What kills the neuron isn't a toxin. It's a shape — and the shape copies itself.
Alzheimer's and Parkinson's sound like wear and tear: parts get old, parts break. But wear and tear can't explain two odd facts. The damage always starts in the same small patch and then moves on in an almost fixed order. And the neurons that die first are always the same few kinds — while the ones right next door can be untouched. This one goes down to the molecules, and what you find looks less like rust and more like contagion. By the time you notice your memory slipping, it has been running quietly for twenty years.
A protein is basically a chain of amino acids, and everything it does depends on folding into one particular 3-D shape — like origami, where the same sheet becomes a crane or a paper plane depending only on the folds. Folding it wrong isn't rare at all. Cells misfold things constantly, and clean them up constantly.
The dangerous part is that some wrong shapes have an extra trick: they work as a template. A misfolded copy grabs a normal neighbour and bends it into the same wrong shape. One becomes two, two become four. Amyloid-β and tau in Alzheimer's, α-synuclein in Parkinson's — three unrelated proteins, all running the same script: misfold → clump into small soluble oligomers → grow into ever-longer fibrils → pile up into plaques or inclusions (Lewy bodies, in Parkinson's).
That explains the strange shape of the timeline. It starts unbelievably slow — forming that first "seed" is hard, and this stretch is called the lag phase. Then nucleation happens and everything speeds up. Amyloid pathology in Alzheimer's has typically been accumulating for around twenty years before the first symptom. Which makes the word "onset" badly misleading: that moment isn't the beginning, it's the point where the pile finally tipped over.
The same dynamics have a twin in AI that people are now studying seriously: model collapse. Once a model is trained on content generated by itself (or by models like it), small distortions get copied forward as a template, drift compounds, and diversity eventually caves in. Note this is a match in dynamics, not in mechanism — proteins template by physical contact, models template through data recycling. But the skeleton is shared: when a system feeds its own output back in as input, error amplifies itself — nothing visible for a long time, then an avalanche.
If misfolding happens daily, why are you fine when you're young? Because the cell runs an entire quality-control and recycling crew: chaperones give a bad fold a second chance, small hopeless ones go to the proteasome to be shredded, big clumps get swallowed by autophagy and digested in lysosomes, and outside the cell microglia clean up. So the disease is usually less about making too much and more about not clearing it. Age slows every one of those lines down. proteostasis glia
Genetics points the same way. The two commonest risk genes in Parkinson's, GBA and LRRK2, both work on lysosomes and vesicle trafficking — jobs on the waste-disposal line, not on the α-synuclein production line.
And now the single most misread fact in this whole field: more plaque ≠ worse disease. Some people are packed with plaque and stay sharp until their last day. The explanation that currently holds up best is that a mature plaque is closer to sealed landfill, possibly even protective, and that the genuinely toxic species is the soluble oligomer still in transit — small, mobile, able to lodge in a synapse and take it apart. Meanwhile the thing that actually tracks cognitive decline, in both time and place, is a different protein: tau. Where tau pathology is heavy is where the tissue thins and the function goes first.
If this were wear and tear, you'd expect damage to bleed outward from a spot like rust on metal. It doesn't.
Tau pathology almost always starts in one small patch of entorhinal cortex, jumps to the hippocampus, and then spreads out across the neocortex — a sequence regular enough to be used as a staging system (Braak staging): look at the tissue and you can say which stop the disease has reached. Parkinson's is stranger still. α-synuclein shows up first in the dorsal vagal nucleus at the very bottom of the brainstem and in the olfactory bulb, then climbs level by level. By the time it reaches the substantia nigra in the midbrain and your hand starts to shake, more than half of those dopamine neurons are already dead.
The rule for the jumps is the whole point: it doesn't spread to whoever is nearby, it spreads to whoever it is wired to. A region far away anatomically, but joined by a direct fibre bundle, goes down earlier than the neighbour sitting right against it with no connection. There's also a bolder and more contested thread: α-synuclein may start in the nerves of the gut and travel up the vagus into the brain — if that holds, Parkinson's first stop isn't in the brain at all. vagus nerve
"Spreads along the wiring" can be written down as maths. Researchers take the whole-brain wiring map (which region connects to which, and how thickly — the connectome) and run a diffusion equation on it — the same family of tools as PageRank and as message passing in graph neural networks — using today's pathology to predict where the tissue will thin years from now. This isn't a metaphor: the predicted atrophy patterns line up decently with real follow-up scans. The topology of the brain network is itself the script the disease follows.
In the same patch of tissue, one population is wiped out and the one beside it is nearly untouched. Look at what the dead ones have in common and the answer turns out to be very physical: they are the ones with the biggest bills.
The dopamine neurons of the substantia nigra are the extreme case. A single one of these cells carries an axonal tree that, unrolled, is measured in metres in a human brain, with on the order of a million terminals, and it is almost unmyelinated (myelin is the insulating sleeve around an axon that makes signalling far cheaper). Worse, it doesn't wait to be told: it fires rhythmically on its own, like a pump that never switches off, calcium washing in and out, mitochondria permanently at full load. basal ganglia The entorhinal cells that go first in Alzheimer's fit the same profile: long-range projections, high metabolic upkeep. hippocampus & entorhinal
So neurodegeneration isn't a block of metal rusting evenly. It's more like a power grid where the most heavily loaded lines burn out first. That also clarifies what drugs can and can't do. Levodopa replaces the dopamine that's missing once the cells are gone; it can hold the symptoms back for years without rescuing a single neuron. neurotransmitter systems
The anti-amyloid antibodies are a different bet. They do clear plaque, and the two phase-3 trials reported in 2023 did slow cognitive decline by roughly a quarter to a third. Read that carefully: slower, not stopped — and the price includes a risk of brain swelling and microbleeds. Put it all together and you get an honest interim verdict: amyloid looks like the trigger, tau looks like the bullet — and the hard part is finding it during the twenty years before anything is pulled.
"A shape that copies itself" shows up in several fields that have nothing to do with each other: