LSMA

LONDON SCHOOL OF MULTIOMICS & AI

Free course

Free course

0 of 73 lessons

Lesson 5 of 10

Module 1 · Understanding omics

Genomics: reading the variants

Genomics: reading the variants

Genomics: reading the variants

A tumour genome is a record of everything that has gone wrong in its DNA. Genomics finds those changes and asks which of them drive the cancer.

⏱ 12 min readModule 1 · Understanding omicsFree
Lesson at a glance5 steps · about 12 min
  1. Germline and somatic
  2. Four kinds of change
  3. Variant allele fraction
  4. Drivers, passengers and mutational burden
  5. Genome, exome or panel

Germline and somatic

A germline variant is one you were born with, present in every cell, including your blood. A somatic variant arose later in one cell and was passed to its descendants, so in cancer it is in the tumour but not in normal tissue. Cancer studies often sequence a tumour–normal pair from the same person: variants in both are germline; variants only in the tumour are somatic.

This lesson builds on How sequencing works: the reads, quality scores and BAM files described there are the raw material for every variant below.

A variant is any position where a person's DNA differs from the reference genome. The word "germline" comes from the germ cells, the egg and sperm: a germline variant was in the egg or sperm you grew from, so it is copied into every cell of your body and can be passed to your children. An inherited harmful variant in BRCA1, which raises the risk of breast and ovarian cancer, is a germline example. A somatic variant cannot be inherited by children, because it is not in the egg or sperm.

Why sequence normal tissue at all? Every person carries several million germline differences from the reference genome, nearly all of them harmless. Sequencing the tumour alone, you would see all of those as well as the tumour's own changes, with no way to tell them apart. The matched normal sample, usually blood, shows which variants the person already had, and subtracting them leaves the somatic ones. Leaving it out is a common mistake: an inherited variant can then be reported as a tumour mutation.

Germline findings matter in their own right in cancer care. If a harmful BRCA1 variant turns up in the normal sample as well as the tumour, it was inherited: the patient's relatives may carry it too, and genetic counselling is usually offered. Laboratories therefore plan in advance how such findings are reported.

Four kinds of change

Variants are grouped by size, from a single letter up to whole pieces of chromosome. The size matters because each kind is found by a different method and changes the protein in a different way.

Variant typeWhat changesA cancer example
SNV (single-nucleotide variant)One base swapped for anotherBRAF V600E, the commonest of the BRAF mutations found in about half of melanomas
IndelA few bases added or removedEGFR exon 19 deletions in lung adenocarcinoma
CNV (copy-number variant)A whole region gained or lostERBB2 (HER2) amplification in breast cancer
Structural variantLarge pieces rearranged or joinedBCR–ABL1 fusion in chronic myeloid leukaemia

How to read this table

The rows go from the smallest change to the largest. Each example is a real, well-known cancer change, written in the usual shorthand:

  • BRAF V600E describes the change at the protein level: at position 600 of the BRAF protein, the amino acid valine (V) has been replaced by glutamic acid (E). One DNA letter changed, and the protein became permanently switched on.
  • Indel is short for insertion or deletion. An exon is a part of a gene that is kept in the final RNA message and codes for protein. In the EGFR example, a few bases are cut out of exon 19, which again leaves the protein switched on.
  • Amplification means many extra copies of a region. A normal cell has two copies of ERBB2; an amplified tumour may have many more, and makes far more of the protein.
  • A fusion joins parts of two genes. BCR–ABL1 forms when pieces of chromosomes 9 and 22 swap places; the shortened chromosome 22 is known as the Philadelphia chromosome. The drug imatinib was designed to block the fused protein.

Short-read sequencing finds SNVs and small indels well. Copy-number changes are inferred from how many reads fall in a region, and structural variants from reads that span the break, which is where long reads help (How sequencing works).

The type also shapes the effect on the protein. An SNV may change one amino acid, create a stop codon, or change nothing at all, because several codons can name the same amino acid. An indel whose length is not a multiple of three shifts the reading frame, so every codon after it is read wrongly, which usually ruins the protein; this is called a frameshift. An indel of exactly three bases, or a multiple of three, adds or removes amino acids but keeps the rest of the protein in frame. The common EGFR exon 19 deletions are of this in-frame kind, which is why the protein still works, and works too well.

Variant allele fraction

At a variant position, some reads carry the reference base and some the variant. The share carrying the variant is the variant allele fraction (VAF).

An allele is one version of a sequence at a position. Most of your chromosomes come in pairs, one from each parent, so most positions have two copies. If one copy carries a variant and the other does not, the variant is heterozygous, and in a pure sample about half of the reads would carry it. That is why a germline heterozygous variant has a VAF near 0.5, whatever the tumour content of the sample.

VAF is easy to compute, but its meaning depends on context. In a germline sample, a VAF near 0.5 suggests a heterozygous variant and a VAF near 1.0 a homozygous one, carried on both copies. In a tumour sample, VAF also reflects how many of the cells are tumour and how many copies of the region each cell has, as the example below shows. A VAF of 0.30 means little on its own until you know which kind of sample it came from.

Worked example · VAF and tumour purity

36 of 120 reads carry the variant → VAF = 36 ÷ 120 = 0.30

A tumour sample also contains normal cells. If 60% of cells are tumour (the purity) and the mutation is on one of two gene copies in every tumour cell:

expected VAF = purity ÷ 2 = 0.60 ÷ 2 = 0.30

The numbers match, so the mutation is probably in all tumour cells: an early event. A VAF well below purity ÷ 2 suggests a subclone, present in only some tumour cells.

Step by step. Imagine 10 cells from this sample. At 60% purity, 6 are tumour cells and 4 are normal. Each cell has 2 copies of the gene, so there are 10 × 2 = 20 copies in all. Each tumour cell has 1 mutant copy, giving 6 mutant copies; the normal cells have none. The share of mutant copies is 6 ÷ 20 = 0.30. Dividing purity by 2 is a shortcut for exactly this count.

What the answer means. When the observed VAF matches purity ÷ 2, every tumour cell carries the mutation, so it happened in the ancestor cell before the tumour spread out: it is clonal. Had the observed VAF been 0.06 instead, the mutation would be in 0.06 ÷ 0.30 = 0.2 of the tumour cells, about one in five: a subclone that arose later. Two cautions: the shortcut assumes the region has its normal two copies, and with 120 reads the VAF itself carries some chance variation, so small differences should not be over-read.

Genomics · LSMA free course Ten cells from a sample at 60% purity: six tumour cells, each with one mutant and one normal copy of the gene, and four normal cells with two normal copies. Of the 20 copies, 6 carry the variant, so the VAF is 0.30. 1 · 60% PURITY: 6 TUMOUR, 4 NORMAL CELLS variant copy normal copy Tumour cells · 1 variant copy each Normal cells · no variant 2 · COUNT THE COPIES 20 copies in all · 6 carry the variant VAF = 6 ÷ 20 = 0.30 same as purity ÷ 2 = 0.60 ÷ 2
Each cell holds two copies of the gene. Only tumour cells carry the variant, and only on one copy, so the variant's share of all copies is half the share of tumour cells. Schematic, not to scale.
Go deeper: when the two-copy assumption breaks

The shortcut purity ÷ 2 assumes every cell has two copies of the region. Tumours often break that rule, and the VAF moves with it.

Take the same sample: 60% purity, 10 cells, 6 of them tumour. Suppose each tumour cell carries the mutation and has also lost its normal copy, so it holds a single, mutant copy. Mutant copies: 6 × 1 = 6. Normal copies: 4 normal cells × 2 = 8. The VAF is 6 ÷ 14 = 0.43, well above 0.30. Now suppose instead the mutant copy was duplicated to replace the lost one, which is called copy-neutral loss of heterozygosity. Each tumour cell then holds two mutant copies: 12 mutant copies out of 20, a VAF of 0.60. A VAF above purity ÷ 2 is therefore a hint to check the copy number of the region before drawing conclusions about clonality.

Drivers, passengers and mutational burden

A tumour may carry thousands of somatic mutations, but only a handful give it an advantage: the drivers. The rest are passengers. Recurrence across patients, position in a known cancer gene and the predicted effect on the protein help tell them apart. Tumour mutational burden, mutations per megabase sequenced, also matters: a burden of 10 or more per megabase is one regulatory threshold for immunotherapy.

What the terms mean

  • Drivers and passengers. A driver changes the cell's behaviour in a way that helps it grow or survive, for example by switching on a growth signal (BRAF V600E) or switching off a brake. A passenger happened to arise in the same cell lineage and was carried along, without effect.
  • Recurrence. If the same position is mutated in many unrelated patients' tumours, chance is an unlikely explanation; it is more likely that the change helps tumours grow. BRAF V600E is the classic case: one exact change, seen again and again.
  • Mutational burden. A megabase is one million bases. If a test reads 1.5 megabases and finds 15 somatic mutations, the burden is 15 ÷ 1.5 = 10 mutations per megabase. Tumours with many mutations make many abnormal proteins, which gives the immune system more to recognise; that is the reason burden is linked to how well immunotherapy can work. Melanomas and lung cancers in smokers, shaped by sunlight and tobacco smoke, are among the tumours with the highest burdens.

Telling drivers from passengers is often the slowest part of interpreting a tumour. A typical analysis first keeps the variants that change a protein, then checks whether each falls in a gene known to be involved in cancer, and finally looks up whether the same change has been seen in other tumours. Public databases of cancer mutations collect that recurrence evidence. Even then, some variants stay of uncertain significance until more patients or laboratory experiments settle them.

The threshold of 10 mutations per megabase comes from the US Food and Drug Administration's 2020 approval of the immunotherapy pembrolizumab for solid tumours with high mutational burden. It was based on the KEYNOTE-158 trial, analysed by Marabelle and colleagues (2020, The Lancet Oncology), with burden measured by one specific gene panel.

Burden numbers are not fully interchangeable between tests. They depend on how many megabases the panel reads, which kinds of mutation are counted and how germline variants are removed. When you cite a burden value, give the assay that produced it, and be cautious about applying one test's cut-off to another's results.

Genome, exome or panel

These three options trade breadth for depth. With a fixed budget you can read a little of everything, or a lot of a little.

Whole genome
  • All 3.1 billion bases
  • Finds every variant type
  • Most data and highest cost
Whole exome
  • Protein-coding regions, about 1–2% of the genome
  • Cheaper, or deeper for the same cost
  • Misses most non-coding and structural changes
Gene panel
  • Tens to hundreds of chosen cancer genes
  • Very deep, often several hundred reads per base
  • The usual choice in clinical testing
The file trail
  • FASTQ: reads and quality scores
  • BAM: reads aligned to the genome
  • VCF: called variants

Why does a panel go so deep? Depth matters most when a variant is rare in the sample. At a VAF of 0.05, only 1 read in 20 carries the variant. At 30× coverage that is one or two reads, too few to trust; at 500× it is about 25 reads, which is clear evidence. Deep panels are therefore good at finding small subclones and mutations in samples with low purity, which are common in clinical biopsies.

The file trail runs in that order. A VCF (Variant Call Format) file is a plain-text table with one line per variant: the chromosome, the position, the reference base, the variant base, and measures of how well the call is supported. It is far smaller than the BAM file it came from, and it is the file most analyses of variants start from.

Choosing between the three is a question about what you need to find. A rare-disease search with no strong suspect gene often starts with an exome or a genome, because the cause could be anywhere. A cancer test that only needs to know whether a few drug-relevant genes are mutated is usually a panel. One pitfall when reading results: a panel can only report on the genes it covers, so "no mutation found" says nothing about a gene outside it.

04812Chromosome−log10 p1357911131517192122genome-wide significance
Genomics at population scale: a genome-wide association study tests millions of variants at once, which is why its significance line sits so high. Illustrative data.

A genome-wide association study (GWAS) is different from tumour sequencing. It compares germline variants between many people with a disease and many without, and asks which variants are more common in one group. In this Manhattan plot, each dot is one variant, placed along the chromosomes from left to right; its height is −log10 of the p-value, so higher dots mean stronger evidence. The usual genome-wide threshold is a p-value of 5 × 10−8, which is about 7.3 on this scale. It is set so strictly because, with a million or more tests, a looser threshold such as 0.05 would let tens of thousands of variants through by chance.

Key point

A variant call is an estimate, not a fact. It depends on depth, purity and the caller's settings. Always ask how many reads support a variant.

Take two calls with the same VAF of 0.10. One rests on 3 of 30 reads; the other on 30 of 300. Three reads could be sequencing errors, PCR copies or a mapping problem. Thirty independent reads at the same position are much harder to explain away. The VAF alone does not tell you which situation you are in; the read counts do.

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. 45 of 150 reads carry the variant. What is the VAF?

2. A variant is in both the tumour and matched blood. It is most likely:

3. HER2 amplification in breast cancer is which type of change?

Up next

Lesson 6

Epigenomics

Ask BioTutor

Create a free website with Framer, the website builder loved by startups, designers and agencies.