The ISR downstream target ATF4 represses long-term memory in a cell type-specific manner

Abstract

The integrated stress response (ISR), a pivotal protein homeostasis network, plays a critical role in the formation of long-term memory (LTM). The precise mechanism by which the ISR controls LTM is not well understood. Here, we report insights into how the ISR modulates the mnemonic process by using targeted deletion of the activating transcription factor 4 (ATF4), a key downstream effector of the ISR, in various neuronal and non-neuronal cell types. We found that the removal of ATF4 from forebrain excitatory neurons (but not from inhibitory neurons, cholinergic neurons, or astrocytes) enhances LTM formation. Furthermore, the deletion of ATF4 in excitatory neurons lowers the threshold for the induction of long-term potentiation, a cellular model for LTM. Transcriptomic and proteomic analyses revealed that ATF4 deletion in excitatory neurons leads to upregulation of components of oxidative phosphorylation pathways, which are critical for ATP production. Thus, we conclude that ATF4 functions as a memory repressor selectively within excitatory neurons.

Keywords: integrated stress response; learning and memory; protein synthesis; synaptic plasticity.

Fig. 1.

ATF4 deletion in excitatory neurons facilitates LTM formation. (A) Schematic of the ISR pathway. The central component of the ISR, eIF2α, is phosphorylated by four dedicated ISR kinases, which leads to a decrease in global protein synthesis but a paradoxical increase in translation of the Atf4 mRNA. (B) The breeding strategy used to generate forebrain excitatory neuron-specific Atf4 knockout mice. (C) Control (Atf4^+/+:Camk2a-Cre) and mice in which ATF4 is deleted in excitatory neurons (Atf4^fl/fl:Camk2a-Cre; here defined as Atf4^Ex cKO). (D) Combined ISH/IHC shows selective deletion of Atf4 (green dots) in the CaMKIIα-positive excitatory neurons (depicted by “red” staining) in the CA1 region of Atf4^Ex cKO mice compared to the control group. (E) Quantification of Atf4 mRNA levels in control and Atf4^Ex cKO mice (t = 16.55, degrees of freedom (df) = 4.202, n = 5/group, P < 0.0001, unpaired t test with Welch’s correction; each point in the graph represents means per mouse). (F) Protocol for weak MWM test. (G and H) Enhanced spatial LTM acquisition in Atf4^Ex cKO mice (Day 5: n = 10/group, P = 0.0471, two-way ANOVA with Tukey’s post hoc comparisons) and preference for the target quadrant on the probe test under a weak training paradigm (n = 10/group, P = 0.0130, two-way ANOVA with Tukey’s post hoc comparisons). The Atf4^Ex cKO mice spent significantly more time than the baseline by a one-sample t test (P = 0.0048 for Atf4^Ex cKO and P = 0.2667 for control mice), indicating that weak training was sufficient to form an LTM. (I) The swimming distances of the two genotypes were comparable (mean distance for control: 12.56 ± 0.25 m and Atf4^Ex cKO: 11.70 ± 0.30 m; t = 2.08, df = 17.51, n = 10/group, P = 0.0525, unpaired t test with Welch’s correction). (J) The time required to find the visible platform was similar between control and Atf4^Ex cKO mice (t = 0.4238, df = 17.66, n = 10/group, P = 0.6769, unpaired t test with Welch’s correction). (K) Schematic of the weak contextual fear conditioning (CFC) protocol (1-foot shock; 0.35 mA for 1 s). The percentage of freezing was calculated for 5 min 24 h posttraining. (L) Long-term contextual fear memory is enhanced in Atf4^Ex cKO mice compared to the control group (n = 8/group, On test day: P = 0.0003, two-way ANOVA with Bonferroni’s post hoc comparisons). (M) Time course of normalized synaptic changes induced by high-frequency stimulation (100 Hz for 1 s) in control (n = 7 slices obtained from five mice) and Atf4^Ex cKO mice (n = 7 slices obtained from six mice). (N) Quantification of average LTP recording after stimulation (last 10 min of recording). Average LTP for control = 106.9 ± 4.2 and Atf4^Ex cKO = 118.1 ± 6.1; the asterisk (*) indicates a significant increase from the baseline 100 as determined by a Wilcoxon test (P = 0.0469). Data are mean ± SEM. *P < 0.05, ***P < 0.001, ****P < 0.0001, n.s. = not significant.

Fig. 2.

Deletion of Atf4 in inhibitory neurons has no effect on LTM formation. (A) Schematic of the breeding strategy used to generate inhibitory neuron-specific Atf4 knockout mice. (B) Diagram of the two genotypes of mice Atf4^+/+:Gad2-Cre (control) and Atf4^fl/fl:Gad2-Cre (Atf4^In cKO) used in experiments. Combined ISH/IHC showing selective deletion of Atf4 (green dots) in the GAD67-positive inhibitory neurons (depicted by red staining) of Atf4^In cKO mice compared to the control group. (C) Quantitation of Atf4 mRNA levels in control and Atf4^In cKO mice (t = 9.168, df = 5.461, n = 5/group, P = 0.0002, unpaired t test with Welch’s correction; each point in the graph represents means per mouse). (D and E) Deletion of Atf4 in the inhibitory neurons does not affect memory acquisition, and mice from both genotypes showed similar preference for the target quadrant on the probe test (n = 10/group, P = 0.9786, two-way ANOVA, two-way ANOVA with Tukey’s post hoc comparisons). The control and Atf4^In cKO mice did not spend significantly more time in the target quadrant than the baseline (one-sample t test, P = 0.1357 for control and P = 0.1714 for Atf4^In cKO), indicating that the weak training was insufficient to induce a robust LTM in both groups. (F) Latency to find the visible platform is similar between control and Atf4^In cKO mice (t = 0.2631, df = 11.67, n = 9 for control and n = 10 for the Atf4^In group, P = 0.7971, unpaired t test with Welch’s correction). (G) Long-term contextual fear memory is similar between control and Atf4^In cKO mice (n = 8 mice/group, On test day: P = 0.5449, two-way ANOVA with Bonferroni's post hoc comparisons). Data are mean ± SEM. ***P < 0.001, n.s. = not significant.

Fig. 3.

Deletion of Atf4 in astrocytes has no effect on LTM formation. (A) Schematic of the breeding strategy used to generate astrocyte-specific Atf4 knockout mice. (B) Diagram of the two genotypes of mice Atf4^+/+:Gfap-Cre (control) and Atf4^fl/fl:Gfap-Cre (Atf4^Astro cKO) used in experiments. Combined ISH/IHC showing selective deletion of Atf4 (green dots) in the GFAP-positive astrocytes (depicted by red staining) of Atf4^Astro cKO mice compared to the control mice. (C) Quantitation of Atf4 mRNA levels in control and Atf4^Astro cKO mice (t = 4.025, df = 6.897, n = 5/group, P = 0.0052, unpaired t test with Welch’s correction; each point in the graph represents means per mouse). (D and E) Deletion of Atf4 in the astrocytes does not affect memory acquisition, and mice from both genotypes showed similar preference for the target quadrant on the probe test (n = 7/group, P = 0.1727, two-way ANOVA with Tukey’s post hoc comparisons). The control and Atf4^Astro cKO mice did not spend significantly more time in the target quadrant than the baseline (one-sample t test, P = 0.3943 for control and P = 0.0868 for Atf4^Astro cKO), indicating that the weak training was insufficient to form a robust LTM in both these groups. (F) The latency to find the visible platform is similar between the control and Atf4^Astro cKO (t = 1.093, df = 11.83, n = 7 mice/group, P = 0.2960, unpaired t test with Welch’s correction). (G) Long-term contextual fear memory is similar between Atf4^Astro cKO and control mice (n = 7 mice/group, On test day: P = 0.2474, two-way ANOVA with Bonferroni's post hoc comparisons). Data are mean ± SEM. **P < 0.01, n.s. = not significant.

Fig. 4.

Selective molecular changes in mice in which Atf4 is deleted in excitatory neurons. (A) Schematic of synaptosome fractionation protocol from the mouse hippocampus (generated using BioRender.com). (B) Venn diagram of the common (1,232) and unique (69) protein hits identified in the proteomic analyses of the synaptosome lysates obtained from control and Atf4^Ex cKO mice. (C) Top 10 up-regulated proteins in the synaptosomes of Atf4^Ex cKO mice (based on foldchange relative to the control mice). (D) Western blot from the cytoplasm and synaptosome fractions of control and Atf4^Ex cKO mice. The blots were probed with antibodies against Homer1, Homer2, and Homer3. β-actin was used as a loading control. Cytoplasmic fraction was run as a control to confirm the purity of the synaptosome fractionation; an antibody against cyclin-dependent kinase 5 (CDK5) was used as a marker of cytoplasmic fraction. The bar graphs in the Right panel represent the densitometric quantification of the bands for each protein (n = 3/group; for Homer3: t = 4.953, df = 2.006, P = 0.0382; for Homer2: t = 0.2585, df = 3.939, P = 0.8090; for Homer1: t = 0.1136, df = 2.308, P = 0.9187, unpaired t test with Welch’s correction). (E) Pathway enrichment analysis of the 69 differentially abundant proteins in the synaptosome of Atf4^Ex cKO mice. (F) Heatmap of the OXPHOS pathway proteins that are up-regulated in the synaptosome of Atf4^Ex cKO mice. (G) Primary hippocampal pyramidal neurons were cultured from the samples obtained from control and Atf4^Ex cKO mice. The DIV21 neurons were cotreated with TMRE (red) and NeuO (green) and mitochondrial membrane potential (ΔΨm) was determined from the intensity of the signals. Merged images for four different cells per genotype are shown as representatives. Bar graph (on the Right) showing the quantification of the TMRE signals. Each dot indicates the intensity of the TMRE signal obtained from a single cell. t = 3.980, df = 18.04, P = 0.0009, unpaired t test with Welch’s correction. (H) The deletion of Atf4 in the DIV21 neurons of Atf4^Ex cKO was confirmed by a western blot using an ATF4 antibody. CaMKIIα expression indicated the activation of Cre in these cells. β-actin was used as a loading control. The bar graph (on the Right) shows the densitometric quantification of the ATF4 bands (t = 3.819, df = 2.678, n = 3/group, P = 0.0385, unpaired t test with Welch’s correction). Data are mean ± SEM. *P < 0.05, ***P < 0.001, n.s. = not significant.