DNA MODIFICATION AND BEHAVIORAL MEMORY
Methylation of cytosine bases in DNA has been shown to be essential for a striking variety of memory tasks in numerous species (Day et al., 2013; Feng et al., 2010; Lockett, Helliwell, & Maleszka, 2010; Lubin, Roth, & Sweatt, 2008; Maddox et al., 2014; Miller et al., 2010; Miller & Sweatt, 2007; Monsey et al., 2011; Nikitin, Solntseva, Nikitin, & Kozyrev, 2015). Additionally, emerging evidence has revealed a critical role for Tet enzymes and active DNA demethylation machinery in memory formation (Kaas et al.,
2013; Rudenko et al., 2013). This section reviews the role for DNA methylation and demethylation in different memory systems and highlights the potential involvement of these mechanisms in cognitive disease states.
Spatial and fear memory systems
In rodent model systems, a classic test for memory formation and storage is contextual fear conditioning, in which an animal learns that a novel context (conditioning box) is associated with delivery of a mild aversive electric shock through the floor grid. This learning manifests itself as an increase in the time spent immobile (freezing) upon reexposure to the conditioning chamber, which is extremely robust and can last for the lifetime of an animal. This form of learning is dependent, in part, on the hippocampus, where it requires transcription of genetic information and de novo protein synthesis (Alberini, 2008; Alberini, Milekic, & Tronel, 2006; Davis & Squire, 1984). Early studies investigating the role of DNA methylation in learned behavior used the nucleoside analogue DNMT inhibitors 5-aza-deoxycytidine or zebularine to block DNA methylation in the hippocampus either immediately or 6 h after contextual fear conditioning (Miller & Sweatt, 2007). Strikingly, DNMT inhibitors produced a robust impairment in longterm memory (tested at 24 h after initial training), but only when delivered immediately after training. These results are similar to the effects observed with protein synthesis inhibitors and indicated for the first time that DNA methylation was a crucial mechanism in learned behavior. Similarly, infusion of DNMT inhibitors (zebularine or the small molecule inhibitor RG108) directly into the hippocampus before training produced a robust deficit in long-term contextual fear memory (Lubin et al., 2008). Similarly, DNMT inhibition in the lateral amygdala, a brain region critical for cued fear memory, impairs cued fear memory consolidation (Monsey et al., 2011). Importantly, these effects do not seem to be an off-target effect of DNMT inhibitors, as conditional deletion of DNMT1 and DNMT3a in the adult forebrain resulted in impaired contextual fear memory consolidation and impaired spatial memory on the water maze (Feng et al., 2010).
Whereas the initial formation of contextual fear memories is dependent upon the hippocampus, subsequent memory maintenance, remote storage, and retrieval also engages areas of the prefrontal cortex (Bero et al., 2014; Frankland & Bontempi, 2005; Frankland, Bontempi, Talton, Kaczmarek, & Silva, 2004; Lesburgueres et al., 2011; Rajasethupathy et al., 2015; Tse et al., 2011). Consistent with its role in these processes, disruption of the DNA methylation machinery in the cortex has time-dependent effects on memory storage (Miller et al., 2010). Thus, delivery of DNMT inhibitors into the prefrontal cortex shortly after contextual fear conditioning has no effect on subsequent memory maintenance, whereas DNA inhibition in the prefrontal cortex at remote time points after memory formation (eg, 1 month) produces significant degradation of a previously established memory. Similarly, another report demonstrated that long-term object place memory, which is dependent on both the hippocampus and perirhinal cortex, requires DNMT activity in both brain regions (Mitchnick, Creighton, O’Hara, Kalisch, & Winters, 2015). However, further analysis suggested dissociable molecular roles for DNMTs in each brain region. In the hippocampus, knockdown of DNMT3a (but not DNMT1) recapitulated the place learning deficits observed with global DNMT inhibition. Conversely, knockdown of DNMT1 (but not DNMT3a) in the perirhinal cortex impaired long-term memory (Mitchnick et al., 2015). These results suggest that different components of the fear and spatial memory circuitry may engage distinct epigenetic mechanisms at different times (eg, immediately after experiences or at a significant delay) and for different purposes (eg, for recent or remote memory storage).
Consistent with observations that DNA modification is critical for memory formation and storage, several studies have found altered DNA methylation in memory circuits after fear conditioning. For example, formation of contextual fear memories results in hypermethylation of the promoter for the Ppplcc gene, which codes for a subunit of the memory-repressive gene protein phosphatase 1 (Miller & Sweatt, 2007). Conversely, the same experience results in a hypomethylation of the promoter for the memory- promoting gene Reln and varied effects on methylation of unique isoform promoters at the Bdnf gene locus (Lubin et al., 2008; Miller & Sweatt, 2007; Mizuno, Dempster, Mill, & Giese, 2012). In the hippocampus, these effects seem to be highly dynamic, returning to prememory baseline levels as soon as 24 h after fear conditioning (Miller & Sweatt, 2007). In contrast, cortical changes in DNA methylation have been observed to endure for longer periods, consistent with the behavioral role of cortical structures in memory storage. Thus, Ppp3ca, a gene that codes for the catalytic subunit of the memory-suppressing gene calcineurin (a calcium-sensitive protein phosphatase), undergoes promoter hypermethylation and gene repression after fear conditioning, and this change is stable for at least 30 days after memory formation (Miller et al., 2010).
Reports have also revealed a key role for 5hmC in learning and memory. Deletion of Tetl results in a significant reduction in 5hmC levels in the cortex and hippocampus, consistent with the involvement of this enzyme in 5-methyl cytosine (5mC) hydroxylation (Rudenko et al., 2013). Moreover, Tetl knockout mice display impaired extinction of contextual fear and spatial memory, suggestive of an inability adapt to new behavioral contingencies (Rudenko et al., 2013). Conversely, viral overexpression of Tetl in the hippocampus increases 5hmC levels and results in impaired long-term contextual fear memory (Kaas et al., 2013). Together, these reports also suggest that Tet1 is a central regulator of immediate early genes such as Fos, Npas4, Arc, and Egrl, all of which are induced during memory formation. Tet1 overexpression results in an upregulation of these key plasticity/memory genes (Kaas et al., 2013), whereas Tet1 deletion produces a decrease in these transcripts in both hippocampus and cortex (Rudenko et al., 2013).
In contrast to hippocampal Tetl manipulations, Tetl knockdown in the prefrontal cortex does not alter memory formation or extinction (Li et al., 2014). However, expression of Tet3, which is regulated by neuronal activity and extinction training in cortical neurons, is required for normal fear memory extinction. Short hairpin RNA—mediated knockdown of Tet3 in the infralimbic prefrontal cortex (PFC) results in the maintenance of fear- related freezing behavior despite extinction training (Li et al., 2014). To determine how fear learning and extinction altered 5hmC levels in the PFC, Li and colleagues used 5hmC- immunoprecipitation sequencing, which allowed genome-wide characterization of experience-dependent 5hmC changes. Intriguingly, although the initial fear conditioning did not produce substantial alterations in 5hmC content or localization, extinction training induced profound 5hmC reorganization. Whereas 5hmC was predominantly clustered in intronic and intergenic regions in control animals, extinction training produced a shift in 5hmC peaks in favor of 5' untranslated regions and coding sequences in DNA (Li et al., 2014). These findings suggest that a “permissive” epigenetic state is established by Tet3 after extinction learning, possibly as a way to establish an epigenetic memory that will alter future experience-dependent gene transcription (Baker-Andresen, Ratnu, & Bredy, 2013; Li, Wei, Ratnu, & Bredy, 2013; Li et al., 2014). Furthermore, these results highlight the partially overlapping but distinct functions of Tet family members in the genesis and maintenance of fear-related memories.
Another piece of the DNA methylation/demethylation puzzle is the immediate early gene Gadd45b, which despite not being a direct mediator of DNA oxidation, is nevertheless critical for demethylation of cytosine bases in DNA (Guo, Ma, et al., 2011; Ma, Guo, et al., 2009; Ma, Jang, et al., 2009; Niehrs & Schafer, 2012; Sultan & Sweatt, 2013). Gadd45b levels are acutely upregulated in the hippocampus and amygdala in mice after exposure to new contexts, including those paired with shock (Leach et al., 2012; Sultan, Wang, Tront, Liebermann, & Sweatt, 2012). However, the role of Gadd45b in memory formation and maintenance is less clear, as different groups have observed opposite effects of Gadd45b deletion on memory capacity (Leach et al., 2012; Sultan et al., 2012). Sultan and colleagues found that Gadd45b knockout mice exhibit an increase in longterm fear and spatial memory performance, which is consistent with the enhanced hippocampal LTP observed in these mice (Sultan et al., 2012). In contrast, Leach and colleagues reported a decrease in contextual fear memory in Gadd45b knockout animals (Leach et al., 2012). The cause of this discrepancy is not clear, although the use of different background strains and different training conditions may have influenced these results. Nevertheless, these results support a role for the DNA demethylation enzyme Gadd45b in fear memory formation.