All articles

The non-coding genome in a clinical exome era.

Most of the genomeSequencing of all of a person's DNA, coding and non-coding regions alike. lies outside the exons, and a growing list of diagnoses sits there. A case should record what its test could not see.

A girl of twelve with a congenital limb malformation and no other features has had an exome and an array, both negative. Her father and his sister have the same malformation. The pedigree is dominant and the segregation is clean. There is a known gene whose coding sequence is normal in all three.

The cause, found later in a research genome, is a single-base change in an enhancer a long way from that gene, in a region no exome captures and no array resolves. Three affected relatives had been sequenced across the only part of the genome where the answer could not be.

What the field knows

The exome targets the one or two in a hundred bases that code for protein, and most Mendelian diagnoses are found there, which is why the exome has been the clinical workhorse. The remainder of the genome is not inert. Regulatory elements, promoters, enhancers, untranslated regions, non-coding RNA genes and deep intronic sequence each carry variants that cause disease, and the catalogue of such variants has grown as genomes have been sequenced. The limb malformation literature is the best-known example: point mutations and structural changes in a distant enhancer of a developmental gene, first described decades ago (Lettice et al., 2003), cause a classic dominant malformation with the gene's coding sequence intact.

Large genome programmes have measured how much the non-coding genome adds in practice. The 100,000 Genomes Project pilot (Smedley et al., 2021) reported diagnoses that an exome would have missed, among them non-coding and structural variants, alongside the larger share found in coding sequence. Studies of untranslated regions (Whiffin et al., 2020) showed that variants creating upstream start codons are a recurrent and under-recognised cause of loss of function in constrained genes, detectable in genome data and routinely ignored in exome pipelines. Deep intronic variants that create cryptic splice sites account for a measurable share of second alleles in recessive disease, and are visible only in a genome or by RNA.

Interpretation of non-coding variants is harder, and the field says so. There is no reading frame to reason from, population constraint is weaker outside exons, and the functional evidence usually needed to classify a regulatory variant is slow to obtain. The ACMG framework was written for coding variants (Richards et al., 2015), and work on extending it to the non-coding genome (Ellingford et al., 2022) sets out how existing evidence types apply and where new ones are needed. A candidate non-coding variant is therefore a hypothesis that needs more than its position to become a diagnosis.

Where practice breaks

The negative exome is read as a negative genome. The report says that no causative variant was identified, and a reader who does not hold the test's design in mind takes it to mean that the genome was searched. It was not. In a family like the one above, with clean dominant segregation and a known gene whose coding sequence is normal, the negative exome is the strongest possible pointer at the non-coding genome, and it is read as the end of the road.

Nobody records what the test could not see. The limitations paragraph in a report describes the assay in general terms. It does not say, for this case, that the leading candidate gene has a known regulatory region outside the capture, or that the recessive hypothesis lacks a second allele that could be deep intronic. The gap is generic when it should be specific.

The genome is ordered late or by rule rather than by hypothesis. Some services escalate every negative exome to a genome, which spends resources on cases whose hypotheses do not point outside the exons. Others never escalate, which strands the cases whose hypotheses do. In neither case is the decision made from the particular evidence the case holds.

When a genome is done, the non-coding variants arrive unranked. A genome produces millions of non-coding variants per person, and a pipeline that has no way to prioritise them against the phenotype and the candidate genes hands the scientist a list nobody can read. The data were generated and the search did not happen. A genome without a ranking that reads the non-coding variants against the case is an expensive way to confirm the exome.

How Tulparus handles it

Every case records the regions its data cover, from the capture design and the achieved coverage, and every hypothesis is checked against them. A dominant hypothesis on a gene with known regulatory elements outside the covered regions carries a named gap: the regulatory region of this gene was not sequenced. A recessive hypothesis with one allele carries the gap that a deep intronic or regulatory second allele could not be seen. The report, rendered from the case, states these gaps for this patient rather than for the assay in general.

The gap drives the next step. Where a hypothesis points outside the exons, the proposed actions include targeted sequencing of the regulatory region, RNA analysis, or a genome, each with what it would resolve, so that the genome is proposed for the cases whose hypotheses need it and not as a reflex. For the family above, the page would have shown a dominant hypothesis on the known gene, the uncaptured enhancer named as the gap, and a targeted assay or genome proposed to close it.

When genome data arrive, non-coding variants are ranked against the case rather than listed. The candidate channels score them by their position relative to the candidate genes' regulatory annotations, by conservation and constraint where those exist, by segregation in the family, and by the phenotype the gene is already implicated in. A variant in a known enhancer of the leading candidate gene, segregating with the malformation in three relatives, would be at the top of a short list with its evidence laid out and its uncertainties stated.

The classification standard for a non-coding candidate is shown as what it is. The page says which evidence types apply, which have been met, and which, usually functional confirmation, are outstanding. No regulatory variant becomes a diagnosis on position and segregation alone, and the report says so.

What you can do tomorrow

The first change is in how a negative exome is written and read.

  • For each negative exome, write one sentence in the report about the specific regions the leading hypotheses would need and whether the test covered them.
  • Flag the families with clean segregation and a normal coding sequence in a known gene. They are the cases where a genome or a targeted regulatory assay is most likely to pay.
  • Ask your pipeline team whether untranslated-region and deep intronic variants are called and annotated in your genome data, and if they are not, start with the candidate genes.
  • Record the capture design and coverage with the case, so that at reanalysis the gaps are known without re-deriving them.

Sources

  1. Lettice LA, Heaney SJH, Purdie LA, et al. A long-range Shh enhancer regulates expression in the developing limb and fin and is associated with preaxial polydactyly. Human Molecular Genetics, 2003.
  2. Smedley D, Smith KR, Martin A, et al. 100,000 Genomes pilot on rare-disease diagnosis in health care: preliminary report. New England Journal of Medicine, 2021.
  3. Whiffin N, Karczewski KJ, Zhang X, et al. Characterising the loss-of-function impact of 5' untranslated region variants in 15,708 individuals. Nature Communications, 2020.
  4. 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.
  5. Ellingford JM, Ahn JW, Bagnall RD, et al. Recommendations for clinical interpretation of variants found in non-coding regions of the genome. Genome Medicine, 2022.

Three relatives, one enhancer.See the gap the exome left.

A demo takes 20 minutes, screen shared, on synthetic cases. Bring the question from one of your own.