Eight things, forty-two thousand impressions, one big asterisk.
On the evening of May 28, David Sinclair posted a list. Eight items, one tweet, no caveats: exercise, plant-rich diets, semaglutide, caloric restriction, ketamine, omega-3s, plasma, pitavastatin.1 Each one, he wrote, was "associated with decreased epigenetic age" in a new paper he had co-authored with Adiv Johnson. By the time I started writing this, the post had 444 likes, 331 bookmarks, and 42,000 impressions, and the comments had filled with people asking whether they should add ketamine to their stack.
The paper is real. It is a review titled Turning back time: a comprehensive list of interventions that decrease next-generation epigenetic aging clocks in humans, published this month in Frontiers in Genetics.2 It catalogs 41 human studies in which someone took an intervention, had their blood methylation measured before and after, and reported a change in at least one second-generation epigenetic clock score. That is what the paper does. It is also, more or less, all that it does.
The tweet does something different. It compresses a hedged biomarker review into a list that reads like a recipe. The paper says these eight interventions are associated with reduced clock scores. The tweet implies that reducing clock scores reverses aging. The paper says clock movement does not necessarily translate to a change in biological age or lifespan. The tweet does not. The paper's first author is the Head of Scientific Affairs at Tally Health, a company that sells a $229 cheek-swab test built on exactly these clocks. The tweet does not say that either.
This is a piece about the gap between those two things.
What an epigenetic clock is, and what it isn't.
An epigenetic clock is a statistical model. You take a sample of blood, you measure DNA methylation at hundreds or thousands of specific cytosine sites, and you plug those numbers into a regression equation that was trained, originally, to predict a person's chronological age. The first one, published by Steve Horvath in 2013, did this well enough to be uncanny.3 A second generation followed: PhenoAge in 2018, GrimAge in 2019, DunedinPACE in 2022.4,5 These newer clocks were trained not on age itself but on mortality and a panel of clinical biomarkers like C-reactive protein, glycated hemoglobin, and white-cell counts. They predict who will die sooner better than chronological age does.
Notice what just happened. A clock trained on hsCRP, A1c, and white-cell counts is now being marketed as a measurement of "biological age." When a drug that lowers hsCRP and A1c lowers your clock score, what has it actually done? It has lowered the inputs to the regression. Whether it has changed how you will age is a separate, unanswered question.
This is not my objection. It is the field's. The FDA does not recognize epigenetic clocks as a surrogate endpoint for any indication.6 A 2022 Nature feature on the field summarized the consensus position bluntly: clocks are "investigational biomarkers" that "have not yet been thoroughly validated as clinical surrogate endpoints."6 That language reappears, almost verbatim, in the Sinclair and Johnson review itself.2 The hedge is in the paper. It is not in the tweet.
The reliability problem is worse than the validity problem. In 2022, Albert Higgins-Chen and Morgan Levine published a paper in Nature Aging showing that the original clocks were so noisy in test-retest that small intervention effects were inside the measurement error.7 They built a principal-component version of the clocks specifically because the raw output had to be cleaned up before you could trust it to move with treatment. The Sinclair-Johnson review uses the PC-corrected clocks for some studies and the raw clocks for others. It is not always the same clock that moves.
A clock trained on hsCRP and A1c is now being marketed as a measurement of biological age. When a drug that lowers hsCRP and A1c lowers your clock score, what has it actually done?
On the surrogate-endpoint problemThis matters because the field's most respected trial, the CALERIE-2 caloric-restriction study, illustrates the problem perfectly. Dan Belsky's group at Columbia took 185 healthy adults, randomized them to two years of 25% caloric restriction or no restriction, and measured five second-generation clocks before and after.8 One moved. DunedinPACE slowed by roughly two to three percent in the restricted arm. GrimAge did not change. PhenoAge did not change. PC GrimAge did not change. PC PhenoAge did not change. This is the best-designed intervention trial in the entire review. Four of five clocks failed to detect a signal in a randomized study of the most-validated aging intervention in human biology. That should be the headline. Instead it is a footnote.
The methodological floor of the field.
To understand the trial base the Sinclair-Johnson review draws on, it helps to look at the paper that started the whole "clock reversal" claim. In 2019, Gregory Fahy and Stephen Horvath published TRIIM in Aging Cell.9 They gave a cocktail of growth hormone, DHEA, and metformin to nine men aged 51 to 65, for one year. There was no placebo arm. There was no randomization. There were no women. They reported about a 2.5-year reduction in epigenetic age. The paper is still cited as evidence that aging can be reversed.
The TRIIM trial. Nine men, ages 51 to 65, all received growth hormone plus DHEA plus metformin for one year. Stored samples were sent to Horvath's lab after the fact for clock analysis.9
Result: Roughly 2.5 years of "epigenetic age reversal" across several clocks. No placebo comparator. No randomization. No women.
Limitation: By modern evidence standards, this is a single-arm pilot. It would not be admissible as primary evidence in any other drug or biomarker domain. It remains the second-largest year-reduction in the new review.
Then there is the next-cleanest study in the review, CALERIE-2. This is a properly randomized, controlled, well-powered trial.8 The effect was a two to three percent slowing on one clock out of five. Belsky's group, to their credit, has been careful in how they frame this. Mapping that pace change to a mortality benefit requires extrapolation from observational cohorts where DunedinPACE predicts death. The mapping suggests a roughly 10 to 15 percent mortality reduction at the population level. That estimate has never been tested by following the trial participants long enough to die.
CALERIE-2 caloric restriction trial. Healthy non-obese adults randomized to 25% caloric restriction versus ad libitum eating for two years. Five next-generation clocks measured pre- and post-intervention.8
Result: DunedinPACE slowed by 2 to 3 percent in the restriction arm. GrimAge, PhenoAge, PC GrimAge, PC PhenoAge: no significant change. Modeled mortality benefit by extrapolation: roughly 10 to 15 percent.
Limitation: Single clock responded out of five. The mortality estimate is extrapolated from observational cohorts, not measured in trial participants. Real-world adherence to sustained 25% restriction is near zero.
The third anchor of the review is DO-HEALTH, a Swiss-led trial published last year.10 Seven hundred and seventy-seven older adults were randomized in a two-by-two-by-two factorial of vitamin D, omega-3, and home exercise. Three years of follow-up. The largest aging-clock RCT to date. Omega-3 alone shifted PhenoAge, GrimAge2, and DunedinPACE by standardized effects of 0.16 to 0.32 units. In raw terms, that translated to roughly three to four months of "epigenetic age" over three years. Vitamin D alone did not move any clock. Exercise alone did not move any clock. Only the combination of all three had additive effects, and even there the additive PhenoAge benefit was about a tenth of a year over three years.
If the omega-3 result is the strongest supplement signal in the field, the right way to describe it is "two to four months over three years." That is what a well-powered randomized trial of the cleanest intervention in the list actually produced. It is also a number nobody puts in a tweet.
What each item on the list actually is.
Exercise (1). The two studies cited in the review are a 33-person uncontrolled cycling trial and a 24-person frail-elderly nutrition-plus-exercise trial.2 Cycling alone reduced PC GrimAge by about 0.6 years versus the participants' own baselines, with no comparison group. The frail-elderly trial moved PhenoAge but not GrimAge2. Meanwhile, DO-HEALTH found exercise alone did not move any clock over three years.10 Exercise extends life in trials measured in deaths and disability. The clock literature does not strengthen that case. It cluttters it.
Plant-rich diet (2). The signal comes mostly from the DAMA trial, which randomized 219 Italian women to a plant-rich diet or control for two years.11 GrimAge dropped by about 0.66 years in the diet arm. PhenoAge and DunedinPACE were not measured. In a separate Mediterranean-diet trial, the related NU-AGE study, the clock effect was significant in Polish participants but not Italian ones, which is the kind of country-by-country split that makes a generalizable signal hard to claim. DIRECT-PLUS, a 256-person trial of a Mediterranean or "Green-Med" diet in abdominally obese adults, found no significant change across PhenoAge, DunedinPACE, or PC GrimAge.2 Plant-rich eating is supported by dozens of hard-endpoint trials. The clock numbers are inconsistent supporting evidence, not the case for the diet.
Semaglutide (3). This is the most impressive trial in the review, and the one I want to look at most carefully.
Brave Beyond Beauty trial (NCT04019197). Randomized, double-blind, placebo-controlled, 32 weeks. Forty-five patients on semaglutide, 39 on placebo. The study population was adults living with HIV who had developed lipohypertrophy, a metabolic side effect characterized by abnormal fat deposition.12
Result: PCGrimAge reduced by 3.1 years (p=0.007). GrimAge V2 reduced by 2.3 years (p=0.009). PhenoAge reduced by 4.9 years (p=0.004). DunedinPACE slowed by roughly 9 percent (p=0.01). Adjustments made for sex, BMI, hsCRP, and sCD163.
Limitation: Patients lost approximately 10% of body weight. The clocks that moved are most heavily weighted on hsCRP and A1c, which improve with weight loss alone. The HIV lipohypertrophy population has accelerated epigenetic aging at baseline; rolling that back toward normal is not the same as making a healthy adult younger. Not replicated in any non-HIV, non-obese population.
This is the trial Sinclair's tweet is built on, and it is real evidence of something. The question is what. Semaglutide reduces inflammation and metabolic disease, both of which are clock inputs by construction. The trial population started with markedly accelerated clocks because of the underlying HIV-related inflammation. The drug normalized their metabolic state. Whether that constitutes "aging reversal" or "metabolic-disease reversal that happens to register on an inflammation-trained biomarker" is the question the field has not answered. A separate single-arm pilot in 41 patients with HIV-related fatty liver, SLIM LIVER, found similar directional effects on DunedinPACE.13 Same population, no placebo, same caveat.
Caloric restriction (4). Discussed above. CALERIE-2. One clock out of five.8
Ketamine (5). This is the entry that most needs a footnote. The cited study is Dawson et al. 2025 in Translational Psychiatry.14 It is a single-arm pilot with no placebo and no randomization. Twenty patients with major depressive disorder or post-traumatic stress disorder received six ketamine infusions over two to three weeks. Seven epigenetic clocks were measured. After Bonferroni correction for multiple comparisons, exactly one survived: OMICmAge, which dropped by about 1.81 years. PhenoAge and GrimAge2 reached nominal significance before correction. After correction, both were not significant. DunedinPACE, PC PhenoAge, PC GrimAge, and Systems Age were not significant at any threshold.
Ketamine for MDD and PTSD. Six intravenous infusions at 0.5 mg/kg over two to three weeks. No placebo. No randomization. Conducted in collaboration with TruDiagnostic, a commercial epigenetic-age testing company, and Expedition Mental Health, a private ketamine clinic.14
Result: Of seven clocks tested, only OMICmAge survived Bonferroni correction. Movement of 1.81 years on that one clock.
Limitation: Severely depressed and traumatized patients have elevated inflammation, cortisol, and disordered sleep, which inflate inflammation-trained clocks. Treating their depression normalizes those inputs. Ketamine is unblindable due to dissociation, and depression trials are famously placebo-responsive. The result is consistent with treating an acute inflammatory state, not aging.
Omega-3 (6). DO-HEALTH is the strongest study and gets us to a few months of epigenetic age over three years.10 Other trials are smaller. The result is real but small. The breathless framing collapses "negative 0.64 years over three years" into a number people read as "fish oil makes you a year younger."
Plasma (7). The plasma signal rests on Clement et al. 2022, an Aging Cell paper.15 Eighteen elderly adults received intramuscular umbilical cord plasma concentrate for 10 weeks. No control arm. GrimAge dropped 0.82 years against the participants' own baselines. The same 18-person dataset was reanalyzed by Ying et al. in 2024 and shows up as a separate citation, which makes the evidence base look larger than it is. The Conboy lab at UC Berkeley has argued for years that the original parabiosis findings, in which young blood appeared to rejuvenate old mice, are best explained by dilution of old-blood factors, not infusion of young ones.16 If they are right, adding cord plasma should not work. And in the same Sinclair-Johnson review, Borsky et al. 2025 reported that multiple rounds of plasmapheresis in 34 healthy subjects accelerated their epigenetic clocks.2 Plasma manipulation, in this review, moves the clock in both directions. The FDA issued a 2019 safety warning specifically against young-plasma infusions marketed for anti-aging.17
Pitavastatin (8). The cited trial is a substudy of REPRIEVE, a large statin-for-HIV-prevention trial.18 Ninety-nine participants, 65 on pitavastatin and 34 on placebo, with HIV, for two years. Baseline epigenetic age acceleration was about seven years above chronological age, a feature of the HIV population. PCGrimAge: no change. DunedinPACE: rose in placebo, did not rise in pitavastatin. The "effect" is interpreted as preventing further acceleration in an already-accelerated cohort. Pitavastatin was chosen for REPRIEVE because it does not interact significantly with antiretrovirals, not because anyone hypothesized epigenetic-age effects. Whether the result is a class effect of any statin or specific to pitavastatin is unknown. The most parsimonious explanation is statins' well-characterized reduction in systemic inflammation, which is also a GrimAge input.
Forty-one studies, six of them properly randomized, three positive findings sized in months, none replicated in healthy general populations.2,12
The commercial layer behind the byline.
The Frontiers paper has two named authors. Adiv Johnson is the first and corresponding author. The author-contribution statement credits him with conceptualization, data curation, investigation, methodology, project administration, visualization, and writing the original draft. David Sinclair is the second author. He is credited with conceptualization and writing the review-and-editing. In academic terms, Johnson did almost all of the work.
Adiv Johnson's day job is Head of Scientific Affairs and Education at Tally Health, the New York-based company that sells direct-to-consumer epigenetic-age tests built on the second-generation clocks this paper is reviewing. Tally's pricing structure runs $229 for a single test and $129 to $199 per month for a subscription that includes a quarterly retest plus lifestyle recommendations.19 David Sinclair co-founded Tally and is a director, equity holder, and scientific advisor. He has also founded or holds equity in Life Biosciences, EdenRoc Sciences, NucleicAI, and Paradigm88, all disclosed in the paper's conflict-of-interest statement.2The funding line states that "financial support was not received for this work." The acknowledgements credit a Tally Health staffer for revising the paper's figures.
This is not a hidden conflict. It is disclosed. But disclosure is the floor, not the ceiling. The relevant question is whether a literature in which the most prominent reviewer and the company selling the test are entangled this thoroughly can be relied on to mark its own homework. The track record is informative.
The relevant question is whether a literature in which the most prominent reviewer and the company selling the test are entangled this thoroughly can be relied on to mark its own homework.
On the Tally Health bylineIn 2003, Sinclair reported that resveratrol activated the SIRT1 enzyme and extended life in yeast. In 2004, he co-founded Sirtris Pharmaceuticals around this finding. In 2008, GlaxoSmithKline acquired Sirtris for $720 million. Within two years, multiple replication efforts at Pfizer and Amgen showed that the original SIRT1 activation assay was an artifact of the fluorescent peptide substrate used in the screening assay; the compounds did not directly activate SIRT1.20 Sirtris's lead compound was discontinued. By 2013, GSK had dissolved Sirtris as a separate unit. Resveratrol itself has subsequently failed to extend lifespan in normally-fed mice and has not produced longevity benefits in human trials.
The NMN chapter followed a similar arc. Sinclair promoted nicotinamide mononucleotide as a longevity supplement through his own commercial ventures and his bestselling 2019 book Lifespan. Charles Brenner, the discoverer of the nicotinamide-riboside salvage pathway, published an extensive scientific critique of Sinclair's framing in Archives of Gerontology and Geriatrics in 2022.21 Notably, when nicotinamide riboside was tested in a clinical trial for epigenetic-age effects (Orr et al. 2024), it appeared in the new review's table of non-significant results. The same paper that catalogs the eight things that "work" also catalogs the things that did not work, including some that Sinclair himself spent years promoting. The tweet did not mention those either.
What the paper does not say, and what the tweet does not show.
It is worth pausing on what the Frontiers review actually catalogs versus what Sinclair's tweet selects from it. The full review describes 41 trials. Of those, the authors classify 23 as showing at least one significant clock reduction and 18 as null. The null table is long. It includes nicotinamide riboside, resveratrol in some trials, rapamycin, metformin in some configurations, fisetin, and several supplement combinations.2 The fact that rapamycin and metformin are missing from the tweet is striking, because these are the two molecules longevity influencers spend the most time talking about. Their absence is a finding. The author chose not to highlight it.
Five of eight signals come from disease populations.
Semaglutide: HIV lipohypertrophy. Pitavastatin: HIV. Ketamine: severe depression or PTSD. One omega-3 trial: mild cognitive impairment. Plasma: a small uncontrolled group of elderly. Generalizability to healthy adults is untested.
The effect sizes are months, not years.
DO-HEALTH's omega-3 result, the strongest supplement signal, is roughly three to four months of "epigenetic age" reduction across three years of daily supplementation. CALERIE's caloric-restriction signal is a two to three percent slowing of one pace clock.
Weight loss confounds the metabolic signals.
Semaglutide trial participants lost about 10% of body weight. The clocks that moved most are weighted heavily on inflammation and glucose-handling markers, which improve with weight loss alone. The paper does not adjust for this in its narrative framing.
The clocks are not validated as surrogate endpoints.
The FDA does not recognize epigenetic age as a clinical endpoint. The paper itself says clocks "have not yet been thoroughly validated as clinical surrogate endpoints." Moving the biomarker has not been shown to change what the biomarker is supposed to predict.
There is a deeper omission. Almost every clock that moved in the positive trials had its training inputs improved by the intervention. Semaglutide reduces hsCRP and A1c. GrimAge is built on hsCRP and A1c. Statins reduce LDL and inflammation. GrimAge is built on inflammation. Ketamine, in this trial, treated depression and likely reduced acute inflammation. OMICmAge is sensitive to inflammation. The clocks were trained to predict mortality from a panel of biomarkers. Drugs that move those biomarkers move the clocks. That is not evidence of aging reversal. It is evidence that the regression is doing what regressions do.
To know whether semaglutide or statins or ketamine are actually slowing aging, in the sense that matters, you would need a trial in which the intervention was randomized in a general population, the clocks moved, and then the people randomized to the intervention lived longer or got sick less. No such trial has reported. The Sinclair-Johnson review acknowledges this. The tweet implies otherwise.
A biomarker review, not a recipe.
The Frontiers paper itself is a reasonable, if narrative, review. It catalogs 41 human studies of next-generation epigenetic clocks. Its written conclusions are hedged and accurate: clocks are investigational biomarkers, effect sizes do not directly translate to lifespan, the field is in early days. Read the abstract and you will not find a claim that any of these eight interventions extends human life.
The tweet is the problem. A list of eight things, posted to a million-follower account, read by people who will not click through to the paper, will be interpreted as a recipe. The list collapses a noisy, small, mostly-uncontrolled trial base into a stack you can buy. It puts ketamine, with one pilot in 20 depressed patients, alongside exercise, which is supported by hard-endpoint trials in hundreds of thousands of people. It puts umbilical-cord plasma, with 18 elderly participants and no control group, alongside caloric restriction, which is the closest thing aging biology has to a foundational intervention. Treating these as comparable is the marketing.
The commercial overlay matters. The first author works for the company that sells the test that this entire literature props up. The second author co-founded that company. That does not mean the science is wrong. It means the framing is conflicted, and the framing is what people see.
I rate this Insufficient Data on the underlying biology and the framing as Marketing Hype. The interventions that have hard-endpoint evidence (exercise, plant-rich eating) had it before this review and will have it after. The interventions that do not (ketamine, plasma, pitavastatin in a general population) are not made stronger by movement in a biomarker that has never been shown to causally drive the outcome it is built to predict. If anyone tells you this paper proves anti-aging works, ask them which trial, in which population, was randomized, blinded, and followed to a hard endpoint. The answer is none of them.
The Sinclair tweet is a marketing compression of a hedged biomarker review. Eight interventions, mostly in disease populations, mostly small, mostly uncontrolled, all measured against a clock that the FDA does not accept as a surrogate endpoint. Eat plants, move your body, see a doctor about your statin. Don't buy the cheek swab.
- 1. Sinclair DA. Post on X, May 28, 2026: "Pleased to share a new paper on medicines, supplements & other factors associated with decreased epigenetic age with Dr Adiv Johnson @JoinLifespan..." x.com/davidasinclair/status/2060128617753981321
- 2. Johnson AA, Sinclair DA. Turning back time: a comprehensive list of interventions that decrease next-generation epigenetic aging clocks in humans. Frontiers in Genetics. 2026;17:1836446. doi:10.3389/fgene.2026.1836446.
- 3. Horvath S. DNA methylation age of human tissues and cell types. Genome Biology. 2013;14(10):R115.
- 4. Lu AT, Quach A, Wilson JG, et al. DNA methylation GrimAge strongly predicts lifespan and healthspan. Aging. 2019;11(2):303-327.
- 5. Belsky DW, Caspi A, Corcoran DL, et al. DunedinPACE, a DNA methylation biomarker of the pace of aging. eLife. 2022;11:e73420.
- 6. Couzin-Frankel J. Turning back time with epigenetic clocks. Nature news feature. 2022. doi:10.1038/d41586-022-00077-8.
- 7. Higgins-Chen AT, Thrush KL, Wang Y, et al. A computational solution for bolstering reliability of epigenetic clocks. Nature Aging. 2022;2(7):644-661.
- 8. Waziry R, Ryan CP, Corcoran DL, et al. Effect of long-term caloric restriction on DNA methylation measures of biological aging in healthy adults from the CALERIE trial. Nature Aging. 2023;3(3):248-257.
- 9. Fahy GM, Brooke RT, Watson JP, et al. Reversal of epigenetic aging and immunosenescent trends in humans. Aging Cell. 2019;18(6):e13028. Single-arm pilot, n=9 men aged 51-65, growth hormone plus DHEA plus metformin, 1 year, no placebo.
- 10. Bischoff-Ferrari HA, et al. Individual and additive effects of vitamin D, omega-3 and exercise on DNA methylation clocks of biological aging in older adults from the DO-HEALTH trial. Nature Aging. 2025. RCT, n=777, 3-year factorial of vitamin D, omega-3, and home exercise.
- 11. Fiorito G, et al. DAMA trial: dietary intervention for breast cancer prevention. n=219 Italian women, 24 months. GrimAge reduction -0.66 years in plant-rich diet arm. GeroScience. 2021.
- 12. Corley MJ, et al. Semaglutide slows epigenetic aging in a randomized trial of HIV-associated lipohypertrophy (NCT04019197). Nature Communications. 2026. (preprint: medRxiv 2025.07.09.25331038). RCT, n=84 completers, 32 weeks.
- 13. Stein JH, et al. Semaglutide and epigenetic aging in the SLIM LIVER study (ACTG A5371, NCT04216589). npj Aging. 2026. Single-arm, n=41, 24 weeks, HIV with MASLD.
- 14. Dawson K, et al. Epigenetic aging and DNA methylation biomarker changes following ketamine treatment in patients with MDD and PTSD: a pilot study. Translational Psychiatry. 2025;15:452. Single-arm pilot, n=20, NCT05294835. Only OMICmAge survived Bonferroni correction.
- 15. Clement J, Yan Q, Agrawal M, et al. Umbilical cord plasma concentrate has beneficial effects on DNA methylation GrimAge and human clinical biomarkers. Aging Cell. 2022;21(10):e13696. n=18 elderly, 10 weeks, intramuscular injection, no control arm.
- 16. Mehdipour M, Skinner C, Wong N, et al. (Conboy lab) Rejuvenation of three germ layers tissues by exchanging old blood plasma with saline-albumin. Aging. 2020;12(10):8790-8819. The dilution-of-old-blood hypothesis.
- 17. US Food and Drug Administration. Statement from FDA Commissioner Scott Gottlieb cautioning consumers against receiving young donor plasma infusions promoted as treatment for varying conditions. February 19, 2019.
- 18. Corley MJ, et al. Effect of pitavastatin on epigenetic aging biomarkers in people with HIV: pilot substudy of the REPRIEVE trial (NCT02344290). Clinical Infectious Diseases. 2026;81(6):e560-e567. n=99, 24 months.
- 19. Tally Health pricing. tallyhealth.com. Single test $229; subscription $129-$199/month with quarterly retests (accessed May 2026).
- 20. Pacholec M, Bleasdale JE, Chrunyk B, et al. SRT1720, SRT2183, SRT1460, and resveratrol are not direct activators of SIRT1. Journal of Biological Chemistry. 2010;285(11):8340-8351. Industry replication that showed the original Sirtris assay was a fluorophore artifact.
- 21. Brenner C. A science-based review of the world's best-selling book on aging. Archives of Gerontology and Geriatrics. 2022. PMC9669175. Critique of Sinclair's framing of NMN and resveratrol in Lifespan.