From cellular to fear memory: An epigenetic toolbox to remember

Throughout development, the neuronal epigenome is highly sensitive to external stimuli, yet capable of safeguarding cellular memory for a lifetime. In the adult brain, memories of fearful experiences are rapidly instantiated, yet can last for decades, but the mechanisms underlying such longevity remain unknown. Here, we showcase how fear memory formation and storage – traditionally thought to exclusively affect synapse-based events – elicit profound and enduring changes to the chromatin, proposing epigenetic regulation as a plausible molecular template for mnemonic processes. By comparing these to mechanisms occurring in development and differentiation, we notice that an epigenetic machinery similar to that preserving cellular memories might be employed by brain cells so as to form, store, and retrieve behavioral memories.


Introduction
Memories of the past have a fundamental role in life, providing individuals with a framework to structure their present and future behavior on the canvas of previous experiences.Understanding how the brain converts temporary external stimuli into long-lasting changes is a fundamental question of neuroscience, and generations of scientists have been confronted with the challenge to reconcile the transient nature of synapse-based events with the persistence of memory [1e4].During development, programs of gene expression initially established in response to extracellular signals are maintained across multiple rounds of cell division by means of changes to the chromatin structure referred to as epigenetic modifications [5e7].In essence, epigenetic mechanisms e defined as the perpetuation of altered gene activity states in the context of the same DNA sequence [6] e sit at the core of "cellular memories" that arise during lineage development.By ensuring the persistence of cell fate trajectories even in the absence of the initial signals, the newly established epigenetic landscape locks cell identity into a specialized function for life, as Conrad Waddington (1905e1975) had likely already envisioned when portraying the developmental history of a cell in his renowned illustration [8,9].
Similarly, in the adult central nervous system, synaptic inputs initiate signaling cascades that triggerspecific transcriptional programs in response to environmental stimuli [10,11], which, as a growing body of evidence over the past two decades has shown, are regulated by epigenetic mechanisms [12e15].Nevertheless, whether epigenetic mechanisms constitute the molecular equivalent for long-term memory storage is still a matter of debate.This question has been best studied in the context of fear learning (Figure 1), which is one of the longest-lasting forms of memory [16].In particular, Pavlovian fear conditioning e the learning of associations between a neutral and a painful, dangerous, or threatening stimulus (Figure 1a) e offers an ideal paradigm to access the sequence of molecular events underlying long-term memory storage [17].
In this review we summarize the most recent studies in the field of epigenetic regulatory mechanisms of fear learning, focusing on DNA methylation, histone modifications, and higher-order chromatin organization (Figure 1b).In particular, by drawing parallels with the epigenetic principles of cellular memory, we explore the hypothesis that epigenetic mechanisms could be coopted by the nervous system in the adult brain so as to register memories of past experiences using chromatin as a template.Finally, we outline emerging tools and future challenges for pushing the boundaries of epigenetic memory research even further.For advances in the role of histone variants, noncoding RNAs, and epitranscriptional modifications in memory processes, we refer the reader to further publications [18e21].
The Janus-faced property of DNA methylation for memory formation and storage DNA methylation (DNAme) on the fifth position of cytosine (5 mC), which in vertebrates mainly occurs at CpG sites [22], is a key layer of epigenetic regulation during development and throughout life [23].Indeed, for any given cell type, the genome-wide distribution of DNAme often correlates with its transcriptional state and reflects specific differentiation patterns.By virtue of its ability to be maintained across cell replication, DNA methylation is traditionally viewed as the bona fide epigenetic mark for cellular memory [22,24].
Notwithstanding, as emerged in more recent years, DNA methylation is also a highly dynamic mechanism, whose homeostasis reflects the interplay between 3 distinct pathways: a) the establishment of de novo methylation marks, carried out by the DNA methyltransferases (DNMTs) DNMT3A and DNMT3B complexed with DNMT3L; b) the maintenance of existing methylation patterns across DNA replication, ensured by the activity of DNMT1; and c) the erasure of DNA methylation, which predominantly occurs through a cascade of oxidation reactions mediated by the Teneeleven translocation (TET) family of enzymes [23].
By being at once dynamic and stable, DNA methylation attracted the interest of neuroscientists since the early days of the neuroepigenetics field.Indeed, pioneering work demonstrated that in the adult brain patterns of Epigenetic mechanisms in fear learning a) Using Pavlovian fear conditioning in mice to study the "lifespan" of a memory.During encoding, a neutral stimulus, such as a novel context, is paired with an aversive stimulus, such as foot shock.A memory of the event begins to form within the population of neurons with higher excitability at the time of training.The mouse is then returned to its home cage, and the new memory trace is stabilized into a more long-lasting form.Upon subsequent exposure to the same context, the stored memory gets retrieved and the mouse exhibits conditioned fearful responses, such as freezing.For a time-limited period, the retrieved memory becomes labile and needs to be reconsolidated to be stored further.Such further storage can be overcome by re-exposing the mouse repetitively to the contextual cues without foot shock, a process called memory extinction.b) Several levels of epigenetic regulation have been shown to accompany the encoding, storage, retrieval, and extinction of fear memories.Highlighted in the boxes are the epigenetic mechanisms discussed in this review, alongside the most recent studies from the field with icons representing the individual memory phases they focused on.lncRNA, long noncoding RNA; TADs, Topologically Associated Domains, C, Cytosine; 5 mC, 5-Methylcytosine; 5hmC, 5-Hydroxymethylcytosine; 5 fC, 5-Formylcitosine; DNMT, DNA Methyltransferase; TET, Ten Eleven Translocation; HAT, Histone Acetyltransferase; HDAC, Histone Deacetylase; HDACi, HDAC inhibitor; KMT, Lysine Methyltransferase; KDM, Lysine Demethylase.
DNA methylation rapidly change upon neuronal activity in response to physiological and environmental stimuli, including cognitive processes [25,26].Importantly, learning-induced bi-directional changes in DNA methylation levels were observed at promoters of plasticity-related genes (becoming de-methylated) and memory suppressor genes (gaining methylation) starting from 1 h and up to 4 weeks after memory encoding [27e29].In addition, the pharmacological inhibition of DNMTs in the rat hippocampus either immediately after contextual fear conditioning (CFC) or one month later was shown to impinge on the formation and maintenance of long-term memory, respectively [25,27].Likewise, similar impairments in memory-related hippocampal functions were observed upon disruption of DNMT activity using targeted genetic approaches, such as conditional knockout mice or brain region-specific knockdowns [30e33].Finally, the manipulation of the TET pathway also affects memory performances, and it does so in a highly isoform-specific manner: constitutive TET1 knockout mice display normal fear memory formation but impairment of its extinction, whereas the loss of TET2 in adult excitatory neurons enhances the recall of fear memories [34,35].Together, these findings show that manipulating DNA methylation either in the entire brain or in a brain region/cell-type-specific manner changes the genomic responses to neuronal activity as well as memory performances.
Recent technological advances now offer the possibility to investigate the role of DNA methylation by targeting exclusively the neuronal ensemble that encodes a specific memory trace, also commonly referred to as engram.Broadly, an engram can be defined as a group of neurons that 1) are activated by a specific learning experience, 2) become modified by this experience, 3) are later re-activated by the re-exposure to the same experience, and 4) thus store the content of the learned experience [36,37].Importantly, restricting the overexpression of DNMT3a2 e an isoform previously identified to regulate cognitive processes [32] e to engram cells of the dentate gyrus (DG) increased engram reactivation at recall and strengthened fear memory recall [38].Hence, dynamics of DNA methylation within the engram itself are likely crucial for a successful memory retrieval, but where precisely across the genome DNA methylation changes remains to be investigated.
A first attempt in this direction comes from a study showing that, following novel environment exploration, the activated ensemble is characterized by differential methylation specifically occurring at genomic regions harboring bistable DNA methylation states, suggesting that variability in DNA methylation levels may account for why only certain cells react to a given environmental stimulation [39].Intriguingly, this hypothesis resonates well with the observation that during development the oscillatory DNA methylation dynamics typically found at the pluripotency stage facilitate key lineage decisions by allowing the same external signal to have different transcriptional outputs, eventually resulting in the emergence of different cell lineages starting from the same pool of stem cells [40].Along these lines, it would be interesting to explore whether and the extent to which the balance in DNA methylation governs the allocation of fear memory in the brain, especially in light of the observation that DNMT3a2 overexpression in a sparse and random population of cells prior to CFC did not bias the recruitment of individual neurons to the fear memory trace [38].Whether memory allocation relies on other DNMTs rather than DNMT3a2, or on other epigenetic marks rather than DNA methylation e with dynamic posttranslational histone modifications (PTMs) being a likely candidate e remains to be investigated.

Histone acetylation as mnemonic markers on the chromatin
With a plethora of enzymatic factors involved in their deposition, removal, or reading, histone posttranslational modifications (PTMs) serve as a signal integration template between genes and the environment in complex biological processes ranging from stem cell differentiation to immunological responses [6,41].In the adult brain, early efforts focused on the role of histone acetylation, with pioneering work showing that learning triggers the recruitment of histone acetyl transferases (HATs) and the increase of histone acetylation at the promoter of synaptic plasticity and memory-associated genes, whilst histone deacetylases (HDACs) such as HDAC2 negatively affect memory processes [42e44].
In the last five years, several studies further refined the role of histone acetylation in memory formation and storage, using either pharmacological, behavioral, or epigenetic editing tools.In parallel, three important papers started to shed light on the upstream regulatory mechanisms of histone acetylation, which revealed an important metabolic contribution (Box 1).Multiple research efforts have explored the consequences of manipulating acetylation levels e either by directly inhibiting the activity of HDACs with HDAC inhibitors (HDACis) or by replacing the pharmacological intervention with a pure behavioral alternative e with the goal of ameliorating memory and rescuing cognitive impairments [45,46].Indeed, several different types of HDACis were found to improve performances in CFC and extinction learning, as well as to rescue memory in mouse models of Alzheimer's disease [47e49].Mechanistically, it is interesting to highlight that despite being administered systemically and lacking any inherent target specificity, HDACi treatment elicits electrophysiological, transcriptional, and epigenetic changes only when applied jointly with CFC [43,47].In particular, HDACi treatment was found to enhance H3K27ac levels at genes involved in synaptic communication that were already acetylated by CFC, suggesting that the amelioration of behavioral responses is likely due to a reinforcement action by the HDACi [47].To explain this phenomenon, the idea of cognitive epigenetic priming has been proposed [46], purposely evoking a widely used concept in developmental studies, epigenetic priming [50,51].Epigenetic priming describes the state of chromatin regions in pluripotent cells that are neither silenced nor fully active, but are instead epigenetically bookmarked for rapid gene activation in response to signaling and developmental cues [50,51].In developing sensory neurons, for example, immediate early genes (IEGs) are embedded into a unique chromatin signature carrying H3K27ac on promoters but repressive H3K27me on gene bodies [52].Such epigenetic signature prevents inappropriate transcription in response to non-relevant stimuli, but at the same time primes IEGs for fast induction following appropriate stimuli [52].Similarly, glucocorticoid exposure during hippocampal neurogenesis induces long-lasting changes in DNA methylation that prime target genes for an enhanced responsivity to future stress exposures [53].Although based on only a few lines of evidence, it appears that both histone acetylation and DNA methylation are epigenetic mechanisms which the brain appears to have co-opted from development for its adult functioning.
These advancements notwithstanding, a more finegrained level of investigation is still missing, namely addressing the causal link between the epigenetic modification per se and the storage of fearful experiences as long-term memory.Thanks to the development of transcriptional and epigenetic engineering technologies, it has in the meantime become possible to achieve a precise transcriptional and epigenetic control only at genomic sites of interest.A first example along these lines used engineered zinc finger proteins (ZFPs) fused to the p65 transcriptional activation domain to upregulate the expression of the Cdk5 gene in the mouse hippocampus, resulting in long-term fear memory attenuation [54].Furthermore, in a rat model of adolescent alcohol abuse, CRISPR-based epigenomic editing was shown to improve anxiety by targeting histone acetylation markers at a specific enhancer that responds to synaptic activity [55].To date, epi-editing technologies have not been exploited yet in fear memory studies, and whether the sitespecific manipulation of histone acetylation in a defined brain region e or even, only in its engram cells e would affect memory performances is one of the next open questions in the field.

Histone methylation: Bivalency at play
Similar approaches could also be used to investigate the contribution of other histone PTMs that have been less extensively studied in the neuroepigenetics of memory such as histone methylation.Considered to be more durable and stable relative to histone acetylation, the effects of histone methylation on gene expression depend on the specific residue and the degree (i.e., mono, di, tri) of methylation.Namely, H3K4me3 is associated with genes that are either poised for activity or actively transcribed, whereas H3K27me3 is a marker of repressed chromatin, and H3K4me1 of silent or active enhancers [56,57].For activating methylation marks, global levels of H3K4me3 were found to be elevated in the hippocampus and broader domains of H3K4me3 established at the promoters and super-enhancers of learning-associated genes by the histone lysine methyltransferases (KMTs) KMT2A and KMT2B [58,59].For repressive methylation marks, an siRNA-mediated knockdown of the KMT EZH2 in rat hippocampus was shown to reduce fear memory retrieval by affecting H3K27me3 levels [60].
So far, patterns of H3K4me3 and H3K27me3 have been investigated separately from each other.Yet, whether and the extent to which these markers co-exist in neurons activated by learning would be highly interesting to investigate in light of embryonic stem cell (ESC) differentiation.There, the promoters of key developmental genes are simultaneously enriched for both activating (H3K4me3) and repressive (H3K27me3) marks forming so-called bivalent domains.These have Box 1.The role of the metabolic epigenetic axis in memory.
Acetyl-CoA is the metabolic substrate of HATs to generate histone acetylation by transferring the acetyl-group from acetyl-CoA to the lysine residues of histones [88].In neuronal nuclei, circulating acetate derived from alcohol consumption was found to be captured and turned into acetyl-CoA by the chromatin-bound acetyl-CoA synthetase 2 (ACSS2).In particular, ACSS2 has been shown to bind to the promoter of memoryrelated genes alongside the HAT CREB-binding protein (CBP) in the mouse hippocampus, suggesting a key role in the regulation of histone acetylation at these genomic sites [89,90].Indeed, upon CFC, mice constitutively lacking ACSS2 show a reduction in the levels of H3K9ac and H4K5actwo markers associated with learningand in the expression of activity-dependent genes, as well as a deficit in the formation of longterm fear memory [89].Noteworthily, the same effects were also observed when blocking ACSS2 using a small molecule inhibitor via systemic administration, a result that showcases how the acetyl-CoA pathway could be amenable for pharmacological interventions targeting persistent memories of traumatic events [91].In the future, it will be interesting to extend these lines of research to other metabolic substrates that also fuel epigenetic mechanisms involved in memory processes, for example S-adenosylmethionine (SAM) for DNA methylation.
been proposed to maintain genes in a "poised" state, maintaining repression in the absence of differentiation signals, but at the same time allowing for rapid activation in response to external stimuli via the removal of H3K27me3 [61,62].As bivalent domains are also found on specific genes in adult brain cells [63,64], it is tempting to speculate that a similar "poising" mechanism also occurs for rapidly induced learning and memory genes.Emerging methods for single-cell sequencing of histone marks may shed light on this intriguing possibility in the future.

Towards an integrated view: Multi-omics profiling of the memory engram
A crucial step towards understanding the molecular foundations of long-term memory storage is e the nowadays possible e integration of omics approaches in cell type-specific manner, in particular of engram cells [65].Recently, several studies have begun to profile different modalities of the engram's epi-transcriptional landscape throughout various memory stages e from the initial learning event to the preservation of the fear experience over time until its final retrieval e, producing insightful results [66e68].
By combining activity-dependent genetic labeling of engram cells in the mouse hippocampus with fluorescent-activated nuclei sorting, Marco and colleagues analyzed transcriptional changes, chromatin accessibility, and three-dimensional (3D) genome architecture over the lifespan of a 5-day-long fear memory [68].Although memory formation led to an extensive chromatin reorganization characterized by an increase in enhancer accessibility, these gained-open regions did surprisingly not match changes in gene expression, indicating a potential priming mechanism.Then, during the consolidation period, the newly established epigenetic landscape was maintained and further stabilized by means of new promoter-enhancer interactions which in turn led to a modest transcriptional activity, likely functional to facilitate memory expression at the time of recall (Figure 2a).When the memory ensemble was reactivated by memory retrieval 5 days post-encoding, primed engram cells finally underwent more robust transcriptional changes, resulting in the upregulation of genes involved in protein synthesis and synaptic morphogenesis [68].Although the temporal stability of such epi-transcriptional program has not been addressed beyond the 5-day experimental setup, it is tempting to speculate that it could also be maintained over longer periods of time.Indeed, when a similar study investigated the transcriptional signature of prefrontal cortex engram cells activated by the recall of a remote fear memory at single-cell resolution, profound alterations of gene expression signatures were found up to 14 days after encoding, both in neurons as well as in astrocytes and microglia cells [66].
A further important confirmation that stimulus-induced epigenetic modifications in the brain persist over time comes from the transcriptomic and epigenomic profiling of hippocampal neurons activated by a novel context exploration or kainic acid injection [67].In this experimental model, neuronal activity triggered rapid changes in both gene expression and chromatin organization at the level of enhancer-promoter interactions and transcription factor (TF) binding site accessibility, but at later times, only the epigenetic alterations remained [67].
Drawing parallels between epigenetic mechanisms for memory storage in the brain and in other systems, the sequence of molecular events occurring in neuronal cells activated by a fearful experience, a novel environment, or elevated neuronal activity are not only similar with one another, but also with intracellular responses upon differentiation or inflammation observed in immune cells [69,70].For example, treatment of murine epidermal stem cells with imiquimod (IMQ) e a known inflammatory agent e induced fast transcriptional changes and increased chromatin accessibility at specific enhancers, which acquired H3K4me1 and H3K27ac marks [70].Following IMQ withdrawal, transcriptional activity returned to baseline, while the open chromatin configuration remained, likely through the coordinated action of histone markers and homeostatic TFs (Figure 2b).Upon further inflammatory challenge, a robust transcriptional response was promptly reinstated leading to enhanced tissue inflammation [70], akin to a fear memory recall event triggering epi-transcriptional program for longterm memory storage [66,68].Thus, it appears that to retain a memory of inflammation epidermal stem cells rely on a similar epigenetic toolbox as used by brain cells to remember a fearful experience (Figure 2).What dictates the specificity of the outputs in these cases remains unknown, but likely lies in the interplay between the epigenetic landscape and the transcriptional state of each individual cell type.

Future strategies to disentangle the epigenetic basis of memory function
The application of epigenetic research tools to fear conditioning or other learning paradigms has significantly changed our interpretation of how the brain stores and retrieves memories of past experiences.Traditionally thought to be an exclusive property of synapses, it is now clear that mnemonic processes also have a significant impact on chromatin.Here, we have showcased how modulations of the epigenetic landscape at the level of DNA methylation, histone modifications, and 3D chromatin structure parallel and may indeed underlie the cellular processes behind fear memory formation, persistence, and retrieval.To push the boundaries of neuroepigenetics research on memory even further, two major challenges must be met in the future: 1) achieving a composite and refined portrait of the epigenetic machinery for long-term memory; 2) addressing causality by means of precise functional epigenetic manipulations.
Memory processes are highly dynamic, involving different cell types and multiple brain areas across an extensive timespan [71e74].Nevertheless, most studies to date have only captured snapshots of the epigenetic mechanisms of memory, focusing on individual chromatin modifications and/or brain regions without accounting for cell type diversity and temporal resolution.State-of-the-art technologies such as singlecell bisulfite sequencing (scBS-seq), single-cell Cleavage Under Targets and Tagmentation (scCUT&Tag), single-cell assay for transposase accessible chromatin (scATAC-seq), and single-cell Hi-C (scHi-C) are now enabling to profile, respectively, DNA methylation, histone PTMs and TFs occupancy, chromatin accessibility and 3D chromatin organization from individual cells [75e78].For some of these methods, the option to combine transcriptomic and epigenomic profiling of the same cell in a single readout is already technically possible, and by so-called spatial-omics approaches, even preserving a cell's positional information within the brain can be achieved [79e82].With these possibilities on the horizon, research in neuroepigenetics is poised to achieve hitherto unmatched levels of precision.ened, but chromatin is kept in a poised state with the retention of H3K4me1 and some cases H3K27ac.Once the inflammation memory is later recalled by a second inflammatory event, primed chromatin sites become rapidly transcriptionally activated.Although specific histone PTMs and chromatin conformation changes were not assessed in a and b, respectively, it is interesting to note that H3K4me1 and H3K27ac have been found at the boundaries of 3D chromatin loops in the adult brain, where they co-regulate the expression of genes important for spatial memory [87].TF, transcription factor.
As the resolution of omics techniques continues to improve, the need arises to also infer causality between epigenetic states, transcriptional activity, and behavioral responses by means of functional validation.The explosion of epigenomic editing tools, and in particular those based on CRISPR/dCas9 e an enzymatically inactive (dead) variant of Cas9 e opens up the possibility of testing the relevance of specific chromatin perturbations at the genomic site(s) of interest, both in vitro and in vivo [83].Versions of the system already exist where dCas9 has been fused to HATs or HDACs, Tet1 or DNMT3a, writers and erasers of H3K4me and H3K9/K27me, to mediators of chromatin looping or TF domains, and it is likely that dCas9 combining different epigenetic effectors at once will be engineered soon [84,85].Moreover, a further level of precision can be achieved by controlling the activity of the dCas9-based systems both in time and space, for example by using cell type-specific expression constructs or optogenetically and chemically inducible approaches [86].Doing so will allow for the interrogation of the mnemonic capacity of specific loci within specialized cell types at precise moments pre-and post-learning.At the same time, such experimental efforts will constitute precious resources to further explore the therapeutic potential of epigenetic mechanisms as biomarkers and drug targets for memory disorders.
In conclusion, it is exciting to think outside the box of the adult central nervous system and notice that, in order to store memories of our past, the brain might coopt similar molecular mechanisms e epigenetic in nature e that maintain cellular memory throughout development in other organs.To future research, the challenge of exploring this captivating hypothesis further.

*
. Chen MB, Jiang X, Quake SR, Südhof TC: Persistent transcriptional programmes are associated with remote memory.Nature 2020, 587:437-442.Long-term contextual fear memory induces complex gene expression programs in engram cells from medial prefrontal cortex.The activityspecific transcriptional alterations persist for weeks after the learning event and occur not only in neuronal cells but also in astrocytes and microglia cells.
Albert J, Lipinski M, Lopez-Cascales MT, Rowley MJ, Martin-Gonzalez AM, Del Blanco B, Corces VG, Barco A: Immediate and deferred epigenomic signatures of in vivo neuronal activation in mouse hippocampus.Nat Neurosci 2019, 22:1718-1730.Status epilepticus upon kainic acid stimulation triggers robust transcriptional and chromatic changes in hippocampal excitatory neurons.While transcriptional profiles return to baseline levels within 24 h, some chromatin changes and interactionsin particular, those driven by AP1remain up to 48 h after neuronal activation.68 * * .Marco A, Meharena HS, Dileep V, Raju RM, Davila-Velderrain J, Zhang AL, Adaikkan C, Young JZ, Gao F, Kellis M, et al.: Mapping the epigenomic and transcriptomic interplay during memory formation and recall in the hippocampal engram ensemble.Nat Neurosci 2020, 23:1606-1617.In engram cells of the hippocampus, memory encoding is characterized by increased chromatin accessibility at enhancers without cooccurring transcriptional changes.The newly established chromatin landscape is maintained during memory consolidation and results in the upregulation of specific synapses-related genes at the time of memory recall.69.Ostuni R, Piccolo V, Barozzi I, Polletti S, Termanini A, Bonifacio S, Curina A, Prosperini E, Ghisletti S, Natoli G: Latent enhancers activated by stimulation in differentiated cells.Cell 2013, 152:157-171.70 * .Larsen SB, Cowley CJ, Sajjath SM, Barrows D, Yang Y, Carroll TS, Fuchs E: Establishment, maintenance, and recall of inflammatory memory.Cell Stem Cell 2021, 28:1758-1774.e8.Characterization of the changes in chromatin accessibility, histone modifications and transcription factor binding occurring during an inflammation process, from the initial responses to its attenuation up to a second inflammatory event.