Aging, read as information

The Palimpsest

One model of aging asks whether cells slowly forget who they are, one imperfect copy at a time.

Scroll ↓
Scene 02

Your identity is a message

Every cell in your body carries the same genome, the same three billion letters. So what makes a neuron a neuron and a liver cell a liver cell? Not the letters. It is the annotations: a layer of chemical methylation marks laid over the DNA that decide which genes are switched on. That pattern is the message. Read correctly, it spells you.

625 sites · each on or off · together, one codeword
Scene 03

Every division is a noisy copy

When a cell divides, the pattern must be re-written onto its daughter. The copy is good, but not perfect. In 1948 Claude Shannon showed what happens when a message crosses a noisy channel over and over: retained information leaks away along a predictable curve, drifting toward a coin-flip.

generation 0 bits retained 625 / 625
Scene 04 · the crux

There is no backup

A photocopier can compare each new page to a pristine original. Ordinary epigenetic maintenance has no known genome-wide youthful backup: the machinery often copies the marks that are currently there, corruption and all. Copy from an accurate reference, and the message can be corrected. Copy only the copy, and errors can be inherited and compounded.

copy from the original625 bits
copy the copy · reality625 bits
generation 0 · same noise per copy, opposite fate
Scene 05

Protect the paper. Protect the readers.

The model separates two preventive levers. Protect the paper: preserve accurate chromatin anchors before they are corrupted. Protect the readers: keep the local read–write machinery able to propagate their signal. Both can slow further loss; neither manufactures truth that is already gone. The slider shows an engineered consensus memory with fixed anchors—not evidence that natural blood aging literally melts at one threshold.

0.06
pattern intact
100%
state A state B anchor · sequence-pinned
Scene 06

Repair depends on reference fidelity

Visible drift and recoverability are not the same thing in this simulation. A heavily drifted cell with correct anchors can rebuild; a nearly normal-looking cell with corrupted anchors can confidently propagate the wrong state. There is no demonstrated universal biological cliff here. The lesson is narrower: internal repetition helps only while its reference remains accurate, and late repair needs an external bank cleaner than the state it is correcting. Press attempt recovery to see the contrast.

Cell A · looks wrecked
heavy drift · anchors intact
true identity ,
Cell B · looks fine
slight drift · anchors corrupted
true identity ,
Both cells re-read their marks from the anchors outward.
Scene 07

Targeted repair, then hold

Late intervention needs a cleaner reference. In the simulations, a nanoparticle or hypothetical nanobot is a delivery operator, not a demonstrated device: it revives failed readers or carries banked information into a bounded region. The actual spatial simulations find that too little radius misses damage, in-domain radius repairs it, and over-broad radius overwrites neighboring domains; endogenous maintenance then holds the restored state. The control below is a separate qualitative schematic of too little, bounded repair, and loss of specificity. It does not calculate spatial radius or simulate a device.

0.34
Bounded schematic window: restoration without loss of specificity.
Scene 08 · up close

The asymmetry that traps the cell

Why might a cell fail to fix itself? Its copier, DNMT1, is faithful to the present: it can reproduce a mistaken mark. The simulation gives sequence-tethered anchors a corrective role inspired by proteins such as CTCF, but that clean truth-clamp is a model assumption, not a complete molecular account. Self-maintenance can hold surviving information; corrupted anchors require a cleaner external reference.

faithful copying propagates the error · the anchor overwrites it with truth

Some text may remain unreadable. Some may be genuinely erased.

That boundary defines the practical program: protect accurate substrate and functioning readers early; when internal references are no longer trustworthy, repair from a cleaner external bank and then hold the result. The model says when those strategies can work. It does not yet show that they slow aging or restore health in an organism.

Our own analysis · mixed evidence

What we actually found

The model now has both negative and positive external checks. Human blood does not show the proposed mean-level collective melting signature. A small read-level signal in very old T cells remains unresolved. Separately, public mouse skin data move in the restoration direction predicted for an external-reference intervention, but do not establish mechanism or organismal benefit.

01
Shape leans relaxation, not melting

The age-drift trajectory is front-loaded and decelerating, a relaxation toward a floor, not the late-accelerating runaway the melting picture predicts. But much of that front-loading is developmental — restricted to the adult aging window the edge nearly vanishes — so it is suggestive, not decisive, and we lean on the composition-inert test below.

02
The one melting-like signal is unresolved

Sorted CD4 T-cells, ages 18–25 against 82–86 (GSE79798), showed a small +0.008 read-level disorder gap, consistent across every old donor. We first read it as an aged-DNA artifact because it rises with CpG density. But with real genomic annotation the gap concentrates in Polycomb domains — exactly where the companion paper's meltable phase lives. CpG-dense concentration is the shared signature of both aged-DNA error and predicted melting, so it can't tell them apart. At n=3 vs 3 this stays unresolved; a bisulfite-free nanopore test is the decider.

read-level disorder gap · old minus young
cross-kit variance (loose bound)
+0.008
0
0.02
0.04
03
Mouse OSKM result points youthward

In public mouse WGBS (GSE231658), old skin treated with OSKM moves both independently defined methylation-aging arms youthward across all 19 autosomes: +0.0077 on sites that gain methylation with age and +0.0299 on sites that lose it. All five treated-animal summaries are positive. The weaker-arm label permutation is exploratory (p=0.0952), and bulk cross-tissue data cannot distinguish within-cell repair from composition, selection, or another nonuniform mechanism.

04
Simulated memory has a finite-horizon threshold

The first one-dimensional mechanism fails as a long-lived memory. In two dimensions, no run failed by generation 4,000 at copy noise 0.10; at 0.20 the median lifetimes were 318, 614, and 2,725 generations as system size grew; at 0.22 every run failed. This supports robust collective memory below a measured threshold, not indefinite aging arrest.

Plain takeaway

The strongest current claim is conditional: protecting accurate substrate or functioning readers can slow simulated information loss; cleaner external references can repair it within targeting limits. Human blood argues against universal collective melting, while mouse OSKM data support the predicted restoration direction without proving mechanism, healthspan, lifespan, medicine, or nanobots. The bisulfite-free elderly test and an independent reprogramming cohort remain decisive next steps.

Data & sources

What this is built on

Datasets analyzed
GSE79798
Sorted CD4 T-cell WGBS · Jenkinson et al., Nat Genet 2017
GSE128731
Bisulfite library-prep benchmark · Zhou et al., Sci Rep 2019
GSE186458
WGBS cell-type atlas · Loyfer et al., Nature 2023
PRJNA339792
Whole-blood pilot cohort
GSE261769
Methylation age dataset · Mboning et al., Front Bioinform 2024
GSE231658
Mouse tail epidermis aging direction and old-skin OSKM WGBS
Key theory & methods
Landau et al. 2014 · read-level discordance (PDR)
Landan et al. 2012 · epipolymorphism
Houseman et al. 2012 · cell-composition deconvolution
Horvath 2013 · Hannum et al. 2013 · methylation clocks
Ocampo et al. 2016 · Lu et al. 2020 · partial reprogramming as external reference
Shannon 1948 · Landauer 1961 · Eigen 1971 · Toom 1980 · information, erasure cost, error threshold, fault-tolerant memory
Credits & honest footing

This piece synthesizes existing science and reports our analysis of public DNA-methylation data. The evidence is mixed: a negative human-blood result, an unresolved elderly signal, and a directionally positive but exploratory mouse OSKM result. None is a wet-lab therapy, proven cure, medicine, or nanobot demonstration. The framing draws on David Sinclair's Information Theory of Aging, the Dodd–Sneppen model of bistable chromatin maintenance, the Jenkinson–Feinberg information theory of DNA methylation, Claude Shannon's 1948 A Mathematical Theory of Communication, and Manfred Eigen's error catastrophe. The visuals are deliberate simplifications; real methylation is not a tidy square grid, and the anchor and delivery dynamics are model cartoons.

Open, testable predictions

A cell's recoverability should track its surviving accurate reference set more closely than average methylation drift. A cleaner external reference should restore information bidirectionally rather than merely demethylating. And repair should show a spatial targeting window: too narrow misses damage; too broad overwrites neighboring domains. These remain hypotheses, not organismal findings.

Next discriminators: a primary native-ONT elderly cohort, an independent reprogramming cohort, and direct measurement of reference survival versus recoverability.

Manuscripts in preparation