Epigenetics refers to heritable changes in gene expression and chromatin structure that occur without alterations to the underlying DNA sequence. The central problem epigenetics solves is this: every neuron, glial cell, and hepatocyte in the body shares an identical genome, yet each expresses a radically different set of genes. The information that specifies cell identity, and that allows a fixed genome to respond to a changing environment, is therefore not written in the sequence itself but layered on top of it — hence epi-genetics. These marks are mitotically heritable (a cell's daughters remember which genes were on), but unlike mutations they are chemically reversible, which is precisely what makes them attractive drug targets.

Why this matters for neurology specifically: the brain is the most epigenetically dynamic organ in the body. Neurons are post-mitotic and must persist for a lifetime, so they cannot rely on cell division to refresh their regulatory state; instead they continuously rewrite chromatin in response to neural activity. Activity-dependent methylation and demethylation underlie memory consolidation and synaptic plasticity, which is why mutations in the epigenetic machinery so often produce intellectual disability, autism, and epilepsy rather than a single organ-restricted defect.

Why CpG, and why islands? Methylation in mammals falls almost exclusively on cytosines that sit immediately 5' of a guanine (a CpG dinucleotide). This is mechanistically important because CpG is a palindrome — the complementary strand also reads CpG — so a methylated CpG/CpG pair carries the same mark on both strands. After replication the daughter duplex is hemimethylated (old strand methylated, new strand bare), and this hemimethylated state is the substrate that lets the cell copy the pattern faithfully. Most CpGs across the genome are methylated and depleted by evolution (because methyl-C deaminates to thymine and is mutated away), but CpG islands — dense, unmethylated CpG clusters at ~70% of promoters — are protected and kept open. Pathological hypermethylation of a promoter island recruits methyl-CpG-binding proteins (e.g., MeCP2) and the histone deacetylases they tether, collapsing the local chromatin into a silent state.

Writing, maintaining, and erasing — and why the division of labor matters clinically:

• De novo methyltransferases (DNMT3A, DNMT3B) install brand-new patterns during early development and gametogenesis. Because they set the developmental program, germline DNMT3A variants cause a growth/intellectual-disability overgrowth disorder (Tatton-Brown-Rahman syndrome).
• DNMT1, the maintenance methyltransferase, recognizes the hemimethylated CpG after replication and restores full methylation, preserving the pattern through every cell division. DNMT1 acts in post-mitotic neurons too, where mutations cause adult-onset neurodegeneration (HSAN1E; ADCA-DN) — a clue that maintenance methylation is needed for neuronal survival, not just for dividing cells.

Erasing methylation is not passive. Active demethylation is driven by TET enzymes (TET1-3), which iteratively oxidize 5-methylcytosine to 5-hydroxymethylcytosine and further oxidized forms that are then swapped out by base-excision repair for an unmodified cytosine. 5-hydroxymethylcytosine is not merely an intermediate — it is unusually abundant and stable in neurons, marking actively used neuronal genes, which is why TET activity is tightly coupled to synaptic plasticity and learning.

Genomic DNA is packaged by wrapping approximately 147 base pairs around a histone octamer (two each of H2A, H2B, H3, H4) to form the nucleosome — the fundamental repeating unit of chromatin. Packaging is not merely a storage solution for fitting two meters of DNA into a nucleus; it is itself a regulatory layer. DNA wrapped tightly on a nucleosome is physically occluded from transcription factors, so the cell controls gene access by controlling how nucleosomes are marked and positioned.

The histone code — why modifications, not just compaction: The unstructured N-terminal tails of histones protrude from the nucleosome and are decorated with more than 100 distinct post-translational modifications (acetylation, methylation, phosphorylation, ubiquitination). These marks act in two ways. First, some directly change chromatin physics — acetylation of a lysine neutralizes its positive charge, weakening the electrostatic grip of the histone on the negatively charged DNA backbone and loosening the fiber. Second, and more importantly, marks function as a combinatorial signaling code: 'reader' proteins with specialized domains (bromodomains read acetyl-lysine, chromodomains read methyl-lysine) dock onto specific marks and recruit downstream machinery. The same chemical group can mean different things in different contexts — for example, methylation activates at H3K4 but represses at H3K27 — so it is the position and combination of marks, not any single one, that is interpreted.

Chromatin remodeling — moving the nucleosomes themselves: Marks change how nucleosomes are read; remodelers change where nucleosomes sit. ATP-dependent remodeling complexes hydrolyze ATP to slide, evict, or restructure nucleosomes, exposing or hiding promoters and enhancers. This is an active, energy-consuming process precisely because nucleosome positioning must be fought against the thermodynamic tendency of DNA to stay wrapped.

Why do these genes dominate the neurodevelopmental disorder gene lists? Chromatin regulators are pleiotropic master switches: a single complex governs hundreds of developmental genes simultaneously, so haploinsufficiency for one subunit deregulates entire transcriptional programs in neural progenitors rather than disabling one pathway. This explains the recurring clinical picture across these disorders — intellectual disability with a recognizable facial gestalt and multi-system involvement — and it explains why these conditions, unlike classic enzyme deficiencies, leave a genome-wide methylation fingerprint (an episignature) that can be read diagnostically.

Genomic imprinting breaks the usual rule that a person's two alleles of a gene are functionally equivalent. For a small set of genes, expression depends entirely on which parent the allele came from: one parental copy is epigenetically silenced in the germline and stays silent for life, so the gene is functionally hemizygous even though two physical copies are present. This is biologically counterintuitive — natural selection generally favors diploid backup — and the leading explanation is the parental-conflict (kinship) hypothesis: paternally expressed genes tend to drive fetal growth (the father's genetic interest is a large offspring drawing maximally on maternal resources), while maternally expressed genes tend to restrain growth (the mother's interest is conserving resources across multiple pregnancies). IGF2 (paternal, pro-growth) opposed by H19/CDKN1C (maternal, growth-restraining) at 11p15.5 is the textbook embodiment of this tug-of-war, and it explains why imprinting disorders so often present as overgrowth or undergrowth syndromes.

The imprint is laid down by methylation of imprinting control regions (ICRs) during gametogenesis — a sex-specific erasure-and-reset that is the reason the same DNA region can carry a 'maternal' versus 'paternal' instruction. Humans have roughly 100 imprinted genes, clustered at loci such as 15q11-13, 11p15.5, 7q32, and 20q13; clustering reflects shared regional control by a single ICR.

Because one parental allele is already silenced, losing the active allele leaves no functional copy — so imprinting disorders show parent-of-origin inheritance that looks non-Mendelian. The defining demonstration is that Prader-Willi and Angelman syndromes arise from the same 15q11-13 deletion, distinguished only by which parent transmitted it — paternal deletion gives PWS, maternal deletion gives AS Knoll et al. 1989. This single observation established genomic imprinting as a cause of human disease and reframed how we read pedigrees for these conditions.

A practical corollary follows from the four routes that can knock out the active allele — deletion, uniparental disomy (UPD), imprinting-center defect, and a point variant — they converge on the same expression outcome but are detected by different assays, which dictates the diagnostic algorithm below.

X-chromosome inactivation (XCI) exists to solve a dosage problem. The X chromosome carries over a thousand genes; the Y carries few. Without compensation, XX females would express twice the X-linked gene dose of XY males, which is lethal in the analogous case of autosomal trisomy. XCI silences one entire X in each female cell to equalize output, and the deep reason it produces clinical consequences is how it does so: the choice of which X to silence is made independently and randomly in each cell early in embryogenesis, then locked in.

Initiation — an RNA that acts in cis: Silencing is launched by the XIST long non-coding RNA, transcribed from the future inactive X. Critically, XIST works in cis — it physically coats the very chromosome that produced it, spreading along it and recruiting Polycomb complexes that deposit H3K27me3, followed by DNA methylation and heterochromatinization into the condensed Barr body. This cis-restriction is why only one X (the one expressing XIST) is silenced while its homolog stays active.

Random choice plus mitotic memory equals a mosaic: The choice is random with respect to parental origin, but once a cell commits, all of its descendants inherit the same inactive X. A female is therefore not uniform but a clonal patchwork — roughly half her cells express the maternal X, half the paternal. For an X-linked disorder this mosaicism is the disease modifier: a heterozygous female is a mixture of mutant-expressing and normal-expressing clones, and her phenotype reflects the ratio.

That ratio is usually near 50:50 but can become skewed by chance (few founder cells at the time of choice), by selection (clones expressing a deleterious active allele are outcompeted), or by structural X abnormalities. Skewing toward inactivation of the mutant allele protects a carrier; skewing toward the normal allele unmasks disease — the mechanism behind manifesting carriers of conditions such as Duchenne muscular dystrophy and OTC deficiency, and behind the wide phenotypic range of MECP2-related Rett syndrome in girls.

Diagnostics — why methylation is read, not sequenced: For imprinting disorders the pathology is often not a changed base but a changed mark, so sequencing can be completely normal while the gene is silenced. Methylation assays read the mark directly. The clinical workhorse, MS-MLPA (methylation-specific multiplex ligation-dependent probe amplification), is powerful because it reports two things at once: copy number and methylation at the same probes. A single MS-MLPA at 15q11-13 therefore distinguishes the mechanisms that look identical clinically — a deletion drops copy number, while UPD or an imprinting-center defect leaves copy number normal but methylation abnormal — and that distinction drives recurrence-risk counseling, since a deletion or UPD is usually sporadic whereas an imprinting-center microdeletion can be inherited and carries a high recurrence risk. Bisulfite-based methods (methylation-specific PCR, bisulfite sequencing) add base-level resolution by chemically converting unmethylated cytosines to uracil so that methylation status is read as a sequence difference.

From single loci to genome-wide episignatures: The conceptual leap of the last decade is that disorders of the chromatin machinery do not just dysregulate one gene — they perturb methylation across hundreds of loci in a reproducible, disorder-specific pattern. Profiling these patterns on genome-wide arrays (Illumina EPIC/850K) yields an episignature, a molecular fingerprint. In a systematic evaluation across dozens of Mendelian neurodevelopmental syndromes, distinct and highly accurate blood-derived episignatures were defined and shown to resolve clinically ambiguous cases and reclassify variants of uncertain significance Aref-Eshghi et al. 2020. This matters because it provides functional evidence: a VUS in a chromatin gene that reproduces the known episignature is shown to actually disrupt the protein's job, satisfying ACMG functional criteria — something sequence data alone cannot do.

Therapeutic frontiers — reversibility as the whole point: Because marks are written and erased by enzymes rather than fixed in sequence, they are druggable in a way mutations are not. Alzheimer's disease illustrates both the rationale and the difficulty: the aging brain shows global DNA hypomethylation alongside locus-specific hypermethylation, and a loss of H4K16 acetylation near APP and PSEN1 shifts expression of amyloid-pathway genes. HDAC inhibitors aim to restore acetylation and reopen these loci — but the same broad, genome-wide action that makes them potent also makes them blunt, hitting thousands of genes nonspecifically. That specificity problem is exactly what CRISPR epigenome editors are designed to solve: a catalytically dead dCas9 is steered by a guide RNA to a single locus and fused to an effector (DNMT3A to methylate/silence, or TET1 to demethylate/activate), changing expression at one address without ever cutting or altering the DNA — an especially attractive strategy for imprinting disorders and for silencing toxic repeat expansions.