The phenotype matched one gene closely: an epileptic encephalopathy with a particular seizure semiology and a movement disorder. The gene was clean on sequencing, on coverage and on copy number. The case was closed as unexplained. Three years later a paper described the same presentation in children with de novo truncating variants in a paralogue, a gene in the same family, with the same domain structure and an overlapping expression pattern.
The child had one. It had been in the exome from the start, filtered out at the first step as a variant in a gene with no disease association. Nobody had done anything wrong by the rules of the time, and the rules had cost the child three years.
What the field knows
Genes in the same family share more than sequence. Paralogues arise by duplication and often retain function, expression and, when disrupted, phenotype, so that a disorder caused by one member of a family frequently has a counterpart caused by another. The channelopathies are the plainest example: closely related sodium and potassium channel genes each cause epilepsies that overlap in presentation, and the order in which they were discovered reflects when a family happened to be sequenced, not any biological precedence. Lal and colleagues (2020) showed that gene family information predicts which as yet unassociated genes are likely to cause neurodevelopmental disorders and which positions within them are likely to be intolerant of variation.
Protein domains generalise across the family. A missense change at a position that is pathogenic in one paralogue is, more often than chance, pathogenic at the aligned position in another, and the regions where pathogenic variants cluster in one gene mark the regions to watch in its relatives (Pérez-Palma et al., 2020; Wiel et al., 2019). Tolerance to variation, measured across the aligned family rather than one gene, is a stronger signal than any single gene's constraint when the single gene has few known variants.
Beyond the family, genes that act together cause related phenotypes. Methods that walk the protein interaction network from known disease genes to their neighbours (Köhler et al., 2008) were among the first to show that a candidate's connection to a phenotype-matched gene is evidence in its own right, and pathway membership has since been used in most gene discovery pipelines. The principle is the same as for paralogues: the phenotype points at a function, and more than one gene carries it.
Where practice breaks
The first filter in most clinical pipelines keeps variants in known disease genes and discards the rest. The filter is reasonable for a first pass, since the unassociated genes hold mostly noise. It is a problem when the case stays unresolved and nobody goes back, because the filter removed the candidates whose only fault was being discovered later.
Family knowledge lives in a few heads. A scientist who has worked on channelopathies knows which genes are related; a scientist who has not does not, and no system supplies it. The knowledge is real and it is unevenly distributed, which means the same case gets a different search depending on who picks it up. Which paralogues were considered for a given case is recorded nowhere, so a second scientist cannot tell whether the family was searched or skipped.
Where paralogues and pathway neighbours are considered, they are considered as a list of gene names, without the domain alignment that would say whether this particular variant falls where the family's pathogenic variants fall. A truncating variant in a paralogue and a missense change in a tolerant region of the same gene are handled alike, when one deserves a hypothesis and the other does not.
A paralogue candidate is a novel gene candidate, and the standard for a novel gene is different from the standard for a known one. Without that standard written down, the paralogue is either overclaimed in a report or, more often, never mentioned because nobody is sure what can be said. A finding with nowhere to go in the report is a finding that is quietly dropped.
How Tulparus handles it
The family of a phenotype-matched gene is a candidate channel. When a gene matches the phenotype well and has no qualifying variant, its paralogues and its pathway neighbours are searched for variants in this patient, and each candidate found is labelled with the route it came by: paralogue of the matched gene, aligned to a position where pathogenic variants cluster in the family, or member of the same pathway, with the knowledge release that defined the family and the pathway.
The domain alignment is shown. A variant in a paralogue is displayed at its aligned position across the family, with the known pathogenic variants of the other members at their positions, so that a scientist can see whether this change sits where the family's disease sits or in a region that tolerates variation. Constraint is shown for the gene and for the aligned family, so that a gene with few known variants borrows what its relatives know.
A paralogue candidate is held to the novel gene standard. The hypothesis records that the association is unestablished, lists what would establish it, a second unrelated patient, a functional result, a match in a network, and proposes those steps. The report, if it mentions the gene at all, says exactly this and no more.
The record shows which families and pathways were searched and when, so that the search can be repeated, and so that a later knowledge release that adds a gene to a family or a pathway reopens the cases whose matched gene it is related to. For the child above, the paralogue would have been proposed on the day the matched gene came back clean, with the truncating variant, its position, and a note that the gene was unassociated and what would change that.
What you can do tomorrow
The step most services skip is the second look at the filtered-out genes, and it can be done by hand for the cases where the phenotype points somewhere specific. It takes an hour per case, and it is the hour most likely to find something in a case that has been looked at twice already.
- For every unresolved case where one gene matched the phenotype closely, list that gene's paralogues from a public gene family resource and look at the patient's variants in them before closing the case.
- Align any candidate missense variant to the family and check whether the position is one where pathogenic variants cluster in a relative. The public domain tolerance resources do this in a browser.
- Record which families and pathways were searched in the case note, by name, so that the next person knows what was done.
- Write down your service's standard for reporting a variant in an unassociated gene before you need it, so that a paralogue finding has somewhere to go.
Sources
- Lal D, May P, Perez-Palma E, et al. Gene family information facilitates variant interpretation and identification of disease-associated genes in neurodevelopmental disorders. Genome Medicine, 2020.
- Pérez-Palma E, May P, Iqbal S, et al. Identification of pathogenic variant enriched regions across genes and gene families. Genome Research, 2020.
- Wiel L, Baakman C, Gilissen D, Veltman JA, Vriend G, Gilissen C. MetaDome: pathogenicity analysis of genetic variants through aggregation of homologous human protein domains. Human Mutation, 2019.
- Köhler S, Bauer S, Horn D, Robinson PN. Walking the interactome for prioritization of candidate disease genes. American Journal of Human Genetics, 2008.
- Karczewski KJ, Francioli LC, Tiao G, et al. The mutational constraint spectrum quantified from variation in 141,456 humans. Nature, 2020.