Cellular Reprogramming: Rewinding a Cell Without Erasing It

Ageing looks like damage: oxidised proteins, worn telomeres, mutations that accumulate and cannot be un-accumulated. Some of it is exactly that, and that part is hard to imagine undoing. But part of a cell’s age sits somewhere else — in a layer of chemical marks laid over the genes, deciding which ones the cell reads. That layer was built to be rewritten. Four proteins can rewrite it, and when they do, the tests we use to read a cell’s age fall to zero. Partial reprogramming is the attempt to run that rewrite part of the way, then stop before the cell forgets what it is.

Last in a series on the biology of ageing: gene editing, telomeres, epigenetics, the Blue Zones, and cellular reprogramming.

Four proteins, and a grown cell goes back

The experiment is from 2006. Takahashi and Yamanaka took mouse fibroblasts — ordinary skin cells, fully committed to being skin — and forced them to make four proteins: Oct3/4, Sox2, c-Myc and Klf4. The field calls the set OSKM, after the initials.

All four are transcription factors: proteins that bind DNA and switch other genes on or off. Here they are the cell’s own controls, pushed to levels an adult skin cell never sees. Under that push the cells reverted to an embryonic-like state, called pluripotent: able to give rise to any tissue in the body.

Cells made this way are induced pluripotent stem cells, iPS cells for short. They are routine reagents now, grown in laboratories everywhere. As established as biology gets.

The proof that it worked is a tumour

That first paper demonstrated pluripotency the way the field still does. It injected the reprogrammed cells into mice and waited. The mice grew tumours, and inside them was tissue from all three germ layers — the three sheets an early embryo divides into, from which every organ comes.

A growth like that is called a teratoma, and it is not the experiment going wrong. It is the experiment’s proof: only a cell with the full range of an embryonic cell can build all three germ layers at once.

So the standard assay for complete reprogramming — the accepted evidence that a cell went all the way back — is that it grows a tumour containing several kinds of tissue. Everything about partial reprogramming follows from that.

The clock does not slow down. It resets.

The second finding comes from epigenetic clocks.

An epigenetic clock is not a clock in any mechanical sense. It is a regression model: a formula fitted to data. Take a few hundred particular positions on the genome — CpG sites, where a C sits next to a G along the strand, and where the cell tags DNA with methyl groups. Measure how heavily each is tagged, weight the measurements, add them up, and the output is a number in years. The layer being read is the one this series covered earlier: methylation, the marks that set which genes are available to the cell.

Horvath’s 2013 clock reads many human tissues with one formula. On ordinary tissue it predicts chronological age with a median error of 3.6 years — close enough to be useful, far enough from perfect to be a real measurement rather than a trick.

Then run it on stem cells. On embryonic stem cells it returns an age close to zero, which is what it should do. On iPS cells from adults it returns an age close to zero too, with no detectable difference between the two. The adult cell’s age did not come down a little. It went to the floor.

Stopping partway

A pluripotent cell is not a younger version of the cell you started with. It is a cell that has forgotten what it was, and inside a living animal such a cell is either useless or a tumour. Leave all four factors on continuously in a normal mouse and you often get teratomas, or a dead animal within days.

The picture the field uses is Waddington’s landscape. A hillside seen from the side, with a plateau near the top and a ball on it. The ball is a cell, and its height is not age or health — it is how many different things that cell could still become. Below the plateau the hillside runs into forking valleys, and each fork is a commitment: this way lies skin, that way nerve, that way muscle. A ball on a valley floor is an adult cell doing its job, held there by ridges — the identity barrier that stops a skin cell drifting into something else.

OSKM pushes the ball all the way back up, over every ridge, onto the plateau. From there it can roll down again — sometimes into a different valley, which means the wrong tissue, and sometimes into the tumour. Partial reprogramming is one sentence: push the ball part of the way up the wall, stop, and let it fall back into the valley it came from, carrying — if the idea holds — a clock that has been wound back.

Two places the picture misleads. The height of the hill is developmental potential, not age; there is no age axis in the drawing at all, so whether a ball halfway up the wall is a younger cell or merely a confused one is the question the picture cannot answer. And “part of the way up” is not a place you could stop and stand: in an animal it is a length of time, and the difference between rejuvenation and a teratoma is largely how long the factors stay on.

So: stop partway, on a schedule. In 2016 Ocampo and colleagues used mice carrying an inducible OSKM cassette — the four genes built into the genome, inert until a drug switches them on. The switch ran in cycles: two days on, five days off.

The two dosing regimes came apart at once. Continuous dosing caused heavy weight loss and high mortality after four days. Cyclic dosing did neither.

Then the lifespan result, in a progeroid strain — mice bred to age abnormally fast. These were LAKI mice, which express a truncated form of the protein lamin A. On the cyclic schedule, median lifespan rose by 33% and maximum lifespan by 18%. Median is the age by which half the group has died; maximum is how long the longest-lived animals lasted. The comparison was n = 15 treated against n = 18 controls, at p < 0.0001 — the gap is very unlikely to be an accident of which mice landed in which group.

Cyclic dosing helped ordinary mice too, not only progeroid ones: it improved recovery from pancreatic and muscle injury in twelve-month-old animals.

Waddington’s landscape: the ball is a cell, and its height is how many different things it could still become. Pink is the full trip to pluripotency, whose standard proof is a teratoma; cyan is partial reprogramming. Sources: Takahashi & Yamanaka 2006; Ocampo et al. 2016.

How narrow is the window? Narrow enough to be counted in days. Yücel and Gladyshev’s 2024 review collects what is known:

More than two days of continuous expression can be lethal in mice.
Six days of OSKM in cardiac tissue drives heart regeneration.
Twelve days kills the animal.

Ocampo’s mice made the same point from the other side. Animals carrying two copies of the cassette instead of one, on the same two-on five-off schedule for eight weeks, developed teratomas in liver, kidney and pancreas.

Three factors, not four

c-Myc is the awkward member of the set. It is an oncogene — a gene that, driven hard enough, causes cancer — and it shortens the lifespan of mice. It is also not required to start reprogramming.

So Lu and colleagues dropped it and kept the other three: Oct4, Sox2 and Klf4, or OSK. They delivered the genes with AAV, an adeno-associated virus used to carry genes into cells, aiming at retinal ganglion cells — the neurons whose fibres make up the optic nerve. The genes stayed silent until switched on with doxycycline.

What OSK did in those mice:

restored youthful methylation patterns;
promoted axon regeneration after optic nerve crush;
improved vision, both in a microbead model of glaucoma and in ordinary twelve-month-old mice.

One detail matters more than the headline results. The effect required TET1 and TET2, the DNA demethylases — the enzymes that strip methyl marks off DNA. Remove them and the benefit goes. Removing methyl marks is what does the work.

On safety, such as it is: after 10 to 18 months of continuous OSK induction, tumour incidence did not rise.

That is a single-laboratory result in mice, and mice are not people. But it changed what the field was trying to build. Not a reset button. A dimmer switch.

The lifespan number, and how it gets mangled

In 2024 Cano Macip and colleagues treated mice that were already old: male C57BL/6J mice, 124 weeks of age — about 2.4 years, near the end of a normal mouse life. An inducible OSK system was delivered systemically by AAV9, so the whole animal was treated rather than one organ, and doxycycline was cycled one week on, one week off.

The headline result is about remaining lifespan: the time an animal has left, counted from the day dosing began. Median remaining lifespan was 8.86 weeks in controls and 18.5 weeks in treated animals. That is a 109% increase, at p < 0.05 by log-rank — the standard test for whether two survival curves really differ.

Alongside it: frailty scores fell from 7.5 to 6, methylation clocks read the treated liver and heart as younger, and no gross teratomas were found.

Now the mangling. Popular coverage renders this as “reprogramming doubled the lifespan of mice”. It did not.

Dosing began at 124 weeks. Only a sliver of life remained at that point, and the 109% is 109% of the sliver. Count the same animals from birth instead, and total median lifespan went from about 133 weeks to 142.5 — roughly 7%.

Both figures are correct: same animals, same experiment, same survival curve. All that changes is where the stopwatch starts. Only one of the two numbers travels.

One survival curve, measured two ways: +7% on lifespan counted from birth, +109% on the lifespan still remaining when dosing began at week 124. Cano Macip et al. (2024), Cellular Reprogramming 26:24–32; p < 0.05, Mantel-Cox log-rank.

What is established, and what is not

Sorted by how much weight each claim will bear.

Established. OSKM converts somatic cells — any cell of the body that is not an egg or a sperm — to pluripotency, and pluripotent cells read as epigenetically newborn. Replicated, routine, not in dispute.

Preliminary. Every rejuvenation result above. They are mouse studies, mostly single-laboratory, with tens of animals rather than hundreds. And one of the strongest, Ocampo’s lifespan curve, comes from a progeroid strain — a model of one specific molecular lesion, not of ordinary ageing. Helping a mouse with a truncated lamin A is not the same as slowing normal ageing.

Contested. What an epigenetic clock actually measures.

Remember how those clocks were built. Someone fitted methylation data to people’s known chronological ages and kept the sites that fitted best. The clock is therefore a correlational predictor: selected for tracking age, not for causing it. Moving a correlational predictor is not the same as moving the thing it predicts.

An analogy, with its limit attached. Grey hair predicts age well; dyeing it moves the predictor and changes nothing underneath. Reprogramming is plainly doing more than dye, but the shape of the worry is the same.

Yücel and Gladyshev put it plainly: rejuvenation and clock reversal “should not be used interchangeably”.

There is also a category of ageing that reprogramming does nothing about. Somatic mutations — errors in the DNA sequence itself, accumulated over a lifetime — are damage in the ordinary sense, and rewriting the marks on a corrupted sequence does not correct it. Ageing plausibly has an epigenetic component that is addressable and a damage component that is not.

Human data comes in two parts.

In a dish, yes. Gill and colleagues transiently reprogrammed fibroblasts from donors aged 38 to 53: factors on for thirteen days, then off, stopped short of pluripotency. The cells came back roughly thirty years younger on epigenetic and transcriptomic measures, and they were still fibroblasts.

In a person, no efficacy data exists. A 2024 survey found every reprogramming company still preclinical. Altos Labs, Retro Biosciences and NewLimit are research organisations with large budgets, not firms sitting on a therapy.

The first human study began recruiting in March 2026. Its shape is worth stating exactly, because it will be reported as more than it is:

eighteen participants;
Phase 1;
AAV-delivered OSK, injected into one eye, for optic neuropathies;
endpoints: adverse events, dose-limiting toxicities and liver enzymes.

Every one of those endpoints is a safety measure. The trial asks whether the treatment is tolerated, not whether it works.

What to take away

The interesting claim here is not about longevity. It is about causation.

This series treated the epigenome as the layer deciding which genes a cell reads. Partial reprogramming is the strongest evidence that the layer is partly instructive rather than merely descriptive: the marks are not only a record of what the cell became. Pushing them changes a cell’s functional state, in living tissue, in the direction of younger.

That is a result about mechanism, and today it is only that. A therapeutic version needs three further things:

the factors delivered to the right cells;
expression held inside a window measured in days;
a reliable off switch.

All three are unsolved, and any one of them is fatal alone. Treat any product sold on the strength of these papers as fiction. Treat the papers themselves as the most interesting thing in ageing biology.