The Second Lock

In one person in ten the immune system attacks its own body. The 2025 Nobel Prize went to the discovery of the cells that usually stop it. What they explain, and why the newest treatments try to reset the system rather than suppress it.

The immune system is built to make mistakes. Each T cell carries a receptor assembled by shuffling gene segments at random, and the Nobel Committee puts the number of different receptors a body could in theory make at more than 10¹⁵.[1] That is how it can recognise a virus nobody has seen before. It also means that, every day, the body makes some cells whose receptors fit its own tissues: the thyroid, the lining of the gut, the cells in the pancreas that make insulin, the sheath around nerves. Most of the time nothing happens. The question this essay is about is why not, and what goes wrong in the people for whom something does.

There are a lot of them. In 2023 Nathalie Conrad and colleagues went through the health records of 22 million people in the UK and counted 19 of the commonest autoimmune diseases, from type 1 diabetes and coeliac disease to rheumatoid arthritis, lupus, multiple sclerosis and the autoimmune diseases of the thyroid. Together they affected 10.2 per cent of the population over the study period: 13.1 per cent of women and 7.4 per cent of men.[2] Most of these diseases are treated by turning the whole immune system down, for life, and accepting the infections that come with it.

On 6 October 2025 the Nobel Prize in Physiology or Medicine went to Mary Brunkow, Fred Ramsdell and Shimon Sakaguchi “for their discoveries concerning peripheral immune tolerance”: the finding that the body has a second line of defence against its own immune system, a class of cells that holds the others back.[3] This essay tells how that was found, which of the claims now made about autoimmunity are established and which are still hopes, and what a small model, run here, can show about why a disease of this kind can behave like a switch, and why the most striking new treatment tries to flip it back.

10.2%of 22 million people in the UK had at least one of 19 common autoimmune diseases, 2000–2019: 13.1% of women, 7.4% of men

32×higher risk of multiple sclerosis after infection with Epstein–Barr virus, in a cohort of more than 10 million US soldiers

22/24patients with lupus, scleroderma or myositis met the trial’s targets six months after one infusion of engineered T cells, with no other drugs

200+clinical trials involving regulatory T cells listed by the end of 2025. No treatment based on them is yet approved

The basics

A key that fits a million locks

The immune system has two parts. The innate part acts first, within hours: cells such as neutrophils and macrophages react to features shared by whole classes of microbes, such as the walls of bacteria or the genetic material of viruses, attack what they can, and raise the alarm. The adaptive part is slower, taking days, but it is specific and it remembers. Its cells are lymphocytes, B cells and T cells, and each carries its own receptor. B cells make antibodies, free-floating copies of their receptor that bind whole molecules on a microbe’s surface. T cells see something different. Other cells cut proteins into short fragments and display them on their surface in molecules called HLA (MHC in other animals), and a T-cell receptor reads the fragment together with the HLA molecule that holds it. Killer T cells destroy infected cells. Helper T cells direct the others, and among other things they allow B cells to make better antibodies. Because HLA molecules decide which fragments a person’s T cells are shown, variants of the HLA genes are linked with many autoimmune diseases; a review marking fifty years of that research calls HLA molecules central to protective immunity and to “deleterious, disease-causing autoimmune reactivity”.[4]

The usual image for how a receptor recognises its target is a lock and a key: the receptor’s shape fits the target and nothing else. It is right that recognition depends on shape. It is misleading in three ways, and all three matter for autoimmunity.

First, one key fits many locks. The body could in theory make 10¹⁵ different T-cell receptors, but at any one time a person carries fewer than 10⁸, and these have to cover more than 10¹⁵ possible combinations of fragment and HLA. Each receptor must therefore recognise many targets. In 2012 Andrew Sewell’s group in Cardiff measured how many for one receptor, taken from a T cell of a patient with type 1 diabetes that recognises a fragment of the insulin precursor. It recognised more than a million different fragments, and one of them, which differed from the insulin fragment at 7 of its 10 positions, was more than a hundred times more potent.[5] Sewell’s conclusion was that the natural target of a receptor is almost never its best fit, and that this cross-reactivity, the price of covering every possible microbe, “provides a plausible cause for autoimmune disease”.[6]

Second, the fit is a matter of degree, not yes or no. A receptor can bind its target strongly or weakly, and, as the next section shows, the body’s first safeguard removes the strong binders to the self and lets the weak ones through. Third, the fit does not decide what happens. A T cell that meets its target also needs a second signal, which comes from cells that have sensed danger, and that is how the innate part steers the adaptive part. Meeting the target “in an environment deficient in costimulation” tends to switch the T cell off instead, a state called anergy.[7] So the locks in this essay’s title are not the receptor’s. They are the two safeguards that stop the keys which fit the body’s own tissues from doing harm.

The first lock

An examination in the thymus

By the 1980s immunologists understood the first answer. T cells mature in the thymus, a small organ behind the breastbone, and there they are tested. A young T cell whose receptor binds strongly to the body’s own proteins is told to die. This is called central tolerance, and it removes most of the dangerous cells before they ever reach the blood.[1]

There is an obvious difficulty: the thymus is not the pancreas or the thyroid, so how can it test cells against insulin or thyroid proteins? Part of the answer came in 2002, when Mark Anderson, Diane Mathis, Christophe Benoist and colleagues showed that a protein called AIRE makes cells in the thymus produce proteins that normally belong to other organs, projecting what they called “an immunological self shadow”. People with a defective form of AIRE develop autoimmune disease in many organs, and so do mice without it.[8]

But the examination is not perfect and cannot be. A cell whose receptor binds a self protein only weakly passes, and so does a cell whose target is not in the shadow, or is shown too faintly. Some self-reactive T cells always reach the body. For a long time the dominant view was that this did not much matter. The Nobel Committee describes the situation in the 1990s plainly: “many researchers were convinced that immune tolerance only developed due to potentially harmful immune cells being eliminated in the thymus”.[3]

The second lock

A guard nobody believed in

The first clue had been sitting in the literature since 1969. Yasuaki Nishizuka and Teruyo Sakakura, at the Aichi Cancer Center in Nagoya, removed the thymus from newborn mice expecting a weaker immune system. If they did it three days after birth, the mice developed autoimmune disease instead: of the ovaries, the stomach, the thyroid.[9] Taking away the organ that makes T cells made the immune system attack the body. Something leaving the thymus a little later than the first T cells, it seemed, was needed to keep them in order.

The idea of cells that suppress other immune cells was not new, and by the 1980s it had a bad name. “Suppressor T cells” had been a fashionable subject in the 1970s, and a gene region called I-J was said to encode their key features. When gene sequencing improved it turned out that the I-J region did not exist, and in the Nobel Committee’s words “the notion of suppressor T cells faced growing scepticism until the field eventually collapsed”.[9] Shimon Sakaguchi, who had joined Nishizuka’s laboratory after medical school in Kyoto, kept working on it anyway. In the early 1980s he showed that T cells from normal mice, injected into mice that had lost their thymus, protected them from the disease. Then he spent more than a decade looking for a surface marker that would let him separate the protecting cells from the rest.[1]

The paper that came out of that search, in 1995, is a model of a clear experiment. About 10 per cent of the helper (CD4) T cells in an adult mouse carry a protein called CD25, a part of the receptor for the growth factor interleukin-2. Sakaguchi’s group removed those cells and injected the rest into mice that had no T cells of their own. “All recipients spontaneously developed histologically and serologically evident autoimmune diseases”: thyroiditis, gastritis, inflammation of the insulin-making islets, of the salivary glands, the adrenals, the ovaries, the kidneys and the joints. Putting the CD25 cells back within a short time prevented all of it, in proportion to the dose. The same mice also rejected foreign skin grafts more strongly, which showed that these cells damp down immune responses in general, not only against the self.[10]

Many immunologists were still not convinced, and with reason: CD25 is also carried by ordinary T cells when they are activated, so it was not a clean label. What settled the matter came from an unexpected direction.

The scurfy mouse

The twentieth gene

In the 1940s, at Oak Ridge in Tennessee, where mice were being bred to study radiation as part of the Manhattan Project, some male pups were born with scaly, flaking skin, a huge spleen and swollen lymph nodes. They lived only a few weeks. The strain was named scurfy. Because only males fell ill and healthy mothers passed it on, the mutation had to be on the X chromosome.[1] By the 1990s it was known that the mice were being destroyed by their own T cells: the males died 16 to 25 days after birth, with “overproliferation of CD4+CD8− T lymphocytes, extensive multiorgan infiltration and elevation of numerous cytokines”.[11]

Mary Brunkow and Fred Ramsdell, at a biotech company near Seattle that was looking for drugs against autoimmune disease, decided to find the gene. Mapping narrowed it down to about 500,000 letters of DNA in the middle of the X chromosome, which held 20 candidate genes, and they compared them between healthy and scurfy mice one by one. It was the twentieth. The gene was new, a member of the forkhead family of genes that switch other genes on and off, and they called it Foxp3.[1] In scurfy mice a frameshift mutation makes the protein without its forkhead domain, the part that binds DNA.[11] They suspected that a rare and serious disease of baby boys, IPEX, which is also inherited on the X chromosome and also involves autoimmune attack on many organs, was the human version, and with samples gathered from paediatricians around the world they found mutations in the human gene. The papers came out together in Nature Genetics in January 2001.[1]

The connection was made two years later. In 2003 Shohei Hori, Takashi Nomura and Sakaguchi showed that Foxp3 “is specifically expressed in naturally arising CD4+ regulatory T cells”, and that putting the gene into ordinary T cells turned them into cells that behaved like regulatory ones.[12] Other groups soon showed the same. The cells Sakaguchi had chased for twenty years now had a defining gene, and the scurfy mice and the boys with IPEX had a diagnosis: they are what a body looks like when it has no regulatory T cells at all.

The Committee’s summary is short: the laureates “identified the immune system’s security guards”.[3] These cells work in several ways at once. They act on other T cells by direct contact, they release calming signals such as IL-10 and TGF-β, and they soak up interleukin-2, the growth factor the other T cells need to multiply.[9] They also make sure that a response winds down once an infection is over. The thymus is the first lock on the immune system; regulatory T cells are the second, and they are the reason that the self-reactive cells which get past the first one usually do no harm.

Running it here

A model small enough to read

An autoimmune disease often looks less like a defect than like a state. It may start after an infection and then go on for decades after the infection is gone. It can stop for months and flare again. Some people carry the same risk genes all their lives and never become ill. A model with two quantities and a brake is enough to show how this can happen, and what that means for treatment.

The model tracks two things: the number of self-reactive T cells, measured against their resting level, and the amount of self protein released by damaged tissue. The cells are made at a slow, steady rate and die within days. When they see their target they multiply, and the more damage they cause, the more target there is for them to see. That loop is the heart of it. The brake, standing for regulatory T cells, divides how fast the cells multiply; like the cells in Sakaguchi’s 1995 experiment, it does not care what they are reacting to. Finally an infection can arrive, for three weeks, carrying a protein that looks like the self protein, so that the self-reactive cells see more of their target without any tissue being damaged. The numbers are chosen to make the behaviour clear, not fitted to any disease; the method note gives the equations.

Fig. 1 — The same infection in four people who differ only in the strength of their brake. Self-reactive T cells, as multiples of their resting level, on a log scale, in the essay’s toy model. The grey band is a three-week infection with a look-alike protein, four times as abundant as the self protein at rest. With no brake (black) the cells rise without needing it. With a weak brake (pink) they sit at rest until the infection, then climb and stay high long after it ends. With an ordinary (teal) or a strong brake (dashed) they rise during the infection and settle back.

Figure 1 shows the switch. With no brake at all, the scurfy mouse or a boy with IPEX, the cells rise to 90 times their resting level with or without an infection: there is no healthy state to stay in. With a weak brake there are two stable states, healthy and sick. The person stays healthy for as long as nothing pushes hard enough, then a three-week infection pushes the cells past a tipping point, and from there the loop keeps itself going, at 70 times the resting level, long after the infection has been cleared. With an ordinary brake the same infection makes the cells rise to 3.9 times their resting level, and then they settle back. Nothing in the model remembers the infection. What persists is the loop.

Fig. 2 — Which combinations of brake and trigger leave a lasting disease. For each strength of the brake, the smallest three-week look-alike infection (as multiples of the self protein at rest) after which the cells are still high a year later. Shaded: lasting disease. With a brake weaker than 1.4 the disease comes with no trigger at all. Between about 5.7 and 7.2 a sick state still exists, but no three-week infection, however strong, is enough to reach it; beyond 7.2 it does not exist.

Figure 2 maps the whole space. A weak brake (2 on the model’s scale) is tipped over by a look-alike infection only 1.2 times the self protein at rest; an ordinary brake (4) needs 4.3; somewhat stronger, and no three-week infection can do it. Two things follow, and both match what is seen in patients. Risk is a combination: the same infection is harmless in one person and the start of a lifelong disease in another, depending on a brake that genes, sex and age can all shift. And the infection does not have to persist. Once the loop has started, removing the trigger does not stop it.

Who and why

Genes, sex, viruses

The UK study shows how uneven the burden is. Of the people newly diagnosed between 2000 and 2019, 63.9 per cent were women.[2] The diseases also differ in how they have moved over time, and here the numbers are more modest than the headlines that followed the study. Age and sex taken into account, new cases of any of the 19 diseases together rose by only 4 per cent between 2000–02 and 2017–19. Coeliac disease, Sjögren’s syndrome and Graves’ disease roughly doubled, while pernicious anaemia and Hashimoto’s thyroiditis fell by about a fifth (Figure 3). Some of the rises may reflect better testing and awareness rather than more disease, and a study of health records cannot tell the two apart, although for coeliac disease a 2018 review judged the rise to be real.[13] The authors note a socioeconomic gradient for several diseases, seasonal patterns for childhood type 1 diabetes and vitiligo, and differences between regions, and read them as signs of environmental causes.[2]

Fig. 3 — Not one trend but several. New diagnoses in 2017–19 compared with 2000–02, after allowing for age and sex, for the diseases named in the abstract of Conrad et al. (2023), with 95% confidence intervals, on a log scale. All 19 diseases together rose by 4%.

Why women? Sex hormones are thought to play a part, and so is having two X chromosomes, which carry many immune genes. In 2024 Howard Chang’s group at Stanford proposed a further mechanism. In female cells one X chromosome is silenced by a long RNA molecule called Xist, which gathers many proteins around itself, and several of those proteins are known targets of autoantibodies in lupus and related diseases. When the team made male mice produce Xist, the males made autoantibodies and developed worse disease in a model of lupus, and their T and B cells came to resemble females’. Patients with autoimmune diseases had antibodies against several parts of the complex.[14] The finding drew a sharp objection: the proteins in the complex are ordinary nuclear proteins present in both sexes, so it is not clear why packaging them around Xist should matter.[15] It is a promising lead, not yet an explanation.

The strongest case for a trigger is multiple sclerosis. Kjetil Bjornevik, Alberto Ascherio and colleagues used the blood samples that the US military stores from more than 10 million young adults on active duty. Of the 955 who developed MS during their service, almost all had been infected with Epstein–Barr virus, the virus of glandular fever, and infection multiplied the risk of MS by 32. Cytomegalovirus, which spreads in a similar way, did not raise it. A marker of nerve damage rose in the blood only after infection with Epstein–Barr. The authors concluded that the virus is “the leading cause of MS”.[16] It is not a sufficient cause: the great majority of adults carry the virus, and very few of them develop MS. That is what the model would predict of a trigger that tips only people with a weak enough brake.

A mechanism was published days later. William Robinson’s group at Stanford took antibodies made by expanded B cells in the spinal fluid of MS patients and found that some bound both a protein of the virus, EBNA1, and a protein of the brain’s support cells, GlialCAM, with high affinity. Mice immunised with the viral protein had worse disease in a model of MS.[17] This is molecular mimicry, the look-alike trigger of the model, seen in real antibodies.

The sharpest lesson in caution comes from a different kind of claim. In 2022 a team led by Luis Barreiro and Hendrik Poinar compared ancient DNA from people in London and Denmark who died before, during and after the Black Death, and reported that variants in immune genes changed in frequency fast enough to suggest natural selection, among them a variant of the gene ERAP2 that helped cells control the plague bacterium, and that the protective variants overlap with ones linked today to a higher risk of autoimmune disease.[18] The story was irresistible: we pay in autoimmunity for our ancestors’ survival. In 2025 Alison Barton, Iain Mathieson and colleagues, David Reich among them, re-examined the data and found that after an error in the allele frequencies was corrected none of the four highlighted variants passed the original thresholds, that the enrichment in immune genes appeared even when the labels before and after the plague were shuffled at random, because the immune genes had been captured and sequenced in different batches, and that the ERAP2 variant showed no significant change.[19] The general idea, that past epidemics selected for strong immune responses which now misfire, is still plausible. That paper is not the evidence for it.

Releasing the brake

The price of fighting cancer

The best evidence that the brakes are real comes from medicine that takes them off. Cancer immunotherapy works by blocking checkpoint molecules such as CTLA-4 and PD-1, which hold T cells back; CTLA-4 is also one of the markers of regulatory T cells. Tumours, for their part, attract regulatory T cells, which shield them from attack, and drugs that deplete those cells inside tumours are now in trials.[9]

Releasing the brake has a known price. In the trial that established the combination of the two checkpoint drugs in advanced melanoma, treatment-related side effects of grade 3 or 4, severe or life-threatening, occurred in 55.0 per cent of patients on the combination, against 16.3 per cent on the PD-1 blocker alone and 27.3 per cent on the CTLA-4 blocker alone.[20] These drugs’ characteristic side effects are inflammations of healthy organs, the colon, the liver, the skin, the hormone glands: the body’s own tissues attacked once the brakes are released. Oncologists now manage autoimmune disease as a routine part of curing cancer, which is the same lesson as the scurfy mouse, learned at the bedside.

Resetting the switch

One infusion, no drugs

If autoimmunity can be a state that maintains itself, then the usual treatment has a built-in limit. Drugs that dampen the immune system lower the activity but do not change which state the system is in, so when they stop, the disease comes back. The alternative is to knock the system so far down that the loop is broken, and let it rebuild.

That is the idea behind the most surprising result in the field. CAR-T cells are a patient’s own T cells, engineered to carry a receptor that recognises a chosen target; against CD19, a protein on B cells, they had been used for years to treat B-cell lymphoma and leukaemia. In lupus, B cells make the antibodies against DNA and nuclear proteins that cause much of the damage. In 2022 Andreas Mackensen, Georg Schett and colleagues in Erlangen gave CD19 CAR-T cells to five young patients with severe lupus that had resisted several drugs. All five were in remission three months later. B cells came back after about 110 days, and those that came back were naive, cells that had not yet met their targets, and the disease did not return with them.[21]

In 2024 the same group reported 15 patients with lupus, myositis or systemic sclerosis, followed for a median of 15 months. All met the standard measures of response for their disease, and every patient stopped all immunosuppressive treatment. B cells were absent for 112 days on average. The side effects were mostly mild: ten patients had grade 1 cytokine release syndrome, the fever and inflammation that follow CAR-T infusion, and one was admitted to hospital with pneumonia.[22] By 2025 their first 11 lupus patients had been followed for a median of two and a half years, up to four, and all were in remission without drugs; one had needed treatment for a flare.[23] The first formal trial, CASTLE, published in January 2026, treated 24 patients with a standardised CAR-T product, zorpocabtagene autoleucel, and met its endpoints: 22 reached the predefined targets at 24 weeks, nine of ten with lupus in remission, with no serious cytokine release and no neurotoxicity, and none needing steroids or other immunosuppressants.[24]

Fig. 4 — Dampening versus resetting, in the toy model. A person with a weak brake and established disease. Dashed: the cells’ growth halved for six months, as by an immunosuppressive drug. Pink: the cells killed at a high rate for 110 days, the length of B-cell absence after CD19 CAR-T cells, then left to regrow. Teal: the same depletion in a person with a strong brake. The grey bands are the depletion and, on day 400, a second look-alike infection.

Figure 4 shows why this might work, and where it might fail. Dampening halves the cells’ growth: they fall from 70 to about 41 times the resting level, the disease continues at a lower level, and when the drug stops they climb back. Deep depletion is different in kind. While the cells are absent the damaged tissue stops releasing its proteins, the loop has nothing to feed on, and the cells that grow back afterwards start from the healthy side of the tipping point. In the model the depletion has to be deep: an extra death rate of less than 0.18 per day, kept up for the same 110 days, leaves the sick state in place. That is the Erlangen group’s reasoning in plain form: their aim, in their words, is “resetting aberrant autoimmunity in these diseases through deep depletion of B cells”.[22]

The same figure shows the limit. The reset does not change the brake. When the same person meets a second look-alike infection, the disease comes back exactly as before. With a strong brake, the same infection passes. In the real patients the question is still open: one flare in eleven patients over two and a half years is encouraging, and the trials are too young to say what happens over decades.

The caveats are serious. None of these studies had a control group; the patients were the most severe cases, which makes the results striking but also leaves room for regression from an extreme. CAR-T treatment requires chemotherapy beforehand to make room for the new cells, it is expensive, and it carries a risk of infection while the B cells are gone. Newer approaches try to remove those obstacles. In October 2025 a Chinese group reported in a letter to the New England Journal of Medicine five lupus patients treated with CAR-T cells made inside the body, by lipid particles carrying the CAR’s messenger RNA into T cells, with no chemotherapy.[25] The company behind it reports complete loss of circulating B cells and falls in disease activity within three months.[26] Five patients and a short follow-up show that the method can work, not that it resets anything.

Strengthening the lock

Making more guards

The other half of Figure 4 points at the Nobel laureates’ own target: a stronger brake. The simplest way to strengthen it is to give regulatory T cells their growth factor. They depend on interleukin-2 more than other T cells do, so small doses favour them. In a double-blind trial in China, 60 patients with active lupus received low-dose interleukin-2 or placebo on top of standard treatment. The primary endpoint, the response at 12 weeks, was missed narrowly (55 versus 30 per cent responding, p = 0.052); at 24 weeks the difference was 66 versus 37 per cent, and 7 of 13 patients with lupus kidney disease reached complete remission against 2 of 12 on placebo.[27] An engineered version of interleukin-2 that acts mainly on regulatory T cells, rezpegaldesleukin, improved eczema scores by 53 to 61 per cent in a 393-patient trial in 2025, against 31 per cent on placebo, by the company’s account.[28] Eczema is an inflammatory disease, not an autoimmune one, but the trial shows that a drug aimed at regulatory T cells can work in people.

Others take regulatory T cells out, multiply them in the laboratory and put them back, or give them an engineered receptor that sends them to one organ, a transplanted kidney or liver, for instance, or a receptor for the protein that sets off coeliac disease. The Nobel Committee counted more than 200 clinical trials involving regulatory T cells at the end of 2025, and also wrote that “Treg cell therapeutics have not yet reached the clinic”.[9] One of the laureates, Ramsdell, is a scientific adviser to a company developing them.[3]

Before it starts

The years before diagnosis

Most autoimmune diseases begin long before anyone is diagnosed. The US Department of Defense keeps tens of millions of blood samples from military personnel, and in 2003 John Harley’s group tested those of 130 people who later developed lupus. In 115 of them, 88 per cent, at least one lupus autoantibody was already present before the diagnosis, up to 9.4 years earlier and 3.3 years on average, against 3.8 per cent of matched controls. The antibodies appeared in a predictable order, one kind after another, while the people still had no symptoms.[29] Type 1 diabetes is now classified by stages along the same lines: two or more autoantibodies against the islets of the pancreas are stage 1, even with normal blood sugar; abnormal blood sugar is stage 2, which carries a 75 per cent risk of diagnosis within four to five years.[30] In the language of the model, these people are near the tipping point, or past it, before the damage shows. That window is where prevention can work.

Finding them takes screening. Between 2015 and 2019 paediatricians in Bavaria offered an autoantibody test to children aged about two to five during routine check-ups. Of 90,632 children screened, 0.31 per cent had presymptomatic type 1 diabetes, and 24.9 per cent of those developed clinical diabetes within three years. Only two of them had diabetic ketoacidosis, the dangerous acid build-up of an unrecognised diabetes. Mothers’ stress rose when their child was found positive and fell back within a year.[31]

That only helps if something can then be done, and for type 1 diabetes one drug now can. In a trial of 76 relatives of patients who had such antibodies and abnormal blood sugar, a single 14-day course of teplizumab, an antibody that acts on T cells, delayed diagnosis from a median of 24.4 months to 48.4 months; diabetes was diagnosed in 43 per cent of the treated group and 72 per cent of the placebo group.[32] The US Food and Drug Administration approved it in November 2022, saying that it “does not prevent or cure T1D” but can delay the need for insulin.[30] It has since been extended to children from one year of age, and in June 2026 to children aged 8 to 17 who have just been diagnosed, to slow the loss of their own insulin; the label carries a boxed warning about reactivation of viruses, Epstein–Barr among them.[33] About two years without insulin, for a child, is a lot.

Rheumatoid arthritis has the same kind of warning: antibodies against citrullinated proteins and rheumatoid factor, often with joint pain, before any joint is visibly inflamed. The APIPPRA trial, in the UK and the Netherlands, gave 213 such people weekly injections of abatacept, a drug that blocks the second signal described above, or placebo, for a year. During the treatment year 6 per cent of the abatacept group developed arthritis against 29 per cent on placebo; after two years the figures were 25 and 37 per cent, and the improvements in pain and quality of life had not lasted.[34] The pattern is the one teplizumab shows, and the one the model predicts if the brake is left as it was: stepping in before the loop takes off delays it, and whether it prevents it for good is still open.

One trial points at prevention for everyone. VITAL gave 25,871 Americans, men over 50 and women over 55, vitamin D (2,000 units a day), omega-3 fatty acids, both or neither, for a median of 5.3 years. Confirmed autoimmune disease was diagnosed in 123 people on vitamin D and 155 on its placebo, a reduction of 22 per cent (hazard ratio 0.78, 95% CI 0.61–0.99, p = 0.05); omega-3 made no statistically significant difference.[35] It is one trial, in older people, with an effect at the edge of significance: a reason for further trials, not yet proof.

The other route is the trigger. If Epstein–Barr virus is required for multiple sclerosis, a vaccine against it should prevent most cases, and vaccines are in early trials. Coeliac disease is the case where the trigger is already known. It develops in genetically susceptible people and is set off by eating gluten, and a gluten-free diet is an effective treatment in most of them.[13] Here the essay’s model does not fit: the loop needs its trigger to keep going, and when the trigger is removed the disease settles. Whether the trigger can be made harmless in advance is less clear. Guidelines say that the age at which babies first eat gluten does not affect the risk, on the basis of several trials; in one more, the EAT study, none of 488 infants given gluten in larger amounts from four months developed coeliac disease by age three, against 7 of 516 (1.4 per cent) fed in the usual way. That was a secondary outcome with few cases, and the authors present it as a strategy to test, not to adopt.[36]

Prevention has its own costs. It means treating healthy people, some of whom would never have become ill, so the drugs have to be very safe, and teplizumab’s label carries a boxed warning. Screening gives children and their parents a diagnosis years before any symptom. And so far every drug trial has shown delay, not prevention. But the window is real, and for type 1 diabetes it is now being used.

Coda

What is known and what is hoped

The Nobel discoveries are among the best established facts in immunology: three decades of work, thousands of laboratories, two human diseases and a mouse strain that show what happens without the cells. What is built on them is at very different stages, and it is worth keeping them apart.

Established

Regulatory T cells and FOXP3 are needed to prevent lethal autoimmunity in mice and in humans. Releasing immune brakes to fight cancer causes autoimmune disease in a large share of patients. Autoimmune diseases affect about one person in ten in the UK, and women more than men.

Strong evidence

Epstein–Barr virus is required for almost all multiple sclerosis, and molecular mimicry is at least one route. Autoantibodies appear years before diagnosis in lupus, type 1 diabetes and rheumatoid arthritis. Teplizumab delays type 1 diabetes by about two years.

Striking, uncontrolled

One infusion of CD19 CAR-T cells produces drug-free remission in severe lupus and related diseases, lasting years in a small group so far. No randomised trial has reported.

Early or contested

Low-dose interleukin-2 and engineered regulatory T cells as treatments; abatacept and vitamin D to prevent disease; Xist as the reason women are affected more; the Black Death as the origin of today’s risk variants; CAR-T cells made inside the body.

The model in this essay adds one idea to the list, and it is an idea, not a finding: that an autoimmune disease can be a state that sustains itself, which a strong enough push can start and a deep enough reset can stop, without anything in the person’s genes having changed. If that is right for some of these diseases, two kinds of treatment have to be combined: one that breaks the loop and one that strengthens the brake so that the next infection does not start it again. The first is now being tested. The second is what the 2025 prize was for.

Open threads

Where this could go

The randomised trials. The pivotal CAR-T trials in lupus are the test of everything in the reset section. The question to watch is not the remission rate at six months but the flare rate after five years, and whether it depends on anything that can be measured beforehand.

Measuring the brake. The model treats regulatory strength as one number. In people it is many: the number of regulatory cells, their stability, their access to interleukin-2, the tissue they sit in. A measure that predicts who will be tipped over by an infection would turn prevention, as with teplizumab, into a general strategy.

A vaccine against Epstein–Barr virus. If the virus is required for multiple sclerosis, preventing infection should prevent most of the disease. Vaccines are in early trials. Because MS appears years after infection, proving it would take a long and large study, of the kind the military cohort made possible by accident.

On method and tools

This essay was researched and written with Claude Opus 5.5 (Anthropic): human specification, editorial direction and critical review; machine research, drafting, computation and figure generation. The question came from Luis: what has been found about autoimmune disease, the subject of the 2025 Nobel Prize, and what can be shown. The account of the discoveries follows the Nobel Committee’s press release and its popular and scientific backgrounds, which were read in full, and the abstracts of the original papers by Sakaguchi (1995), Brunkow (2001) and Hori (2003). For Conrad et al., Dendrou et al., Wooldridge et al., Sewell, Schwartz, Bjornevik et al., Lanz et al., Anderson et al., Dou et al., Klunk et al., Larkin et al., Mackensen et al., Müller et al. (2024 and 2026), He et al., Herold et al., Arbuckle et al., Ziegler et al., Cope et al., Hahn et al., Logan et al. and Lebwohl et al. the figures and quotations come from the abstracts; Barton et al. and Taubmann et al. were read in their open-access text. Andrade’s objection is reported from a summary, and the in vivo CAR-T and rezpegaldesleukin figures come from company press releases, as the text says.

The model is a toy, written for this essay and not fitted to any disease. In scripts/tolerance.py, self-reactive T cells E (in units of their resting level) and self antigen A follow dE/dt = s − (d + k)E + pE·H(A + I)/(1 + b)·(1 − E/100) and dA/dt = qE − cA, with H(x) = x²/(25 + x²), s = d = 0.1 per day, q = c = 0.2 per day and p = 1 per day. The brake b stands for regulatory T cells; I is a three-week look-alike infection; k is an extra death rate during depletion (1 per day for 110 days). The equations are solved with a fourth-order Runge–Kutta method in steps of 0.05 days; a disease counts as lasting if E is above 20 a year after the infection. The numbers are in docs/tolerance-results.json. The model has one cell type, no B cells or antibodies, no tissue specificity and no regulatory cells that grow or shrink with the response; its units are arbitrary and its values should not be compared with patients. It is there to show a mechanism. The cover is drawn from the same model.

Authored by: Luis Matos Ferreira — Physicist, Developer, Writer

Related essays on this blog

1. The Complete Parts List — what biology can enumerate, what it can predict, and how to tell a demonstrated result from an announced one.
2. The Second Reader — what AI is and is not doing in medicine.
3. The Fathers Who Stayed — another small model run in the open, on why male lineages collapsed after farming.

Sources

1. The Nobel Committee for Physiology or Medicine, “They understood how the immune system is kept in check”, popular science background to the Nobel Prize in Physiology or Medicine 2025 (text by Ann Fernholm, October 2025), nobelprize.org.
2. N. Conrad, S. Misra, J. Y. Verbakel et al., “Incidence, prevalence, and co-occurrence of autoimmune disorders over time and by age, sex, and socioeconomic status: a population-based cohort study of 22 million individuals in the UK”, The Lancet 401, 1878–1890 (2023); figures from the abstract, doi.org.
3. The Nobel Assembly at Karolinska Institutet, press release, 6 October 2025, nobelprize.org.
4. C. A. Dendrou, J. Petersen, J. Rossjohn & L. Fugger, “HLA variation and disease”, Nature Reviews Immunology 18, 325–339 (2018), doi.org.
5. L. Wooldridge, J. Ekeruche-Makinde, H. A. van den Berg et al., “A single autoimmune T cell receptor recognizes more than a million different peptides”, Journal of Biological Chemistry 287, 1168–1177 (2012), doi.org.
6. A. K. Sewell, “Why must T cells be cross-reactive?”, Nature Reviews Immunology 12, 669–677 (2012), doi.org.
7. R. H. Schwartz, “T cell anergy”, Annual Review of Immunology 21, 305–334 (2003), doi.org.
8. M. S. Anderson, E. S. Venanzi, L. Klein et al., “Projection of an immunological self shadow within the thymus by the Aire protein”, Science 298, 1395–1401 (2002), doi.org.
9. The Nobel Committee for Physiology or Medicine, “Scientific background to the Nobel Prize in Physiology or Medicine 2025”, for the 1969 thymectomy experiments of Nishizuka and Sakakura, the history of suppressor T cells, the mechanisms of regulatory T cells and the state of clinical development, nobelprize.org.
10. S. Sakaguchi, N. Sakaguchi, M. Asano, M. Itoh & M. Toda, “Immunologic self-tolerance maintained by activated T cells expressing IL-2 receptor alpha-chains (CD25). Breakdown of a single mechanism of self-tolerance causes various autoimmune diseases”, Journal of Immunology 155, 1151–1164 (1995), pubmed.ncbi.nlm.nih.gov.
11. M. E. Brunkow, E. W. Jeffery, K. A. Hjerrild et al., “Disruption of a new forkhead/winged-helix protein, scurfin, results in the fatal lymphoproliferative disorder of the scurfy mouse”, Nature Genetics 27, 68–73 (2001), doi.org.
12. S. Hori, T. Nomura & S. Sakaguchi, “Control of regulatory T cell development by the transcription factor Foxp3”, Science 299, 1057–1061 (2003), doi.org.
13. B. Lebwohl, D. S. Sanders & P. H. R. Green, “Coeliac disease”, The Lancet 391, 70–81 (2018), doi.org.
14. D. R. Dou, Y. Zhao, J. A. Belk et al., “Xist ribonucleoproteins promote female sex-biased autoimmunity”, Cell 187, 733–749 (2024), doi.org.
15. F. Andrade, “Opinion: How does XIST promote sex bias in autoimmune diseases?”, Frontiers in Immunology 15, 1399408 (2024), doi.org.
16. K. Bjornevik, M. Cortese, B. C. Healy et al., “Longitudinal analysis reveals high prevalence of Epstein-Barr virus associated with multiple sclerosis”, Science 375, 296–301 (2022), doi.org.
17. T. V. Lanz, R. C. Brewer, P. P. Ho et al., “Clonally expanded B cells in multiple sclerosis bind EBV EBNA1 and GlialCAM”, Nature 603, 321–327 (2022), doi.org.
18. J. Klunk, T. P. Vilgalys, C. E. Demeure et al., “Evolution of immune genes is associated with the Black Death”, Nature 611, 312–319 (2022), doi.org.
19. A. R. Barton, C. G. Santander, P. Skoglund, I. Moltke, D. Reich & I. Mathieson, “Insufficient evidence for natural selection associated with the Black Death”, Nature 638, E19–E22 (2025), doi.org.
20. J. Larkin, V. Chiarion-Sileni, R. Gonzalez et al., “Combined nivolumab and ipilimumab or monotherapy in untreated melanoma”, New England Journal of Medicine 373, 23–34 (2015), doi.org.
21. A. Mackensen, F. Müller, D. Mougiakakos et al., “Anti-CD19 CAR T cell therapy for refractory systemic lupus erythematosus”, Nature Medicine 28, 2124–2132 (2022), doi.org.
22. F. Müller, J. Taubmann, L. Bucci et al., “CD19 CAR T-cell therapy in autoimmune disease — a case series with follow-up”, New England Journal of Medicine 390, 687–700 (2024), doi.org.
23. J. Taubmann et al., long-term follow-up of the efficacy and safety of CD19 CAR T-cell therapy in systemic lupus erythematosus (in German), Zeitschrift für Rheumatologie 84, 612–620 (2025); figures from the abstract, doi.org.
24. F. Müller, M. Hagen, A. Wirsching et al., “CD19 CAR-T cells for treatment-refractory autoimmune diseases: the phase 1/2 CASTLE basket trial”, Nature Medicine 32, 1142–1151 (2026), doi.org.
25. Q. Wang, Z. X. Xiao, X. Zheng et al., “In vivo CD19 CAR T-cell therapy for refractory systemic lupus erythematosus”, New England Journal of Medicine 393, 1542–1544 (2025), correspondence, doi.org.
26. MagicRNA, press release on the publication of its in vivo CAR-T data in lupus (2025); a company statement, not independently checked, biospace.com.
27. J. He, R. Zhang, M. Shao et al., “Efficacy and safety of low-dose IL-2 in the treatment of systemic lupus erythematosus: a randomised, double-blind, placebo-controlled trial”, Annals of the Rheumatic Diseases 79, 141–149 (2020), doi.org.
28. Nektar Therapeutics, “REZOLVE-AD phase 2b study of rezpegaldesleukin meets primary and key secondary endpoints in patients with moderate to severe atopic dermatitis”, press release, June 2025, biospace.com.
29. M. R. Arbuckle, M. T. McClain, M. V. Rubertone et al., “Development of autoantibodies before the clinical onset of systemic lupus erythematosus”, New England Journal of Medicine 349, 1526–1533 (2003), doi.org.
30. US Food and Drug Administration, “FDA approves first drug that can delay onset of type 1 diabetes” (approval of 17 November 2022, FDA Update of 1 January 2023), fda.gov.
31. A.-G. Ziegler, K. Kick, E. Bonifacio et al., “Yield of a public health screening of children for islet autoantibodies in Bavaria, Germany”, JAMA 323, 339–351 (2020), doi.org.
32. K. C. Herold, B. N. Bundy, S. A. Long et al., “An anti-CD3 antibody, teplizumab, in relatives at risk for type 1 diabetes”, New England Journal of Medicine 381, 603–613 (2019), doi.org.
33. US Food and Drug Administration, “FDA approves new indication for Tzield (teplizumab) for certain pediatric patients with recently diagnosed stage 3 type 1 diabetes”, 15 June 2026, fda.gov.
34. A. P. Cope, M. Jasenecova, J. C. Vasconcelos et al., “Abatacept in individuals at high risk of rheumatoid arthritis (APIPPRA): a randomised, double-blind, multicentre, parallel, placebo-controlled, phase 2b clinical trial”, The Lancet 403, 838–849 (2024), doi.org.
35. J. Hahn, N. R. Cook, E. K. Alexander et al., “Vitamin D and marine omega 3 fatty acid supplementation and incident autoimmune disease: VITAL randomized controlled trial”, BMJ 376, e066452 (2022), doi.org.
36. K. Logan, M. R. Perkin, T. Marrs et al., “Early gluten introduction and celiac disease in the EAT study: a prespecified analysis of the EAT randomized clinical trial”, JAMA Pediatrics 174, 1041–1047 (2020), doi.org.