Molecular Genetics and Genomics Molecular Biology

The stability explanation for lncRNA discovery bias is not testable with current half-life coverage

Tani, Hidenori
Springer Nature · 2026-10-26 · DOI 10.1007/s00438-026-02526-9

Abstract

Long non-coding RNAs (lncRNAs) are nominated as candidate disease regulators from steady-state RNA-seq differential-expression (DE) catalogs, and short-lived lncRNAs have been proposed to be systematically missed because rapid decay lowers steady-state abundance. We asked how much of that bias it can account for, using 18 half-life strata drawn from 16 metabolic-labeling datasets across 13 human cell lines and 17 uniformly processed disease cohorts (16 TCGA tumor-versus-normal contrasts, one non-alcoholic fatty liver cohort; recount3), matching the half-life cell line to the tumor tissue where possible. Under the pre-filter used here the abundance floor leaves only 24% of annotated lncRNAs testable, and downsampling BRCA recovers just 18% of its full-cohort DE lncRNAs at n = 5 per group. Its upstream driver is not resolved. Across six tissue-matched half-life datasets — four cell lines — the median half-life–abundance correlation is ρ = 0.195 (cluster-bootstrap range − 0.003 to 0.293, descriptive rather than a formal interval at so few clusters); dropping the weakest match gives 0.143. Among the lncRNAs carrying a half-life we find no excess exclusion of the short-lived relative to the long-lived; but half-lives are measurable only where a transcript is expressed, and only in cell lines, so that comparison is confined to a subset selected on the quantity at issue. The limiting factor is coverage: only 10.0% of annotated lncRNAs, and 4.4% of those below the filter, have any measured half-life — under 1% counting tissue-matched cell lines only. We provide SIEVE, open-source code producing these detectability figures for a given design, and specify the data the hypothesis needs.

Read at publisher →