Distinct Roles of the β1-Class Integrins at the Developing and the Mature Hippocampal Excitatory Synapse

Deletion of β1 integrin from the forebrain excitatory neurons using emx1–cre

The integrins are heterodimeric membrane proteins that consist of an α and a β subunit. They include two major classes, one with a common β1 subunit and the other with a common α_v subunit. In the nervous system, both classes of integrins are widely expressed during development and in the adult, in regions including the cerebral cortex and the hippocampus (Pinkstaff et al., 1998, 1999). To study the in vivo function of β1 integrins in the nervous system, we used a conditional allele of the β1 integrin gene in which the first coding exon was flanked by a pair of loxP sites (Graus-Porta et al., 2001). In previous experiments, this conditional allele was introduced into a cre line whose expression is driven by the neuronal enhancer element of the nestin gene promoter. This resulted in deletion of the β1 integrin gene throughout much of the nervous system and led to severe developmental defects in cerebral cortical lamination and cerebellar foliation (Graus-Porta et al., 2001; Blaess et al., 2004). To study more specifically the functions of β1-class integrins in excitatory neurons, we used another cre line whose expression is driven by the emx1 gene promoter by virtue of an internal ribosome entry site element engineered into the emx1 locus (Gorski et al., 2002). In contrast to the nestin promoter-regulated cre, the expression of cre in this strain is restricted to excitatory neurons and glial cells and their precursors derived from the dorsal forebrain neuroepithelium but is absent from the inhibitory interneurons, which originate from a different source in the ventral pallium. When crossed to a Rosa26 reporter line (Soriano, 1999), the emx1–cre induces expression of the lacZ reporter gene in excitatory neurons in both the cerebral cortex (data not shown) and the hippocampus (Fig. 1a–c), whereas no detectable level of β-galactosidase immunoreactivity was observed in inhibitory interneurons (Fig. 1d–f), confirming previous results (Gorski et al., 2002). This restriction of recombination to the excitatory neurons and the glia has allowed us to study the cell-autonomous requirement for β1 integrins in the development and function of excitatory synapses, particularly in the hippocampus.

Defective cortical but normal hippocampal lamination in β1 integrin/emx1–cre mouse

To obtain β1 integrin mutants, we made use of a null allele of β1 integrin gene (Stephens et al., 1995) and generated heterozygous animals with one null allele and one conditional allele. This ensures the efficiency of deleting the β1 integrin gene because only one allele needs to be recombined in each cell in the presence of the Cre recombinase. We confirmed the deletion of β1 integrin gene by examining its protein expression. We found that, at late embryonic stage, no β1 integrin immunostaining was detected in the radial glia or neurons in the developing cerebral cortex or hippocampus (supplemental Fig. 1, available at www.jneurosci.org as supplemental material). Western blot analysis further confirmed the loss of β1 integrin protein from the cerebral cortex and hippocampus (Fig. 1g). The emx1–cre mutants were born at a normal Mendelian ratio but showed decreased postnatal viability, presumably as a result of seizures caused by cortical lamination defects (see below). Nonetheless, a significant number of these animals survived to adulthood, permitting both anatomical and electrophysiological analyses at the adult stage.

Nissl staining of the cerebral cortices of these mutant animals indicated that there were severe neuronal layering defects. In control animals, the cortex is normally organized into distinct layers with the cell-sparse layer I positioned closest to the pial surface (Fig. 2a,c). In mutant animals, however, no distinct cortical layers were obvious except for layer I, whose border with the other cortical layers was also severely distorted (Fig. 2b,d). Neuronal birthdating experiments further revealed that, in some areas, neurons normally destined for the deeper layers had apparently invaded the superficial layers, whereas in other areas neurons normally destined for the superficial layers were trapped in the deeper layers (Fig. 3a–d). These lamination defects were also confirmed by staining for layer specific markers such as ER81 and Otx-1 (Fig. 3, e,f and g,h, respectively).

In contrast to the cerebral cortex, we observed a relatively normal organization of the neuronal layer in the hippocampal formation. In both the CA1 and CA3 regions of the mutants, the pyramidal neuron cell bodies were tightly packed in a single layer in their normal location (Fig. 2, compare f, e), although occasional defects in the organization of granule cells in the dentate gyrus were observed (data not shown), as has been reported after deletion with nestin–cre (Forster et al., 2002). This suggests that the β1 integrins are not essential for the migration of pyramidal neurons to their normal location in the hippocampus. In addition, because the dendritic fields of the CA1 and CA3 neurons also appeared to be of relatively normal size (Figs. 2e,f, 4) and their thickness was comparable between the control and the mutant (p > 0.7), this suggests that the overall differentiation of these neurons was not compromised. This thus allowed us to investigate the potential role(s) of the β1 integrins in the differentiation and function of the hippocampal pyramidal neurons in a relatively normal in vivo cellular environment.

Normal overall differentiation of dendrites and synapses in β1 integrin/emx1–cre hippocampus

In developing hippocampal neurons in culture, the β1 integrins are localized to the soma as well as the processes of the cells (supplemental Fig. 2, available at www.jneurosci.org as supplemental material). To investigate whether the β1 integrins play a role in the differentiation of hippocampal neuronal processes in vivo, we first stained for the expression of the dendritic marker MAP2. We found that the pyramidal neurons in the CA1 region of the mutant hippocampus had relatively normal dendritic architecture (Fig. 4a,b). The mutant dendrites extended normally into the dendritic field and had similar intensity of MAP2 staining compared with those from the control. Similar patterns were also observed in the CA3 region of the hippocampus (data not shown). Together with the relatively normal size of the dendritic fields (Fig. 2g), these results indicate that the overall differentiation of pyramidal neuron dendrites was not affected by the loss of β1 integrins. To further examine the dendritic morphology of these neurons, we performed Golgi staining on these brains. Consistent with the observations on MAP2-stained neurons, we found that the impregnated mutant CA1 pyramidal neurons also had relatively normal morphologies with well developed apical and basal dendritic branches (Fig. 4c,d). Closer examination revealed that, not only did the dendrites have relatively normal morphologies (Fig. 4e–h, top), they were also decorated by regularly spaced dendritic spines at an apparently normal density (Fig. 4e–h, bottom). Thus, the β1 integrins do not seem to be required for the normal overall differentiation of pyramidal neuron dendrites and spines in the hippocampus.

To determine whether the β1 integrins regulate synaptic differentiation in vivo, we next examined the expression pattern of a number of synaptic markers in the hippocampus. We first examined the expression of synaptotagmin, a marker of all presynaptic terminals, in the hippocampus by immunohistochemistry. We observed a relatively normal pattern and intensity of synaptotagmin staining in the CA1 region of the mutant hippocampus (Fig. 5a,b). Similar results were also observed in the CA3 region of the hippocampus (data not shown). Staining of synaptophysin, another protein present in all presynaptic terminals, also revealed similar results. At high resolution, the synaptophysin-positive puncta showed similar densities and intensities in the CA1 dendritic fields of the control and mutant animals (Fig. 5c,d). To further investigate synaptic differentiation, we next used antibodies specific to the postsynaptic compartments of excitatory synapses. We found that the mutant hippocampus had a relatively normal density and intensity of puncta positive for the AMPA subclass of glutamate receptor subunit GluR1 (Fig. 5e,f). Similar results were also obtained for the NMDA subclass of glutamate receptor subunit NR2A (data not shown). To investigate the differentiation of inhibitory synapses, we examined the staining of GAD65 in the axonal terminals of inhibitory neurons and observed relatively normal pattern of expression in the mutants (Fig. 5g,h). Thus, these results suggest that the overall differentiation of excitatory synapses in the hippocampus was not perturbed by the loss of β1 integrins in excitatory neurons and glial cells.

Impairment of LTP but not basal transmission in β1 integrin/emx1–cre hippocampus

To determine whether the loss of β1-class integrins from excitatory neurons during early development affects the function of adult hippocampal synapses, we examined the electrophysiological properties of the Schaffer collateral–CA1 synapses in hippocampal slices derived from emx1–cre-targeted animals. Recording of fEPSPs from hippocampal slices indicated that basal synaptic transmission, as reflected by the input–output curve, was normal at the mutant synapses (Fig. 6a). No significant difference was observed in fEPSP slopes induced by a wide range of stimulation intensities (Kolmogorov–Smirnov test, p > 0.5). Thus, the overall efficacy at CA1 synapses was similar between the mutant and the control.

We next examined the role of β1 integrins in long-term synaptic plasticity. emx1–cre-targeted mutants exhibited significant defects in LTP (Fig. 6b). In wild-type slices, tetanic stimulation (two times for 100 Hz, 1 s) induced robust LTP that lasted for at least 60 min. The fEPSP slope was 36.9 ± 7.4% above baseline level 60 min after the tetanus. In mutant slices, the same stimulation protocol induced synaptic potentiation that declined after the initial 10 min. The fEPSP slope at the end of the experiments was 11.9 ± 10.6% above baseline, significantly smaller than in the wild-type control (p < 0.01). These results are consistent with previous data implicating RGD-binding integrins in the expression of LTP (Bahr et al., 1997; Staubli et al., 1998; Kramar et al., 2003; Nagy et al., 2006). They are also consistent with studies that implicate several β1-binding α-class integrins in hippocampal LTP (Chun et al., 2001; Kramar et al., 2002; Chan et al., 2003). This suggests that the β1 integrins may be a major class of integrins involved in LTP at the hippocampal synapses. Because emx1–cre induces β1 integrin deletion in excitatory but not inhibitory neurons, our results further argue that the β1 integrins are most likely required in the excitatory neurons for mediating their function in LTP, although they do not exclude the possibility that the presence of β1 integrin may also be required in glial cells.

Presynaptic and postsynaptic changes as consequences of β1 integrin deletion

The LTP deficits we observed in the β1 integrin mutants could be attributable to impairments in presynaptic neurotransmitter release, postsynaptic responsiveness, or both. To probe presynaptic function, we first analyzed PPF, measured as the ratio of fEPSP slopes in response to two successive stimulation pulses at various IPIs. Although no change was observed in PPF at most IPIs, the ratio of fEPSP slopes increased when the IPI became very short (10 ms; p < 0.05) (Fig. 6c). These results raise the possibility that the β1 integrins may alter synaptic response to HFS.

To test this idea directly, we recorded the responses of the CA1 synapses to a brief train of HFS (100 Hz, 40 pulses) in the presence of the NMDA receptor antagonist APV (100 μm). After an initial increase, the fEPSP slopes exhibited a continuous decline over time in wild-type slices (Fig. 6d). Consistent with the increase in PPF at the IPI of 10 ms, the response to the second stimulus in the train was increased in the mutant slices. To better illustrate the differences in the decay phase of HFS response between wild-type and mutant slices, we plotted all of the data normalized to the response to the second stimulus (Fig. 6d). The fEPSP slope recorded from the mutant synapses declined more rapidly than that from the control synapses (Kolmogorov–Smirnov test, p < 0.05). The rate constant (τ), obtained by fitting the decay phase of the plots with an exponential equation, was 20.0 ± 3.9 stimuli for the control but 10.4 ± 1.2 stimuli for the mutant synapses (p < 0.02). To exclude the possibility that repetitive stimulation may recruit inhibition in the mutant slices, we performed the same experiment in the presence of the GABA_A antagonist bicuculline (10 μm). The CA3 regions of the hippocampal slices were removed to prevent excessive excitation and afterdischarges induced by bicuculline. Under these conditions, synaptic responses to HFS were still impaired in β1 integrin mutants (supplemental Fig. 3a, available at www.jneurosci.org as supplemental material). The brief HFS is effective in depleting the readily releasable pool (RRP) of synaptic vesicles but is too fast to affect the reserve pool of vesicles. Thus, a faster decline of response to HFS implies a smaller RRP at mutant synapses.

The increase in PPF at a short IPI observed at β1 integrin mutant synapses could be explained by a decrease in probability of transmitter release (P_r). To directly investigate potential changes in P_r, we performed whole-cell recording of NMDA currents in the presence of MK-801 [(+)-5-methyl-10,11-dihydro-5H-dibenzo [a,d] cyclohepten-5,10-imine maleate] (5 μm), an irreversible, open channel blocker for NMDA receptor. The NMDA currents were pharmacologically isolated by blocking AMPA and GABA_A receptors with CNQX (20 μm) and bicuculline, respectively. The currents were monitored at a holding potential of −40 mV once every 10 s, with the amplitude set to ∼100 pA. After a stable baseline was obtained, stimulation was paused for 5 min when MK-801 was perfused into the slices. When stimulation was resumed in the presence of MK-801, the amplitude of NMDA currents progressively declined (Fig. 7a,b). Scaling of the current at the end of experiment (Fig. 7a, #3) superimposed with that at the beginning (Fig. 7a, #2), suggesting that the effective concentration of MK-801 remained constant during the course of the experiments (Fig. 7c). The time constant of the progressive decline of NMDA currents is inversely related to the P_r (Hessler et al., 1993; Rosenmund et al., 1993). The decay of the averaged responses appeared significantly slower in the mutant compared with wild-type slices (Fig. 7a,b), suggesting a lower P_r in the mutant. A double-exponential fit of the decline revealed a fast and a slow component. The fast decay constant was significantly slower in β1 integrin mutants compared with wild-type animals (21.9 ± 2.5 vs 7.8 ± 0.9 stimuli; p < 0.01), whereas the slow time constant was not statistically different between genotypes (46.2 ± 5.06 vs 52.6 ± 5.7 stimuli; p > 0.1) (Fig. 7d). An alternative measure is the number of stimuli to attain 50% block of the initial EPSC amplitude. This number for the mutants was significantly higher, with a mean of 21.6 ± 3.8 stimuli compared with that of the wild types, which had a mean of 10.1 ± 1.7 stimuli (p < 0.05) (Fig. 7d). Collectively, these data suggest that the P_r is reduced in mice lacking β1 integrin.

Synaptic vesicles in the nerve terminals undergo a cycling process in which the depleted docked vesicles are replenished by the reserved pool of vesicles and/or through endocytosis (Dobrunz and Stevens, 1997; Pyle et al., 2000; Richards et al., 2000). When a prolonged, LFS is applied, the depletion is usually faster than the replenishment, leading to a gradual decline of EPSP slopes, or synaptic depression (Geppert et al., 1994, 1997). To investigate potential effects of β1 integrin deletion on the reserved pool of synaptic vesicles, we examined the synaptic responses to LFS (14 Hz), again in the presence of APV (Fig. 6e). No significant difference was observed initially, but, as the stimulus number further increased, the fEPSP slope of the mutant synapses decreased more rapidly (Fig. 6e). Statistical analysis revealed a significantly faster decline of synaptic responses to LFS in the mutant slices (Kolmogorov–Smirnov test, p < 0.05), suggesting a smaller size of the reserved vesicle pool and/or defects in its mobilization in neurons lacking integrin β1. The same experiment was repeated in the presence of bicuculline, and the mutant slices still showed an impaired response compared with the wild-type slices (supplemental Fig. 3b, available at www.jneurosci.org as supplemental material). Thus, the differences in LFS responses observed between genotypes are unlikely to be attributable to the differential recruitment of the inhibitory network after repetitive stimulation.

The deficits in LFS response raised the possibility that the impairment in LTP maintenance may be attributable to accumulating vesicular depletion when LTP was monitored by stimulation at 0.033 Hz. We addressed this by performing experiments in which stimulation was interrupted after LTP induction. After tetanic stimulation, the resulting LTP was measured for a period of 10 min, after which, monitoring stimulation was stopped for 10 min. After resuming stimulation, the field responses remained potentiated in wild-type slices with a mean fEPSP slope of 150 ± 12% (supplemental Fig. 3c, available at www.jneurosci.org as supplemental material). In contrast, the responses recorded from mutant slices decayed back to baseline to a mean value of 107 ± 10%, indicating that the failure in LTP maintenance at β1 integrin-deficient synapses was not the result of vesicle depletion.

To determine whether deletion of integrin β1 could also elicit postsynaptic changes, we measured the input–output curves of NMDA receptor-mediated field EPSPs. Slices were incubated in Mg²⁺-free ACSF containing CNQX (10 μm) to block AMPA receptors. We found that the NMDA fEPSP slopes of the mutant synapses were consistently lower than those of the control synapses over a range of stimulation intensities (Fig. 6f). Statistical analysis of the two plots indicated that NMDA receptor activity at the mutant synapses is significantly lower compared with the wild-type synapses (Kolmogorov–Smirnov test, p < 0.01). These results are consistent with recent reports in which integrin ligation was found to increase the amplitudes of NMDA receptor-mediated synaptic currents in adult hippocampal slices (Lin et al., 2003; Bernard-Trifilo et al., 2005). The impairment in NMDA receptor function could contribute to the LTP deficits we observed in integrin β1 mutant slices. Together, our results show that the loss of the β1-class integrins during embryonic hippocampal development impacts both the presynaptic and postsynaptic properties of the adult hippocampal excitatory synapses.

Distinct synaptic deficits in β1 integrin/CaMKIIα–cre hippocampus

The presynaptic defects we observed in the emx1–cre mutants suggest that absence of the β1 integrins during synaptic development results in synaptic deficits that persist into adulthood. Conversely, pharmacological studies have implied that the RGD-binding integrins acutely regulate LTP at adult hippocampal synapses. This raises the possibility that the β1 integrins may play different roles during development and in the adult. To test this, we made use of another cre line, CaMKIIα–cre, which induces recombination in the hippocampal pyramidal neurons at a later stage than emx1–cre, at ∼2–4 weeks postnatally (Tsien et al., 1996; Xu et al., 2000; Sonner et al., 2005). When this cre line was used to delete the β1 integrin gene, we found that, as expected, it did not lead to any neuronal lamination defects in either the cerebral cortex (data not shown) or the hippocampal formation (Fig. 8a,b). This is consistent with the fact the cre gene expression in this line comes on postnatally after neuronal migration is completed in both the cerebral cortex and the hippocampus. The dendritic fields of the CA1 and CA3 regions in the CaMKIIα–cre-targeted hippocampus were also of relatively normal size, suggesting normal overall dendritic differentiation. Indeed, staining for the dendritic marker MAP2 showed that the mutant CA1 pyramidal neurons had dendritic processes with similar morphology and MAP2 immunoreactivity to those of control animals (Fig. 8c,d). In addition, the general pattern of staining in the hippocampus for the synaptic marker synaptotagmin was also relatively normal (Fig. 8e,f). Thus, consistent with our results using emx1–cre, the β1 integrins do not seem to be required for the overall dendritic and synaptic differentiation by the CA1 and CA3 pyramidal neurons.

To determine whether the loss of β1-class integrins from mature synapses affects their synaptic physiology, we next performed recording on the Schaffer collateral–CA1 synapses using hippocampal slices from the CaMKIIα–cre animals. We found that, as expected, the basal transmission at these synapses was relatively normal (data not shown). Moreover, the mutant synapses showed similar PPFs over a range of interpulse intervals as the wild-type synapses (ANOVA test, p > 0.05 in all points), indicating a normal sensitivity to calcium and normal P_r (Fig. 9b). Similar to what was observed in the emx1–cre mutants, the synapses of the CaMKIIα–cre mutants also exhibited a significantly faster decline of synaptic response to LFS (Kolmogorov–Smirnov test, p < 0.01) (compare Figs. 9d, 6e). This suggests that the mobilization of the reserved pool of vesicles requires the continuous presence of β1 integrins at the mature hippocampal synapses. However, unlike those in the emx1–cre mutants, the synapses in the CaMKIIα–cre mutants displayed completely normal response to HFS (Kolmogorov–Smirnov test, p > 0.05) (Fig. 9c), suggesting that the size of RRP and the release of vesicles at these synapses were normal. These results suggest that the continuous presence of β1 integrins at the hippocampal synapses is essentially dispensable for the maintenance of their normal presynaptic properties, particularly for the regulated release of RRP vesicles, once the synapses have matured. The defective presynaptic response to HFS we observed in the emx1–cre mutants was thus likely attributable to the loss of the β1-class integrins during synaptic development.

In addition to deficits in their response to LFS, the CaMKIIα–cre mutant synapses exhibited deficits in LTP very similar to those observed in the emx1–cre mutants (Fig. 9a). The fEPSP slope 60 min after the tetanus was 52.1 ± 8.0% for wild-type but 7.4 ± 7.3% for CaMKIIα–cre slices (p < 0.01). Because the β1 integrin gene is deleted in the CaMKIIα–cre mutants only after postnatal week 2–4, these results indicate that the continuous presence of β1-class integrins at the mature hippocampal synapses is required for normal LTP. Thus, the β1 integrins appear to play an active signaling role in the process of LTP expression at the mature hippocampal synapse.

Integrin function in neurite outgrowth and synaptic differentiation

Through binding the ECM, the integrins provide cells with both a biochemical and a mechanical link to the ECM (Hynes, 2002). This unique property has been postulated to serve a key role in the processes of cell and growth cone migration. Indeed, in Drosophila and Caenorhabditis elegans, integrin mutants exhibit axon pathfinding or fasciculation defects (Baum and Garriga, 1997; Hoang and Chiba, 1998). In vertebrates, in vitro observations have long argued for a role of the integrins in promoting neurite outgrowth on ECM substrates (Tomaselli et al., 1993; DeFreitas et al., 1995; Varnum-Finney et al., 1995; Weaver et al., 1995). They may also mediate the responses of axon growth cones to certain guidance cues such as semaphorin 7A (Pasterkamp et al., 2003). However, data for such roles in vivo have been mostly lacking. Instead, strong evidence argues for a role of integrins in neuronal migration. For example, the α3 and α_v integrin receptors seem to regulate distinct aspects of neuronal migration during corticogenesis (Anton et al., 1999; Dulabon et al., 2000; Sanada et al., 2004; Schmid et al., 2004). The β1 integrins are also critical for lamination and foliation of the cerebral and cerebellar cortices (Graus-Porta et al., 2001). We observed similar phenotypes in the layering of the cerebral cortex and the dentate gyrus using emx1–cre-targeted deletion of β1 integrin. Thus, although the in vivo role of integrins in vertebrate axonal guidance remains to be established, these molecules are clearly important for neuronal migration.

The relatively normal cellular organization in the hippocampus of β1 integrin/emx1–cre mutants provides a unique opportunity to investigate its function in synaptic differentiation. Studies in this area have so far been focused primarily on the peripheral NMJ. At the Drosophila NMJ, the integrins have been found to regulate the number of synaptic boutons but play a minimal role in regulating synaptic ultrastructure or miniature EPSP (Beumer et al., 1999, 2002; Rohrbough et al., 2000). At the mouse NMJ, β1 integrin expression in the skeletal muscle but not in motor neurons is required for the proper terminal growth of the axons (Schwander et al., 2004). In the hippocampus, we observed that β1 integrin-deficient pyramidal neurons appeared to express normal overall densities of dendritic spines along the apical and basal dendrites. The overall densities of both excitatory and inhibitory synapses also appear normal. In addition, the basal transmission at the Schaffer collateral–CA1 synapses appears unaffected in β1 integrin/emx1–cre mice. In an independent study using β1 integrin/CaMKIIα–cre mice, however, basal synaptic transmission was found to be reduced, particularly at low-stimulation strengths (Chan et al., 2006). The reason for this discrepancy remains unclear. Regardless, β1 integrins do not seem to regulate overall synaptic density in the hippocampus, nor do they seem to be required for the general differentiation of the synapses.

Integrin function in the developmental maturation of presynaptic physiology

By taking advantage of the temporal difference in cre expression in the emx1–cre and CaMKIIα–cre lines, we identified a specific role of β1 integrins in presynaptic physiology and pinpointed their likely requirement to a time window during early development. We found that two forms of short-term synaptic plasticity, PPF at a short interpulse interval and synaptic response to HFS, are impaired in the emx1–cre mutants but not in the CaMKIIα–cre mutants. Because the only difference between these two cre lines in inducing recombination in pyramidal neurons is a temporal delay for CaMKIIα–cre, our results point strongly to a requirement for β1 integrins at an early stage of development in the proper presynaptic function of adult synapses, although we cannot rule out the possibility that they may act in glial cells of the adult hippocampus to regulate this process.

Two populations of synaptic vesicles have been described at the nerve terminal: the RRP and the reserved pool (Sudhof, 2000). Of them, only the RRP vesicles, equivalent to those morphologically docked at presynaptic membranes, are immediately available for release (Schikorski and Stevens, 1997; Murthy and Stevens, 1999). A brief HFS depletes most vesicles in the RRP but essentially spares the reserved pool and is therefore a good measure of the RRP size. We show that HFS induces a faster decline of fEPSP slopes in the emx1–cre mutant synapses, suggesting that expression of β1 integrins in early development is critical for controlling the number of docked vesicles at adult hippocampal synapses. Additional work using quantitative electron microscopy should provide additional evidence for this proposal.

The increase in PPF at a short IPI at the CA1 synapses of emx1–cre mutants reflects a decrease in P_r. This is supported by our finding using progressive, activity-dependent blockade of the NMDA receptor, a reliable assay for P_r. When NMDA currents are monitored over time in the presence of the open channel blocker MK-801, the amplitudes of the EPSCs gradually decrease at a rate dependent on the P_r. We observed that the mutant synapses exhibited a significantly slower decay in this assay, suggesting that deletion of β1 integrin decreases the P_r. In marked contrast, a previous report showed that application of RGD peptides or an antibody against β3 (but not β1) integrin to cultured hippocampal neurons resulted in an increase in the P_r, measured by the same method. These different observations may reflect a functional difference in the roles of β1 and β3 integrins. Alternatively, they may be attributable to differences between the in vitro cultures analyzed in the previous report and the genetically modified slices used in our study. It is interesting to note that RGD peptides attenuate the increase in transmitter release induced by mechanical stretch or hypertonic solution at the NMJ, suggesting that blockade of RGD-binding integrin function reduces the P_r at the NMJ (Chen and Grinnell, 1995; Kashani et al., 2001).

The alterations in the release kinetics of the RRP vesicles at the β1 integrin-deficient synapses is reminiscent of a recent finding that the level of phosphatidylinositol-4,5-biphosphate at presynaptic sites impacts synaptic vesicle trafficking and the size of RRP (Di Paolo et al., 2004). Interestingly, talin, an integrin cytodomain binding protein, has been found to regulate the activity of phosphatidylinositol kinase type Iγ as well as presynaptic vesicle trafficking (Di Paolo et al., 2002; Ling et al., 2002; Morgan et al., 2004). It would be of interest to explore in the future whether the integrins may be involved in this pathway, perhaps in the recruitment of talin and phosphatidylinositol kinase type Iγ to the synapse.

Integrin function at mature synapses and in learning and memory

At the Drosophila NMJ, the integrins are involved in regulating synaptic transmission and plasticity (Rohrbough et al., 2000). In the mammalian hippocampus, an acute involvement of the RGD-binding integrins in LTP maintenance has also been observed (Bahr et al., 1997; Staubli et al., 1998; Kramar et al., 2003). Several β1-binding α-class integrins, including α3, α5, and α8, have also been implicated in the induction and/or consolidation of hippocampal LTP (Chun et al., 2001; Kramar et al., 2002; Chan et al., 2003). We found that genetic deletion of the β1-class integrins from the excitatory pyramidal neurons using two different cre lines resulted in similar LTP defects both closely resembling those observed in previous pharmacological studies. Because β1 integrin is the known exclusive binding partner for α3, α5, and α8 integrin, the RGD-binding α3β1, α5β1, and α8β1 integrins may thus be the major class of integrins involved in mediating signal transduction at the hippocampal synapses. Because the postnatal deletion of β1 integrin in the CaMKIIα–cre mutants resulted in very similar phenotypes to those after embryonic deletion using emx1–cre, our results also point to a specific postnatal involvement of β1 integrin in LTP. Consistent with our finding, a recent study reported that postnatal deletion of β1 integrin using CaMKIIα–cre resulted in the impairment of LTP (Chan et al., 2006). Interestingly, these mice also exhibit normal PPF and normal response to HFS. These results, together with our finding that PPF and HFS responses are impaired in the emx1–cre mutants, suggest that early expression of β1 integrin is required for some (P_r, vesicle docking) but not other (LTP) functions of the adult synapses.

LTP has long been regarded as a plausible cellular substrate of learning and memory. The requirement for β1 integrins in LTP suggests that they may also be involved in the processes of learning and memory. Consistent with this, mice with reduced expression of the α3, α5, and α8 integrin subunits have been found to display defects in spatial memory (Chan et al., 2003). Expression of the integrin-associated protein gene has also been found to be upregulated during memory formation (Huang et al., 1998), suggesting an orchestrated response of integrin-related pathways during learning and memory. Interestingly, mice with postnatal deletion of β1 integrins exhibit normal hippocampal-dependent spatial and contextual memory but are impaired in working memory (Chan et al., 2006). Thus, future studies that elucidate the signaling cascade triggered by integrin engagement at the synapses may not only clarify the specific role of β1 integrins during LTP but also provide new insight into the processes of learning and memory.