Myelin: The Insulation That Practice Wraps Around a Skill

Introduction

A mouse is put in a cage with an unfamiliar running wheel. Not the smooth one it has used before. This one has rungs spaced at irregular intervals, so it cannot simply run. It has to work out a stride, place its feet, correct itself, and go again, over and over, until the pattern stops being a problem. This is motor skill learning in the plainest sense there is.

Two and a half hours later, in the white matter beneath its cortex, cells that make myelin are already being born at an accelerated rate [1].

Not two and a half weeks. Hours. And when the same research group blocked the birth of those cells, leaving every strand of existing myelin untouched, the animals' learning fell apart inside roughly the same two to three hour window.

That result, and the larger one it came from, is the reason this article exists [2]. Search for myelin now and you will find a great deal of accurate material about what it is made of, how it speeds up nerve conduction, and what goes wrong in multiple sclerosis. What you will struggle to find on the first page of any search engine is the thing the field has spent the last decade establishing: myelin is not fixed hardware. It changes with what you do. In animals, that change is not a side effect of learning. It is part of the machinery.

There is a catch, and it is a big one, and most popular writing on this topic skips it entirely. The evidence that new myelin is required for learning comes almost entirely from mice. The strongest study of human white matter turnover points somewhere subtly different, and it deserves a section of its own. This article gives it one.

So here is what follows. What myelin actually is, and why calling it insulation is only about half right. Why the brain seems to care more about when a signal arrives than about how fast it travels. The twenty-year chain of experiments that took the field from "activity might matter" to "block this and the animal cannot learn". What happens when you run the same question in humans, and why the imaging everyone cites cannot see what people think it sees. Sleep, which doubles the supply of precursor cells. The lifespan, which turns out to be far longer than the usual "it finishes at 25" line suggests. And a fair reckoning with the popular book that put myelin in front of millions of readers and got the arrow right while getting the certainty badly wrong.

Some of this is settled. A lot of it is genuinely open. The open parts are marked as open, because a subject this widely repeated deserves better than a confident summary.

What Myelin Actually Is

Start with the physical object, because almost every later argument depends on getting it right.

A neuron sends its signal down a long thin process called an axon. Myelin is a sheath wrapped around that axon, and the crucial thing to understand is that it is not a coating applied from outside. It is a piece of another cell. A glial cell extends a flat sheet of its own membrane, spirals it around the axon many times, then squeezes almost all the cytoplasm and water out from between the layers so they pack tightly together [5]. What is left is a dense, lipid-rich, many-layered structure with very low electrical capacitance.

In the central nervous system, the brain and spinal cord, the cell that does this is an oligodendrocyte. One oligodendrocyte can send out many processes and myelinate segments on dozens of different axons at once. In the peripheral nervous system, the nerves running to your limbs and organs, the job belongs to Schwann cells, and each Schwann cell makes one segment on one axon [4]. That difference sounds like trivia. It is not, and it comes back later when we talk about why the two systems repair themselves so differently.

The wrapping is not continuous. It comes in segments called internodes, separated by short bare gaps called nodes of Ranvier. Those gaps are where the electrical action happens. Between them the axon is insulated, so instead of the signal crawling along the membrane point by point, it jumps from node to node. This is saltatory conduction, from the Latin for leaping.

The payoff is large. Myelination raises conduction velocity from under a metre per second to somewhere in the range of fifty to a hundred metres per second, and it does this without making the axon thicker [3]. That last clause is the important one. There is another way to make an axon fast, which is to make it enormous. Squid do exactly this, with a giant axon you can see without a microscope. Vertebrates found a cheaper solution, and myelin appears to have been invented independently more than once in evolution as an alternative to the giant-axon approach [8]. A brain full of squid axons would not fit in a skull.

The relationship between geometry and speed has been worked out in considerable detail since the 1980s, and it is not a single knob [7]. The number of wraps matters. The length of each internode matters. The diameter of the axon matters. And, as a more recent line of work shows, the length of the bare node between segments matters too. Hold onto that. It becomes the most interesting part of the story.

If you want the underlying cell biology of how the sheath grows and stays attached, that has been reviewed thoroughly [6]. For our purposes the summary is enough: a glial cell wraps a neuron in its own membrane, leaves regular gaps, and the signal leaps between the gaps.

None of this is new. All of it appears on every page currently ranking for the word myelin. The interesting part starts with what that sheath is doing besides insulating.

The Insulation That Also Feeds the Wire

Here is a fact that reframes the whole subject and almost never makes it into general coverage. The oligodendrocyte does not only wrap the axon. It supplies it with fuel.

Two papers published in Nature within a few months of each other in 2012 established this from different directions. The first showed that oligodendrocytes run on glycolysis and that this is how they maintain both myelin and long-term axonal integrity [9]. The second identified the delivery route: the main lactate transporter in the central nervous system is heavily concentrated in oligodendroglia, and disrupting it produced axon damage and neuron loss in both animal and cell culture models [10]. The authors of that second paper are explicit that this support works through mechanisms independent of myelination.

Think about what that means structurally. A long axon is an extraordinarily demanding piece of biology. It can run a metre or more, it fires constantly, and it is a long way from its own cell body and its own protein-making machinery. The cell wrapped around it turns out to be a support system as well as a cable jacket. Myelin, on this view, is life support that happens to also be fast.

This matters for how you read the rest of the article. When myelination is disrupted in the experiments described below, two things are being disrupted at once: how quickly signals travel, and how well the axon is being kept alive. Careful papers say so. Popular summaries almost never do. The relationship between the two has been reviewed at length [61], and it is one of the reasons a simple "more myelin equals better performance" model is too crude to survive contact with the literature.

Speed Is Not the Point. Timing Is.

Now for the part that changed the shape of this article.

Ask most people what myelin is for and they will say speed. It makes signals faster. That is true, and it is incomplete in a way that matters enormously for anyone interested in learning a skill.

Consider a problem the brain has to solve. Your thalamus sits deep in the middle of the brain and sends signals out to the cortex, which is spread across a much larger surface. Some of those journeys are short. Others are considerably longer. If conduction velocity were uniform, signals would arrive at very different times depending on how far they had to go, and any process that needs inputs to arrive together would be in trouble.

They do not arrive at different times. Work in mouse thalamocortical slices found that latency from thalamus to a cortical cell is remarkably constant, typically around two milliseconds, regardless of distance [11]. The mechanism turned out to be regional myelination. Conduction velocity on the first leg, from the thalamic nucleus out to the white matter, was roughly ten times faster than on the second leg, from the white matter into layer IV of cortex. Histochemical staining matched exactly: the first leg was heavily myelinated, the intracortical stretch much less so. Fast on the long variable stretch, slow on the short consistent one, and the total arrives on time. In these mice the physiology and the myelination pattern developed in parallel and were complete around the fourth postnatal week.

The brain is not maximising speed. It is equalising arrival.

Once you see the problem that way, a series of otherwise puzzling findings snap into focus. Node and internode properties in myelinated axons are tuned specifically to adjust action potential timing [13]. Conduction time regulation looks like a general design principle across systems rather than an incidental consequence of insulation [12].

And then there is the finding about nodes, which is the single most elegant result in this literature.

Measurements in rat optic nerve and cerebral cortical axons found that the length of the node of Ranvier, that small bare gap, varies over a 4.4-fold range in the optic nerve and an 8.7-fold range in cortex [14]. Crucially, node length varies far less along a single axon than it does between different axons, which is what you would expect if it were being set deliberately per axon rather than fluctuating at random. Modelling predicted that these differences alter conduction speed by around twenty per cent, comparable to what you get from changing the number of myelin wraps or the internode length.

Here is the part that makes it beautiful. For a given change in conduction speed, the amount of membrane the cell has to build or remove at the node is more than two hundred and seventy times smaller than the amount it would have to change in the sheath. If you were designing a nervous system and you wanted a fast, cheap, reversible way to tune when a signal arrives, you would not rebuild the insulation. You would adjust the gap.

For a coordinated skill, this is the more satisfying explanation. Playing a chord, forming a syllable, catching a ball thrown at your chest: these are not problems of raw transmission speed. They are problems of getting many signals to converge at the right moment. A system that can tune arrival times cheaply is a system built for exactly that. If you want the neuron-to-neuron layer of the same story, we cover it separately in our piece on how neurons communicate.

Twenty Years of Evidence That Activity Changes Myelination

The idea that neuronal activity influences myelination did not arrive fully formed. It was assembled over roughly two decades, and the sequence is worth walking through, because each step closed a loophole the previous one left open.

**1996: block the firing, block the myelination.** Researchers used neurotoxins to manipulate the electrical activity of neurons in culture. Tetrodotoxin, which blocks action potentials, inhibited myelination. Alpha-scorpion toxin, which increases firing, enhanced it [15]. The link between electrical activity and myelin formation was established, though the mechanism was not.

**2011: the messenger.** A study in Science showed that action potentials control local protein synthesis in the myelinating cell and drive the initial events of myelination [16]. Now there was a plausible molecular route from firing to wrapping.

**2014: stimulate a specific circuit in a behaving animal.** This is the study that converted a laboratory phenomenon into something with behavioural teeth. Using optogenetics, researchers stimulated the premotor cortex of awake, behaving mice [17]. Stimulation triggered a proliferative response in oligodendrocyte precursor cells, promoted the generation of new oligodendrocytes, and increased myelination in deep cortical layers and the white matter beneath. The stimulated animals showed improved motor function of the corresponding limb. Then came the necessary control: when the researchers blocked oligodendrocyte differentiation and the myelin changes, the behavioural improvement did not happen.

**2015 to 2019: which axons, and how many sheaths.** Work in zebrafish, where you can watch individual sheaths form in a living transparent animal, showed that neuronal activity biases which axons get selected for myelination in the first place [18], and that synaptic vesicle release regulates how many sheaths an individual oligodendrocyte ends up making [19]. Calcium transients in the forming sheath were then identified as the local signal that tells a nascent sheath to grow or retract [20] [21]. A signalling pathway involving endothelin was shown to mediate experience-dependent myelination [22].

**And the sensory version.** Changing visual input dynamically modulated myelination and measurably altered conduction velocity in the optic nerve [23]. Not just structure. Function.

Every one of these findings comes from animals. Mice, rats, zebrafish. That is stated here once and it applies to the whole section. It will be stated again, because the temptation to quietly promote a mouse result into a human claim is the single most common failure in writing about this topic.

The Experiment That Blocked Learning

Correlation is cheap. The studies above show that activity changes myelination. The obvious next question is whether that change is doing any work, or whether it is a byproduct.

In 2014 a study in Science answered it directly [2]. Adult mice learned to run on a complex wheel with irregularly spaced rungs, a task that requires genuine motor skill rather than just running. Production of new oligodendrocytes accelerated briefly while they learned. The researchers then used a genetic manipulation, deleting the transcription factor myelin regulatory factor in oligodendrocyte precursor cells, to block the production of new oligodendrocytes during adulthood.

The design of that manipulation is what makes the result convincing. It blocked new oligodendrocytes without affecting pre-existing oligodendrocytes or pre-existing myelin. The animals kept all the myelin they already had. They simply could not make more.

They could not master the wheel.

Two years later a follow-up study sharpened the timing dramatically [1]. The researchers identified an enzyme, Enpp6, whose messenger RNA marks newly forming oligodendrocytes, which let them watch differentiation as it happened rather than counting cells days later. Within two and a half hours of the mice meeting the complex wheel, production of Enpp6-expressing immature oligodendrocytes had accelerated in the subcortical white matter. Within four hours it had accelerated in the motor cortex. When new oligodendrocyte formation was blocked, learning was impaired within the same two to three hour window.

Then the same animals kept improving over the following week, accompanied by secondary waves of precursor proliferation and differentiation. The authors concluded that new oligodendrocytes contribute to both the early and the late stages of motor skill learning.

Sit with the timescale. Whatever this process is, it is not a slow structural adaptation that shows up after months of dedicated practice. It begins during the first session.

The finding then generalised beyond motor skills. A 2020 study found that water maze learning promotes oligodendrogenesis and new myelination in cortex and associated white matter tracts [24]. Preventing the learning-induced increase, again without touching existing oligodendrocytes, impaired consolidation of both water maze and contextual fear memories. The same study found something more specific and more interesting: contextual fear learning normally increases the coupling between hippocampal sharp wave ripples and cortical spindles, a signature of memory consolidation during rest, and that increase was abolished when oligodendrogenesis was suppressed. New myelin was not just present during consolidation. Something about the coordination between hippocampus and cortex depended on it. Our article on how sleep consolidates spaced learning covers the ripple and spindle machinery in more detail.

A separate 2020 study looked at fear memory over longer intervals [25]. Fear learning drove precursor cells in the medial prefrontal cortex to proliferate and differentiate. Mice engineered so they could not form new myelin showed impaired recall of remote memories but not recent ones. The distinction is important. Whatever new myelin is doing, it appears to matter more for holding a memory over time than for forming it in the first place.

A third 2020 study complicated the simple picture in a useful way [26]. Learning a forelimb reach task first transiently suppressed oligodendrogenesis, and only afterwards increased precursor differentiation, oligodendrocyte generation and sheath remodelling. Learning is not a monotonic myelin-adding process even in mice. The same paper reported that mature, already-existing oligodendrocytes could generate new sheaths, something the authors describe as previously considered controversial. That claim should be treated as recent and not yet settled, but if it holds it changes the picture substantially.

And a study on chemotherapy showed the reverse consequence: loss of adaptive myelination contributes to the cognitive impairment that can follow methotrexate treatment [27]. When the process is damaged, thinking suffers.

This is as strong a causal case as this kind of neuroscience produces. Manipulate the process, and the behaviour changes. Leave existing myelin intact, and the effect persists, so it is the new myelin doing the work.

In mice.

What Happens in Humans

Now the harder half.

The human evidence for training-related white matter change starts with a 2009 study that has been cited thousands of times [28]. Volunteers trained on a complex visuo-motor skill, and diffusion imaging afterwards detected a localised increase in fractional anisotropy in the white matter underlying the intraparietal sulcus. The authors described it, carefully, as the first evidence for training-related changes in white matter structure in the healthy adult human brain.

Earlier work had found something related in musicians [29]. Piano practice correlated with fibre tract organisation, and, intriguingly, the regions where the correlation appeared differed depending on whether the practice had happened in childhood, adolescence or adulthood. Childhood practice correlations were the most extensive and included the pyramidal tract. This is correlational data from people who chose to practise, so it cannot tell you which way the causal arrow runs, but the age-band pattern is suggestive of developmental windows.

The bridge between rodent mechanism and human imaging came from a study that ran both in the same animals [30]. Seventy-two adult rats were assigned to skilled reaching, unskilled reaching, or caged control. After eleven days of training, post-mortem diffusion MRI showed significantly higher fractional anisotropy in the white matter beneath sensorimotor cortex, specifically in the hemisphere contralateral to the trained limb. Immunohistology on a subset of twenty-four animals, eight per group, showed increased myelin staining in that same contralateral white matter and not in the untrained hemisphere. Within the trained hemisphere, myelin staining density correlated with how fast the animal had learned.

That is the closest anyone has come to demonstrating that the diffusion signal people measure in humans reflects myelin. It is good evidence. It is also, once again, rats.

The problem with the imaging

Here is where honesty costs the argument something, and where nearly every popular article stays silent.

Fractional anisotropy is not a measurement of myelin. It is a summary statistic describing how directionally constrained water diffusion is inside a voxel of tissue, and a great many things affect it: axon density, axon diameter, fibre crossing within the voxel, membrane permeability, and yes, myelination. A 2013 paper in NeuroImage laid this out with unusual directness, and its title names the problem: white matter integrity, fibre count and other fallacies [32]. The phrase "white matter integrity" gets used constantly and means, on inspection, almost nothing specific.

A broader review of what structural imaging during learning can and cannot show reached a compatible conclusion [31]: real changes are being detected, and the leap from those changes to a specific cellular mechanism is not one the data support on their own.

So when you read that six weeks of juggling practice "increased white matter integrity", the accurate version is narrower. Six weeks of practice was followed by a localised change in a diffusion measure, in a region plausibly related to the task, and that measure is influenced by myelination among several other things. That is still a real and interesting finding. It is not a myelin measurement.

The carbon-14 problem

And then there is the study that should change how anyone writes about this.

Above-ground nuclear weapons testing in the 1950s and early 1960s raised atmospheric carbon-14 sharply, and the test ban treaty then let it fall in a well-characterised curve. Carbon from the atmosphere enters the food chain, and when a cell divides and copies its DNA it incorporates carbon at the concentration prevailing at that moment. The DNA of a cell therefore carries a date stamp. Researchers used this to determine when human brain cells were born [33].

The findings for oligodendrocytes were not what the mouse literature predicts.

The number of oligodendrocytes in the human corpus callosum is established in childhood and remains stable afterwards. The population turns over at roughly one cell in three hundred per year. Meanwhile the myelin itself is exchanged at a high rate. The authors concluded that oligodendrocyte turnover contributes minimally to myelin modulation in human white matter, and that the modulation is instead likely carried out by mature oligodendrocytes.

Read the mouse work and the human work side by side and the tension is obvious. In mice, blocking the birth of new oligodendrocytes blocks learning. In humans, new oligodendrocytes are barely being born at all after childhood, yet myelin is clearly being remodelled.

There are at least three ways to resolve this and the field has not settled on one. It may be that mice and humans genuinely differ, with rodents leaning on new cells and humans on remodelling by existing ones. It may be that the human corpus callosum, which is where the carbon dating was done, is not representative of cortical regions where learning-related change is expected. Or it may be that sheath remodelling by mature oligodendrocytes is the shared mechanism all along, in which case the 2020 finding that mature oligodendrocytes can generate new sheaths [26] becomes far more significant than it first appeared.

Anyone who tells you which of these is correct is ahead of the evidence. The reviews that treat this question seriously present it as open [52] [53] [55].

Myelin Is Patchier Than the Diagrams Suggest

Every textbook diagram shows an axon evenly wrapped along its length, segment after segment, like sausages on a string. Real cortical axons do not look like that.

Researchers reconstructed single axons of pyramidal neurons in mouse neocortex from high-throughput electron microscopy and built high-resolution maps of where myelin actually sits [39]. Individual neurons turned out to have distinct longitudinal myelin profiles. Neurons in the superficial layers were the most varied, including a pattern nobody had described before: myelinated segments interspersed with long unmyelinated stretches. The authors' interpretation is that the distribution of myelin along an axon is an integral feature of a neuron's identity, and may have evolved as a way of modulating long-distance communication in cortex.

That is a genuinely different mental model. Myelin is not a uniform jacket. It is a pattern, and the pattern is specific to the cell.

A related surprise concerns which cells get wrapped. The intuitive assumption is that myelin exists for long projection axons carrying signals across the brain. Using light and electron microscopy of mouse neocortex, researchers found that a surprisingly large fraction of cortical myelin ensheathes the axons of inhibitory neurons instead, specifically parvalbumin-positive basket cells [40]. Half the myelin in layer 2/3. A quarter in layer 4. And the authors point out that because this myelin is patchily distributed over short path lengths, it is unlikely to meaningfully speed up when those spikes arrive. Whatever it is doing on those cells, it is probably not about velocity.

And sheath length is not purely dictated by the axon either. The prevailing assumption was that axons supply the molecular cues that set sheath length. Growing oligodendrocytes on microfibres, with no neurons involved at all, showed that they still produce sheaths of characteristic lengths, which means length is at least partly an intrinsic property of the oligodendrocyte [38].

Put these together with the node-length work and the tuning picture becomes rich. The nervous system has at least four separate levers it can pull to change when a signal arrives: how many wraps, how long each internode is, how long the bare node between them is, and which stretches of the axon get wrapped at all. The reviews describe this as a pattern being sculpted rather than a cable being thickened [59] [60] [66].

Sleep Doubles the Supply

If new oligodendrocytes matter, then the supply of precursor cells matters, and the supply turns out to be sensitive to something you might not expect.

A 2013 study profiled which genes oligodendrocytes were actively translating after sleep, after spontaneous waking, and after forced waking [41]. Hundreds of transcripts differed between the states, and the split was not random. Genes involved in phospholipid synthesis and myelination, and genes promoting precursor proliferation, were transcribed preferentially during sleep. Genes implicated in apoptosis, cellular stress response and precursor differentiation were enriched during waking.

The researchers then confirmed the functional consequence directly using BrdU labelling. Precursor cell proliferation doubles during sleep. It correlates positively with time spent in REM sleep specifically.

The reasoning behind why they looked is worth noting, because it is a nice piece of scientific inference. Glutamate released from neurons, through neuron-to-precursor synapses, can inhibit precursor proliferation. Glutamatergic transmission is higher during waking than during sleep. So the prediction was that sleep should release the brake, and it did.

This is mouse work. Whether human oligodendrocyte precursors behave the same way has not been established with anything like the same directness. But it fits neatly alongside the much larger body of evidence that sleep is when consolidation happens, and it gives that story a cellular limb it did not previously have. We treat the memory side of this in how sleep consolidates spaced learning, and the reason that spacing practice out beats cramming it together in the spacing effect.

Experience of other kinds also feeds into this system. Social isolation impairs adult myelination in the prefrontal cortex of mice [42], and a separate study identified a critical period during which social experience shapes oligodendrocyte maturation and myelination [43]. Myelination is responsive to the environment in a broader sense than practice schedules alone.

Across a Lifetime

The line you will see repeated everywhere is that the brain finishes myelinating around age twenty-five. It is one of those claims that is close enough to something true to be very hard to dislodge, and wrong in every direction that matters.

Start with the animal data, because it is the most direct. Longitudinal two-photon imaging of oligodendrocytes in living mouse cortex found that myelination is an inefficient and extended process, with half of the final complement of oligodendrocytes generated after four months of age [36]. Oligodendrocytes that successfully integrated were extremely stable, and they gradually shifted the overall pattern of myelination by forming new sheaths on unmyelinated and sparsely myelinated axons. Sensory enrichment clearly increased oligodendrocyte integration. It did not change the length of sheaths that already existed.

That last detail is worth holding onto, because it argues against the simplest version of the popular story. Enrichment did not thicken or extend what was there. It added new sheaths in new places.

A companion study imaging the same system over much longer periods found that total myelin coverage along individual axons kept progressing up to two years of age in mice, which is most of a mouse lifespan [37]. After the peak, gradual oligodendrocyte death and myelin degeneration set in, with pronounced internode loss and myelin debris accumulating inside microglia.

The precursor cells that make all this possible are distributed across the adult brain in a striking pattern, tiling it and maintaining spacing through mutual repulsion [35], and myelin remodelling in the healthy adult central nervous system was documented directly in the same period [34].

For humans, the best longitudinal data comes from a study of 103 people aged five to thirty-two, each scanned at least twice, for 221 scans in total [44]. All ten major white matter tracts examined showed nonlinear developmental trajectories. Projection and commissural tracts were mostly finished by late adolescence. Association tracts, the long-range connections that link distant cortical regions, showed within-subject maturation continuing after adolescence.

So the twenty-five figure is not invented. Some tracts do settle around then. But association tracts, which are exactly the ones you would expect to matter for complex integrated skills, keep changing into the thirties in this dataset.

Human myelination is also unusually protracted compared with other primates, which appears to be a species-level feature rather than an accident [45]. Structural change across the human lifespan has been mapped extensively [46], and quantitative MRI methods designed to be more specific than diffusion measures show widespread age-related microstructural differences [47]. Our article on how ageing changes memory covers the cognitive side of that arc.

The developmental picture pairs naturally with what happens to synapses over the same period, which we cover in synaptic pruning and the adolescent brain. Grey matter is being cut back aggressively while white matter is still being built. Two opposite-looking processes, running at once, both of them a form of refinement.

When Myelin Fails

A short section, because this article is not about disease, and because what myelin does is easiest to see when it is gone.

In demyelinating conditions, of which multiple sclerosis is the most familiar, the sheath is damaged and conduction along affected axons degrades. Signals slow, become unreliable, or fail. Because myelinated tracts run everywhere, the consequences depend entirely on which tracts are affected, which is why the clinical picture varies so much between people.

The central nervous system does attempt repair. Precursor cells can migrate to damaged areas and differentiate into oligodendrocytes that lay down new myelin, and understanding why this remyelination often fails, and how it might be encouraged, is a major research programme in its own right [48]. This is also where the difference between oligodendrocytes and Schwann cells becomes concrete: the peripheral nervous system regenerates considerably better than the central nervous system, and the properties of the myelinating cells are part of the reason.

One finding from the learning literature connects the two worlds. In the mouse study on motor learning and remyelination, learning a forelimb reach task after a demyelinating injury enhanced remyelination, from both newly generated and surviving oligodendrocytes [26]. Timing mattered: immediately after demyelination, neurons were hyperexcitable, learning was impaired, and the behavioural intervention gave no benefit at all. Only after partial remyelination restored function did learning start to help.

That is an interesting result and it is a mouse result. It is not a treatment, it is not advice, and nothing in this article should be read as guidance about any medical condition. The relationship between white matter and cognition in the clinical literature has been reviewed for readers who want that angle properly [62] [63]. Myelin abnormalities also turn up in unexpected places, including developmental conditions [67].

The Book That Was Half Right

In 2009 a journalist published a popular book that introduced myelin to an enormous general audience, and its central slogan, that practice makes myelin and myelin makes perfect, is probably how most people first encountered the word.

It deserves credit. It picked a real mechanism, years before the causal experiments described above existed, and it was directionally correct. Practice does change myelination, at least in animals, and the change is not incidental. That is a considerable thing to have got right in 2009 based on the evidence available then.

What has not held up is the certainty, and three specific claims are worth separating out.

**"Deep practice thickens the myelin sheath, and that thickening is what talent is."** No study demonstrates sheath thickening as the mechanism of human expertise. The human data are diffusion measures, which as discussed above are not myelin measures [32]. Meanwhile the animal work that does measure myelin directly points at new sheaths in new places rather than thickening of existing ones [36], and node geometry appears to be a cheaper and possibly more important lever than wrap count anyway [14].

**"More practice means more myelin means more skill, in a straight line."** Even in mice it is not a straight line. Learning first suppressed oligodendrogenesis before increasing it [26]. And the practice literature itself does not support the linear story. The deliberate practice framework that underpins the popular account was proposed in 1993 [49], and a 2014 meta-analysis across music, games, sports, education and the professions found that deliberate practice explains a considerably smaller share of the variance in performance than the popular version claims [50]. If practice does not explain performance as completely as advertised, then a mechanism for practice cannot explain talent as completely as advertised either.

**"You can build myelin deliberately, with the right regimen."** There is no human intervention with evidence behind it for this. Not a supplement, not a diet, not a training protocol. The honest position is that myelination responds to activity in animals, that human white matter changes with training in ways we can detect but not fully interpret, and that nobody has demonstrated a method for building myelin on purpose in a person.

Correcting a popular account is only worth doing if you are fair to it, so here is the fair version. The book pointed at something real and got there early. The field then did the hard work, and what it found is more conditional, more species-specific, and more interesting than the slogan. If you want the evidence on expertise itself rather than its cellular substrate, we cover it in the psychology of expertise, and the automaticity that practice eventually produces in how habits become automatic.

What This Does and Does Not License

It would be easy to end a piece like this with a list of practice tips. That would be dishonest. The biology does not support them, and what it does support is narrower, though more useful for being narrower.

**It licenses taking rest seriously as part of practice rather than as its absence.** Precursor proliferation doubling during sleep in mice [41], and consolidation depending on hippocampal and cortical coordination that itself depends on new myelin [24], together make a reasonable case that the interval after practice is doing work. This converges with a much larger and much more human body of evidence on spacing and consolidation, and that convergence is the reason to take it seriously, not the mouse study on its own.

**It licenses patience with the difficult early part.** In mice, the response begins within hours of first encountering a genuinely difficult task, and it is the difficulty that seems to matter. The wheel with irregular rungs, not the smooth one. That fits the broader finding that conditions which make practice feel harder often produce better retention, which we cover in desirable difficulties.

**It does not license a target.** There is no number of hours, no threshold of wraps, no measurable myelin state you should be aiming for. Nobody can measure your myelination, and if they could, the reading would not tell you what to do next.

**It does not license despair about age.** Association tracts are still maturing into the thirties in longitudinal human data [44], precursor cells persist across the adult brain [35], and remodelling continues in adults [34]. Nothing here says the window closes.

**It does not license buying anything.** No supplement, food or product has been shown to build myelin in a healthy person.

The most useful thing this literature offers is not a technique. It is a correction to a mental model. Practising something is not only strengthening connections between neurons, the process we describe in long-term potentiation. It is also, at least in animals and probably in you, changing the wiring that carries those signals, adjusting when they arrive rather than merely how strongly they fire. Two systems, working on different timescales, both changed by the same practice session.

What Is Settled and What Is Not

Because this subject is repeated so widely and so loosely, it is worth being explicit about the confidence attached to each claim.

**Settled.** Myelin is wrapped glial membrane, made by oligodendrocytes centrally and Schwann cells peripherally. It raises conduction velocity by orders of magnitude without requiring a thicker axon, by forcing the signal to leap between nodes of Ranvier. Oligodendrocytes also support axons metabolically, and axons degenerate when that support is withdrawn. Neuronal activity influences myelination, shown pharmacologically in 1996 and optogenetically in 2014. In adult mice, blocking the production of new oligodendrocytes impairs motor skill learning and memory consolidation while leaving existing myelin intact. Human myelination continues well past childhood, with association tracts finishing last.

**Reasonably supported but not proven.** That the same mechanism operates in humans. That the diffusion changes measured after human training reflect myelination specifically. That sheath geometry is tuned to equalise arrival times as a general principle rather than in the specific systems where it has been shown.

**Genuinely contested.** Whether human white matter plasticity is carried by new oligodendrocytes or by remodelling from mature ones, where the mouse work and the human carbon-dating work point in different directions [33]. Whether mature oligodendrocytes can generate new sheaths at all, described as controversial by the authors who reported evidence that they can [26]. Whether there are sensitive periods for human myelination and how firm their boundaries are. What diffusion imaging is actually measuring [32].

**Not supported.** Any claim that a specific practice regimen, supplement or diet builds myelin in humans. Any claim that myelination is the explanation for talent. Any claim that myelination stops at a fixed age.

The field itself has been moving fast enough that reviews from different years emphasise different things, and reading several of them together gives a better sense of the trajectory than any one of them alone [51] [54] [56] [57] [58] [64] [65].

What has not changed since the beginning of this article is the mouse and the wheel. It meets a problem it cannot solve by running. Within two and a half hours, cells beneath its cortex are already turning into myelin makers. Whatever the human version of that turns out to be, something in the physical structure of a brain starts moving early, and it starts moving because the task was hard.