Brain Energy and Metabolism

The bill: what the brain actually costs

Begin with the arithmetic, because it is the fact from which everything else in this page follows.

Per gram, the brain is roughly an order of magnitude more expensive than the average of the rest of you. That is an extraordinary metabolic commitment for a body to make, and evolution does not make such commitments casually. It is also not a fixed feature of being human at all ages: in early childhood the brain's share of the body's energy budget is considerably higher than in adults, at a period when the brain is comparatively large relative to the body and is engaged in the extremely costly business of building and pruning its connections. The metabolic demand of the growing brain is significant enough that it plausibly constrains how fast a child's body can grow.

It is worth pausing on why this is surprising. The brain does not move, it does not secrete in bulk, it does not filter litres of fluid. Compared with a working muscle it appears to be doing nothing at all. And yet it is one of the most metabolically demanding tissues in the body, hour after hour, whether you are running a marathon of mental arithmetic or lying still in a dark room. Something invisible is consuming an enormous amount of energy continuously. Identifying what that something is, is the intellectual core of this subject.

No reserve tank: why minutes matter

The muscles keep glycogen. The liver keeps glycogen. Adipose tissue is a vast store of chemical energy that can be drawn on for weeks. The brain, alone among the body's great consumers, keeps essentially nothing.

There is a small amount of glycogen in astrocytes, which we will come back to, but it is trivial against demand: it could sustain the brain for a matter of minutes at most, and it is not a meaningful reserve. Neurons themselves store almost no fuel. This means the brain is entirely dependent on continuous delivery, by the blood, of glucose and oxygen, through the vessels that supply it and across the blood-brain barrier.

Note also what this implies about the vasculature. To keep an organ with no reserves supplied at ten times the average tissue's rate, you need a delivery system of remarkable density and remarkable responsiveness. The brain gets around fifteen per cent of the blood the heart pumps, and its capillary network is dense enough that almost no neuron is far from a vessel. The regulation of that flow is not a passive matter, and it turns out to be one of the more elegant pieces of physiology in the body, which is the subject of a later section.

Where the money goes: the pump that never stops

Now to the central question. An organ that does not move is consuming a fifth of the body's energy. On what?

The answer requires a short detour through how neurons signal, which is treated fully under the action potential and ion channels and electrolytes. A neuron at rest is not electrically neutral. It maintains a voltage across its membrane, typically around minus seventy millivolts inside relative to outside, and it does so by holding ions at wildly different concentrations on the two sides. Sodium is concentrated outside the cell; potassium is concentrated inside it. These are not equilibria. They are gradients, held away from equilibrium, and like water held behind a dam they represent stored energy.

Signalling spends that stored energy. When an action potential fires, voltage-gated sodium channels open and sodium rushes in down its gradient, which is what makes the membrane potential swing positive. Then potassium channels open and potassium rushes out down its gradient, restoring the voltage. Similarly, when a neurotransmitter opens receptor channels at the synapse, ions flow down their gradients and produce a postsynaptic potential. In every case, the signal is the controlled release of stored gradient energy. The dam gates open, water flows, work is done.

And here is the point. Every one of those events leaves the cell with slightly more sodium inside and slightly less potassium inside than it started with. The gradients are a little flatter. The dam is a little lower. If nothing intervenes, after enough signalling the gradients equalise, the membrane potential collapses to zero, and the neuron can never fire again. The cell is not damaged; it is simply out of ammunition.

This yields the sentence that the whole subject turns on: thinking is expensive because signalling is expensive, and the cost of signalling is not the signal but the cleaning up afterwards. The action potential is, in energetic terms, close to free at the moment it happens: it is a controlled discharge of energy already stored. The bill arrives immediately afterwards, when the pump must restore the gradient so that the neuron is ready to do it again. A brain that fires a great deal is a brain that must pump a great deal, and pumping burns ATP, and ATP is regenerated by burning glucose with oxygen, and glucose and oxygen must be delivered by the blood without interruption. The entire chain, from the metabolic bill down to the clinical urgency of stroke, hangs from the electrochemistry of a membrane.

Two further costs follow from the same logic. Neurotransmitter release requires calcium entry, and that calcium must then be pumped back out or sequestered, again at the cost of ATP. And released neurotransmitter, glutamate above all, must be cleared from the synaptic cleft, a job performed largely by astrocytes using transporters that are themselves driven by the sodium gradient, which means the astrocyte in turn must run its own pumps to restore what the uptake spent. The bill for one synaptic event is therefore paid by several cells.

The energy budget of Attwell and Laughlin

The above is a qualitative argument. In 2001, David Attwell and Simon Laughlin turned it into an accounting exercise, in a paper in the Journal of Cerebral Blood Flow and Metabolism that has shaped how the field thinks about the question ever since. Their method was direct: take what is known about the biophysics, work out how many ions cross the membrane per action potential and per synaptic event, work out how many ATP molecules the pump must therefore hydrolyse to put them back, multiply by the number of neurons and synapses and by realistic firing rates, and see what total falls out and how it divides.

Their conclusion was that the great majority of the grey matter's energy expenditure is attributable to signalling and to the restoration of ion gradients it necessitates, rather than to the cell's basic housekeeping such as protein synthesis and membrane maintenance. Within that signalling cost, the largest components were the postsynaptic effects of neurotransmitter release, the reversal of ion movements at the synapse, and the action potentials themselves, with the mere maintenance of the resting potential taking a considerably smaller share than one might guess.

The budget also gives a principled reason for why functional imaging works at all. If the metabolic cost is driven by synaptic activity, then measuring metabolism or blood flow in a region is measuring, in effect, how much synaptic traffic that region is handling. This connects directly to the finding, discussed on the brain imaging methods page, that the fMRI signal tracks local synaptic and dendritic processing more closely than it tracks neuronal output firing.

The baseline is the story, not the task

There is a striking and under-appreciated implication of all this, made forcefully by Marcus Raichle and Debra Gusnard in a 2002 commentary in PNAS. The enormous energy consumption of the brain is almost entirely a baseline. It is the cost of the brain simply being on.

When a person performs a demanding cognitive task, the metabolic increase in the regions recruited by that task is, in relative terms, small: a modest percentage above the local resting rate. And those regions are a small proportion of the whole brain. The net effect on total cerebral energy consumption is therefore, against a baseline that large, close to noise. The brain of a person doing nothing in particular is consuming very nearly as much energy as the brain of a person concentrating hard.

This has three consequences worth stating explicitly.

The astrocyte-neuron lactate shuttle: a live dispute

So far, so settled. Now to a part of the field where the honest answer is that we do not know, and where a great deal of teaching material states as fact something that is genuinely under argument.

The question is simple to state: when a neuron needs fuel in a hurry, where does it get it from? The obvious answer is that it takes up glucose from the blood directly, and neurons certainly can and do. But an influential alternative was proposed in the 1990s, principally by Luc Pellerin and Pierre Magistretti, and it is called the astrocyte-neuron lactate shuttle.

It is an elegant model. It explains why glycolysis appears to run ahead of oxygen consumption during brief activation. It gives the astrocyte, long treated as mere scaffolding, a central metabolic role. It fits with the observation that astrocytes hold the brain's only meaningful glycogen store. And it has been taught in textbooks and lecture theatres for two decades.

It is also, in its strong quantitative form, contested. A substantial body of work argues that neurons express glucose transporters and hexokinase perfectly adequate to take up and use glucose directly, that they do so under most conditions, and that the evidence for lactate being the dominant neuronal fuel during activation rests heavily on cell-culture preparations and on modelling assumptions that may not hold in the intact brain. Some studies have reported patterns of glucose uptake that appear to favour neurons rather than astrocytes. The dispute is not about whether lactate can be a fuel for neurons, which is agreed, nor about whether astrocytes can produce and export it, which is also agreed. It is about the quantitative significance in vivo: whether the shuttle carries a major share of the neuron's fuel in a working brain, or whether it is a real but supplementary pathway alongside direct neuronal glucose uptake.

Ketones: a real alternative fuel, with caveats

The brain is usually described as glucose-dependent, and under ordinary conditions that description is accurate. But it is not the whole picture, and the exception is important physiology rather than a diet-industry talking point.

Fatty acids, the body's main long-term energy store, largely cannot serve the brain: they are bound to albumin in the blood and do not cross the blood-brain barrier in useful quantities. That would seem to leave the brain helpless during a fast, entirely dependent on a glucose supply that a starving body would struggle to maintain, and it would predict that a human being should suffer catastrophic cerebral failure within a couple of days without food. That plainly does not happen, and the reason is ketones.

When carbohydrate is scarce, whether through fasting or sustained carbohydrate restriction, the liver takes fatty acids and converts them into small, water-soluble molecules called ketone bodies, chiefly beta-hydroxybutyrate and acetoacetate. These, unlike the fatty acids they came from, are carried across the blood-brain barrier by monocarboxylate transporters, and once inside they can be fed into the mitochondria of neurons and used to make ATP. The transporters are upregulated with sustained exposure, so the brain's capacity to use ketones increases over days and weeks of adaptation. In prolonged fasting, ketones can come to supply a large share of the brain's energy needs.

Two caveats keep this honest. First, the substitution is not total: some glucose remains necessary, and the body ensures a supply through gluconeogenesis, which is the manufacture of glucose from substrates such as glycerol and amino acids. Certain cells and certain biosynthetic pathways in the brain require glucose specifically, so ketosis reduces the brain's glucose requirement substantially but does not abolish it. Second, and importantly, none of this establishes that a ketogenic diet is beneficial for a healthy brain. It establishes that the brain has a well-developed backup fuel system, which is a fact about physiology, not an endorsement of a way of eating.

Neurovascular coupling: how blood follows thought

The last piece of the story closes a loop back to how we see any of this in the first place.

Given an organ with no fuel reserve and a demand that varies from region to region moment by moment, the delivery system must be responsive. It is. When a local population of neurons becomes more active, blood flow to that specific patch of tissue increases within a second or two. This is neurovascular coupling, and it is not a passive consequence of the tissue running short: the vessels do not simply dilate because oxygen fell. They are told to dilate.

The signalling is genuinely local and involves several routes. Active neurons release messengers, including nitric oxide and various metabolites, that relax the smooth muscle of nearby arterioles. Astrocytes, whose end-feet wrap around the vessels, are important intermediaries and release vasoactive substances of their own in response to synaptic activity. Pericytes, contractile cells sitting on the capillaries, appear to participate in fine-grained control. The result is a system in which the pattern of blood flow across the brain tracks the pattern of neural activity across the brain, with a lag of a second or two and a smoothness of a few seconds.

There is a clinical dimension too. If neurovascular coupling is impaired, the brain's supply no longer matches its demand, and this appears to be an early feature of several conditions, including small vessel disease and possibly the earliest stages of some dementias. An organ with no reserve depends utterly on its delivery system being not merely intact but responsive, and the failure of responsiveness is a distinct kind of problem from the failure of the plumbing.

Myths about brain fuel

Related topics