The famine without the slowdown

In 1935 a Cornell nutritionist named Clive McCay underfed young rats and watched them outlive their well-fed cage-mates by a wide margin. It is the oldest reproducible result in the biology of ageing, and it has held up in yeast, worms, flies, mice, and, with the usual caveats, rhesus monkeys. Cut the calories – not the nutrients, just the energy – and the animal ages more slowly. Calorie restriction is the benchmark every other longevity intervention gets measured against.

For most of those ninety years there was a tidy explanation attached. A restricted animal runs cooler and slower: its resting metabolic rate drops, its body temperature drops, it makes less of the oxidative by-product that was supposed to be the wear-and-tear of living. The “rate of living” idea goes back to Max Rubner in 1908 and Raymond Pearl in the 1920s, and it is intuitive enough to survive in every gym conversation about “slowing your metabolism.” The famine works, the story went, because it turns the engine down.

Nobody wanted to actually live that way, so from the late 1990s the field went looking for a calorie-restriction mimetic: a molecule that would trip the same protective program without the hunger. Metformin was nominated. Resveratrol was nominated and mostly withdrawn. Rapamycin was the one that stuck, and it stuck for a specific reason worth holding onto: in 2009 the NIA’s Interventions Testing Program started mice on it at 600 days old – twenty months, roughly a sixty-year-old human – and still extended their lives. Late start, real effect. That is the bar.

On 2 September 2026 Danica Chen’s group at UC Berkeley published a paper in Nature with a title that says exactly what it did: late-life semaglutide treatment slows ageing and extends lifespan in female mice. The design deliberately echoes the rapamycin precedent. Healthy female mice, aged twenty months, put on injectable semaglutide – one cohort for three months of deep phenotyping, a second cohort for the rest of their lives. Alongside them ran the classic arm: mice fed 24% fewer calories. And alongside both, untreated controls.

The lifespan result is the headline and it is a real one. Median lifespan in the semaglutide arm ran about 100 days past the controls, a gain of roughly 12%. In a mouse that has already lived twenty months, that is a lot of extra mouse. Within three months of starting the drug the treated animals were more exploratory in the open field, held glucose better, ran longer on the treadmill, and remembered spatial layouts better than their untreated peers. Stem cells in the bone marrow and in neurogenic regions of the brain showed better regenerative capacity. Inflammatory markers came down. When the group read the transcriptome, several of the canonical hallmarks of ageing had moved the young direction.

The lifespan number is what the press ran with. The finding this site cares about is the comparison arm, because the comparison arm is where the biology is. Semaglutide and calorie restriction landed in the same neighbourhood on nearly every ageing readout, with the drug a notch ahead on exploration, spatial memory, and glucose handling. Chen went in asking a clean question – is a GLP-1 agonist just calorie restriction with a needle? – and the answer that came back is that it is not, and the place the two diverge is the engine.

The calorie-restricted mice did what calorie-restricted mice have always done: their metabolic rate fell. The semaglutide mice, eating less and losing weight, kept theirs. Same protective program, same lifespan dividend, and only one of the two arms turned the engine down. If the slowdown were the cause of the benefit, the arm without the slowdown should have had no benefit. It had at least as much.

That is a ninety-year-old assumption taking a direct hit. The rate-of-living story has been wobbling for two decades anyway – naked mole-rats and birds burn hot and live long, and the free-radical theory has not aged well – but here is a controlled experiment in which the metabolic slowdown is cleanly separated from the outcome it was supposed to produce. The slowdown looks like a passenger. Whatever calorie restriction is actually doing to an old animal, a peptide acting on the GLP-1 receptor can apparently do it while the mitochondria keep the lights on. Chen’s own phrasing is that the drug taps a pathway independent of calorie restriction. The more interesting reading is that both may be tapping something downstream of food intake that has nothing to do with how much fuel gets burnt.

Why would a gut hormone’s receptor be wired into ageing at all? Because it was never only a gut hormone. GLP-1 receptors sit on hypothalamic neurons, on hippocampal neurons, on microglia, on vascular endothelium, on the heart, and on the immune cells that drive the low-grade inflammation of age. The lizard the molecule came from needed a signal that could carry a whole organism through a months-long fast, not one that only talked to the pancreas.

So the mouse findings sort into things the drug plausibly does directly and things that ride along with eating less. The reduced neuroinflammation and the recovered stem-cell activity in the brain are the direct-action candidates: central GLP-1 signalling dampens microglial activation, and the same receptor is the reason the class keeps showing up in Parkinson’s and Alzheimer’s trials. The glucose and adipose improvements are the ride-along candidates, the ones calorie restriction gets too. The paper does not fully untangle which is which, and it would be dishonest to pretend it does. What it establishes is that the two lists are not the same list.

Anyone who has followed this class in humans will trip on one line: the old mice on semaglutide had better muscle function. The whole reason muscle preservation is a live research programme is that people on these drugs lose lean mass alongside fat, and the industry is racing to bolt a myostatin blocker onto the weight loss. How can the mouse and the human disagree?

They may not. The mouse readouts are endurance and grip, which are measures of what the muscle does; the human alarm is over DXA lean mass, which is a measure of how much muscle there is. Those are different things, and the field has already learned the hard way that bigger is not stronger. An old mouse whose muscle is less inflamed, better perfused, and better fuelled can perform better on less tissue. The interesting hypothesis this raises is that the lean-mass loss in people might be a size story and the ageing benefit a quality story, and the two could coexist in the same animal. That is a hypothesis, not a finding. But it is the right shape of question, and it is a more productive one than “the drug eats your muscle.”

Now the sceptical half. This is one paper from one lab, and it is a strong paper, but the things it did not do are exactly the things a lifespan claim needs.

The single-sex point is not a formality. Sex differences in mouse lifespan interventions are the rule, not the exception: rapamycin extends female lifespan more than male at most doses, several mimetics work in one sex only, and the Interventions Testing Program exists in part because single-site, single-sex results kept failing to replicate. A female-only result is a female result until someone runs the males. The NIA, which paid for the work, said so in its own release: this does not show that semaglutide slows ageing or extends lifespan in people.

The pair-feeding question is the one to watch when the full methods are picked over. Semaglutide makes mice eat less. If the calorie-restriction arm was a fixed 24% cut rather than a group matched to what the drug-treated mice actually ate, then the two arms differed in intake as well as in metabolic rate, and the clean “same restriction, different engine” reading softens a little. It does not disappear. The metabolic divergence is real either way. It just becomes a slightly less tidy story.

There is no human lifespan trial of semaglutide and there will not be one for a long time; the drug is too new and the endpoint too slow. What exists is SELECT: more than seventeen thousand adults with cardiovascular disease and obesity but no diabetes, randomised to semaglutide 2.4 mg or placebo and followed for about three years. The trial was built to measure heart attacks and strokes, and it found a 20% reduction in those. Below the primary endpoint sat a secondary one that got less attention: all-cause death fell by about 19% too, and the reduction was not confined to cardiovascular death. Non-cardiovascular deaths, including infections, came down as well.

That is not a longevity trial and it should not be read as one. A three-year mortality reduction in sick, heavy, middle-aged people is a treatment effect, not a slowing of the clock. But it is the human observation that sits closest to what the mouse paper is claiming, and it points the same direction. When the epidemiology and the mechanism agree, that is the moment a field earns the right to design the real experiment.

Settled: semaglutide started late in life extends median lifespan and improves function in female mice; calorie restriction does the same; the two arms diverge on metabolic rate; the rate-of-living explanation for calorie restriction has one more hole in it, and it is a well-placed one.

Open: males; other strains; other labs; whether the benefit holds when the restriction arm is pair-fed to the drug; which of the ageing readouts come from central receptor action and which from eating less; whether any of it translates to a human on a weekly pen for thirty years. The mouse has said something genuinely new about what calorie restriction is. It has not said anything about how long you will live.