Cell publishes a guide to methods for detecting rare mutations in healthy tissues
Cell publishes a guide to methods for detecting rare mutations in healthy tissues
On October 15, the SMaHT network, a US National Institutes of Health program, published a comparative study examining which DNA sequencing methods are suited to each type of rare genetic change.
After conception, cells divide and pass changes in their DNA to their descendants. This produces cell lineages with slightly different genomes within the same body. A sample of healthy tissue contains a mixture of DNA from thousands of these cells. Detecting a rare somatic mutation, a change that arose in the body's cells after conception, requires distinguishing it from sequencing errors.
The authors assembled four benchmark systems: two mixtures of cell lines, fibroblasts and clones derived from those same cells, and tissues from four deceased donors. They established reference sets of genetic changes for each system. These sets showed which methods detect different types of variants and which miss them.
Large DNA rearrangements are particularly difficult to detect with short reads. These capture small fragments and lose the connection between regions on opposite sides of a rearrangement or repeat. Long reads preserve that connection by spanning longer stretches of DNA. In a benchmark test where altered DNA molecules accounted for 5–20% of the total, long reads detected about 80% of rearrangements at an average coverage of 60 reads per site. Short reads detected roughly 30%, even at 180 reads per site.
Another source of error arises when a person's DNA is compared with a reference genome. A conventional reference consists of a single sequence, whereas a pangenome includes variants from many people's genomes. In a test with an average of 500 reads per site, switching to a pangenome increased precision in identifying single DNA base substitutions from 79% to 89% and raised the proportion of reference variants detected from 48% to 59%. The advantage was particularly pronounced in genomic regions that are difficult to align reliably to a reference.
The August review of genomic mosaicism asked how a particular mutation and the clone that grew from it could be linked to tissue function. The SMaHT benchmark adds a prerequisite: first, we need to establish which changes each method can distinguish from errors and where it misses the signal.
Long reads address one part of this methodological picture. Deep sequencing of DNA from mixed tissue measures the fraction carrying a variant, duplex sequencing reduces errors when detecting rare point substitutions, and single-cell analysis shows how changes are distributed among cells. SMaHT plans to compile a catalog of somatic mutations in 10–15 tissue types from 150 donors to compare these changes across tissues and stages of life.