Fifty Variants Nobody Can Classify, in an Ordinary Genome

The limiting step in genomic medicine is no longer reading a genome; it is knowing what an ordinary one contains. Sequencing is now cheap enough that health systems are evaluating it as a screening tool in adults and newborns, which inverts the usual clinical logic: instead of starting with a sick patient and hunting for a cause, you start with a variant list and ask whether anything on it matters. Without a denominator, every flag looks like a finding.

A group at the University of Exeter has now measured that denominator at scale, analysing short-read whole-genome sequences from 879,461 adults in the UK Biobank and the All of Us research programme, neither of which recruited on the basis of disease. The typical genome holds 4.3 to 5.3 million variants relative to the reference, roughly 1.5 percent of them rare. Narrowed to curated monogenic disease genes, the average person carries one to three rare variants ClinVar rates pathogenic or likely pathogenic, about 50 of uncertain significance, and, across 28 polygenic scores, a top-10-percent position for three conditions.

Why it matters: If population-wide genomic screening goes ahead, these numbers set the background rate of alarm. Most of what sequencing finds in a healthy adult is carrier status, uncertainty or a high percentile, and the burden of acting falls on people who are not ill.

The survey, by Wright et al. at the University of Exeter Medical School, appears in Genome Medicine. It was funded by the Medical Research Council, Wellcome and the NIHR Exeter Biomedical Research Centre; the authors declare no competing interests.

Most of a genome is unremarkable, and measurably so

A median of 98.4 percent of a person's single-nucleotide variants and short insertions or deletions are common, meaning a minor allele frequency above 1 percent; 1.34 percent are rare and 0.26 percent ultrarare. Position matters as much as frequency: only 0.54 percent fall in protein-coding sequence, and 35 percent lie outside any annotated gene with no predicted consequence. Restricted to the clinically preferred transcript of each gene, the median genome carries 9,600 missense variants, 28 of which score above 0.7 on the pathogenicity predictor REVEL, and 229 predicted loss-of-function variants, 165 of them high confidence.

Larger rearrangements behave similarly: among calls above 10,000 base pairs in 587,475 people, the median genome has 68 copy-number variants. From 2,334 parent-offspring trios, the average genome also acquires a mean of 61 variants absent from both parents, 75 to 80 percent of them on the paternal chromosome; the count climbs 1.17 per year of the father's age at birth against 0.38 per year of the mother's.

Pathogenic is a label on a variant, not a prediction about a person

ClinVar held about 4 million variants at the snapshot used here, roughly 300,000 of them germline pathogenic or likely pathogenic, and 8 to 10 appear in an average genome. Restricting to 4,262 genes with curated monogenic disease associations cuts that to a mean of 1.9 to 2.6. Most are heterozygous in recessive-disease genes, which is carrier status rather than disease, and the commonest are alleles already known to be frequent in particular populations, such as HFE p.Cys282Tyr at 7 to 8 percent in European-like participants.

Dominant conditions give the more uncomfortable number. Limited to rare variants in genes linked only to autosomal dominant disease, about 1 in 8.6 people carry a pathogenic or likely pathogenic variant, and 1 in 2 to 3 a rare high-confidence loss-of-function variant. Since these are volunteers well enough to attend a research centre, the authors read the excess as incomplete penetrance, modifiers and calls that are not truly loss-of-function.

Uncertainty is abundant, and unevenly distributed

Variants of uncertain significance numbered about 2 million in ClinVar, roughly seven times the pathogenic count, yet within a genome they occur 30 to 90 times more often: about 180 per person in curated disease genes, or 50 once restricted to rare ones. European-like participants carry more pathogenic calls and are likelier to get a diagnosis; non-European-like participants carry notably more uncertainty, which the authors attribute to differing allele frequencies, thin reference data and unequal access to testing.

Polygenic scores add a second kind of inflation. Across 28 pre-calculated scores, the average participant was in the top 10 percent for three conditions and the top 5 percent for 1.4. Among 450,209 European-like UK Biobank participants, about 1 in 5 were in the top 1 percent for at least one score; among 35,844 non-European-like participants it was about 1 in 3, which the authors call likely artefactual and a symptom of poor transfer between populations. Run 100 scores and roughly 99 percent of people would be in the top 5 percent for something.

What the study can't say yet

This is a census of variants, not a test of what they do. No health records were used to ask whether any variant actually caused disease, so the counts bound the problem without resolving it. The cohorts are confined to the UK and the United States, and variation present only in indigenous and geographically isolated populations is absent from existing databases entirely. Both also carry a healthy-volunteer bias, visible in their finding of near-complete chromosomal homozygosity in only 1 in 8,100 people, fewer than anticipated.

Other numbers are bounded by the technology. Everything rests on short-read sequencing at roughly 30-fold depth, mostly from blood, so inversions, mobile-element insertions and tandem repeats are largely invisible. With mean recruitment ages of 57 and 52 years, some large copy-number calls are likely age-related clonal haematopoiesis rather than inherited variation. The new-variant count comes from trios whose members were all old enough to enrol, which enriches for young parents, so 61 sits toward the low end of published estimates. On reporting, the authors are direct: This raises questions about the clinical value of reporting such findings outside of a diagnostic context, due to high carrier frequencies and incomplete penetrance.

Quick questions

Does carrying a pathogenic variant mean you will get the disease? Usually not. Most are heterozygous variants in recessive-disease genes, which leaves the carrier healthy, and even variants in dominant-acting genes are often non-penetrant in adults well enough to join a biobank.

Why would sequencing produce more uncertainty for some people than others? Classification depends on having seen a variant before in a comparable population, so where reference data are thin it stays uncertain rather than being resolved as benign.

What is the one-line takeaway? Across nearly a million healthy adult genomes, disease-associated variation is ordinary rather than exceptional, which is the background any screening programme would read against.