A universal stochastic mechanism of aging from single-cell DNA methylation patterns.

Tarkhov et al. 2024. Nature Aging,   December 2024. Andrei E. Tarkhov and Vadim Gladyshev (Harvard Medical School, Brigham and Women's).

What they did. Single-cell DNA methylation data from multiple tissues across age in mouse and human.
They decomposed methylation changes into two components: a coordinated component (CpG sites whose methylation shifts in correlated patterns across cells, suggesting shared regulatory drivers) and a stochastic component (CpG sites drifting independently, suggesting noise-like accumulation).

Key findings.

1. Both components rise with age. Stochastic drift is the larger numerical contributor; coordinated drift is the smaller but biologically more interpretable component.
2. The coordinated component localises preferentially to PRC2-bound, developmentally regulated, bivalent chromatin regions. These are the same regions Horvath-clock CpGs draw from.
3. The stochastic component is distributed more uniformly across the genome.
4. A model in which methylation is driven by a stochastic process plus a smaller deterministic drift recapitulates the empirical age-methylation relationship across tissues and species.

Relevance to CAD and Covolution. An important empirical anchor for covolution/CAD. The reason: CAD requires exactly the decomposition Tarkhov provides. 1) A structured scaffold (coordinated CpGs at regulatory boundaries) plus 2) a noise field (stochastic CpGs in the bulk). The decomposition is empirical. CAD claims the coordinated component dominates the boundary regions of the attractor landscape. The stochastic component dominates the bulk is what they found, post hoc.

The complication.
Tarkhov and Gladyshev interpret their data within a stochastic-process-with-drift framework, not an attractor-decay framework. Their preferred interpretation is closer to Meyer/Maklakov/Schumacher: aging clocks measure stochastic drift through a structured landscape, and the structure is the developmental regulatory architecture.

This is compatible with CAD but not identical to it.
The disambiguation is whether the structure is passive (a fixed landscape through which stochastic drift accumulates) or active (an attractor maintained by feedback whose maintenance fails with age).