Hypothesis

Circular ANRIL (circANRIL) transcripts, generated from the 9p21.3 locus, act as molecular sponges that sequester the histone demethylase Jmjd3 and specific microRNAs, thereby modulating the balance between H3K27me3 deposition and removal at the CDKN2A/B promoter in a tissue‑dependent manner. This mechanism explains why blood shows a straightforward age‑related increase in p16^INK4a expression while Alzheimer’s brain exhibits reduced CDKN2A transcription despite elevated exon‑2 CpG methylation.

Mechanistic rationale

1. circANRIL biogenesis rises with epigenetic disorder – Within‑read methylation variability (RD) at PRC2 targets increases quadratically with age [4]. Elevated RD promotes alternative splicing and backsplicing, favoring circANRIL production over linear ANRIL.
2. circANRIL binds Jmjd3 via exposed KDM6 domains – The circular structure preserves RNA‑protein interaction motifs that are obscured in linear ANRIL. By sequestering Jmjd3, circANRIL limits H3K27me3 demethylation, reinforcing PRC2‑mediated repression.
3. circANRIL also sponges miR‑204 and miR‑199a‑3p – These microRNAs normally target Jmjd3 mRNA for degradation. Their sequestration stabilizes Jmjd3 protein levels, creating a buffered system where changes in disorder translate into modest, tissue‑specific shifts in demethylase activity.
4. Tissue‑specific expression of RNA‑binding proteins (RBPs) – In peripheral blood, high levels of hnRNPL favor linear ANRIL, keeping Jmjd3 relatively free and allowing age‑dependent H3K27me3 loss → p16^INK4a up‑regulation. In neurons, elevated FUS and TDP‑43 promote circANRIL accumulation, enhancing Jmjd3 sequestration and maintaining higher H3K27me3 despite rising CpG methylation, thereby suppressing CDKN2A transcription.

Testable predictions

• Prediction 1: circANRIL levels will correlate positively with global epigenetic disorder (RD) in both blood and brain, but the slope will be steeper in neuronal tissue. Quantify circANRIL by RNase R‑resistant RT‑qPCR and RD by whole‑genome bisulfite sequencing across age‑stratified samples; expect a significant interaction term (tissue × RD) in a linear model (p < 0.01).
• Prediction 2: Knock‑down of circANRIL in primary human neurons using antisense oligos targeting the backsplice junction will increase Jmjd3 protein, reduce H3K27me3 at the CDKN2A promoter, and elevate p16^INK4a mRNA, reversing the Alzheimer‑like expression pattern. Controls using scrambled oligos should show no change.
• Prediction 3: Overexpression of circANRIL in hematopoietic stem cells will decrease Jmjd3 activity, increase H3K27me3 at CDKN2A, and blunt the age‑related rise in p16^INK4a, measurable by flow cytometry for senescence‑associated β‑galactosidase.

Falsifiability

If circANRIL manipulation fails to alter Jmjd3 occupancy, H3K27me3 levels, or p16^INK4a expression in the predicted direction across both tissues, the hypothesis is refuted. Similarly, if circANRIL abundance does not track with epigenetic disorder (RD) or shows no tissue‑specific bias, the core premise collapses.

Broader implication

This model links two hallmarks of aging—epigenetic disorder and non‑coding RNA regulation—to explain divergent CDKN2A/B trajectories in peripheral senescence versus neuronal resilience, offering a mechanistic bridge between GWAS SNPs at ANRIL and tissue‑specific disease risk.