A role for calcium-permeable AMPA receptors in synaptic plasticity and learning

Abstract

A central concept in the field of learning and memory is that NMDARs are essential for synaptic plasticity and memory formation. Surprisingly then, multiple studies have found that behavioral experience can reduce or eliminate the contribution of these receptors to learning. The cellular mechanisms that mediate learning in the absence of NMDAR activation are currently unknown. To address this issue, we examined the contribution of Ca(2+)-permeable AMPARs to learning and plasticity in the hippocampus. Mutant mice were engineered with a conditional genetic deletion of GluR2 in the CA1 region of the hippocampus (GluR2-cKO mice). Electrophysiology experiments in these animals revealed a novel form of long-term potentiation (LTP) that was independent of NMDARs and mediated by GluR2-lacking Ca(2+)-permeable AMPARs. Behavioral analyses found that GluR2-cKO mice were impaired on multiple hippocampus-dependent learning tasks that required NMDAR activation. This suggests that AMPAR-mediated LTP interferes with NMDAR-dependent plasticity. In contrast, NMDAR-independent learning was normal in knockout mice and required the activation of Ca(2+)-permeable AMPARs. These results suggest that GluR2-lacking AMPARs play a functional and previously unidentified role in learning; they appear to mediate changes in synaptic strength that occur after plasticity has been established by NMDARs.

Figure 1. Conditional deletion of GluR2 in cKO mice.

In situ hybridization with a GluR2 probe (black) demonstrates marked loss of GluR2 from the CA1 region of the dorsal hippocampus and from cortex layer III, but not from other brain regions, in the GluR2-cKO mice. ( A–G ) Representative images of GluR2 mRNA expression from brains of 8-week-old animals as evidenced by the pattern of in situ hybridization of a GluR2 probe. Shown are: ( A ) Whole brain of control animal; ( B ) Whole brain of GluR2-cKO animal; ( C ) Dorsal hippocampus and overlying cortex of control animal; ( D ) Dorsal hippocampus and overlying cortex GluR2-cKO animal; ( E ) Ventral hippocampus and amygdala of control animal; ( F ) Ventral hippocampus and amygdala of GluR2-cKO animal; ( G ) Dorsal hippocampus and overlying cortex of GluR2 global KO animal; ( H ) Percent GluR2 positive cells in brains taken from mice at 6 and 8 weeks age in dorsal hippocampus CA1 region, dorsal CA3 region, dorsal dentate gyrus (DG), cortex layer III, and in ventral hippocampus CA1 region. ( C–G ) (i) Low power images (magnification = 4×), ( C–D ) (ii) high power image of cortex (magnification = 10×), ( C–D ) (iii) and ( E–F ) (ii) high power image of hippocampus CA1 region (magnification = 10×), ( E–F ) (iii) high power image of amygdala adjacent to ventral hippocampus (magnification = 10×). All images couterstained with Nissl stain (blue). A description of the approach used to generate the GluR2-cKO and Glur2-KO mice as well the molecular characterization of these mice is shown in Fig. S1.

Figure 2. GluR2-cKO mice exhibit loss of GluR2 protein in CA1 and unchanged neuronal numbers.

( A–I ) Tissue from GluR2-cKO, control, and GluR2 global KO animals were immunohistochemically labeled for GluR2 (red) and counterstained with DAPI (blue). Confocal micrographs were obtained of the DG, CA3 and CA1 regions in the hippocampus. Scale = 50 µm. ( J–K ) Tissue from GluR2-cKO and control animals were immunohistochemically labeled for the neuronal marker NeuN for stereological estimation of the neuronal population. ( L ) Quantification of neuronal populations in the DG, CA3 and CA1 regions using the Stereo Investigator. Values as mean ± s.e.m. Note that the knockout of GluR2 did not lead to changes in the expression of other glutamate receptor subunits (See Fig. S2 and Table S1).

Figure 3. LTP is enhanced in GluR2-cKO mice.

(A) HFS stimulation-induced LTP is enhanced in slices from GluR2-cKO mice. LTP was induced at time = 0 with two, 1 second long trains of 100 Hz stimulation (inter-train interval = 10 sec). (B) A 30 second long train of 5 Hz stimulation induces significantly larger LTP in slices from GluR2-cKO mice. 45 minutes post-5 Hz stimulation (delivered at time = 0) fEPSPs were potentiated to 168±9% of baseline in control slices (gray symbols, n = 5) and were potentiated to 198±9% of baseline in slices from GluR2-cKOs (black symbols, n = 5). (C) A short train of 5 Hz stimulation (5 sec) induces similar levels of LTP in control (gray symbols, n = 5) and GluR2-cKO slices (black symbols, n = 5). (D) LTD is normal in GluR2-cKO slices. LTD was induced using a 15 minute long train of 1 Hz stimulation (indicated by the bar). The magnitude of LTD seen 60 minutes after the start of 1 Hz stimulation is similar in wild type (gray symbols, n = 6) and GluR2-cKO slices (black symbols, n = 5). Baseline synaptic transmission was altered as predicted in the GluR2-cKO mice (see Fig. S3).

Figure 4. LTP can be induced in GluR2-cKO mice at both hyperpolarized and depolarized membrane potentials.

(A) A Hebbian LTP induction protocol (10 Hz presynaptic fiber stimulation paired with postsynaptic depolarization) induces LTP in cells from both control (gray circles, n = 7 cells) and GluR2-cKO slices (black triangles, n = 7 cells). (B) 10 Hz stimulation at resting membrane potential failed to induce LTP in pyramidal cells from both wild type (n = 14 cells from 4 mice) and GluR2-cKO mice (n = 11 cells from 3 mice). (C) A 10 Hz stimulation paired with postsynaptic hyperpolarization has no effect on synaptic transmission in control cells (open symbols, n = 12 cells) but induces LTP in cells from GluR2-cKO mice (black symbols, n = 11 cells). The traces shown in A, B, and C are superimposed EPSPs recorded during baseline and 30 minutes post-pairing in a control and GluR2-cKO cell. Scale bars are 20 milliseconds and 5 mV. (D) Histograms show the amount of potentiation present 30 minutes post-pairing in control (gray bars) and GluR2-cKO cells (black bars).

Figure 5. Deletion of GluR2 in CA1 impairs context fear.

(A) Controls and GluR2-cKO mice showed similar increases in context fear immediately after 1 (controls n = 13, GluR2-cKO n = 13) or 5 (controls n = 21, GluR2-cKO n = 12) shocks. (B) Controls showed more context fear than GluR2-cKO mice 24-hours after training. This difference was observed in both the 1 and 5 shock groups. Baseline freezing levels from the training session (prior to shock delivery) are shown for comparison. (C) To examine the time course of memory loss we tested separate groups of animals at three different time points after training. Memory was not reduced in GluR2-cKO mice immediately after training (controls n = 10, GluR2-cKO n = 12) but was impaired at 2 (controls n = 9, GluR2-cKO n = 14) and 24 (controls n = 21, GluR2-cKO n = 12) hours. (D) Excitotoxic hippocampus lesions made 1 day after training produced amnesia for context fear in both controls (sham n = 21, lesion n = 8) and GluR2-cKO mice (sham n = 15, lesion n = 13) (E) The ability to form a long-term memory of the context in the absence of shock was examined using a context pre-exposure (PE) procedure. PE produced robust context learning in control mice (n = 11) relative to non-exposed animals (n = 15). In contrast, pre-exposed GluR2-cKO mice (n = 12) showed less freezing than control mice and were not different from cKO mice not exposed to the context (n = 12) (F) Auditory fear conditioning produced equivalent freezing increases in controls (n = 15) and cKOs (n = 18) during the white noise relative to baseline. GluR2-cKO mice were impaired during the context test conducted the next day. Error bars represent ± SEM and * indicates statistical significance (p<.05).

Figure 6. Long-term spatial memory is impaired in GluR2-cKO mice.

( A ) Controls (n = 16) and GluR2-cKO (n = 12) mice traveled the same distance in the watermaze to the find the platform across training days. ( B ) Control mice spent more time in the target quadrant then the other quadrants during the watermaze probe test. In contrast, GluR2-cKO mice spent an equivalent amount of time in all quadrants during the probe test. ( C ) Controls (n = 13) made a higher percentage of correct responses than GluR2-cKO mice (n = 15) on the reference memory version of the radial arm maze. ( D ) Controls showed a reduction in the number of errors (re-entries) across days on the reference memory version of the radial arm maze while GluR2-cKO mice did not ( E ) The percentage of correct responses on the working memory version (win-shift) of the radial arm maze was the same in controls (n = 11) and GluR2-cKO mice (n = 11) ( F ) Controls and GluR2-cKO mice showed equivalent reductions in the number of errors made across days in the working memory version of the radial arm maze. Error bars represent ± SEM and * indicates statistical significance (p<.05).

Figure 7. Knockout of GluR2 in CA1 does not affect immediate early gene expression or seizure susceptibility.

(A) Confocal micrographs of cFOS-positive cells (green), detected by immunohistochemical staining, counterstained with nuclear marker DAPI (blue) in the CA1 of control and GluR2-cKO mice prior to training (‘no training’), following training (‘learning)’, and following kainate-induced seizures (KA) (n = 3 animals per group per genotype). Scale bar = 30 mm. (B) Quantitative analysis of cFOS-positive cells in the CA1 using stereology in control and GluR2-cKO mice that had no training or following training. Values shown as mean ± s.e.m. (C) Quantitative analysis of cFOS-positive cells in the CA1 using stereology in control and GluR2-cKO mice that had received a single i.p. injection of 25 mg/kg KA. Values shown as mean ± s.e.m. (D) Comparison of seizure susceptibility in control and GluR2-cKO animals. Seizure score is defined as the total seizure score divided by the number of observations and plotted as a cumulative frequency step graph – see experimental methods regarding seizure score scale. As the data were not normally distributed they were analyzed by Mann-Whitney. The Benjamini-Hochberg FDR correction was applied to correct for the multiple t-tests performed. No significant difference between curves was observed.

Figure 8. NMDAR and GluR2-lacking AMPAR make distinct contributions to learning.

(A) Experimental design. Animals were trained in context A on day 1 and context B on day 2. Mice were then tested for freezing in context B and context A on days 3 and 4 (B) The NMDAR antagonist CPP was given prior to training in context A in one group of mice and prior to context B in a second group of mice. CPP was more effective at blocking learning in context A than context B (saline n = 8, CPP n = 8). (C) Deletion of GluR2 was more effective at blocking learning in context A than context B (control n = 26, GluR2-cKO n = 23) (D) Learning in context A was impaired in GluR2-cKO mice and animals receiving CPP (controls saline n = 16, GluR2-cKO saline n = 7, controls CPP n = 21, GluR2-cKO CPP n = 13) while learning in context B was unaffected by either of these manipulations. (E) GluR2-cKO were trained in context A and then received injections of saline, CPP or CPP +IEM-1460 (a Ca²⁺-permeable AMPAR antagonist) before subsequent training in context B. Injections of CPP did not impair learning in context B while injections of CPP + IEM-1460 produced significant deficits (saline n = 10, CPP n = 10, CPP + IEM-1460 n = 11). (F) Wild-type mice were trained in context A and then received injections of saline, CPP or CPP +IEM-1460 before subsequent training in context B. Injections of CPP did not impair learning in context B while injections of CPP + IEM-1460 produced significant deficits (saline n = 8, CPP n = 8, CPP + IEM-1460 n = 8). Error bars represent ± SEM and * indicates statistical significance (p<.05).