Experiments were performed according to the institutional guidelines of the Max Planck Society and of the animal core facility (IBF) of the Heidelberg University. These guidelines adhere to the German Animal Welfare Act: Regulation for the Protection of Animals Used for Experimental or Other Scientific Purposes (Animal Welfare Regulation Governing Experimental Animals (TierSchVersV). Animal numbers for molecular and histological experiments were recorded under the protocol MPI/T-6/06; 15/08; 20/; 28/11. Genetic manipulations and behavioral experiments were licensed by the Regional Council in Karlsruhe, Germany (35-9185.81/G-4/02 and 35-9185.81/G-71/10). Efforts were made to minimize the number of animals used.

For the generation of Gria1^−/−/2^ΔFb, Gria1^−/−/2^QFb, and Gria1^−/−/Tg8.1 mice the following gene-targeted and transgenic mouse lines were used as founder lines:

Mice were genotyped by tail-PCR with specific primers. Indicated below are the names of primers, primer sequences, and the approximate lengths of the amplified gene fragments.

Coronal 70–100 μm thick vibratome sections were analyzed using different primary antibodies as described (Shimshek et al., 2005, 2006). Anti-Cre recombinase (1:3,000, polyclonal, gift from G. Schuetz; licensed from Covance, RRID:AB_11220031), anti-GluA1 (1:600, polyclonal, RRID:AB_390157) and anti-GluA2 (1:50, polyclonal, Millipore, RRID:AB_2336198) in combination with secondary anti-mouse (RRID:AB_2336176) and anti-rabbit (RRID:AB_2313567) antibodies coupled to horseradish-peroxidase (Vector Laboratories, each 1:600) or with biotinylated secondary antibodies (Vector Laboratories, 1:600, RRID:AB_2313581; RRID:AB_2313606) and ABC-peroxidase kit (Vector Laboratories, RRID:AB_2336827) were used.

Mouse brains were removed and both hippocampi were isolated. Total protein was prepared and immunoblots were performed as described (Mack et al., 2001). Antibodies used: anti-GluA1 (1:2,000, polyclonal, Millipore, RRID:AB_390157), anti-GluA2 (1:800, monoclonal, Millipore, clone L21/32, RRID:AB_10806492), anti-GluA3 (1:1,000, monoclonal, Millipore, clone 3B#, RRID:AB_2113897), anti-GluA4 (1:400, polyclonal, Millipore, RRID:AB_310095), anti-GluN1 (1:600, polyclonal, Millipore, RRID:AB_2112158), anti-β actin (1:40,000, monoclonal, Sigma, clone AC-15, RRID:AB_476744); secondary goat anti-rabbit (RRID:AB_2336198) and goat anti-mouse coupled to horseradish-peroxidase (1:15,000, Vector; RRID:AB_2336171). Data are presented as mean ± SEM. Western blot quantification was statistically evaluated by analysis of variance (ANOVA) measurements followed by Holm-Sidak's multiple comparison and Bonferroni post-hoc tests (Prism 6, RRID:SCR_002798; IGOR Pro, RRID:SCR_000325).

Brains were removed from deeply anesthetized mice (halothane; age P42) and transverse hippocampal slices (250 μm) were prepared and incubated for 30 min at 37°C in artificial CSF (ACSF) containing (in mM): 125 NaCl, 25 NaHCO₃, 2.5 KCl, 1.25 NaH₂PO₄, 1 MgCl₂, 25 D-glucose, 2 CaCl₂; bubbled with 95% O₂/5% CO₂ (pH 7.4). Patch pipettes were pulled from borosilicate glass capillaries and had resistances of 4–7 MΩ when filled with (in mM) 125 Cs-gluconate, 20 CsCl, 10 NaCl, 10 HEPES, 0.2 EGTA, 4 MgATP, 0.3 Na₃GTP, 100 μM spermine, and 2.5 mM QX-314 (pH 7.3, 290–305 mOsm). All chemicals were obtained from Sigma. Series resistances and input resistances were continuously monitored by measuring peak and steady-state currents in response to hyperpolarizing pulses (−5 mV; 20 ms). Liquid junction potentials were corrected. Synaptic currents were activated between −70 and +40 mV in 10 mV steps by stimulating the Schaffer collateral/commissural fibers in str. radiatum 150 μm away from the CA1 cell body with a glass electrode filled with 1 M NaCl. AMPAR currents were recorded in presence of 50 μM D-2-amino-5-phosphonopentanoic acid (D-AP5; Tocris), 10 μM bicuculline methiodide (Sigma) and 1 μM CGP 55845 (Tocris). Single traces were analyzed and illustrated. The rectification index (RI) is given as the current ratio at +40 and −60 mV. Data are presented as mean ± SEM. Statistical significance was evaluated by a two-tailed, unpaired Student's t-test.

Pairing-induced LTP was induced by pairing low frequency stimulation (120 pulses, 0.67 Hz) with postsynaptic depolarization to 0 mV for 3 min as published in Chen et al. (1999). Monopolar stimulation electrodes were placed in the str. radiatum and in the str. oriens. The former was used to induce LTP, whereas the latter activated the control pathway. Excitatory postsynaptic currents (EPSCs) were elicited by activation of the two pathways (0.2 Hz) and were recorded for 20 min at –70 mV after the LTP-induction. The following intra- and extra-cellular solutions were used: Intracellular (in mM): 120 CsGluconate, 10 CsCl, 8 NaCl, 10 HEPES, 10 phosphocreatine, 0.2 EGTA, 4 MgATP, 0.3 NaGTP. The pH was set to 7.24 with CsOH and osmolarity was analyzed (295–310 mOsm). Extracellular (in mM): 124 NaCl, 26 NaHCO₃, 2.5 KCl, 1.25 NaH₂PO₄, 4 MgSO₄, 4 CaCl₂, 10 glucose. All chemicals were obtained from Sigma. Statistical analysis was done by a two-tailed paired Student's t-test.

Potentiation of hippocampal field excitatory postsynaptic potentials (EPSPs) was induced by tetanic stimulation as previously published. In all these studies Vidar Jensen and Øivind Hvalby performed the experiments under the same conditions and at the same E-Phys. setups (Feldmeyer et al., 1999; Zamanillo et al., 1999; Mack et al., 2001; Jensen et al., 2003; Shimshek et al., 2006). To standardize tetanization strength in different experiments, the tetanic stimulation strength was set in response to a single shock at intensity just above the threshold for generating a population spike. Synaptic efficacy was assessed measuring the slope of the fEPSP in the middle third of its rising phase. Six consecutive responses (1 min) were averaged and normalized to the mean value recorded 4–7 min prior to tetanic stimulation. In some experiments D-AP5 (50 μM, Sigma) was present during the recordings. Statistical significance of LTP levels between tetanized and non-tetanized pathways were calculated by Student's paired two-tailed t-test. LTP levels between genotypes were evaluated by linear mixed model statistical analysis (SAS 9.2, RRID:SCR_008567).

SWM was administrated in the rewarded alternation task on an elevated T-maze (Deacon et al., 2002; Reisel et al., 2002). The T-maze consisted of a start arm (47 × 10 cm) and two identical goal arms (35 × 10 cm) with 10 cm high walls made out of black-painted wood. Mice were kept on diet at 85–90% of the starting body weight and were habituated to the investigator and the T-maze 2 days before testing. For the test, each trial consisted of a sample run followed by a choice run; the two separated by 15 s. During each run a food reward (30 μl sweetened, condensed milk; 4% fat, 10% fat-free dry milk, 27% sugar) was available in a food vial at the end of both arms. On the sample run the choice arms was blocked and the mouse picked up the reward in the sample arm. For the choice run, both arms of the T-maze were open and mice were rewarded for choosing the choice arm and unrewarded when choosing the previously visited sample arm. Correct choices in the choice runs of eight trials per day (four trials in the morning and four trials in the afternoon) were pooled and monitored as daily “block” performance. Behavior was statistically evaluated by analysis of variance (ANOVA) measurements followed by Holm-Sidak's multiple comparison and Bonferroni post-hoc tests (Prism 6, RRID:SCR_002798; IGOR Pro, RRID:SCR_000325).

Acquisition of SRM was performed with mice kept on a strict food diet (remain to 85–90% of the starting body weight) on an elevated Y-shaped maze with prominent extra-maze cues as previously described (Shimshek et al., 2006). In brief, the Y-maze consisted of three identical arms without walls (arms: 50 × 10 × 0.5 cm; angle: 120°; height: 110 cm) made of black painted wood. Mice were trained in 10 sessions per day (inter-trial interval of 10–15 min; 10 sessions in total) to find a milk reward (30 μl sweetened milk) at the end of a designated target arm (marked by a checkerboard pattern as extra-maze cue). The other two arms were assigned as starting position in a pseudo-random order (no more than three successive starts from the same arm with equal numbers of starting positions per day). On a given trial, the mouse was placed at the distal end of the starting arm and the initial entering of one of the other two arms was evaluated as correct (target arm) or incorrect (other start arm) trial. During the initial two sessions, exploring the maze and consuming the bait in the target arm (including entering and re-entering of all arms) was allowed to habituate to the spatial reward location. From session three on, the mouse was removed from the Y-maze when entering the wrong arm. To avoid any olfactory, visible or tactile cue inside the setup directed to a particular arm, the Y-maze was rotated by 120° in random direction between each trial. Mice were trained in two daily blocks of five trials (one in the morning, the other in the afternoon) for 10 days (100 trials in total). Successful trials were recorded and pooled as daily performance. Data represent mean ± SEM. Behavior was statistically evaluated analysis of variance (ANOVA) measurements followed by Holm-Sidak's multiple comparison and Bonferroni post-hoc tests (Prism 6, RRID:SCR_002798; IGOR Pro, RRID:SCR_000325).

Restoration of endogenous Ca²⁺-permeable AMPARs in GluA1-deficient mice (Gria1^−/−) was achieved in Gria1^−/−/2^ΔFb and Gria1^−/−/2^QFb mice either by Cre-mediated deletion of the Gria2^2lox gene (Shimshek et al., 2006) or Cre-mediated removal of a loxP-flanked selection marker in Gria2^neo. The presence of the neo gene in the targeted Gria2 gene attenuates the expression of the GluA2 Q/R-site editing-deficient, hypomorphic Gria2^neo allele (Feldmeyer et al., 1999). The αCaMKII promoter-driven transgene Tg^Cre4 (Mantamadiotis et al., 2002) was used to provide specific Cre expression in principal neurons of the forebrain.

The Gria1^−/−/2^ΔFb mice are viable in contrast to Gria1/2 double knockout mice, which die shortly after birth (V. Mack, personal observation). Gria1^−/−/2^QFb mice were also viable, but the epileptic phenotype observed in mice with forebrain-specific, heterozygous GluA2(Q) expression (Gria2^QFb; Krestel et al., 2004), persisted in the absence of GluA1. However, life expectancy of Gria1^−/−/2^QFb mice was increased, thus permitting behavioral analysis; more than 60% of Gria1^−/−/2^QFb mice reached P60 (Supplementary Figure 1) compared to <40% of Gria2^QFb mice (Krestel et al., 2004).

To quantify and to visualize the expression of AMPAR subunits, we determined the hippocampal expression pattern of GluA1–3. As expected from the Cre expression pattern of Tg^Cre4 mice, the hippocampal GluA2 expression was abolished in Gria1^−/−/2^ΔFb. In Gria1^−/−/2^QFb mice the GluA2 signal was reduced and accumulated at somatic sites as it does in Gria1^−/− mice (Figure 1A). The normalized protein levels in immunoblots of hippocampal extracts confirmed the absence of GluA2 in Gria1^−/−/2^ΔFb. The strong GluA2 reduction in Gria1^−/−/2^QFb compared to Gria1^−/− mice (40.8 ± 10.4% vs. 92.8 ± 6.1%, mean ± SEM, p < 0.005; Figure 1B) was more pronounced than described for the expression of the modified Gria2^neo gene (Feldmeyer et al., 1999). The reduction and lack of GluA2 was accompanied by a substantial reduction of GluA3 in Gria1^−/−/2^ΔFb (74.3 ± 4.6%) and Gria1^−/−/2^QFb mice (61.7 ± 4.5%) compared to control (98.7 ± 5.4%) and Gria1^−/− mice [100.8 ± 6.4%; F_{(3, 16)} = 13.88; p = 0.0001; Holm-Sidak pairwise comparison at p < 0.03 for Gria1^−/−/2^QFb, resp. p < 0.0006 for Gria1^−/−/2^ΔFb vs. Gria1^−/− and control]. In Gria1^−/−/2^QFb mice GluA3 levels seemed to be more reduced. However, the difference between GluA3 levels in Gria1^−/−/2^ΔFb and Gria1^−/−/2^QFb did not reach statistical significance (p = 0.22). The levels of the NMDAR subunit GluN1 [F_{(3, 9)} = 0.952; p = 0.456] were not altered. Similarly, the amount of the GluA4 subunit, which is not expressed in mature hippocampal pyramidal neurons, was unchanged [F_{(3, 15)} = 1.177; p = 0.352].

In order to show that the remaining GluA3 and the activated GluA2(Q) subunits form Ca²⁺-permeable AMPARs in hippocampal pyramidal cells, we performed whole-cell recordings in acute hippocampal slices of Gria1^−/−/2^ΔFb and Gria1^−/−/2^QFb mice. In accordance with previous studies of homomeric GluA3 AMPARs (Boulter et al., 1990), we observed an increased AMPAR-mediated conductance in hippocampal brain slices of Gria1^−/−/2^ΔFb (8.02 ± 0.84 pA/V, n = 10) compared to Gria1^−/− (5.13 ± 0.9 pA/V, n = 7; p < 0.05). In Gria1^−/−/2^QFb, the presence of the higher conducting GluA2(Q)-containing AMPARs (Feldmeyer et al., 1999) generated excitatory postsynaptic currents (EPSCs, which were similar to those of AMPARs in slices of wild-type mice (wild type: 12.72 ± 1.96 pA/V, n = 15 vs. Gria1^−/−/2^QFb 11.53 ± 3.45 pA/V, n = 3; p = 0.58). Importantly, both genotypes expressed Ca²⁺-permeable AMPARs in CA1 pyramidal cells, as indicated by rectification indices (RIs) of current-voltage relationships (Figure 1C; Burnashev et al., 1992). In wild-type mice the RI was close to 1 (1.45 ± 0.17), since GluA2 renders AMPARs impermeable for Ca²⁺ (Burnashev et al., 1992). The GluA2-deficiency in Gria1^−/−/2^ΔFb was confirmed by the RI increase (15.7 ± 6.8 vs. 1.45 ± 0.17, p < 0.05), which is characteristic of Ca²⁺-permeable AMPARs lacking the GluA2 subunit (Washburn and Dingledine, 1996). The smaller, but still significant RI increase in Gria1^−/−/2^QFb (2.12 ± 0.15, p < 0.05) can be explained by the presence of two Gria2 alleles (Gria2 and Gria2^ΔECS) leading to a mixed AMPAR population containing Ca²⁺-permeable GluA2(Q) and Ca²⁺-impermeable GluA2 receptors.

The mixed AMPAR population could also be monitored in brain slices from Gria1^−/−/2^QFb mice by a small but significant amount of NMDAR-independent LTP (1.12 ± 0.03 vs. 1.00 ± 0.02, p < 0.01) measured in the presence of the NMDAR antagonist D-AP5 (Supplementary Figure 2) as previously reported in for heterozygous Gria2^+/ΔECS mice (Feldmeyer et al., 1999). In Gria1^−/−/2^ΔFb mice the remaining AMPARs resulted in five-fold reduced currents (Figure 1C) and LTP was completely blocked in the presence of the NMDAR antagonist D-AP5 (1.04 ± 0.03 vs. 1.01 ± 0.04, p = 0.53; Supplementary Figure 2) as described before for forebrain-specific GluA2 knockout mice (Gria2^ΔFb) mice (Shimshek et al., 2006).

Activity-induced changes in synaptic responses at CA3-to-CA1 synapses were assessed in acute brain slices of adult mice using cellular- and field-recordings (Figures 2, 3). In slices of control mice, low frequency stimulation (0.67 Hz for 3 min) at presynaptic sites in str. radiatum paired with depolarization (at 0 mV) in voltage-clamp, whole cell recordings of hippocampal CA1 neurons elicited a robust and long-lasting potentiation of excitatory postsynaptic currents (EPSCs) compared to the un-paired control pathway in str. oriens (pairing-induced LTP after 20 min vs. control pathway, wild type = 2.28 ± 0.20 vs. 1.24 ± 0.13, p < 0.01). Consistent with previous observations (Jensen et al., 2003), CA1 neurons in slices of adult Gria1^−/− mice did not express a significant pairing-induced LTP (1.18 ± 0.06 vs. 1.00 ± 0.14, p > 0.3; Figure 2). However, after additional genetic removal of GluA2 the remaining AMPARs in pyramidal CA1 neurons lacking both, GluA1 and GluA2, were sufficient to produce pairing-induced LTP in hippocampal slices of Gria1^−/−/2^ΔFb mice (1.67 ± 0.1 vs. 1.13 ± 0.1, p < 0.01). Moreover, expression of GluA2(Q) in Gria1^−/−/2^QFb mice enabled potentiation of CA1 EPSCs (2.07 ± 0.14 vs. 1.05 ± 0.09, p < 0.01) which was similar to pairing-induced LTP of control mice (Figure 2).

In hippocampal field recordings using the tetanization paradigm (100 Hz, 1 s) we could also monitor LTP in Gria1^−/−/2^ΔFb and Gria1^−/−/2^QFb mice (Figure 3A). Field excitatory postsynaptic potentials (fEPSPs) in the tetanized pathway were significantly increased when compared to the non-tetanized control pathway (normalized fEPSP slopes 45 min after LTP induction), both in Gria1^−/−/2^ΔFb (1.19 ± 0.05 vs. 1.04 ± 0.03, p < 0.01) and in Gria1^−/−/2^QFb (1.21 ± 0.04 vs. 1.00 ± 0.02, p < 0.01; Figure 3A).

Similarly, we obtained an LTP rescue in Gria1^−/− mice that express a transgenic, PDZ motif-truncated and GFP-tagged GluA1(TG) mutation (Freudenberg et al., 2013a) in excitatory neurons of the forebrain (Gria1^−/−/Tg^8.1). In these mice the αCaMKII promoter-driven transgene Tg^aCaMKII-tTA (Mayford et al., 1996) permits the cell-type specific GFP-GluA1(TG) expression (Figure 3B, left). Hippocampal LTP of Gria1^−/−/Tg^8.1 mice was well-developed (1.29 ± 0.04 vs. 1.01 ± 0.02; p = 0.01), but was still significantly reduced (p < 0.05) when compared to LTP of wild-type mice (1.47 ± 0.05 vs. 1.03 ± 0.01; p = 0.01; Figure 3B, right). Importantly, the field-LTP in Gria1^−/−/Tg^8.1 mice reached a potentiation level that was monitored in slices of Gria1^−/−/2^ΔFb and Gria1^−/−/2^QFb mice and that was achieved by the transgenic, full-length GFP-GluA1 expression (Mack et al., 2001). But despite the higher transgenic GFP-GluA1(TG) expression levels compared to GFP-GluA1 (Freudenberg et al., 2013a) and comparable LTP, the SWM of in Gria1^−/− mice was only observed in GFP-GluA1-, but not in GFP-GluA1(TG)-expressing GluA1 knockout mice (Schmitt et al., 2005; Freudenberg et al., 2013a,b). Similarly, the forebrain-specific GluA2 knockout mice (Gria2^ΔFb) developed regular levels of LTP but showed strong SWM deficits (Shimshek et al., 2006).

The lack of SWM in Gria1^−/−/Tg^8.1 mice and Gria2^ΔFb, despite the presence of partial or full LTP (Shimshek et al., 2006; Freudenberg et al., 2013b) led us to study the SWM performance of Gria1^−/−/2^ΔFb and Gria1^−/−/2^QFb mice. We tested Gria1^−/−/2^ΔFb and Gria1^−/−/2^QFb mice together with control and Gria1^−/− mice as negative control in the rewarded alternation task on a T-maze. Control mice alternated efficiently and visited the previously blocked target arm in the test run (77.6 ± 3.1) while Gria1^−/− performed not different to chance level (53.4 ± 1.6) as reported (Reisel et al., 2002).

Regardless of the activated expression of endogenous Ca²⁺-permeable AMPARs and the restored LTP at CA3-to-CA1 synapses of Gria1^−/−/2^ΔFb and Gria1^−/−/2^QFb mice, both lines displayed in the rewarded alternation task on the elevated T-maze a blunted SWM comparable to that of Gria1^−/− mice (Gria1^−/−/2^ΔFb, 53.6 ± 2.3; Gria1^−/−/2^QFb, 57.5 ± 2.8; five blocks of eight trials, correct trials in %; mean ± SEM; Figure 4A). Repeated measures of two-way ANOVA revealed a main effect of genotype [F_(3/44) = 22.51, p < 0.0001] and block [F_{(4, 176)} = 2.61; p < 0.04], but any genotype-by-block interaction [F_{(12, 176)} = 0.73; p > 0.72]. Post-hoc Student-Newman-Keuls comparison identified significant differences (p < 0.05) for all three mouse models (Gria1^−/−, Gria1^−/−/2^ΔFb, Gria1^−/−/2^QFb) vs. control mice while mutants did not differ among each other. In addition, one sample t-test to the theoretical mean of 50% SWM performance (chance level) revealed significant values (p < 0.05) for control mice while all mice of the GluA1-deficient mouse lines did not perform differently from chance level.

Gria1^−/−/2^ΔFb mice were also tested to learn a fixed location of an arm in a Y-maze in 10 blocks of 10 trials each (Figure 4B). Gria1^−/−/2^ΔFb mutants were able to find the milk reward in the designated target arm efficiently and similar as control littermates (block 10: 90.0 ± 5.8 vs. 99.0 ± 1.0, p > 0.19), supporting the finding of Gria1^−/− mice that the SWM is not a prerequisite for the formation of SRM (Reisel et al., 2002; Sanderson et al., 2009). However, as also observed in Gria2^ΔFb mice (Shimshek et al., 2006), SRM acquisition in Gria1^−/−/2^ΔFb mice was delayed and showed a lower success rate on day 4, 5, and 6 (p < 0.05, p < 0.001, p < 0.001, respectively; Bonferroni post-hoc test) indicating a specific role of GluA2 for certain behaviors.

DS and BS are currently employed by Novartis Pharma AG, Basel, Switzerland. However, this work was completed whilst DS and BS were employed at the Max Planck Institute for Medical Research. The other authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.