Aging, read as information

The Palimpsest

Your cells are not wearing out. They are slowly forgetting 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. Your epigenome cannot, there is no master copy stored anywhere. The machinery copies the marks that are currently there, corruption and all. Copy from the original, and the message stays perfect forever. Copy the copy, and errors are inherited and compound. Life does the second thing.

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

Anchors hold the line

So why don't cells dissolve into noise within a week? Because a sparse set of sites are pinned to the DNA sequence itself, proteins like CTCF and boundary elements that do not drift. Local feedback lets them re-impose their signal on wavering neighbors. Below a noise threshold, the pattern heals faster than it decays. Drag the noise up and watch it hold, then, past a point, melt all at once.

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

The point of no return

With age the anchors themselves erode, and not because the DNA changes. Methylation creeps over the binding site, and a reader like CTCF cannot grip a methylated motif: it bounces off, and the anchor goes dark. The sequence is still there; nothing can read it. And here is the cruel part: recoverability does not track how damaged a cell looks. A heavily-drifted cell whose anchors survive can be rebuilt. A barely-drifted cell whose anchors have gone bad is already lost, it just doesn't show it yet. Press attempt recovery and watch which one comes back.

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

Reprogramming has a safe dose

If the original text is still down there, you can re-read it. A pulse of Yamanaka factors coaxes old cells back toward youth without changing their DNA. But it is an inverted-U. Too little does nothing; a measured dose rejuvenates; too much erases identity outright, a cell that forgets it was ever a cell becomes a tumor. Slide the dose and find the window.

0.34
Rejuvenation window, identity restored, age reset.
Scene 08 · up close

The asymmetry that traps the cell

Why can't a cell simply fix itself? Because its copier, DNMT1, is faithful to the present, not to the truth, it duplicates whatever mark it lands on, mistake and all. An anchor protein like CTCF is different: it clamps a site back to the value the sequence dictates, ignoring the drift. Self-maintenance can only hold what remains; only a reference from outside the drifting copy can restore what's lost.

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

The text was never erased. We only lost the ability to read it.

That is the hopeful edge of this picture. Aging, in the information view, is lost access to instructions that are still physically present, a palimpsest, not a blank page. Which is exactly why re-reading can, in principle, make an old cell young again.

Our own analysis · a negative result

What we actually found

Everything above is the model. So we went looking for its signature in real human blood, reanalyzing public whole-genome methylation data for the tell-tale melting: within-cell disorder that should rise with age at composition-inert, fully-methylated sites. We did not find it.

01
No late-life melting

The age-drift trajectory is front-loaded and decelerating, a relaxation toward a floor, not the late-accelerating runaway the melting picture predicts.

02
The one melting-like signal is an artifact

Sorted CD4 T-cells, ages 18–25 against 82–86 (GSE79798), showed a small +0.008 read-level disorder gap. But it rose with CpG density, the fingerprint of per-base copy error, and it sat inside the pure lab-noise floor measured on identical DNA run through different bisulfite kits and sequencers (GSE128731: +0.022 to +0.049). A gap smaller than the instrument's own noise is not a biological signal.

read-level disorder gap · old minus young
lab-noise floor
+0.008
0
0.02
0.04
Plain takeaway

In blood, aging looks like per-site information decay plus shifting cell composition, with no detectable melting to reverse. This is a cross-sectional negative result at the resolution public data allows. The definitive single-cohort test is still pending.

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
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, in the section above, reports our own analysis of public DNA-methylation data. That analysis is a negative, in-model result: it is not a wet-lab therapy and not a proven cure. 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 dynamics shown are a cartoon of a subtler biology.

The one open, testable prediction

A cell's recoverability should track its surviving set of stable anchors, not its average methylation drift, the quantity today's "aging clocks" measure. Two cells with identical clock ages could differ completely in whether they can be rejuvenated. This is pre-registered and unproven; it is what would distinguish this model from a standard correlational clock.

The only new empirical claim here is a cross-sectional negative result: we did not detect a melting signature in human blood. The model's positive prediction above remains a hypothesis, not a finding.

Manuscripts in preparation