Epige­netics

The genome is often described as a blueprint, a complete set of instructions for building an organism. The metaphor is not entirely wrong. The genome does contain the information required to specify every protein the organism can make. But it is wrong in a way that matters enormously for understanding what a human being actually is.

A blueprint specifies a fixed output. The same blueprint, followed in the same way, produces the same building. The genome does not work like this. Every cell in the human body, from a neuron in the prefrontal cortex to an insulin-secreting beta cell in the pancreas to a keratinocyte in the skin, contains exactly the same 3 billion base pairs of DNA sequence. The sequence is identical. The cells are profoundly different in their shape, their function, their protein content, their lifespan, and their response to signals. A beta cell secretes insulin. A neuron fires action potentials. A keratinocyte produces keratin. All three are reading the same genomic text, but they are reading different parts of it, at different times, at different intensities.

The question of how the same sequence produces hundreds of different cell types, and how the specific identity of each cell type is maintained across decades of continuous cell division, is the central question of epigenetics. The answer involves a layer of information sitting above the DNA sequence, written not in nucleotides but in chemical modifications to the DNA itself and to the histone proteins around which DNA is wound. These modifications do not change the sequence. They change what the sequence does. The epigenome is not a second genome. It is the instruction set that tells the genome how to read itself.

The most extensively studied epigenetic modification is DNA methylation: the addition of a methyl group to the fifth carbon of a cytosine base, producing 5-methylcytosine. In mammals, methylation occurs almost exclusively at CpG dinucleotides, positions in the genome where a cytosine is immediately followed by a guanine. The exceptions are CpG islands: stretches of several hundred to several thousand base pairs with high CpG density, typically located near the promoters of genes, that are generally unmethylated in active genes.

DNA methylation is carried out by a family of enzymes called DNA methyltransferases (DNMTs). DNMT3A and DNMT3B establish new methylation patterns, the de novo methyltransferases. DNMT1 is the maintenance methyltransferase: it recognises hemi-methylated DNA produced after replication, where one strand carries the original methylation pattern and the new strand does not, and methylates the new strand to match the original. This is the mechanism by which epigenetic patterns are copied when a cell divides. The methylation pattern on the parental strand serves as a template for re-establishing the same pattern on the daughter strand, just as the base sequence of the parental strand serves as a template for replicating the sequence.

In general, methylation of CpG islands near gene promoters is associated with gene silencing: the methylated cytosines are recognised by proteins that recruit chromatin-compacting complexes, producing a local chromatin environment that excludes the transcriptional machinery. Unmethylated CpG islands near promoters are associated with gene activity. DNA methylation patterns can be used to identify cell type, to classify tumours, and to estimate biological age.

Adrian Bird at the University of Edinburgh identified in 1992 the proteins that read methylated DNA: the methyl-CpG-binding domain (MBD) proteins, which bind selectively to methylated CpGs and recruit gene-silencing complexes. Bird's work provided the molecular mechanism by which a chemical mark on DNA is translated into a change in gene expression: methylation recruits readers, readers recruit silencers, silencers compact chromatin, compacted chromatin excludes transcription factors.

DNA in eukaryotic cells is wound around protein spools called histones, approximately 146 base pairs of DNA wrapped 1.65 times around a core of eight histone proteins in a structure called the nucleosome. The 2 metres of DNA in a human cell nucleus is packaged into roughly 30 million nucleosomes, linked by short stretches of linker DNA, forming a beaded chain that is further folded and compacted into higher-order chromatin structures.

The histone proteins have unstructured N-terminal tails that protrude from the nucleosome core and are subject to a remarkable variety of post-translational modifications: methylation, acetylation, phosphorylation, ubiquitination, and others, each modification occurring at specific amino acid residues. These modifications constitute what has been called the histone code: a combinatorial language in which specific combinations of modifications at specific positions on specific histones create binding platforms for specific regulatory proteins, determining whether the local chromatin is in an open, transcriptionally permissive state or a closed, repressive state.

Histone acetylation, the addition of an acetyl group to lysine residues on histone tails by enzymes called histone acetyltransferases (HATs), generally promotes gene expression by neutralising the positive charge of lysine, reducing the affinity of the histone for the negatively charged DNA and loosening chromatin structure. Enzymes that remove acetyl groups, histone deacetylases (HDACs), generally promote gene silencing. Histone methylation at different positions has different effects: methylation of H3K4 is associated with active promoters, while methylation of H3K9 and H3K27 is associated with silencing. The same chemical modification, methylation, produces opposite regulatory effects depending on which amino acid it occurs on.

X-inactivation is the process by which one of the two X chromosomes in every cell of a female mammal is transcriptionally silenced. Mary Lyon proposed in 1961 that one X chromosome in each cell was permanently inactivated early in development, the choice of which X to inactivate being random and subsequently maintained clonally through all cell divisions. This is the Lyon hypothesis, now confirmed in full. The inactivated X chromosome becomes the Barr body, a compact, darkly staining structure visible in the nucleus. Its inactivation is maintained by a non-coding RNA called XIST, which is transcribed specifically from the inactive X and coats it in a molecular blanket of RNA that recruits silencing complexes across the entire chromosome.

The consequence is that every female mammal is a mosaic: different cells in the body have silenced different X chromosomes, the maternal or the paternal copy chosen randomly during early development. In female cats with the orange and black tortoiseshell coat pattern, the two differently pigmented cell clones are directly visible as patches of colour on the fur, each patch derived from a cell in which one or the other X chromosome was inactivated. The tortoiseshell pattern is a macroscopic demonstration of epigenetic mosaicism.

Genomic imprinting is the phenomenon by which specific genes are expressed exclusively from either the maternally or paternally inherited chromosome, not from both. Approximately 100 genes in the human genome are imprinted. The imprint is carried by differential DNA methylation of the maternal and paternal alleles.

The evolutionary explanation for genomic imprinting is contested but the most widely accepted account is the parental conflict hypothesis, proposed by David Haig at Harvard in the 1990s. Paternally imprinted genes (expressed only from the maternal copy) tend to be growth-restricting. Maternally imprinted genes (expressed only from the paternal copy) tend to be growth-promoting. Haig argued that this reflects a genomic conflict between maternal and paternal interests in mammalian pregnancy: the paternal genome benefits from maximising the transfer of maternal resources to the current offspring, while the maternal genome benefits from distributing resources more evenly across current and future offspring. The imprinting patterns reflect an evolutionary arms race between the two parental genomes, fought out at the level of epigenetic gene regulation in every cell of every mammalian embryo.

The process by which a single fertilised egg generates 37 trillion cells of more than 200 distinct types is fundamentally an epigenetic process. The genome does not change. The epigenome is systematically remodelled as development proceeds, with each cell fate decision involving a change in the pattern of DNA methylation and histone modification that commits the cell to a specific developmental trajectory and excludes the alternatives.

Conrad Waddington proposed his famous epigenetic landscape metaphor in 1957: a ball rolling down a hillside divided by ridges into valleys, each valley representing a possible cell fate. As development proceeds, the ball rolls into one valley or another at each branching point, and the ridges between valleys grow higher, making it progressively harder for the cell to return to a more primitive state or adopt a different fate. The landscape is not given in advance. It is created by the gene regulatory network as it operates during development.

The power of Waddington's metaphor is demonstrated by the experiments of Shinya Yamanaka and colleagues at Kyoto University, published in 2006. Yamanaka showed that fully differentiated adult cells, specifically mouse skin fibroblasts, could be reprogrammed into induced pluripotent stem cells (iPSCs) by introducing just four transcription factors: Oct4, Sox2, Klf4, and c-Myc. These four proteins, produced at sufficient levels, were enough to push the ball back up the epigenetic landscape, erasing the accumulated epigenetic marks of differentiation and restoring a pluripotent state. Yamanaka received the Nobel Prize in Physiology or Medicine in 2012. The result demonstrated that cell identity is not a property of the DNA sequence. It is a property of the epigenome, and the epigenome can be rewritten.

The most extensively studied human case of environmental effects on the epigenome involves a precisely defined episode of severe nutritional deprivation: the Dutch Hunger Winter of 1944 to 1945, when a Nazi blockade of food supplies to the western Netherlands reduced daily caloric intake in the affected population to as low as 400 to 800 calories for several months.

The children born to women who were pregnant during the famine were studied systematically over subsequent decades. Those who had been in utero during the famine showed elevated rates of obesity, cardiovascular disease, type 2 diabetes, and mental health disorders compared to siblings born before or after the famine. Critically, these effects differed depending on the trimester of pregnancy during which famine exposure had occurred. Children exposed during the first trimester, when early developmental programming of the epigenome is occurring, showed stronger metabolic effects than those exposed later.

Molecular studies of individuals exposed to the Dutch Hunger Winter famine in utero found differences in DNA methylation at specific genomic loci compared to unexposed siblings, most notably at the IGF2 gene, a growth factor gene subject to genomic imprinting. Bastiaan Heijmans and colleagues at Leiden University published these findings in 2008, six decades after the famine, demonstrating that the epigenetic differences established during fetal development had persisted for six decades. The environmental exposure had written instructions onto the genome that six decades of subsequent experience had not erased.

The Dutch Hunger Winter data are important but require careful interpretation. The persistent health differences between famine-exposed and unexposed individuals are established. The epigenetic differences at specific loci are established. The causal chain from the specific epigenetic differences to the specific health outcomes has not been fully demonstrated. What the Dutch Hunger Winter evidence shows with confidence is that the nutritional environment during fetal development has lasting consequences for gene regulation and health outcomes that persist for decades.

The most controversial area of epigenetics research concerns whether epigenetic changes acquired in response to experience in one generation can be transmitted to subsequent generations through the germline. Transgenerational epigenetic inheritance (TEI) has been robustly demonstrated in plants and in certain invertebrates. In mammals, including humans, the evidence is more complicated and more contested, for a fundamental mechanistic reason.

During the formation of the germ cells, the developing sperm and eggs undergo a process of epigenetic reprogramming: a near-complete erasure of the existing DNA methylation patterns and histone modifications, followed by re-establishment of a new epigenetic landscape appropriate to the developing organism. This reprogramming occurs at two stages in mammalian development. If reprogramming is complete, epigenetic marks acquired by parents during their lifetime should be erased before they are transmitted to offspring.

Michael Meaney and colleagues at McGill University demonstrated in a series of papers from 1997 onward that maternal care behaviour in rats, specifically the amount of licking and grooming a mother rat provides to her pups in the first week of life, has lasting effects on the epigenetic state of the glucocorticoid receptor gene in the hippocampus of the pups, effects that persist into adulthood and influence the pups' own stress responses and maternal behaviour. However, the transmission mechanism was behavioural, not epigenetic in the strict sense: the experience of being well-cared-for produced an epigenetic change in the pup's own brain, which influenced the pup's own later behaviour, which produced the same epigenetic change in the next generation. The information was not transmitted through the germline. It was transmitted through experience.

The distinction between experience-driven epigenetic change and germline epigenetic inheritance matters enormously for the relationship of these findings to evolutionary biology. True germline transmission would imply a mechanism by which acquired characteristics could be inherited, a Lamarckian mode of inheritance that the neo-Darwinian synthesis excluded on mechanistic grounds. The current evidence for this in mammals is limited, contested, and does not in any established case involve the transmission of a specific environmentally induced epigenetic mark through the germline across multiple generations with clear phenotypic effects. The behavioural transmission of stress responses demonstrated by Meaney and colleagues is remarkable and important. It is not Lamarckian inheritance.

One of the most striking applications of epigenetic analysis to human biology is the epigenetic clock: the observation that DNA methylation patterns change in a highly predictable, tissue-specific manner with age, such that the methylation state of a set of specific CpG sites can be used to predict a person's chronological age with accuracy approaching 3 to 5 years.

Steve Horvath at UCLA published in 2013 a multi-tissue epigenetic clock based on the methylation status of 353 CpG sites that predicted chronological age across all tissues and cell types examined, including brain, heart, muscle, kidney, liver, and blood, with a correlation of 0.96 between predicted and actual age. The Horvath clock and subsequent improved versions permit the quantification of epigenetic age acceleration: the difference between a person's methylation-predicted age and their chronological age. Individuals whose epigenetic age exceeds their chronological age show elevated risks of age-related diseases and mortality, independent of known risk factors.

The epigenetic clock reveals that ageing has a molecular signature that is written progressively into the epigenome, measurable with precision, and modifiable. In 2020, David Sinclair and colleagues at Harvard demonstrated partial epigenetic reprogramming in mouse retinal ganglion cells, reversing epigenetic ageing as measured by the clock and restoring visual function in aged and injured mice. The prospect of epigenetic rejuvenation is now an active area of research, though whether the epigenetic clock causes ageing phenotypes or merely reflects them remains a central unresolved question.

Cancer is a disease of the genome, but it is equally a disease of the epigenome. The epigenetic landscape of a cancer cell is profoundly abnormal: globally reduced DNA methylation across the genome, combined with focal hypermethylation at specific CpG islands near tumour suppressor gene promoters, producing aberrant silencing of genes that would normally constrain cell proliferation.

The global hypomethylation of cancer cell genomes was first described by Andrew Feinberg and Bert Vogelstein at Johns Hopkins in 1983, one of the first observations that cancer involved systematic epigenetic change rather than only genetic mutation. Hypomethylation of repeated sequences, which are normally heavily methylated and silenced, can reactivate transposable elements, contributing to genomic instability in cancer cells. The focal hypermethylation of tumour suppressor gene promoters silences critical growth-control genes by an epigenetic mechanism equivalent in its functional effect to a mutation: the gene is there, the sequence is intact, but it cannot be expressed.

The reversibility of epigenetic changes, unlike mutations, makes them attractive therapeutic targets. Epigenetic drugs, agents that inhibit DNA methyltransferases or histone deacetylases, have been approved for the treatment of specific haematological malignancies. The DNMT inhibitors azacitidine and decitabine cause global demethylation of the cancer cell genome, reactivating silenced tumour suppressor genes and restoring growth control. These drugs demonstrate that epigenetic alterations in cancer cells are pharmacologically accessible.

The epigenetic regulation of gene expression is not carried out by DNA methylation and histone modification alone. A large and functionally diverse class of RNA molecules that do not encode proteins plays critical roles in establishing and maintaining epigenetic states.

MicroRNAs (miRNAs) are small RNA molecules of approximately 22 nucleotides that bind to complementary sequences in messenger RNAs and direct their degradation or translational repression, providing a post-transcriptional layer of gene regulation. Approximately 1,000 miRNA genes are encoded in the human genome, and each miRNA can regulate multiple target genes simultaneously. The discovery of miRNAs by Victor Ambros and Gary Ruvkun in C. elegans in 1993 and 2000 respectively, recognised with the Nobel Prize in Physiology or Medicine in 2024, fundamentally extended the understanding of how gene expression is regulated.

Long non-coding RNAs (lncRNAs) are RNA molecules greater than 200 nucleotides in length that carry out structural and regulatory roles in chromatin organisation and gene regulation. XIST, the RNA responsible for X chromosome inactivation, is a lncRNA. HOTAIR, a lncRNA transcribed from the HOXC gene cluster, regulates the expression of the HOXD cluster on a different chromosome by serving as a scaffold for the recruitment of repressive chromatin-modifying complexes. The discovery that RNA molecules could serve as guides for chromatin-modifying machinery to specific genomic locations provided an elegant solution to the problem of how writers of epigenetic marks are targeted to the right place.

Epigenetics reveals that the relationship between an organism and its genome is not the relationship between a building and its blueprint. It is more like the relationship between a performance and a score.

A musical score contains the information from which a performance can be produced. But the score does not determine the performance in the way that a blueprint determines a building. The same score can produce radically different performances depending on the performer, the instrument, the tempo chosen, the dynamics applied at each phrase. A human genome is like a score in this sense: it specifies the proteins that can be made, but not the pattern in which they will be made, not the intensity at which each gene will be expressed in each cell type at each moment in development and response to environment.

Nature is the genome sequence, invariant from conception across the lifetime. Nurture is not opposed to nature. Nurture operates on nature by writing instructions into the epigenome, instructions that modify the output of the genome without changing the genome itself. The environment does not bypass the genome. It annotates it.

The sequence is the same in every cell. The annotation is different in every cell type, different between individuals, different across a lifetime, and responsive, within limits, to experience. The organism, at the molecular level, is not its genome. It is its genome being read.