Epigenetics: The Latch

The non-volatile memory tier the transcription loop was missing

The loop the genome chapter left volatile

DNA and the Living Lattice shows the cell’s regulatory cycle as a flux graph: the Mediator “nose” finds the sensed pattern at its source port, the lit locus drives Pol II, the transcript leaks as a modon coin, the ribosome catches and translates it, and the new protein re-tunes the smell the nose was matching. That loop is complete as written — and it is entirely volatile. It describes expression moment-to-moment. Cut the power — divide the cell — and nothing in the loop tells the daughter which loci were open. Yet a liver cell stays a liver cell through a hundred divisions, carrying a sequence identical to a neuron’s.

From the dna section the genome stores the code but it has no temporary modification ability, and yet that’s how cell’s work, they have differentiation, a lineage that holds across divisions. That silencing that keeps a transposon quiet for a lifetime, is non-volatile memory of an expression state, written once and copied forward through division. Epigenetics, DNA methylation, the histone marks, and chromatin compation are the cell’s flash: the layer that latches “which loci are open” into a form that survives the power cycle of mitosis. The framework can say what the latch is physically, and even put a number on it.

The latch is a deep-closed node, wrapped to its lifetime

The genome chapter reads a resting gene as a closed node — a store held coherent, drive port near zero — and transcription as the node sliding open at one locus. Health is “the capacity to slide”: keep most of the store closed and coherent, open exactly the loci the moment needs. Epigenetics is what sets the slide’s detents — what holds a locus closed (or poised open) when the volatile nose is not actively pushing it.

The framework’s own memory grammar says how a state is held: persistence is boundary-wrapping depth, read in time as ring-down lifetime. A more heavily wrapped topology hands its breath across more paired seams before it leaks, so it lives longer. Lay the epigenetic marks on that one axis and the biology’s own hierarchy of persistence falls out as a wrapping ladder:

Mark	Persistence	Wrapping reading
Histone acetylation	seconds–minutes	the lightest tag; an easily-opened node (RAM-like)
Histone methylation (H3K4/K9/K27)	hours–days	more shells; deeper detent
DNA methylation (5mC)	cell-generations, mitotically heritable	a covalent stamp-edit; a deep store (flash-like)
Heterochromatin / imprinting	lifelong, survives meiosis	the heaviest coil; the longest-lived topology

Wrapping DNA around a nucleosome, and nucleosomes into heterochromatin, is — in the chapter’s own language — adding boundary shells around the polar channel. Heavier wrap → deeper-closed node → longer ring-down. This is the reading the lifetime-of-the-stamp chapter wants, made cellular: the cell tunes a mark’s lifetime to the persistence it needs by tuning how heavily it wraps. Acetylation is the loose tag it can pull in seconds; constitutive heterochromatin is the coil it means to keep. Life has found many ways to wrap substrate energy into long-lasting modon-topologies, wrapped more heavily when they need to last a long time — that sentence is the chromatin packaging hierarchy, read through the node grammar already in the paper.

Propagation: reader–writer copying as boundary-matching

A latch that could not be copied would be useless: the daughter would lose the state at the first division. The mechanism that copies it is the framework’s own boundary-matching — “opposites attract,” the restoration of a broken coherence — run as a write rather than a read.

• Maintenance methylation. After replication a CpG is hemimethylated: the mark sits on the parental strand, the new strand is bare. That asymmetry is a coherence mismatch — a half-closed boundary. The maintenance methyltransferase (DNMT1) reads the hemimethylated state and heals it by writing the mark onto the daughter strand, restoring symmetry. The mark is copied because the hemimethylated topology is a mismatch the cell actively closes — structurally the same move the [4Fe-4S] cluster makes when it reads channel integrity, but here it writes the channel back to match.
• Histone reader–writers. PRC2 reads the H3K27me3 mark and writes the same mark on the neighbouring nucleosome; HP1 reads H3K9me and recruits the writer that extends it. A mark templates its own propagation along the fiber. This is a self-templating coherence-match — the same logic, one rung up from the strand.

So the structural feedback loop that propagates a trait is a self-templating boundary-match: a parent modon-topology forcing the daughter to match it. It is strong precisely because it rides a heavily-wrapped, long-lifetime topology rather than the volatile nose-RAM — the strength of the inherited signal is the wrapping depth that carries it.

The covalent mark as a stamp-edit — the one computed observable

Here the chapter can stop re-describing and compute. A covalent mark is a chemical modification of an aromatic base, and the codon-stamp model already turns a base into a substrate-displacement field \phi_B(\vec r). The principal DNA marks are all C5-substituents on cytosine (\text{C}\to\text{5mC}\to\text{5hmC}\to\text{5fC}\to\text{5caC}), so the model can be extended to them with one new feature lobe and the edit measured in the same metric as the codon matrix (scripts/epigenetics/methyl_stamp.py).

The chemistry fixes the most important fact before any number is computed. 5-methylcytosine still base-pairs as cytosine — the methyl sits at C5, which projects into the major groove, and leaves the entire Watson–Crick face (N4 donor, N3 acceptor, O2 acceptor) untouched. So the mark is invisible to the genetic letter and is read only as a feature in the groove channel — exactly the outward signal the genome chapter says the polar channel projects to protein readers. In one line: a covalent epigenetic mark is a stamp-edit that rewrites the channel address the nose smells, without touching the ROM sequence underneath it. That is what it means, in this framework, for a mark to be heritable yet silent.

The computation makes the direction precise, and corrects a naïve guess. One might expect methylation to nudge cytosine’s stamp toward thymine, since 5mC and T share the C5-methyl. It does not. With the per-base metric normalized so that, in the natural alphabet, d(\text{C},\text{T}) = 0.855 and a purine/pyrimidine transversion d(\text{C},\text{A}) = 1.04 sets the large end:

d(\text{C}, \text{5mC}) \approx 0.23, \qquad d(\text{5mC}, \text{T}) \approx 0.97 .

Methylation moves the stamp a real distance off cytosine — roughly a quarter of the full \text{C}\!\to\!\text{T} edit at the default amplitude, and between 0.14 and 0.51 of it across the parameter sweep — but away from thymine, not toward it: d(\text{5mC},\text{T}) is larger than d(\text{C},\text{T}) itself, because 5mC carries all of cytosine’s face plus a new major-groove feature. The edit is orthogonal to the \text{C}\!\to\!\text{T} axis. The biology’s \text{C}\!\to\!\text{T} connection is a different event — the spontaneous deamination of 5mC, which does rewrite the WC face and is the origin of the CpG mutational hotspot (below). The framework now separates the two moves the field has always lumped under “methylation chemistry”: the write moves the stamp into the groove; the decay collapses it onto T.

In a CpG context — the dinucleotide methylation actually targets — methylating the single C shifts the local two-base channel stamp by \approx 0.10, and the three-base contexts (TCG, GCG, ACG) by 0.07–0.09. These sit inside the synonymous-codon stamp-distance range (0.05–0.32): a single methyl mark perturbs the channel by an amount comparable to a silent codon swap, in the same metric — small, but not nothing, and read in the groove the framework already routes to readers.

A sensitivity sweep on the one new parameter (the C5-substituent amplitude) confirms what is robust: across amplitudes 1.0–4.0, d(\text{C},\text{5mC}) rises monotonically from 0.12 to 0.44 while d(\text{5mC},\text{T}) stays near or above d(\text{C},\text{T}) throughout. The magnitude rides on the amplitude exactly as the codon matrix rides on its one overall scale; the direction (a real edit, into the groove, away from T) does not.

Finally, the TET active-demethylation pathway \text{C}\to\text{5mC}\to\text{5hmC}\to\text{5fC}\to\text{5caC}\to\text{C} comes out as a closed trajectory in stamp space — a cycle of edits the cell drives the base around and pays for, net displacement zero. The individual leg lengths are model-dependent and not claimed; the closed-loop form is the framework-native reading of why demethylation is an energetic cycle rather than a simple erasure.

The CpG irony: the confound that became the signal

The codon-stamp ClinVar test came back negative: the apparent excess of damaging silent transversions dissolved confound by confound, and the residual signal lived only at CpG — “marks it mutational rather than structural.” But CpG is precisely the methylation site, and 5mC’s deamination to thymine is why CpG is a \text{C}\!\to\!\text{T} mutational hotspot. The confound that defeated the codon-stamp prediction is the fingerprint of the very mark this chapter computes. What was noise to the stamp metric is the epigenetic signal here: the same data the codon-stamp chapter had to discard as “mutational” is the deamination leg of the demethylation cycle above. The framework does not get the codon-stamp prediction back — it gets an honest account of why that prediction’s residual lived where it did.

Pure topological inheritance: prions

The cleanest case for “a coherence-match change that propagates a trait” carries no nucleic-acid change at all. A prion is a self-templating protein conformation: the misfolded form’s surface templates the conversion of the native form, and the state is inherited by daughter cells. Yeast [PSI⁺] and [URE3] are heritable elements with zero DNA edit — pure structure copying structure. In the framework’s reading this is boundary-matching stripped to its essence: a modon-topology copying itself by surface match, the same opposites-attract closure that runs the maintenance-methylation loop, with the template and the copy both made of protein. Prions are the strongest support case because they remove DNA from the argument entirely and leave only the topological inheritance the framework claims is primary.

The germline reset as a coherence-domain reset

Note

This subsection is the heavily-wrapped tail — flagged speculative, kept cautious.

Most marks are erased and reset in the germline: a coherence reset deeper than the domain fission the genome chapter reads in ordinary mitosis. The framework’s prediction is an ordering, not a mechanism of transmission: only the deepest-wrapped, longest-lifetime topologies — imprinted loci, some transposon silencing, the marks carried by small RNAs — should survive the reset, because survival through a coherence reset is exactly what wrapping depth buys. Transgenerational heritability, where it is real, should therefore track mark lifetime / wrapping depth, and the marks that make it across should be the heaviest-coiled ones. The framework does not claim a strong transgenerational channel exists where the biology says marks are cleared; it claims that whatever survives will be the most heavily wrapped, and that this ordering is testable against the catalogue of escapers.

Predictions and falsification

1. Methylation is an orthogonal groove-edit, not a C→T nudge — and the reader inventory already tests it. The substrate stamp of 5mC sits a clear distance off C (\sim\!0.14–0.51 of d(\text{C},\text{T})) but farther from T than C is: the mark adds a feature to the major-groove channel and leaves the Watson–Crick face that fixes the genetic letter untouched. The framework’s claim is therefore geometric — a reader that contacts the major groove at the CpG has a direct handle on the edit and should respond to it; a reader whose specificity comes only from the Watson–Crick edge has no direct handle and should be blind to it. This is largely a retrodiction, and it holds. The dedicated methyl readers — MeCP2, MBD1, MBD2 — grip the 5-methyl directly in the major groove through a conserved arginine “stair” motif¹ — exactly the major-groove readout the edit predicts. (The degenerate MBD3 and the repair glycosylase MBD4 are not methyl readers, and the framework does not count them.) Two sharpenings the inventory forces, both already contained in the stamp picture:

   • The framework predicts response, not sign. An added groove feature says a major-groove reader will react; whether binding is enhanced or inhibited depends on whether the feature completes or clashes with that protein’s contact. Both classes exist and neither falsifies the edit: Yin et al.’s methyl-SELEX survey of 542 human transcription factors sorts them into methyl-minus, methyl-plus, and lesser classes, with the extended homeodomain family the canonical methyl-plus case — HOXB13 makes a direct hydrophobic contact to the 5-methyl — while CTCF is methyl-inhibited at the specific motif positions where an aspartate reads the unmodified cytosine². Sensitivity tracks the groove geometry; the sign is set by the local contact, exactly as a stamp-edit into the channel should be.
   • A second, weaker channel runs through shape, not direct contact. The same edit that adds the major-groove feature also widens the major groove, narrows the minor groove, and stiffens the step³. So a minor-groove or shape reader with no direct contact on the methyl can still be moved — indirectly. The clean blind class is therefore the reader that takes its specificity from the Watson–Crick edge alone; the indirect movement of shape readers is not a failure of the prediction but its prediction-2 companion (the base-step stiffness edit) surfacing in a second reader population.

2. Single-mark channel perturbation is in the synonymous-codon range. A CpG methyl shifts the local channel stamp by \sim\!0.07–0.10. The prediction is that methylation’s effect on DNA charge-transport efficiency and on base-step stiffness should be of the same relative size as a synonymous codon substitution’s, and should scale with the number of CpGs in the read tract — a measurable, already-instrumented comparison (electrochemistry on methylated vs unmethylated duplexes of matched sequence).

3. Persistence tracks wrapping depth. Across the mark hierarchy, the erasure timescale of a mark should order with its boundary-wrapping depth (acetyl < histone-methyl < 5mC < heterochromatin), and the marks that survive the germline reset should be drawn from the heavy-wrapped end. A mark that persisted across generations while being lightly wrapped, or a heavily-wrapped mark that erased in seconds, would falsify the wrapping=lifetime reading.

4. The demethylation cycle is an energetic loop, not an erase. TET-driven \text{5mC}\to\text{5hmC}\to\text{5fC}\to\text{5caC}\to\text{C} is a closed trajectory in stamp space; the framework predicts the cell pays a loop cost to return the base to C rather than reaching it by the shortest edit, which is the deamination shortcut it actively suppresses (TDG excises the deaminated product). The signature is that the enzymatic path and the chemical-decay path reach the same endpoint (C or T) by different routes of different length.

Honest assessment

What is solid: the organizing claim — that epigenetics is the non-volatile memory tier the genome chapter’s volatile loop was missing, that a mark’s persistence is its wrapping depth read as ring-down lifetime, and that reader–writer copying is boundary-matching run as a write. These are re-descriptions; they are faithful to the existing framework and they reorganize known biology, but they are not derivations and predict no new number on their own.

What is computed: the stamp-edit a C5 mark makes on the codon-stamp field. The result that survives the one new parameter is directional — 5mC is a genuine edit into the major groove, orthogonal to and farther from thymine, with a CpG-context channel shift in the synonymous-codon range. This reuses the codon-stamp model’s calibrated per-base inputs and its single outstanding caveat (one unset overall scale); it adds one more placeholder (the C5 amplitude) and reports the sweep over it.

What is not solid: every absolute magnitude (two unset amplitudes now), the \text{5hmC}-vs-\text{5mC} ordering (parameter-dependent and not claimed), and the entire transgenerational tail, which the framework treats only as a wrapping-depth ordering over the marks the biology says survive. The strongest near-term test is prediction 1 — and checked against the known reader inventory it already holds as a retrodiction: the dedicated 5mC readers grip the methyl in the major groove, Watson–Crick-edge readers are blind to it, and the major-groove transcription factors split into methyl-enhanced and methyl-inhibited classes exactly as an added groove feature predicts (sensitivity fixed by geometry, sign by the local contact). What would turn this into a novel test is the framework’s quantitative ask — that a reader’s methyl-sensitivity scale with the computed groove-stamp edit, and that the WC-edge/major-groove split be drawn cleanly across the full inventory rather than read back from the well-known exemplars.

Putting the Section in Context

The genome chapter built the cell’s regulatory loop and left it volatile. This chapter adds the tier that makes the loop a cell rather than a momentary computation: the latch that writes “which loci are open” into long-lived, heavily-wrapped modon-topologies and copies them forward by boundary-matching. Read back against the three jobs, the mark is an anti-lock edit — it diversifies the channel address without changing the locking machinery that reads it — wrapped by lock-pole structure (the nucleosome coil, the heterochromatin) to whatever lifetime the state must keep. The sequence is the ROM, the nose is the RAM, and the methyl-and-chromatin layer is the flash the framework had drawn around without naming. Life writes its persistent choices the same way it writes its hydrogen spectrum and its B-DNA pitch — into the substrate, wrapped exactly as heavily as the choice needs to last.

Footnotes

1. Ohki, Y. et al., “Solution structure of the methyl-CpG binding domain of human MBD1 in complex with methylated DNA,” Cell 105, 487–497, 2001; Ho, K.L. et al., “MeCP2 binding to DNA depends upon hydration at methyl-CpG,” Molecular Cell 29, 525–531, 2008. A conserved arginine pair H-bonds the CpG guanines while contacting the two 5-methyls; MeCP2 in fact reads ordered major-groove hydration around the methyls rather than a bare hydrophobic bump.

2. Yin, Y. et al., “Impact of cytosine methylation on DNA-binding specificities of human transcription factors,” Science 356, eaaj2239, 2017; Hashimoto, H. et al., “Structural basis for the versatile and methylation-dependent binding of CTCF to DNA,” Molecular Cell 66, 711–720, 2017. Methyl-plus binding concentrates in the extended homeodomain family; CTCF’s sensitivity is position-specific, not a blanket block.

3. Rao, S. et al., “Systematic prediction of DNA shape changes due to CpG methylation explains epigenetic effects on protein–DNA binding,” Epigenetics & Chromatin 11, 6, 2018.