David Sterling: Megan, good to be back — I'll let you lead this one, you've been sitting on something.
Megan Skiendel: I have, yeah. I want to start with a number, actually: fifteen years. That's the onset gap between two sisters — identical twins, same DNA sequence, histologically confirmed Alzheimer's in both — one diagnosed in her late 60s, one not until 83.
David Sterling: And you'd expect what — two, maybe three years of variation at most, given identical genomes.
Megan Skiendel: That was my assumption. Fifteen breaks it completely. And the reason it breaks it is — I mean, this took me a minute to really land on — the DNA sequence isn't the whole story.
David Sterling: The G. Martin paper in PNAS in 2005 established this formally. Identical twins accumulate measurable divergence in DNA methylation over their lifetimes. They start with essentially the same epigenetic state. By their 70s, that's not true anymore.
Megan Skiendel: And DNA methylation — just to put this plainly — it's chemical marks on the DNA that control which genes get switched on or off. The underlying code doesn't change. The volume knobs drift.
David Sterling: That drift is epigenetic drift. Gradual, cumulative, no sequence change required.
Megan Skiendel: Think of it like two identical houses — same blueprint, built the same year. But one sits on a cliff, takes the weather, nobody oils the hinges. The other's inland, better maintained. Seventy years later, same blueprint. Two completely different structures.
David Sterling: Chronological age is the year the blueprint was drawn. Biological age is the condition of the building.
Megan Skiendel: And those two numbers — chronological, biological — they can diverge substantially. For twins they start at exactly the same point. The divergence is the story.
David Sterling: Frankly, that's why the twin setup is the best natural experiment we have for separating what genes determine from what they don't.
Megan Skiendel: Which brings us to the question that I don't think is obvious at all — if the sequence is locked, what's actually driving the drift?
David Sterling: Three things, actually. Not one. The first is epigenetic drift — the slow accumulation of methylation changes at CpG sites across the genome. No sequence change. Just the marks shifting, gene expression profiles pulling apart. Starts post-twinning, accelerates with age.
Megan Skiendel: CpG sites being the specific addresses where the methylation actually lands.
David Sterling: Correct. That's the shared molecular ledger all three mechanisms write into. Now — the second engine is the one that's, I think, underappreciated. Stochastic cellular damage. Random replication errors, oxidative damage, protein misfolding. Completely independent of what you eat or where you live. Molecular coin flips.
Megan Skiendel: Wait — so even if two twins lived together their whole lives, same apartment, same meals, same commute — they'd still diverge?
David Sterling: Yes. Because random errors don't care about the environment. A polymerase makes a copying mistake in one cell, not the other. That's not lifestyle. That's noise baked into the biology.
Megan Skiendel: Honestly, that's the part that quietly demolishes a certain kind of wellness promise. If a meaningful fraction of divergence is just — random molecular damage — you cannot optimize your way out of it.
David Sterling: Well, that's the tension the research doesn't resolve. The 2021 Genome Biology work suggested that for young, healthy twins, methylation divergence may be largely stochastic post-twinning processes — not environmentally driven. So the proportion that's actually modifiable is... contested. We don't know if it's thirty percent, sixty percent.
Megan Skiendel: Which means the third engine — the exposome, all the diet, stress, infection history, movement — that piece could be smaller than the headline version of the story implies.
David Sterling: It could be. The exposome is real — it continuously reshapes the epigenome, that's documented — but how much of the twin divergence it explains versus the stochastic component? Genuinely unresolved. The research flags it explicitly.
Megan Skiendel: So picture a specific case — imagine one twin spends twenty years working night shifts, the other doesn't. The exposome says that stress and circadian disruption should show up in the methylation record. But you're saying even if we controlled for that perfectly, some divergence was already happening anyway, from day one after the egg split.
David Sterling: From the moment the zygote divides, each daughter cell is running its own stochastic error process. The PNAS paper from 2005 — G. Martin — established that divergence is measurable and that it accelerates with age. The early divergence was always there. The exposome layering on top of it makes attribution genuinely hard.
Megan Skiendel: That's the tension I want to come back to. Because if you can't cleanly separate noise from signal in the divergence data, that has real consequences for whether these clocks — GrimAge, PhenoAge — are actually telling you something actionable.
David Sterling: The fact is, the clocks are still better predictors than chronological age even with that uncertainty baked in. That's the paradox — the instrument works at the population level even if we can't fully decompose what's driving the signal.
Megan Skiendel: Right — but that's actually what I want to crack open. Because 'works at the population level' is doing a lot of lifting there. Picture Clara. She's 62, sitting in a waiting room for a standard check-up. Her chart says 62. Routine. But if you ran her blood through a GrimAge analysis, you might find her methylation profile reads closer to 72.
David Sterling: Ten years of epigenetic age acceleration. That's not noise. That's a meaningful mortality signal.
Megan Skiendel: And nobody in that waiting room knows it. Her doctor is scheduling her like a 62-year-old.
David Sterling: That gap exists because — and this is worth tracing — the clocks weren't always built to catch it. Steve Horvath's first-generation clock, the foundational one, was calibrated to chronological age. Trained to say: here's how old your DNA methylation pattern looks, against a population reference. It wasn't designed to predict death or disease. It measured accumulated biological aging. Full stop.
Megan Skiendel: So the Horvath clock tells you *where* you are, not — actually, wait — not *what's coming*.
David Sterling: Correct. Morgan Levine and colleagues changed the target. PhenoAge was trained not on chronological age but on a composite of mortality-linked clinical biomarkers. Outperforms first-generation clocks for predicting all-cause mortality, cancers, Alzheimer's-related outcomes, healthspan. Different objective function, better predictive power.
Megan Skiendel: And GrimAge goes further still.
David Sterling: GrimAge was trained directly on time-to-death outcomes. It incorporates methylation-based surrogates for smoking pack-years, plasma protein levels — so it's essentially reading the biological residue of a lifetime's exposures in the methylation pattern itself. That's why it consistently outperforms first-generation clocks. It's asking the right question: when, not just how old.
Megan Skiendel: So back to Clara. If her GrimAge reads 72 — that ten-year gap carries, what, roughly an 8 to 15 percent higher all-cause mortality risk per five-year unit. Which means — I mean, two units of acceleration, that compounds.
David Sterling: The 8 to 15 percent figure is the systematic review estimate. And the research flags it immediately — study heterogeneity, limited evidence quality, possible publication bias. So take the precision loosely. The direction is robust. The exact coefficient is not.
Megan Skiendel: Which lands differently when you consider the 2023 Journal of Clinical Medicine case — male monozygotic twins, both presenting with acute coronary syndrome nearly simultaneously, same occlusion site, concordant coronary anatomy. Genetics picked the location. But timing and severity? Still subject to non-genetic influences. The clock can't separate those out.
David Sterling: That's exactly where DunedinPACE fills a different slot. It's not measuring where you are — it's estimating the *rate* of aging. Which means if you run an intervention and want to know whether it's actually slowing the process, DunedinPACE is the instrument. Not GrimAge, not PhenoAge.
Megan Skiendel: And the bigger question — which honestly we haven't touched yet — is why Clara's chart still just says 62. Why chronological age is still the operating variable everywhere that matters: insurance, clinical trials, healthcare allocation. That's the part that's going to cost people, and we need to get into it.
David Sterling: And that's not a rhetorical question — it's costing people right now. The status quo isn't neutral. Insurers are already making age-based decisions that misclassify individuals. Clinicians are scheduling screenings on a birth-year calendar. That's not a gap waiting to be filled — that's active misclassification running at scale.
Megan Skiendel: But isn't the counterargument just — the evidence will catch up? Trials take time, adoption takes time.
David Sterling: That's the inertia story. I don't think it's the right one. The liability problem is different — and it's harder. A population-level predictor is not the same as an individual-level verdict. GrimAge says Clara reads 72. Fine. But if her insurer raises her premium on that basis and she dies at 95, who carries that call?
Megan Skiendel: Okay, so — who decides which twin gets into the clinical trial?
David Sterling: That's exactly it. Eligibility criteria are written in chronological age. A 70-year-old whose PhenoAge reads 55 gets excluded from a geriatric intervention. A 60-year-old whose GrimAge reads 75 gets enrolled in a low-risk arm. The clock would sort them correctly. Birth year gets it backwards.
Megan Skiendel: And the 70-year-old who's epigenetically 55 — does she get better rates, or does the underwriter just assume the clock is broken?
David Sterling: Honestly — I think the asymmetry is real. Insurers adopt tools that reduce their exposure. A clock that flags high-risk individuals above their chronological age gets used. A clock that exonerates someone? That's a harder sell internally, because the liability runs the other way.
Megan Skiendel: That's — I mean, that's a governance problem, not a science problem.
David Sterling: Both. The Alzheimer's twin case makes the science case cleanly — a 15-year onset gap in genetically identical individuals. If chronological age carried real predictive weight, that gap is essentially impossible. It exists. So the burden of proof runs both directions: clocks must earn individual-level precision, but chronological age has already failed. We just haven't updated the ledger.
Megan Skiendel: Wait — so who's actually blocking the update? Is it the FDA? Actuarial standards bodies? Because someone is holding the door shut.
David Sterling: All of them, actually — and not cynically. Personalized aging trajectories require individual-level precision, and the clocks aren't there yet for a single-person verdict. A prediction that works across ten thousand people in a cohort — that's a population instrument. Telling Clara specifically that she has two years less than she thinks? The confidence interval on that is still too wide to stake a clinical decision on.
Megan Skiendel: Which means the people who most need recalibration — the ones whose biological age is running ten years hot — are the ones the system is most likely to miss. Right now. Not eventually.
David Sterling: And that's the stochastic argument's sharpest edge, actually. If a meaningful share of the divergence is random molecular noise — not exposome, not behavior — then the ceiling on intervention is lower than anyone selling a longevity protocol wants to admit.
Megan Skiendel: Which is — honestly, that's what strikes me most. Because the counter-argument isn't 'noise doesn't matter.' It's almost the opposite.
David Sterling: Right — if stochastic damage compounds, the logic flips. You're not trying to reverse a decade of accumulated errors. You're slowing the rate before they stack. Early, sustained investment. That's precisely where DunedinPACE earns its place — it measures rate, not position. Am I aging faster or slower than I was six months ago? That's a question it can actually answer.
Megan Skiendel: So the irreducible noise isn't an argument against intervening. It's an argument for intervening earlier, before the noise has time to compound into something you can't walk back.
David Sterling: That's — well, that's the version of the story I can actually defend. Which parts of your trajectory are movable, which aren't — precision medicine doesn't have that answer yet. But the question itself is now askable in a way it wasn't before GrimAge, before PhenoAge, before DunedinPACE existed.
Megan Skiendel: And whether that knowledge feels like freedom or just a more detailed map of a wall — I genuinely don't know which way that cuts for most people.
David Sterling: Mm. The 2005 PNAS paper settled one question cleanly: identical twins prove aging isn't a fixed shared clock. That much is done. Everything downstream of it is still open.
Megan Skiendel: That's the one thing I'll actually take out of this. The clock isn't fixed. That's modest, but it's real. Appreciate the push on the stochastic piece — that's the part I'd have soft-pedaled.
