RNA Length and Ageing: Why Long Genes Suffer Most

Most ageing research asks which genes change. A different question turned out to be more productive: does anything about gene expression change systematically, regardless of what the genes do?

The answer appears to be yes, and the organising variable is length. Long genes are transcribed less efficiently in old tissue than in young tissue, short genes relatively more, and this shift occurs broadly rather than in a handful of pathways. It is a structural bias in the machinery rather than a change in a regulatory programme, which makes it a genuinely different kind of finding.

The observation

A 2022 study sequenced transcriptomes from 17 mouse tissues across the lifespan and looked for features that changed with age independently of gene identity. [1]

Transcript length emerged. In middle-aged animals, median transcript length had declined in roughly 60% of tissues; in old animals, in around 75%. Extending the analysis to rats, killifish and human tissue showed the same direction of effect, with roughly 80% of tissues showing an age-dependent relative decrease in long transcripts. In human samples, the effect was most pronounced in the brain. Single-cell data indicated the shift was reasonably consistent across cell types rather than driven by a change in tissue composition, which is an important control — much of what looks like ageing in bulk tissue is really a change in which cells are present.

The authors also examined eleven interventions previously shown to extend lifespan in mice. Seven were associated with an increase in relative long-transcript abundance. That is suggestive rather than conclusive, but it is the kind of observation that makes a phenomenon worth pursuing.

Why length would matter

The mechanistic intuition is straightforward once stated, and it is what makes this finding more than a curiosity.

Transcription is a processive operation. RNA polymerase II binds near the start of a gene and travels the entire length of it, synthesising RNA as it goes. Human genes vary enormously in length — from a few hundred base pairs to over two million.

Now introduce randomly distributed obstacles. DNA damage accumulates stochastically across the genome, and transcription-blocking lesions stall the polymerase. A polymerase transcribing a two-kilobase gene has a small chance of encountering a lesion. A polymerase transcribing a two-megabase gene has a thousand times more opportunity to hit one.

So if lesion density rises with age, the output of long genes falls disproportionately, purely as a consequence of geometry. No regulatory decision is required. The bias falls out of the arithmetic.

The mechanism, and how well it holds

The intuition would be unpersuasive on its own. What makes the finding credible is that it was tested directly and by a different group.

Work published in 2023 used nascent RNA sequencing combined with polymerase II ChIP-seq in ageing mouse liver, and found that in two-year-old animals approximately 40% of elongating RNA polymerases were stalled. [2]

The stalling was attributable to endogenous DNA damage, and it skewed transcriptional output in a gene-length-dependent manner, as the geometric argument predicts.

Two further points from that work matter. The transcriptional stress accounted for the majority of age-related gene expression changes in mostly postmitotic organs, and it disproportionately affected pathways central to ageing biology — nutrient sensing, autophagy, proteostasis, energy metabolism, immune function and stress resilience. The effect was also evolutionarily conserved from nematodes to humans.

Other groups have added to the picture rather than contradicting it. Separate work found that transcriptional elongation speed increases with age, and that interventions which slow it are associated with extended lifespan — a related but distinct fidelity problem. A 2026 analysis focusing on mouse brain reported that diminished transcriptional activity and changes in splicing both contribute to length-biased rewiring of the ageing transcriptome, implicating chromatin regulators and splicing factors alongside the damage mechanism. [3]

The phenomenon has acquired a name in the review literature: gene length-dependent transcription decline. [4] It has also been reported in Alzheimer's disease, where long neuronal genes are especially vulnerable, which may partly explain why the human effect is strongest in brain.

What it does not establish

Several caveats deserve stating plainly, and the original authors stated most of them themselves.

The imbalance is not specific to ageing. DNA damage produces it in progeroid models, and heat shock produces it through premature transcriptional termination. Anything that impedes processive transcription will generate a length bias. Observing the pattern in old tissue therefore does not prove it is a driver of ageing rather than a consequence of it.

Causation remains open. Showing that long-transcript decline accompanies ageing, and that some lifespan-extending interventions attenuate it, is consistent with a causal role but does not demonstrate one. The intervention data involves seven of eleven compounds, which leaves four that did not.

Most of the direct mechanistic work is in mice, and much of it in liver. The human data is largely cross-sectional transcriptomics rather than the nascent-RNA and polymerase-occupancy measurements that established the mechanism in rodents.

And the practical consequences of the imbalance are inferred rather than measured. Long genes are enriched for certain functions — neuronal signalling and synaptic architecture prominently — so a length bias would be expected to hit those functions hardest. Demonstrating that this is what actually degrades tissue function with age is a further step nobody has taken.

What it would mean if it holds

Three implications follow, and they are worth distinguishing from the finding itself.

It would suggest that a meaningful share of the transcriptional disorder of ageing has a single upstream cause, which is a more tractable target than a diffuse list of dysregulated pathways. Interventions that reduce transcription-blocking damage, improve transcription-coupled repair, or alter elongation dynamics would in principle act broadly rather than on one pathway.

It offers a possible measurement. Length-associated imbalance is computable from transcriptomic data that many studies already generate, which makes it a candidate readout for whether an intervention is affecting something systemic. Whether it performs well enough to serve as a biomarker is unestablished.

And it would partly explain why the brain and other postmitotic tissues age as they do. Cells that do not divide cannot dilute damage through replication, and they express an unusually high proportion of very long genes.

Where this sits

This is basic biology, and it does not change clinical practice. There is no test to order, no intervention that targets transcript length imbalance, and no evidence that any commercially available product affects it.

What it offers is a better conceptual frame. Ageing research has long been organised around lists — hallmarks, pathways, differentially expressed genes. A finding that cuts across all of those, driven by a physical property of the genome rather than a regulatory programme, suggests that some of what has been catalogued as many separate problems may be fewer problems seen from multiple angles.

That is worth knowing even though nothing follows from it on Monday morning. Most durable clinical progress starts as an observation nobody can act on yet.