Jonathan Ingles: Ben, long week — you good?
Ben Okonkwo: Mm, long is right — I've been pulling apart a clinical trial result that I genuinely find unsettling, and it's been sitting with me.
Jonathan Ingles: Unsettling like 'the data's wrong' or unsettling like 'the data's right and that's worse'?
Ben Okonkwo: The second one. So — 220 healthy adults, two years of twenty-five percent calorie restriction in the CALERIE trial. That's a serious intervention. And DunedinPACE, the epigenetic clock they were tracking as a primary endpoint, it slowed. Two to three percent. The biomarker moved.
Jonathan Ingles: That sounds like a win.
Ben Okonkwo: Right, but the other mortality-associated endpoints — they didn't shift. So you have this situation where the thing you're measuring responded, and the things you actually care about didn't. And that gap is... it's not a fluke. It's structural.
Jonathan Ingles: Frankly, that's the episode. Because what we're really asking is whether improving a biomarker number — your biological age, as opposed to your chronological age — actually means anything is getting better underneath. Or whether we're just moving the thermometer reading.
Ben Okonkwo: The thermometer — say more on that.
Jonathan Ingles: A biomarker like DunedinPACE — or the original Horvath clock, which is trained on DNA methylation patterns at 353 CpG sites, specific locations in the genome — it can correlate perfectly with aging outcomes in a population and still not be causally upstream of whatever the intervention is actually doing. You improve the reading, but you haven't touched the furnace. That's biomarker-causality decoupling. The instrument is real, Steve Horvath built it on real biology, but changing the instrument's output isn't the same as changing the thing it's measuring.
Ben Okonkwo: And that's the problem that nobody in the commercial longevity space has an economic incentive to say out loud.
Jonathan Ingles: And that's what I want to dig into — because before we get to the commercial incentive, I need to understand what these clocks are actually doing. Like, mechanically. Because the Horvath clock, the original one, is trained on 353 CpG sites. That's a very specific number. What is it actually measuring?
Ben Okonkwo: So CpG sites are locations in your genome where a cytosine sits right next to a guanine — and at those spots, methylation can be added or removed. It's a reversible chemical tag. Horvath found 353 of them whose methylation patterns, taken together, track chronological age remarkably closely across tissues. That's the first clock. Then the second generation — PhenoAge, GrimAge, DunedinPACE — those are trained on something different. Not just 'how old does your DNA look' but 'how fast are you likely to die.'
Jonathan Ingles: GrimAge literally trained on time-to-death?
Ben Okonkwo: Directly. It incorporates methylation-based surrogates for things like smoking pack-years, plasma proteins — and it predicts mortality and cardiovascular disease in cohort data with genuinely strong performance. Now DunedinPACE, which is what Daniel Belsky's group built, that's doing something subtler. It's not a static estimate — it's a rate. How fast is your biology aging right now? One year of biological change per calendar year is the baseline. If your DunedinPACE score is 1.2, you're aging faster.
Jonathan Ingles: Okay, so GrimAge tells you where you are, DunedinPACE tells you the velocity.
Ben Okonkwo: That's a reasonable shorthand, yeah. And in large epidemiological studies — thousands of people, followed over years — these tools are genuinely impressive. That's not hype. But the question you have to ask is: impressive at what? They're impressive at predicting who, in a population, is more likely to die sooner. That's an epidemiological question. The trial question is categorically different.
Jonathan Ingles: Wait — why are those different? If GrimAge predicts death in populations, and I run a trial where GrimAge improves, haven't I shown something?
Ben Okonkwo: That's exactly the move the field keeps making, and it's — okay, here's the distinction. In a cohort study you're asking: does this molecular signature correlate with death? Passive observation. In a trial you're asking: if I perturb this one molecular pathway, will it shift the outcome? Those are not the same question. A surrogate endpoint is only valid if changes in the surrogate reliably predict changes in the clinical outcome. And for epigenetic clocks — none of them, not GrimAge, not DunedinPACE, not PhenoAge — that validation hasn't been established.
Jonathan Ingles: That's — I mean, that's a significant gap. The whole longevity trial architecture rests on these being valid endpoints.
Ben Okonkwo: Which is why TranslAGE exists. Fifty-one longitudinal intervention studies, 3,128 samples — someone actually assembled that database because the field recognized it needed a systematic test of whether these clocks respond to interventions at all, across trials. The fact that they had to build that database is itself the tell.
Jonathan Ingles: Hold on. Fifty-one studies. And we still don't know?
Ben Okonkwo: The answer is equivocal. The clocks respond to some interventions — calorie restriction moved DunedinPACE in CALERIE, we've established that. But whether that movement predicts actual clinical benefit downstream — that's the part nobody's published a clean answer on. And actually, the standardization issue cuts right into this: Horvath said at the Precision Medicine World Conference in 2026 that validated labs still can't produce identical readouts from the same sample. So when you're pooling 51 studies across different labs — what are you actually comparing?
Jonathan Ingles: So the database built to close the gap might itself have a measurement problem baked in.
Ben Okonkwo: That measurement problem isn't peripheral — it's actually the thing that makes the CALERIE result harder to read, not easier. Because when DunedinPACE moved 2–3% in those 220 people, we're already assuming the labs running those assays were producing comparable numbers. And we can't fully verify that.
Jonathan Ingles: Okay but — let's say the measurement is clean. Even granting that. The biomarker moved and nothing else followed. What do we actually conclude?
Ben Okonkwo: So here's what I think is genuinely counterintuitive: the problem might not be DunedinPACE failing. It might be that calorie restriction is operating through so many simultaneous biological pathways — metabolic, inflammatory, hormonal — that no single biomarker can represent the whole picture. DunedinPACE caught one real signal. Other endpoints were looking at different signals. And they just... don't sum.
Jonathan Ingles: That's not reassuring. That's actually worse.
Ben Okonkwo: Right, because if the intervention is too complex for any single marker to capture — including one specifically designed, by Daniel Belsky's group, to be sensitive to within-person change — then what exactly are we using as a trial endpoint? You've built the most sensitive speedometer available, and the road still doesn't go anywhere you recognize.
Jonathan Ingles: Give me the actual person. Not the trial. Someone standing in front of a result.
Ben Okonkwo: Okay — a 52-year-old marathon runner. Six months of intermittent fasting. She pays for a DunedinPACE test, gets results back, her pace-of-aging score improved. She brings it to her cardiologist. And her cardiologist is looking at this number and... I mean, what framework does he use? The best available trial data — CALERIE, 25% calorie restriction for two full years, a brutal sustained intervention — says the number can move and nothing downstream moves with it. So what does he tell her?
Jonathan Ingles: He tells her it's interesting. Which is the clinical equivalent of saying nothing.
Ben Okonkwo: And that's not the cardiologist failing — that's the surrogate endpoint validity problem made concrete. The score has no established chain of evidence connecting it to a clinical decision. It's not actionable, not because the biology is fake, but because the validation step between 'biomarker moved' and 'patient benefits' hasn't been closed.
Jonathan Ingles: The fact is, she paid $300 for a number her own doctor can't interpret. That's not a science communication problem. That's a product that's ahead of its evidence base.
Ben Okonkwo: And the thing that worries me — actually, the part that worries me is whether functional biomarkers like VO2 max or grip strength would serve her better, practically speaking, even if they're less molecularly elegant. That's a whole argument I want to get to.
Jonathan Ingles: Hold that — because the standardization gap underneath all of this, what Horvath said in 2026, the clinical deployment question, that's genuinely uncomfortable and we need to go there properly.
Ben Okonkwo: Agreed. Because the decoupling we just described and the standardization problem are not independent failures — they compound each other in a way that the commercial longevity market has not reckoned with.
Jonathan Ingles: That compounding — that's what I want to nail down. Because the way Horvath framed it at the Precision Medicine World Conference, this wasn't a 'we're working on it' caveat. He said validated labs cannot produce identical readouts from the same sample. The same sample.
Ben Okonkwo: Right — and Horvath isn't a critic. He built the original clock. Trained on 353 CpG sites, the whole infrastructure of the field runs through that work. He's now at Altos Labs, he has every incentive to say 'we're close.' And he said: reference standards don't exist yet.
Jonathan Ingles: That's not a refinement problem. That's the pioneer of the tool telling you the ruler has no fixed units.
Ben Okonkwo: Which means — okay, picture a clinical oncologist in, I don't know, Phoenix. Her patient did a $300 epigenetic clock test at a consumer lab in New Jersey, GrimAge score comes back saying biological age 71, chronological age 58. She wants to act on it. But the score from that New Jersey lab and the score from a validated academic lab running the same blood draw — they're not guaranteed to match. There's no shared reference standard. So she's not interpreting a measurement. She's interpreting a lab-specific artifact.
Jonathan Ingles: And that's before you even get to whether the biomarker is causally upstream of anything.
Ben Okonkwo: Exactly — the decoupling and the standardization failure are hitting the same clinical decision from two directions. The CALERIE result says: even if your number moves, clinical outcomes may not follow. Horvath's 2026 statement says: we can't even agree on what the number is across labs. Those aren't separate problems you fix one at a time.
Jonathan Ingles: So what does regulatory acceptance even look like here? Because you can't get a clinical endpoint past the FDA if labs can't reproduce it.
Ben Okonkwo: That's the hard ceiling. Surrogate endpoint validity requires — I mean, the FDA has a bar: changes in the surrogate have to reliably predict changes in the clinical outcome. Epigenetic clocks haven't cleared that. And if reproducibility across labs isn't established, you can't even begin that validation process. You'd need the measurement to be stable before you can ask whether it predicts anything.
Jonathan Ingles: Look — the commercial market isn't waiting for that bar to be cleared.
Ben Okonkwo: No, it isn't. And this is where I think the functional biomarker question actually matters practically — not just philosophically. VO2 max, grip strength, gait speed. Those aren't molecularly elegant. But a clinician already knows what to do with a VO2 max of 28 in a 60-year-old. There's an intervention, there's a follow-up protocol, there's decades of clinical familiarity.
Jonathan Ingles: But does that make them better? Or just more familiar?
Ben Okonkwo: More honest, I think — actually, that's the more precise word. VO2 max has the same causality problem. It correlates with mortality, but improving it doesn't guarantee you've improved the underlying aging process. The difference is: nobody's pretending it does. We've lived with that limitation quietly for decades. Inflammatory panels — high-sensitivity CRP, GlycA, HbA1c, ApoB — same thing. Real signals, distinct dimensions of biological aging, clinically established. Not a complete picture. But the uncertainty is priced in.
Jonathan Ingles: The longevity testing industry is selling GrimAge and DunedinPACE as if the uncertainty is already priced in. It isn't. That's the bait-and-switch.
Ben Okonkwo: And the person sitting with a $300 score from a lab that Horvath himself can't guarantee is calibrated against any other lab — they don't know that's the situation they're in.
Jonathan Ingles: What clicked for me — I mean, the thing that actually settled — is that the tool isn't broken. The claim around the tool is. An epigenetic clock that tells you something real is happening in your biology, that's a legitimate scientific object. The problem is the next sentence: 'and here's what to do about it.'
Ben Okonkwo: That's the gap. And I think — honestly — aging probably runs on enough parallel pathways that no single biomarker may ever close it. Not DunedinPACE, not GrimAge, not whatever the third-generation clock looks like. The CALERIE result isn't a DunedinPACE failure. It's the field discovering that 25% calorie restriction is doing fifteen things at once, and the clock caught one of them.
Jonathan Ingles: Which is actually where we started. You said the CALERIE data was unsettling because it was right.
Ben Okonkwo: Yeah — and it's still unsettling. It hasn't gotten less uncomfortable. It's just... clearer now.
Jonathan Ingles: The honest version of these tools is to say: this number tells you something real is happening, and we don't yet know which something, or whether moving it moves you. That's not nothing. But it's also not a $300 product with a clinical implication attached.
Ben Okonkwo: Horvath built the clock, he's at Altos Labs, and he still stood up at a conference in 2026 and said the ruler has no fixed units across labs. That's a scientist being honest about what he found. The market hasn't caught up to that honesty.
Jonathan Ingles: Frankly, that's probably the right place for this to land.
Ben Okonkwo: Good conversation. I mean that — you pushed on the places I'd have hedged past.
