Glial Cells

What glial cells are

Glial cells, collectively called glia, are the cells of the nervous system that are not neurons. They do not generate the electrical impulses that carry information, and for that reason they were long treated as secondary. Their name, coined in the nineteenth century, comes from the Greek word for glue, reflecting the early belief that they simply filled the spaces between neurons and held the tissue together.

The truth is far richer, and the reason is worth stating plainly, because it explains everything that follows. Neuronal signalling is chemically dirty and metabolically expensive. It works by throwing ions across a membrane and squirting transmitter into a gap, and both the ions and the transmitter have to be removed afterwards or the next signal will be corrupted and the cell may be killed. Neurons cannot do this for themselves: they are busy, they store almost no fuel, and they have no direct access to the bloodstream. So the brain evolved a second population of cells to hold the conditions steady while the first population computes. Glia supply neurons with fuel, build and maintain the insulating myelin that speeds nerve signals, instruct the barrier that protects the brain from the chemical turbulence of the blood, clear away debris and fight infection, and keep the ionic and transmitter balance around neurons inside the narrow limits that signalling requires. A neuron stripped of its glia would not merely be inconvenienced. It would fail within minutes.

How many glia are there?

A figure repeated in textbooks for decades held that glia outnumber neurons by about ten to one. Careful modern counting has corrected this. When the whole human brain is counted, glia and neurons turn out to be present in roughly similar numbers, close to a one-to-one ratio overall. The old ten-to-one claim was an overestimate that spread widely before it could be checked.

The ratio also varies sharply from region to region. In the cerebral cortex glia are more numerous than neurons, whereas in the cerebellum, which is packed with tiny granule neurons, neurons vastly outnumber glia. So the honest summary is that glia are about as common as neurons across the brain as a whole, with the balance tipping one way or the other depending on where you look.

Why the older figure was so wrong is instructive. Early estimates were made by counting cells in small samples of tissue and scaling up, a method that is easily thrown off by how unevenly cells are distributed. When newer techniques allowed the whole brain to be counted in a more even and systematic way, the inflated ratio collapsed. The lesson is not that glia are unimportant, far from it, but that a memorable number should not be trusted simply because it appears in many textbooks.

The main types of glia

Glia are not a single kind of cell but a family of several, and they have almost nothing in common with one another beyond not being neurons. Each type has its own developmental origin, its own shape, and its own job; microglia are not even of neural descent. Lumping them under one word is a historical accident, and the table below is best read as a list of four unrelated professions that happen to share a workplace.

Astrocytes: the cells that stop the brain poisoning itself

Start with the danger, because the danger is what astrocytes exist to answer.

Brain tissue is packed so densely that the extracellular space between cells is measured in tens of nanometres. It is a very small volume, and small volumes concentrate whatever is put into them. Neuronal signalling puts two things into that volume, thousands of times a second, and both are toxic in excess.

The first danger: potassium

An action potential ends by letting potassium ions rush out of the neuron. That is how the membrane repolarises. But the potassium has to go somewhere, and where it goes is the sliver of fluid outside the cell. Now consider what a neuron's resting voltage actually depends on. The resting potential sits close to the equilibrium potential for potassium, and that equilibrium potential is set by the ratio of potassium inside the cell to potassium outside it. Raise the concentration outside and the ratio falls, and the equilibrium potential for potassium becomes less negative. The membrane potential follows it. The neuron is now sitting closer to threshold than it should be, so a smaller input will fire it, and its neighbours, bathed in the same fluid, are in the same predicament. Every one of them fires more readily, dumps more potassium, and depolarises the rest a little further. That is a positive feedback loop, and its endpoint is synchronised, uncontrolled firing across a whole population of neurons.

Astrocytes prevent this. Their membranes are studded with potassium channels and they are extraordinarily permeable to potassium, so potassium accumulating in the extracellular space flows into them down its gradient. Better still, astrocytes are coupled to each other through gap junctions into a continuous network, so potassium taken up at a busy synapse can be passed sideways through the syncytium and released somewhere quieter, or dumped at an endfoot against a blood vessel. The local excess is not just absorbed, it is dispersed. This is potassium spatial buffering, and it runs continuously, invisibly, in the background of everything the brain does.

The second danger: glutamate

Glutamate is the brain's main excitatory neurotransmitter, which means almost every excitatory synapse releases it, and it must be removed from the cleft within milliseconds. If it is not, two things go wrong. The trivial problem is blurring: transmitter that lingers spills onto neighbouring synapses and the signal loses its address. The serious problem is lethal.

Glutamate acts on several receptor types, and one of them, the NMDA receptor, admits calcium. Calcium inside a neuron is not merely a charge carrier; it is a second messenger, and the cell keeps its internal calcium concentration thousands of times lower than the fluid outside precisely because calcium is used to command things. Sustained glutamate means sustained NMDA activation, which means calcium floods in and stays in. High intracellular calcium switches on proteases, lipases, and endonucleases, damages mitochondria, and generates free radicals. The neuron is dismantled by its own enzymes. This is excitotoxicity: death by over-stimulation.

Stroke makes the point unanswerable. When blood flow stops, ATP fails, the sodium-potassium pump stops, ion gradients collapse, and the astrocytic glutamate transporters, which depend on those gradients, stop working and can even run backwards. Glutamate accumulates in the extracellular space instead of being cleared from it, and the neurons around the infarct, many of which still have a blood supply, are killed by glutamate rather than by lack of oxygen. The dying core is a metabolic event. The expanding penumbra around it is a chemical one, and it is what happens when astrocytic clearance fails.

The mechanism: transporters and the glutamate-glutamine cycle

Astrocyte processes wrap the synapse and carry high-affinity glutamate transporters that pull glutamate out of the cleft against a steep concentration gradient, using the energy stored in the sodium gradient to do it. The astrocyte does not merely store what it takes up. It converts it. The enzyme glutamine synthetase, which neurons do not have but astrocytes do, adds ammonia to glutamate and turns it into glutamine, which is inert as a transmitter: it does not bind glutamate receptors and it cannot excite anything. The astrocyte then exports the glutamine to the neuron, which converts it back into glutamate and repackages it into vesicles ready for the next release.

That loop is the glutamate-glutamine cycle, and it is elegant for a reason worth spelling out. The brain does not want to synthesise its transmitter from scratch after every impulse, which would be slow and expensive, and it cannot afford to leave the used transmitter lying about, which would be fatal. So it hands the dangerous molecule to a neighbouring cell, which neutralises it into a safe carrier, and hands the safe carrier back. The astrocyte is both the sink and the supply line.

Only now does the standard sentence, that astrocytes regulate the chemical environment around neurons, mean anything at all. It means: they hold extracellular potassium low enough that neurons do not fire uncontrollably, and they hold extracellular glutamate low enough that neurons do not die.

The tripartite synapse: an astrocyte with a vote

Everything above describes an astrocyte cleaning up after a synapse. The next step is stranger. The astrocyte is not only listening to the synapse; it is answering back, and what it says can decide whether the synapse is capable of learning.

Anatomy first. A fine astrocytic process does not sit near the synapse, it envelops it, sheathing the presynaptic terminal and the postsynaptic membrane so closely that the cleft is effectively a three-walled chamber. That process is not a passive wrapper. It carries receptors for the transmitter passing through the cleft, so it detects when the synapse is active, and it detects which synapse, because its processes are compartmentalised finely enough to respond at individual contacts.

What it does with that information is chemistry, not electricity. An astrocyte cannot fire an action potential; it has no fast voltage-gated sodium channels to do so. Its signal is calcium. Transmitter binding to astrocytic receptors triggers a rise in intracellular calcium, which propagates as a slow wave through the cell and, through gap junctions, into neighbouring astrocytes. The astrocyte network therefore has a signalling language of its own: slower than the neuronal one by orders of magnitude, diffuse rather than point to point, and running underneath the electrical traffic rather than in it.

And the calcium wave has an output. It drives the release of gliotransmitters: signalling molecules released by a glial cell rather than a neuron, principally ATP, glutamate, and D-serine. These act back on the synapse and change how strongly it transmits. A synapse, on this account, is not two neurons and a gap. It is three cells, and the third one has a vote.

D-serine: the fact that changes the picture

Now the payoff, and it is the single most striking thing known about glia.

The NMDA receptor is the central molecular device of synaptic plasticity. It is the coincidence detector that underlies long-term potentiation, the cellular process most closely tied to learning: it opens only when the presynaptic cell has released glutamate and the postsynaptic cell is already depolarised, and the calcium it then admits is the trigger for strengthening the connection. If a synapse is going to learn, the NMDA receptor is how.

And glutamate alone will not open it. The NMDA receptor has a second, separate binding site, the co-agonist site, which must also be occupied before the channel will conduct. The molecule that occupies it at many synapses is D-serine, and D-serine is supplied by astrocytes.

Two cautions, because this is an active research area and the page should not overstate it. The relative contributions of D-serine and glycine at the co-agonist site vary between brain regions and receptor subtypes, and the precise machinery by which astrocytes release gliotransmitters is still argued over. What is not in dispute is the structural fact that the NMDA receptor requires a co-agonist, that D-serine occupies that site at many synapses, and that astrocytes are a principal source of it. The direction of the finding is secure even where the details are not.

Metabolic coupling: astrocytes at the blood vessel

Look at where an astrocyte puts its ends. One set of processes wraps synapses. Another set terminates in flattened plates, the endfeet, plastered against the outer surface of the brain's capillaries, covering nearly the whole vessel. A single astrocyte therefore touches the blood supply at one end and the signalling machinery at the other.

That arrangement is not a coincidence of packing. It is a supply chain. Glucose crosses from the blood into brain tissue through dedicated transporters, and astrocytes, sitting at the vessel wall, take up a large share of it. Astrocytes also store the brain's only meaningful fuel reserve, a small quantity of glycogen, which neurons do not keep. And astrocytes signal in the other direction too: their calcium activity is linked to the dilation and constriction of the small vessels they wrap, so active tissue can call for more blood. This coupling between local neural activity and local blood flow is the physical basis of functional brain imaging: an fMRI scan does not see neurons firing, it sees the blood response that follows, and astrocytes are part of the machinery that produces it.

Whether astrocytes also hand fuel directly to neurons is the contested part, and it should be reported as contested. On a widely discussed but still-debated model, the astrocyte-neuron lactate shuttle, astrocytes take up glucose, metabolise it, and pass lactate to neurons as a supplementary fuel during bursts of activity, with the glutamate they are simultaneously clearing acting as the trigger. The model is attractive because it ties the two jobs together, and it has considerable experimental support. But how much of the brain's working energy actually travels by that route, and whether neurons prefer lactate to glucose under normal conditions, remain genuinely disputed. State the shuttle as a leading hypothesis, not as a settled fact. What is not in dispute is the geometry: astrocytes sit between the capillary and the synapse, and nothing metabolic reaches most neurons without passing them. See brain energy metabolism for the wider picture.

How astrocytes induce the blood-brain barrier

It is often said that astrocytes "help form" the blood-brain barrier, which is true and almost useless, because it leaves out what they do. Here is what they do.

The barrier itself is not made of astrocytes. It is made of the endothelial cells that line the brain's capillaries, and its essential feature is that those cells are stitched to one another by tight junctions, protein seals that close the gaps between adjacent cells so that nothing can slip between them. Everywhere else in the body, capillary endothelium is deliberately leaky. In the brain it is not, and the difference is not something the endothelial cells decide for themselves.

They are instructed. Astrocytic endfeet cover the vessel and secrete signalling molecules that act on the endothelium, and those signals drive the expression of the tight-junction proteins and the specific transporters that let glucose and certain amino acids across while excluding almost everything else. Endothelial cells taken out of the brain and grown alone gradually lose their barrier properties; cultured with astrocytes, or with astrocyte-conditioned medium, they recover much of it. The barrier is therefore an induced phenotype, maintained by continuous glial instruction, not a fixed property of the vessel.

Two consequences follow immediately. First, damage the astrocytes and the barrier weakens, which is why astrocyte injury is part of the pathology in so many neurological conditions, and why a leaky barrier is a common downstream sign rather than a separate disease. Second, this is a case of a glial cell not supporting neurons directly but building the infrastructure on which every neuron in the brain depends. It is also, incidentally, why so many drugs that work beautifully in a dish do nothing in a living brain: they never get in.

Oligodendrocytes and Schwann cells: the myelin makers

Some of the most important glia are those that make myelin, the fatty sheath that wraps around axons and lets them conduct signals at high speed. Two different cell types share this job, working in two different parts of the nervous system.

What myelin actually does, and why "the impulse jumps" is wrong

The sheath is laid down in segments, and between consecutive segments the axon membrane is left bare over a short gap. Those gaps are the nodes of Ranvier, and they are not incidental. They are where the entire mechanism lives.

Now the mechanism, in the order in which it happens. Voltage-gated sodium channels, the proteins that actually produce an action potential, are concentrated at the nodes and are essentially absent under the myelin. So the insulated stretch of axon, the internode, cannot fire. It has nothing to fire with. What it can do is conduct passively, and myelin makes it very good at that: wrapping the axon in many layers of fatty membrane greatly increases the electrical resistance of the axon wall, so less current leaks sideways out of the fibre, and it reduces the wall's capacitance, so less current is wasted charging the membrane instead of travelling along it. The result is that when a node fires, the current it produces spreads down the internode quickly and with little loss, and arrives at the next node still strong enough to push it past threshold. That node fires, and does the same again.

So the impulse is regenerated only at the nodes. Nothing jumps. The action potential is destroyed and rebuilt at every node, and between nodes there is no action potential at all, only a rapidly spreading passive current. Record from a myelinated axon and the spike appears at successive nodes with the internodes silent, which is what gave rise to the word saltatory, from the Latin for leaping. It is a description of the appearance, not the mechanism, and it has been mistaken for one for a century. The neuron page sets out the same account from the neuron's side.

Consequence one: myelin is a metabolic saving

Consequence two: why demyelination is so disabling

Here the mechanism pays for itself, because it explains something that "myelin speeds signals up, so losing it slows them down" cannot. Losing myelin does not return an axon to the unmyelinated condition. It leaves it worse off than an axon that was never myelinated at all.

The reason is the channel distribution. An unmyelinated axon carries sodium channels along its whole length, so it can regenerate the impulse anywhere, everywhere, continuously. It is slow and expensive, but it is robust. A myelinated axon has made a bargain: it has stripped the sodium channels out of the internodes and concentrated them at widely spaced nodes, because the insulation will carry the current across the gaps. Strip the insulation off and that bargain collapses. The current now leaks out across long stretches of bare, channel-free membrane, and there is nothing in those stretches to rebuild it. The depolarisation arriving at the next node may be too small to reach threshold, and if it is, the impulse simply stops.

That is why demyelinating disease is so destructive. In multiple sclerosis, in which the immune system attacks central myelin, conduction along affected fibres slows, becomes erratic, or blocks outright, and the symptoms follow the tracts involved: disturbances of vision, sensation, movement, and coordination. It is not a mild degradation of an otherwise working system. It is the removal of a component the axon has reorganised itself to depend on.

Microglia: how a cell chooses which synapse to eat

Microglia are the resident immune cells of the central nervous system, and they have a different origin from the other glia, arising from the same lineage as the body's immune cells rather than from nervous tissue. Small and highly mobile, they patrol the brain constantly, their fine processes forever extending and retracting as they survey their surroundings for signs of trouble. There is nothing metaphorical about this: a microglial cell samples the entire volume of tissue around it every few hours.

When they detect injury, infection, or dying cells, microglia spring into action. They move toward the problem, engulf and digest debris, dead cells, and invading pathogens, and release signals that coordinate the brain's response to damage. In this they are both the cleaners and the defenders of nervous tissue, which matters because the blood-brain barrier keeps most of the body's ordinary immune cells out. The brain has to run its own immune system, and microglia are it.

Pruning, and the question everyone skips

Microglia also eliminate synapses. During development, and to a lesser extent throughout life, they engulf and destroy weak or surplus connections, which is how the brain refines an initially exuberant tangle of wiring into an efficient circuit. That much is usually stated and then abandoned, and the abandonment is where the explanation fails, because it leaves the central question unasked: how does a microglial cell, blundering about among a hundred trillion synapses, know which ones to eat? It cannot read the circuit. It has no way of knowing what any connection is for.

It does not need to. It is told.

Synapses that are weak or inactive become tagged with complement proteins. Complement is an ancient part of the immune system, far older than the brain, and its ordinary job elsewhere in the body is to coat a bacterium or a damaged cell in a molecular label that says: destroy this. The developing brain reuses that machinery. Complement proteins are deposited on synapses, and they are deposited preferentially on the ones that are not being used. Microglia carry receptors for complement. They engulf what carries the tag.

Two things follow. The first is that pruning is not tidying, it is sculpting: circuits are shaped by subtraction, and the pattern that survives is a record of what was used. It is the physical counterpart of the rule that connections which fire together are kept, which is the same logic that runs through neuroplasticity and long-term potentiation.

The second is that a mechanism this powerful is dangerous when it misfires. If complement can mark a synapse for destruction, then inappropriate complement deposition, or over-eager microglia, will destroy synapses that should have been kept. Excessive or misdirected synaptic elimination has become one of the leading hypotheses linking microglial biology to neurodevelopmental and neurodegenerative disease, and it is an active field rather than a closed one. What has changed permanently, however, is the status of the cell. Microglia are not a clean-up crew that arrives after the interesting work is done. They are one of the instruments by which the brain's wiring is decided.

Ependymal cells: lining the ventricles

Ependymal cells form a thin lining over the walls of the brain's internal cavities, the ventricles, and the central canal of the spinal cord. These cavities are filled with cerebrospinal fluid, the clear liquid that cushions the brain, carries away waste, and helps distribute nutrients and signals.

Ependymal cells help produce this fluid and keep it moving. Specialised ependymal cells of the choroid plexus secrete the fluid; the rest of the lining carries tiny hair-like projections, cilia, whose coordinated beating drives circulation through the ventricular system. In this quiet but essential way, ependymal cells maintain the fluid environment that bathes and protects the entire central nervous system.

The cerebrospinal fluid they tend does more than cushion the brain against knocks. By carrying nutrients in and waste products away, and by helping distribute chemical signals, it forms part of the brain's own internal plumbing, and it is central to how the brain clears the metabolic waste that accumulates during waking, a process that appears to run most effectively during sleep. A healthy flow depends on the ependymal lining working properly, and blockages in the ventricular system raise pressure inside the skull. Modest as they seem, ependymal cells are part of the machinery that keeps the brain's environment stable.

Active partners, not glue

The story of glia is, in large part, the story of how neuroscience corrected its own first impression, and the correction is now deep enough that it changes the question. It is not that glia turned out to be more useful than expected. It is that several of the things the brain most obviously does cannot be located in neurons at all.

Assemble the mechanisms. Whether a neuron survives an episode of intense activity is decided by whether an astrocyte cleared the glutamate. Whether a cortical circuit fires in an orderly way or slides towards synchronised discharge depends on astrocytic potassium buffering. Whether a given synapse can undergo the NMDA-dependent strengthening that underlies learning depends on whether an astrocyte supplies D-serine. Whether a signal reaches the far end of an axon in three milliseconds or thirty, and at what metabolic cost, is set by an oligodendrocyte. Which synapses survive development is decided by complement tags and the microglia that read them. And whether any of these cells receives fuel at all depends on a barrier that astrocytes instruct the vasculature to build.

None of these is a support function in any ordinary sense of the phrase. They are constitutive: remove them and there is no computation to support. The correct picture is not neurons served by an inert scaffold, but a single tissue in which two great families of cells hold different parts of one job, and the older habit of ranking them tells you more about the history of the microscope than about the brain.

Related topics