Epigenetic alterations and loss of proteostasis

In LON-2.1 you watched aging edit the source. Genomic instability lets somatic mutations pile up as repair falls behind, and telomere attrition runs a counter down toward zero. Both are damage to the letters themselves, the kind of change you first traced in S7.1. This lesson covers the other two hallmarks in that pair, and their defining feature is the opposite. The letters can be perfectly intact and the cell still fails. Epigenetic alteration corrupts the settings that decide which genes run. Loss of proteostasis lets the products of those genes, the proteins, pile up broken. One is config rot. The other is a failing garbage collector. Neither touches the genome sequence, and that is exactly what makes them slippery.

Epigenetic alterations: the config drifts

Recall from S8.3 the layer that sits above the transcription-factor logic. DNA methylation and histone marks decide whether a gene is physically reachable, open euchromatin or locked-down heterochromatin, without editing a single base. That lesson also flagged the property this one cashes in: the marks drift with age.

Here is the beat. Across a lifetime the epigenome loses its crispness. The genome-wide trend is a broad loss of methylation (global hypomethylation) alongside, at the same time, gains of methylation at specific promoters and CpG islands (focal hypermethylation). Heterochromatin that kept dangerous regions silent loosens, and stretches that should stay off begin to leak, including the transposable elements S6 warned you about. The net effect is a cell that gradually mis-sets which genes are on. Identity blurs. The cell keeps its genome but forgets, a little, what kind of cell it was supposed to be.

Why does it drift at all? Derive it from the maintenance mechanism in S8.3. After each division the new DNA strand starts bare, and a maintenance enzyme reads the old strand and re-writes the matching methyl marks. That copy is faithful but not perfect, and here is the crucial asymmetry. The DNA letters have a stored ground truth to correct against (the complementary strand, proofreading, and the repair systems of S7.2), while the methylation pattern has no external reference of record. Nothing holds a pristine baseline the cell can diff against and restore. So small per-copy errors are never caught. They accumulate, division after division, with no reset. That is the derivation of drift. High-fidelity copying without correction is still lossy over enough iterations.

For a programmer this is config rot: settings that slowly corrupt because they are copied forward forever and never reconciled against a known-good baseline. The DNA is the committed source. The marks are the environment config layered on top. Copy that config by hand on every deploy with no canonical file to reconcile against, and a fleet drifts into a hundred subtly different states.

Now the failure edge, because an analogy without its limit is a bug. Config rot implies two false things here. First, that a known-good baseline exists somewhere to diff against and roll back to. In a cell there is no stored pristine epigenome, the young pattern is gone, not archived. Second, that every drifted value is a defect to revert. Epigenetic drift is a mix of stochastic copy error and genuine, adaptive response to real signals (diet, stress, inflammation), so not every changed mark is a bug. Some of it is the system doing its job in a changed body. Read drift as a loss of fidelity with no rollback, not as clean corruption you could check out away.

Before we turn drift into a clock, make the underlying switch concrete. Drag the promoter methylation up and watch the gene cross its silencing threshold from ON to OFF. Then reframe what you are looking at. Aging does not move that slider on purpose. It nudges it by accident, one imperfect copy at a time, until a gene that should be on has drifted across the line.

The toggle shows one gene at one moment. Drift is what happens to a whole pattern over a lifetime of divisions. Read this toy model and trace the mechanism, do not expect to run it (in-browser execution is a later phase). The point is the asymmetry with DNA. With no reference to correct against, faithful-but-imperfect copying decays toward noise, and it does so at a steady, readable rate.

# A toy model of WHY methylation marks drift with age.
# Each division, a maintenance enzyme re-copies the pattern from the
# old strand. The copy is faithful but not perfect, and unlike the DNA
# letters there is no stored reference and no repair to correct against.

import random

def redivide(marks, error_rate):
    # marks is a list of booleans, one per CpG site (True means methylated).
    # Copy the pattern forward, flipping each site with a small probability.
    return [
        (not m) if random.random() < error_rate else m
        for m in marks
    ]

def fidelity_after(n_sites, divisions, error_rate):
    original = [True] * n_sites
    marks = list(original)
    for _ in range(divisions):
        marks = redivide(marks, error_rate)
    matches = sum(1 for a, b in zip(marks, original) if a == b)
    return matches / n_sites

# No correction step exists, so errors are never reversed on purpose.
# The pattern decays away from its young state at a steady rate, which
# is why the drift is regular enough to read as an age estimate later.
print(round(fidelity_after(1000, 0, 0.002), 3))    # 1.0    at birth
print(round(fidelity_after(1000, 200, 0.002), 3))  # ~0.72  mid life
print(round(fidelity_after(1000, 600, 0.002), 3))  # ~0.55  old, toward noise (0.5)

# Real drift is not this symmetric: aging shows global LOSS of methylation
# plus focal GAINS at specific promoters. The lesson here is only the
# asymmetry with DNA. Copying without a reference is lossy over a lifetime.

Because that decay is steady, it is predictable, and predictable is measurable.

Loss of proteostasis: the collector falls behind

Now follow the genes downstream to their products. Proteostasis (protein homeostasis) is the balance a cell holds across the whole life of its proteins: making them, folding them into shape (S4.2), shipping them, and destroying them when they wear out or come out wrong. It is not one machine but a budget. Synthesis on one side, quality control and disposal on the other, kept roughly balanced.

The quality-control side has three moving parts, all from the spine. Chaperones are proteins that help other proteins fold correctly and try to refold ones that have slipped, gripping sticky intermediates so they do not clump (recall from S4.2 that folding buries hydrophobic residues on the inside, and misfolding exposes them). When refolding fails, two disposal routes clear the protein. The ubiquitin-proteasome system tags a doomed protein with a small marker called ubiquitin and feeds it into the proteasome, a barrel that shreds it into short peptides that are later recycled back into amino acids (this is the targeted degradation you met in S8.4). Autophagy is the bulk route: the cell wraps larger junk and whole aggregates in a membrane and delivers them to the lysosome to be digested.

Here is the aging beat. With age every arm of this system weakens. Chaperone capacity falls, proteasome throughput drops, autophagy slows. Meanwhile the rate of protein damage does not fall, so the balance tips. Misfolded proteins that once got refolded or shredded now linger, and lingering is dangerous because of a mechanism you can derive. A misfolded protein exposes the hydrophobic surface that should be buried. Exposed hydrophobic patches stick to each other. So misfolded copies clump into aggregates, and the aggregates themselves clog the proteasome and soak up chaperones, cutting quality-control capacity further. That is a positive feedback loop. A little loss of proteostasis makes the next loss easier. Damage that a young cell would clear without noticing becomes self-amplifying in an old one.

The programmer's frame is a failing garbage collector. Chaperones are the runtime trying to repair a corrupted object in place. The proteasome and autophagy are the collector reclaiming what cannot be saved. In a young cell the collector keeps pace with corrupted objects as they appear. In an old cell it falls behind, and un-collectable objects accumulate like a leak of memory you can never free.

The failure edge is sharp. A garbage collector frees by reachability, reclaiming objects that nothing references. Proteostasis turns on conformation, whether a protein is folded right, which reachability cannot see, so a misfolded protein can be fully in use and still be the problem. And a collector treats garbage as inert, dead weight off the heap. Protein aggregates are often not inert. Some species are actively toxic, a gain of harmful function, and some of the worst end up outside the cell entirely, not just space taken up inside one process. The analogy buys you the throughput-versus-load intuition and then stops. Do not push it to "the junk is harmless as long as we ignore it."

The bridge: this is Alzheimer's and Parkinson's

The protein aggregation of ordinary aging is not a separate thing from neurodegenerative disease. It is the same phenomenon, sped up and pinned to specific proteins. In Alzheimer's disease, fragments called amyloid-beta, cut from a larger precursor protein, clump into plaques outside neurons, while a full protein called tau tangles up inside them. In Parkinson's disease, a protein called alpha-synuclein aggregates into deposits (Lewy bodies) inside neurons, and the dopamine-making cells that carry them die off.

Why neurons, more than most cell types? Derive it from what you already know. Most neurons are post-mitotic, meaning they stopped dividing early and last a lifetime. A dividing cell has a crude but real defense against aggregates: it can split its damaged load between two daughters and dilute it. A neuron cannot. Whatever it fails to clear, it keeps, for decades, on one cell. Add high metabolic demand and long, thin projections that are hard to supply, and the post-mitotic neuron becomes where a slow, lifelong proteostasis deficit shows up first and worst.

Keep the honesty sharp, because this is a field thick with overclaiming. The leading account, that amyloid-beta buildup triggers the Alzheimer's cascade, is a model, not a settled law. Anti-amyloid antibodies do clear plaques and slow decline modestly, which supports a causal role but does not prove amyloid is the whole story. That modest benefit is exactly what you would expect if soluble oligomers, tau, inflammation, and vascular factors all contribute too.

Two hallmarks, one shared shape. The genome can be letter-for-letter fine and the cell still ages, because the settings that read it drift and the machinery that maintains its proteins falls behind. Config that no one resets, and a collector that cannot keep up. In LON-2.3 you add the energy and signaling layer (nutrient sensing, mitochondria, and senescence) and start to see why these hallmarks form a wired network, not a checklist.