Subcellular localization of human long non-coding RNAs is an isoform-resolved property masked by gene-level analysis

The function of a long non-coding RNA (lncRNA) is tightly coupled to its subcellular localization, yet localization is almost always annotated at the gene level while most lncRNA genes produce multiple isoforms. Reusing tissue-specific dominant-isoform switch calls for 268 lncRNA genes (GTEx long-read), I assigned each isoform a cytoplasmic-nuclear relative concentration index (CN-RCI) measured from ENCODE subcellular fractionation RNA-seq across 15 human cell lines (1,262 isoforms localized). Of the 163 genes with both tissue-dominant isoforms localized, 45 (27.6%) are masked: the gene-level CN-RCI sits in the neutral zone while the gene's isoforms occupy opposite compartments. That unfiltered figure is not itself evidence of within-gene heterogeneity, since exchangeable dispersion produces 25% under two null models. Requiring each isoform's compartment call to be reproducible across cell lines, and scoring the nulls on the same genes, 21 of 98 genes are masked (21.4%) against 3.1% expected, and 8 of 54 (14.8%) against 0.3% under the strictest criterion (both p = 0.001). The measurement reproduces in RKO cells from an unrelated laboratory, protocol and quantifier (Spearman ρ = 0.71), in BLaER1 cells (ρ = 0.80), under an independent quantifier, and against proximity labelling, which never fractionates cells (ρ = 0.56). The unchanged pipeline places 48 of 49 control isoforms in their established compartments, and recovers the reported pattern at four of five genes whose individual isoforms have been localized without sequencing. Tissue switching does not preferentially cross compartments, and the compartment differences are not explained by Alu content, 3'-end usage, structural mode or translation.