Search tips
Search criteria 


Logo of nihpaAbout Author manuscriptsSubmit a manuscriptHHS Public Access; Author Manuscript; Accepted for publication in peer reviewed journal;
J Neurosci. Author manuscript; available in PMC 2008 May 30.
Published in final edited form as:
PMCID: PMC2408376

β1-Integrins Are Required for Hippocampal AMPA Receptor-Dependent Synaptic Transmission, Synaptic Plasticity, and Working Memory


Integrins comprise a large family of cell adhesion receptors that mediate diverse biological events through cell–cell and cell–extracellular matrix interactions. Recent studies have shown that several integrins are localized to synapses with suggested roles in synaptic plasticity and memory formation. We generated a postnatal forebrain and excitatory neuron-specific knock-out of β1-integrin in the mouse. Electrophysiological studies demonstrated that these mutants have impaired synaptic transmission through AMPA receptors and diminished NMDA receptor-dependent long-term potentiation. Despite the impairment in hippocampal synaptic transmission, the mutants displayed normal hippocampal-dependent spatial and contextual memory but were impaired in a hippocampal-dependent, nonmatching-to-place working memory task. These phenotypes parallel those observed in animals carrying knock-outs of the GluR1 (glutamate receptor subunit 1) subunit of the AMPA receptor. These observations suggest a new function of β1-integrins as regulators of synaptic glutamate receptor function and working memory.

Keywords: integrins, AMPA receptors, basal synaptic transmission, LTP, working memory, synaptic plasticity


A key feature underlying nervous system function is the intricate coordination of signaling at synapses. Dynamic changes in the strength of signaling between the presynaptic and postsynaptic terminals in response to experience underlie memory formation through the actions of a plethora of molecules. Cell adhesion molecules, which are involved in mediating cell–cell and cell–extracellular matrix interactions, have been implicated in synaptic function, synaptic plasticity, and memory formation (for review, see Benson et al., 2000; Dityatev and Schachner, 2003; Yamagata et al., 2003).

One family of important cell adhesion receptors is the integrins. Integrins are transmembrane heterodimers of α- and β-subunits (Giancotti and Rouslahti, 1999; Coppolino and Dedhar, 2000; Schwartz, 2001; van der Flier and Sonnenberg, 2001). Studies of integrins from tissues other than the nervous system have revealed several unique features of these molecules. First, integrins transduce biochemical signals bidirectionally. In the “outside-in” signal transduction pathway, integrins can activate a number of cytoplasmic tyrosine kinases in response to extracellular ligand binding, including FAK (focal adhesion kinase), Src, and PI3 (phosphatidylinositol 3-kinase), thus leading to the activation of several dominant signaling pathways such as the MAPK (mitogen-activated protein kinase) and Akt/protein kinase B (PKB) cascades. In the “inside-out” pathway, physiological events within a cell can modulate the affinity of integrins for their ligands through an induced, transmembrane conformational change (Tadokoro et al., 2003). Second, integrins interact physically with extracellular molecules and the cytoskeleton, thus providing a physical link between the external environment of the cell and intracellular components that dictate its architecture. Although these remarkable features are not fully understood mechanistically, they provide enormous potential power to these molecules for regulating synaptic functions that underlie animal behavior.

Indeed, accumulating evidence has shown that many integrins are localized to synapses (Einheber et al., 1996; Pinkstaff et al., 1999; Prokop, 1999; Chan et al., 2003) and that they are involved in synaptic function and memory formation. For example, in vitro studies have shown that integrin activation produces rapid changes in spontaneous electrical activity and voltage-gated Ca2+ currents on Lymnaea motoneurons (Wildering et al., 2002) or enhancement of fast excitatory transmission in rat hippocampal slices (Kramar et al., 2003). Long-term potentiation (LTP) generated at the hippocampal CA3:CA1 synapses becomes unstable in the presence of integrin antagonists (Bahr et al., 1997; Staubli et al., 1998; Chun et al., 2001; Kramar et al., 2002). A direct role of integrins in memory formation was first demonstrated in the Drosophila olfactory memory mutant, Volado, which is impaired in the expression of an α-integrin (Grotewiel et al., 1998). Furthermore, mice triply heterozygous at the α3-, α5-, and α8-integrin genes are defective in paired-pulse facilitation (PPF), LTP, and spatial memory (Chan et al., 2003). Despite the broad support for the integrin family of molecules in synaptic function and behavior, the roles for individual integrin subunits in different aspects of neuronal physiology and types of memory remain unclear.

In the present study, we examined the effects of a postnatal and forebrain-specific knock-out of the β1-integrin gene on synaptic physiology and memory formation. Our results show that basal excitatory synaptic transmission through AMPA receptors and LTP in the hippocampus are impaired in these β1-integrin mutants, but PPF remains intact. Behavioral studies revealed a selective deficit in a hippocampal-dependent working memory task but not in multiple variations of hippocampal-dependent spatial memory or conditioned fear memory tasks. Interestingly, a similar dissociation between LTP and spatial memory and a parallel deficit in working memory have also been observed in knock-outs of the AMPA receptor subunit glutamate receptor subunit 1 (GluR1), suggesting a regulatory link between integrin signaling and function of the AMPA receptor.

Materials and Methods

Mice and genotyping

The floxed β1-integrin and α-calcium/calmodulin-dependent protein kinase II–cAMP response element (CaMKII–Cre) mice were kindly provided by E. Fuchs (Rockefeller University) and S. Tonegawa (Massachusetts Institute of Technology, Cambridge, MA), respectively. Animals were maintained as outcrossed stocks to C57BL/6 (The Jackson Laboratory, Bar Harbor, ME). Floxed β1-integrin knock-outs and the control littermates were obtained by crossing Cre/+; f(β1)/+ to +/+; f(β1)/+ animals. PCR genotyping of tail DNA used the following primers: CreF2 (5′-GGCGTTTTCTGAGCATACCTGGAA-3′) and CreR2 (5′-CACCATTGCCCCTGTTTCACTATC-3′) for Cre recombinase; β1–5 forward (For) (5′-CGCAGAACAATAGGTGCTGAAATTAC-3′) and β1–5 reverse (Rev) (5′-TGACACTGAGAACCACAAACGGC-3′) for the 5′ loxP insertion in the β1-integrin gene; β1–3For (5′-CGGCTCAAAGCAGAGTGTCAGTC-3′) and β1–3Rev (5′-CCACAACTTTCCCAGTTAGCTCTC-3′) for 3′ loxP insertion in the gene.

Mice were housed in conventional animal cages and maintained on a 12 h light/dark cycle. All animals were handled and treated during the experiments in ways approved by the Baylor College of Medicine Institutional Animal Care and Use Committee and according to national regulations and policies.

Western blotting and immunohistochemistry

For Western blotting, freshly dissected hippocampi were homogenized in buffer containing 137 mm NaCl, 20 mm Tris-HCl, pH 8.0, 10% glycerol, 1% NP40, 0.1% SDS, and 0.1% sodium desoxycholate. Twenty-five micrograms of the hippocampal protein extract was fractionated by 7.5% SDS-PAGE. The separated proteins were transferred to Immobilon-P membranes (Millipore, Bedford, MA). The membrane was incubated with a 1:500 dilution of anti-phospho-GluR1 [serine 831 (S831)] antibody (Upstate Biotechnology, Lake Placid, NY), 1:500 anti-phospho-GluR1 [serine 845 (S845)] antibody (Upstate Biotechnology), 1:400 anti-GluR1 antibody (Chemicon, Temecula, CA), 1:200 anti-GluR2/3 antibody (Upstate Biotechnology), or 1:100 anti-actin antibody (Chemicon) for 1 h at room temperature, washed three times, and then incubated for 30 min at room temperature with a horseradish peroxidase-conjugated secondary antibody (Jackson ImmunoResearch, West Grove, PA). Each blot was developed with ECL Western Blotting Detection Re-agent (GE Healthcare, Little Chalfont, UK). Signals detected in Western blots were scanned and quantified using Scion Image software (Scion, Frederick, CA). The net signal (total signal — background) for each animal was normalized against the corresponding signal for actin, and the average signal was calculated for each genotype. The relative expression level of each AMPA receptor subunit was expressed as the ratio of the average signal for each genotype relative to the Cre; +/+ control animals.

Immunohistochemical experiments were performed as described previously (Chan et al., 2003). Animals were anesthetized with isoflurane and perfused with 4% paraformaldehyde in 0.1 m phosphate buffer for 10 min. The dissected brain was separated into two hemispheres and postfixed in 4% paraformaldehyde for an additional 1 h at 23°C. Cryosections of the brain were then incubated in 7.5% goat serum in PBS [(in mm) 137 NaCl, 2.7 KCl, 10 phosphate buffer, pH 7.2] containing 0.15% Triton X-100 for 1–2 h at 23°C and then with 150 μl of diluted primary antisera [1:100 anti-GluR1 antibody (Chemicon), 1:100 anti-GluR2/3 antibody (Upstate Biotechnology), or 1:500 anti-NR1 antibody (Upstate Biotechnology)] overnight at 4°C. Sections were washed twice at 23°C with PBS plus 0.2% Triton X-100, pH 7.4 (PBST) for 10 min and then incubated with a biotinylated secondary antibody for 1 h. The sections were washed twice in PBST and finally incubated with Vectastain ABC reagent (Vector Laboratories, Burlingame, CA) with high salt (0.5 m NaCl) to reduce background staining. The immunostaining was visualized by reacting sections with 1 mg/ml diaminobenzidine/0.3% H2O2 for 5–10 min. After two washes in distilled water, the sections were mounted on slides with AquaMount (Lerner Laboratories, Pittsburgh, PA).

Electron microscopy of CA1 area

Animals were anesthetized with isoflurane (50 mg/kg) and perfused transcardially with 0.1 m phosphate buffer, pH 7.2, followed by a fixative containing 2.5% glutaraldehyde, 2.0% formaldehyde in 0.1 m cacodylate buffer with 2 mm CaCl2. After postfixation of the dissected brain overnight at 4°C, the hippocampal CA1 region was dissected and fixed again with 1% OsO4 in 0.1 m cacodylate buffer for 1 h. After washing, the tissue fragments were stained with 1% uranyl acetate for 1 h. The tissue was then dehydrated with an ethanol series and embedded in Glauert Med embedding resin (Electron Microscopy Sciences, Hatfield, PA). Ultra-thin sections were cut from the stratum radiatum and visualized with a Hitachi (Yokohama, Japan) 7500 digital electron microscope.

Hippocampal slice preparation and electrophysiology

Extracellular field recordings were obtained from the area CA1 stratum radiatum. Stimulation was delivered with a bipolar Teflon-coated platinum electrode and recording was obtained with the use of a glass microelectrode filled with artificial CSF (ACSF) [(in mm) 125 NaCl, 2.5 KCl, 1.24 NaH2PO4, 25 NaHCO3, 10 d-glucose, 2 CaCl2, and 1 MgCl2] (resistance, 1–4 MΩ). Input/output curves were generated by single-pulse stimulation of Schaffer collateral inputs with 2.5–28.5 mA in 2.5 mA increments. The 100 Hz stimulation consisted of two trains of 100 Hz frequency stimulation for 1 s, with each train separated by a 20 s interval. The saturating LTP protocol consisted of two trains of 100 Hz frequency stimulation for 1 s, with each train separated by a 20 s interval, repeated four times with each pair of stimulations separated by 6 min. Stimulus intensities were adjusted to give population EPSPs (pEPSPs) with slopes that were ≤50, 85, or 100% that of the maximum determined from an input/output curve. Potentiation was measured as the increase of the mean pEPSP after tetanic stimulation normalized to the mean pEPSP for the duration of the baseline recording. Experimental results were obtained from those slices that exhibited stable baseline synaptic transmission for a minimum of 30 min before the delivery of the LTP-inducing stimulus. Two-way ANOVA was used to determine statistical differences in synaptic transmission, PPF, and LTP. Depolarization was performed by means of the delivery of three groups of five trains of 100 Hz stimulation, with each train separated by 20 s and each group separated by 2 min. Total depolarization (the integral of the tetanic depolarization response) was calculated as the total area of high-frequency-induced depolarization during the entire 1 s application of 100 Hz stimulation.

Behavioral assays

Water maze

Briefly, mice were handled extensively for 2 weeks before training in the water maze. The animals were kept in individual cages during training and allowed to acclimatize to the water maze room for 1 h before the start of the experiment on each day. Each trial began by placing the mouse into the water facing the wall of the maze, and the animal was allowed to search for the platform until the animal climbed onto the platform or when a maximum of 60 s had elapsed. The animal was allowed to remain on the platform for 20 s before it was returned to its cage. Four trials were performed each day with an intertrial interval (ITI) of 60 min. Trials were balanced for starting positions and began at the 12, 3, 6, or 9 o'clock position of the pool. The platform location remained constant for any individual mouse for the duration of the training, but different animals were trained with the platform in different positions to avoid quadrant bias. Animals were trained for 5 d at the same time on each day. Twenty-four hours and/or 1 week after the last training trial, a probe trial (or transfer test) was administered in which the platform was removed from the pool, and animals were placed in a quadrant opposite to the location of the training platform and allowed to swim for 60 s. The time the mice spent searching for the platform in each quadrant and the number of times the mice crossed the virtual platform location was measured. The same procedure was performed in the “pattern completion” version of the probe trial, except that three of the four extramaze cues present on the surrounding walls of the water maze room had been removed.

Fear conditioning

During training, each mouse was placed inside a conditioning chamber (26 × 22 × 18 cm; San Diego Instruments) in the presence of white light and background noise generated by a small fan for 2 min before the onset of a conditioned stimulus (CS) (an 85 dB tone), which lasted for 30 s. A 2 s unconditioned stimulus (US) footshock (0.6 mA) was delivered immediately after the termination of the CS. Each mouse remained in the chamber for an additional 60 s, followed by another CS–US pairing. Each mouse was returned to its home cage after another 30 s. The test for contextual fear memory was performed 2 and 24 h after training by measuring freezing behavior during a 5 min test in the conditioning chamber. Freezing was defined as a lack of movement for 2 s or longer in each 5 s interval. Cued fear memory was tested in the presence of red light, coconut odor, and the absence of background noise. In addition, a triangular black box was put inside the conditioning chamber, and the grid floor was covered to present context different from that experienced during training. Each mouse was placed in this novel context for 3 min, 24 h after training to record baseline freezing after which they were exposed to the CS for another 3 min. Freezing behavior was recorded and processed by San Diego Instruments Photobeam Activity System (San Diego Instruments, San Diego, CA) software throughout each testing session.

T-maze nonmatching-to-place working memory task

The working memory task was performed as described by Reisel et al. (2002). The T-maze was made of wood and was elevated to a height of 1 m from the floor. It is consisted of a start arm (47 × 10 cm) and two identical goal arms (35 × 10 cm), each covered with the bedding used in the animals' home cage environment. Animals were restricted to 1 h of food access per day, which resulted in animals maintaining at 85% of their free-feeding weight. The animals were habituated to the maze and to drinking the food reward (50 μl of 50% sweetened, condensed milk) from a small dish (2 cm diameter) for 5 d before testing.

Each testing trial consisted of a sample run followed by a choice run. On the sample run, food reward was placed ~3 cm from the end on both goal arms, but access to one of them was blocked by a wooden block. A mouse was put inside an opaque cylindrical tube and transported from the home cage to the end of the start arm, where it was released facing the experimenter end wall of the start arm. The animal was allowed to find and drink the food reward in the unblocked goal arm. The animal was then collected into the transfer tube and returned to the end of the start arm. The time required for this transfer was <5 s. and represented the minimum time lapse between the sample and the choice runs but was not included in the actual timing of the delay time. For the choice run, the wooden block was removed, and the animal was released with or without delay to choose either goal arm. The animal was allowed to drink the food reward if it chose the previously unvisited arm. The animal was then returned to the home cage. Four trials with a 20 min ITI were performed on each animal per day for 18 d. The reward arm for each trial was assigned pseudorandomly, i.e., two times for each goal arm each day but with a different order for each animal. Working memory was expressed as the ratio of correct choices over total choices binned for 16 trials. For the statistical analysis, two-way ANOVA was performed followed by post hoc Scheffe's test.

Expression assay for surface receptors

Surface biotinylation and pull-down of cell surface proteins were performed as described previously (Thomas-Crusells et al., 2003). Freshly dissected hippocampal slices (400 μm) were placed in mesh chambers, washed twice in ice-cold ACSF for 5 min, and incubated with 100 μm sulfosuccinimidyl 2-(biotinamido)-ethyl-1,3-dithiopropionate (NHS-SS-biotin; Sigma, St. Louis, MO) in ACSF for 45 min. The biotinylated slices were then washed twice with 1 μm lysine (Sigma) in ACSF for 5 min to block the excess nonbound NHS–SS–biotin. To extract total proteins, two biotinylated slices were homogenized in 200 μl of cell lysis buffer [1% Triton X-100, 0.1% SDS, 1 mm EDTA, 50 mm NaCl, 20 mm Tris, pH 7.5, and protease inhibitor cocktail (Roche Diagnostics, Indianapolis, IN)] and incubated with rotation at 4°C for 30 min. The extract was then centrifuged for 5 min at 13,800 × g to remove extra debris. To precipitate the surface-biotinylated proteins, 400 μg of protein extract from the supernatant was incubated with 40 μl of UltraLink Immobilized Streptavidin beads (Pierce, Rockford, IL) in a total 800 μl of ACSF/protease inhibitor volume overnight at 4°C with gentle mixing. The precipitates were pulled down and washed twice with 500 μl of ACSF/protease inhibitors by brief centrifugation and, finally, resuspended in 40 μl of SDS-PAGE buffer. To compare the relative amount of surface biotinylated proteins, one-tenth of the input extract (40 μg) and precipitates (4 μl) were loaded for Western blotting analysis, as described above.


Generation of forebrain-specific, knock-out of the β1-integrin gene

To define the role of β1-integrin in adult CNS function, we generated a β1-integrin postnatal and spatial knock-out [f(β1)KO] by crossing a forebrain and excitatory neuron-specific Cre transgene, driven by the α-CaMKII promoter, into the genome of animals homozygous for a floxed β1-integrin allele [f(β1)]. Cre-induced deletion of exon 2, which was flanked by loxP sites, predicted early termination of translation (Fig. 1A) (Tsien et al., 1996; Raghavan et al., 2000). We reasoned that β1-integrin would be a good candidate for studying integrin function in the CNS, because it is the only known subunit partner for several α-integrins, including α3, α5, and α8. PCR analyses of genomic DNA isolated from the hippocampus of these and control animals confirmed the occurrence of Cre-dependent recombination at exon 2 in the f(β1)KO animals (Fig. 1 B). The presence of PCR products representing the nonrecombined floxed β1-integrin allele was also observed in the knock-outs because glial cells, which do not express the Cre protein, make up >30% of the cell population of the hippocampus (Geisert et al., 2002). β1-Integrin expression in the forebrain is extremely low and cannot be reliably detected using standard protocols for RNA in situ hybridization (Pinkstaff et al., 1998) (data not shown).

Figure 1
Cre-dependent recombination of the floxed β1-integrin allele. A, Three alleles of the β1-integrin gene are depicted. WT, Wild-type allele; Floxed, allele with loxP sequences flanking exon 2 in the same orientation; Recombined, allele with ...

We did not observe any alterations in CNS gross anatomy in mice deficient in β1-integrin (data not shown) (see Fig. 7A). To determine whether the postnatal loss of β1-integrins produces structural defects in synapses, we studied >700 axospinous and axodendritic synaptic contacts from the stratum radiatum of the hippocampal CA1 region using electron microscopy. No detectable morphological differences were observed comparing synapses of f(β1)KO [Cre/+;f(β1)/f(β1)] animals with Cre (Cre/+; +/+) control animals (Fig. 2A). We compared the distribution of synapse size, determined by the length of the postsynaptic density, as well as the density of synapses in a standardized area for animals 6 months of age for the two groups. There was no statistical difference between the two groups in either the density [Cre/+, 3.22 ± 0.33 synapses/10 μm2; f(β1)KO, 3.19 ± 0.29 synapses/10 μm2; p = 0.89, t test] or the length [Cre/+, 0.23 ± 0.003 μm; f(β1)KO, 0.22 ± 0.003 μm; p = 0.35, t test] of the synapses (Fig. 2B), indicating that synaptic structures remain intact in the f(β1)KO animals.

Figure 2
Normal synaptic ultrastructure in β1-integrin knock-outs. A, Representative electron micrographs of axospinous (top panels) and axodendritic (bottom panels) synaptic contacts in the hippocampal CA1 stratum radiatum region of Cre control (left ...
Figure 7
Expression of AMPA and NMDA receptor subunits in β1-integrin knock-outs. A, Immunohistochemistry was performed to compare the expression pattern of GluR1, GluR2/3, and NR1 subunits in the hippocampus of f(β1)KO and wild-type (WT) animals. ...

Impaired basal synaptic transmission in f(β1)KO mice

We began physiological studies on the roles of β1-integrin in the adult CNS by examining the basal synaptic transmission in the Shaffer collateral pathway of the hippocampus using field recordings of hippocampal slices (Fig. 3). The f(β1)KO slices showed a significant reduction in evoked field EPSPs (fEPSPs) across a range of stimulus intensities (Fig. 3A). In contrast, when measured in the same fEPSP trace, there was no difference in fiber volley amplitude in the mutant mice compared with the two parental controls genotypes, indicating an equivalent transfer of the presynaptic stimulus into an axonal depolarization (Fig. 3B). These results, combined as the slope of the fEPSP versus the fiber volley amplitude (Fig. 3C), clearly indicate that the loss of β1-integrin results in a significant decrease in the overall basal synaptic transmission in hippocampal slices across a broad range of stimulus intensities. Moreover, this reduced synaptic transmission appears to be the result of a decrease in the evoked postsynaptic fEPSP. For instance, a fiber volley produced with a stimulation of 20 mA in the control slices elicited an fEPSP with a slope of ~1.1 mV/ms, whereas the fEPSP elicited by the same fiber volley amplitude in the f(β1)KO slices was approximately one-half of this value. Because the fiber volley is one measure of neuron presynaptic function, these results are consistent with a deficit of the β1-integrin mutants in postsynaptic responses to equal pre-synaptic stimuli. Because the majority of the fast excitatory synaptic transmission under the recording conditions is AMPA receptor mediated, these results also suggest that AMPA receptor function may be compromised by the absence of β1-integrins. As an additional test of this idea, we also measured the fEPSPs in the mutant and control slices in the absence of Mg2+ to release the NMDA receptor block and in the presence of CNQX to block AMPA receptor potentials. Under these recording conditions, no significant differences were observed in overall synaptic transmission between the f(β1)KO and the control animals (Fig. 3D), suggesting that synaptically evoked NMDA receptor-mediated currents were normal in area CA1 of mutant mice. Overall, then, the most straightforward interpretation of our results is that the decrement in synaptic connectivity is attributable to a decrease in AMPA receptor function and not a decrease in synaptic glutamate release.

Figure 3
Impaired AMPA receptor synaptic transmission in β1-integrin knock-outs. A, The fEPSP plotted against the intensity of stimulus applied to the Schaffer collateral. Loss of β1-integrin had a significant effect on slope of the fEPSP (postsynaptic ...

Hippocampal synaptic plasticity in f(β1)KO mice

We next investigated the effects of β1-integrin knock-out on synaptic plasticity. PPF is a transient form of presynaptic-dependent short-term plasticity believed to be the result of enhanced probability of synaptic vesicle release. This phenomenon occurs with the facilitation of a closely timed second stimulus primarily as a result of residual calcium remaining from the first stimulation (Wu and Saggau, 1994). PPF in the f(β1)KO mice was normal (Fig. 4A), consistent with normal presynaptic function observed in the unaltered fiber volley amplitude at increasing stimulus intensity (Fig. 3B). In contrast, NMDA receptor-dependent LTP in the f(β1)KO mice was reduced to approximately one-half of the level found in the controls using two trains of high-frequency stimulation (HFS) (100 Hz) at a stimulus intensity that generated 50% of the maximum synaptic response (Fig. 4B). This reduction of LTP was not attributable to a deficit in the total amount of postsynaptic depolarization induced by HFS. Comparisons of the total depolarization envelope during successive 100 Hz/s stimulus trains showed no significant difference between f(β1)KO and control slices (Fig. 4C). This shows that the total conductance through all ion channels during HFS was similar between the mutant and control. In addition, the deficit in LTP remained in the f(β1)KOs even when saturating amounts of HFS (8 trains of 100 Hz stimulation) were used to generate the LTP at stimulus intensities that elicited equal fEPSPs (controls, 50%; (β1)KOs, 85% of max) (Fig. 4D). Together, these results indicate that f(β1)KOs have a deficit in NMDA receptor-dependent LTP. It remains unclear, however, whether this deficit is attributable to an impairment in the molecular machinery underlying hippocampal LTP induction or whether the deficit is a consequence of the impaired synaptic transmission we observed in our studies of baseline synaptic connectivity. Additional studies will be required to distinguish between these two possibilities. Nevertheless, these studies indicate that hippocampal synaptic plasticity deficits occur in β1-integrin-deficient animals.

Figure 4
Synaptic plasticity in β1-integrin knock-outs. A, PPF in f(β1)KOs [Cre; f(β1)/f(β1), n = 26] and controls [Cre; +/+, n = 24 and +; f(β1)/f(β1), n = 24]. The loss of β1-integrin expression had no ...

Hippocampal-dependent spatial learning and fear memory are normal in f(β1)KO Mice

Subsequently, we examined the memory capabilities of f(β1)KO mice using a battery of behavioral tests. We first measured the behavior of the f(β1)KO animals in rotarod and open-field tests and found no detectable difference in initial performance or improvement over multiple trials, indicating that the f(β1)KOs had normal locomotor and exploratory activities as well as normal motor learning (data not shown). Subsequently, we investigated spatial memory performance in the water maze using a training protocol consisting of four training trials per day. The f(β1)KO animals were indistinguishable from control animals for escape latency (Fig. 5A) or path lengths (data not shown) and improvement with training in this hippocampal-dependent task. Memory examined 24 h or 1 week after the last acquisition trial in a probe trial was identical for all three genotypes (Fig. 5B,C). Recently, certain mutants have been reported as deficient in a “pattern completion” version of the probe trial, in which all but one of the obvious spatial cues present during training were removed, but normal when all of the cues were present (Nakazawa et al., 2002). When tested under this more difficult condition, the f(β1)KOs still showed memory indistinguishable from the control animals (Fig. 5D).

Figure 5
β1-integrin knock-outs exhibited normal spatial memory and conditioned fear memory. A, Acquisition of spatial memory plotted as escape latency as a function of training day in the hidden version of the water maze. No significant difference was ...

We next examined the performance of f(β1)KOs along with controls in a pavlovian fear conditioning paradigm. Animals were presented with two tone–foot shock pairings inside a sound proof chamber. Memory of the context, a hippocampal-dependent memory, was measured by placing animals inside the same chamber in which they had received the training 2 or 24 h later and recording the percentage of time spent in a fearful posture (freezing). The f(β1))KOs and the control animals exhibited a high and equivalent level of freezing responses at both 2 and 24 h after training (Fig. 5E). To measure memory of the tone cue, which is hippocampal independent, we placed the animals after training into a chamber with distinct contextual cues and measured their fear response to the tone that was paired with shock during training. The percent time that the f(β1)KOs and the controls spent in a fearful posture before the tone onset during the memory test (pre-CS) was low, with no significant differences between groups (Fig. 5F). In addition, the percentage of time that all genotypes spent freezing during the tone presentation was high, with no significant differences between groups. Therefore, despite a deficit in basal hippocampal synaptic transmission and LTP, the f(β1)KO animals surprisingly have intact hippocampal-dependent spatial as well as contextual and cued fear memories.

Working memory is impaired in f(β1)KO mice

The impaired basal synaptic transmission in the hippocampus of the f(β1)KOs suggests that the function of synaptic AMPA receptors may be compromised in the absence of β1-integrins. Interestingly, knock-out mice of the GluR1 subunit of the AMPA receptors have been reported to exhibit normal spatial memory and deficient LTP, similar to the phenotypes exhibited by the β1-integrin animals discussed above (Zamanillo et al., 1999). Moreover, a recent study has shown that GluR1-deficient mice are impaired in working memory in a nonmatching-to-place T-maze test (Reisel et al., 2002). Because of the similar behavioral and physiological phenotypes of the GluR1 and the f(β1)KO animals, we tested the f(β1)KOs in the same, nonmatching-to-place T-maze test used previously for GluR1 KOs.

In this nonmatching-to-place T-maze assay, each trial consists of two runs: a sample run and a choice run (Fig. 6A). For the sample run, one of two target arms of the T-maze is randomly blocked so that the animal is forced to obtain a food reward from the unblocked target arm. For the choice run, both target arms are unblocked, and the animal is free to move into either arm but is rewarded only if the previously unsampled arm is chosen. After initiating training, all animals performed at the random (50%) level. The animals were then trained to a criterion of 75% correct choices over 16 trials in the choice run, with no delay between the two runs. After reaching criterion, they were tested with no delay between the sample and choice runs, or with delays imposed between the two runs. The animals must remember the arm chosen during the sample run over the delay to obtain the food reward and to be scored as having made a correct choice. We expressed working memory as the percentage of correct choices the animal makes during the choice runs.

Figure 6
β1-integrin knock-outs exhibited impaired working memory. A, Schematic diagram of the nonmatching-to-place T-maze assay. Each trial consists of a sample run and a choice run. For the sample run, one of two target arms of the maze is blocked, and ...

The f(β1)KO animals were able to learn and perform as well as the controls, reaching the criterion after 6 d of training like the controls. After reaching criterion, they also performed equivalently to the controls when the choice run was made with no delay after the sample run (Fig. 6B). However, the f(β1)KOs performed significantly poorer than the control animals with a 10 s delay. This performance was also significantly different from their performance with no delay. The controls, in contrast, showed no decrease in performance with a 10 s delay. When we extended the delay to 1 min, all groups performed near the random choice level, which was significantly lower than their performance with no delay, indicating the transient nature of this type of working memory. It is noteworthy that there was no significant difference between the performance of the f(β1)KOs with a 10 s and 1 min delay. These results, which were replicated in a second independent experiment with a different set of animals, along with those of Zamanillo et al. (1999), show that working memory is more labile in the absence of the β1-integrins and the GluR1 subunit of the AMPA receptor. They therefore support the hypothesis that β1-integrins regulate AMPA receptor functions for synaptic transmission, synaptic plasticity, and working memory.

Expression of AMPA and NMDA receptor subunits is normal in the f(β1)KOs

To investigate whether the expression of the AMPA receptors is altered by the loss of β1-integrins, we performed immunohistochemistry to study the distribution and abundance of GluR1 and GluR2/3 subunits in the f(β1)KO brains (Fig. 7A). We also looked at the distribution of the NR1 subunit of the NMDA receptor, the other major excitatory neurotransmitter receptor in the hippocampus. No detectable difference was observed in either the pattern or level of the immunohistochemical signal between the f(β1)KO and control animals for these receptor subunits. Results showing the immunohistochemical signals in the hippocampus are shown in Figure 7A, but no difference was observed throughout the brain. These results therefore suggest that β1-integrin KO does not affect the abundance or distribution of AMPA or NMDA receptor subunits.

Protein phosphorylation of AMPA receptors is a critical modification that influences receptor function. Because integrins signal by activating multiple protein kinases, it seemed possible that the loss of β1-integrin would influence the phosphorylation state of the receptor. In particular, phosphorylation at S831 and S845 of GluR1 has been shown to potentiate the ion channel function of the receptor (Roche et al., 1996; Barria et al., 1997; Derkach et al., 1999; Banke et al., 2000). Furthermore, mutations at these two sites produced LTP deficits and normal spatial reference memory (Lee et al., 2003). We therefore measured the level of phosphorylation of these two sites in the f(β1)KOs by immunoblotting hippocampal protein extracts and probing these with phosphorylation site-specific antibodies against S831 and S845. As shown in Figure 7, B and C, there was no detectable difference in signal intensity from the f(β1)KOs compared with the controls.

Finally, we examined the level of expression of GluR1 and its level of phosphorylation at S831 and S845 on the cell surface. To do this, we conducted surface biotinylation of hippocampal slices and measured the expression levels from the biotinylated fractions by immunoblotting. As shown in Figure 7, D and E, neither the level of total GluR1 nor the level of phosphorylation at S831 and S845 showed a significant difference between the f(β1)KO and the control animals. Together, these results suggest that the steady-state expression level of AMPA receptor subunits and the phosphorylation level of S831 and S845 are not regulated by β1-integrins.


Using a genetic approach to remove β1-integrin expression in the postnatal forebrain, we demonstrated that β1-integrins are required for three aspects of adult hippocampal function: basal synaptic transmission, synaptic plasticity, and working memory. Surprisingly, the f(β1)KOs did not have a deficit in hippocampal-dependent spatial or contextual memory.

In the absence of β1-integrins, fEPSPs were dramatically reduced at the CA3–CA1 synapses. This reduction is most consistent with a postsynaptic defect in AMPA receptor synaptic transmission because it was accompanied by normal fiber volley amplitude and normal PPF. In addition, the majority of the fast excitatory synaptic transmission is mediated by AMPA receptors. The large deficit in output responses (>50%) observed in the f(β1)KOs requires a change in AMPA receptor-mediated synaptic transmission. Furthermore, altered NMDA receptor function was not detected by measuring the NMDA receptor-dependent synaptic transmission in the presence of AMPA receptor blockers. These results are consistent with the observation reported recently that infusion of an integrin ligand into rat hippocampal slices causes a slow (minutes) increase in fast excitatory transmission mediated by AMPA receptors (Kramar et al., 2003). The combined physiological results therefore suggest that integrin activation causes an increase in AMPA receptor function, and integrin inactivation causes a decrease in AMPA receptor function.

How might β1-integrins regulate AMPA receptor functions? Our EM studies have shown that the basic synapse structure is not disrupted in the absence of β1-integrins. Western blotting and immunohistochemical results suggested that the loss of β1-integrin function does not lead to a global redistribution or loss of AMPA receptor subunits at steady state. However, it is possible that our assays do not have the resolution needed to detect subtle changes that might exist in a subset of synapses. It is also possible that such changes might occur only after integrin activation. Accumulating evidence has demonstrated that protein phosphorylation is critical in modulating AMPA receptor-mediated transmission by altering either channel properties (Benke et al., 1998; Banke et al., 2000; Poncer et al., 2002) or abundance at the synaptic surface through receptor trafficking (Shi et al., 1999; Hayashi et al., 2000; Lu et al., 2001). For example, phosphorylation at S831 of GluR1 by CaMKII and at S845 by PKA have been shown to potentiate the response of the receptor to glutamate (Roche et al., 1996; Barria et al., 1997; Derkach et al., 1999; Banke et al., 2000), and phosphorylation of S845 might also be important for the internalization and recycling of the receptors (Ehlers, 2000). Interestingly, activation of integrins has been shown to stimulate a variety of signaling molecules. One potential candidate for regulating AMPA receptors is the tyrosine kinase Src. In fact, direct activation of endogenous Src in rat CA1 hippocampal neurons has been reported to potentiate AMPA receptor currents through an increase in the receptor conductance (Lu et al., 1998). More recently, Kramar et al. (2003) have also shown in rat hippocampal slices that the enhancement of AMPA receptor potentials by integrin-activating peptides can be blocked by the Src inhibitor PP2. However, it remains unknown whether Src regulates AMPA receptors directly.

Recent studies have also shown that AMPA receptor functions are also regulated by receptor interaction with the cytoskeleton. Drugs that disrupt actin filaments reduce AMPA receptor-mediated synaptic transmission (Kim and Lisman, 1999), LTP (Kim and Lisman, 1999; Krucker et al., 2000), and the internalization of AMPA receptors (Zhou et al., 2001) in hippocampal neurons. Shen et al. (2000) have shown that disruption of actin filaments or the interaction between GluR1 and 4.1N, a cytoskeleton-associated protein, decreases the cell surface expression of GluR1 in heterologous cells, suggesting that 4.1N may link AMPA receptors to the actin cytoskeleton. The intimate association of integrins with actin and other cytoskeletal proteins may allow them to modulate the interaction of AMPA receptors with components of the cytoskeleton and thus the synaptic functions of the receptor.

Our results once again confirm a role for integrins in LTP at CA3–CA1 synapses. Because the expression of LTP requires AMPA receptors, our findings offer a possible integrin-dependent regulatory mechanism for AMPA receptor function in LTP. Furthermore, although some LTP was still detected in the f(β1)KOs, potentiation was completely abolished in the GluR1−/− mutants (Zamanillo et al., 1999), suggesting that β1-integrins probably have a modulatory rather than an on/off, switch role for LTP.

Our behavioral results have uncovered a hitherto unknown function for integrins and revealed a notable similarity between the f(β1)KO and GluR1−/− mice. Both mutants exhibit normal spatial reference memory in the water maze despite suffering dramatic deficits in hippocampal CA3–CA1 LTP. In fact, the same dissociation between LTP and spatial reference memory was also observed in animals heterozygous for the α3-integrin null allele or simultaneously for α3 and α8 null alleles (Chan et al., 2003). In addition, the β1-integrin knock-out and GluR1−/− animals were both impaired in the same nonmatching-to-place working memory task, which has also been shown to be hippocampus dependent (Olton et al., 1979; Deacon et al., 2002). One possible explanation is that LTP in the hippocampal CA1 area is not required for spatial reference memory but is critical for working memory. A distinct difference between these two tasks is the persistence requirement for memory. In the water maze, the mice are trained to locate a fixed platform location over a period of days. In contrast, working memory must be kept on line for tens of seconds in the working memory task. It is conceivable that structural plasticity, which involves changes in synaptic number and morphology, accounts for the formation of spatial reference memory, whereas short-term synaptic plasticity, without associated structural plasticity, underlies working memory. Another possibility is that a requirement of NMDA receptor/β1-integrin/AMPA receptor-dependent LTP can be substituted by other LTP mechanisms for spatial reference memory but not for working memory. Nevertheless, NMDA receptor/β1 integrin/AMPA receptor-dependent LTP is dispensable for spatial reference memory but not for working memory. It is interesting that, like the deficit in LTP, the magnitude of impairment in working memory also differs between the f(β1)KO and GluR1−/− mice. The GluR1−/− mice exhibit a complete loss of working memory even with no delay between the sample run and the choice run. The f(β1)KO mice, in contrast, showed a memory deficit only after a delay was imposed between the two runs. This may indicate that β1-integrins are involved in the stability rather than the formation of working memory. One model for the instability based on the AMPA receptor function is that working memory requires a certain level of AMPA receptor potentiation, which decays rapidly and thus is short lasting. In the absence of β1-integrins, a lower level of potentiation caused by the impaired AMPA receptor potentials results in a shorter time interval to fall below the threshold required for the expression of working memory.


This work was supported by National Institutes of Mental Health Grant MH60420 (R.L.D.) and Baylor College of Medicine Mental Retardation Research Center Grant HD24064. R.L.D. is the recipient of the R.P. Doherty-Welch Chair in Science at the Baylor College of Medicine. Members of the Davis laboratory thank Dr. J. N. P. Rawlins and Robert Deacon (University of Oxford, Oxford, UK) for significant assistance in establishing the T-maze working memory assay. We also thank Dr. Susumu Tonegawa (Massachusetts Institute of Technology, Cambridge, MA) for supplying the αCaMKII–Cre animals.


  • Bahr BA, Staubli U, Xiao P, Chun D, Ji ZX, Esteban ET, Lynch G. Arg-Gly-Asp-Ser-selective adhesion and the stabilization of long-term potentiation: pharmacological studies and the characterization of a candidate matrix receptor. J Neurosci. 1997;17:1320–1329. [PubMed]
  • Banke TG, Bowie D, Lee H, Huganir RL, Schousboe A, Traynelis SF. Control of GluR1 AMPA receptor function by cAMP-dependent protein kinase. J Neurosci. 2000;20:89–102. [PubMed]
  • Barria A, Derkach V, Soderling T. Identification of the Ca2+/calmodulin-dependent protein kinase II regulatory phosphorylation site in the alpha-amino-3-hydroxyl-5-methyl-4-isoxazole-propionate-type glutamate receptor. J Biol Chem. 1997;272:32727–32730. [PubMed]
  • Benke TA, Luthi A, Isaac JT, Collingridge GL. Modulation of AMPA receptor unitary conductance by synaptic activity. Nature. 1998;393:793–797. [PubMed]
  • Benson DL, Schnapp LM, Shapiro L, Huntley GW. Making memories stick: cell-adhesion molecules in synaptic plasticity. Trends Cell Biol. 2000;10:473–482. [PubMed]
  • Chan CS, Weeber EJ, Kurup S, Sweat JD, Davis RL. Integrin requirement for hippocampal synaptic plasticity and spatial memory. J Neurosci. 2003;23:7107–7116. [PubMed]
  • Chun D, Gall CM, Bi X, Lynch G. Evidence that integrins contribute to multiple stages in the consolidation of long term potentiation in rat hippocampus. Neuroscience. 2001;105:815–829. [PubMed]
  • Coppolino MG, Dedhar S. Bi-directional signal transduction by integrin receptors. Int J Biochem Cell Biol. 2000;32:171–188. [PubMed]
  • Deacon RM, Bannerman DM, Kirby BP, Croucher A, Rawlins JN. Effects of cytotoxic hippocampal lesions in mice on a cognitive test battery. Behav Brain Res. 2002;133:57–68. [PubMed]
  • Derkach V, Barria A, Soderling TR. Ca2+/calmodulin-kinase II enhances channel conductance of alpha-amino-3-hydroxy-5-methyl-4-isoxazolepropionate type glutamate receptors. Proc Natl Acad Sci USA. 1999;96:3269–3274. [PubMed]
  • Dityatev A, Schachner M. Extracellular matrix molecules and synaptic plasticity. Nat Rev Neurosci. 2003;4:456–468. [PubMed]
  • Ehlers MD. Reinsertion or degradation of AMPA receptors determined by activity-dependent endocytic sorting. Neuron. 2000;28:511–525. [PubMed]
  • Einheber S, Schnapp LM, Salzer JL, Cappiello ZB, Milner TA. Regional and ultrastructural distribution of the α8 integrin subunit in developing and adult rat brain suggests a role in synaptic function. J Comp Neurol. 1996;370:105–134. [PubMed]
  • Geisert EE, Jr, Williams RW, Geisert GR, Fan L, Asbury AM, Maecker HT, Deng J, Levy S. Increased brain size and glial cell number in CD81-null mice. J Comp Neurol. 2002;453:22–32. [PubMed]
  • Giancotti FG, Rouslahti E. Integrin signaling. Science. 1999;285:1028–1032. [PubMed]
  • Grotewiel MS, Beck CD, Wu KH, Zhu XR, Davis RL. Integrin-mediated short-term memory in Drosophila. Nature. 1998;391:455–460. [PubMed]
  • Hayashi Y, Shi SH, Esteban JA, Piccini A, Poncer JC, Malinow R. Driving AMPA receptors into synapses by LTP and CaMKII: requirement for GluR1 and PDZ domain interaction. Science. 2000;287:2262–2267. [PubMed]
  • Kim CH, Lisman JE. A role of actin filament in synaptic transmission and long-term potentiation. J Neurosci. 1999;19:4314–4324. [PubMed]
  • Kramar EA, Bernard JA, Gall CM, Lynch G. Alpha3 integrin receptors contribute to the consolidation of long-term potentiation. Neuroscience. 2002;110:29–39. [PubMed]
  • Kramar EA, Bernard JA, Gall CM, Lynch G. Integrins modulate fast excitatory transmission at hippocampal synapses. J Biol Chem. 2003;27:10722–10730. [PubMed]
  • Krucker T, Siggins GR, Halpain S. Dynamic actin filaments are required for stable long-term potentiation (LTP) in area CA1 of the hippocampus. Proc Natl Acad Sci USA. 2000;97:6856–6861. [PubMed]
  • Lee HK, Takamiya K, Han JS, Man H, Kim CH, Rumbaugh G, Yu S, Ding L, He C, Petralia RS, Wenthold RJ, Gallagher M, Huganir RL. Phosphorylation of the AMPA receptor GluR1 subunit is required for synaptic plasticity and retention of spatial memory. Cell. 2003;112:631–643. [PubMed]
  • Lu ML, Roder JC, Davidow J, Salter MW. Src activation in the induction of long-term potentiation in CA1 hippocampal neurons. Science. 1998;279:1363–1367. [PubMed]
  • Lu W, Man H, Ju W, Trimble WS, MacDonald JF, Wang YT. Activation of synaptic NMDA receptors induces membrane insertion of new AMPA receptors and LTP in cultured hippocampal neurons. Neuron. 2001;29:243–254. [PubMed]
  • Nakazawa K, Quirk MC, Chitwood RA, Watanabe M, Yeckel MF, Sun LD, Kato A, Carr CA, Johnston D, Wilson MA, Tonegawa S. Requirement for hippocampal CA3 NMDA receptors in associative memory recall. Science. 2002;297:211–218. [PMC free article] [PubMed]
  • Olton DS, Becker JT, Handelmann E. Hippocampus, space and memory. Behav Brain Sci. 1979;2:313–366.
  • Pinkstaff JK, Lynch G, Gall CM. Localization and seizure-regulation of integrin beta 1 mRNA in adult rat brain. Brain Res Mol Brain Res. 1998;55:265–276. [PubMed]
  • Pinkstaff JK, Detterich J, Lynch G, Gall C. Integrin subunit gene expression is regionally differentiated in adult brain. J Neurosci. 1999;19:1541–1556. [PubMed]
  • Poncer JC, Esteban JA, Malinow R. Multiple mechanisms for the potentiation of AMPA receptor-mediated transmission by α-Ca2+/calmodulin-dependent protein kinase II. J Neurosci. 2002;22:4406–4411. [PubMed]
  • Prokop A. Integrating bits and pieces: synapse structure and formation in Drosophila embryos. Cell Tissue Res. 1999;297:169–186. [PubMed]
  • Raghavan S, Bauer C, Mundschau G, Li Q, Fuchs E. Conditional ablation of beta1 integrin in skin. Severe defects in epidermal proliferation, basement membrane formation, and hair follicle invagination. J Cell Biol. 2000;150:1149–1160. [PMC free article] [PubMed]
  • Reisel D, Bannerman DM, Schmitt WB, Deacon RM, Flint J, Borchardt T, Seeburg PH, Rawlin JN. Spatial memory dissociations in mice lacking GluR1. Nat Neurosci. 2002;9:868–873. [PubMed]
  • Roche KW, O'Brien RJ, Mammen AL, Bernhardt J, Huganir RL. Characterization of multiple phosphorylation sites on the AMPA receptor GluR1 subunit. Neuron. 1996;16:1179–1188. [PubMed]
  • Schwartz MA. Integrin signaling revisited. Trends Cell Biol. 2001;12:466–470. [PubMed]
  • Shen L, Liang F, Walensky LD, Huganir RL. Regulation of AMPA receptor GluR1 subunit surface expression by a 4.1N-linked actin cytoskeletal association. J Neurosci. 2000;20:7932–7940. [PubMed]
  • Shi SH, Hayashi Y, Petralia RS, Zaman SH, Wenthold RJ, Svoboda K, Malinow R. Rapid spine delivery and redistribution of AMPA receptors after synaptic NMDA receptor activation. Science. 1999;284:1811–1816. [PubMed]
  • Staubli U, Chun D, Lynch G. Time-dependent reversal of long-term potentiation by an integrin antagonist. J Neurosci. 1998;18:3460–3469. [PubMed]
  • Tadokoro S, Shattil SJ, Eto K, Tai V, Liddington RC, de Pereda JM, Ginsberg MH, Calderwood DA. Talin binding to integrin beta tails: a final common step in integrin activation. Science. 2003;302:103–106. [PubMed]
  • Thomas-Crusells J, Vieira A, Saarma M, Rivera C. A novel method for monitoring surface membrane trafficking on hippocampal acute slice preparation. J Neurosci Methods. 2003;30:159–166. [PubMed]
  • Tsien JZ, Chen DF, Gerber D, Tom C, Mercer EH, Anderson DJ, Mayford M, Kandel ER, Tonegawa S. Subregion- and cell type-restricted gene knockout in mouse brain. Cell. 1996;87:1317–1326. [PubMed]
  • van der Flier A, Sonnenberg A. Function and interactions of integrins. Cell Tissue Res. 2001;305:285–298. [PubMed]
  • Wildering WC, Hermann PM, Bulloch AG. Rapid neuromodulatory actions of integrin ligands. J Neurosci. 2002;22:2419–2426. [PubMed]
  • Wu LG, Saggau P. Presynaptic calcium is increased during normal synaptic transmission and paired-pulse facilitation, but not in long-term potentiation in area CA1 of hippocampus. J Neurosci. 1994;14:645–654. [PubMed]
  • Yamagata M, Sanes JR, Weiner JA. Synaptic adhesion molecules. Curr Opin Cell Biol. 2003;15:621–632. [PubMed]
  • Zamanillo D, Sprengel R, Hvalby O, Jensen V, Burnashev N, Rozov A, Kaiser KM, Koster HJ, Borchardt T, Worley P, Lubke J, Frotscher M, Kelly PH, Sommer B, Andersen P, Seeburg PH, Sakmann B. Importance of AMPA receptors for hippocampal synaptic plasticity but not for spatial learning. Science. 1999;284:1805–1811. [PubMed]
  • Zhou Q, Xiao M, Nicoll RA. Contribution of cytoskeleton to the internalization of AMPA receptors. Proc Natl Acad Sci USA. 2001;98:1261–1266. [PubMed]