DNA Methylation

DNA methylation is the covalent addition of a methyl group (–CH₃) to the 5-carbon of a cytosine ring, producing 5-methylcytosine (5mC), catalyzed by enzymes called DNA methyltransferases (DNMTs) using S-adenosylmethionine (SAM) as the methyl donor. In vertebrates the overwhelming majority of methylation occurs at CpG dinucleotides — a cytosine immediately followed by a guanine in the 5′→3′ direction, linked by a phosphodiester bond (hence "CpG," to distinguish it from a C–G base pair). The human genome is globally CpG-depleted (only about 20% of the expected frequency, a legacy of deamination-driven mutation over evolutionary time) except in CpG islands: GC-rich, CpG-dense stretches roughly 200–2000 bp long that sit at the promoters of about 60–70% of human genes, including nearly all housekeeping genes. Methylation is not simply "on" or "off" at the level of the whole genome — it is patterned. CpG islands at active gene promoters are typically kept unmethylated, while CpGs in gene bodies, intergenic regions, and transposable elements are heavily methylated. This produces a counterintuitive but well-established rule: promoter methylation silences, gene-body methylation is often compatible with — or even positively associated with — active transcription. The functional logic is that promoter methylation blocks the assembly of the transcription pre-initiation complex and recruits repressive chromatin readers, whereas gene-body methylation may help suppress spurious transcription initiation from within actively transcribed genes and modulate splicing. Because CpG is a palindrome (the complementary strand reads 5′-CpG-3′ too, just read the other way), a fully methylated CpG site has methyl groups on both strands. After DNA replication, each daughter duplex is hemimethylated — old strand methylated, new strand not — and this asymmetry is precisely the substrate that the maintenance machinery recognizes, which is the molecular key to heritability discussed in Mechanism.

Step 1 — Establishing the mark (de novo methylation). During early embryogenesis and germ cell development, the genome undergoes near-genome-wide erasure and re-establishment of methylation. DNMT3A and DNMT3B, guided by their catalytically inactive partner DNMT3L and by chromatin context (H3K36me2/3 recruits DNMT3B; unmethylated H3K4 is required, since the DNMT3 ADD domain is blocked by H3K4 methylation), lay down new methylation patterns on previously unmethylated DNA. This happens during implantation-stage development and during gametogenesis, setting the patterns that distinguish maternal and paternal imprinted alleles.

Step 2 — Maintaining the mark through replication. This is the crux of heritability. When a fully methylated CpG (5mC on both strands) is replicated by DNA polymerase, the new strand is synthesized with ordinary unmethylated cytosine. The result is a hemimethylated duplex. DNMT1, the maintenance methyltransferase, has strong biochemical preference for hemimethylated DNA (its cofactor UHRF1 specifically recognizes hemimethylated CpG via its SRA domain and recruits DNMT1) and methylates the new strand to restore full, symmetric methylation — copying the parental pattern faithfully onto both daughter chromosomes. This happens at the replication fork, coupled to PCNA and replication timing, ensuring the mark is propagated at every S phase without needing to re-derive it from any other information source.

Step 3 — Reading the mark. Methylation silences transcription through two overlapping mechanisms:

• Direct interference: methyl groups projecting into the major groove can sterically block binding of certain transcription factors (e.g., CTCF, whose binding is blocked by methylation at specific sites — this is exploited in genomic imprinting at the H19/Igf2 locus).
• Indirect, reader-protein mediated silencing: proteins with a methyl-CpG-binding domain (MBD), notably MeCP2, MBD1, MBD2, and MBD4, bind 5mC and recruit corepressor complexes containing histone deacetylases (HDACs) and other chromatin-modifying enzymes. This connects DNA methylation directly to Chromatin and Gene Expression: methylated DNA recruits machinery that removes activating histone acetylation and promotes formation of compact, transcriptionally silent heterochromatin.

Step 4 — Erasing the mark (active demethylation). Methylation is reversible. TET (ten-eleven translocation) enzymes (TET1/2/3) iteratively oxidize 5mC to 5-hydroxymethylcytosine (5hmC), then to 5-formylcytosine and 5-carboxylcytosine; these oxidized derivatives are excised by thymine-DNA glycosylase (TDG) and repaired by base excision repair, replacing them with unmodified cytosine. Passive demethylation also occurs when DNMT1 maintenance fails over successive replication rounds (dilution). Both mechanisms are critical during the two waves of global demethylation in mammalian life: in the pre-implantation embryo (erasing most parental methylation, except at imprinted loci) and in primordial germ cells (erasing imprints before sex-specific re-establishment).

Step 5 — Consequences at the whole-locus scale. Several classic biological phenomena reduce to this system operating at specific loci:

• X-inactivation: one X chromosome in every female mammalian cell is transcriptionally silenced, with CpG islands on the inactive X becoming densely methylated to lock in silencing after an initial non-methylation-based establishment step (Barr body formation).
• Genomic imprinting: roughly 100–200 genes in humans are expressed from only one parental allele, determined by germline-inherited "imprints" — differentially methylated regions (DMRs) set up during oocyte or sperm maturation and then maintained through all subsequent somatic divisions.
• Transposon silencing: roughly half the human genome consists of transposable elements; most are kept permanently methylated and silent, preventing genomic instability from their movement.

Griffith (1928) and Avery–MacLeod–McCarty (1944) established that heritable phenotypic information is carried in DNA at all, the necessary prerequisite for any theory of DNA-based epigenetic memory — Griffith's transforming principle in pneumococcus, later shown by Avery's group to be DNA itself, not protein. This is background context: it does not directly address methylation, but it fixes the arena (DNA carries heritable information) in which the methylation story unfolds.

The identification of 5-methylcytosine itself dates to Rollin Hotchkiss (1948), who used paper chromatography to separate calf thymus DNA hydrolysates and found a minor spot migrating differently from the four standard bases — subsequently identified as 5-methylcytosine, roughly 1% of total bases. This was the first direct chemical evidence that DNA carries a "fifth base" beyond A, T, G, C.

Holliday and Pugh, and independently Riggs (1975), proposed the theoretical model that made sense of scattered observations: they hypothesized that a maintenance methyltransferase acting specifically on hemimethylated DNA (produced transiently after each round of replication) could copy methylation patterns from parent to daughter strands, providing a mechanism for clonal, heritable, sequence-independent gene silencing — years before DNMT1 was purified or cloned. This was a rare case of a mechanistic model preceding its enzymological confirmation, later vindicated when DNMT1 was purified (Bestor and Ingram, 1983) and shown biochemically to prefer hemimethylated substrates by roughly 5- to 30-fold over unmethylated DNA, exactly as the model demanded.

Adrian Bird's discovery of CpG islands (1985–1986) used a restriction-enzyme approach: enzymes such as HpaII cut only unmethylated CpG sites, while its isoschizomer MspI cuts regardless of methylation status. By comparing digestion patterns, Bird's group showed that the genome contains discrete, GC-rich, CpG-dense regions that remain unmethylated in essentially all tissues and are preferentially located at gene 5′ ends — establishing CpG islands as a defining structural feature of active mammalian promoters, still visualized today by the "HpaII tiny fragment" (HTF island) method that gave islands their original name.

Bisulfite sequencing (Frommer, Clark, and Frommer's group, 1992) solved the central technical problem: methylation cannot be seen by ordinary Sanger sequencing because 5mC base-pairs identically to unmethylated C. Sodium bisulfite treatment deaminates unmethylated cytosine to uracil (read as T after PCR) but leaves 5-methylcytosine unreacted (still read as C), because the methyl group sterically and electronically protects the ring from the bisulfite reaction. Sequencing bisulfite-converted DNA and comparing it to the untreated reference sequence reveals, base by base, exactly which cytosines were methylated. This single method — now scaled to whole-genome bisulfite sequencing (WGBS) — converted DNA methylation from an inferred property into a directly, quantitatively readable feature of any genome, and remains the gold-standard assay.

Direct proof of causal gene silencing came from experiments transfecting in-vitro methylated versus unmethylated copies of the same reporter gene (e.g., work by Walter Doerfler in the 1980s using adenovirus and viral promoter constructs, and later classic experiments with methylated versus unmethylated herpes thymidine kinase genes) into cultured cells: methylated copies of an identical DNA sequence were stably and heritably silenced across cell divisions, while unmethylated copies remained active — demonstrating that methylation itself, not merely correlation with an already-silent state, causes silencing.

Knockout mouse experiments completed the causal chain in vivo. Li, Bestor, and Jaenisch (1992) showed that Dnmt1-null mouse embryos have globally hypomethylated genomes, aberrant imprinted gene expression, and die around mid-gestation — proving DNMT1 is essential for maintaining methylation and for normal development. Okano, Bestor, and Li (1999) showed Dnmt3a and Dnmt3b double knockouts fail to establish de novo methylation and die shortly after birth, confirming the distinct de novo versus maintenance division of labor predicted by the Holliday–Riggs model decades earlier.

Example 1 — Reasoning through an imprinting disorder (Angelman vs. Prader-Willi syndrome). Chromosome 15q11–13 contains a cluster of imprinted genes, including UBE3A (expressed only from the maternal allele in neurons; the paternal copy is silenced by methylation-associated imprinting) and SNRPN (expressed only from the paternal allele; the maternal copy is silenced). Consider a patient with a deletion of the maternal copy of 15q11–13.

• The maternal UBE3A allele (the only one normally expressed in neurons) is now missing entirely.
• The paternal UBE3A allele is present but is epigenetically silenced (as it always is, in every normal individual).
• Net result: zero functional UBE3A protein in neurons → Angelman syndrome (severe intellectual disability, ataxia, inappropriate laughter, seizures).

Now consider a patient with the same-sized deletion, but on the paternal chromosome 15.

• The paternal SNRPN (and neighboring genes normally paternally expressed) are lost.
• The maternal copies of those genes are present but silenced by the maternal imprint.
• Net result: loss of paternally-expressed gene products → Prader-Willi syndrome (hyperphagia, obesity, hypotonia, mild intellectual disability).

The key quantitative/logical point: identical deletion, opposite phenotype, depending purely on the parent of origin — a result that is completely inexplicable under classical Mendelian genetics (which assumes both alleles are equivalent) and is only explicable once you know that a DNA methylation mark, laid down in the germline, makes one parental allele functionally "invisible" regardless of its sequence. About 70% of Angelman cases arise from maternal deletion, ~2–5% from uniparental disomy (both copies of chromosome 15 inherited from the father, so no maternal UBE3A exists at all), and a further fraction from direct mutations in the imprinting center itself that fail to erase/reset methylation correctly in the germline — each mechanism converging on the same molecular deficit: no active maternal UBE3A.

Example 2 — Quantitative reasoning with a cancer methylation biomarker panel. Suppose a colorectal cancer screening test measures methylation at the promoter CpG island of a tumor suppressor gene (e.g., a MLH1-like locus) using quantitative bisulfite pyrosequencing, reporting percent methylation (fraction of alleles methylated at that CpG, averaged across the cell population sampled). Given data from a hypothetical validation cohort:

Reasoning through this: a normal cell population is a mixture of cells with essentially unmethylated promoters (hence ~4% reflects assay noise plus rare stochastic methylation), and gene expression is at its full baseline. As the lesion progresses from polyp to carcinoma, an increasing fraction of cells in the population carry the fully-methylated, silenced allele — 91% average methylation in the carcinoma sample most likely reflects a population in which nearly all cells have both CpG-island alleles hypermethylated (biallelic silencing), consistent with the near-total loss of expression (3% residual, likely from rare heterogeneous subclones or stromal contamination in the biopsy, since epithelial tumor purity is rarely 100%). This matters clinically because for a mismatch-repair gene, biallelic silencing by promoter hypermethylation produces the same functional loss-of-heterozygosity phenotype as a two-hit sequence mutation (per Knudson's two-hit hypothesis), but is detectable non-invasively and is, unlike a mutation, in principle reversible — the rationale behind DNA methyltransferase inhibitors (e.g., azacitidine) used clinically in myelodysplastic syndrome and acute myeloid leukemia, which force passive demethylation by trapping DNMT1 on DNA during replication and depleting functional enzyme, allowing silenced tumor suppressor genes to re-express.

Cancer diagnostics and therapy. Aberrant promoter hypermethylation silencing tumor suppressors (MLH1 in Lynch-syndrome-like sporadic colorectal cancer, BRCA1 in some breast/ovarian cancers, CDKN2A/p16 across many tumor types) is now a routine clinical biomarker, assayed by bisulfite pyrosequencing or methylation-specific PCR; blood-based "liquid biopsy" methylation panels (e.g., Galleri-type multi-cancer early detection tests) exploit tumor-specific methylation signatures in circulating cell-free DNA. DNMT inhibitors (azacitidine, decitabine) are FDA-approved standard-of-care agents for myelodysplastic syndrome and certain leukemias.

Reproductive medicine and imprinting disorders. Assisted reproductive technologies (IVF, ICSI) are under active study for possible associations with rare imprinting disorders (Beckwith-Wiedemann syndrome, Angelman syndrome via imprinting-center defects), because the window of germline methylation reprogramming overlaps with the period of oocyte/embryo manipulation.

Forensic and age-estimation science. Age-associated, tissue-specific methylation drift at specific CpG sites ("epigenetic clocks," e.g., Horvath's multi-tissue clock) allows estimation of biological age from a DNA sample, used in forensic age estimation and studied as a biomarker of aging and mortality risk.

Agriculture and livestock cloning. Somatic cell nuclear transfer (cloning, as in Dolly the sheep) frequently fails or produces abnormal offspring in large part because the donor nucleus's somatic methylation pattern is not properly reprogrammed to an embryonic state — understanding methylation reprogramming has directly improved cloning and stem-cell reprogramming efficiency (relevant to induced pluripotent stem cell technology, where residual "epigenetic memory" methylation from the donor cell type can bias differentiation potential).

Conservation and rewilding genomics. Methylation profiling is increasingly used alongside genomic sequencing to assess stress, environmental exposure, and population health in endangered species, since methylation patterns can shift in response to environmental conditions within a generation, offering a faster-responding biomarker than sequence-based diversity metrics alone.

Ask the tutor: How does DNA methylation interact with histone modifications to establish stable heterochromatin, and how does this connect to X-chromosome inactivation?