TOPIC 30 · PHASE D CLINICAL/FRONTIER

Neuroplasticity & Rehabilitation

The brain can change for life — but which way does it change by default?

2026-08-06 · BigCat

One rule turns you into a pianist — and the same rule finishes off a hand after a stroke.

"Neuroplasticity" has been flattened into a motivational slogan: the brain can change, so nothing is too late. The real thing is far colder. Plasticity isn't a reward. It's a neutral accounting rule — connections that get used are reinforced, connections that don't are dismantled. It doesn't care what you want. That uncooperative hand after a stroke is being sentenced by exactly this rule, and what rehabilitation medicine has spent decades doing is, bluntly, wrestling it back before the rule writes the hand off for good.

// 01

The brain recovers on its own for a while — then stops

When a vessel blocks, the tissue at the centre is dead within minutes: the infarct core. Around it sits a ring of tissue that is starved but not yet dead — the penumbra. That ring is what emergency treatment is racing for.

The interesting part comes over the following weeks. Injury pushes the surviving tissue nearby into an unusual state: the inhibition that normally stops things from growing is lifted, growth genes that are otherwise mostly a fetal affair switch back on, and new synapses sprout fast. In effect the adult brain drops back, locally and temporarily, into a stretch of development. The window is clean in animal work; in people it corresponds roughly to the first one to three months — the same amount of training bought during that stretch pays off far more. (How connections strengthen and weaken: synaptic plasticity)

sensitive window plasticity function regained stroke 1 mo 3 mo 6 mo 12 mo how growth-permissive the brain is function regained
The weeks of peak plasticity land right where function is only starting to climb

There's also a much-quoted near-law, the proportional recovery rule: many patients end up regaining about 70% of what they could in principle have regained. It sounds uncanny — but a later wave of work argued the 0.7 is largely a statistical artifact. Using "maximum possible recovery" as the denominator mechanically squeezes the scatter onto a line, and ceiling effects pile on top. Read it as a rough group-level pattern, never as a prognosis for one person.

// 02

Use it or lose it — and losing it is the default

The scene every rehab ward knows: the affected hand can still move a little, and the patient uses only the good one, all day, every day. That's not laziness. Edward Taub found the mechanism in monkeys last century — cut the sensory nerves to one arm and the animal stops using it entirely, even though the pathways that drive the muscles are intact. The reason is mundane: in the weeks after the injury, every attempt fails, feels clumsy, and gets solved faster by switching hands. So the brain learns one thing very quickly — don't use it. Once that learning sets, "can it move" and "will it be used" have come apart for good. This is learned non-use.

this hand fails use the good one its territory shrinks CIMT: restrain the good hand every lap around the loop shaves a little more off "can"
Learned non-use is a self-fulfilling loop — therapy has to cut one of its links

Run the mechanism backwards and the therapy falls out: constraint-induced movement therapy — mitt or sling on the good hand, several hours a day of real tasks with the affected one (pouring water, turning pages, opening doors), the movements broken into steps that are just within reach, with immediate feedback on every bit of success. The EXCITE trial in 2006 confirmed it genuinely improves arm function in subacute patients.

But "earlier and harder is better" is wrong, and two trials found that out the hard way. In 2009, VECTORS found that doubling the dose of very early constraint therapy did worse than the standard dose. In 2015, AVERT started high-intensity out-of-bed activity within 24 hours of stroke and found worse outcomes at three months. The window is open, but the tissue is also at its most fragile.

At the other end the problem is the exact opposite — the dose is usually far too small. Researchers have gone and counted the actual movements in rehab sessions: an upper-limb session averages only around thirty useful repetitions, while animal work needs hundreds before cortical maps visibly move. (How commands travel from cortex to muscle: the motor system & cerebellum)

// 03

Maps get redrawn — including badly

The cortex carries a body map: a patch for the fingers, a patch for the face, laid out side by side. Old textbooks drew it as a fixed little homunculus, as if stamped in at the factory. Michael Merzenich's monkey experiments demolished that impression: use a finger more and its cortical territory expands; sew two fingers together so they can only move in lockstep and their two territories fuse into one. The proportions of the map are being rewritten by how much each part gets used.

hand in daily use face thumb index middle ring little hand unused for months weeks to months later face (expanded) thb ind mid rng lit same strip of cortex, territory reallocated by use
What gets used takes ground; what doesn't gets eaten by its neighbours

Stroke rehab bets everything on this. Randolph Nudo's monkey work showed that after an infarct, if there's no training, the surviving hand territory around the lesion keeps shrinking — while targeted hand training not only holds that ground but can push it outward.

Mirror therapy is a trick built on the same logic: put a mirror between the two hands and have the person watch the reflection of the good hand moving. The brain receives visual evidence that the affected hand is moving normally, and the silent territory gets a chance to light up again. Ramachandran first used it against phantom limb pain; in stroke hemiparesis there is now moderate-quality evidence that it improves motor function.

But a map that can be redrawn can also be drawn wrong. Some professional musicians develop focal dystonia: decades of intense practice with several fingers moving in strict lockstep smear finger territories that should stay separate into one blur, so moving one finger drags the neighbours along — practice itself ruined the map. Phantom limb pain and central sensitisation in chronic pain are the same story. Plasticity never guarantees the outcome is good.

AI cross-read

Take a trained large model and fine-tune it (keep training briefly on data for one new task so it gets good at that thing) and its internal representations get redrawn much like a cortical map — the new task's "territory" grows. The price is identical: the standard side effect is catastrophic forgetting, where learning the new costs you the old. Biology's countermeasures are to keep the rewriting local and to make most of it wait for sleep before it sets; engineering's are to mix old data back in or to let only a small slice of parameters move. Being changeable and being ruinable are two faces of one property on both sides.

// 04

Plastic for life — with the brakes on

A child's brain really is easier to change; that part isn't an illusion. Hubel and Wiesel's classic experiment: cover one eye of a kitten for a few weeks early in development and that eye's cortical territory is permanently taken over by the other — the sight never comes back. The same deprivation in an adult cat does almost nothing. The human counterpart is amblyopia, which becomes much harder to treat once the window has passed. Windows like this are critical periods.

The counterintuitive part: a critical period closes not because the capacity has been used up, but because the brain actively applies a brake. At a certain stage of development, a mesh of sugar-proteins called a perineuronal net shrink-wraps certain neurons and fixes the wiring already in place, and myelin carries growth-inhibiting signals of its own. The evidence is hard: enzymatically digesting those nets in animals reopens the critical period in adults. (How these windows open and shut: neural development)

So adult plasticity comes into focus as what it actually is: real, but expensive. It demands enough repetitions, difficulty pinned to the edge of what you can do, attention genuinely present (zoned-out passive repetition barely counts), and sleep to lock in the day's changes. As for whether the adult human hippocampus still grows new neurons — two groups using different methods reached opposite conclusions in 2018 and it is still unsettled, so any product selling "adult neurogenesis" is running ahead of the evidence.

In the end, "the brain is plastic" is not "you can change whatever you want". It's a rule with windows, conditions and costs, not an incantation.

AI cross-read

Neural networks lose plasticity too, and it's been quantified repeatedly: train a network across a long sequence of tasks and its ability to learn new ones visibly degrades — some units saturate and stop moving, a brake the system grew for itself. The engineering fixes read like a pointed echo of the biology: periodically reset a small fraction of units, inject noise into the weights, or regularise parameters back toward a range where they can still move — the same idea as digesting perineuronal nets to reopen a critical period. And both hit the same wall: turn plasticity back up and the old, settled competence starts coming loose with it.

🌀 Crossing over · interdisciplinary echoes

"What you do repeatedly remakes you into the kind of person who does it" — several old traditions wrote this rule down long ago, and remarkably they got the fine print right too:

// Going deeper

If it really is "use it or lose it," does "keep your brain busy to stave off dementia" follow?
The direction is right, but far narrower than the slogan. Education, complex work and social life do correlate with later symptom onset — the mechanism is called cognitive reserve (Topic 25) — but that looks more like "the pathology accumulates anyway, you just hold out longer" than like keeping the disease out. And plasticity transfers very poorly: you get good at what you practise. Brain-training apps mostly make you good at the app, and evidence for transfer to everyday cognition has stayed thin. The interventions with genuine cross-domain effects are the whole-brain conditions — physical activity, sleep.
If critical periods can be reopened with drugs or enzymes, could an adult learn a language like a child?
Locally, animal work has done it. But what reopens isn't only learning capacity — it's instability. Things got locked precisely because they were already in use. An adult with critical periods reopened at scale might learn fast while existing skills and memories wash out around them. So the clinical direction actually being tried is narrow and targeted: loosen the brake only in the circuit that needs remodelling, only for the stretch of time when training is happening.
Rehab insists attention must be present — so why do robot-assisted and passive movement work less well?
Plasticity isn't triggered by how many times the muscle moved; it's triggered by the nervous system actually solving a problem. When a machine completes the movement for you, the error is erased externally — and so is the internal "didn't make it" signal that drives the rewriting (an old friend from Topic 1). That's why good rehab design keeps difficulty parked at "just barely out of reach": assistance supplies only the last sliver, and the rest has to be earned.
If plasticity is neutral, why does chronic pain get worse the longer it lasts?
Because pain pathways obey the same accounting rule. A persistent pain signal running the same route over and over strengthens those synapses and lowers the threshold, until mild stimulation — or none at all — is enough to set the route off. That's central sensitisation. At that point pain is no longer faithfully reporting tissue damage; it's a state the system has learned. Which is also why blocking the pain alone often isn't enough: what has to be shifted is the learning that got written in.

// Further reading