Neuroinflammation: Why Being Sick Makes It Impossible to Think Straight

Introduction

You know the feeling before you can name it. You sit down with something you have read a hundred times and the sentences slide off. You reach for a word you have used since you were nine and it is simply not there. You reread the same paragraph four times and retain nothing. Somebody asks you an easy question and you can feel yourself taking too long.

Then you notice you are a bit hot.

Almost everybody has had this. It arrives with flu, with a heavy cold, with food poisoning, sometimes in the twelve hours after a vaccination. And almost everybody explains it the same way, roughly: my body is busy fighting something, so there is less left over for thinking. Like a laptop running a big update in the background.

That explanation is wrong in an interesting way. Not because energy has nothing to do with it. Because it has the causation backwards.

Your brain is not being starved of resources by the infection. Your immune system is actively sending it instructions, and one of those instructions is to stop doing expensive cognitive work. The fog is not what is left over after the fight. The fog is part of the fight.

This article is about how that instruction travels, what it does when it arrives, and how researchers proved it was an instruction at all rather than a side effect. It runs from a 1988 paper that nobody outside the field reads to a set of experiments where healthy volunteers were injected with a piece of bacterial cell wall and then asked to remember a list of words. It ends at a live argument about what happens when the instruction fails to switch off.

The word for the whole process is neuroinflammation. It is a badly behaved word, and one section near the end is about the researchers who want to retire it. But it points at something real, and the real thing is much stranger than brain fog.

The Experiment That Separated the Fever From the Fog

Start with the result that makes everything else make sense, because it is the one almost nobody outside the field has heard.

In 1992, Stephen Kent and colleagues published a short piece in a pharmacology journal arguing that sickness behavior should be treated as a target for drug development [1]. Buried in the argument was an experimental dissociation that ought to be famous.

If you give an animal interleukin-1, one of the immune system's main alarm signals, two things happen. It develops a fever. And it stops moving, stops eating, stops exploring and stops socialising. The obvious reading is that the second thing is caused by the first. It feels awful because it is hot, the way you feel awful in a heatwave.

Kent and colleagues blocked interleukin-1 signalling inside the brain and watched the two come apart.

The behaviour went away. The fever stayed.

Read that again, because it is the hinge of this entire subject. The lethargy and the withdrawal and the cognitive slowdown are not downstream of the fever. They are generated separately, by a separate signalling route, and they can be switched off on their own while the temperature carries on climbing.

Which means the fog is not a symptom in the ordinary sense. A symptom is what damage feels like. This is closer to a setting.

The framework built around that finding has held up for more than thirty years and is usually credited to Robert Dantzer and Keith Kelley, who spent that time working out how a signal in the blood becomes a change in behaviour [2] [3]. Their term for it is cytokine-induced sickness behavior. It is a coordinated response, not a collection of malfunctions, and the cognitive part is not an afterthought [4].

Once you see it that way, an obvious question follows. Why would a body do this on purpose?

The Man Who Said Sick Animals Were Not Broken

The answer starts four years earlier, with a veterinarian.

Benjamin Hart spent his career watching animals. In 1988 he published a review in a neuroscience journal with a claim that sounds trivial until you sit with it: the behaviour of a sick animal is not the animal failing to be well [5]. It is an organised survival strategy, as much a part of fighting the infection as the fever itself.

Look at what a sick animal actually does. It stops eating. It stops drinking as much. It stops grooming. It withdraws from the group. It becomes still, curls up, and sleeps in a different way. It loses interest in mating, in territory, in anything that involves effort with a delayed payoff.

Every item on that list looks like debility. Hart's argument was that every item on that list is doing work.

That inversion is harder to accept than it sounds, because it runs against how we normally read a body. We are trained to see reduced function as loss. A limp is loss. A cough is loss of control over the airway. Sickness behavior looks like the same category of thing and is not.

Fever is metabolically enormous. Raising and holding core temperature costs a great deal, and the calories have to come from somewhere. Stop foraging, stop grooming, stop patrolling territory, and you free up the budget. Curl up and you cut heat loss, so the same fever costs less to maintain. Stop eating and you withhold iron and other nutrients that invading bacteria need as badly as you do.

None of this is a choice. It is not the animal deciding to rest. It is a program running, and the animal experiences it as an overwhelming disinclination to do anything.

Hart's paper reframed a whole field. It also created the question this article is really about. If the immune system can reach into behaviour that precisely, how does it reach into thought?

Twenty Men and One Injection

Animal work can only take you so far. A mouse cannot tell you that words feel further away than usual. At some point somebody has to do this to a person.

In 2001, Abraham Reichenberg and colleagues did. Twenty healthy men, double-blind crossover design, each of them tested twice. On one occasion they received a low dose of endotoxin, a fragment of the outer membrane of Salmonella bacteria that the immune system reacts to violently even though it cannot cause an infection. On the other occasion they received salt water. Then they completed psychological questionnaires and neuropsychological tests at one hour, three hours and nine hours [6].

The endotoxin sessions produced worse mood, more anxiety, and measurably worse performance on memory tasks. Nobody was infected. Nobody had a virus. There was no illness to speak of beyond a few hours of feeling rough. The immune signal on its own was enough.

This became a paradigm. Since 2001 a series of laboratories have run versions of it, and the reason it matters so much is that it inverts the usual problem in this area. Ordinarily you find sick people and measure their thinking, and then spend the rest of your life arguing about what else was different about them. Here the inflammation is the only thing that changed, it changed because somebody administered it, and each person is compared against themselves.

Karen Krabbe and colleagues ran twelve healthy young men through the same design in 2005 and found the same neuropsychological signature [7]. Jan-Sebastian Grigoleit and colleagues added the piece that turns a finding into a mechanism: dose dependence. Eighteen participants at one dose, sixteen at double that dose, and the neurobehavioural effects scaled with the amount given [8].

Effects that scale with dose are not moods. They are pharmacology.

It is worth pausing on how strange that is. A substance was put into an arm. It never reached a neuron in any meaningful quantity. And within an hour the person holding that arm was worse at remembering a list of words, in proportion to how much had been given. Whatever is happening, it is happening through a signalling chain, and the chain has a gain control.

Two things have to be said plainly about this literature, because they are the sort of thing that gets quietly dropped when these studies are summarised for a general audience.

The samples are small. Twenty people. Twelve people. Sixteen people. These are demanding studies that involve putting a substance into healthy volunteers and monitoring them closely for a day, and they are not going to be run on thousands. The consistency across laboratories is what carries the weight, not any single result [9].

And the samples were overwhelmingly male. Read the cohorts above and you will notice every one of them is men. That was a deliberate design choice about hormonal variability, and it left a hole. Work published in 2024 found meaningful sex differences in the transcriptional response to an acute inflammatory challenge [10], and studies in women have since found effects the male cohorts were never positioned to see, including a reduction in empathy for another person's psychological pain [11]. The classic results are not wrong. They are just narrower than they look.

There is also a fair criticism of the whole paradigm, which is that a clean injection of bacterial cell wall is not the flu [12]. It is faster, it is more synchronised, and it lacks a live pathogen doing its own things. What makes it convincing anyway is that the picture it produces matches what turns up in people with real infections. When researchers followed patients through acute bacterial infection and into recovery, they found the same coupling of inflammatory markers to cognitive dysfunction and low mood [13].

That chart is the honest picture of the evidence base. Every bar is a real published cohort, and not one of them reaches twenty five people. Keep it in mind for the rest of this article. The findings replicate, which matters enormously. But nobody here is working with large numbers.

What Goes, and What Stays

If you have had this, you already know the deficit is oddly specific. You do not forget your own address. You do not lose your vocabulary. What you lose is the ability to do anything with them.

The research says exactly that, and the pattern is the most useful thing in this article.

A systematic review by Jessica Bollen and colleagues gathered the experimental inflammation studies in healthy people and looked at what actually moved [14]. The reliable casualties are attention, speed and the formation of new memories. What holds up is what you already know.

Consider what that dissociation means. Crystallised knowledge sits in cortical networks that are already built and require no new construction to use. Reading this sentence and holding its beginning in mind while you get to the end is a different job, done by circuits that have to actively maintain and update information moment to moment. That second job is the one that fails.

Leonie Balter and colleagues showed how selective it gets. Twenty healthy men, mean age twenty four, double-blind and placebo-controlled within subjects. Low-grade inflammation did not blanket-degrade visual attention. It hit some components and left others intact [15].

Lena Brydon and colleagues went looking for where the slowing comes from. Sixteen healthy men received a typhoid vaccination, which produces a mild inflammatory response without making anybody meaningfully ill, and were scanned. Inflammation was associated with altered activity in the substantia nigra, a dopamine-producing structure, and with psychomotor slowing [16].

Neil Harrison and colleagues used the same vaccination model twice in 2009. Once to show that inflammation shifts activity in the subgenual cingulate and changes its connectivity with reward circuitry, tracking the drop in mood [17]. Once to trace how the body's internal sensing systems generate the experience of feeling ill in the first place [18]. Earlier work had already established that the mood drop after an inflammatory challenge is mediated by the cytokine response rather than by knowing you had been injected with something [19].

Put Brydon's finding and Harrison's together and a picture forms. The mood drop and the slowing are not separate misfortunes that happen to arrive on the same afternoon. They are the same signal arriving at different addresses. One arm of it lands in circuitry that sets how much anything feels worth doing. Another lands where movement and speed get organised. You experience the combination as a general flatness, but the machinery underneath is specific.

In animals the same selectivity appears at the level of working memory specifically. Nathan Sparkman and colleagues showed in mice that interleukin-6 is a necessary participant in the disruption of working memory after an immune challenge, not merely a bystander marker [20].

Here is the table version of the pattern.

That last pair of rows is why the experience is so specific and so frustrating. Nothing has been erased. The library is intact and the librarian has gone home.

It also explains a small oddity most people notice and never mention. You can usually still do the thing you have done ten thousand times. Driving a familiar route, cooking something you know, holding a conversation about nothing in particular. It is the unfamiliar that becomes impossible, because unfamiliar means building something new, and building is the part that has been suspended.

How a Signal in Your Blood Reaches a Brain That Is Sealed Off

There is an obvious objection to everything above, and it is a good one.

The brain is protected by the blood-brain barrier, a lining of tightly joined cells that controls what crosses from the bloodstream into neural tissue [21]. It exists precisely to keep circulating molecules out. Cytokines are large proteins. So how does a signal produced by immune cells in your throat end up changing what your hippocampus does?

For years this was the strongest argument against the whole idea. The answer turned out to be that the question contained a false assumption. The barrier is not a wall with nothing on the other side. It is an interface, and it participates.

There are several routes, they operate at once, and no single one is the answer.

William Banks and colleagues showed that some cytokines cross by saturable transport. There are dedicated carrier systems that move specific molecules across, in limited quantities, in a regulated way [22]. Not a leak. A door with a keyhole. Later inventories of this interface have kept adding to the list of ways traffic moves across it [23] [24].

That detail matters more than it looks. A transport system has a capacity, and a capacity means the response saturates. It is one of the reasons the effect is graded rather than a switch, and one of the reasons a small immune signal produces a small fog rather than none at all.

Second, the barrier is not uniform. In a handful of places called circumventricular organs it is deliberately thin, because those regions exist to sample the blood. Signals reaching those areas can influence neighbouring tissue without anything having to breach anything.

Third, and this is the elegant one, the barrier makes its own signal. The endothelial cells that line brain blood vessels carry receptors for inflammatory molecules on their blood-facing side. When those receptors are engaged, the cell manufactures prostaglandin E2 and releases it into the brain side. Nothing crossed. The message was received on one face of the wall and re-broadcast from the other. Work by Ridder and colleagues showed that disabling a specific signalling molecule inside those endothelial cells prevented both fever and lethargy in mice [25]. Clifford Saper and colleagues mapped the prostaglandin circuitry that follows [26].

Fourth, and probably fastest, there is a nerve. Linda Watkins and colleagues proposed in 1995 that immune-to-brain communication runs partly along the vagus nerve rather than entirely through the bloodstream [27]. Lisa Goehler and colleagues then characterised it as a genuine sensory pathway: vagal afferents behave like chemical sensors, detecting inflammation in peripheral tissue and reporting it to the brainstem electrically [28]. Your body has a hardwired line for this, and it runs in both directions, which is why stimulating the vagus is being explored as a way to damp inflammation down [29].

Now the restraint. It is tempting to summarise all of this as "inflammation opens the blood-brain barrier", and that sentence is on a great many web pages. It is not what the evidence says for ordinary illness.

Aravinthan Varatharaj and Ian Galea reviewed what actually happens to the barrier during systemic inflammation and found the picture is one of modulation rather than breach: region-specific, graded, and mostly not a matter of the barrier failing [30]. Microglia, the brain's own immune cells, turn out to play both parts. Koichiro Haruwaka and colleagues showed in mice that they initially move to blood vessels and protect barrier integrity, and only later contribute to its breakdown if inflammation persists [31].

So the honest version is this. When you have flu, your blood-brain barrier is doing its job. It is also relaying a message, which was always part of its job.

Eight People in a Scanner

Everything so far is a mechanism inferred from behaviour, blood and animal tissue. It leaves a gap. Can you see anything actually happening inside a living human head?

In 2015, Christine Sandiego and colleagues tried. They gave healthy volunteers endotoxin at one nanogram per kilogram of body weight and imaged them with positron emission tomography using a tracer that binds to a protein which increases when microglia become activated. They measured an average increase in binding of forty six percent, plus or minus eight, across the brain regions examined [32].

Eight people.

That number deserves to sit on its own line, because forty six percent is the kind of figure that gets repeated without its denominator. It is eight participants. It is one tracer, and the protein it binds is an imperfect proxy for what microglia are doing. It is a first demonstration rather than a settled measurement.

There is a temptation, reading a number like that, to treat the imaging as the moment the argument was proved. It was not. The argument was already carried by thirty years of converging behavioural, pharmacological and lesion evidence. The scan is corroboration, and corroboration from a study of eight people is worth exactly what it is worth.

With those caveats attached, it is still the closest thing anyone has to a photograph of this process in a living person who was perfectly healthy an hour earlier. Something changed inside the skull, it changed within hours of an immune signal in the arm, and it was visible from outside.

Down at the Synapse

Now go one level lower. What is a cytokine actually doing to the cells that make memory?

The relevant molecule is interleukin-1 beta, and the relevant process is long-term potentiation, the strengthening of connections between neurons that follows repeated use. Long-term potentiation is as close as neuroscience gets to a physical correlate of learning something.

Vereker and colleagues, working with Marina Lynch's group, showed that interleukin-1 beta inhibits long-term potentiation in rats, and traced the inhibition to the activation of stress-related signalling kinases inside the neuron [33]. The cell is not damaged. Its capacity to strengthen a connection is being suppressed by an incoming instruction.

This is the point where a lot of writing about neuroinflammation goes wrong, so here is the correction, and it is the single most under-reported finding in the field.

Interleukin-1 is not the enemy of memory. Inbal Goshen and colleagues demonstrated a dual role in mice. Block interleukin-1 signalling entirely, or knock out its receptor, and hippocampal memory gets worse, not better. Raise it well above normal and memory also gets worse [34].It is an inverted U. You need a small amount of this molecule to form memories at all. Illness pushes you off the top of the curve.

That shape is worth holding onto, because it kills a very common framing. Inflammation is not an intruder in the brain. The molecules involved are already there, already doing necessary work, at concentrations the system is tuned for. What illness changes is not the cast. It is the volume.

Which means nothing foreign has been introduced. A molecule your hippocampus uses every day has been turned up past its useful range, and the process it normally supports stops working. Raz Yirmiya and Inbal Goshen's review remains the best account of how far this reaches into plasticity and learning [35]. The interleukin-1 family is central to immune regulation generally, which is why the same molecule turns up in so many places [36].

The second mechanism is about new cells. The adult hippocampus continues producing new neurons, and there is good evidence that this ongoing neurogenesis matters for forming new memories. Two independent groups published in 2003 that inflammation suppresses it. Michelle Monje and colleagues showed in rats that blocking inflammation restores hippocampal neurogenesis [37]. Christine Ekdahl and colleagues showed the same relationship from the other direction [38].Two labs, same year, opposite approaches, same conclusion. That is about as solid as this field gets.

Other cytokines do their own jobs. Interleukin-6 knockout mice show a blunted sickness response, which is how you know that molecule is load-bearing rather than incidental [39]. Tumour necrosis factor alpha given systemically produces acute cognitive dysfunction in mice on its own [40]. Immune challenge in mice reliably produces measurable cognitive impairment alongside the inflammatory markers [41], and different brain regions differ in how vulnerable they are [42]. Recent work continues to map the specific signalling that links inflammation in the body to synapse loss in the brain [43] [44].

None of that is one molecule acting alone. The signal is a chorus, the members of the chorus have overlapping jobs, and removing any single one changes the sound without silencing it. That is why you will see different papers name different cytokines as the important one. They are all partly right.

Every result in the last two paragraphs is from mice or rats. That labelling is not a formality. Julie Lasselin and colleagues compared endotoxin-induced sickness behavior in rodents and humans directly and found the overlap is real but partial [45]. Some things transfer. Some do not. Anybody who tells you a mouse result is a human fact is skipping a step.

The Cells That Carry It Out

The brain has its own resident immune cells, called microglia, and for most of the twentieth century they were assumed to be janitorial. Cleanup crew. Interesting only when something had gone badly wrong.

That view collapsed around 2011.

Rosa Paolicelli and colleagues showed that microglia physically prune synapses during normal brain development, and that this pruning is necessary for circuits to mature correctly [46]. Dorothy Schafer and colleagues showed the pruning is guided by neural activity and by complement proteins, molecules borrowed from the immune system's oldest branch, so that less active connections get tagged and removed [47]. Amanda Sierra and colleagues found microglia shaping adult hippocampal neurogenesis by clearing newborn cells that fail to survive [48].

Think about what kind of cell that makes them. Not guards standing at a perimeter waiting for trouble. Something closer to a maintenance crew that is always on site, always making small structural decisions, and already holding the tools that could do harm if the instructions changed.

Which means immune cells are not visitors to the brain. They are part of how it builds and maintains itself [49] [50].

This is why "inflammation in the brain" is a poor mental image. Picture inflammation and you probably picture a swollen ankle: hot, red, and clearly a problem. What happens in your head during flu is nothing like that. A population of cells that spends its life adjusting synapses shifts its behaviour, and the adjustment goes further than usual for a while.

The same fact explains why this can go badly wrong when it persists. Cells that sculpt connections for a living are exactly the cells you would not want stuck in an altered state. That is the difference between a week of flu and the chronic conditions where neuroinflammation genuinely does damage [51].

Why Thinking Feels Expensive

Return to the folk explanation from the introduction, the one about resources. It is wrong about the mechanism. It is not wrong about the economics.

John Kealy and colleagues found that acute inflammation shifts brain energy metabolism, in both mice and humans [52]. The acute phase response really is expensive: work in animals has measured the elevated metabolic rate that comes with fever and sickness behavior directly [53]. Energy constraint is increasingly treated as an organising principle for how bodies allocate effort under stress [54].

But the interesting finding is about what specifically changes in your decision-making, and it is not what most people would guess.

Amelia Draper and colleagues gave healthy volunteers endotoxin and had them make choices that traded off effort against reward. Sensitivity to effort changed. Sensitivity to reward did not [55].Sit with what that means for the experience. You have not stopped wanting things. The thing you want has not become less appealing. What has changed is the price tag on getting there. Reading the paper still seems worth doing. Reading the paper now costs more than you have.

That is a much better description of a sick day than "I couldn't be bothered".

Related work fills in the picture. Naomi Eisenberger and colleagues found that endotoxin reduced ventral striatum responses to reward cues [56]. Harrison and colleagues found inflammation shifts the balance of learning toward punishment and away from reward [57]. Jennifer Felger and Michael Treadway have argued that dopamine is the common thread running through the motivational changes [58].

This is also, incidentally, why sick people are difficult company and then feel guilty about it afterwards. The threshold for anything requiring effort has moved, including the effort of being pleasant. Nothing about your values changed. The accounting did.

There is also a biochemical route that pulls in the same direction. Inflammation activates an enzyme that diverts tryptophan away from serotonin production and down the kynurenine pathway instead, and blocking that enzyme prevents depressive-like behaviour after an immune challenge in mice [59] [60]. The competition for transport into the brain is specific enough that dietary amino acids can interfere with it [61]. Which is one of the several reasons this literature keeps colliding with research on depression [62] [63]. That is a large subject in its own right, and it is not this article's, but the overlap is not a coincidence: some of the same signalling underlies both. Similar arguments run through work on inflammation and pain [64], which is part of why pain and memory interfere with each other the way they do.

The Sleep You Cannot Get

Being ill wrecks sleep, and everybody knows it. What is less obvious is that this is also part of the program rather than an accident of a blocked nose.

Luca Imeri and Mark Opp laid out the case that the immune system actively restructures sleep during infection, and that the restructuring is functional rather than incidental [65]. James Krueger has spent decades arguing that some cytokines are ordinary sleep regulators that happen to rise during illness, not foreign agents disrupting sleep from outside [66].

The restructuring is specific. Deep slow-wave sleep tends to increase. Rapid eye movement sleep tends to be suppressed. Jeremy Borniger and colleagues found part of the mechanism in mice: peripheral endotoxin rapidly silences a population of neurons in the lateral hypothalamus that promote REM sleep [67]. Interleukin-1 beta and tumour necrosis factor alpha are both implicated in the sleep changes that follow [68].

There is a rough logic to which stage gets sacrificed. Deep slow-wave sleep is the most restorative and the least metabolically demanding to run. Dreaming sleep is neither. If the body is rationing, the choice is not hard to guess.

This matters for thinking, because sleep is when a great deal of memory consolidation happens. The work your brain does while you sleep is not optional for retaining what you learned. So illness hits memory twice. Encoding is impaired while you are awake, and the overnight processing that would normally rescue some of it is running a different program.

Fever is doing something similar. It is not the infection overheating you. It is a regulated setpoint change, and there is good evidence it makes immune cells work better [69]. Uncomfortable and useful at the same time, like most of this list.

The Same Fog Without a Single Germ

If the fog were caused by a virus doing something specific to neurons, it would not appear when there is no virus. It appears constantly when there is no virus.

Major surgery produces it. So does a serious physical injury. So, in a milder and briefer form, does a vaccination, which is the basis of the typhoid studies described earlier. In all of these cases there is no pathogen in the body at all. There is tissue damage or an immune provocation, an inflammatory response, and then the familiar cognitive shutdown.

This is sometimes called sterile inflammation, and it is the strongest single argument that the immune response rather than the infection is what drives the effect.

It is also the cleanest way to test the claim. If you believe a virus is doing something specific to neurons, you have to explain why a knee replacement produces the same mental state. If you believe the immune response is doing it, you do not have to explain anything. The prediction was already made.

The clinical end of this axis is delirium, the acute confusional state that turns up in hospitals, especially in older patients after surgery or during severe infection. Colm Cunningham and Alasdair MacLullich made the argument that delirium sits at the extreme end of the same spectrum as ordinary sickness behavior rather than being a categorically different thing [70]. Cunningham's group had already shown that systemic inflammation produces acute behavioural and cognitive change and can accelerate an underlying neurodegenerative process [71]. Delirium is now a well characterised clinical entity in its own right [72].

The reason the same mechanism can produce a mild fog in one person and a hospital emergency in another is largely about what state the brain was in beforehand. Hugh Perry and colleagues developed the idea of microglial priming: in a brain already carrying some pathology, the resident immune cells respond to a systemic infection far more strongly than they otherwise would [73] [74]. Clive Holmes and colleagues found in people with Alzheimer's disease that episodes of systemic inflammation were associated with faster cognitive decline [75].

Notice what that claim is and is not. It is that infection can accelerate the course of an existing disease. It is not that infection causes the disease.

The distinction is easy to lose and expensive to lose. A great deal of alarming writing about neuroinflammation runs the two together, so that a reader who felt foggy with a chest infection comes away worrying about dementia. The finding is about brains that already had a disease process running when the infection arrived. It says nothing about a brain that did not.

At the far end sits sepsis. Theodore Iwashyna and colleagues followed 1,194 patients across 1,520 hospitalisations for severe sepsis in a nationally representative prospective cohort and found substantial long-term cognitive impairment and functional disability afterwards [76]. That is a very large study by the standards of everything else in this article, and it describes the ceiling of the same axis: what happens when the immune response is overwhelming rather than proportionate [77].

The animal work at that end is dramatic and needs handling carefully. Liya Qin and colleagues showed that a single high systemic dose of endotoxin in mice produced chronic neuroinflammation and progressive neurodegeneration lasting many months [78]. That is a real and important result about mice given a large dose. It is not a description of what your cold did to you, and it should never be quoted as though it were.

The same logic connects this to two neighbouring subjects. The cognitive effects that follow anaesthesia and surgery involve the same inflammatory machinery, and so do the persistent problems after a traumatic brain injury, where the tissue damage itself sets off a sterile inflammatory response that continues long after the impact.

Two branches, one controller, two outputs. The diagram makes the Kent result visible: cut the line to the right-hand output and the left-hand one keeps running.

The Order the Evidence Arrived In

The idea that illness reshapes thought is old. The evidence that it does so through a specific signalling system is recent, and it arrived in a particular order.

Notice how much of that list is about plumbing. Two of the nine entries are somebody working out how a signal physically gets from one compartment to another. That is what the middle of a research programme usually looks like, and it is the part that never makes it into the popular version of a discovery.

What is striking about that sequence is how late the human evidence arrives. The framework was built on animals in the late eighties and early nineties, and the first proper human experiment came more than a decade later. The imaging came fourteen years after that. This is a field that spent a long time confident about a mechanism it could not yet see.

When It Does Not Switch Off

Everything so far describes a program that runs and then stops. For most illnesses, in most people, that is what happens. The fog lifts, usually within days of the infection clearing.

Sometimes it does not, and this is where the research becomes genuinely contested. It deserves careful handling, because a lot of people reading a page like this are reading it because their own fog did not lift.

The clearest recent data come from COVID-19. Felicia Ceban and colleagues ran a systematic review and meta-analysis including 81 studies. Twelve or more weeks after diagnosis, the pooled proportion reporting fatigue was 32 percent with a confidence interval of 27 to 37, and the pooled proportion showing cognitive impairment was 22 percent with a confidence interval of 17 to 28. Both estimates carried heterogeneity above 98 percent [79].

That last number is not a technicality. Heterogeneity that high means the included studies disagree with each other far more than they agree. The pooled percentage is a summary of a very noisy literature, and anybody who quotes "22 percent" without it is quoting half a result.

There is real biological evidence that something persists. Anthony Fernández-Castañeda and colleagues showed that even mild respiratory COVID, with no infection of the brain itself, produced lasting glial reactivity and disrupted myelin in mice and in human tissue [80]. Michelle Monje and Akiko Iwasaki set out the case for a neuroinflammatory account of the persistent syndrome [81]. Chris Greene and colleagues used contrast-enhanced MRI and found evidence of blood-brain barrier disruption in patients with long COVID and cognitive complaints, alongside sustained systemic inflammation [82].

And there is a serious result pulling the other way on magnitude. Adam Hampshire and colleagues invited 800,000 adults in England to complete an online cognitive assessment. 141,583 started it and 112,964 completed it. They found measurable cognitive deficits associated with prior infection, and the deficits were small [83].Both of those things can be true. A real, biologically grounded effect can also be a small effect on average, while being severe for a minority. What is not yet settled is whether persistent fog is ongoing neuroinflammation, the aftermath of neuroinflammation that has since resolved, or several different conditions being grouped under one name because they share a symptom. Similar arguments run through the literature on myalgic encephalomyelitis and chronic fatigue syndrome [84], and psychoneuroimmunology as a field spent the pandemic years arguing about exactly this [85].

This article is not going to tell you which answer is right, because nobody currently knows.

What can be said is narrower and more useful. The mechanism described in the rest of this piece is real, it is well evidenced, and it explains the ordinary case cleanly. Whether the same mechanism stays switched on in some people, and why, is an open research question with serious work on both sides and no verdict. Anyone offering you a confident answer to that is ahead of the data.

Why It Evolved, and the Argument Nobody Has Settled

Go back to Hart's claim that the behaviour is a strategy. A strategy for what, exactly?

The traditional answer is energy conservation. Fever is expensive, so the body suspends everything else to pay for it. That account fits the metabolic data well [53] [52] and it has been the standard explanation since 1988.

In 2015, Keren Shakhar and Guy Shakhar proposed a different one [86]. They pointed out that the behaviour does something else obvious: it takes an infectious individual out of circulation. Withdrawing from the group, losing interest in social contact and staying still all reduce the chance of passing the infection to relatives. Since relatives share genes, a behaviour that protects them can be selected for even at a cost to the individual running it.

They named the idea after Eyam, the English village that cut itself off during a plague outbreak in 1666 to keep the infection from spreading to neighbouring settlements.

The comparison is doing real work, not just decorating the hypothesis. Eyam's quarantine was a deliberate decision by people who understood what they were choosing. The claim is that evolution arrived at something functionally similar without anybody deciding anything, by making infected animals want to be left alone.

Neither account excludes the other, and both make predictions. The social one has gathered supporting evidence. Naomi Eisenberger and colleagues found that an inflammatory challenge induces feelings of social disconnection in humans [87]. Small inflammatory challenges are enough to shift social behaviour specifically [88]. Being inflamed changes how you read sickness cues in other people [89], and the social dimension has become a research programme in its own right [90] [91] [92].

There is also a pointed objection to reading too much into it. Jaouad Bouayed and Torsten Bohn argued that whatever the behaviour evolved for, it is plainly not sufficient to limit the spread of a modern respiratory virus, since people with mild symptoms keep moving through crowded environments regardless [93].

The honest position is that this is unresolved. The behaviour probably does several jobs at once, which is normal for anything shaped by evolution.

It is worth noticing what both accounts agree on, though, because the agreement is the part that touches you. Under either story, the shutdown is not a failure. It is expenditure, spent on something. The disagreement is only about which ledger it is charged to.

The Word Itself Is Under Dispute

Here is something the popular coverage of this topic almost never mentions. A number of the researchers who work on neuroinflammation would like the word retired.

Their objection is that it has been stretched to cover everything from a mild transient change in glial signalling to the destructive chronic pathology of a neurodegenerative disease. When one word covers both, it stops carrying information. Damon DiSabato, Ning Quan and Jonathan Godbout made this argument directly in a paper whose title says it plainly: the devil is in the details [94].

The related complaint concerns how microglia are described. Older literature sorts them into "resting" and "activated", or into M1 and M2 states borrowed from macrophage biology. Rosa Paolicelli and a large group of co-authors argued in 2022 that this vocabulary does not survive contact with modern single-cell data, and that microglia occupy many states rather than two [95]. You will still find the M1 and M2 framing everywhere, including in papers published this year.

The complaint is not pedantry. Words in science do work, and a word that covers everything does none. When a term stretches from "a few days of feeling stupid" to "cells are dying", the reader who meets it in both places has no way of knowing which one they have met.

This matters for a reader rather than only for specialists, because the ambiguity in the word is exactly what makes the internet frightening on this subject. Search the term and you will get pages about Alzheimer's disease, multiple sclerosis and chronic illness, because those are what most published work on neuroinflammation is about. The transient version you experienced last winter shares a name with them and very little else.

There is a further dispute about the mechanism of the fatigue itself. The mainstream account has cytokines acting more or less directly on the circuits that set motivation. Klaas Enno Stephan and colleagues proposed an alternative: that fatigue is the brain's own read-out of losing control over the body's internal state, a metacognitive inference rather than a direct chemical effect [96]. It is a minority position. It is also a serious one, and it makes different predictions.

A small recent result sits interestingly alongside it. Justine Schmidt and colleagues found that placebo effects still improve sickness symptoms during systemic inflammation [97]. Expectation continues to shape the experience even while the underlying immune signal is running full tilt.

What This Does Not Mean

A page like this one can be read badly, so it is worth being explicit about the limits.

The transient cognitive changes described here resolve for the overwhelming majority of people, along with the illness that produced them. That is the standard course. The research on persistent post-infectious impairment is about a minority, it is unsettled, and nothing in it says that ordinary post-flu fog is a warning sign.

The relationship between systemic inflammation and neurodegenerative disease runs in a specific direction in the evidence discussed above. Holmes and colleagues found that inflammatory episodes were associated with faster decline in people who already had Alzheimer's disease [75]. Perry's group described a brain already carrying pathology responding more strongly to a systemic challenge [74]. Neither is a finding about healthy people getting colds.

Most of the mechanistic detail comes from rodents, and the parts that have been checked in humans have been checked in small groups, mostly of young men [45] [10]. The human work continues, and the recent studies are steadily better designed [98] [99] [100], but the field is not sitting on a mountain of human data.

And the mood literature that keeps intersecting this one is its own subject with its own controversies [101] [102]. Overlapping mechanisms are not the same as identical conditions.

Conclusion

The next time it happens, you will recognise it. The sentence that will not stay in your head, the word that has gone missing, the strange difficulty of an ordinary decision. And then, some hours later, the temperature.

What is happening is not your brain failing. It is your brain receiving a message.

Immune cells somewhere in your body detected a problem and released signalling proteins. Those proteins reached your brain along several routes at once: carried across by dedicated transporters, sensed by the lining of your blood vessels which then manufactured a fresh signal on the inside, and reported electrically up the vagus nerve before your blood levels had risen much at all. Your hypothalamus received them and started a program. Your temperature went up. Interleukin-1 in your hippocampus rose past the narrow range where it helps memory work and into the range where it suppresses it. Effort started to cost more. Your sleep architecture changed shape.

None of that is damage. It is a system doing what it evolved to do, whether that was to save the calories for a fever or to keep you away from your family while you were contagious. Kent's experiment is the proof: block the signal inside the brain and the fog lifts while the fever burns on. Two outputs, one controller. They come apart cleanly, which is exactly what you would expect from a program and not at all what you would expect from damage.

There is a real version of this that does damage, and it is what most of the published literature on neuroinflammation is actually about. Sepsis. Chronic disease. A brain already carrying pathology meeting a systemic infection. Those are serious, and they are not what happened to you in February.

What happened to you in February was a set of instructions arriving, being carried out, and then being withdrawn. The executive systems in your prefrontal cortex came back online. So did everything else. The library was never on fire. The lights were turned down for a few days, deliberately, by something that was trying to keep you alive.

Given how it felt, that is a surprisingly good outcome.