Lesson 6 of 10
Module 1 · Understanding omics
The epigenome decides which parts of the genome a cell may read. Cancer rewires these switches as surely as it mutates DNA, and unlike mutations, the switches can be reset by drugs.
Change without changing the sequence
Epigenetic changes alter how active a gene is without changing its DNA sequence, and are passed on when a cell divides. The best-studied are DNA methylation and modifications to the histone proteins DNA is wrapped around.
The prefix epi- means "on top of": these are marks sitting on top of the sequence, not changes to the letters themselves. A helpful picture is a piece of sheet music with pencil notes added. The notes stay the same, but the pencil marks say which passages to play and which to skip.
The DNA in a human cell is about two metres long if stretched out, yet it fits in a nucleus a few thousandths of a millimetre across. It fits because it is wound around histones, small proteins that act as spools. One spool, made of eight histone proteins with DNA wrapped around it, is called a nucleosome. Packed nucleosomes form chromatin, the mix of DNA and protein that makes up chromosomes. Where chromatin is tightly packed, the machinery that reads genes cannot reach the DNA; where it is loose, it can.
In Genomics you looked at changes to the DNA letters themselves. This lesson looks at the layer above them: the marks that decide which of those letters are read in a given cell.
Why does this layer matter? Almost every cell in your body carries the same genome, yet a neuron and a liver cell behave very differently. Much of that difference is epigenetic: each cell type keeps a different set of genes open and a different set packed away. One practical consequence follows for data. Epigenomic profiles are specific to a tissue and a cell type, so a methylation profile measured in blood tells you about blood cells, not directly about the liver or a tumour elsewhere.
DNA methylation
Methylation is mostly added to cytosines followed by a guanine, called CpG sites. Many promoters sit in CpG-rich CpG islands; a heavily methylated promoter usually means the gene is off. Cancer shows genome-wide loss of methylation alongside gains at specific promoters that silence tumour suppressors. In some colorectal cancers, the DNA repair gene MLH1 is silenced this way, with no mutation in the gene.
It helps to know what a normal pattern looks like before reading a cancer one. In healthy human cells, most CpG sites scattered across the genome are methylated, while the CpG islands at promoters are mostly unmethylated, which leaves those genes free to be switched on. Cancer turns this pattern upside down in two places at once: the scattered sites lose methylation and some islands gain it.
Location matters as much as level. Methylation at a promoter usually goes with a silent gene, but methylation inside the body of a gene often goes with an active one. A common mistake is to read every methylated site as "gene off". Before you interpret a result, check where the CpG sits relative to the gene.
Unpacking the terms
- Methylation is the addition of a methyl group, one carbon atom with three hydrogens (CH3), to the base cytosine (C). The result is called 5-methylcytosine.
- In CpG, the "p" stands for the phosphate that links the C to the next base, a G, along the same strand. It distinguishes a C followed by a G from a C paired with a G on the opposite strand.
- A promoter is the stretch of DNA immediately before the start of a gene, where the machinery that copies the gene into RNA binds. Methylation there blocks some of the proteins that switch the gene on and attracts proteins that pack the chromatin tightly.
- A tumour suppressor is a gene whose normal job is to restrain cell growth or repair damage. Losing it, by mutation or by silencing, removes a brake.
- The pattern is kept through cell division because an enzyme copies the methylation onto each new DNA strand. This is why a silenced gene stays silent in all of a cell's descendants.
In the MLH1 example, the gene's sequence is perfect but its promoter is heavily methylated, so no MLH1 protein is made. MLH1 is part of the system that corrects copying mistakes in DNA, so these tumours build up many small errors. The result is the same as if the gene had been mutated, but only a methylation test, or a test of MLH1 protein, would show why.
Worked example · the beta value
Methylation arrays measure methylated (M) and unmethylated (U) signals at each CpG, combined into a beta value from 0 to 1. An offset of 100 keeps it stable when both signals are weak.
M = 1,800 U = 200 → 1,800 ÷ 2,100 = 0.86
About 86% of DNA copies in the sample are methylated at this site.
Step by step. First add up the bottom of the fraction: 1,800 + 200 + 100 = 2,100. Then divide: 1,800 ÷ 2,100 = 0.857, which rounds to 0.86. To see why the offset is there, imagine a site where both signals are tiny, M = 6 and U = 2. Without the offset, β would be 6 ÷ 8 = 0.75, a confident-looking number built on almost no signal. With it, β = 6 ÷ 108, about 0.06, which is pulled towards zero and much less likely to be mistaken for strong methylation.
What the answer means. A methylation array is a glass chip that measures hundreds of thousands of chosen CpG sites at once. A beta value near 0 means almost no copies are methylated at that site; near 1, almost all are. A value in the middle, such as 0.5, has more than one possible cause: one of the two copies in every cell could be methylated, or the sample could be a mix of methylated and unmethylated cells, such as tumour mixed with normal tissue. As with VAF in Genomics, purity affects what you see.
In a heatmap like this, each row is one CpG site and each column one sample, and the colour shows the beta value. "Clustered" means the rows and columns have been reordered so that similar ones sit side by side; the tree drawn along the edge, called a dendrogram, shows which samples are most alike. The clustering is given no labels. That tumours group with tumours and normals with normals shows the methylation pattern alone carries the difference.
Two cautions apply when you read a heatmap like this on real data. First, samples processed on different days or arrays can cluster by batch rather than by biology, so check that the groups are not simply the processing dates. Second, a tissue sample is a mix of cell types, and each type has its own pattern. Two groups can differ because their cell mix differs, for example more immune cells in one, not because any single cell has changed.
Go deeper: how sequencing sees a methyl group
A sequencing machine reads only A, C, G and T, so a methylated C looks just like an ordinary one. The usual trick is a chemical called sodium bisulfite. It turns unmethylated C into U (uracil), which is copied as T during PCR. Methylated C is protected and stays C. The method was introduced by Frommer and colleagues in 1992.
After treatment, the reads are compared with the reference. At each CpG, a read showing C was methylated there; a read showing T was not. The share of reads still showing C is the methylation level at that site. One known blind spot: bisulfite cannot tell 5-methylcytosine from a related mark, 5-hydroxymethylcytosine, which is common in the brain. Both stay as C, so the two are reported together unless an extra method is used.
Histone marks and open chromatin
Tags on histone tails change how tightly DNA is wound. They are named by histone, position and tag: H3K27ac is histone H3, lysine 27, acetylated.
Each histone protein has a loose tail that sticks out of the nucleosome. Enzymes add and remove small chemical tags on these tails. In the name, K is the one-letter code for the amino acid lysine; me3 means three methyl groups and ac an acetyl group. Acetylation removes the positive charge that helps the tail hold on to DNA, which is negatively charged, so the DNA loosens. Methylation of histones can mean on or off, depending on exactly which position carries it.
The enzymes that handle these tags are often described in three groups. Writers add a tag, erasers remove it, and readers are proteins that recognise a tag and bring in other machinery, such as complexes that pack or open chromatin. This is why a mark has an effect at all: the tag itself is small, but the reader proteins that bind it do the work. Mutations in writer, eraser and reader genes are common in cancer, which links this layer back to the genome.
| Mark | Where it is found | Effect |
|---|---|---|
| H3K4me3 | Promoters of active genes | Active |
| H3K27ac | Active enhancers and promoters | Active |
| H3K27me3 | Genes silenced by Polycomb proteins | Repressive |
| H3K9me3 | Tightly packed heterochromatin | Repressive |
How to read this table
Each row is one mark. Notice that lysine 27 appears twice with opposite effects: acetylated (H3K27ac) it marks active regions, triply methylated (H3K27me3) it marks silenced ones. The position alone does not tell you the effect; you need the tag as well. Three terms are new here:
- An enhancer is a stretch of DNA that boosts the activity of a gene, sometimes from many thousands of bases away. Enhancers are often specific to one cell type.
- Polycomb proteins are a family of proteins that keep genes switched off, notably genes that control development; they add the H3K27me3 mark.
- Heterochromatin is chromatin packed so tightly that genes within it are rarely read. Loosely packed, readable chromatin is called euchromatin.
ChIP-seq and CUT&RUN pull down DNA carrying one chosen mark. ATAC-seq uses an enzyme that cuts only loosely packed DNA, so its reads mark open chromatin.
All three methods end with sequencing, and their reads pile up into peaks where the mark or the opening is found (The central dogma, as data). In ChIP-seq (chromatin immunoprecipitation followed by sequencing), an antibody that recognises the chosen mark pulls out the nucleosomes carrying it, together with their DNA. CUT&RUN does the same job the other way round: the antibody guides a cutting enzyme to the mark inside intact nuclei, and only the DNA cut free is sequenced, which needs far fewer cells. In ATAC-seq (assay for transposase-accessible chromatin), the enzyme, a transposase, inserts sequencing adapters wherever it can reach the DNA. It cannot get between packed nucleosomes, so its reads collect in open regions, such as active promoters and enhancers.
On real data, the main result of these methods is a list of peaks: one row per region, with its chromosome, start and end position and a score for how strongly reads are enriched there. Software called a peak caller decides where a pile of reads is high enough to count. Comparing peak lists between samples, for example tumour and normal, shows which promoters or enhancers have opened or closed.
Two pitfalls are worth knowing early. ChIP-seq needs a control, usually a sample of the same chromatin sequenced without the antibody, called the input; some regions give extra reads in any experiment, and without a control they look like false peaks. In ATAC-seq, a large share of reads can come from mitochondrial DNA, which is not packed into nucleosomes and so is easy for the enzyme to reach. These reads are usually removed before peaks are called.
The methods in this section each have a founding paper: ATAC-seq in Buenrostro and colleagues (2013, Nature Methods) and CUT&RUN in Skene and Henikoff (2017, eLife). Reference maps of histone marks and open chromatin across many human tissues come from the Roadmap Epigenomics Consortium (2015, Nature) and the ENCODE project.
The table's "effect" column describes strong associations, not fixed rules. A mark tends to be found at active or silent genes, but it does not prove the gene's state on its own. Peak results also depend on the antibody's quality and the software used to call peaks, so state both when you cite a result.
Why it matters in the clinic
Epigenetic marks are reversible, so they are drug targets: azacitidine and decitabine are used in myelodysplastic syndromes and acute myeloid leukaemia. Methylation patterns also fingerprint a tumour's cell of origin, and methylation profiling is now used to classify brain tumours.
Myelodysplastic syndromes are bone-marrow disorders in which blood cells form abnormally, and acute myeloid leukaemia is a fast-growing cancer of the same blood-forming cells. Azacitidine and decitabine are built into DNA in place of cytosine and trap the enzymes that copy methylation onto new strands. With each cell division, methylation is diluted, and some silenced genes can be switched back on. A mutation cannot be undone by a drug in this way; a methylation mark can.
The fingerprint idea works because every cell type has its own methylation pattern, and a tumour keeps much of the pattern of the cell it started from. A computer classifier compares a tumour's pattern with reference patterns from many known tumour types. You will see this in the case study, where a brain tumour is classified partly by its methylation.
Methylation can also predict how a tumour will respond to treatment. In glioblastoma, an aggressive brain tumour, the promoter of the DNA repair gene MGMT is methylated in some patients. MGMT removes the damage caused by the chemotherapy drug temozolomide, so when the gene is silenced the drug works better. Testing MGMT promoter methylation is a routine part of the work-up for these tumours.
A newer use is in blood tests: tumours shed small fragments of DNA into the bloodstream, and their methylation patterns can hint at whether a cancer is present and where it started. One caution for all of these uses is that marks also change with age, tissue type and how a sample is handled, so a result must be compared with suitable normal samples of the same kind.
Key point
A gene can be switched off without any mutation. Sequencing DNA alone misses tumour suppressors silenced by methylation.
A second example: CDKN2A, which encodes the p16 protein, is a brake on cell division. In many cancers it is lost by deletion, but in others its promoter is methylated and the gene is silent though intact. A DNA panel would report CDKN2A as normal in the second group; a methylation test, or a stain for p16 protein, would show that the brake is gone.
Further reading · Module 1Sources for this module and the rest of the course are gathered on one page. · All reading for this course
Check your understanding
1. M = 700, U = 200. What is the beta value?
2. Which mark is the usual marker of active enhancers?
3. What does ATAC-seq measure?
Up next
Lesson 7
Bulk transcriptomics