NMDA receptors inhibit synapse unsilencing during brain development

Abstract

How the billions of synapses in the adult mammalian brain are precisely specified remains one of the fundamental questions of neuroscience. Although a genetic program is likely to encode the basic neural blueprint, much evidence suggests that experience-driven activity through NMDA receptors wires up neuronal circuits by inducing a process similar to long-term potentiation. To test this notion directly, we eliminated NMDA receptors before and during synaptogenesis in single cells in vitro and in vivo. Although the prevailing model would predict that NMDA receptor deletion should strongly inhibit the maturation of excitatory circuits, we find that genetic ablation of NMDA receptor function profoundly increases the number of functional synapses between neurons. Conversely, reintroduction of NMDA receptors into NR1-deficient neurons reduces the number of functional inputs, a process requiring network activity and NMDA receptor function. Although NMDA receptor deletion increases the strength of unitary connections, it does not alter neuronal morphology, suggesting that basal NMDA receptor activation blocks the recruitment of AMPA receptors to silent synapses. Based on these results we suggest a new model for the maturation of excitatory synapses in which ongoing activation of NMDA receptors prevents premature synaptic maturation by ensuring that only punctuated bursts of activity lead to the induction of a functional synapse for the activity-dependent wiring of neural circuitry.

Considerable evidence links NMDA receptor activation to the maturation of excitatory circuitry during brain development (1–3). Most studies, however, have relied on widespread pharmacological inhibition or broad genetic deletion of NMDA receptors to explore their involvement in synaptic development (4–8). To avoid indirect effects that such manipulations might have by generally altering network activity (9–11), we abolished NMDA receptor protein expression in sparsely distributed cells in the hippocampus by introduction of CRE recombinase to neurons in a floxed NR1 mouse (NR1^fl/fl) (12). Mosaic deletion permitted simultaneous paired whole-cell recordings from CRE-expressing and untransfected neighboring cells to provide a rigorous, quantitative, and internally controlled comparison of the physiological effects of NR1 deletion.

Results

First, we biolistically transfected a plasmid encoding CRE recombinase into cells in organotypic hippocampal slice cultures from NR1^fl/fl mice. The low efficacy of this transfection technique leads to NR1 deletion in only a few cells in the slice. A GFP reporter in the construct identified CRE-expressing cells. Two weeks after transfection, synaptic NMDA currents were strongly reduced in GFP⁺ neurons but not in their untransfected neighbors (Fig. 1a, I_transfected/I_control = 0.23 ± 0.03, n = 41 pairs, P < 10⁻¹¹). As a control, transfection of CRE on its own into wild-type slices did not significantly affect NMDA or AMPA receptor-mediated synaptic currents [NMDA: I_transfected/I_control = 1.5 ± 0.2, n = 13 pairs, P = 0.27; AMPA: I_transfected/I_control = 1.4 ± 0.3, n = 13 pairs, P = 0.21; see supporting information (SI) Fig. S1 a–c]. This confirms the efficacy and specificity of our technique.

Fig. 1.

NMDA receptor deletion results in a strong increase in the number of AMPA receptor-containing synapses. (a) Scatter plot of NMDA EPSCs of pairs of simultaneously recorded neurons in NR1^fl/fl slice culture from CRE-IRES-GFP transfected and neighboring control cells 12–17 days after transfection (n = 41 pairs, P < 10⁻¹⁰). (Inset) Representative traces: black, control cell; green, CRE-IRES-GFP-expressing cell. (b) Scatter plot of AMPA EPSCs from the same cells (n = 41 pairs, P < 10⁻⁵). (c) Bar graph showing the relative change in AMPA and NMDA EPSCs between control and CRE-transfected neurons from the pairs in a. Error bars indicate SEM. (d) Cumulative distribution of mEPSC interintervals from control (n = 16) and CRE-expressing (n = 15) neurons (P < 0.005). Above is an example recording of mEPSCs from a CRE-expressing cell. (e) Cumulative distribution of mEPSC amplitudes from control (n = 16) and CRE-expressing (n = 15) neurons (P = 0.65). Above is an example recording of mEPSCs from a control cell. (f) Bar graph showing average paired-pulse ratio in control and CRE-expressing cells. Above are representative traces.

If NMDA receptor function were necessary for synaptic maturation, or for simply maintaining existing functioning synapses, then NR1 knockdown would be expected to reduce synaptic AMPA currents in CRE-transfected cells. Instead, NR1-deleted neurons actually exhibited a pronounced, ≈3-fold enhancement of synaptic transmission mediated by AMPA receptors (Fig. 1 b and c; I_transfected/I_control = 3.19 ± 0.35, n = 41 pairs, P < 10⁻⁵). Because synaptic currents in transfected neurons showed no difference in the paired-pulse-ratio (PPR), a measure of presynaptic release (Fig. 1f; PPR transfected, 1.77 ± 0.10; PPR control, 1.83 ± 0.08; P = 0.62), these effects are likely to be postsynaptic in origin. This increase in synaptic transmission onto NR1 knockout cells could be due to enhanced quantal amplitude or to a greater number of functional synaptic inputs. To test this we compared the amplitude and frequency of AMPA receptor-mediated action potential-independent miniature excitatory postsynaptic currents (mEPSCs) between transfected and untransfected cells. NR1-deficient neurons displayed no change in quantal amplitude (transfected: 10.2 ± 0.3 pA, n = 15; control: 9.6 ± 0.4 pA, n = 16; P = 0.65) but a strong increase in the frequency of spontaneous events [Fig. 1 d and e; interevent interval (iei) transfected: 2.0 ± 0.3 s, n = 15; iei control: 4.1 ± 0.7 s, n = 16; P < 0.005]. These results suggest that NR1 deletion in postnatal slice culture triggers a net increase in the number of functional synapses made onto cells lacking NMDA receptor function. These can represent either completely new synaptic contacts or, as we will argue below, the unsilencing of preexisting synapses. It is unlikely that NMDA receptors directly limit synaptic strength, because quantal amplitude was unchanged.

Based on these findings it is possible that NMDA receptors negatively regulate synaptic input to pyramidal neurons. If this is true, reintroduction of NMDA receptors to cells in a global NMDA receptor knockout might be sufficient to reduce the number of synapses made onto these cells. Toward this end, we rescued NMDA currents in CA1 pyramidal neurons by biolistic transfection of NR1-GFP into slice cultures prepared from NEX-CRE; NR1^fl/fl mice, which lack NR1 in forebrain pyramidal neurons, similar to a forebrain NR1 knockout mouse previously described (5). Six to 10 days after transfection, all GFP⁺ neurons exhibited robust synaptic NMDA EPSCs, whereas untransfected cells had no detectable synaptic NMDA current, confirming the efficacy of the rescue (Fig. 2a; control: 0.9 ± 0.3 pA; NR1-transfected: 33 ± 4 pA, n = 27 pairs; P < 10⁻⁷). Consistent with the inhibitory role of NMDA receptor on synapse maturation, NR1-expressing cells had significantly reduced synaptic AMPA currents as compared with neighboring untransfected, knockout neurons (I_transfected/I_control = 0.52 ± 0.08, n = 27 pairs, P < 0.005; Fig. 2 b and c), without a change in the paired-pulse ratio (PPF control: 2.1 ± 0.1; PPF NR1-transfected: 2.3 ± 0.2; n = 8, P = 0.31; Fig. 2g). Incubating the cultures in a mixture of NMDA receptor antagonists or eliminating action potentials with TTX blocked the reduction of AMPA currents by NR1 reexpression (NMDA mixture: I_transfected/I_control = 1.23 ± 0.13, n = 16 pairs, P = 0.48; TTX: I_transfected/I_control = 1.11 ± 0.17, n = 14 pairs, P = 0.53; Fig. 2c). Importantly, this indicates that this process requires network activity that engages NMDA receptor function. Analysis of mEPSCs in rescued and knockout neurons showed that transfected cells exhibited significantly fewer spontaneous events (iei control: 3.7 ± 0.9 s, n = 9; iei NR1-transfected: 8.9 ± 2.3 s; n = 9, P < 0.001; Fig. 2d) but that they were of comparable amplitude (control: 12.7 ± 1.6 pA, n = 9; NR1-transfected: 12.3 ± 1.5 pA, n = 9; P = 0.72; Fig. 2e), indicating that NR1-rescued neurons actually received fewer functional inputs than their NR1-deficient neighbors. These results support the notion that NMDA receptor function during network activity is both necessary and sufficient to limit the number of functional synapses made onto pyramidal neurons.

Fig. 2.

Reintroduction of NMDA receptors to NR1 knockout cells from NEX-CRE;NR1^fl/fl slice cultures reduces AMPA receptor-containing synapses in an activity-dependent manner. (a) Scatter plot of NMDA EPSCs in simultaneous paired recordings from control (NEX-CRE;NR1^fl/fl) and NR1-GFP-transfected neurons 5–10 days after transfection (n = 25 pairs, P < 10⁻⁷). (Inset) Representative traces: black, control cell; green, NR1-GFP-expressing cell. (b) Scatter plot of AMPA EPSCs in the same cells (n = 25 pairs, P < 0.005). (c) Bar graph showing the relative change in AMPA and NMDA EPSCs between control and NR1-GFP-transfected neurons from the pairs in a and in separate experiments when the slices were incubated in a mixture of NMDA receptor antagonists (NMDAR block) (n = 16 pairs, P = 0.46) or TTX (n = 14 pairs, P = 0.48). Error bars indicate SEM. (d) Cumulative distribution of mEPSC interintervals from control and NR1-GFP-expressing neurons (n = 9 each, P < 0.001). Above is an example recording from an NR1-GFP-expressing cell. (e) Cumulative distribution of mEPSC amplitudes from control and NR1-GFP-expressing neurons (n = 9 each, P = 0.72). Above is an example recording from a control cell. (f) Bar graph showing average paired-pulse ratio in control and NR1-GFP-expressing cells. Above are representative traces.

Because cells in organotypic slice culture do not receive the same patterns of activity as neurons in the intact brain, deletion or rescue of NMDA receptors in vitro might lead to nonphysiological changes in synaptic maturation; if true, then NMDA receptors could still be required for functional integration of neurons into their local circuit in vivo. To test this possibility, we injected a CRE-GFP-expressing adenoassociated virus (AAV-CRE-GFP) into NR1^fl/fl embryos at embryonic days 14.5–16.5 to infect progenitor cells of pyramidal neurons, deleting NR1 well before synapse formation (13). As early as AMPA-receptor-mediated synaptic currents could be routinely recorded [postnatal days 3–4 (P3–P4)], GFP⁺ neurons exhibited profoundly reduced synaptic NMDA receptor currents (Fig. S2a). As a preliminary control, injection of this virus to wild-type embryos did not alter synaptic currents (Fig. S1 d–f). We then targeted GFP⁺ neurons and their uninfected neighbors in injected NR1^fl/fl animals for simultaneous whole-cell recordings in acute hippocampal slices at a time point when much of the excitatory circuitry in CA1 has already formed (P12–P19). Whereas NMDA currents were nearly absent in GFP⁺ neurons (I_infected/I_control = 0.05 ± 0.02, n = 32 pairs, P < 10⁻⁹; Fig. 3a), AMPA currents were almost three times as large, similar to the increase observed by deletion of NR1 in slice culture (I_infected/I_control = 2.74 ± 0.36, n = 33 pairs, P < 10⁻⁷; Fig. 3 b and c). Also, there was no associated change in the paired-pulse ratio (infected: 1.99 ± 0.14; control: 2.16 ± 0.15; n = 15, P = 0.41) or a difference in the rectification of synaptic AMPA receptors from that observed in wild-type animals (rectification index: 0.89 ± 0.04, n = 10) (14), ruling out an insertion of calcium-permeable AMPA receptors. These results confirm that NMDA receptors are not required for synaptic maturation during the normal course of brain development and that they actually limit functional synaptic input to developing pyramidal neurons. To further investigate the origin of the increase, we recorded mEPSCs in prenatally NR1-deleted neurons and control cells. NR1 knockout cells exhibited a large increase in mEPSC frequency (iei infected: 2.0 ± 0.2 s, n = 22; iei control: 4.6 ± 0.8 s, n = 21, P < 0.001; Fig. 3d) as well as a very slight increase in mEPSC amplitude (infected: 10.6 ± 0.3 pA, n = 22; control: 9.9 ± 0.4 pA, n = 21; P = 0.013; Fig. 3e). This suggests that NMDA receptors are primarily involved in restricting the number of functional synapses pyramidal neurons make during the natural course of brain development.

Fig. 3.

NMDA receptor deletion in utero increases the number of AMPA receptor-containing synapses. (a) Scatter plot of NMDA EPSCs of pairs of simultaneously recorded neurons in acute slices from AAV-CRE-GFP-infected and neighboring control cells (n = 32 pairs, P < 10⁻⁹). (Inset) Representative traces: black, control cell; green, CRE-expressing cell. (b) Scatter plot of AMPA EPSCs from the same cells (n = 33 pairs, P < 10⁻⁷). (c) Bar graph showing the relative change in AMPA and NMDA EPSCs between control and CRE-expressing neurons from the pairs in a. Error bars indicate SEM. (d) Cumulative distribution of mEPSC interintervals from control (n = 21) and NR1-deficient (n = 22) neurons (P < 0.001). Above is an example recording of mEPSCs from CRE-expressing cell. (e) Cumulative distribution of mEPSC amplitudes from control (n = 21) and NR1-deficient (n = 22) neurons (P < 0.05). Above is an example recording of mEPSCs from a control cell. (f) Bar graph showing average paired-pulse ratio in control and CRE-expressing cells. Above are representative traces.

NMDA receptors could regulate synaptic input by restricting the total number of presynaptic partners of each postsynaptic cell or by limiting the number of functional contacts made by each presynaptic partner. To discriminate between these two possibilities we compared the amplitude of spontaneously occurring unitary EPSCs (sEPSCS) between AAV-CRE-GFP infected and uninfected control cells. The size of these currents is related to the average number of functional contacts each presynaptic neuron makes onto the postsynaptic cell (15). In infected neurons, spontaneous synaptic AMPA currents were significantly larger and more frequent than in neighboring control cells (sEPSC amplitude infected: 13.4 ± 0.6 pA, n = 12; control sEPSC amplitude: 11.8 ± 0.70 pA, n = 11, P < 0.05; sEPSC iei infected: 1.6 ± 0.2 s, n = 12; sEPSC iei control: 2.8 ± 0.4, n = 11, P < 0.001; Fig. 4 a and b), in agreement with the enhancement of evoked EPSCs onto these cells. To separate action-potential-dependent unitary sEPSCs from mEPSCs, we then applied TTX to block voltage-gated sodium channels. In control, uninfected neurons, blockade of action potentials reduced the frequency of spontaneous events (iei before TTX: 2.8 ± 0.5 s; iei after TTX: 4.7 ± 1.4 s; n = 10, P < 0.001) but did not significantly affect their amplitude (before TTX: 11.8 ± 0.7 pA; after TTX: 10.4 ± 0.7; n = 10, P = 0.53; Fig. 4c), indicating that unitary connections in wild-type pyramidal neurons of this age range generally involve the release of a single quantum of neurotransmitter (15). In contrast, TTX reduced both the frequency (iei before TTX: 1.6 ± 0.2 s; iei after TTX: 2.2 ± 0.4 s; n = 10, P < 0.001) and amplitude of sEPSCs in NR1-deficient cells (before TTX: 13.5 ± 0.6 pA; after TTX: 10.2 ± 0.3 pA; n = 10, P < 0.01; Fig. 4d). This demonstrates that the average unitary connection onto NR1 knockout cells, but not onto neighboring control neurons, involves more than one functional contact.

Fig. 4.

NR1 deletion in utero results an increase in the strength of unitary connections without any significant changes in cell morphology. (a) Cumulative distribution of sEPSC amplitudes from control (n = 11) and AAV-CRE-GFP-infected (n = 13) neurons (P < 0.05). (b) Cumulative distribution of sEPSC interevent intervals from control (n = 11) and CRE-expressing (n = 13) neurons (P < 0.001). (c) Cumulative distribution of sEPSC and mEPSC amplitudes (after application of TTX) in control neurons (n = 10, P = 0.53). (Inset) Representative traces before (black) and after (gray) application of TTX. (d) Cumulative distribution of sEPSC and mEPSC amplitudes (after application of TTX) in CRE-expressing neurons (n = 10, P < 0.01). (Inset) Representative traces before (thick line) and after (thin line) application of TTX. (e) Representative confocal stacks of 20-μm stretches from secondary apical dendrites of a CRE-expressing cell and a control cell. (f) Representative confocal stacks of CRE-expressing and control cells. (g) Average spine density in control (n = 8) and CRE-expressing (n = 10) neurons (P = 0.88). (h) Average number of dendritic branch points in control (n = 8) and CRE-expressing (n = 9) neurons (P = 0.99). (i) Average dendritic length in control (n = 8) and CRE-expressing (n = 9) neurons (P = 0.87).

We next investigated the structural basis for the increase in synaptic input to NR1 knockout cells. NR1-deficient neurons could have a greater number of postsynaptic spines, a more elaborate dendritic tree to accommodate more synapses, or fewer “silent synapses” without a change in basic cellular structure. We filled infected and uninfected neurons in acute slices with fluorescent dyes and obtained high-resolution confocal reconstructions of the recorded neurons. Surprisingly, we did not observe a change in spine density (control: 21 ± 1 spines per 20 μm, P = 8; CRE: 21 ± 1 spines per 20 μm; n = 10, P = 0.88), or a change in the overall dendritic structure of NR1 knockout neurons (Fig. 4 e–i). This result excludes a requirement for NMDA receptors in the normal morphological development of pyramidal neurons. Without gross changes in morphology, it would appear that a sizeable fraction of silent synapses in NR1 knockout cells have acquired AMPA receptors. In other words, the results suggest that elimination of NMDA receptors during synaptogenesis drives premature recruitment of AMPA receptors to nascent synapses, which would normally have expressed only NMDA receptors.

Competition between afferent axons plays a crucial role in determining the final synaptic architecture of the neuromuscular junction (16). Thus, we sought to test whether the regulatory role of NMDA receptors on synaptic maturation relies on a comparison of activity between neighboring control and NR1-deficient neurons. We took advantage of the ability to densely infect regions of the hippocampus by using a high-titer AAV-CRE-GFP virus (10¹² to 10¹³ pfu). If competition between neighboring pyramidal cells mediated by NMDA receptor activity influences synapse maturation, then in regions where nearly all cells are infected by CRE-expressing virus and lack NMDA receptors there should be no increase in synaptic input, as compared with sparsely infected regions. We injected AAV-CRE-GFP virus into the hippocampus of postnatal (P0) NR1^fl/fl mice to delete NMDA receptors as early as possible after birth. Consistent with the results presented above, postnatal in vivo elimination of NMDA receptors enhanced AMPA currents (I_infected/I_control = 1.97 ± 0.30, n = 32 pairs, P < 0.0005; Fig. 5 a and b) in paired recordings in slices with sparse infection. We then compared mEPSC frequency between infected neurons in regions of slices with very sparse and very dense CRE-GFP expression (see Methods and Fig. S2 c and d). In disagreement with the notion that the effects of NR1 deletion rely on synaptic competition, mEPSC frequency was increased to a similar extent in NR1-deficient neurons whether or not most of their neighbors lacked NMDA receptors as well (Fig. 5c). Because viral infection is necessarily variable and incomplete, we also compared mEPSC frequency in CA1 pyramidal neurons in a global, forebrain NR1 knockout (NEX-CRE;NR1^fl/fl mice used above) and their NR1^fl/fl control littermates. Unlike the dense viral infection, we found that mEPSC frequency did not differ between control and NEX-CRE;NR1^fl/fl mice (Fig. S3e). This result suggests that competition may play a role, but because global NR1 knockouts have severely compromised viability (they have reduced body weight, and none survive pass weaning) we cannot rule out compensatory effects. Additionally, the NEX promoter drives CRE expression as early as E11.5 (17), as compared with the postnatal viral infection, which could also account for the differences. Future work, perhaps with specific inducible CRE lines of mice (16), should be able to carefully separate the cell-autonomous role of NMDA receptor function from any part it has in comparing the afferent activity of competing axons.

Fig. 5.

Postnatal NR1 deletion increases the number of AMPA receptor-containing synapses but does not depend on synaptic competition between neighboring cells. (a) Scatter plot of NMDA EPSCs of pairs of simultaneously recorded neurons in acute slices from postnatally AAV-CRE-GFP-infected and neighboring control cells (n = 32 pairs, P < 10⁻⁸). (Inset) Representative traces. (b) Scatter plot of AMPA EPSCs from the same cells (n = 31 pairs, P < 0.0005). (c) Cumulative distribution of mEPSC interevent interval between control (n = 12) and CRE-expressing cells in slices with sparse (n = 12) or dense (n = 26) infection. (d) Model for the role of NMDA receptors in synaptic maturation. During development, modest activity through NMDA receptors at silent synapses prevents the constitutive trafficking of AMPA receptors to the PSD. This mechanism ensures that synapses become functional only after strong or correlated activity, when enough calcium entry through these NMDA receptors overrides the inhibitory pathway and drives AMPA receptor insertion (like LTP). When NMDA receptors are genetically deleted, the inhibitory signal is absent, and AMPA receptors traffic to the PSD in the absence of any NMDA receptor activity.

Finally, we examined whether NR1 deletion in mature animals influences excitatory circuitry in the hippocampus. We bred NR1^fl/fl mice to a late-expressing CA1-specific CRE line (12, 18) and compared the amplitude and frequency of quantal events between knockout and control animals. Consistent with previous findings, deletion of NR1 in mature animals (>P60) had no effects on any of the basic parameters of AMPA receptor-mediated synaptic transmission (Fig. S4).

Discussion

Taken together, our results suggest a number of surprising conclusions. First, it is unequivocal that synaptic maturation cannot absolutely require NMDA receptor activation, because in its absence fully functioning synapses are readily formed. Second, because NR1 deletion actually enhances the number of functional synapses, the primary role of NMDA receptor activity during early brain development is to limit the number of functional synaptic inputs. Third, because NR1 deletion does not increase spine number, it is likely that NMDA receptors regulate the unsilencing of nascent synapses rather than controlling the generation of entirely new spines.

Many studies have aimed at ascertaining the precise role of NMDA receptors in the development of cortical circuitry. Most have relied on eliminating NMDA receptor function with NMDA receptor antagonists applied in vivo to various brain regions or in vitro to hippocampal slice cultures. In vivo, NMDA receptor antagonists potently alter afferent patterning in visual areas (7) and can promote remodeling of thalamic neurons (4). In dissociated neurons, NMDA receptor antagonism has been reported to limit synapse unsilencing assayed by immunohistochemical techniques (2), and, in the autaptic preparation, chronic AP5 application reduced quantal amplitude but did not diminish evoked AMPA EPSC amplitude (19). In organotypic hippocampal slice culture, one study demonstrated that NMDA receptor antagonism delays the AMPAfication of CA1 synapses (3), whereas another showed that blocking NMDA receptors can actually limit synaptic connectivity by constraining dendritic growth and presynaptic elaboration (8). Although antagonists offer reversible elimination of NMDA receptor function, one confounding factor is that widespread blockade of NMDA receptors has also been reported to massively reorganize and cluster NMDA receptors in neurons, which could have various downstream effects (20).

Alternatively, several studies have eliminated NMDA receptors genetically by targeting the NR1 locus. The overall brain architecture of the global NR1 knockout at birth is normal, and some excitatory AMPA receptor-containing synapses do form in the brainstem. However, patterning in the somatosensory pathway is completely disrupted, an effect observed in the barrel field even when the NR1 deletion is limited to cortex but not thalamus (5). Additionally, cortex-specific NR1 or NR2B deletion has been shown to decrease spine density and alter synapse function in layer 2/3 barrel neurons (21, 22). Finally, a recent study used RNAi to knock down NR1 expression in hippocampal pyramidal neurons in organotypic slices and found that compromised NMDA receptor expression decreased net AMPA receptor-mediated input (23).

Although this literature on the role of NMDA receptors in regulating synapse maturation is complex and perhaps conflicting, the various approaches and different preparations used could easily explain the differences. However, both global pharmacological blockade and widespread genetic deletion suffer from a loss of specificity: they cannot separate the cell-autonomous role of NMDA receptors from the indirect effects on network activity associated with complete elimination of NMDA receptor function (10). Mosaic genetic deletion, the technique used in this study, has the unique advantage that NMDA receptor activity is eliminated only sparsely and selectively within the hippocampal network. This very focal manipulation should interfere only with the cell-autonomous involvement of NMDA receptors in synaptic development and not perturb network function.

Activation of NMDA receptors can lead to either increases or decreases in synaptic strength depending on the magnitude of incoming activity, better known as long-term potentiation (LTP) and long-term depression (LTD) (24). By completely eliminating NMDA receptors during development, activity-dependent modulation of synaptic strength in either direction is abolished. Based on our results, we suggest the following model (Fig. 5d): ongoing activity during early development normally stimulates NMDA receptors at silent synapses to an extent that actually limits the constitutive trafficking of AMPA receptors to the PSD, perhaps by engaging a pathway shared with LTD (Fig. 5d). Although some constitutive insertion can occur, this inhibitory mechanism ensures that synapses primarily gain AMPA receptors and become functional after strong or correlated activity, when enough calcium entry through these NMDA receptors overrides the inhibitory pathway and drives an LTP-like mechanism. When NMDA receptors are genetically deleted, the inhibitory pathway is lost, and AMPA receptors traffic to the PSD even in the absence of any NMDA receptor activity. To fully separate the role of LTD and LTP in synapse development during brain ontogeny it will be necessary to find mutations, optimally in the NMDA receptor itself, that eliminate the ability to express either LTD or LTP, but not both. Our findings point to a crucial role for NMDA receptors in synaptic development but demonstrate that their primary role is to limit, rather than to promote, synapse maturation.

Methods

Slice Preparation and Recording.

Transverse hippocampal slices 300–400 μm thick were cut from NR1^fl/fl mice on a Leica vibratome in normal ACSF. After at least 1 h of incubation at room temperature slices were transferred to a submersion chamber on an upright Olympus BX51 microscope, and CA1 pyramidal cells were visualized by IR differential interference contrast microscopy. The extracellular solution contained 119 mM NaCl, 2.5 mM KCl, 26 mM NaHCO₃, 1 mM Na₂PO₄, 11 mM glucose, 4 mM CaCl₂, 4 mM MgCl₂, 0.1 mM picrotoxin, or 0.01 mM GABAzine and was saturated with 95% O₂/5% CO₂. The intracellular solution contained 135 mM CsMeSO₄, 8 mM NaCl, 10 mM Hepes, 0.3 mM Na₃GTP, 4 mM MgATP, 0.3 mM EGTA, 5 mM QX-314, and 0.1 mM spermine. For anatomical analysis, Alexa Fluor 488 or 555 (0.5 mM; Invitrogen) was included in the internal solution. CA3 axons were stimulated with low-resistance monopolar glass pipettes containing ACSF placed in stratum radiatum. Cells were recorded with 3- to 5-MΩ borosilicate glass pipettes. Series resistance ranged between 8 and 25 MΩ and was not compensated for. Experiments in which series resistance changed by >20% were excluded from analysis.

GFP⁺ neurons were identified by epifluorescence microscopy. All paired recordings involved simultaneous whole-cell recordings from one GFP⁺ neuron and a neighboring GFP⁻ neuron. The stimulus was adjusted to evoke a measurable, monosynaptic EPSC in both cells. AMPA currents were measured at a holding potential of −60 mV, and NMDA EPSCs were measured at +40 mV and at 150 ms after the stimulus, at which point the AMPA EPSC has completely decayed. This separation of the two components of the EPSC was confirmed in some experiments by applying either D-AP5 or NBQX to isolate the AMPA or NMDA conductances, respectively.

Paired-pulse ratios were measured by giving two pulses at a 40-ms interval and taking the ratio of the two peaks of the EPSCs from an average of 15–20 sweeps. Rectification indices were calculated as the ratio of the slopes of the two lines connecting average EPSC values at −60, 0, and +40 mV. mEPSCs were acquired in the presence of 500 nM TTX and semiautomatically detected by offline analysis using in-house software in Igor Pro (Wavemetrics). At least 75–150 events were collected for each cell. For recording sEPSCs, slices were bathed in a high-Ca²⁺/low-Mg²⁺ buffer, including a low dose of the potassium channel blocker 4-amino-pyridine, which elevates spontaneous firing and release probability (15). This strongly increases the number of action-potential-dependent spontaneous currents against the background of mEPSCs.

All paired recording data were analyzed statistically with a two-tailed paired Student t test. For analysis of cumulative distributions the Kolmogorov–Smirnov test was used. For all other analyses an unpaired t test was used. All errors bars represent standard error measurement.

For the experiment in Fig. 5c, we tested whether the density of GFP expression accurately reported the extent of NR1 deletion. We screened slices for high GFP signal under low magnification, and then pyramidal neurons at the center of the region with highest GFP signal were randomly sampled for the presence of NMDA currents, without first determining whether the targeted neuron was GFP⁺ (for pictures see Fig. S2). Under these conditions a high proportion of sampled neurons (26/32, ≈81%) exhibited undetectable NMDA EPSCs, confirming the efficacy of the deletion.

Organotypic Slice Culture.

Cultured slices were prepared and transfected as previously described (25). Briefly, hippocampi were dissected from P5–P9 NR1^fl/fl, NEX-CRE;NR1^fl/fl, or wild-type animals and cultured for 2–3 days before biolistic transfection with GeneGun (Bio-Rad). For NR1 deletion NR1^fl/fl slices were transfected with pCSCG CRE-IRES-GFP and cultured for an additional 12–17 days before recording. For NR1 rescue, slices from NEX-CRE;NR1^fl/fl animals were transfected with pCI NR1-GFP (26) and cultured for 5–10 days before recording. For recording from organotypic slices, ACSF was supplemented with 10 μM 2-chloroadenosine to dampen epileptiform activity, and GABA_A receptors were blocked by a combination of picrotoxin (0.1 mM) and bicucculine (0.02 mM) or GABAzine (0.01 mM). For experiments in Fig. 2c, slices were incubated in control, TTX (2 μM), or a mixture of NMDA receptor antagonists (50 μM D-AP5, 20 μM CPP, and 40 μM MK-801). Medium was changed every other day, and drugs were added fresh for each replacement.

In Utero Viral Injection.

Timed pregnant NR1^fl/fl or wild-type mice (E14.5–E16.5) were anesthetized, and the abdomens were cleaned with 70% ethanol. For each, a midline incision was made and the uterus was exposed. The cerebral vesicle of each embryo was transilluminated with a fiber optic source, and 1–2 μl of AAV-GFP-CRE solution prepared at 10⁸ to 10⁹ pfu/μl in 1× PBS with 0.04% Trypan Blue was injected into the lateral ventricle with a beveled glass micropipette. After the procedure, the uterus was placed back into the abdominal cavity and the abdominal cavity was filled with prewarmed 1× PBS. The abdominal wall and skin were sutured to allow the embryos to develop further. The mother was placed on a 37°C plate until recovery from the surgery. The entire surgical procedure was completed within 45 min.

In Vivo Postnatal Viral Injection.

NR1^fl/fl mice were injected on the day of birth with concentrated AAV-GFP-CRE viral solution (≈10¹² to 10¹³ pfu). Newborns were anesthetized on ice and then stabilized in a custom ceramic mold before being injected with 4.2–9.2 nl of viral solution at six sites targeting the hippocampus intracerebrally using Nanoject (Drummond Scientific) and a beveled glass injection pipette. Animals recovered immediately after injection and were used for recording 13–19 days afterward.

Anatomical Analysis.

CA1 pyramidal cells were filled with Alexa Fluor 488 or 555 dyes through the patch pipette for at least 5–10 min. After recording, slices were fixed in 4% PFA for 30 min at room temperature, then washed at least three times with PBS. Slices were imaged in PBS by using a Zeiss confocal laser scanning microscope with 488 or 543 laser lines. For dendritic analysis, 3D stacks of each neuron were taken by using a ×25 water immersion objective, and the 2D projections were imported into Neurolucida software running AutoNeuron (MBF Bioscience). The dendritic tree was semiautomatically traced and then analyzed by using NeuroExplorer (MBF Bioscience). For spine analysis, 3D stacks of two to three 20-μm dendritic stretches of each neuron from secondary apical dendrites were collected by using a ×40 water immersion lens, and spines were counted in 3D projection mode by using Zeiss software. All anatomical analysis was done blind to the genotype of the cell. For statistical analysis Student's t test was used.