsequent ultrafractionation.

Ultrasynaptic fractionation was performed according to Feligioni et al. (Feligioni et al., 2006). Synaptosomes were pelleted by centrifugation (16,000 g, 5 min, 4°C) and resuspended in 300 µl of 0.32 M sucrose, 0.1 mM CaCl₂. An aliquot was removed and kept as ‘total’. Synaptosomes were diluted 1∶10 in ice-cold 0.1 mM CaCl₂ and mixed with an equal volume of 2× solubilization buffer (2% Triton X-100, 40 mM Tris-HCl, pH 6, 4°C). After a 30-min incubation at 4°C, the insoluble material (synaptic junction) was pelleted by centrifugation (40,000 g, 30 min, 4°C). The supernatant (non-synaptic synaptosomal protein; NSSP) was decanted and the proteins precipitated with six volumes of acetone at −20°C overnight and then centrifuged (18,000 g, 30 min, −15°C). The synaptic junction pellet (containing the insoluble postsynaptic density and the presynaptic active zone) was resuspended in ten volumes of 1× solubilization buffer (1% Triton X-100, 20 mM Tris-HCl, pH 8, 4°C), incubated for 30 min at 4°C then centrifuged (40,000 g, 30 min, 4°C). The pellet represents the postsynaptic density (PSD). The proteins in the supernatant (presynaptic fraction, AZ) were precipitated in 6 volumes of acetone at −20°C overnight then centrifuged (18,000×g, 30 min, −15°C). All pellets (NSSP, AZ and PSD) were resuspended in 5% SDS, quantified by a BCA assay and loaded on SDS-PAGE gels for electrophoresis and western blotting. Polyclonal anti-NR2B and monoclonal anti-synaptophysin, anti-PSD95, anti-NR1 antibodies were from Synaptic Systems. The monoclonal anti-panSHANK antibody was from the UC Davis/NIH NeuroMab Facility; the anti-NR2A antibody was from BD Biosciences. Exposures were optimized in order to avoid saturation of the signal in any of the lanes (P12 and adult were performed in the same western blot and quantified at the same exposures for a given experiment to allow comparison across age for a given antibody). Protease inhibitors were used in all purification steps.

Eupergit C1Z methacrylate microbeads (1 µm diameter; Röhm Pharmaceuticals, Darmstadt, Germany) were blocked with glycine or conjugated with affinity-purified goat anti-rabbit-IgG antibodies (IgG; Sigma) as previously described (Burger et al., 1991; Fattorini et al., 2009). Affinity-purified anti-VGLUT1, -VGLUT2 and -VGAT antibodies (Synaptic Systems) were conjugated with the covalently bound secondary antibodies, generating IgG-coated anti-VGLUT1, anti-VGLUT2 or anti-VGAT beads. Beads coated with glycine or secondary antibodies only were used as negative controls (mock beads). Aliquots of the LS1 fraction (300 µg protein; Huttner et al., 1983) obtained from osmotic lysis of cortical synaptosomes prepared from either adult (3–6 months old) or P6 mice were incubated for 2–4 h at 4°C in phosphate-buffered saline (PBS) with bead preparations (50–70 µl settled beads in 500 µl final volume) under constant rotation. After centrifugation at 1000 g for 1 min and repeated washes in PBS, corresponding amounts of bead pellets and supernatant fractions were subjected to SDS-PAGE on 10% polyacrylamide gels. Gels were electrophoretically transferred to nitrocellulose membranes and immunoblotted with polyclonal anti-synaptophysin or monoclonal anti-NR1 antibodies. Quantification of the recovered immunoreactivity was performed by densitometric analysis of the fluorograms and by data interpolation into a standard curve of mouse brain LS1 fraction run in parallel. Amounts of synaptophysin and NR1 associated with immunoisolated vesicles were expressed as a percentage of the total input of LS1 fraction added to the samples. Experiments were repeated at least three times. Membranes were also probed with affinity-purified anti-VGLUT1, -VGLUT2 and -VGAT antibodies to confirm the immunoisolation of glutamatergic and GABAergic synaptic vesicles.

Subcellular fractions were prepared from 5–6-week-old male Sprague-Dawley rat forebrain, and synaptic vesicles were purified through controlled-pore glass chromatography (Huttner et al., 1983). Purified synaptic vesicles (USV) were recovered by high-speed centrifugation and resuspended in 0.3 M glycine, 5 mM Hepes (pH 7.4) at a protein concentration of 1 mg/ml.

Custom macros in ImageJ (NIH, Bethesda, MD) were used to conduct automated image analysis. Maximum intensity z-projections were made then filtered using a Gaussian blur. Intensities were measured along axon lengths to generate line profiles. Measurements were transferred to Matlab (Mathworks, Natick, MA) for quantification using custom-written functions. A threshold of 1–2 standard deviations above the mean intensity was used to define an NR1 punctum or hot-spot. For experiments involving axon–HEK-293 cell contact, puncta distributions were compared within and outside contact regions. A MAP2 signal that was 2.5 standard deviations above the mean intensity was used to blindly identify presynaptic contact with dendrites. For colocalization and cross-correlation, each channel was analyzed independently while blind to all other channels.

For live imaging, NR1 movements were tracked in ImageJ then further analyzed in Matlab. Axons were included if at least one punctum moved within the field of view. Movements were defined as instantaneous velocities >0.1 µm/s. Run times correspond to the length of time that instantaneous velocities were continuously >0.1 µm/s. Because some runs began before or continued after the imaging period, long runs may be truncated and under-represented.