Research Article

Rapid Spine Delivery and Redistribution of AMPA Receptors After Synaptic NMDA Receptor Activation

See allHide authors and affiliations

Science  11 Jun 1999:
Vol. 284, Issue 5421, pp. 1811-1816
DOI: 10.1126/science.284.5421.1811


To monitor changes in α-amino-3-hydroxy-5-methyl-4-isoxazole propionate (AMPA) receptor distribution in living neurons, the AMPA receptor subunit GluR1 was tagged with green fluorescent protein (GFP). This protein (GluR1-GFP) was functional and was transiently expressed in hippocampal CA1 neurons. In dendrites visualized with two-photon laser scanning microscopy or electron microscopy, most of the GluR1-GFP was intracellular, mimicking endogenous GluR1 distribution. Tetanic synaptic stimulation induced a rapid delivery of tagged receptors into dendritic spines as well as clusters in dendrites. These postsynaptic trafficking events required synapticN-methyl-d-aspartate (NMDA) receptor activation and may contribute to the enhanced AMPA receptor–mediated transmission observed during long-term potentiation and activity-dependent synaptic maturation.

Excitatory synaptic transmission in the vertebrate central nervous system is mediated by activation of AMPA- and NMDA-type glutamate receptors. Repetitive synaptic activity transiently activates NMDA receptors and triggers long-lasting plasticity (1), expressed, at least in part, as an increase in AMPA receptor function (2, 3). The molecular basis for activity-induced changes in AMPA receptor function is not known and may include changes in channel conductance (4), possibly after receptor phosphorylation (5), or delivery of AMPA receptors to synapses, as has been documented during development (6). We investigated if an increase in AMPA receptor number at synapses may occur rapidly during NMDA receptor–dependent synaptic plasticity.

AMPA receptors are oligomers formed by a combination of four different subunits, GluR1 to 4 (GluRA to D) (7). A substantial proportion of endogenous AMPA receptors in hippocampal neurons have the GluR1 subunit (8). We constructed a recombinant GluR1 tagged with green fluorescent protein (GFP) at the putative extracellular NH2-terminus (GluR1-GFP; Fig. 1A) (9). This protein was expressed in human embryonic kidney (HEK) 293 cells; extracts showed a single band by protein immunoblotting of the expected molecular mass (Fig. 1B). Whole-cell recordings from GluR1-GFP–transfected HEK 293 cells showed inwardly rectifying responses to puffed agonist (Fig. 1C) (7). Cotransfection of GluR1-GFP with wild-type GluR2 yielded responses with no rectification (Fig. 1C), indicating effective hetero-oligomerization between GluR1-GFP and GluR2, as homomeric GluR2 can produce little current (7).

Figure 1

Expression and functional analysis of GluR1 fused with GFP (GluR1-GFP). (A) Schematic drawing of expected transmembrane topology of GluR1-GFP. N, NH2-terminus; C, COOH-terminus. (B) Protein immunoblotting of membrane fractions of rat forebrain and HEK cells transfected with GluR1-GFP or GluR1, probed with antibody to GluR1 COOH-terminus. The increase in molecular mass in GluR1-GFP is comparable with the molecular mass of GFP (27 kD). There was no other band detected at lower molecular mass. (C) (Left) Current-voltage plots of whole-cell responses from HEK 293 cells transfected with GluR1-GFP alone (•,N = 3) and with GluR1-GFP with GluR2 (○,N = 3). Responses are normalized to values at –60 mV. (Right) Representative responses at membrane potentials from –80 to +60 mV (20-mV steps). One millimolar kainic acid was puff applied in the presence of 100 μM cyclothiazide.

GluR1-GFP was introduced into neurons with Sindbis virus expression system (10, 11). In hippocampal dissociated cultured neurons (Fig. 2) (12), GluR1-GFP showed distribution throughout the dendritic tree with expression levels in dendrites approximately three times that of endogenous GluR1 (13). Immunostaining for surface (Fig. 2D) (14) recombinant receptor displayed a punctate distribution that colocalized with surface labeling of endogenous GluR2 (Fig. 2D) as well as with a presynaptic marker (synapsin 1; Fig. 2D). Whole-cell responses to caged glutamate showed greater rectification in GluR1-GFP–expressing neurons, indicating functional delivery of homomeric GluR1-GFP to the surface (Fig. 2B) (15).

Figure 2

Expression of GluR1-GFP in dissociated cultured hippocampal neurons is targeted to synapses. (A) Neuron 1 day after infection expressing GluR1-GFP. Scale bar, 30 μm. (B) Whole-cell responses to uncaged glutamate (10 ms, 40 μM CNB-glutamate). (Left) Current-voltage relations plotted, normalized to values at –60 mV. •, infected cell (n = 8); ○, uninfected cell (n = 8). (Right) Responses from uninfected (top) and infected (bottom) neuron at different holding potentials (−60 mV to +60 mV, 20-mV steps). (C) Dissociated cultured neurons expressing plain GFP were fixed in nonpermeabilizing (NP) or permeabilizing (P) conditions (14), stained with antibody to GFP, and imaged with filters for GFP (top) or antibody to GFP (Texas Red) (bottom). The same gray scale was used on all images. Scale bar, 2 μm. (D) Surface expression of GluR1-GFP. Immunostaining with antibodies to GFP (polyclonal for double staining with GluR2 and monoclonal for synapsin I, both at 1:100; Clontech) under nonpermeabilized conditions reveals punctate pattern of expression. This punctate pattern colocalizes with endogenous GluR2 (left) detected with antibody against extracellular domain of GluR2 (10 μg/ml; Chemicon International) and synapsin I (Syn I) (1:100) (immunostained after permeabilizing conditions) (right). Scale bar, 5 μm.

We next examined the distribution of GluR1-GFP in neurons of organotypic hippocampal slice cultures (16) because such a preparation can display robust long-term potentiation (LTP) (Fig. 3A). Two to 3 days after focal viral infection of the CA1 region (17), several hundred neurons showed GluR1-GFP expression (Fig. 3B) distributed throughout the apical and basal dendritic trees. To quantify the subcellular distribution of recombinant receptor in these slice cultures, we used postembedding immuno-gold electron microscopy with an antibody directed against GFP (18, 19). We examined apical dendritic areas in infected regions (Fig. 3, E and F). Most of the labeling in dendrites was located intracellularly in the dendritic shaft (88%). The remainder was largely (9%) on the dendritic shaft surface, with little (2%) in spines, and only three grains (0.4%) were found at postsynaptic densities, sites of synaptic contact. This distribution pattern was in general similar to that of endogenous GluR1, detected with an antibody to GluR1 in comparable dendritic regions from (noninfected) postnatal day 10 brain tissue (Fig. 3F). Most of the endogenous GluR1 labeling in dendrites was also intracellular in the dendritic shaft (71%), with 20% on the dendritic shaft surface and 8% in surrounding spines [3% was in postsynaptic densities (PSD)] (20).

Figure 3

GluR1-GFP expression in organotypic hippocampal slice culture is primarily intracellular. (A) Organotypic slice culture displays LTP in CA1. (Top) Plot of field excitatory postsynaptic potentials recorded in CA1 region of organotypic slice (mean ± SEM, N = 5; tetanus: time = 0). (Bottom) Whole-cell recording of synaptic responses from neuron expressing GluR1-GFP. One hundred synaptic stimuli (2 Hz) were paired with depolarization (0 mV) where indicated. (Insets) Representative responses obtained before and after LTP induction. Scale bar, 0.3 mV, 10 ms for top traces; 20 pA, 10 ms for bottom traces. (B) Expression of GluR1-GFP in pyramidal cells 2 days after infection, imaged with TPLSM. Scale bar, 20 μm. (C and D) Apical dendrite of CA1 pyramidal cells expressing either GluR1-GFP (C, top and bottom) or plain GFP (D, top and bottom). Scale bar, 5 μm. (E) Immuno-electron microscopic image of dendrite expressing GluR1-GFP. Postembedding immunolabeling was performed with antibody to GFP (18–20). Immunogold particles were mainly distributed inside dendrite. Scale bar, 0.3 μm. (F) Distribution of immunogold particles in different dendritic compartments for GluR1-GFP (left) and endogenous GluR1 (right). (Inset) Location of immuno-labeling. See (20) for details. (G) Surface expression of GluR1-GFP assessed with fluorescent immunostaining. Cells infected with GluR1-GFP were fixed and stained under nonpermeabilized (left) and permeabilized (right) conditions with antibody to GFP (Texas Red detection). Images detected in GFP channel (top) or antibody to GFP channel (bottom). Note immunostaining along dendritic membrane (bottom, left) with no detectable spines. Scale bar, 5 μm.

To monitor changes in the distribution of GluR1-GFP, we used time-lapse two-photon laser scanning microscopy (TPLSM) (21) and examined neurons in organotypic slices 2 to 3 days after infection. High-resolution optical stack images of dendritic regions revealed that the GluR1-GFP signal was fairly homogeneous (Fig. 3, B and C) along the dendrite and largely restricted from dendritic spines, consistent with the immuno-gold electron microscopic analysis (Fig. 3F) and in contrast to the distribution of plain GFP, which displayed numerous spines (Fig. 3D). To test the effect of synaptic activity on receptor distribution, we placed a small glass-stimulating electrode near (5 to 15 μm) a group of dendrites labeled with GluR1-GFP (22). In the absence of evoked activity, the GluR1-GFP distribution pattern was stable for hours with no signs of photodamage and little bleaching (see below; for example, Fig. 6A). However, delivery of a brief tetanic stimulus, which was sufficient to induce LTP in these slices (Fig. 3A), produced a rapid redistribution of GluR1-GFP (Figs. 4 to 6).

Two kinds of changes in GluR1-GFP distribution were detected after tetanic stimulation: delivery to spines and clustering in the dendritic shaft. Delivery of GluR1-GFP was measured in 38 spines from five experiments (Fig. 4). In about half of these spines (17 of 38), the amount of fluorescence at the corresponding location in images obtained before a tetanus was near background (23) (termed “empty” spines, Fig. 4A, arrow a), whereas in the remaining 21 of 38 spines (termed “active” spines, Fig. 4A, arrow b), there was a detectable amount of GluR1-GFP before tetanus. The intensity distribution of these analyzed spines is shown in Fig. 4, B and C. The GluR1-GFP signal at “empty” spines increased from 200 ± 43 AU to 1737 ± 235 AU [measured in arbitrary fluorescence units (AU); mean ± SD, N = 17] after a tetanic stimulus. At “active” spines, the increase was from 1023 ± 101 AU to 2210 ± 235 AU (mean ± SD, N = 21). Such increases in GluR1-GFP signals at spines were never observed in the absence of tetanic stimulation (24).

Figure 4

Tetanic stimulation induces spine delivery and clustering of GluR1-GFP. (A) Column 1, GluR1-GFP expression in apical dendritic region. Stimulation electrode was placed in nearby region (∼5 to 10 μm from top left corner, outside imaged region). Column 2, region near stimulation electrode (top and middle: two different magnifications of same region) and another region (bottom) imaged before tetanus. a and b denote locations of interest. Column 3, same regions imaged 30 min after tetanic stimulation. Arrows mark regions a and b in column 2. Column 4, surface GluR1-GFP assessed with antibody to GFP immunostaining in nonpermeabilized fixation conditions. At region showing redistribution (top and middle), immunostaining detected increased GluR1-GFP on dendrite and spine-like structures. Scale bars, 2 μm. (B) Quantification of GluR1-GFP signal intensity of spines before and after tetanus. Spines were identified in images obtained 15 min after tetanus. Fluorescence was integrated over two to three optical sections containing spine and also from equivalent places before tetanus. Background fluorescence was determined in nearby regions of similar size without any obvious structure and was subtracted from all measurements. Spines were selected from five independent experiments carried out in identical experimental conditions. Data from the individual spines are connected by lines. Units are arbitrary fluorescence units (AU). Imaging parameters were identical before and after tetanus. (C) The same data plotted as histograms. Bin width, 200 AU.

In addition to spine delivery, tetanic stimulation produced clustering of GluR1-GFP in the dendritic shaft. The clustered receptor could be seen at the base of a spine (Fig. 5A, arrow) or with no detectable delivery to spines (Fig. 5A, arrowhead). Clustering was quantified by computing an index of an autocorrelation function (R 50%) calculated over a region of interest before and after tetanus (Fig. 5B) (25). In the absence of stimulation, this index changed little over time, on average increasing 5.4 ± 6.6% (mean ± SD, N = 20, randomly chosen dendrites) between two observation periods separated by 15 min. However, upon tetanic stimulation, 27 dendritic segments from 18 experiments became clustered (R 50% decreased by 17.8 ± 1.6%, mean ± SD) (26). Dendritic regions showing spine delivery of GluR1-GFP generally showed clustering of receptor (R 50% decreased by 18.3 ± 2.6% at the 10 dendrites analyzed above, showing delivery of GluR1-GFP to spines after tetanus).

Figure 5

Tetanic stimulation induces clustering of GluR1-GFP. (A) TPLSM images of dendrite before (20 min, left; 10 min, middle) and after (right) tetanic stimulation. Note marked clustering of signal in dendrite, including at the base of a spine (arrow) and at regions without obvious spines (arrowhead). In this example, spine (arrow) became only slightly brighter (10% increase in fluorescence) and no other spines emerged. Scale bar, 2 μm. (B) Autocorrelation function computed over boxed region in (A) at 20 min (•) and 10 min (▵) before and 15 min after (○) stimulation (25). Distance at which function decays to 50% (R 50%) was used as a measure of clustering. A decrease in R 50% value indicates cluster formation. (C) Dendrites with clustering have more surface GluR1-GFP. Images were taken before and after tetanic stimulation and after immunostaining with antibody to GFP under nonpermeabilized fixation condition. Amount of GluR1-GFP on surface versus total amount of GluR1-GFP is plotted against the ratio ofR 50% value before and after the tetanus for dendritic segments that appeared (•) or did not appear (○) to show clustering. A larger value for anti-GFP/GFP ratio indicates that more tagged receptor reached the surface of membrane. A decrease in R 50% value (indicating clustering) is correlated with a larger anti-GFP/GFP ratio (R = −0.78; P < 0.01).

We wished to determine if tetanus-induced redistribution of GluR1-GFP included delivery to the surface. We first established a method using TPLSM to image surface recombinant receptor in fixed slices (Fig. 3G) (14). The distribution and quantification of GluR1-GFP with these methods (13.3 ± 0.9% on surface) generally agree with values obtained with immuno-gold electron microscopy (9% on surface, Fig. 3F). Regions examined in live tissue during stimulation were analyzed for surface distribution after fixation. Regions in which GluR1-GFP had undergone clustering with tetanic stimulation showed a greater amount of receptor at the surface (Figs. 4A, column 4, and 5C; 18.6 ± 0.2 % on surface), although most of the receptor still remained intracellular. At spines that showed GluR1-GFP delivery after a tetanus (including previously “empty spines”), surface GluR1-GFP could also be detected (Fig. 4A). This indicates that some of the GluR1-GFP delivered into spines after a tetanus reached the spine surface, suggesting their contribution to an increase in synaptic transmission.

To determine whether the redistribution of GluR1-GFP by tetanic stimulation requires synaptic activation of NMDA receptors, we conducted experiments with (d,l)-2-amino-5-phosphono valeric acid (APV), a reversible NMDA receptor antagonist (Fig. 6). With APV in the bath, tetanic stimulation produced no clear redistribution of GluR1-GFP (neither spine delivery nor clustering; Fig. 6A). After washing APV for 45 min, another tetanus was delivered at the same site. Now spine delivery and clustering could be detected (see Fig. 6, no APV, −7 and 15 min). Ensemble averages from several experiments in which spine delivery and clustering were monitored in the presence and subsequent absence of APV are shown in Fig. 6B. These results show that both clustering and spine delivery of GluR1-GFP require synaptic activation of NMDA receptors. These experiments also demonstrate that the effect of tetanic stimulation is not due to direct depolarization of dendrites by the current passed through the stimulating electrode, because such effects would not be blocked by APV.

Figure 6

NMDA receptor antagonist reversibly blocks tetanus-induced redistribution of GluR1-GFP. (A) Images of apical dendritic segments obtained at different times during experimental period. (Top) In the presence of 100 μMd,l-APV, tetanus produces little change in fluorescence pattern. (Bottom) After 45 min of drug wash, tetanus at the same site produces both delivery to spines and clustering of GluR1-GFP. In this example, spine delivery persisted at 50 min after tetanus in only two of five spines and clustering reverted to smooth distribution. Time stamps are in minutes relative to tetanus. Scale bar, 5 μm. (B) Ensemble averages from experiments carried out as in (A). (Left) Spines showing increased GluR1-GFP fluorescence after tetanus in no APV (tetanus 2) were measured before and after tetanus in APV (tetanus 1). N = 12 spines from four experiments. For details of quantification, see legend toFig. 4B. (Right) Dendritic regions showing clustering in no APV (tetanus 2) were measured before and after tetanus in APV (tetanus 1).N = 7 dendrites from seven experiments. For details of quantification, see (25).

In this study, we showed that the GFP-tagged GluR1 receptor is electrophysiologically functional and mimics a number of cell-targeting properties of endogenous receptors. In dissociated neurons, the protein is delivered to synapses in the absence of evoked activity. In contrast, in slices given no stimulation, a large fraction of the recombinant GluR1-GFP, as well as endogenous GluR1, is found in the intracellular dendritic compartment and excluded from synapses. This difference may explain the observed difficulty with which LTP is generated in dissociated neurons (27). This intracellular pool is within 1 to 2 μm of synapses and thus could be rapidly delivered to synaptic sites during plasticity.

Indeed, we found that GFP-tagged receptors in hippocampal slice neurons were rapidly recruited to dendritic spines after a tetanic stimulus (Figs. 4 and 6). Immunostaining indicated that at least some of the recruited GluR1-GFP reached the spine surface. The delivery to the dendritic shaft surface may also represent synaptic delivery, as shaft synapses (or short “stubby” spines) are more common in young tissue (28). The spine delivery of the tagged AMPA receptor required synaptic NMDA receptor activation, providing a strong link between receptor recruitment and activity-induced forms of plasticity. These results provide direct evidence showing rapid effects of synaptic activity on postsynaptic membrane trafficking.

In about half of the spines detected with GluR1-GFP after tetanus, there was no fluorescence at the corresponding region before tetanic stimulation. On the basis of their length (0.95 ± 0.17 μm) and a previous study (29), these spines are not likely to have been generated after tetanic stimulation. Our previous study indicated that tetanic stimuli do not generate short spines, but rather such stimuli generate filopodial structures that are typically >3 μm in length (29). In view of these observations, it is likely that GluR1-GFP was delivered to existing spines and not newly formed spines, although such a possibility cannot be excluded by our results. If receptors were delivered to existing “empty” spines, these could represent “silent synapses”: synapses with only NMDA receptors that gain AMPA receptors during LTP (3).

Our results build on a number of studies suggesting that the delivery of AMPA receptors to synapses contributes to activity-dependent plasticity. Inhibition of membrane fusion processes in the postsynaptic cell blocks LTP (30). Furthermore, the COOH-termini of AMPA receptor subunits GluR2 and GluR4c bindN-ethylmaleimide–sensitive fusion protein, a protein involved in membrane fusion processes (31). Vesicular organelles, possibly undergoing exocytosis and endocytosis, have been detected with electron microcopy in spines (32). And last, dendrites can display a calcium-evoked exocytosis of trans-Golgi–derived organelles that is mediated by the calcium/calmodulin-dependent protein kinase II, an enzyme thought to mediate LTP (33). Other postsynaptic mechanisms, such as an increase in conductance of AMPA receptors (4, 5), may also occur in parallel. Our results also do not rule out a contribution by presynaptic modifications.

In addition to the spine delivery of GluR1-GFP, tetanic stimulation induced the formation of clusters of the tagged receptor within dendrites. These structures may be related to the spine apparatus, membranous structures at the base of spines (32) that appear to contain AMPA receptors (34). The entry of calcium through synaptic NMDA receptors may cause nucleation of AMPA receptor–containing membranes close to active synapses. Once formed, such sites may serve several functions. These sites may replenish those receptors delivered to spines during plasticity. Additionally, they may serve as a “synaptic tag” (35), providing a docking site for AMPA receptors synthesized at distant sites. Last, they could provide a site for local AMPA receptor synthesis (36). In these capacities, such clusters could represent a structural modification serving as a long-lasting memory mechanism.

  • * To whom correspondence should be addressed. E-mail: malinow{at}


Stay Connected to Science

Navigate This Article