Abstract

Layers of memory

Epigenetic changes are “mitotically and/or meiotically heritable changes in gene function that cannot be explained by changes in the DNA sequence”, leading to the propagation of heritable changes in phenotype [1]. Epigenetic regulation can be either relatively stable (such as X inactivation, imprinting, silencing or boundary activities) or dynamic [2–4]. Where epigenetic regulation is dynamic, it is often described as epigenetic memory: a heritable change in gene expression or behavior that is induced by a previous stimulus. The stimulus can be either developmental or environmental. Memory occurs by multiple mechanisms, but often requires chromatin-based changes such as DNA methylation, histone modifications or incorporation of variant histones [5]. DNA methylation can template its own inheritance through methylation of hemimethylated sites following DNA replication by maintenance DNA methylases [6]. However, most histone modifications are not heritable and the extent to which any can template their own inheritance is still somewhat contentious. For the purposes of this review, we will focus on the epigenetic mechanisms that require chromatin changes, but we do not mean to suggest that it is clear that these changes are the source of epigenetic information and inheritance.

Here we briefly discuss three types of epigenetic memory that utilize related mechanisms over different time scales: 1) cellular memory, mitotically heritable transcriptional states established during development in response to developmental cues, 2) transcriptional memory, mitotically heritable changes in the responsiveness of organisms to environmental stimuli due to previous experiences and 3) transgenerational memory, meiotically heritable changes in the gene expression and physiology of organisms in response to experiences in the previous generations (Fig. 1) [7–11].

Open in a separate window

Figure 1

Types of epigenetic memory

A. Cellular memory: During early development, tissue-specific transcription factors establish different transcriptional programs. These programs are maintained through mitotic cell division later in development through the action of epigenetic regulators such as the Trithorax, which catalyzes methylation of histone H3, lysine 4 to promote sustained expression, and Polycomb, which catalyzes methylation of histone H3, lysine 27 to promote stable repression.

B. Transcriptional memory: Environmental changes induce changes in gene expression. Following such an experience, some genes remain poised for faster reactivation for several generations. Upon gene activation, genes can experience a more robust secondary transcriptional response. Nuclear pore proteins (Nups) and specific histone marks at the nuclear pore (yeast) or nucleoplasm (higher eukaryotes) are required to establish and inherit this poised state.

C. Transgenerational Memory: The parental experiences can impact the behavior of the offspring. In this example, environmental stress reduces maternal LG-ABN, which reduces the stress-tolerance of pups into adulthood. Low LG-ABN alters the expression of stress regulators, leading to greater stress sensitivity. This altered gene expression requires changes in DNA methylation and histone acetylation and is inherited in future generations (F1 and F2).

Cellular Memory

Developmental signals induce changes in gene expression and chromatin structure [7]. Such changes define cell identity and potential for differentiation and they can be maintained through subsequent cell divisions, even in the absence of these signals. One of the best examples of this phenomenon is the developing Drosophila embryo, in which expression of homeotic genes is established by transient expression of the segmentation transcription factors. After these factors turn over, the expression patterns of many genes, including the homeotic genes, is maintained through many cell divisions. This cellular memory requires the Trithorax and Polycomb group (PcG) proteins (Fig. 1A) [12,13]. The Trithorax complex mediates methylation of histone H3 on lysine 4 (H3K4me) and is required to maintain genes in an active state [12,13]. The PcG proteins have two biochemically characterized repressive complexes: PRC1 and PRC2. PRC2 methylates histone 3 lysine 27 (H3K27me) at genes targeted for silencing. PRC1 binds H3K27me and induces spreading of structural changes in the chromatin [14,15]. This histone mark has been proposed to act as a repressive “bookmark” during mitosis where it is maintained through cell division and transmitted through DNA replication in the absence of the initial stimuli [16,17].

Bookmarking can also contribute to the maintenance and inheritance of active chromatin states. For instance, when cells enter mitosis, transcription is abruptly repressed and RNA polymerase II (RNAPII) is displaced during chromatin condensation [18,19]. The RNAPII transcription factor TFIID is largely removed from gene promoters by phosphorylation of histone H3 on threonine 3 (H3T3) during mitosis. However, TFIID is selectively retained in the promoters of certain active genes, bookmarking them for future expression in G2 [20,21]. The mechanism proposed for this phenomenon involves high levels of methylation of H3K4, which inhibits H3T3 phosphorylation. This allows TFIID and the phosphatase PP2 to be retained at these promoters through mitosis, counteracting chromatin compaction [22]. The role of histone modifications in bookmarking may be related to reader-protein binding complexes that contribute to epigenetic inheritance and the reestablishment of post-mitotic transcription programs [17,23,24]

Development can also be regulated by environmental conditions. For example, flowering in certain plants requires previous exposure to cold, a phenomenon called vernalization. During this process, plants become competent to flower only after prolonged exposure to the cold winter, an epigenetic change that ensures that flowering occurs under favorable conditions in spring [25,26]. In Arabidopsis thaliana, flowering is controlled by the expression of the FLOWERING LOCUS C (FLC) gene, which encodes a transcription repressor that prevents flowering. In the autumn, FLC expression is high, preventing flowering. Extended exposure to cold represses FLC transcription and the FLC mRNA gradually decreases during winter and stays low as the temperatures rise in the spring [27–29]. Extended cold induces the VERNALIZATION INSENSITIVE 3 (VIN3) gene, which interacts with the Polycomb homologue VERNALIZATION 2 (VRN2) to promote methylation of H3K27 at the FLC locus and reduce its expression [29,30]. Methylation of H3K9 and H3 de-acetylation are also required for full repression of FLC [29]. This process also involves the noncoding RNAs COLDAIR and COOLAIR [31–33]. These two noncoding RNAs have been proposed to bind and recruit PcG to FLC. Thus, environmental cues can impact developmental timing through a mechanism involving chromatin modification. The repression is mitotically stable through a large number of cell divisions in the absence of the inducing signals creating a new epigenetic state [25,26]. This suggests that the mitotic epigenetic memory involves positive feedback loops where the repressive chromatin modifications recruit the chromatin modifying complexes themselves to maintain FLC in a repressive state [25,26,34]. Also, it is important to note that not all plants species exhibit the same vernalization behavior. For instance, some species flower during cold exposure because the chromatin changes are not mitotically maintained, suggesting this to be an adaptive response to climate changes [25,26].

Transcriptional memory

Cells and organisms must respond to changes in their environment to adapt and survive. In general, such responses involve changes in transcription. Work from several organisms suggests that these responses can be quantitatively or qualitatively altered by previous experience through epigenetically heritable mechanisms. For example, previously expressed genes are frequently primed for re-activation, a phenomenon called transcriptional memory (Fig. 1B). This mechanism requires changes in chromatin structure and a physical interaction with nuclear pore proteins. Such a mechanism allows cells to mount a more rapid or robust transcriptional response to an environmental challenge that they have previously experienced [35,36].

The nuclear pore complex (NPC) is a large molecular portal that penetrates the nuclear envelope to facilitate nuclear-cytoplasmic trafficking. Nuclear pore proteins interact with particular parts of the genome in numerous species [37–39]. The targeting of such loci to the yeast NPC involves transcription factor binding to cis-acting DNA “zip codes” and promotes both stronger transcription and epigenetic states such as chromatin boundaries and transcriptional memory [35,38,40–47].

One well-established model for transcriptional memory at the NPC is the inducible INO1 gene in budding yeast. Upon activation, the INO1 gene moves from the nucleoplasm to the nuclear periphery through interaction with the nuclear pore complex (NPC) [48]. After repression, INO1 remains associated with the NPC for 3–4 future generations [35,44,45]. Thus, maintenance of recently repressed INO1 at the NPC represents an epigenetic state. While at the nuclear periphery, recently repressed INO1 is poised for transcriptional reactivation. The positioning of INO1 at the NPC is mitotically inherited, although the molecular mechanism responsible for re-establishing it after cell division still is not fully understood. The yeast GAL genes are also maintained at the nuclear periphery after repression exhibiting a similar faster rate of transcriptional reactivation [35]. However, unlike INO1 memory, GAL gene memory involves formation of an intragenic loop in association with the NPC-associated factor Mlp1 [49].

Two different mechanisms target INO1 to the NPC, one when the gene is active and another when the gene is recently repressed. The targeting of active INO1 to the NPC is controlled by two cis-acting DNA zip codes called Gene Recruitment Sequences (GRS) [44]. However, the targeting of recently repressed INO1 to the NPC is controlled by a different cis-acting DNA zip code called the Memory Recruitment Sequence (MRS) [45]. Mutation of the MRS sequence specifically disrupts the interaction of recently repressed INO1 with the NPC and causes a strong defect in the rate of reactivation [45,46]. This mutation affects neither the initial activation rate of INO1 nor its localization at the nuclear periphery when active, suggesting that two different molecular mechanisms promote targeting of active INO1 versus recently repressed INO1 [45].

In addition to the MRS element, INO1 memory requires the interaction with components of the NPC and particular chromatin changes (Fig. 2A) [35]. Mutations in a number of nuclear pore or NPC-associated proteins impact localization of INO1 to the nuclear periphery [45]. Several of these proteins, including Nup100, are specifically required for localization of recently repressed INO1 [45]. Based on ChIP experiments, Nup100 physically interacts with the INO1 promoter after repression and nup100Δ mutants exhibit slower INO1 transcriptional reactivation rates [45].

Open in a separate window

Figure 2

Transcriptional Memory is conserved between organisms

A. Model for yeast INO1 transcriptional memory. Upon repression, INO1 remains associated with the nuclear pore complex in the population of cells for three to four generations. This interaction requires a cis-acting DNA element called the MRS and leads to an altered chromatin state involving the incorporation of H2A.Z and dimethylation of histone H3, lysine 4 by COMPASS (H3K4me2). These changes, and effector complexes such as Set3C are required to allow binding of poised, preinitiation RNA Polymerase II to the promoter. B. A conserved mechanism primes IFN-γ-induced genes for faster or stronger reactivation in HeLa cells. After exposure to IFN-γ, the nuclear pore protein Nup98 (homologous to yeast Nup100) binds to the promoters of genes with memory. This interaction occurs in the nucleoplasm, near PML bodies. Similar to the yeast INO1 gene, genes that exhibit IFN-γ memory maintain H3K4me2 and poised RNAPII on their promoters. These similarities suggest the possibility of recruiting similar chromatin effector complexes.

The ultimate output of INO1 memory is to allow binding of the preinitiation complex and RNAPII to the promoter (Fig. 2A) [45,46]. After repression, RNAPII remains bound to the INO1 promoter in a poised, preinitiation form that is not phosphorylated on the carboxy terminal domain. Mutations that disrupt transcriptional memory (eg. mutations in the MRS, nup100Δ) lead to loss of RNAPII from the recently repressed INO1 promoter [45,46]. This suggests that INO1 transcriptional memory may prime INO1 for reactivation by bypassing the rate-limiting step in transcriptional activation, recruitment of RNAPII.

Another type of epigenetic memory that leads to greater stress resistance after exposure to a previous stress is also connected to the NPC. Many yeast genes induced by oxidative stress are activated more rapidly in cells that have previously experienced salt stress [47]. This effect persists for up to four generations after the initial stress. The nuclear pore protein Nup42 is required for this faster induction and loss of Nup42 leads to greater sensitivity to oxidative stress [47]. Interestingly, the promoters of genes that exhibit salt-induced memory are enriched for a DNA element similar to the INO1 MRS [45,47]. Thus, various examples of transcriptional memory in yeast involve interactions with the nuclear pore complex and similar cis-acting elements.

The histone variant H2A.Z is highly enriched at gene promoters and has conserved roles in transcriptional regulation [50]. In budding yeast, H2A.Z can have both a role in transcriptional gene activation or repression [51–53]. H2A.Z nucleosomes are often found in regions flanking a nucleosome-free region (NFR) near the transcription start site [51]. In fact, a nucleosome-free region is sufficient in certain contexts to induce H2A.Z incorporation [54]. This suggests that H2A.Z can affect nucleosome positioning and chromatin structure, perhaps explaining the divergent, gene-specific effects of H2A.Z on transcription [53].

H2A.Z is found in most inducible promoters, and facilitates faster induction [52,55]. For instance, oleate-reponsive genes require H2A.Z for rapid and robust activation [52,56]. Hence, the incorporation of non-canonical histone variants such as H2A.Z can produce specialized chromatin domains that influence the rates of transcriptional induction.

H2A.Z is required for INO1 transcriptional memory; mutants lacking H2A.Z do not retain INO1 at the nuclear periphery after repression, fail to recruit RNAPII and show a defect in the rate of reactivation [35,45]. After repression of INO1, H2A.Z is incorporated into a particular nucleosome in the promoter [35,45]. Incorporation of H2A.Z into the recently repressed INO1 promoter requires the MRS sequence and the Nup100 protein [45]. Furthermore, insertion of the MRS sequence alone at ectopic sites leads to incorporation of H2A.Z, suggesting that the MRS sequence is sufficient to promote incorporation of H2A.Z [35,45]. Thus, specific chromatin structural changes at the INO1 promoter are required for INO1 transcriptional memory.

Aspects of INO1 transcriptional memory are evolutionarily conserved and also occur in human cells. The interferon gamma (IFN-γ)-induced class II major histocompatibility gene HLA-DRA is much more rapidly and robustly induced if cells have previously been exposed to IFN-γ [36]. This response persists for at least four mitotic generations and is associated with dimethylation of H3K4 in the HLA-DRA promoter [36]. This behavior is widespread; of the ~650 genes that are induced by IFN-γ, ~250 exhibit faster or stronger activation upon subsequent treatment with IFN-γ [46]. The molecular mechanism of IFN-γ-induced memory is very similar to the molecular mechanism of transcriptional memory of the INO1 gene in yeast [45,46]. HLA-DRA memory requires a physical interaction with the nuclear pore protein Nup98, which is homologous to yeast Nup100 [45,46]. As in yeast, a poised, preinitiation form of RNAPII binds to promoters of genes that exhibit IFN-γ memory (Fig. 2B). However, unlike yeast INO1, the HLA-DRA gene does not localize at the NPC, but instead interacts with Nup98 in the nucleoplasm [46,57,58](Fig. 2B).

Both INO1 memory in yeast and IFN-γ-induced memory in HeLa cells require dimethylation of H3K4 in promoters [36,46]. In yeast, mutations that block ubiquitylation of histone H2B or methylation of H3 lysine 4 also disrupt INO1 transcriptional memory. Furthermore, the Set3C histone deacetylase complex, recognizes H3K4me2, is also required for INO1 transcriptional memory, suggesting that it might be an important reader of this mark (Fig. 2).

Transgenerational memory

Meiosis, gametogenesis and embryogenesis are associated with dramatic global changes in chromatin structure and transcription [10]. Despite this, previous experiences can impart epigenetic changes in gene expression in subsequent generations. For example, in Drosophila, the transcription factor dATF-2 binds to heterochromatin and is required for H3K9 methylation [59]. Upon heat or osmotic shock, Stress-Activated Protein Kinases (SAPKs) phosphorylate dATF-2 and the protein is released from heterochromatin. This leads to loss of H3K9me2, and increased transcription [59]. The change in localization of dATF-2 is transmitted to the next generation [60,61]. If the stress is applied over multiple generations, the effect persists over several additional generations before gradually returning to the original state [59]. This suggests that the transgenerational memory effect is adaptive, helping the organism adjust to the environment, and that the new induced chromatin state is unstable [59]. Disruption of heterochromatin may improve tolerance to variable or challenging environmental conditions. The heterochromatin that is lost during stress might be reestablished through reprograming in germ cells or early embryogenesis [59]. dATF-2 phosphorylation and function is conserved in mammals and Schizosaccharomyces pombe, suggesting that epigenetic reprogramming by environmental stimuli is a highly conserved mechanism [59,62,63].

Transgenerational epigenetic memory maintains proper germline development in Caenorhabditis elegans. During early embryogenesis, germline blastomeres cells (P cells) contain high levels of H3K4me2 [64]. P cells undergo asymmetric cell division into primordial cells and Z2/Z3 cells and the primordial cells lose the H3K4me2 histone mark at specific target genes. This mark is also removed in the germline pole cells in Drosophila, suggesting a similar epigenetic phenomenon [64]. In C.elegans, the protein responsible for the erasure of the specific H3K4me2 histone mark is the demethylase protein SPR-5, which belongs to the LSD1/KDM1 family of demethylases [65,66]. Loss of Spr-5 results in germline mortality and a high level of H3K4me2, stably maintained across the germline in future generations [67,68]. The propagation of this mark results in the misregulation of gamete and meiosis-specific genes [67,68]. In addition, the absence of SPR-5 activity causes a transgenerational accumulation of H3K4me2 that eventually saturates a second mechanism of chromatin remodeling in the Z2/Z3 cells, resulting in the increased retention of H3K4me2 in the primordial cells increasing future generations sterility [67,68]. This suggests that the resetting of H3K4me2 is required to prevent inappropriate transgenerational epigenetic memory from being transmitted from one generation to the next and it is essential for germline maintenance [67,68].

C. elegans individuals that are wild type descendants of ancestors that are mutant for the highly conserved COMPASS H3K4 methyltransferase increase their longevity up to three generations compared to descendants from pure wild type worms [69–71]. In addition, when genetically wild type males are crossed with COMPASS mutant hermaphrodites, they are longer lived for up to three generations compared to wild type; this longer lifespan is dependent on the H3K4me3 demethylase retinoblastoma binding protein related (RBR-2). This epigenetic memory appears to be specific to the epigenetic changes of the COMPASS complex; manipulation in other longevity related pathways such as insulin or mitochondrial signaling and other effector complexes do not increase transgenerational inheritance of longevity [69,70]. These findings suggest that manipulation of chromatin structure and histone modifications such as methylation of H3K4 can impact the physiology and development of future generations [72].

Epigenetic phenomena also have important impacts on human morphology, physiology and health. For example, in many organisms including humans, genomic imprinting, the exclusive expression of a single parental allele (maternal or paternal), can lead to phenotypic differences between genetically identical individuals [73]. Imprinted genes have parental allele-specific epigenetic modifications that are maintained following fertilization when the genome is reprogrammed. Therefore, mutations in imprinted genes lead to phenotypes if the mutant allele is expressed. For example, a mutation in a paternally-expressed gene will lead to a phenotype in offspring who inherited it from their fathers, but not those who inherited it from their mothers. Such “epimutations” lead to human diseases such as Prader-Willi syndrome or Angelman syndrome [11,74]. Thus, the parental origin of individual chromosomes must impart heritable epigenetic marks that survive meiosis, fertilization, embryogenesis and development.

Environmental factors experienced in one generation can impact the behavior of unborn offspring in mammals (Fig. 1C). For example, environmental stresses such as high exposure to predators reduces maternal care in female rats, as measured by licking/grooming and arched-back nursing (LG-ABN; Fig. 1C) [75]. Pups reared under conditions of low maternal protection and LG-ABN are more fearful and more sensitive to environmental stresses. These pups exhibit less LG-ABN with their offspring than normal pups, even in the absence of environmental stressors and this behavior is passed on to future generations (Fig. 1C) [75].

Pups reared under low LG-ABN conditions show reduced levels of glucocorticoid receptor (GR) in the hippocampus, which controls expression of approximately three hundred genes and is responsible for dampening the stress response [76,77]. Low LG-ABN leads to a decrease in the binding of the transcriptional activator NGFI-A, an increase in DNA methylation and histone deacetylation and reduced expression of the GR gene over the first week of life [76,78]. These epigenetic marks persist and dictate GR expression for the rest of the animal’s life [76,78]. However, the phenotypic and molecular effects of stress are reversible: inhibition of histone deacetylases leads to decreased DNA methylation, increased expression of GR and reduced stress-sensitivity (Fig. 1C) [76]. Thus, the environmental experiences of the maternal generation can result in epigenetic changes in behavior that require changes in chromatin structure. These changes can maintain the phenotype into adulthood and for subsequent generations [75].

Paternal behavior and environment can also impact the physiology of the offspring through epigenetic mechanisms [79]. Male mice consuming a low-protein diet father offspring with decreased hepatic levels of cholesterol esters and altered hepatic expression of lipid/cholesterol biosynthesis genes [80]. Likewise, blood glucose levels and pancreatic function in rats and mice are affected by paternal diet; high calorie diet leads to β-cell dysfunction in the female progeny [81,82]. Although the molecular mechanism by which paternal diet impacts future generations is not entirely clear, DNA methylation and expression of several genes involved in lipid metabolism is impacted by paternal diet [80]. For example, an enhancer for the lipid regulator PPARα was more highly methylated in the livers of low-protein offspring. However, the mechanism is still unclear; global DNA methylation in sperm was not dramatically affected by paternal diet and the altered methylation of the PPARα enhancer was not observed in sperm. Regardless, it is clear that paternal environmental factors can impact the physiology of offspring through epigenetic mechanisms.

Concluding remarks

It is important to integrate the concept of epigenetic memory into our thinking about the phenotypes that are subject to evolutionary selection [10]. Plastic responses to heterogeneous environmental conditions are one of the most common phenomena characterizing the living world [83]. Epigenetic variation and plasticity is an integral part of how organisms develop and interact with their environment [83]. The examples mentioned in this brief review demonstrate that epigenetic memory can modify phenotype to impact fitness. To understand biology and evolution, we must understand the role of epigenetic mechanisms in regulating behavior, physiology and gene expression over varying time scales. How, and to what extent, epigenetic mechanisms impact evolutionary fitness remains a fascinating question for the future. It is also important to consider that, depending on the context, memory may be both adaptive and harmful. For example, parental experience can either improve or diminish the health of offspring. Likewise, although transcriptional memory induced by interferon gamma may improve the response of cells to potential infection, it might also contribute to pathological states of inflammation. Future work must illuminate both the molecular basis for epigenetic regulation of phenotype and the ways in which it can be either adaptive or pathological.

Footnotes

References