Mutation spectra and signatures
Two single-base substitutions can have entirely different causes: a C>T and a C>A rarely come from the same process. Counting a sample’s mutations by substitution type gives its spectrum; adding the neighbouring bases turns the spectrum into a fingerprint that points to ageing, UV light, tobacco or repair defects, and also exposes bench artefacts such as FFPE and oxidative damage.
Six substitution classes and Ti/Tv
A base-pair substitution can be described from either strand: G>A on one strand is C>T on the other. By convention the pyrimidine (C or T) is used as the reference, which folds all single-base substitutions into six classes: C>A, C>G, C>T, T>A, T>C, T>G. C>T and T>C are transitions, the other four are transversions, and Ti/Tv is the ratio of the two counts.
Ti/Tv is a common QC metric for germline calling: empirical values for germline SNVs are about 2.0–2.1 genome-wide and 2.8–3.0 in exomes, and a clearly lower value suggests many false positives. Those values apply to germline variants only. The Ti/Tv of somatic mutations depends on the processes behind them and has no fixed “normal” value.
The widget shows the six-class fractions for every TCGA-LAML patient with at least 3 SNVs, plus the cohort’s Ti/Tv. C>T dominates, consistent with spontaneous deamination of methylated cytosine, a clock-like process that accumulates with age. Each patient has only about ten mutations, so single-sample fractions swing widely; read the cohort as a whole rather than any one patient.
96 contexts and mutational signatures
Six classes are coarse. Taking one base on each side into account gives 6 classes × 16 flanking combinations = 96 categories: the SBS96 spectrum. Each mutational process leaves its own pattern over these 96 categories, a mutational signature:
- SBS1: C>T at CpG, from spontaneous deamination of 5-methylcytosine, accumulating with age;
- SBS2 / SBS13: C>T and C>G in a TCW context, linked to APOBEC cytidine deaminases;
- SBS4: mainly C>A, linked to tobacco;
- SBS7: C>T at dipyrimidines, linked to UV light;
- SBS6, SBS15 and others: linked to mismatch-repair deficiency.
Signatures are usually extracted by non-negative matrix factorisation (NMF): the samples × 96 matrix is split into a few signatures and each sample’s exposure to them, and each signature is matched to the COSMIC catalogue by cosine similarity. In maftools, trinucleotideMatrix fetches the flanking bases (it needs the reference genome as a BSgenome package), extractSignatures runs the NMF and compareSignatures matches against COSMIC. COSMIC also lists signatures thought to be sequencing artefacts, such as SBS45, linked to 8-oxoG.
Common pitfalls
- Judging somatic calls by germline Ti/Tv values. Somatic Ti/Tv depends on the mutational processes.
- Reading FFPE C>T as biology. Rule out artefact first: look at the VAF distribution, orientation bias and sample type.
- Single-sample signature analysis on few mutations. With few mutations, every exposure is unstable.
- Treating a signature name as a cause. A signature is a statistical pattern pointing to a likely process; it needs clinical context and other evidence.