A discovery about the lung
A real experiment in mice, published in Nature in 2026 · about fifteen minutes to walk through · who did it
By the day of the flu their lungs looked recovered — no difference could be detected in how they breathed, or in the tissue, apart from one kind of immune cell still settled in it. What put it there had happened a month earlier, and had been over for weeks.
Then both were given the same flu
Two lungs, drawn alike. Both are schematic. On the measures taken before the flu — breathing, and the look of the tissue — no difference was detected between the two groups; they were alike in what was measured, not randomly assigned. Before the flu, two differences were found: the cell at step 1, and, at step 6, what the macrophages had left ready to read.
People differ enormously in how badly a lung infection hits them, and often it is the response doing the harm, not the microbe itself. Which makes an old question worth asking again: could an infection you had long ago change how you survive a completely different one later? Early in the COVID-19 pandemic it was asked again, when reports suggested that regions where worm infections are common were seeing milder disease.
This team had a narrower reason to look. They already knew that this worm, on its way through the lungs, makes one particular kind of immune cell in the lungs multiply — and that the numbers settle back to normal afterwards. So the question became small enough to test: once everything looks normal again, do those cells still carry anything?
This is the story of what that one thing left behind. Five things to know first, then six steps through the discovery, then what it might mean for us.
Groundwork
The terms you need before the steps. None of them takes longer than a sentence or two, and you can come back to them at any point.
Air goes down a windpipe that branches like a tree into smaller and smaller airways, ending in millions of tiny air sacs called alveoli. The sacs are where oxygen crosses into your blood. They are delicate, and there is a lot of surface to protect.
Too small to see, and not really alive on its own. A flu virus gets inside your cells and makes them build copies of it. The cells are wrecked in the process, which is part of why you feel terrible.
A parasite — an animal that lives in another animal. The one in this story travels through the lungs on its way to the gut. The immune system attacks it and expels it. Within a couple of weeks it is gone.
A large immune cell that engulfs and digests whatever does not belong: debris, dead cells, invaders. Many settle into a single organ and stay there for months, maintaining the tissue. This study is about one such group, the ones that sit along the lung's airways and the nerves beside them.
Your defences damage your own tissue while they work. Usually that is a fair trade. In a severe lung infection it can be most of the problem — you can be harmed more by the response than by the infection itself.
These five are standard biology, not findings from the paper. Everything in the six steps that follow comes from the paper and says where; the pictures are drawn for this page from its figures.
The discovery
Six steps, in the order the researchers took them. Each has a picture you can play with and an explanation you can read three ways — for kids, for everyone, or for scientists. Pick one and it stays picked.
Two words, used throughout: a trained mouse or lung is one that was infected with the worm a month earlier, a trained cell is one taken from such a mouse, and an untrained one is any that was not — those two words, and no others, for the two groups.
Step 1 of 6
Thirty-three days, to scale. Drag along it — most of what matters here is the long stretch where nothing shows.
A timeline of thirty-three days with two lanes of mice: the trained lane above, the untrained lane below. In the trained lane a worm travels through the lungs for the first few days and is gone by about day 14. Nineteen days follow in which nothing visible happens in either lane. On day 33 both lanes receive the same dose of the same flu, given into the nose. The drawing labels the dose in the laboratory's own units.
The days run to scale. Within a single day, the order events are drawn in carries no meaning. The worm's path through the lungs is an illustration of the route it takes. That last row counts the ones settled into the lung tissue; the lung's overall count of that cell was back to normal. It is the difference the researchers reported at this point; steps 4 and 6 find others. It is not the macrophage from the groundwork — that cell is named at step 4.
Some of the mice were infected with a worm. It travelled through their lungs, and then it was gone. A month later, when those mice looked well again — though one kind of immune cell had settled into their lungs in about twice the usual numbers — every mouse in the study was given a dangerous flu.
Two groups of mice. One group had been infected with a lung-travelling worm about a month earlier; the other had not. By the time of the flu the worm was long gone and those lungs had recovered — the researchers measured breathing and examined the tissue and could find no difference from lungs that had never been infected. That is not the same as there being none: one kind of immune cell had settled into the lung tissue in roughly twice the usual numbers, even though the lung's overall count of that cell had gone back to normal.
Then both groups got the same dose of a flu that is normally lethal to mice. No other difference was detected in what the researchers measured. That is the design: make the two groups as alike as you can, so that whatever happens next can be traced to the one thing you changed. The animals were not randomly assigned and the study was not blinded, so the worm is the likeliest explanation of what follows rather than a proven one.
~600 N. brasiliensis L3 larvae subcutaneously, 4–5 weeks' rest, then intranasal H1N1 PR8. At day 35, in mice given no influenza, no difference was detected between trained mice and naive — never worm-infected — animals by lung histology (ED 1c,d) and across five whole-body plethysmography measures (Fig. 1f–j, n = 8, all NS).
Not identical, though. Total eosinophil frequency had returned to baseline, but trained lungs held roughly twofold more CCR3+ eosinophils (ED 6c), ~95% of them tissue-resident rather than circulating (ED 6d). Those cells matter at step 5. "Recovered" here means no detected difference in what was measured — not a return to a naive state.
Step 2 of 6
Then the flu, and ten days of watching. Drag through them and watch the two numbers come apart.
Two charts over ten days. Survival: the trained group stays at one hundred per cent throughout; the untrained group falls to half on day seven and to zero on day eight. Body weight: the untrained group falls past twenty per cent lost by day seven; the trained group dips to about seven per cent lost and then levels off without returning to its starting weight.
Where the paper disagrees with itself. The Results text says the trained mice “rapidly returned to their baseline weight by 8 days”; the published curve flattens near 93%. The abstract says the untrained mice “succumbed by days 5–6”; the published survival curve holds at 100% through day 6, falls to half on day 7 and reaches zero on day 8. Both pictures here are drawn from the figures.
Every mouse that had been infected with the worm lived. Every mouse that had not, died — on the seventh and eighth days.
All of the trained mice survived. All of the untrained mice died within about a week.
The trained mice lost weight too — but far less of it, and then they stopped losing and held steady while the untrained mice kept dropping.
100% survival in N. brasiliensis-conditioned mice; all naive controls reached the clinical end point — the abstract says days 5–6, the Fig. 1c curve leaves 100% on day 7 and reaches zero on day 8. Controls lost >20% of body weight within the first week. Survival by Kaplan–Meier, P = 0.0054; weight by Mann–Whitney, P = 0.03.
A second departure from the text. Results states that trained mice "rapidly returned to their baseline weight by 8 days". Read off Fig. 1b, the trained mean plateaus near 93% of starting weight from about day 6 through day 9 and does not reach 100%.
Step 3 of 6 — the centre of it
You would expect the mice that lived to have beaten the virus. Each colour below shows one thing in the tissue, the way a microscope filter does. Switch them off one at a time and see whether they did.
Nobody could measure a difference in virus between the two lungs, on any of the three days it was counted. On those days, clearing the virus is not what the worm did for these mice. What changed was what the infection did to them.
This is the damage: torn, collapsed tissue across much of the field.
And these crowd it: neutrophils, the fastest and roughest of the immune cells. They spill sticky webs called NETs, made of their own DNA. The webs are meant to trap microbes, and are thought to damage the tissue around them as well; here they were seen where the damage was worst.
Both fields are illustrations, built from the paper's findings. What the researchers established was a failure to detect a difference in virus, which is a weaker statement than saying the two were equal — so the particle counts here differ slightly. The air sacs behind both lungs are identical, so nothing in them carries meaning.
Nobody could find a difference in how much flu was in the lungs that lived and the lungs that died. So beating the virus is not how they survived — not on any of the days anyone counted it. They just didn't hurt themselves as badly while they fought it.
Switch to the virus on its own and the two lungs look alike. The researchers measured the amount of live virus on three separate days and could not detect a difference between the two groups. On those days, clearing the virus is not what the worm did for these mice.
Switch to damage and the two lungs stop looking alike at all. One is torn up; the damage in the other is slight. Switch to neutrophils and you can see what crowds the damaged lung — the immune system's fastest responders, spilling webs of DNA that trap microbes and are thought to damage the tissue around them.
That is the whole idea the paper turns on. There are two ways to survive an infection: clear the microbe, or stay well despite it. The virus was counted on three days and did not differ. Early clearance and antibodies were never measured. The paper reads this as the second one and calls it tolerance.
Which leaves one obvious question hanging. If it wasn't the virus that was different, what was?
Infectious viral titres showed no differences at 1, 3 or 6 d.p.i. (Fig. 1e). Protection tracked instead with reduced immunopathology and amplified type 2 cues. Control lungs showed severe acute lung injury — pneumonia, vascular damage, haemorrhage (Fig. 1d) — with marked neutrophilia (Fig. 2k, counted) and prominent airway NET signal (Fig. 2q, imaged, not quantified).
No increase was detected in lymphocyte numbers or CD8 activation, arguing against a major adaptive contribution; antibody responses were not measured.
Step 4 of 6
So they measured a great deal of what the lung was doing, and here it is, sorted by which way it went. Several of these are signals: small molecules one cell releases to tell the cells around it what to do. The middle column — where no difference could be detected — is the one to look at.
Choose any measurement to see what it is.
Position gives the direction of a change: anything in the right-hand column went up. Position is the only measurement here, and a size appears only where one was taken. Dead-cell clearing is missing here because it never was.
In the lung that lived, the mess stayed small and the lung could mend. In the other lung, a crowd of neutrophils — the immune system's fastest and roughest responders — rushed in, and that lung was wrecked.
Both lungs held about the same number of macrophages — bigger, longer-lived cells that clear debris and help tissue mend. Same number in both. What differed was how they behaved.
The tidy version of this would be that the trained lungs were simply calmer. They were not. They had far fewer of the swarming cells that can wreck tissue, and more of the signals that go with repair — but one of the signals that went up is a pro-inflammatory one, and in two more — one inflammatory, one antiviral — no difference could be detected at all.
So the honest summary is not "less inflammation". It is a different mix: far fewer of the cells that crowd a damaged lung, and more of the signals that go with repair.
The number of the macrophages themselves barely changed. Whatever the worm left behind, it changed what those cells are set up to do, not how many there are.
Trained lungs at 72 h: significantly higher IL-4, IL-5, IL-13, IL-3, IL-9, plus IL-10, eotaxin-1 and MIP-1α. No differences in IL-1β or IFN-α. Marked neutrophilia and abundant airway NET signal in naive mice; eosinophils markedly increased in trained lungs. Nerve- and airway-associated macrophage (NAM) abundance comparable between cohorts. Threefold increase in acetylated α-tubulin; proSP-C significantly higher; KRT5+ cells described as emerging, not quantified.
Not listed above: efferocytosis. The abstract names it, and the chromatin and expression data support a programme — but clearance itself was never assayed, so it is not listed as a measured difference.
Step 5 of 6
How do you show it was the macrophages? You take them away, and you put them somewhere else. Watch the argument build.
One branch is a control. It is there to rule out an objection to the branch above it. The fourth branch has its dose written on it, because that test used a milder flu. What it showed was less illness. Survival against a lethal dose was never tested there. The replacement cells arrive with no origin drawn, because the paper reports them as local and/or bone-marrow-derived and does not separate the two.
How did they know it was the macrophages? They took the macrophages away and let new, untrained ones move in — and the mice died. When they did the same swap in a mouse that never had the worm, nothing changed that they could measure, so it was not the swapping that killed them. Then they moved trained macrophages into an untrained mouse, and that mouse got less sick, from a gentler flu.
Then they found something else. The macrophages had a partner cell: the eosinophil, the same cell that had been sitting in those lungs since the worm. When the scientists took most of the eosinophils away, the protection stopped working — though other things changed too, so it was not a clean test of just one cell.
Everything up to here is a pattern: the trained mice live, and their macrophages are different afterwards. A pattern is not proof. To test it you have to break it.
So the researchers took the trained macrophages out of a trained mouse and let new, untrained ones take their place. Those mice died. The protection depended on the cells that were taken away. It did not live only in them — a later branch shows the eosinophils beside them mattered too. No difference in virus was detected in those mice on the one day it was measured, so the researchers read the deaths as worse damage rather than a bigger infection.
Two checks sit under that one. The same swap in a mouse that had never been infected showed no detectable effect, so the swap itself is not what kills. And when a mouse whose macrophages had been replaced was given a worm of its own, the new cells could be trained from scratch — and every one of those mice lived.
Then the same question from the other side: move trained macrophages into an untrained mouse. That mouse got less sick. The protection travels with the cells — though this test used a gentler flu, so what it showed was less illness, not survival.
There is one more experiment, and it changes what the answer means. The macrophages had help. A second kind of immune cell gathers right beside them — the eosinophils from step 4 — and blocking those takes the protection away as well.
So this is not one cell doing the work. The eosinophil test shifted the neutrophils too, so it cannot say how the labour divides. The authors' own word is that the two cooperate. The trained macrophages also read the gene for eotaxin-1 more often — a gene being one instruction out of the set a cell carries — and eotaxin-1 is the signal that calls eosinophils in. That is the researchers' suggested link rather than a proven one. And blocking the eosinophils changed other things too — neutrophils roughly doubled — so it is not a clean test of a single cell.
NAM-DTR depletion with replacement by unconditioned NAMs abrogated survival, with no difference in viral titres detected at 3 d.p.i. (Fig. 3b,c,i). Niche repopulation in naive mice showed no detectable effect — a repopulation control (Fig. 3k–m). Repopulated NAMs could be trained de novo and restored full protection, including complete survival (Fig. 3n–p). Adoptive transfer reduced weight loss at a sublethal inoculum (Fig. 3q,r).
Anti-CCR3 removed about 70% of eosinophils and the survival benefit with them; no change in NAM abundance was detected but neutrophils roughly doubled, so this is not a clean single-population depletion. The authors' conclusion is that the two populations cooperate. Trained NAMs transcribe more Ccl11; secretion was not measured, and eosinophils are themselves the dominant source of IL-5 here, so the signalling is not one-way.
Step 6 of 6
The worm left a month ago. So why were the macrophages still different? Switch between the two cells and watch what changed.
| Gene | What it is for | Can it be reached? | Is it being read? |
|---|---|---|---|
| Pparg | A master switch for repair programmes | More open | Read more |
| Arg1 | Makes ARG1, a protein that calms tissue down | More open | Read more |
| Stat6 | Part of the pathway the authors name | More open | Read less |
| Ccl11 | Makes eotaxin-1, the signal from step 4 | More open | Read more |
A schematic stretch of DNA. The genes are spaced for legibility; on a real chromosome they sit far apart. Stat6 earns its place on the list: its stretch is more reachable while its gene is read less. Naming a pathway does not mean every part of it is turned up.
The worm was long gone. So why were the macrophages still different? Because of something they kept. A cell keeps its instructions on DNA, a molecule so long it has to be wound up tight to fit inside — and a stretch that is wound tight cannot be read. The worm left some stretches unwound and within reach. Most of them were not being used yet. They were just reachable.
This is the heart of it. The trained cells were not running the repair job at full tilt while the lung was healthy — though a few of those instructions were already being read more than usual. What had mostly changed was which of their instructions were within reach: stretches that sat packed away in untrained cells were open in trained ones — 24,456 of them.
A few of those instructions were already being read more often than in an untrained cell — Arg1 among them. But the big difference was in reachability. Being ready is not the same as being busy.
One way to picture a month-old memory inside a cell: a door left unlocked. What was measured is which stretches are open, not what the cell does with them.
The researchers use the word priming, though for one pathway rather than the whole cell — and it raises a last question worth asking. If what the worm left behind was a door left unlocked, does it have to be a worm that unlocks it?
Bulk RNA-seq and ATAC-seq on NAMs sorted at day 35, again without influenza, separated trained from naive in both transcriptional and accessibility space, with 24,456 regions of increased accessibility against 310 decreased. Most leading-edge genes were associated with accessibility changes rather than steady-state expression.
The authors' phrase "durable epigenetic priming without active pathway engagement" applies to the pro-resolving lipid-mediator pathway specifically, not to the cell as a whole: at rest, trained NAMs upregulated Rxra, Pparg, Arg1, Chil3, Fabp5 and Angptl, while Stat6 and Egr2 were reduced — Stat6 despite increased accessibility at its locus. Motif enrichment implicated a PPARγ-centred architecture.
Why it matters
Nobody is proposing worms as medicine. But that question has a version you can actually test: was it the worm that mattered, or the signals it set off? They tried it both ways.
The experiment the six steps were about.
IL-4, then IL-5 a week later, then three weeks of nothing, then the flu.
The same question from the other side.
Each row is one experiment and its outcome, nothing else. Row 2 is the hopeful one and also the weakest — the line beside it says why.
Maybe you don't need the worm at all — only the signals a body sends while fighting one. Mice given two of those signals, and no worm at all, lost less weight in the first days of a flu weeks later. Nobody checked whether they lived.
Mice given two of the signals the worm provokes — IL-4, then IL-5 a week later, with no worm anywhere — lost less weight early in a flu given three weeks after that. And when those same two signals were blocked while a real worm was doing its training, the later flu went worse.
Together those two results are why the researchers think the signals are what set the protection up — though the hopeful one was measured as early weight loss, with no survival reported. Two injections are not known to do what the worm did.
Recombinant IL-4 complexes followed by rIL-5 seven days later, then a three week rest, protected naive mice against early influenza-induced weight loss (Fig. 5f,g). No survival panel was reported for that cohort. Conversely, dual IL-4/IL-5 blockade during the training window significantly increased later influenza-associated weight loss (Fig. 5d,e), so the blockade does not establish either cytokine as individually necessary.
A note on Fig. 5g. The keys printed inside that panel name the adoptive-transfer groups rather than the cytokine groups its legend describes. The claim here rests on the Results sentence, and that curve is not reproduced.
What about people?
Nobody knows yet. The COVID-era reports that started this question were geographical patterns, and nothing in this paper tests them. What the paper can show is that cells answering to the same description exist in human lungs, and that in one disease they are running something that looks like the trained programme.
No new human experiments. They went back to single-cell atlases that other researchers had already published — lungs from healthy people, from people with COPD (a long-term disease that narrows the airways), and from people with IPF, a disease in which the lung scars — and looked for the trained-mouse signature in them.
In IPF, the cells scored high on that signature. In COPD, they scored in the other direction — and the paper describes those cells as running a different, more inflammatory programme rather than a mirror image of the trained one.
People have cells like these too. But nobody has given anyone a worm and then the flu to find out what happens — and nobody should. This part is a clue, not an answer.
This is a hint, and a small one. No human was given anything: cells in the lungs of people with a scarring disease simply look, in their gene activity, something like the trained mouse cells.
That cuts both ways. What follows is a reading of the pattern, not something the paper tested: a repair programme that saves a lung after a sudden injury might be the same programme that scars it when it runs on and on. The thing that helps in a crisis is not automatically good to have switched on forever.
Meta-analysis of published human lung single-cell atlases across healthy, COPD and IPF cohorts. The trained NAM signature was positively enriched in IPF and negatively enriched in COPD, with COPD NAMs described as a distinct, more inflammatory and migratory state. Nothing here concerns human influenza, and no causal claim is available from an enrichment score.
Nothing on this page is a reason to seek out a worm infection, and nothing here is medical advice. Worm infections cause real and serious disease in millions of people. This is a study in mice about why a lung sometimes survives its own defences — not a treatment, and not a suggestion.
The edges
Whether any of this happens in people is not the only open question here. A good paper leaves a clear edge around what it showed, and this is the rest of that edge — in the authors' terms and ours.
Arg1 is one of the genes the trained cells read more often at step 6. Removing the protein it makes — ARG1, which helps calm inflamed tissue — from these macrophages lost only part of the protection: about half the mice died, against none of the trained and all of the untrained. And the mice built for that experiment lose ARG1 from more kinds of cell than the ones this story is about.
Whether IL-4 and IL-5 work on the macrophages themselves, or reach them through something else, is unresolved. The paper says so itself.
Experiments in this paper use between three and ten mice per group, which is ordinary for the field and still a small number. A result from eight animals carries less weight than the same result from eighty, and none of the numbers on this page should be read as a population figure.
Male mice only, 7 to 16 weeks old, one genetic background, one laboratory influenza strain. Nothing here has been shown in females, in other ages or strains, in other infections, or in people.
The instructions for it sit ready in the trained cells, and the paper names it, but no one counted cells being cleared, so whether the trained macrophages actually cleared more is unknown.
The authors disclose in the paper's reporting summary that the animals were not randomised to groups and that the studies were not blinded. They are the kind of thing reporting standards ask authors to declare, and the authors declared them.
That bounds how far the result reaches. A month after an infection that was over, in a lung that looked recovered, something was still different — and taking it away cost these mice their lives.
That is the calm this page is named for. Not a quiet immune system: the trained lungs were busy, and some of what they made was inflammatory. A calm lung is one that comes through its own defence still able to breathe.