Report

Multiple repressive mechanisms in the hippocampus during memory formation

See allHide authors and affiliations

Science  02 Oct 2015:
Vol. 350, Issue 6256, pp. 82-87
DOI: 10.1126/science.aac7368

Memory consolidation by gene suppression

Storing a persistent memory in the brain involves dynamic gene regulation. However, our knowledge of the target genes controlled during memory formation is limited. Cho et al. used RNA sequencing and ribosome profiling to compare transcription and translational levels in the mouse hippocampus before and after memory formation. Under basal conditions, there was an unexpected translational repression of ribosomal protein-coding genes. Early after learning, specific genes were translationally repressed. Later, suppression of a group of genes resulted from the inhibition of estrogen receptor alpha signaling. Thus, suppression mechanisms in the hippocampus appear to play a major role during memory consolidation.

Science, this issue p. 82

Abstract

Memory stabilization after learning requires translational and transcriptional regulations in the brain, yet the temporal molecular changes that occur after learning have not been explored at the genomic scale. We used ribosome profiling and RNA sequencing to quantify the translational status and transcript levels in the mouse hippocampus after contextual fear conditioning. We revealed three types of repressive regulations: translational suppression of ribosomal protein-coding genes in the hippocampus, learning-induced early translational repression of specific genes, and late persistent suppression of a subset of genes via inhibition of estrogen receptor 1 (ESR1/ERα) signaling. In behavioral analyses, overexpressing Nrsn1, one of the newly identified genes undergoing rapid translational repression, or activating ESR1 in the hippocampus impaired memory formation. Collectively, this study unveils the yet-unappreciated importance of gene repression mechanisms for memory formation.

Storing a persistent memory in the brain involves dynamic gene regulation. This is orchestrated by complex processes, including chromatin alteration and the activity of transcription factors (1), as well as translational controls by mammalian target of rapamycin (mTOR) (2), translation initiation or elongation factors (3, 4), and microRNAs (5). The importance of translational regulators for memory consolidation has been demonstrated by assessing the effect of genetic or pharmacological manipulation on behavioral phenotypes (6, 7). However, our knowledge of the target genes controlled translationally during memory formation remains limited. One of the reasons has been the lack of high-throughput techniques to accurately measure the translation rate of endogenous genes in the brain (8). To date, temporal profiles of the translatome paired with the transcriptome throughout the memory consolidation period are not available, and it remains largely unknown which genes are controlled and how they are regulated during memory consolidation.

Ribosome profiling (RPF) allows sensitive and quantitative measurement of translation at the genomic scale (912). Deep sequencing of the ribosome-protected mRNA fragments (“ribosomal footprints”) yields quantitative information about the mRNAs undergoing translation. When normalized against the mRNA level, which can be measured by RNA sequencing (RNA-seq) in parallel, the translational efficiency (TE) of mRNAs from thousands of protein-coding genes can be determined simultaneously.

To map the comprehensive landscape of translatome and transcriptome during memory formation, we used RPF and RNA-seq in the mouse hippocampus after contextual fear conditioning (13) (Fig. 1A and fig. S1A). The hippocampi were collected from untrained mice (control) and from trained mice at 5, 10, and 30 min and 4 hours after conditioning. The qualities of our RPF libraries were confirmed by their enrichment in coding sequences (fig. S1B) and three-nucleotide periodicity of the RPF reads (fig. S1C), two representative signatures of the in vivo movement of ribosomes in translation (912).

Fig. 1 Translational suppression of ribosomal protein-coding genes in the mouse hippocampus.

(A) Workflow of RPF and RNA-seq from the mouse hippocampus after contextual fear conditioning. (B) The top four clusters of functional annotations associated with the genes that have low or high TE in the hippocampus. The top 1100 (10%) genes with high or low TE were selected for functional annotation analysis using the DAVID program (http://david.abcc.ncifcrf.gov). The “non-membrane–bounded organelle” also includes the ribosome. (C) Box plots showing the TEs of mRNAs that encode all proteins, ribosomal proteins, or mitochondrial ribosomal proteins in hippocampal tissues and mESCs (left) and hippocampal primary culture (right). (D) Western blot analysis of mESCs and mouse tissues. RPL26 and RPS6 are components of the ribosome large and small subunit, respectively. Coomassie staining visualizes the amounts of loaded proteins (bottom). (E) Sucrose gradient sedimentation of mouse tissue lysates. Absorbance at 260 nm was recorded from the sedimented lysates to detect ribosome subunits, monosomes, and polyribosomes.

We first conducted functional annotation analysis using the Database for Annotation, Visualization and Integrated Discovery (DAVID) to learn about the global characteristics of hippocampal translation (table S1 and Fig. 1B). The genes related to the translational machinery itself (i.e., the ribosome) were translationally suppressed in the hippocampus (Fig. 1, B and C, and fig. S2, A and C). Considering the importance of translational regulation, this specific suppression of translational machinery was unexpected. We observed similar suppression in cultured hippocampal neurons (Fig. 1C and fig. S2B) but not in the RPF data set from mouse embryonic stem cells (mESCs) (Fig. 1C and fig. S2C). The mRNA levels for ribosomal proteins were comparable in the hippocampus and mESCs (fig. S2D). Unlike cytoplasmic ribosomal protein genes, mitochondrial ribosomal protein genes did not show cell type–dependent differential regulation patterns (Fig. 1C). Western blotting demonstrated that the ribosomal protein levels (RPL26 and RPS6) were indeed much lower in the hippocampus than in mESCs and other tissues (Fig. 1D). We also estimated actively translating ribosomes by sucrose gradient sedimentation (9, 14). Polyribosomes were not clearly detected in the hippocampus, whereas they were readily observed in the kidney, testis, and liver (Fig. 1E). These data collectively indicate that ribosome biogenesis is suppressed translationally in the hippocampus and that overall translation activity is maintained at a low level as compared with that in other tissues.

To identify reproducible temporal changes of gene expression after contextual fear conditioning, we constructed three biologically independent replicates of the RPF and RNA-seq libraries. From these data, we determined differentially expressed genes (DEGs) by calculating the RPF read counts, which are values that provide better estimates of the ultimate protein output from individual genes than those of RNA-seq (15). DEGs at each time point were defined as the genes whose RPF counts altered significantly from those in untrained controls [log2 RPF fold changes above 0.35 or below -0.35, with a false discovery rate (FDR) of less than 0.1] (fig. S3, A to D, and table S2). We identified a total of 104 DEGs (fig. S3). The DEG groups of the 5- to 30-min time points showed only a few overlapping genes, indicating highly dynamic regulation across these time points (fig. S3, E and F).

At 5 min, DEGs showed either decrease (in 15 genes) or increase (in 10 genes) in RPF levels, mostly without significant alterations in RNA levels (14 of 15 decreased DEGs and 8 of 10 increased DEGs) (fig. S4A). Thus, the first wave of gene regulation after conditioning may be through translation rather than transcription. These changes became undetectable by 30 min in most cases (21 out of 22 translationally regulated DEGs at 5 min). Translationally regulated DEGs also appeared at 10 min (6 out of 8 decreased DEGs) (fig. S4B). At 10 and 30 min, the genes known as immediate early genes such as Fos and Arc were obviously induced at RNA and RPF levels (fig. S3, B, C, and E, and fig. S4, B and C), indicating that our data set is in accordance with previous studies of memory (1). Unlike DEGs at the early stage (5 to 10 min), most of the DEGs at 30 min and 4 hours showed concordant changes at mRNA and RPF levels (fig. S4, C and D), indicating that their regulation is largely dependent on mRNA abundance rather than translational control at these later time points. Down-regulation rather than up-regulation became profoundly dominant over time (a decrease in 31 of 42 DEGs at 30 min and in 48 of 55 DEGs at 4 hours; fig. S3, C to F, and fig. S4, C and D). Most of the down-regulation found at 30 min appeared to continue through 4 hours (24 of 31 decreased DEGs at 30 min; fig. S3F). Our time- course study thus uncovered three major waves of alterations: an initial wave of transient translational regulation at around 5 to 10 min; a second wave of induction of immediate early genes at 10 to 30 min; and the suppression of genes through decrease of mRNA levels after 30 min, which continued through 4 hours.

It is generally believed that repressive gene regulation is relieved upon stimulus so as to allow the gene activation necessary for memory consolidation. Only a few cases are known for the repressive mechanisms that are triggered by learning and play an active role in forming a long-term memory (7, 16, 17). In our data set, however, two types of repressive events were markedly induced after learning.

We first looked into the transcript-level suppression. Whereas previous reports have shown that RNA and protein synthesis at 3 to 4 hours after learning are important for memory formation (18, 19), repressive gene regulation at this time point has not been investigated. To understand the mechanism of molecular changes at the late phase of memory formation, we performed ingenuity pathway analysis (IPA) (20) for the 4-hour DEGs (Fig. 2A, right; table S3; and supplementary text). Estrogen receptor 1 (ESR1/ERα) was identified as the most prominent upstream regulator (Fig. 2A; P value = 9.50 × 10−29 at 4 hours) and was predicted to be inhibited. Half of the decreased DEGs at 4 hours (24 of 48) were putative ESR1 downstream genes, and their down-regulation had already started to be seen at 30 min [Fig. 2, A (left) and B]. Otx2, one of the previously reported targets of ESR1 (21), markedly decreased at 30 min and 4 hours (Fig. 2B). OTX2 was also proposed as an upstream regulator of five down-regulated DEGs found at 4 hours (Fig. 2, A and C). Together, these results suggest that the ESR1/OTX2 axis may play a pivotal role in modulating gene-regulatory networks after learning (Fig. 2C and supplementary text).

Fig. 2 Persistent transcript-level down-regulation after contextual fear conditioning via ESR1 inhibition.

(A) Graphical demonstration of upstream regulators from the IPA for DEGs at 30 min and 4 hours. The y axis shows the identified regulators. The x axis displays the significance, which is –log10 (P value). (B) Scatter plots showing the fold changes of RPF (x axis) and RNA-seq (y axis) of all genes (gray) and the downstream genes of ESR1 (red) in 30-min (left) or 4-hour samples (right). (C) Graphical demonstration of the putative regulatory network of ESR1 and Otx2. Our data suggest that upon contextual fear conditioning, ESR1 signaling is suppressed, resulting in the marked decrease of the downstream genes at 4 hours after learning. (D) qRT-PCR of 8 DEGs in the hippocampal tissues of mice injected with MPP (n = 4) or vehicle (veh; n = 6). The hippocampi were collected 3 hours after injection. The Nrsn1 level was measured as a control. Data are shown as mean ± standard error of the mean (SEM). (E and F) Contextual fear conditioning (E) or object location task (F) with ESR1 agonist administration into the hippocampus. One picogram of PPT, the ESR1 agonist, was bilaterally infused into the hippocampus through cannulae right after learning. (E) Contextual fear memory was assessed the next day by measuring the percentage of time that the mice spent freezing in fear (percentage time freezing). Unpaired t test, *P = 0.0378, n = 12 or 13 mice. (F) Object location memory tested 24 hours after sampling. Unpaired t test, *P = 0.0115, n = 8 or 9 mice. Data are shown as mean ± SEM.

To examine whether ESR1 inhibition indeed down-regulates these DEGs at 4 hours in the hippocampus, we subcutaneously administered the ESR1-specific antagonist methyl-piperidino-pyrazole (MPP) and conducted quantitative reverse transcription polymerase chain reaction (qRT-PCR) on hippocampal mRNAs. We observed marked decreases in Otx2 and other putative ESR1 downstream targets after MPP treatment (Fig. 2D), indicating that these genes are under the control of ESR1 in the hippocampus. Furthermore, we assessed the importance of the ESR1 inhibition for memory formation by applying the ESR1-specific agonist 4,4',4"-(4-propyl-[1H]-pyrazole-1,3,5-triyl) trisphenol (PPT) into the hippocampus right after learning. Memory formation was significantly impaired in PPT-injected mice as compared with vehicle-injected mice in two hippocampus-dependent tasks [contextual fear conditioning (Fig. 2E) and object location task (Fig. 2F)], suggesting that down-regulation of ESR1 signaling is important for memory formation.

Next, we examined the translational repression of specific genes soon after learning (within 10 min). ATF4 is the only established gene undergoing translational repression during memory consolidation (4, 7). We could greatly expand the list of translationally repressed genes (14 DEGs at 5 min and 6 at 10 min). Among these genes, Nrsn1 showed a clear decrease in TE at 5 to 10 min without a change in mRNA level (Fig. 3A). Our Western blot analysis also demonstrated that the Nrsn1 protein level in the hippocampus was significantly reduced after conditioning, as compared with that of untrained controls (Fig. 3B). Moreover, elevating neuronal activity in cultured hippocampal neurons resulted in a decrease of the Nrsn1 protein without affecting its mRNA level (Fig. 3C and fig. S5A). The N-methyl-d-aspartate receptor (NMDAR) (22) antagonist d-aminophosphovalerate (d-APV) blocked this activity-dependent reduction of Nrsn1 protein (Fig. 3C). Activity-dependent translational suppression of the Kv1.1 channel depends on NMDAR and mTOR pathways (23). In contrast, Nrsn1 down-regulation by neuronal activity was not affected by treatment with rapamycin, the mTOR inhibitor (Fig. 3D), suggesting that Nrsn1 is controlled by a mechanism distinct from what controls Kv1.1. In addition to Nrsn1, Mapk6, another gene that was found to be translationally repressed in our analysis (fig. S5B), showed a significant decrease in protein level by neuronal activity (fig. S5C) without a change in mRNA level (fig. S5A). This was also NMDAR-dependent (fig. S5D) and mTOR-independent (fig. S5E). These results suggest the involvement of common upstream mechanisms in repression of 5 min DEGs including Nrsn1 and Mapk6.

Fig. 3 Rapid translational repression of Nrsn1 after conditioning.

(A) Line plots showing the fold changes of RPF and RNA-seq (left) and in the TE (right) of Nrsn1 after conditioning. The x axis (time) is in log scale. (B) Western blot and quantitative analyses on NRSN1 in hippocampal tissues from mice (n = 4 or 5) that underwent fear conditioning (FC) compared with home-cage controls (Ctrl). Unpaired t test, **P value = 0.0055. (C) Western blot of NRSN1 in cultured neurons treated with KCl (40 mM) or KCl/APV (50 μM). One-way analysis of variance (ANOVA), F(2,15) = 8.393, **P = 0.0036, post hoc Tukey’s multiple comparison test, **P < 0.01 (Ctrl versus KCl), *P < 0.05 (KCl versus KCl/APV). (D) Western blot of NRSN1 in cultured neurons treated with KCl (40 mM) or KCl/rapamycin (200 nM). One-way ANOVA, F(2,6) = 6.656, **P = 0.0300, post hoc Tukey’s multiple comparison test, *P < 0.05 (Ctrl/DMSO versus KCl/DMSO, Ctrl/DMSO versus KCl/rapamycin), P > 0.05 (KCl/DMSO versus KCl/rapamycin). (E) Schematic of the object location task, see also (13). (F) Object location memory of mice overexpressing Nrsn1 or green fluorescent protein (GFP) (n = 8 or 9 mice). The discrimination index was calculated from the exploration time for each object [(displaced object – nondisplaced object)/(displaced object + nondisplaced object)]. Unpaired t test, *P = 0.0202. (G) Schematic of the contextual fear conditioning procedures, see also (13). (H and I) The percentage time freezing of mice overexpressing Nrsn1 or GFP, showing the strength of contextual fear memory. Unpaired t test, *P = 0.0148, n = 7 mice per group for long-term memory (H) and P = 0.6632, n = 6 mice per group for short-term memory (I). All data are shown as mean ± SEM.

Some of these genes might be “memory suppressor genes” that need to be down-regulated for memory formation (24). Among the translationally repressed DEGs, we chose Nrsn1 for further examination. Nrsn1 encodes a neuron-specific membrane protein, which has been suggested to interact with tubulin and play a role in vesicle trafficking and neurite outgrowth (25, 26). Given that membrane trafficking (27) and microtubule dynamics (28) are essential processes that underlie memory storage, the regulation of Nrsn1 expression might be relevant for learning and memory. To examine the functional role of Nrsn1 in memory formation, we overexpressed Nrsn1 in the mouse hippocampus using adeno-associated virus (fig. S5F). Nrsn1-overexpressing mice showed deficits in long-term memory formation in an object location task (Fig. 3, E and F) and contextual fear conditioning (Fig. 3, G and H). In contrast, short-term memory tested 1 hour after the conditioning was intact (Fig. 3I), supporting the idea that Nrsn1 specifically interferes with the memory consolidation process without affecting the acquisition or expression of contextual fear memory. Thus, Nrsn1 may act as a suppressor of long-term memory formation, suggesting the physiological significance of its down-regulation after learning (24). In addition, we examined whether Nrsn1 might regulate the 4-hour DEGs by overexpressing Nrsn1 in cultured hippocampal neurons. The expression of these genes was not changed significantly by Nrsn1 overexpression, suggesting that Nrsn1 and the ESR1/OTX2 axis may operate independently (fig. S5G).

In addition to uncovering interesting repressive pathways, our quantitative data will provide a valuable resource for future studies of the regulation of numerous individual protein-coding genes in the hippocampus (tables S2 and S4). We could also examine long noncoding RNAs (lncRNAs); among the 886 lncRNAs detected, 13 lncRNAs showed significant changes (FDR < 0.1, log2 fold change >0.35 or <–0.35) for at least one time point (fig. S6, table S5, and supplementary text).

The integrative analysis of transcriptome and translatome in mouse hippocampal tissue after contextual fear conditioning has uncovered numerous gene-regulatory events during memory formation in vivo. In particular, our study illustrates the potential importance of negative gene regulation in learning and memory.

Supplementary Materials

www.sciencemag.org/content/350/6256/82/suppl/DC1

Materials and Methods

Figs. S1 to S6

Tables S1 to S5

References (2946)

References and Notes

  1. Materials and methods are available as supplementary materials on Science Online.
  2. Acknowledgments: We thank H. Chang, S.-C. Kwon, J. Park, Y. Choi, and J.-E. Park of the V. Narry Kim lab and H. K. Choe for helpful discussion of data analysis; and Y. Kim of the same lab for critical commentary on the manuscript. V.N.K was supported by grant IBS‐R008‐D1 from the Institute for Basic Science from the Ministry of Science, Information, Communication and Technology and Future Planning of Korea. B.-K.K. was supported by the National Honor Scientist Program (2012R1A3A1050385) from the National Research Foundation of the Ministry of Science, Information, Communication and Technology. J.-i.K. was supported by the Pohang Iron and Steel Company (POSCO) Tae Joon (TJ) Park Foundation. High-throughput sequencing data have been deposited in the National Center for Biotechnology Information Gene Expression Omnibus (GEO) database with accession number GSE72064.
View Abstract

Navigate This Article