On 14 October 2026, a paper landed in Nature Aging with a result built for headlines: one gene, switched on in old mice, and they lived 14% longer.[1] Within a day the number was everywhere, usually attached to the phrase “reverse aging.”
The 14% is real and it was measured properly. But it is not the interesting part of this paper, and treating it as the headline gets the story backwards. What makes this study unusual is how the gene was chosen. Nobody had a hypothesis about SRSF1 and longevity. A model did the choosing, and it chose something a human probably would not have.
The screen that picked the candidate
The group — led by George Church at Harvard's Wyss Institute, with the computational side run by Antonio del Sol's team at CIC bioGUNE in Spain — started by building what they call an interpretable transcriptomic age predictor.[1] In plain terms: a model that reads which genes a human skin cell is currently switching on and off, and estimates how old that cell is from the pattern alone. It was trained on gene-expression data from 464 fibroblast samples taken from donors across a wide range of ages.
This is a cousin of the epigenetic clocks you may have seen sold as consumer tests, but with a different readout. Epigenetic clocks read chemical tags on DNA. This one reads the cell's actual activity — its transcriptome. The word doing the heavy lifting is interpretable: the model was built so researchers could ask it which genes it was weighing, rather than just accepting a number.
Then they ran it in reverse. Instead of using the model to score cells, they used it as a judge in a screen: overexpress a gene in an aged human fibroblast, re-read the transcriptome, and ask the model whether the cell now looks younger. Run that across a focused library of candidate genes and let the ranking fall out.
SRSF1 came out on top. It restored youthful signatures across three fibroblast lines from donors aged 55, 65 and 79 — reducing cellular senescence, proteostatic stress and inflammatory signalling, three items straight off the hallmarks-of-aging list[2] — without the cells losing their identity as fibroblasts.
Why a splicing factor is a strange answer
SRSF1 is not a longevity gene in any conventional sense. It is a serine/arginine-rich splicing factor: a protein that helps decide, after a gene is read into RNA, which pieces of that RNA get kept and which get cut out. One gene can yield several different proteins depending on how that editing goes, and SRSF1 is one of the workers making the call.
That is housekeeping, not glamour. Which is exactly why it is interesting. Most of the longevity pipeline — rapamycin, NAD precursors, senolytics, the repurposed drugs people argue about online — came from humans reasoning forward from a mechanism they already believed in. A model with no priors reached instead for the machinery that decides what the cell's messages say. If that result holds, it says something about aging: that part of it may be a slow drift in RNA processing, an accumulating editing error rather than a broken switch.
What actually happened in the mice
The in-vivo arm is the part worth reading slowly. The team packaged SRSF1 into a liver-tropic adeno-associated virus and gave it to aged mice under inducible control — meaning expression could be turned on, rather than running flat out from birth.
Two outcomes were reported. Frailty progression fell by 44%, measured on a mouse clinical frailty index — a validated scoring sheet of visible deficits (coat condition, grip, gait, body condition) that has been used to compare longevity interventions since 2016.[3] And median lifespan rose by 14%.[1]
The frailty number is arguably the better one. Lifespan tells you when an animal died; a frailty index tells you what shape it was in on the way there. A 44% slowdown in accumulating deficits is a healthspan claim, and healthspan is what most people actually want when they say they want to live longer.
The number in the brackets
Now the caveat the headlines dropped. The abstract states the group sizes plainly: n = 9–10 per group.[1]
Nine or ten mice is a small lifespan study. Mouse survival curves are noisy; a couple of animals dying early or late for unrelated reasons can move a median by several percent. This is why the field's gold-standard programme, the NIA Interventions Testing Program, runs survival studies across multiple independent sites with hundreds of animals — the rapamycin result that made the field take geroscience seriously was built that way, in genetically heterogeneous mice at three laboratories.[4]
That does not make the SRSF1 finding wrong. It makes it a first observation rather than a settled effect size. The honest reading is: a 14% median extension was observed in a small cohort, and the number itself should be treated as provisional until someone repeats it at scale. The authors say as much — they frame SRSF1 as a promising geroprotective target, not a finished intervention.
The sentence missing from the coverage
Here is what almost no write-up of this paper mentioned. SRSF1 has a well-documented second identity in the cancer literature. Under its older name SF2/ASF, it was characterised in 2007 as a proto-oncogene: modest overexpression was enough to transform cells and drive tumour formation in a mouse model.[5] It has been studied as an oncogenic driver repeatedly since.
This is not a gotcha, and it does not mean the Harvard result is unsafe. It means the therapeutic window matters enormously, and it explains a design choice that otherwise looks like fussiness: the team used inducible expression rather than permanent overexpression. A gene that rejuvenates cells at one dose and transforms them at another is a dial, not a switch, and the whole question becomes where the safe setting is and whether it stays there for years.
Anyone reading this as a preview of a future injection should hold that thought alongside the 14%. Long-term tumour surveillance in a properly powered cohort is the single study that has to happen next, and it has not happened yet.
One more piece of context, stated openly in the paper's own competing-interests section: several authors, including Church, are named inventors on a patent application covering the use of SRSF1, and two are co-founders with equity in RegulaBio, an early-stage company developing therapeutics for age-related disease.[1] Disclosed conflicts are how science is supposed to work — the declaration is the system functioning, not failing. But it is useful to know that the enthusiasm around a result has a commercial shape to it, especially in a field where reversal claims move money.
How this differs from Yamanaka reprogramming
The obvious comparison is partial reprogramming. In 2016, Juan Carlos Izpisua Belmonte's group showed that cycling the Yamanaka factors in mice could ameliorate hallmarks of aging in vivo — a landmark result that also came with a hard problem: push the factors too far and cells lose their identity entirely, with teratomas as the failure mode.[6] Dosing has been the central engineering challenge ever since.
SRSF1 is pitched as a way around that. The paper's claim is that it produces youthful transcriptional signatures without loss of cellular identity — a fibroblast stays a fibroblast.[1] If that holds up, it is a meaningful structural advantage over Yamanaka-factor approaches, because the thing you are steering around is no longer the cliff edge of de-differentiation.
It trades one risk for another, though. Yamanaka reprogramming risks the cell forgetting what it is. A splicing factor with oncogenic history risks the cell doing what it does too enthusiastically. Neither is obviously the easier problem.
What would make this matter for people
Three things, in order. A replication of the lifespan effect in a properly powered, multi-site cohort — that is the difference between a signal and a number you can quote. Long-term cancer surveillance, given ref 5. And a delivery route: this was liver-tropic AAV in mice, which is a long way from anything a person would receive.
Realistically that is years, and most candidates that look this good in mice do not survive the journey. The honest comparison is that the ceiling on human lifespan has barely moved despite decades of mouse results that looked like this one.
What is different here is the method, and the method is reusable. An interpretable age model used as a screening judge is a tool that can be pointed at the rest of the genome, in other cell types, repeatedly. SRSF1 may or may not become a drug. The screen that found it is the part most likely to still matter in five years.
Common questions
What is SRSF1?
SRSF1 is a serine/arginine-rich splicing factor — a protein that helps decide how RNA is edited after a gene is read, determining which version of a protein a cell produces. In October 2026 a Harvard-led team identified it as a regulator of cellular aging using a machine-learning screen, and showed that inducing it in aged mice reduced frailty progression by 44% and was associated with a 14% extension of median lifespan.
Can SRSF1 reverse aging in humans?
Not today, and nobody has tested it. The work is preclinical: human skin cells in a dish and a small mouse cohort of nine to ten animals per group. It was delivered by liver-tropic AAV gene therapy, not a pill or supplement. SRSF1 was also characterised as a proto-oncogene in 2007, so long-term cancer surveillance is a prerequisite before any human application.
Is there an SRSF1 supplement?
No. SRSF1 is a gene encoding a protein inside your cells, not a compound that can be swallowed. The study raised its expression using an engineered virus under inducible control. Any product marketed as an SRSF1 supplement is not doing what this research did — if you are looking at the evidence-backed end of that shelf, start with what actually has human data.