> ## Content Index > Fetch the complete content index at: https://nexi.fund/llms.txt > Use this file to discover other available public pages before exploring further. # 51 Trials, 16 Aging Clocks, and the $101M Measurement Gap - URL: https://nexi.fund/healthspan-measurement-gap/ - Published: 2026-10-05T20:00:16.000Z - Updated: 2026-10-05T20:00:16.000Z - Description: TranslAGE pooled 51 interventional studies and ran 16 clocks plus 94 further biomarkers across all of them. Chronological-age clocks moved sporadically; one held up everywhere. Mortality-trained clocks all fell and agreed. Gastric bypass cut a metabolic score by 0.43 without moving a general clock. - Author: Nexi.fund Labs - Tags: Biotech & Health, #mode-6, #hook-statistic, #track-A Fifty-one clinical studies. Sixteen epigenetic clocks. Ninety-four further methylation biomarkers. And a field that has just been handed a systematic answer to a question it had been avoiding: do the instruments used to measure aging respond when something actually works? The test ran in *Nature Medicine* on 21 August 2026\. Raghav Sehgal, Daniel Borrus and Albert Higgins-Chen built TranslAGE, a harmonised database of 51 longitudinal interventional studies, then computed the same set of clocks across every one of them. Steve Horvath, who published the first pan-tissue methylation clock in 2013, wrote the accompanying commentary and set the standard: this is what a surrogate endpoint has to do — move when there is genuine clinical benefit, and stay still when there is not. ⚠️ **What the analysis established** Clocks trained to predict mortality or the pace of aging respond consistently across interventions. Clocks trained on chronological age mostly do not. Reliable clocks of the second generation and newer all showed a significant decrease, and they agreed with each other. Only one reliable first-generation clock responded across every intervention in the set. That matters, because $101M of prize money is now being spent against this readout. 51 interventional studies #### Studies in the harmonised set Public and private longitudinal trials, re-analysed so one identical clock panel could be computed across all of them · *Nature Medicine, 2026* 16 clocks per study #### Clocks tested side by side Plus 94 additional DNA methylation biomarkers used to explain what moved underneath each clock · *Nature Medicine, 2026* $10M milestone 2 pool #### Ten teams, $1M each Twenty finalists announced in Salt Lake City on 11 August 2026; ten named awardees, drawn from four countries · *XPRIZE Foundation, 2026* 14 healthy adults enrolled #### Registered plasmid combination trial Early Phase 1, non-placebo-controlled, completed April 2026, with epigenetic biological age as a primary outcome · *ClinicalTrials.gov, 2026* ## The prize just moved from theory to protocol On 11 August 2026, in a Salt Lake City awards ceremony, XPRIZE named 20 finalist teams in its $101M Healthspan competition. Ten of them received a $1M Milestone 2 award, taking total milestone funding to $10M. An up to $81M grand prize follows in 2030 — one of the largest incentive prizes ever offered for a single scientific breakthrough. The important detail sits in the operations. The headline number is the least informative part of it. From 2026 to 2029 the finalists will run clinical trials of up to one year each in adults aged 50 to 90\. The University of Utah's Data Coordinating Center will handle trial and data management across the whole field. The UCSD Stein Institute for Research on Aging runs the shared laboratories, the shared biomarker pipeline, and the shared handling of blood and cell samples. Twenty competing therapies, one measuring system. That is a deliberate design, and it moves the bottleneck. For thirty years the hard problem in aging research was whether any intervention could work. The remaining problem is whether the field can tell, on a shared dataset with a shared assay, that one did. A prize that standardises the ruler is doing something quite different from a prize that funds twenty more therapies — and it is the part of the announcement that deserves the attention. Three of the ten awardees are worth naming, because one of them is further along than the others: 🧬 **Minicircle Inc., USA** — plasmid gene therapy, and the only one of the three with a registered combination trial already completed **Longeveron, Inc., USA** **RETRO-EPIGERNA, China** All twenty finalists are working toward the same endpoints: muscle, cognition and immune function. ## Fourteen healthy adults and one methylation readout Minicircle is the most concrete case in the field, because it has already published the raw material other teams are still theorising about. The Austin-based company works on non-viral, plasmid-based delivery, and its combination trial administers follistatin and klotho as plasmids. Its lead asset, FST-344, is a follistatin gene therapy; a klotho programme sits behind it. The published trial record is short. A follistatin gene therapy study run at the GARM Clinic in Próspera ZEDE, Roatán, Honduras, under Global Alliance for Regenerative Medicine ethics oversight, is registered as NCT06411366 and was sponsored by Minicircle. That study is not a footnote — it is one of the interventions inside the TranslAGE dataset. The follow-on combination trial is NCT07285629: an interventional Early Phase 1 pilot in healthy adults aged 50 to 80, delivering follistatin plus klotho as a non-viral plasmid by subcutaneous injection into abdominal fat, administered outside the United States. Actual enrollment was 14\. Study start 16 December 2025, primary completion 30 April 2026, completion the same day. The primary outcome list includes epigenetic biological age derived from whole-blood DNA methylation, measured by bisulfite sequencing and computed with validated algorithms, reported as change from baseline. Physical function, cognitive function, kidney function, body composition and subjective wellbeing are named alongside it. ⚠️ **Read the design before reading the ambition** Fourteen participants, an Early Phase 1 label, and no placebo arm. The register claims — improve physical function, cognitive function, kidney function, epigenetic age — are stated as potential, not results. Nothing published on this programme demonstrates efficacy in humans for FST-344 or for the follistatin and klotho combination. A pilot of this size can tell you about safety, dose and whether an assay returns a readable signal. It cannot tell you whether aging moved. ## Sixteen clocks, and the one that moved Here is where the TranslAGE result bites. A biomarker is only useful as a surrogate endpoint if it moves in response to a real intervention, and the analysis found that most of the standard panel fails that test — while a specific subset passes it. ### Gen 1: trained on chronological age The first generation of clocks — Horvath1, Horvath2 and Hannum — was trained to estimate how old a person's tissue looks in years. That was a genuine scientific advance. It is also the version of the idea most people outside this field have in mind when they hear the word clock. It is also the weakest choice for measuring an intervention, because the thing you are trying to move is not age in years. It is rate of decline. In the harmonised re-analysis, the reliable first-generation clocks moved sporadically. They ticked up or down in a handful of individual interventions and showed no consistent direction across the dataset. Exactly one of them held up everywhere. PCHorvath1 decreased significantly across all interventions, with a mean effect of −0.06674 at P = 0.029\. One clock, one small mean effect, one p-value just under the conventional threshold. A single non-reliable second-generation biomarker, PhenoAge, also fell, at a mean of −0.06735 and P = 0.0408 — and that result did not survive correction for multiple testing. ### Gen 2 and beyond: trained on mortality and pace of aging The generation that followed was trained on harder targets. PhenoAge, GrimAgeV1 and DunedinPoAm38 were built against mortality risk and physiological decline rather than against the calendar. The updated set extends that lineage: PCHorvath1, PCHorvath2, PCHannum, PCPhenoAge, PCGrimAge, GrimAgeV2 and DunedinPACE. The result is the cleanest finding in the paper. Every reliable second-generation-or-newer biomarker showed a significant decrease. Clocks trained on mortality or on pace of aging also agreed with each other, which matters more than any individual p-value: converging instruments that respond in the same direction are evidence of a real underlying shift, while a single clock moving alone is indistinguishable from noise. ### Where the aggregate result hides the detail The aggregate also hides how specific the responsiveness is. Alongside it, the paper picked out individual interventions and reported which clocks moved. Across the seven it examined, these are five of them: | Intervention | Responsive clock | How the paper classifies it | | ----------------------------- | ----------------------------------- | -------------------------------------- | | **Smoking cessation** | ◐ SystemsAge | Gen 2+ reliable | | **Metformin** | ✔ SystemsAge, PCGrimAge, PCPhenoAge | Gen 2+ reliable | | **Hyperbaric oxygen therapy** | ◐ DunedinPACE | Gen 2+ reliable | | **Vegan diet** | ✔ DunedinPACE, SystemsAge | Gen 2+ reliable | | **Gastric bypass** | ✗ none of the Gen 2 reliable clocks | Metabolic score fell, effect size 0.43 | Selected rows from the published intervention analysis. ✔ multiple clocks responded, ◐ one, ✗ none. The paper classifies SystemsAge and DunedinPACE as explainable generation-X clocks, and counts them here alongside the generation-2 reliable set. Gastric bypass moved a metabolic system score without moving any general reliable clock · *Nature Medicine*, 2026 Gastric bypass is the instructive row. It produced a clear effect on a metabolic system score, and no movement at all in the general reliable clocks. Gastric bypass did not fail. What the row demonstrates is that "biological age" decomposes into distinct systems, and that a therapy acting on one of them may be invisible to a whole-body clock. | Dimension | Gen 1 clocks | Reliable Gen 2+ clocks | | --------------------------------- | -------------------- | --------------------------- | | **Training target** | ✗ Chronological age | ✔ Mortality, pace of aging | | **Response across 51 studies** | ✗ Sporadic, no trend | ✔ Significant decreases | | **Agreement between clocks** | ✗ Divergent | ✔ Consistent | | **Suitability as trial endpoint** | ✗ Unsupported | ◐ Supported, still unproven | Where the two clock families stand after the TranslAGE re-analysis. The last row is the open question, not a settled one · Nature Medicine, 2026 Independent work points the same way. A study published in *NPJ Aging* in July 2026 paired methylation and RNA sequencing data from the US Health and Retirement Study and reported that the most widely used clocks capture substantially different biological processes rather than one shared axis, with scores built from the genes behind each clock sometimes associating more strongly with age-related morbidity and mortality than the clocks themselves. Five clocks, five partial views. A trial that measures six of them and reports the most flattering one is not measuring aging. ## Why the training target decides everything This is not a laboratory quibble. The choice of training target determines whether a trial can be powered, and therefore whether an expensive trial is worth running at all. Suppose a therapy produces a genuine but modest shift in mortality risk. A clock trained on chronological age will barely register it, because chronological age does not change. A clock trained on mortality risk will register it, because the underlying quantity moved. Same intervention, same blood sample, same lab. The difference in measured effect size is not scientific — it is an artefact of which reference the algorithm was fitted against. Run twenty trials on the wrong ruler and you get twenty null results, each of which will be reported as a therapy that does not work. Promising interventions, underpowered cohorts, effect that evaporates at scale: this field has seen that pattern before, whatever the cause. The plausibility shortcut here — assuming it works in mice, so it will probably read out in humans — has cost this field more than any single failed molecule. There is a second-order problem: multiple testing. The PhenoAge result is the clean demonstration. It looked significant at P = 0.0408 and then stopped being significant the moment the analysis accounted for the number of things it had measured. Any multi-team competition sharing one biomarker pipeline will hit the same wall, which is a further argument for the University of Utah coordinating the statistics rather than twenty teams each running their own analysis. ## Is healthspan even the right score? [As we wrote in September](https://nexi.fund/slug/insilico-ai-longevity-2026/), when six aging clocks agreed on Insilico's AI drug, the hard endpoint was still missing. Five weeks before that, a *Nature Medicine* commentary had put the missing endpoint at the centre of the analysis. The measurement problem is now the field's primary scientific problem, and it has a documented failure mode rather than an open debate. Still, some researchers argue the metric itself is unsound. Writing in *Fight Aging!* on 24 September 2026, they argued that healthspan may not be a robust way to measure progress in rejuvenation biotechnology at all — that a composite summary of morbidity and independence can move for reasons unrelated to the underlying biology being targeted. The longer-standing version of the objection, from geroscience trial-methodology work, is that a surrogate marker earns its place only when a change in the marker reliably predicts the magnitude of change in the clinical outcome. Responsiveness is necessary. It is not sufficient. Capital has not waited for the debate to resolve. Reporting compiled by Longevity.Technology and circulated through The Pharmaletter puts investment in longevity biotech at a record $18.4bn in 2025, against $4.7bn in 2024, with $12.1bn already committed by September 2026\. Commentary in *Al Jazeera* in late September argued the boom is running ahead of the science. The consensus-forming effort is now visible as infrastructure. The Biomarkers of Aging Consortium — academic, clinical, industry, regulatory and philanthropic participants — convenes its fourth annual conference on 5 and 6 October 2026 at the Joseph B. Martin Conference Center at Harvard Medical School, with the stated purpose of developing, validating and applying biomarkers of aging so a therapy that changes biological aging can actually be shown to have done so. So the honest summary is two-sided. The measurement instruments are better than their reputation, because the reliable second-generation clocks do respond and they do agree with each other. They are also unproven as regulatory-grade surrogate endpoints, they decompose into systems that a single whole-body score cannot see, and twenty finalists are about to run up to one-year trials against them in adults aged 50 to 90. The XPRIZE decision in 2030 is supposed to produce a winner. Whether the field believes it will turn on the assay rather than the therapy. 📊 **Signals worth tracking** Which clock panel the University of Utah Data Coordinating Center mandates for the 2026-2029 finalist trials — and whether it is restricted to reliable second-generation clocks Whether Stein Institute shared biomarker protocols are published before the first finalist readout, or after Any pre-registration of primary endpoints across the 20 finalist teams, which would prevent post-hoc clock selection The biomarker validation outputs from the Biomarkers of Aging Consortium meeting on 5-6 October 2026 [ Responsiveness of epigenetic aging biomarkers to longevity interventions in humans The TranslAGE database: 51 harmonised interventional studies, 16 clocks, 94 additional methylation biomarkers. The paper every claim in this piece about clock responsiveness comes from. Nature Medicine — Sehgal, Borrus, Higgins-Chen ](https://www.nature.com/articles/s41591-026-04562-9?ref=nexi.fund) Read this before quoting any aging-clock number, including the ones here. The companion News & Views by Steve Horvath is two pages and states the surrogate-endpoint standard more clearly than the analysis does. [ NCT07285629 — Safety and Efficacy of Klotho and Follistatin Gene Therapy Early Phase 1, non-placebo-controlled, 14 healthy adults aged 50 to 80, completed 30 April 2026, epigenetic biological age as a primary outcome. ClinicalTrials.gov — sponsor Minicircle ](https://clinicaltrials.gov/study/NCT07285629?ref=nexi.fund) The trial registration, not the company page, is the source for enrollment, dates and endpoint list. A completed 14-participant pilot answers safety and assay questions. It does not answer the efficacy question. [ 10 Finalist Teams Awarded $1M to Advance in XPRIZE Healthspan 20 finalists announced in Salt Lake City on 11 August 2026, 10 awarded $1M each, coordinated trials in adults aged 50 to 90 running 2026-2029, up to $81M awarded in 2030. XPRIZE Foundation ](https://www.xprize.org/news/10-finalist-teams-awarded-1m-to-advance-in-xprize-healthspan-101-million-race-to-transform-healthy-aging?ref=nexi.fund) The prize structure is the story: $10M of Milestone 1 payments, $10M of Milestone 2 payments, and a single shared trial infrastructure that forces twenty teams onto one biomarker pipeline.