Review

Dynamic DNA methylation: In the right place at the right time

See allHide authors and affiliations

Science  28 Sep 2018:
Vol. 361, Issue 6409, pp. 1336-1340
DOI: 10.1126/science.aat6806

Abstract

The classical model of cytosine DNA methylation (the presence of 5-methylcytosine, 5mC) regulation depicts this covalent modification as a stable repressive regulator of promoter activity. However, whole-genome analysis of 5mC reveals widespread tissue- and cell type–specific patterns and pervasive dynamics during mammalian development. Here we review recent findings that delineate 5mC functions in developmental stages and diverse genomic compartments as well as discuss the molecular mechanisms that connect transcriptional regulation and 5mC. Beyond the newly appreciated dynamics, regulatory roles for 5mC have been suggested in new biological contexts, such as learning and memory or aging. The use of new single-cell measurement techniques and precise editing tools will enable functional analyses of 5mC in gene expression, clarifying its role in various biological processes.

DNA methylation (the presence of 5-methylcytosine, 5mC) is one of the best studied epigenetic systems in mammals, and the mechanisms by which this covalent epigenetic mark are written, read, and perpetuated are well characterized (Figs. 1 and 2) (1, 2). Gene regulation by 5mC was initially studied by analyzing CpG islands (CGIs), which are stretches of DNA sequence enriched in CpG dinucleotides that are found at the majority of mammalian promoters (2). The presence of 5mC at a CGI is usually associated with long-term, stable gene repression (2). However, 5mC patterns outside of promoters and during dynamic biological processes were less understood before the development of epigenomic tools. Whole-genome 5mC profiling has uncovered the dynamics of 5mC at enhancers, gene bodies, and extended transcriptionally inactive partially methylated domains (PMDs), indicating regulatory roles for 5mC in development and disease (35). Although 5mC is broadly accepted as a repressive epigenetic mark, the mechanism of 5mC-mediated transcriptional repression is still not fully understood (1, 2). Several studies suggest that gene regulation by 5mC may involve functional interactions between 5mC and sequence-specific transcription factors (TFs) (69). Here we review recent insights regarding the dynamics of 5mC in development and disease. We further discuss the utility of single-cell methods and new epigenome-editing technologies for further analysis of 5mC function and the discovery of additional DNA base modifications in mammalian genomes.

Fig. 1 DNA methylation signatures of developmental stages and tumors.

This figure summarizes 5mC levels measured from heterogeneous collections of cell types, tissues, or organs. Low global 5mC levels are observed in primordial germ cells and preimplantation embryos. A global gain of 5mC occurs during postimplantation development. In late fetal development, local demethylation occurs at regulatory sequences; in postnatal development, a gain of 5mC occurs at embryonically active regulatory sequences, and a loss of 5mC occurs at regulatory sequences active in differentiated tissues. Neuron subtypes A and B show brain-specific accumulation of mCH, the abundance of which is inversely correlated with transcription activity. In aged tissues, PMDs in late-replication regions resulted from progressive methylation loss; the dashed line indicates that a small percentage of CGIs show hypermethylation in aged tissues (67). The trophectoderm has a low global 5mC level similar to that observed in the preimplantation embryo. Placenta and tumors show pervasive PMDs and CGI hypermethylation. The solid and dashed lines in the tumor panel indicate mCG levels in tumor and matching normal tissues, respectively. The color gradient in the background indicates an increase in the level of 5mC, from the bottom to the top of each panel.

Fig. 2 The function and regulation of DNA methylation in diverse genomic contexts.

(A) Gene promoter CGIs are protected from 5mC by CXXC domain–containing proteins. De novo DNA methyltransferase DNMT3A activity is inhibited by H3K4me3. At distal regulatory elements, 5mC can either prohibit the binding of 5mC-sensitive TFs or interact with TFs that show a preference for methylated binding sequence. DNA methylation at CTCF binding sites regulates chromatin conformation by modulating CTCF binding. In mESCs (top), DNMT3B is recruited to transcribed regions by H3K36me3 and establishes gene-body methylation. In postmitotic neurons (bottom), 5mCH is bound by MeCP2. Pol2, RNA polymerase II. Nucleosomes were omitted from the bottom panel for simplicity. (B) Transcription outputs are correlated with 5mC and 5hmC levels of enhancers (E), promoters (P), and gene bodies (G). *5mCH patterns in hESCs are shown; the 5mCH level is substantially lower in mESCs than in hESCs. †To date, a significant level of 5hmCH was only found in cerebellar granule cells. N.D., not detected.

Dynamic DNA methylation in development

In the mammalian genome, 5mC is predominantly located within the CpG dinucleotide context. In most somatic tissues, the majority (>80%) of CpGs in the genome are methylated, with the exception of CGIs and other gene regulatory sequences that show reduced 5mC levels (3). Indeed, local 5mC depletion is a reliable signature of promoters and enhancers: CpG sites in promoter-associated CGIs are often less than 10% methylated, whereas distal regulatory sequences such as enhancers are commonly marked by levels of 5mC ranging from 10 to 50% (6). CGI methylation shows few dynamics across healthy tissues. By contrast, aberrant CGI hypermethylation is commonly observed in cancers (2). With respect to methylation patterns outside CGIs, direct comparison of whole-genome methylome profiles derived from numerous cell types has revealed that, under normal physiological conditions, 15 to 21% of CpGs vary in methylation status among human tissues (4, 5). These dynamically methylated CpGs cluster into more than 1 million tissue-specific differentially methylated regions (DMRs). DMRs are predominantly distal to transcriptional start sites and overlap with enhancers. Intragenic DMRs that stretch for kilobases (large DMRs) can overlap with tissue-defining “signature genes” as well as with predicted superenhancers, which are clusters of enhancers that function cooperatively (4, 10, 11). These tissue type–specific DMRs are defined by both lineage-specific gain and loss of 5mC during embryogenesis. A study that profiled 5mC dynamics during mouse late fetal development with daily sampling (12) revealed a temporal uncoupling of the loss and gain of 5mC during the development of all tissues analyzed: Enhancers predominantly lose 5mC during fetal development, whereas most de novo 5mC occurs after birth (Fig. 1) (12). In mouse brain cortex, synchronized de novo 5mC is observed during the second postnatal week, primarily targeting enhancers that are embryonically active (12). Although few analyses have been systematically performed across the lifespan of mammals, existing data suggest that, overall, the methylome is stable beyond adolescence (13).

Contrary to what is observed in somatic cells, genome-wide changes of DNA methylation patterns occur during preimplantation mammalian development and in the early embryonic germ line (Fig. 1) (1417). However, these global methylation changes are coordinated with genome-wide changes in histone modifications and reflect a global reprogramming of epigenetic information rather than locus-specific transcriptional regulation. For example, global erasure of DNA methylation in primordial germ cells does not lead to widespread transcriptional activation (16, 17), and numerous genes are expressed during spermatogenesis despite high levels of methylation at their promoters (18).

5mC patterns in differentiated tissues are highly specific and established primarily by site-specific remodeling at regulatory sequences. Sequential waves of gain and loss of 5mC in late fetal to early postnatal development contribute to the majority of 5mC diversity found in differentiated tissues.

Transcription factors direct local 5mC remodeling

The landscape of 5mC in the mammalian genome is shaped by local depletion of 5mC at regulatory sequences (Fig. 2). The depletion of 5mC at CGIs is, in part, determined by the presence of CpG dinucleotide clusters that are specifically recognized and protected from methylation by proteins containing a CXXC zinc-finger domain (19). These proteins include components of the histone H3 lysine 4 (H3K4) methyltransferase complex (19). Trimethylated H3K4 (H3K4me3), in turn, inhibits the activity of the de novo DNA methyltransferase DNMT3A, thus preventing methylation at CGIs (20). Beyond CXXC domain–containing proteins, the depletion of 5mC at CGIs is also dependent on the binding of other specific TFs (19).

Unlike CGI promoters, enhancers have a CpG density comparable to the genomic average. Tissue- or cell type–specific regulatory sequences can be effectively identified by DMRs and are enriched in recognition sites for TFs with tissue-specific functions (4, 5, 21). We speculate that the large quantity of DMRs (20,000 to 100,000 per tissue type) reflects the collective local demethylation induced by specific TF binding. This model has been supported by experiments in mouse embryonic stem cells (mESCs) that use insulator protein CCCTC-binding factor (CTCF) and a transcriptional repressor of neural genes—RE1-silencing transcription factor (REST) (6). The presence of an intact CTCF motif is sufficient and necessary for local methylation depletion. The deletion of REST led to increased 5mC at regions of low methylation normally bound by REST. Notably, such gain of 5mC can be reversed by reintroducing REST.

It has been postulated that “pioneer” TFs, which can direct cellular reprogramming, are likely to induce local epigenetic remodeling (22). One example is ASCL1, whose forced expression directly converts fibroblasts to postmitotic neurons and leads to reduced 5mC at most of its binding sites (22, 23). However, given the small number of TFs tested to date, it is unknown whether the ability to remodel 5mC is indeed restricted to TFs with pioneer activity. In addition, the mechanism of TF-induced local 5mC loss is not understood. TFs may directly recruit cytosine demethylation machinery containing ten-eleven translocation methylcytosine dioxygenase (TET, see below) (24) or, alternatively, the demethylation is mediated by intermediate steps of chromatin remodeling or histone modification.

Transcription factors read DNA methylation

The depletion of 5mC at active regulatory elements suggests that the absence of this repressive mark is essential for active transcription regulation. However, our understanding of the mechanism by which 5mC affects transcription machinery remains incomplete. 5mC readers, including proteins containing a methyl-CpG–binding domain such as MECP2 or MBD proteins, can recognize 5mC and recruit coregulators to methylated DNA (25). In addition, studies suggest that numerous TFs can also directly sense the state of cytosine methylation through the modulation of their binding affinity and specificity (26).

The development of high-throughput assays for analyzing TF-DNA interactions has enabled the identification of many TFs with binding specificity that is altered by the presence of 5mC in recognition sequences (26). A protein microarray study that investigated more than 1500 TFs found more than 40 TFs that can bind methylated motifs (27). Pull-down experiments of methylated and unmethylated random sequences with a recombinant TF protein followed by sequencing [methyl–SELEX (systematic evolution of ligands by exponential enrichment)] concluded that the binding of about 60% of TFs was influenced by 5mC, and the binding of comparable numbers of TFs was enhanced or inhibited by 5mC (28). These studies unexpectedly uncovered that certain interactions between TFs and low-affinity binding sites can be enhanced by the presence of 5mC. Similarly, DNA affinity purification sequencing (DAP-seq) of more than 500 Arabidopsis (plant) TFs determined that >75% of TFs are sensitive to 5mC (29), suggesting that the regulation of TF binding by 5mC is conserved across animal and plant kingdoms.

Although in vitro studies reveal that the majority of TF-DNA interactions can be modulated by 5mC, it is still uncertain how broadly this mechanism contributes to in vivo gene regulatory programs. There are only a handful of reports of 5mC regulating TF binding in a physiological context. In two well-characterized examples, aberrant gain of 5mC at the binding site of insulator protein CTCF is observed in the Igf2/H19 locus in Beckwith-Wiedemann (fetal overgrowth) syndrome and in the PDGFRA locus of isocitrate dehydrogenase mutant gliomas (8, 9). In both cases, hypermethylation of the binding site prevents CTCF binding and results in the disruption of chromatin topological domains, leading to ectopic enhancer-promoter interactions. Intriguingly, 5mC only affects CTCF binding at a specific subset of recognition sequences (30). This selective 5mC sensitivity results from the different biophysical properties of CTCF’s 11 zinc fingers—each motif has a different sensitivity for methylation of specific cytosines (31).

We speculate that the in vivo regulation of TF binding by 5mC is more common than what has been reported in the literature. However, such mechanisms may only affect TF function at a subset of sites that show developmental- or disease-related 5mC dynamics and thus might have been previously overlooked. In addition, the bidirectional regulation between TF binding and 5mC presence poses a challenge in determining whether TF binding is the cause or consequence of 5mC remodeling. We anticipate that new epigenetic editing tools for specific manipulation of 5mC (see below) may greatly facilitate functional studies of TF-5mC interactions.

Intragenic DNA methylation and transcription

Compared with its well-established repressive function at regulatory elements, less is known about 5mC regulation and function(s) at intragenic regions (Fig. 2). Studies with mESCs and a heterologous yeast system found that the establishment of intragenic 5mC is a controlled process carried out by de novo DNA methyltransferase DNMT3B and cotranscriptionally regulated histone modification H3K36me3 (32, 33). Specifically, DNMT3B preferentially binds to H3K36me3 and deposits 5mC within gene bodies.

Although promoter 5mC shows a negative correlation with transcription, early studies of intragenic 5mC function posited the opposite finding (2). Indeed, for certain tissue types, active genes are more methylated than repressed genes. However, this correlation appears to be tissue specific. The interaction between DNMT3B and H3K36me3 leads to greater levels of intragenic 5mC for active genes in mESCs (32). For some tissue types, the positive correlation between 5mC and transcription can be explained by the stratification of genes by chromatin states. At many developmental genes enriched in H3K27me3, intragenic 5mC antagonizes polycomb repression and thus is positively correlated with transcription (34). In PMD-associated tissues such as cultured fibroblasts, genes located in PMDs are both expressed and methylated at low levels (3). Many signature genes of human tissues overlap with intragenic large DMRs, suggesting that intragenic 5mC may regulate tissue-specific gene expression (4). In addition, intragenic 5mC has also been postulated to regulate alternative promoter use and alternative splicing and prevent spurious internal transcription initiation (2, 35, 36).

Human embryonic stem cells (hESCs) and mammalian brain tissues contain abundant non-CG methylation (5mCH, where H represents A, C, or T), although the underlying three-base DNA sequence context (5mCAG versus 5mCAC, respectively) is distinct (3, 13). Intragenic 5mCH is positively correlated with gene expression in hESCs but is negatively correlated with gene expression in the brain (13). In neurons, intragenic 5mCH quantitatively modulates gene expression by interacting with MECP2 and recruiting the nuclear receptor corepressor complex to chromatin (3740). MECP2 contains a methyl-CpG binding domain, and mutations in MECP2 have been linked to Rett syndrome, an autism spectrum disorder (41). Independent studies have reported that MECP2 binds to both 5mCA and 5mCG sites with high affinity (37, 39, 40). CA sites are the most frequently methylated non-CG sites (13). Repression dependent on both MECP2 and 5mCA is particularly critical for long genes (with a length greater than 100 kb), many of which have neuronal functions (39).

Intragenic 5mC (including 5mCH) performs fine-tuning and supportive roles in transcription regulation but is nevertheless functionally important, as suggested by known links between disease and intragenic 5mCA and MECP2 repression (39).

Functions of cytosine hydroxymethylation

The dynamic changes in 5mC patterns have raised questions about the possible mechanisms of 5mC removal. In this context, 5-hydroxymethylcytosine (5hmC) is the most abundant oxidative derivative of 5mC. 5hmC is generated by the TET family of proteins and is an intermediate product of 5mC demethylation mediated by TET in combination with thymine DNA glycosylase–mediated base excision repair (24). Although TET enzymes are highly expressed during the developmental periods of global DNA demethylation, their role in the observed methylation dynamics seems auxiliary and primarily antagonizes the activity of de novo methylation (17, 42). However, TETs are likely to play important roles in local 5mC dynamics at the regulatory sequences. TET binding is enriched in promoters and enhancers and TET is found in complexes with TFs, suggesting involvement in 5mC dynamics after TF binding (24).

Studies also suggest that the majority of 5hmC can be relatively stable, arguing against 5hmC as being solely a short-lived demethylation intermediate and potentially pointing toward an epigenetic or regulatory role for this modification (43). This is also supported by the abundance of 5hmC in the brain (up to 20% of the total amount of 5mCG and 5hmCG) (13, 40). Although sequencing-based approaches have only found 5hmC in CG dinucleotides in mESCs and brain cortical tissue (24), 5hmCH accounted for 38% of the total 5mCH and 5hmCH modifications in cerebellar granule cells, suggesting that 5mCH is also targeted for TET-mediated oxidation (40). Interestingly, 5hmCG and 5hmCA (the main context of non-CG methylation) interact with MECP2 differently and lead to opposite gene regulatory outputs. Both 5mCA and 5hmCA can be bound by MECP2 (39, 40), whereas the oxidation of 5mCG to 5hmCG abolishes the MECP2 interaction and disrupts the function of 5mCG as a repressive mark (Fig. 2B).

Aberrant methylation patterns in diseases and aged tissues

Global changes of methylation patterns are observed in pathology. Tumor methylomes are characterized by global hypomethylation and localized hypermethylation at specific CGIs, many of them associated with tumor suppressor genes (2). Analysis of the whole-genome methylomes of 39 primary tumors in mouse and human shows a pronounced loss of methylation at nuclear lamina–associated PMDs located in late replicating parts of the genome (44). Intriguingly, limited PMD hypomethylation can be detected in healthy tissue and increases with age (44). This loss in DNA methylation observed during aging may be due to the insufficient maintenance of DNA methylation across an increasing number of cell divisions (44), which ties similar methylation changes to both cancer and aging (44, 45). An increase in PMD abundance is also found during the early development of placenta along with comparable CGI hypermethylation, both of which are also observed in cancers (46). The similarity between tumor and placenta methylomes suggests that tumorigenesis may represent a latent developmental epigenomic program.

Despite global DNA hypomethylation, a limited number of genomic loci gain DNA methylation during aging and in cancer. Interestingly, these are sequences that are transcriptionally repressed by the polycomb system (and lack DNA methylation) under normal physiological conditions (47). Similar gains of methylation at distinct polycomb-repressed sequences have also been observed, although to a lesser degree, during placental development (46), suggesting an interesting and mechanistically important interplay between DNA methylation and polycomb-regulated epigenetic systems.

In instances in which DNA methylation is globally reduced as a part of a normal developmental program, no aberrant transcriptional activation is observed, suggesting an alternative or compensatory mechanism of gene expression regulation. The situation is, however, different under pathological conditions in which the global loss of DNA methylation is often associated with an aberrant transcriptional outcome.

Single-cell techniques uncover the diversity of DNA methylation patterns

Developing tissues contain heterogeneous cell populations that are derived from different cell lineages and exhibit various degrees of maturation. Although computational approaches for tissue deconvolution are improving (48), bulk tissue analyses cannot easily distinguish molecular signatures contributed by distinct cell populations. Improvements in single-cell methylome approaches now provide the power to distinguish cellular heterogeneity within a tissue or cell population and identify epigenome fluctuations in a given cell type over time.

Since single-cell reduced representation bisulfite sequencing (scRRBS) and single-cell bisulfite sequencing (scBS-seq) were first reported (49), methodological improvements have increased both throughput and resolution. Datasets containing thousands of single-cell methylomes, such as neuron classification in the human and mouse frontal cortices, have been generated with a well-based nuclei collection method—single-nucleus methylcytosine sequencing (snmC-seq)—or a combinatorial indexing-based method—single-cell combinatorial indexing for methylation analysis (sci-MET) (10, 50). Unlike single-cell RNA sequencing, which captures transcriptomic information, single-nucleus methylome data allowed prediction of cell type–specific regulatory elements with 5mC signatures. Building on single-cell methylome methods, single-cell multi-omics approaches have been developed that can analyze both the transcriptomic and epigenomic profiles from a single cell (49). Studies using these techniques have identified cell-to-cell epigenetic variation in early mouse and human embryos (51, 52), demonstrating the enhancement of resolution enabled by multi-omic single-cell methods.

In addition to sequencing-based single-cell methylome strategies, a fluorescent in vivo reporter of genomic methylation enables microscopic examination of the 5mC status at a single genomic locus with cellular resolution (53). In an alternative strategy, the fluorescent protein gene reporter can be directly inserted into the methylation-sensitive locus, which provides a direct readout of the methylation status in living cells or a whole animal (54). Although single-cell methods are just starting to be applied to the analysis of 5mC functions, studies have already identified 5mC heterogeneity both across or within cell types, providing information that cannot be accessed with bulk-level analyses.

Toward site-specific manipulation of DNA methylation

Functional analyses of 5mC have traditionally relied on pharmacological or genetic perturbations that affect 5mC levels genome-wide (30, 32). Although discoveries regarding 5mC function and regulation have been made with genome-wide perturbation strategies, these methods are associated with pleiotropic phenotypes and inevitably confound the interpretation of results (35). Modern genome-engineering tools have opened up the possibility for a more precise site-specific intervention that would potentially avoid pleiotropic effects of global 5mC disruption (55).

Strategies for targeting 5mC to specific genomic locations have used different programmable DNA binding domains, including zinc fingers, transcription activator–like effectors and cleavage-deficient dCas9. dCas9-TET and dCas9-DNMT3A fusions have been shown to induce effective remodeling of 5mC (Fig. 3) (56). Most studies take a candidate approach to estimate the extent of off-target manipulation. However, off-target de novo methylation activity can be widespread across the entire genome (57) and is not easily detected with steady-state 5mC measurements. Local amplification of regulator concentration through protein (SunTag) or RNA scaffolding (e.g., Casilio) has been incorporated into epigenetic editing to enhance the on-target/off-target ratio (Fig. 3). This scaffold allows multiple copies of epigenetic effectors (e.g., DNMT3A catalytic domain) to be tethered to a genomic site, providing more robust on-target modulation and reduced nonspecific effects (58).

Fig. 3 Comparison of direct fusion- and SunTag protein scaffold–based 5mC editing tools.

(A) Direct fusion-based editing. (B) The SunTag system allows local amplification of effector (DNMT3A) concentration and independent control of dCas9 and DNMT3A expression levels to minimize off-target de novo methylation caused by untethered DNMT3A. The SunTag-based 5mC editing tool (dCas9Sun-D3A) shows comparable levels of on-target methylation at on-target promoters as the direct fusion-based editing tool (dCas9-D3A), but much lower off-target methylation levels. sgRNA, single-guide RNA; NLS, nuclear localization sequence; GB1, solubility tag (protein G B1 domain); scFv-GCN4, single-chain variable fragment antibody against GCN4. The schematics in (A) and (B) are based on 5mC editing tools developed by Liu et al. (56) and Pflueger et al. (58), respectively.

Beyond cytosine base modifications: 6mA, the new kid on the block

5mC and its oxidation derivatives had been considered the only type of mammalian DNA modification until reports of the presence of N6-methyladenine (6mA) in DNA of higher eukaryotes (59). Compelling multimodal evidence has been presented for the existence of 6mA in mouse and human genomes. These findings are supported by the identification of both methyltransferase and demethylase for deposition and removal of 6mA, respectively (60, 61). Interestingly, the genome-wide distribution of 6mA is species-specific and lacks a conserved pattern across the eukaryotic species analyzed to date (59). Reported differences in 6mA distributions between mouse and human may represent tissue-specific patterns rather than species differences, because the tissues used for the mouse and human studies were unique to each species [mESCs and blood DNA, respectively (60, 61)]. Functional analysis of 6mA in mammalian genomes may be most productive when applied to tissue with abundant 6mA, to be determined with surveys of 6mA across tissues and developmental stages.

Outlook

Recent studies have indicated intriguing, possibly regulatory, roles for 5mC dynamics in diverse biological processes and disease states. Animal models, such as mouse fear conditioning, suggest a role for 5mC in regulating learning and memory (62). Epigenome-wide association studies have linked 5mC signatures to many diseases (63). However, many studies use correlative analysis and report only moderate 5mC differences detected from complex tissues. Thus, although interesting, these studies should be considered as exploratory in nature. In addition, meta-analysis of 5mC microarray data has identified a small number (~300 to 400) of CpG sites that are associated with age across multiple tissues, suggesting the existence of a methylome-based clock (45). Further work is needed to link these clock methylation variants with other molecular information (e.g., gene expression) and aging-related phenotypes.

In mitotically inactive mammalian tissues, moderate changes in 5mC dynamics likely reflect cellular heterogeneity either across cell types or within a cell type. In this context, we anticipate that single-cell multi-omic strategies will enhance the specificity and sensitivity of the analysis of 5mC patterns, allowing the determination of cellular identity in tissues (cell atlas) and measurement of alterations pertinent to experimental and environmental perturbations. Effectively, the 5mC dynamics of all cell types in the tissue can be reconstructed with single-cell profiling without cell-type labeling or enrichment strategies, contributing to organism-scale cell-type reference atlas initiatives as well as profiling of disease tissues at a cell-type resolution. (64).

Site-specific methylome editing will serve as a key technique for the study of 5mC function and should pave the way for the development of epigenetic-based strategies for cellular engineering and therapies. Specific editing tools will be crucial for moving beyond correlation, allowing separation of cause from consequence. An important use for methylome editing will be to aid understanding of the impact of 5mC on transcription and its relationship to other types of epigenetic modifications—a subject actively debated. For example, a recent methylation-editing study demonstrated that deposition of 5mC at thousands of gene promoters led to only modest transcriptional alterations (65, 66). With the availability of facile methods for single-cell methylome profiling and editing tools that allow manipulation of 5mC in cells at the right place and time, we anticipate that, in the coming decade, discoveries of new biological functions (or the debunking of proposed ones) for DNA methylation will proceed at an unprecedented pace.

References and notes

Acknowledgments: We apologize to those colleagues whose work we could not cite owing to space limitations. We thank B. A. Wang, J. A. Law, and J. R. Dixon (Salk Institute); S. C. Huang (New York University); and B. Ren (University of California, San Diego) for insightful discussions. Work in the Hajkova lab is supported by MRC funding (MC_US_A652_5PY70) and by an ERC grant (ERC-CoG-648879–dynamic modifications). Research in the Ecker lab is supported by NIH awards 5R21HG009274, 5U24DK112348, 5R01ES025585, 5R21MH112161, 5R01MH112763, and 1U19MH114831. Additional support to J.R.E. is provided by the Office of Naval Research N00014-16-1-3159 and the Cure Alzheimer’s Fund. J.R.E. is an investigator of the Howard Hughes Medical Institute. Competing interests: The authors declare no competing interests.
View Abstract

Stay Connected to Science

Navigate This Article