Studies on various forms of synaptic plasticity have shown a link between messenger RNA translation, learning and memory. Like memory, synaptic plasticity includes an early phase that depends on modification of pre-existing proteins, and a late phase that requires transcription and synthesis of new proteins1,2. Activation of postsynaptic targets seems to trigger the transcription of plasticity-related genes. The new mRNAs are either translated in the soma or transported to synapses before translation. GCN2, a key protein kinase, regulates the initiation of translation. Here we report a unique feature of hippocampal slices from GCN2-/- mice: in CA1, a single 100-Hz train induces a strong and sustained long-term potentiation (late LTP or L-LTP), which is dependent on transcription and translation. In contrast, stimulation that elicits L-LTP in wild-type slices, such as four 100-Hz trains or forskolin, fails to evoke L-LTP in GCN2-/- slices. This aberrant synaptic plasticity is mirrored in the behaviour of GCN2-/- mice in the Morris water maze: after weak training, their spatial memory is enhanced, but it is impaired after more intense training. Activated GCN2 stimulates mRNA translation of ATF4, an antagonist of cyclic-AMP-response-element-binding protein (CREB). Thus, in the hippocampus of GCN2-/- mice, the expression of ATF4 is reduced and CREB activity is increased. Our study provides genetic, physiological, behavioural and molecular evidence that GCN2 regulates synaptic plasticity, as well as learning and memory, through modulation of the ATF4/CREB pathway.
Translation of eukaryotic mRNAs is regulated primarily at the level of initiation3. Binding of the initiator tRNA, Met-tRNAiMet, to the 40S subunit is facilitated by the initiation factor 2 (eIF2) which forms a ternary complex with GTP and Met-tRNAiMet. Although phosphorylation of the α subunit of eIF2 can inhibit general translation4,5, it stimulates the mRNA translation of the transcriptional modulator ATF4 (ref. 6), which inhibits synaptic plasticity and behavioural learning in various phyla7,8,9,10. In view of the need for translation for the modulation of synaptic activity and strong evidence that phosphorylation of eIF2α controls translation of ATF4 mRNA6,11,12, eIF2α kinase(s) may regulate synaptic plasticity. Because GCN2 is the most evolutionarily conserved eIF2α kinase and GCN2 mRNA is enriched in the brain of flies13 and mammals (as well as in liver)14,15 (see Supplementary Fig. 1), we explored the role of GCN2 in synaptic plasticity and behavioural learning.
The GCN2 gene was inactivated by homologous recombination in embryonic stem cells (Supplementary Fig. 1A and Supplementary Information). Hippocampal immunohistochemistry and in situ histohybridization show that GCN2, normally expressed mainly in CA1 and CA3 and also in dentate gyrus, is undetectable in brain slices from GCN2-/- mice (Supplementary Figs 1 and 2).
There were no gross morphological changes in the hippocampus or other regions of the brain of GCN2-/- mice (Supplementary Fig. 3), and basal synaptic transmission in CA1 was unaltered as indicated by the following: first, the relation of fEPSPs to stimulus intensity; second, the size of the fibre volley; third, paired-pulse facilitation (PPF); and fourth, peak response to tetanic stimulation (Supplementary Fig. 4 and Supplementary Information). Normally, a single high-frequency tetanus (100 Hz for 1 s) elicits in the Schaffer collateral/commissural pathway a transient form of long-term potentiation known as early LTP (E-LTP), which decays in 2–3 h and does not require RNA or protein synthesis1. In slices from GCN2-/- mice, a single tetanus induces a robust and sustained L-LTP (Fig. 1a; at 180 min, P < 0.001) and the initial potentiation is greater than in slices from wild-type (WT) mice (Fig. 1a; at 15 min, P < 0.05). This increase was synapse-specific because a control input that received only test stimulation remained stable for the entire experimental session (Supplementary Fig. 5A). GCN2 therefore affects the duration of LTP and its initial amplitude.
Like the L-LTP normally elicited by four tetanic trains, the L-LTP induced by a single tetanus in slices from GCN2-/- mice depends on cAMP-dependent protein kinase (PKA) (Fig. 1b; at 180 min, P < 0.01), new mRNA (Fig. 1b; at 180 min, P < 0.01) and protein synthesis (Fig. 1b; at 180 min, P < 0.001) and is resistant to depotentiation (Supplementary Fig. 5C). As expected, E-LTP elicited in slices from WT mice by a single tetanus was not affected by inhibiting these pathways and could be depotentiated (Supplementary Fig. 5B, C). Anisomycin (a translation inhibitor) and actinomycin D (a transcription inhibitor) not only prevented the persistence of LTP in slices from GCN2-/- mice, but also caused an immediate decrease in the early phase of LTP (Fig. 1b). Similarly to our results, the effects of anisomycin on L-LTP in the Schaffer collateral pathway often show an immediate decrement in the magnitude of potentiation, indicating that protein-synthesis-dependent processes are required early after L-LTP induction16,17. The early effect of actinomycin D indicates that the increased amplitude of initial potentiation might be due to the translation of immediate-early genes whose mRNAs are quickly turned over. Indeed, when actinomycin D is applied 15 min (instead of 30 min) before the onset of tetanization to minimize the effects of steady-state levels of rapidly turning-over mRNAs, the drug did not have an immediate effect (Fig. 1c; at 60 min, P > 0.05). Instead there was a delayed decrease in LTP, which was consistent with the lack of induction of new mRNAs necessary for the maintenance of LTP (Fig. 1c; at 180 min, P < 0.01).
According to these observations, deletion of GCN2 leads to an enhanced response to a single tetanus, resulting in L-LTP instead of E-LTP. Does GCN2 deletion also affect the L-LTP normally induced by repeated tetani? To address this question we examined L-LTP induced in CA1 by two different protocols: tetanic stimulation with four trains at 100 Hz, and forskolin, an activator of PKA18. As expected, in slices from WT mice, four trains elicited L-LTP that persisted for at least 4 h. By contrast, in slices from GCN2-/- mice, the LTP decayed to baseline within 3 h (Fig. 1d; at 240 min, P < 0.01). In slices from WT mice, forskolin elicited the usual L-LTP whereas in GCN2-/- slices the L-LTP was not sustained (Supplementary Fig. 5D). The GCN2 deletion specifically affected LTP because long-term depression (LTD), which is induced by low-frequency stimulation or by incubation with an agonist of group I mGluRs, 3,5-dihydroxyphenylglycine (DHPG)19, was unaltered in GCN2-/- slices (Supplementary Fig. 6 and Supplementary Information).
Activation of GCN2 can inhibit the initiation of translation by eIF2α phosphorylation but, paradoxically, it stimulates the translation of ATF4 mRNA6. We therefore measured eIF2α phosphorylation in hippocampal extracts from WT and GCN2-/- mice and found that it was lower (50 ± 19%) in GCN2-/- mice (Fig. 2a). Consistent with this finding was the observation that ATF4 mRNA was shifted to the lighter polysome fractions of hippocampal extracts from GCN2-/- mice (Fig. 2b, c). In agreement with a weak basal translation of ATF4 mRNA, ATF4 protein was correspondingly lower (49 ± 11%; Fig. 2e). By contrast, β-actin mRNA sedimented predominantly in the heavy polysome fractions, as would be expected for an efficiently translated mRNA (Fig. 2b, d). Thus, GCN2 deletion leads to a decrease in translation of ATF4 mRNA in the hippocampus. In accordance with the inhibition of CREB by ATF4, decreased translation of ATF4 mRNA in GCN2-/- mice was associated with enhanced CREB function: expression of immediate-early genes regulated by CREB (BDNF, c-fos, Egr-1) was 25–35% greater in GCN2-/- hippocampal extracts (Fig. 2f).
To further investigate how synaptic plasticity affects GCN2, we examined the effects of forskolin or four trains at 100 Hz (both induce L-LTP and stimulate CRE-mediated gene expression)20 on GCN2 and eIF2α phosphorylation. Both protocols decreased GCN2 and eIF2α phosphorylation in WT but not in GCN2-/- slices (Fig. 2g and Supplementary Fig. 7A). However, E-LTP elicited by a single train was not associated with a decrease in GCN2 and eIF2α phosphorylation (Supplementary Fig. 7B). GCN2 activity is therefore regulated by two forms of strong stimulation that elicit L-LTP, but not by a weaker stimulation that induces only E-LTP.
The effects of GCN2 deletion on long-term learning and memory were first studied in a fear conditioning paradigm. Fear conditioning by two tone–shock pairings has two components. One is contextual fear conditioning, which associates the training context and the footshock and requires both the hippocampus and the amygdala. The second, which associates the tone and the footshock, requires the amygdala but not the hippocampus21. When tested 1 and 10 days after training, GCN2-/- mice showed a deficit in contextual fear conditioning (Fig. 3a, P < 0.05, and Supplementary Information). By contrast, auditory fear conditioning (tested in a different chamber) was intact (Fig. 3b, P > 0.05, and Supplementary Information).
Next, hippocampus-dependent spatial memory was tested in the Morris water maze22. In the course of training (three times a day, at 30-min intervals) the performance of both groups improved (Fig. 4a, P < 0.001), but WT mice learned faster than GCN2-/- mice (Fig. 4a; at 5 days, P < 0.01). In probe tests performed after the end of training, the platform was removed and the mice were allowed to search for 60 s (Fig. 4b). Unlike WT mice (Fig. 4b, P < 0.001), GCN2-/- mice showed no preference for the training quadrant (Fig. 4b, P > 0.05) and fewer platform crossings (Fig. 4c, P < 0.001). Vision and locomotor functions were equally efficient in WT and GCN2-/- mice, as judged by swimming speed (P > 0.05) and latency of escape to a visible platform (P > 0.05). Thus, GCN2 deletion is associated with a specific impairment of hippocampus-dependent learning and memory.
Because a single tetanus elicits L-LTP in slices from GCN2-/- mice (Fig. 1a), we reasoned that mnemonic processes might be enhanced during weaker conditioning. Indeed, when mice were trained only once (compared with three times) a day, Tukey's test showed that escape latencies on day 5 were shorter for GCN2-/- than WT mice (Fig. 4d; at 5 days, P < 0.02). Enhanced spatial learning by GCN2-/- mice was also evident in the probe tests that were conducted 3 days after the end of training (Fig. 4e). According to repeated-measures analysis of variance (ANOVA), the GCN2-/- mice spent significantly more time in the target quadrant (‘trained’ in Fig. 4e) than WT mice did (P < 0.001). Thus, in agreement with the findings on LTP, memory is enhanced after weak training.
The major finding of this study is that a decrease in threshold for L-LTP in CA1 (in slices) is associated with an improved spatial memory of weak conditioning in GCN2-/- mice. A switch from short-term to long-term plasticity8,10,23,24 is generally associated with enhanced gene expression. Indeed, CREB-dependent gene expression is increased in GCN2-/- mice. Thus, GCN2 could effect long-lasting changes in plasticity by modulating CREB activity. The dependence of the early phase of LTP in GCN2-/- mice on transcription and translation may be due to translation of mRNAs coding for CRE-dependent immediate-early genes (which are upregulated at the basal state). Because they turn over rapidly25,26, these mRNAs are likely to be downregulated during the 30 min of preincubation with actinomycin D, whereas in WT slices they are scarce in the basal state but are induced by repeated tetani. L-LTP in WT slices therefore requires a stronger stimulation and is inhibited only at later times. Thus, two mechanisms underlie L-LTP in GCN2-/- slices: first, translation of pre-existing transcripts immediately increases LTP, and second, increased transcription of specific mRNAs generates persistent L-LTP.
How does GCN2 affect synaptic plasticity and learning? One possible model is based on the translational regulation of ATF4 mRNA through the GCN2-mediated phosphorylation of eIF2α. A pivotal point is that ATF4 represses neuronal CREB activity1,7,8,10. Thus, under basal conditions, when GCN2 and eIF2α are phosphorylated and ATF4 levels are high, CREB-dependent transcription, synaptic plasticity and learning are repressed. By decreasing the phosphorylation of GCN2 and eIF2α, LTP-inducing stimulation would remove this inhibition of synaptic plasticity and memory formation. In this manner, GCN2 regulates the switch from short-term to long-term memory.
We documented a correlation between L-LTP and spatial memory. In accordance with the low threshold for L-LTP and its suppression after strong stimulation, the spatial memory of GCN2-/- mice depended on the intensity of training: it was impaired by strong training and enhanced by weaker training. A likely explanation is that strong stimulation (behavioural or by four trains in slices) potentiates an inhibitory pathway that is facilitated in GCN2-/- mice. The nature of this mechanism will be an important target of future studies and may involve changes in regulation of gene expression and/or synaptic translation. Our results indicate that neurons might have not only a threshold for activation of gene expression but also a second threshold at which too much gene expression blocks synaptic plasticity. Shutting off plasticity could be important under conditions of excessive activity such as seizures. Our results provide genetic evidence that translational control by GCN2 is critical for synaptic plasticity, learning and memory. In addition, they raise the prospect that memory formation is regulated through the translational control of transcription.
Generation of transgenic mice by GCN2.KO4 targeting
GCN2 was deleted by a targeting vector constructed from polymerase chain reaction (PCR) fragments amplified from cloned 129SvEv genomic DNA (see Supplementary Methods). Chimaeric mice derived from GCN2.KO4ex/ + embryonic stem cells were prepared by blastocyst injection and the mutant allele was transmitted through the germline to isogenic 129SvEv mice, which were bred to homozygosity. GCN2-/- mice were phenotypically normal in comparison with their wild-type littermates and were obtained in a mendelian ratio.
In situ hybridization histochemistry
Mouse sense and antisense cRNA probes coding for exon 12 of GCN2 were labelled with [35S]UTP and [35S]CTP (1,250 Ci mmol-1; Amersham), and in situ hybridization histochemistry was performed as reported previously27.
Immunoprecipitation, immunohistochemistry and western blotting
Antibodies against the carboxy-terminal portion (kinase domain) of mouse GCN2 kinase (C-term) and ATF4 have been described6. The antibody against the amino-terminal portion (amino acid residues 1–363; N-term) of human GCN2 was produced as a glutathione S-transferase fusion protein in BL-21, and purified on glutathione–Sepharose (APB). Immunoblotting and immunohistochemistry were as reported6,28. Antibodies against phospho-eIF2α total eIF2α and β-actin were purchased from Cell Signalling and Technology Laboratories.
After decapitation of WT (GCN2+/+) or transgenic (GCN2-/-) age-matched littermates (6–12 weeks old), hippocampal slices 400 µm thick were cut with a vibratome and kept submerged at 27–28 °C. Slices were perfused (at 1–2 ml min-1) with oxygenated (95% O2, 5% CO2) artificial cerebrospinal fluid (ACSF) containing 124 mM NaCl, 2.5 mM KCl, 1.25 mM NaH2PO4, 1.3 mM MgSO4, 2.5 mM CaCl2, 26 mM NaHCO3 and 10 mM glucose. Bipolar tungsten electrodes were placed in CA1 stratum radiatum to stimulate Schaffer collateral and commissural fibres, and extracellular field EPSPs (fEPSPs) were recorded from stratum radiatum with a glass microelectrode (2–3 MΩ, filled with 2 M NaCl). Stimulus (0.1-ms duration) was adjusted to evoke 35–40% maximal fEPSPs at 0.033 Hz. LTP was induced with one or four trains (1 s) at 100 Hz delivered 5 min apart. For LTD experiments, 1 Hz stimulation was applied for 15 min. Forskolin (50 µM; Sigma) or DHPG (50 µM; Tocris) was added to the bath after at least 30 min of stable recording. Anisomycin (40 µM; Calbiochem), actinomycin D (40 µM; Calbiochem) or KT5720 (1 µM; Biomol) was applied for 30 min, or as indicated otherwise, before tetanic stimulation. Statistical analysis used t-tests and two-way ANOVA. All data are presented as means ± s.e.m.; n indicates the number of slices. The experimenter was blind to the mouse genotype.
The experimenter (blind to mouse genotype) compared GCN2-/- and WT littermates (males, 2–4 months old). Training consisted of two pairings of a tone (2,800 Hz, 85 dB, 30 s) with a co-terminating footshock (0.7 mA, 2 s). The first tone started 120 s after animals had been placed in the conditioning chamber, where they remained for a further 1 min after the second pairing, and were then returned to their home cage. Mice were tested 1 and 10 days later for freezing in response to training context in a counterbalanced manner (Supplementary Information).
Morris water maze task
The pool was 100 cm in diameter and the water was rendered opaque by the addition of white tempera. Water temperature was kept at 20 °C. The platform was 4.5 cm in diameter.
Mice were trained three times a day at intervals of 30 min, or once a day over five consecutive days. In each trial the mouse swam until it found the platform, or after 120 s it was guided to the platform; the mouse was then placed on the platform for 10 s before being picked up. At the end of the testing period, a probe trial (60 s) was performed. Statistical analysis was based on univariate and multivariate ANOVA, and between-group comparisons were made by Tukey's test.
Polysome profile analysis and RT–PCR
Hippocampal slices were washed twice with cold PBS containing 100 µg ml-1 cycloheximide, suspended in lysis buffer, homogenized with 15 strokes (7-ml Wheaton Dounce) on ice, and then centrifuged for 2 min at 14,000g. Gradients were prepared and analysed as described29. For detection of ATF4 and β-actin mRNAs, RNA from individual fractions was amplified in one-tube RT–PCR reactions, which were optimized to detect the exponential phase on the amplification curve.
The one-step RT–PCR LightCycler RNA Master SYBR Green kit (Roche) was used to quantify CRE-dependent gene expression, as recommended by the manufacturer. Primers, RT–PCR conditions and normalization procedures were as described30.
We thank E. Kandel, K. Krnjević, K. Rosenblum, E. Landau, R. Blitzer, C. Alberini, Y. Mamane and T. Lubell for comments on the manuscript; Y. Zhang, R. Jungreis and A. Sylvestre for assisting in the production and maintenance of the GCN2-/- mice; and Colin Lister for assistance. This work was supported by grants from the Canadian Institute of Health Research (CIHR) and the Howard Hughes Medical Institute (HHMI) to N.S, a CIHR Group Grant to J.-C.L and W.S; a CIHR grant to N. Seidah; an NIH grant to D.R.; CIHR, Natural Sciences and Engineering Research Council of Canada (NSERC), Volkswagen Foundation and EJLB Foundation grants to K.N.; and a CIHR grant to A.C.C. N.S. is a CIHR Distinguished Scientist and a HHMI International scholar. M.C.-M. is supported by a CIHR postdoctoral fellowship.
About this article
Biology & Philosophy (2017)