Detecting Repeat Expansions from Short Reads: Why the Pathogenic Allele Is Often Longer Than the Read That Should Reveal It

Zetobit · Bioinformatics Insight Series

Detecting Repeat Expansions from Short Reads: Why the Pathogenic Allele Is Often Longer Than the Read That Should Reveal It

For a class of neurological diseases, the mutation is not a changed base — it is a count. And the pathogenic count is frequently larger than the sequencing read, so the very feature that makes the allele disease-causing is the feature that makes it invisible to a standard pipeline.

By Kanna Nandakumar, PhD Zetobit LLC

Huntington's disease, fragile X syndrome, myotonic dystrophy, several spinocerebellar ataxias, and the most common genetic form of ALS share an unusual property: the causal mutation is not a substituted or deleted base but an expansion — a short DNA motif repeated too many times. In HTT, a run of CAG at 40 or more copies produces Huntington's disease in a normal lifespan.¹ In FMR1, more than 200 copies of CGG causes fragile X.² The number is the diagnosis. Where the number sits relative to a threshold is the entire clinical question.

A standard short-read variant pipeline is built to find the wrong kind of thing. It aligns reads to a reference and reports where bases differ — substitutions, small insertions and deletions. A repeat expansion is none of those. It is a length change in a region that already contains the motif, and when the expanded tract runs to hundreds or thousands of bases, it is longer than the 150 bp read that would need to contain it. The pipeline does not throw an error. It simply reports the reference-length repeat, or nothing, and moves on — a false negative in exactly the disorders where the finding would have been decisive.

Why read length is the whole problem

To measure a repeat directly, a read has to span it: cover the repeat plus enough unique sequence on both sides to anchor its position. That works when the repeat is short. Once the tract is longer than the read, no single read can contain it, and direct measurement becomes impossible.³ Pathogenic expansions live precisely in that regime — C9orf72 repeats in ALS begin at roughly 30 copies of a six-base motif and can reach thousands; HTT tracts inherited around 40–50 CAG can somatically expand far beyond.⁴ The alleles with the greatest clinical impact are the ones short reads are least able to see.

Because direct measurement fails, expansion-aware tools reconstruct length from three indirect signals, none of which is a base call: spanning reads fully contain a short repeat and measure it directly; flanking reads start in unique sequence and run into the repeat, giving a lower bound; and in-repeat reads (IRRs) are composed entirely of the motif, anchored to the locus only by their mate on the other end of the fragment.⁵ A large expansion announces itself not as a variant but as a pile of IRRs — far more repeat-only reads mapping to the locus than the reference length could produce.⁶ The evidence is statistical and inferential, and a caller that isn't looking for it discards it as noise.

As a repeat lengthens past the read, the only remaining evidence is indirect: flanking reads that give a lower bound, and in-repeat reads counted against the locus via their anchoring mate. Expansion-aware tools model all three signals; a general variant caller uses only the first.

The threshold problem raises the stakes

Repeat disorders are not simply expanded-or-not. They are graded, and the grades carry different clinical meaning. HTT illustrates the full ladder: up to about 26 CAG is normal; 27–35 is an intermediate allele that usually does not cause disease but can expand in the next generation; 36–39 causes Huntington's with reduced penetrance; and 40 or more is fully penetrant.⁷ FMR1 has its own tiers: typical alleles under about 45 CGG, a premutation from roughly 55 to 200 that confers risk of tremor/ataxia and primary ovarian insufficiency, and full mutation above 200.²

This means a tool cannot merely flag "expanded." It has to estimate the count precisely enough to place an allele on the correct rung — the difference between a reassuring result, a reduced-penetrance diagnosis, and a reproductive-risk finding can be a handful of repeats. And precision degrades exactly as the tract grows: once a repeat is sized only from in-repeat reads, the estimate becomes a confidence interval, not a number. Published tool output for an expanded allele is reported as a range — an estimated size with an interval attached — because the underlying signal cannot pin it down further.⁵

Somatic instability compounds it. These repeats are unstable in the body: a single person carries a distribution of repeat lengths across cells, and in HTT the inherited 40–50 CAG tract can expand somatically past 150 in vulnerable neurons.⁴ The blood measurement a lab returns is one summary of a moving target — accurate for the germline allele, but not the whole biological story.

Why the usual pipeline assumptions break

Beyond read length, three standard assumptions fail in repeat regions. First, PCR amplification biases against long, GC-rich repeats — the C-and-G-heavy tracts in FMR1 and C9orf72 amplify poorly, so the expansion-detection tools are designed for PCR-free WGS specifically to avoid dropping the very alleles they exist to find.⁶ A PCR-based library can lose the expanded allele before sequencing begins. Second, panel and exome capture leave these regions thin or absent, and the read-depth-based logic that expansion tools rely on assumes uniform WGS coverage; the same analysis on targeted data is far less reliable.³ Third, the reference contains only the short, normal-length repeat, so the aligner has no expanded template to match — reads from a large expansion mismap, soft-clip, or scatter rather than aligning cleanly.

The corollary is that "we did WGS" does not mean expansions were assessed. Unless an expansion-aware caller was run, on PCR-free data, against a curated set of repeat loci, the pipeline was structurally incapable of detecting them — and the report's silence on a repeat disorder means nothing was looked for, not that nothing was found.

What actually works

Detecting expansions is a distinct analysis bolted alongside the variant pipeline, not a setting within it:

• Targeted expansion callers on PCR-free WGS. Tools such as ExpansionHunter, STRetch, TREDPARSE, and exSTRa genotype known pathogenic loci by modeling spanning, flanking, and in-repeat evidence together, and can size repeats beyond read length.⁵
• De novo genome-wide scans. ExpansionHunter Denovo, STRling, and superSTR search for anomalous in-repeat-read pileups across the genome to surface expansions at loci not on a predefined list — important because novel disease-associated repeats are still being discovered.⁸
• Orthogonal confirmation. A positive short-read call is a screen, not a final answer; repeat-primed PCR or Southern blot remains the standard for confirming and precisely sizing an expansion, and tool documentation explicitly directs that predicted expansions be validated.³
• Long reads for definitive sizing. Long-read platforms can span an entire expanded tract on a single molecule, giving direct length measurement and resolving complex or interrupted repeat structures that short-read inference cannot.⁹

And, as with the other blind spots in this series, the cheapest safeguard is explicit scope: a clinical pipeline should state which repeat loci it interrogates and by what method, and should report low-confidence or out-of-range sizing rather than defaulting to a reassuring reference-length call. "Expansion suspected — reflex to repeat-primed PCR" is the correct output when the short-read signal is ambiguous. A silent reference-length genotype is not.

The takeaway

Repeat expansion disorders invert the usual logic of variant calling. The pathogenic feature is length, not sequence; the clinical answer is a count placed against a threshold; and the alleles that matter most are the ones too long for a short read to contain. A standard pipeline aligned to a reference will pass over them without complaint. Catching them takes a purpose-built, expansion-aware analysis on the right kind of data — and the discipline to say, plainly, when the read simply wasn't long enough to answer the question.

References

1. Ferguson R, et al. A CAG repeat threshold for therapeutics targeting somatic instability in Huntington's disease (≥40 CAG fully penetrant; ~65+ juvenile onset). Brain. 2024. PMC11068328
2. Trinucleotide CGG repeat diseases (FMR1: normal <45, premutation 55–200, full mutation >200 CGG). Front Genet. 2022. frontiersin.org
3. Detecting short tandem repeats from whole genome sequencing (read-length limits; PCR validation still required). Genomics England. 2026. genomicsengland.co.uk
4. A shared DNA-repeat toxicity threshold in Huntington's disease (inherited 40–50 CAG expanding somatically beyond 150). PMC. PMC12713137
5. Dolzhenko E, et al. Detection of long repeat expansions from PCR-free WGS data (ExpansionHunter; spanning/flanking/in-repeat reads; sizing beyond read length). Genome Res. 2017. PMC5668946
6. Large-scale in silico characterization of repeat expansion variation (IRR evidence; three read types defined). Sci Data. 2020. nature.com
7. HTT intermediate and reduced-penetrance allele ranges (27–35 intermediate; 36–39 reduced penetrance; ≥40 full). PMC. 2024. PMC11555821
8. Computational tools for tandem repeat detection (EHdn, STRling, superSTR for de novo detection; short-read length limits). Brief Bioinform. 2026. academic.oup.com
9. Dolzhenko E, et al. ExpansionHunter: a sequence-graph-based tool to analyze STR variation. Bioinformatics. 2019. academic.oup.com

Zetobit LLC · Bioinformatics Insight Series · Lexington, KY · zetobit.com