Regional and person-associated variation in age-related intronic RNA across human tissues

Background. Altered RNA processing is often proposed as a feature of human ageing. Three different changes are easily confused: changes in the abundance of junction-bearing transcripts, changes in the choice among completed splice junctions, and accumulation of intron-containing RNA. They can have different causes and consequences. The Genomic Failure Hypothesis (GFH) predicts age-related loss of splicing accuracy linked to mitochondrial decline. Methods. From recount3’s uniformly processed GTEx v8 data, we built an auditable junction-level atlas of 33 non-CNS tissues and 13 CNS regions. Age was tested by stratified permutation within Hardy death-circumstance groups, with a depth-weighted count statistic for splice choice. The main outcome was intronic RNA per junction read, a ratio of recount3 intronic read-equivalents to junction reads, tested in 39 non-CNS tissues and 13 CNS regions with technical covariates and HC3 robust intervals. Two pre-defined false-discovery families were used: 26 primary tissues and regions, and 26 later non-CNS tissues. In 19 tissues and regions, a simplified IRFinder-style per-intron retention ratio was measured from base coverage. It was summarised by pooled, equal-weight and gene-balanced indices, and refitted intron by intron with the full covariate set and donor bootstrap. Within-donor contrasts used the same donor’s cerebellum or lung. Person-level variation was estimated as variance components from donor-pair covariances. External tests used 19 recount3 cohorts with living tissue, cultured cells or exact ages, TCGA surgical normals, developmental series and 28 public perturbation experiments. Results. Read-level junction trends were common outside the brain and were consistent for ageing and senescence genes such as CDKN2A (14 tissues). Detected changes in 5′-site splice choice were rare. However, only 38–44% of profiled junctions were informative for that test, so this is not evidence that splice choice is unchanged. Versions of the test centred on the 3′ site or on intron clusters also found changes in only a few tissues (8 of 46 had a call under any test). Benchmarks showed the false-positive rate to be nominal and power for modest effects to be low. Intronic RNA per junction read rose with age in tibial artery (+7.0% per decade, 95% CI 4.2–10.0), skeletal muscle (+4.2%, 2.3–6.1), hippocampus (+7.4%, 2.9–12.2) and caudate (+7.2%, 1.9–12.8), all at primary q ≤ 0.03. Among the later tissues it rose in kidney cortex, spleen, oesophageal mucosa and testis (q ≤ 0.011). It did not rise in cerebellum, spinal cord, substantia nigra or putamen. Relative to the same donor’s cerebellum, hippocampus, caudate, amygdala and anterior cingulate rose by 8.7–9.5% per decade. The pooled hippocampal retention index rose by 7.1% per decade (2.1–12.3), with little sensitivity to any single gene. Equal-weight indices rose less (+1.5%), as did the subset with above-median RNA integrity (+3.7%, p 0.20). Under full adjustment, the per-intron rank measure still rose broadly (82% of introns; 1,783 rising and 25 falling at q ≤ 0.05) and more in longer introns (ρ 0.21; 0.24 given coverage, expression and baseline retention). In the best-covered introns, the log-ratio measure rose 3.6–9.1% per decade in hippocampus, caudate, tibial artery and kidney, but not in cerebellum or lung. The hippocampal rise did not depend on intron position and was not explained by gene-body 3′ bias. Intronic RNA was 27–36% higher in older than younger adults in two living-muscle cohorts, and rose with donor age in one large series of cultured fibroblasts. Of the 19 cohorts examined, most did not show an age effect, and surgical kidney did not reproduce the GTEx magnitude. Within an age group, residual intronic RNA was shared across each person’s tissues. About 18% of residual variance (15–21%) lay between donors within the brain or body compartment, even across extraction batches; about 4% (0.6–6.7%) was shared between brain and body. This person-associated component replicated in multi-region brain cohorts, in repeated living muscle and blood samples, and between TCGA tumours and their adjacent normal tissue. Perturbation experiments gave no contrast at q ≤ 0.05, and none resembled the ageing per-intron pattern beyond ρ ≈ 0.08. Conclusions. The clearest RNA-processing change with age in these data is age-associated accumulation of intronic RNA, consistent with delayed or incomplete splicing. It varies by region and tissue, is strongest in hippocampus and other forebrain regions, tibial artery, muscle and kidney, and persists within donors. Whether it reflects nascent RNA, nuclear detention, retained introns in mature transcripts or composition is not established by bulk RNA-seq. Intronic RNA is also partly a property of the person, present in living and surgical tissue; its origin is unresolved. Mitochondrial and acetyl-CoA explanations remain hypotheses, as does a secondary model of cerebrospinal-fluid melatonin exposure (Supplement S1). Supplement S2 sets out experiments to test them.