Learning and memory
The first indication that DNA methylation may be involved in brain activity came from our laboratory. We found that rat learning variably affects the neuronal DNA methyla- tion levels in different regions of the rat brain (Guskova et al., 1977; Vanyushin et al., 1974, 1977). Two models of learning were used. In the first model the male rats were trained to a simple conditioned movement reflex to a light stimulus. Then DNA samples were isolated from neocortex, hippocampus, and cerebellum, and 5mC levels were analyzed by a thin-layer chromatography method. The 5mC level in the cerebellum DNA was not affected (0.99 ± 0.01 mol % in conditioned animals vs 0.96 ± 0.06 mol % in the control animals). In the neocortex DNA the 5mC level was increased by learning from 1.07 ± 0.07 to 1.45 ± 0.13 mol %. Similarly, in hippocampus the 5mC level was increased from 1.15 ± 0.01 to 1.83 ± 0.04 mol %. Thus, DNA methylation levels were variable in different brain divisions and were differently affected by learning. Specifically, learning did affect the genome methylation in brain divisions known to be involved in memory formation and maintenance. These effects were most pronounced at the early steps of conditioning, whereas at later steps there was only a small increase in 5mC levels. In the second model of learning the rats were trained to a Pavlovian food-conditioning reflex. At the early stages (20 and 50 min) of learning the methylation levels of the neocortex
DNA were increased by 22% and 26%, respectively; those of the hippocampus DNA by 18% and 7%,respectively; and those of the cerebellum DNA by 17% and 15%, respectively. At 24 h the neocortex and hippocampus DNAs were still hypermethylated by ~10%, whereas the cerebellum DNA was not hypermethylated at all. In the active control animals that received equivalent numbers of uncombined stimuli, small hypermeth- ylation was observed in the neocortex and cerebellum DNAs after 20 and 50 min, whereas the hippocampus DNA was not hypermethylated. By 24 h the methylation levels of the neocortex and cerebellum DNAs return to control values. Thus, functional activity of brain structures seemed to be accompanied by reversible changes in DNA methylation. Whether these changes represent a mechanism of the gene expression modulation was unknown at the time, although such suggestion seemed reasonable. Regardless, these data allowed us to first announce that brain (neuronal) DNA is involved in the memory formation (Vanyshin et al., 1974). An investigation of DNA synthesis in the rat brain on learning was then conducted to elucidate the nature of DNA hypermethylated on learning and the mechanisms of its subsequent demethylation (Ashapkin et al., 1983). The rat conditioning training, in both food-reward and passive defense reflex models, was found to be accompanied by an increase in DNA synthesis intensity (incorporation of [3H]thymidine) in the brain cortex, whereas no effect was observed in cerebellum. The DNA synthesis induced was found to be selective in respect to various genome sequences and thus could not be a part of DNA replication. We have suggested it to represent an activated DNA repair synthesis involved in local DNA demethylation and chromatin remodeling. This could explain the observed reversibility of DNA methylation.
For a long time the topic of DNA methylation in relation to neurological memory was abandoned. It reappeared many years later, when the nature of genes involved in memory formation and maintenance was elucidated. The activity-dependent transcription of plasticity-related gene Bdnf was shown to be promoted by active demethylation at particular CpG sites in its promoter in postmitotic neurons from mouse cortex (Martinowich et al., 2003). The KCl treatment of cultured neurons led to membrane depolarization, activation of Bdnf transcription, and a decrease in methylation of CpG sites in its promoter. Elevated Bdnf expression in the Dnmt1~/~ mutant mouse brains was correlated with almost complete demethylation of these CpG sites. A partial dissociation of MeCP2 and more tight association of CREB were shown to occur and may account for induced Bdnf expression. In addition, less H3K9me2 and more H3K4me2 and acety- lated H3 and H4 histones were associated with Bdnf promoter after depolarization. Thus, DNA methylation and chromatin remodeling seemed to play critical roles in regulating gene transcription in response to neuronal activity. Treatment of the hippocampal slices with the Dnmt inhibitor zebularine resulted in demethylation of specific CpG island sequences in promoters of two genes involved in memory, reelin and Bdnf (Levenson et al., 2006). Activation of protein kinase C (PKC) in the hippocampal slices with phorbol-12,13-diacetate led to a rapid demethylation of the same sequence in the reelin promoter and an increase in the Dnmt3a gene expression in Area CA1 of the hippocampus. The expression of immediate early gene c-fos was also significantly increased. These results showed that the expression of Dnmt3a in the hippocampus is regulated by PKC signaling cascade and probably plays a role in synaptic plasticity.
Increased levels of Dnmt3a and Dnmt3b mRNAs were found in the hippocampal area CA1 of rats trained to contextual fear conditioning relative to control animals exposed only to the novel context of the experimental chamber (Miller & Sweatt, 2007). Dnmt inhibitors, 5-aza-dC and zebularine, injected directly into area CA1 immediately after training significantly impaired memory retention 24 h later. Thus, the hippocampal Dnmt activity seems to be necessary for memory consolidation. The methylation level of the protein phosphatase 1 (PP1) gene, known to suppress learning and memory, was greatly increased at 1 h after training. At the same time, the methylation levels of reelin gene, which promotes synaptic plasticity and memory, were significantly decreased and its transcription increased. Hence, some DNA demethylation mechanism acting in an activity-dependent manner during memory consolidation must be present in hippocampal neurons. The methylation levels of both reelin and PP1 returned to control levels within a day of training. Thus, DNA methylation changes after training are both rapid and reversible. Very similar findings were reported for Bdnf (Lubin et al., 2008). Bdnf mRNA levels in the area CA1 of rat hippocampus were increased within 30 min of fear conditioning, still more increased at 2 h, and returned to baseline levels at 24 h. The exon-specific Bdnf mRNA levels were correlated with decreased methylation of respective CpG islands. Thus, DNA methylation controls the exon-specific readout of the Bdnf gene. A surprising complexity in the control of DNA methylation at the Bdnf gene locus, involving decreases and increases in methylation at individual transcription initiation sites, was noted. DNA methylation in the adult hippocampus seemed to play a role in memory consolidation, but not in long-term memory storage.
In a rat model of contextual fear conditioning the immediate early gene Egr1 was found to be demethylated in adult neurons of the neocortex both in trained and active control animals at 1 h, 1 day, and 7 days after training (Miller et al., 2010). Reelin gene was hypermethylated in trained rats in 1 h after training. The hypermethylation was reduced at the later time points. The memory suppressor gene calcineurin (CaN) was not affected shortly after training, but highly hypermethylated at 1 and 7 days. Methylated CpGs were randomly distributed across the analyzed 0.5-kb segment of CaN promoter associated CpG island. An N-methyl-D-aspartate (NMDA) receptor antagonist MK-801 interfered with both acquisition of fear memory and hypermethylation of CaN and reelin at 7 days, without affecting Egr1 methylation. Thus, CaN and reelin hypermethylations seemed to be a specific response to associative environmental signals. An infusion of the NMDA receptor antagonist D-(-)-2-amino-5-phosphonovaleric acid directly into dorsal hippocampus (area CA1) immediately before training interfered with both learning and
CaN and reelin methylations in the dorsomedial prefrontal cortex 7 days after training, indicating that a single hippocampus-dependent learning experience is sufficient to drive lasting, gene-specific methylation changes in the cortex. The fear memory and the methylation status of these genes were persistent at 30 days after training. Rats received intracortical infusions of Dnmt inhibitors at 30 days after training failed to display normal memory. Thus, DNA methylation in the dorsomedial prefrontal cortex is critical for remote memory stability. The same infusions 24 h after training, before the memory became reliant on the dorsomedial prefrontal cortex, did not interfere with fear memory at 2 and 30 days posttraining. Thus, cortical DNA methylation is triggered by a learning experience and is a perpetuating signal used by the brain to promote and maintain remote memories.
As stated previously, proteins of Gadd45 family play a role in the locus-specific DNA demethylation events. An electroconvulsive treatment (ECT) of mature neurons in mice dentate gyrus induced expression of Gadd45b gene (Ma et al., 2009). Spatial exploration of a novel environment, known to activate immediate early genes, also led to significant induction of Gadd45b. Most Gadd45b-positive cells also expressed Arc, a classic activity- induced immediate early gene. In vivo injection of an NMDA receptor antagonist abolished the induced Gadd45b and Arc expression. Adult Gadd45b KO mice seemed anatomically normal and exhibited identical NMDA receptor-dependent induction of immediate early genes at 1 h after ECT. The basal densities of proliferating cells in the dentate gyrus were similar in the wild-type (WT) and KO mice, but after ECT there was a 140% increase in the density of such cells in WT mice and only a 40% increase in KO littermates. Reducing expression of endogenous Gadd45b by a small hairpin RNA largely abolished induced, but not the basal proliferation of adult neural progenitors. Thus, Gadd45b plays an essential role in activity-induced, but not basal, proliferation of neural progenitors in the adult dentate gyrus. ECT markedly increased the total dendritic length and complexity of postmitotic neurons in dentate gyrus of adult mice. This ECT-induced dendritic growth was significantly attenuated in KO mice, whereas the basal level of dendritic growth was unaffected. Thus, Gadd45b is also essential for activity-induced dendritic development of newborn neurons in the adult brain. No significant global DNA demethylation in mature neurons of adult dentate tissue was detected after ECT in vivo. However, significant demethylation was found at specific regulatory regions of Bdnf and Fgfl genes. The basal levels of DNA methylation within these regions were similar in WT and KO mice, whereas ECT-induced DNA demethylation of these regions was almost completely abolished in KO mice. Thus, GADD45b is essential for activity-dependent demethylation and late-onset expression of specific secreted factors in the adult dentate gyrus. Gadd45b expression in hippocampus was upregulated during memory consolidation after contextual learning and associative fear training (Leach et al., 2012; Sultan, Wang, Tront, Liebermann, & Sweatt, 2012). Gadd45b KO mice had no significant changes in baseline behavior, but performed significantly lower compared with WT mice in the long-term memory test (24 h after contextual learning conditioning), but did not differ in short-term memory test (1 h after conditioning) (Leach et al., 2012). Somewhat contradictory results were obtained using a rotarod motor learning model: 24 h after training Gadd45b~/~ mice on the hybrid background demonstrated significantly enhanced performance versus WT, implicating Gadd45b in motor memory consolidation, but not initial acquisition (Sultan et al., 2012). In cue-plus-context fear conditioning a significantly enhanced performance of KO mice compared with WT mice was also found at 24 h with mild and moderate, but not robust training. Whatever the reasons for these differences, both studies support the general view that Gadd45b is transcriptionally regulated by experience and also regulates memory capacity. Recently, an excision repair mechanism of DNA demethylation by Gadd45 proteins was demonstrated (Li et al., 2015). Thus, our earlier suggestion that region-specific DNA demethylation at learning can be mediated through DNA repair-like mechanisms (Ashapkin et al., 1983) gains an experimental support.
DNA methylome changes in the dentate gyrus granule neurons in the adult mouse hippocampus in vivo after synchronous neuronal activation by ECS were analyzed at single-nucleotide resolution by using a next-generation sequencing-based method (Guo et al., 2011). Of -200,000 CpGs analyzed, 1892 and 1158 exhibited activity-induced de novo methylation and demethylation, respectively, at 4 h after ECS. Thus, methylation of at least -1.4% CpGs is rapidly modifiable by neuronal activity in the adult brain. Some methylation changes were reversed by 24 h after ECS, but 31% activity-modified CpGs remained at their modified states. Pretreatment of animals with a highly selective NMDA receptor antagonist 3-(2-carboxypiperazin-4-yl)propyl-1-phosphonic acid abolished ECS-induced changes, confirming these modifications to be neuronal activity dependent. Infusion of either 5-aza-dC or RG108, two Dnmt inhibitors with distinct mechanisms of action, abolished activity-induced de novo methylation with no obvious effect on demethylation. Dnmt3a, but not other Dnmts, was upregulated by ECS, suggesting its potential role in the neuronal activity-induced de novo methylation. In contrast, activity-induced demethylation was abolished in the Gadd45b KO mice, consistent with its role in DNA demethylation. Bisulfite sequencing of five representative regions revealed that activity-induced modifications are highly site specific. Similar ECS-induced CpG modifications were observed in directly fluorescence-activated cell sorting-purified NeuN+ postmitotic neuronal nuclei from the dentate gyrus, indicating that the CpG modifications observed are predominantly neuronal and independent of DNA replication. To further determine whether a physiological paradigm of neuronal stimulation could induce DNA methylation changes, adult mice were subjected to a 3-day course of voluntary running, and highly similar changes were observed. Thus, widespread changes in CpG methylation occur in postmitotic neurons in vivo in response to both chemical and physiological neuronal activation. Analysis of genomic location of activity-modified CpGs revealed a striking exclusion of methylation changes in CpG-dense regions for both activity-induced de novo methylation and demethylation; significant resistance of both gene-associated and intergenic CpG islands to activity-induced methylation changes was observed. Thus, the main targets of activity-induced acute modifications are low-density CpGs. Activity-modified CpGs are underrepresented in the 5' upstream regions (putative promoters), exons and 3' downstream regions of the genes, but slightly enriched in introns. Intergenic CpGs (>5 kb away from any known genes) are most susceptible to methylation changes induced by neuronal activity. Although activity-modified CpGs are enriched in intergenic regions, 1819 activity-modified CpGs were mapped to 1518 genes. The methylation changes located in their 5' upstream regions (putative promoters) were modestly but significantly anticorrelated with changes in expression. In contrast, no significant correlation was detected between methylation changes in other gene parts and their expression. Thus, the activity-induced methylation changes may regulate gene expression in a highly context-dependent manner and may have other roles, besides transcription regulation. The 1518 genes associated with the activity-modified CpGs are significantly enriched in genes that are expressed in the brain. The activity-modified CpGs are also preferentially associated with the alternative splicing variants, suggesting a potential role of DNA methylation changes in regulating alternative splicing in neurons. A gene ontology analysis revealed significant overrepresentations of genes involved in synaptic function, protein phosphorylation, neuronal differentiation, and the calcium signaling pathway. Some of them are enriched also in activity-regulated genes at the mRNA level. Surprisingly, multiple genes encoding Notch signaling components exhibited CpG methylation and expression changes. Identification of the Notch signaling pathway as a novel epigenetic target of neuronal activity in mature neurons supports its emerging role in synaptic plasticity and long-term memory (Pierfelice, Alberi, & Gaiano, 2011). The role of DNA modifications in learning and memory is explored in further detail in Chapters 5 and 8.