A boy of six with focal epilepsy, a hemimegalencephaly on imaging and a segmental skin finding has a blood trio exome. It is negative. The variant caller's default settings expect heterozygous calls near half of reads and homozygous near all; a variant supported by one read in nine in the child, and by none in either parent, is filtered as noise before anyone sees it.
The same variant, in a gene of the growth pathway known for somatic overgrowth disorders, is found four years later at a fraction of one in three in tissue from a resection. It was in the blood exome at a low fraction, in the raw data, below a threshold nobody had set with him in mind.
What the field knows
Mosaicism is the presence of a variant in some cells and not others, arising from a mutation after fertilisation. The fraction of cells that carry it depends on when and where it arose, and so does the fraction of sequencing reads that show it in any given tissue. Biesecker and Spinner (2013) set out the clinical range: from disorders that exist only in mosaic form, because a constitutional mutation would be lethal, through segmental presentations of otherwise familiar conditions, to the parental mosaicism that explains an apparently de novo variant recurring in a second child.
The sequencing consequence is a low allele fraction. A heterozygous germline variant is expected near half of reads; a mosaic one can be anywhere below that, and in blood it is often far below the fraction in the affected tissue. Variant callers tuned for germline analysis apply priors and thresholds that treat low-fraction calls as error, which they usually are, and the mosaic variant is removed with them. Methods built for low-fraction detection exist, developed for tumour sequencing and adapted for mosaic disease, and they trade sensitivity against a higher rate of false calls that then need confirmation.
Studies that looked found more than expected. Cao and colleagues (2019), reviewing a large clinical exome series, found mosaic variants in the proband or a parent in a measurable share of cases, including parental mosaicism that changed recurrence counselling, and most of them had been detected only because the laboratory had lowered its thresholds deliberately. Studies of focal cortical malformations and segmental overgrowth have shown that the causative variant is frequently absent from blood at any detectable fraction and present in the affected tissue, so that the choice of tissue is part of the test design (Jamuar et al., 2014).
Parental mosaicism matters for the family even when the child's variant is a clear germline call. A variant called de novo in the child may be present at a low fraction in a parent's blood or gonads, and the recurrence risk for the next pregnancy is then well above the figure quoted for a new mutation. Finding it requires looking, at the parental reads at the child's variant position, below the germline threshold.
Where practice breaks
The threshold is set once, for everyone. A pipeline tuned to keep false positives low in a typical germline case applies the same allele-fraction floor to a child whose phenotype says mosaic. The case's own features, a segmental skin finding, an asymmetric malformation, a focal lesion, should have changed the threshold, and nothing in the pipeline reads features.
The tissue is wrong and the report does not say so. Blood is drawn because blood is what the service takes. For a brain or skin-limited mosaic disorder, a negative blood result is close to uninformative, and the report that says no causative variant was identified does not say that the relevant tissue was never tested.
Parents are checked for presence and not for fraction. The trio analysis asks whether the child's variant is in the parents and gets the answer no, because the parental calls fall below the germline threshold. The de novo label follows, with its low recurrence risk, and the parent who carries the variant in a fraction of cells is never found.
Low-fraction candidates, when kept, are mistrusted uniformly. A one-in-nine call in the right gene for the phenotype, in a child whose imaging says mosaic, deserves confirmation; a one-in-nine call in a random gene deserves dismissal. A pipeline that cannot tell them apart dismisses both, and a reviewer facing a page of low-fraction noise does the same. The phenotype is what separates the two calls, and it is the one input the filter never receives.
How Tulparus handles it
The case model carries the features that argue for mosaicism, segmental or asymmetric findings, focal lesions, a phenotype known to present in mosaic form, and when they are present the workbench marks the case as one in which low-fraction variants are expected. The low-fraction candidates in genes relevant to the phenotype are then surfaced as a separate list, with their allele fractions, read support and strand balance shown, for a person to judge, rather than being dropped with the noise.
Tissue is part of the hypothesis. A hypothesis of a mosaic disorder confined to brain or skin records that blood is a poor tissue for it, and the proposed next step is a sample from affected tissue where one can be obtained, or from a surrogate such as saliva, buccal cells or cultured fibroblasts where it cannot, with what each would show. The report, rendered from the case, says that the tissue tested may not carry the variant and names the tissue that would.
For every variant called de novo, the parental reads at that position are examined below the germline threshold, and a parent with supporting reads is flagged for confirmation by a sensitive assay. The recurrence conversation then rests on a measurement rather than on the absence of a call.
For the boy above, the one-in-nine call in a growth pathway gene would have appeared on the mosaic candidate list on the first day, beside his imaging and skin findings, with a deep sequencing confirmation proposed and, if negative in blood, a tissue sample named as the next step. No variant would have been reported on a low-fraction call alone; the point is that it would have been seen.
What you can do tomorrow
The raw data usually hold the answer. The work is to look below the default.
- For any case with segmental, asymmetric or focal features, re-examine the variant calls in the relevant genes with the allele-fraction floor lowered, and confirm anything plausible by a sensitive assay.
- For every de novo variant you report, look at the parental reads at that position below the germline threshold, and say in the report whether parental mosaicism was assessed.
- When blood is the wrong tissue for the hypothesis, say so in the report and name the tissue that would answer the question.
- Keep the raw alignments. A mosaic variant that was filtered can be recovered from them years later; one whose reads were discarded cannot.
Sources
- Biesecker LG, Spinner NB. A genomic view of mosaicism and human disease. Nature Reviews Genetics, 2013.
- Cao Y, Tokita MJ, Chen ES, et al. A clinical survey of mosaic single nucleotide variants in disease-causing genes detected by exome sequencing. Genome Medicine, 2019.
- Jamuar SS, Lam ATN, Kircher M, et al. Somatic mutations in cerebral cortical malformations. New England Journal of Medicine, 2014.
- Richards S, Aziz N, Bale S, et al. Standards and guidelines for the interpretation of sequence variants: a joint consensus recommendation of the American College of Medical Genetics and Genomics and the Association for Molecular Pathology. Genetics in Medicine, 2015.