Samples: tumour and matched normal
WES looks for the mutations a tumour acquired during life, but every person is born with tens of thousands of sites that differ from the reference genome. So the tumour is sequenced together with a normal sample from the same patient: variants found in both are germline, variants found only in the tumour are somatic. How pure the tumour piece is, and how it was preserved, end up in every number downstream.
Paired design: in the tumour, not in the normal
A piece of tumour is never just tumour cells: stroma, vessels and infiltrating immune cells come with it. The share of tumour cells is the tumour purity. The matched normal, usually blood for solid tumours, carries only germline variants. Variants present in the tumour and absent from the normal are the candidate somatic mutations.
Blood cancers are the exception: the blood itself contains tumour cells and cannot serve as the normal. TCGA-LAML, the dataset used in this tutorial, used skin biopsies instead (Ley et al., NEJM 2013).
Why purity matters: picture a mutation present in every tumour cell, on one of the two homologous chromosomes, in a region with no copy-number change. At 100% purity half of all DNA copies carry it, so the VAF is about 50%. At 40% purity only 40% of the cells contribute mutant copies, and the VAF is about 20%. In short, the expected VAF is about purity / 2. Copy-number changes and ploidy break this rule; chapter 12 comes back to it.
FFPE: false mutations left by the fixative
Most clinical samples are formalin-fixed and paraffin-embedded (FFPE). Formalin cross-links and breaks DNA, lowering its integrity, and it deaminates cytosine to uracil. PCR copies uracil as T, so the sequencer reports a C>T that was never in the patient (G>A on the complementary strand).
The damage hits one strand of an original molecule at random, which gives it two signatures. First, a low VAF: only a few molecules at any one site are damaged. Second, it shows up almost only in read pairs of one orientation, a read-orientation bias. A true mutation sits on both strands of the original molecule and appears in both orientations.
Common pitfalls
- Treating tumour-only calls as somatic. Without a matched normal, germline variants can only be removed with population databases and a panel of normals, and rare germline variants slip through in large numbers.
- Tumour cells in the normal. Tissue next to the tumour may already hold tumour cells or precursor clones, and blood is a poor normal for a blood cancer. A true mutation that is also seen in the normal gets filtered as germline.
- Judging calls by a VAF cut-off alone. A low VAF can be an FFPE artefact, or a real mutation in a low-purity sample or a subclone. Look at orientation bias and the spectrum as well.