204 - Centenarians, metformin, and longevity | Nir Barzilai, M.D.

Barzilai's empirical compliance strategy: he directs interested patients to send their physicians his two Cell Metabolism papers (2016 mechanism paper, 2020 TAME clinical rationale). In his experience, 100% of physicians who read either paper then prescribe. For the trial, metformin's side-effect-sensitive patients will be tracked as a sub-cohort to look for unique genetic signatures explaining GI sensitivity.

Attia used his own elevated fasting lactate as the deciding signal to stop metformin three years before this episode. The association of higher fasting lactate with worse health at the population level is the same mechanism in reverse: the less healthy an individual is, the higher their resting lactate. Barzilai first described lactate rising universally in metformin users in his 1987 fellowship work with Ralph DeFronzo at Yale.

The offspring data is the key bridge: not only centenarians but their adult children in their 60s-70s also show elevated TSH. This rules out the alternative explanation that TSH rises in centenarians because of age-related thyroid dysfunction — if the offspring also have it, it is more likely a heritable protective variant than tissue failure. Barzilai hopes the offspring longitudinal study will eventually show whether TSH was elevated in centenarians even in their 30s-40s.

Attia stopped metformin three years before this episode, citing elevated fasting lactate as a signal of impaired mitochondrial efficiency. Since then, multiple studies have confirmed the VO2 max blunting and the mixed-but-not-null signal on muscle. The nuanced position: metformin blunts mTOR (bad for hypertrophy), reduces inflammation (good), induces autophagy (good), and has a unique GDF-15 mechanism not replicated by exercise. For most exercising longevity patients, exercise already covers the autophagy and inflammation territory; what is unique to metformin may not be worth the VO2 max cost.

Attia frames this as the core clinical philosophy: secondary prevention is not prevention. Statins after a heart attack, chemo after cancer — these are treatments. The centenarian's lesson is that their aging was slower throughout life, not that they got better treatments afterward. The economic argument (Andrew Scott, Nature paper): a healthy older adult generates economic value through continued productivity and spending, not just by avoiding healthcare costs. The combination of lower end-of-life medical costs (1/3 of dying-at-70 costs) plus extended productive years makes the longevity intervention a net economic positive.

Barzilai's systems biology approach: he uses computational biology to model every bad genotype in interaction with every longevity genotype. The Lp(a)+CETP example: centenarians who have high Lp(a) are almost uniformly also homozygous for a CETP variant that provides strong atherosclerosis protection. This counteracts the Lp(a) risk entirely — and may even allow a net benefit from Lp(a) (which may have infection-fighting properties). Without the CETP, the Lp(a) would likely have killed them before 85.

The twin astronaut (Scott vs Mark Kelly) natural experiment supports this: the ISS astronaut showed large methylation clock changes during his year in space, but they largely reverted after return — suggesting clocks measure environmental stress responses, not underlying biological aging rates. By contrast, the proteomics breakdown signature (collagen, ECM, platelet granulation) is structural and would not revert quickly with dietary manipulation.

Barzilai's team sequenced their first 44 centenarians against ClinVar's 15,000 disease-associated variants. Each centenarian carried only five to six bad variants out of 15,000 — but none manifested those diseases. The perfect-genome hypothesis (absence of bad genes) was partly falsified by the two APOE4/APOE4 centenarians. The correct model is that positive longevity genes (APOE2, CETP, GHR deletion, FOXO) actively override risk variants, not merely their absence.

The mechanism: the exon-3 deletion receptor has lower baseline phosphorylation activity, explaining lower IGF-1 at rest. But on GH exposure it phosphorylates threefold more than wild-type. During puberty — when GH surges — this amplification drives taller-than-normal growth. Post-puberty, with GH declining, the low-baseline phenotype dominates and IGF-1 stays low for decades. This elegantly explains how longevity-promoting low-IGF can coexist with normal or even above-average height.

The phenomenon is classic antagonistic pleiotropy: the same gene variant (or hormone level) is selected for early-life benefit and tolerated despite late-life harm. In the case of IGF-1, muscle growth and tissue repair in youth are net beneficial; the same anabolic signaling in an aged organism with accumulated cellular damage promotes senescent and cancerous cell proliferation. This is why Barzilai's earlier centenarian data showed low IGF-1 among the oldest-old: those with high IGF-1 were selected out by disease and death before 100.

Key design choices: (1) 65-79 age range — young enough that disease hasn't already dominated the biology, old enough that 5-year event rates justify powering the study. (2) Inclusion requires some evidence of aging phenotype (e.g., gait speed under a threshold), to exclude super-healthy future-centenarians who wouldn't generate enough events. (3) The composite endpoint gives one point per event regardless of type, avoiding early stopping for any single disease crossing significance. (4) The trial also banks plasma, DNA, and cells for proteomic and omic sub-studies — the NIA separately granted an additional omics supplement on top of the $50M. (5) Metformin is started at small dose with food after the first bite to minimize GI side effects.

The mechanism for VO2 max blunting is metformin's weak mitochondrial complex-1 inhibition. GDF-15 (a cytokine that rises 3.5-fold on metformin) is proposed as both a compliance biomarker and a candidate mediator of some of metformin's geroprotective effects — it is also a nausea signal, which may partly explain why metformin-takers inadvertently eat less. For highly active individuals, the key issue is that exercise already provides most of the same transcriptomic benefits metformin provides (autophagy, anti-inflammatory pathways) through different mechanisms. The muscle-mass loss from mTOR suppression matters more for people who are already near the physiological ceiling for hypertrophy.

Attia's critique of methylation clocks: fasting glucose and vitamin D both factor into some clocks, making them trivially manipulable day-to-day. Barzilai's more measured response: the twin astronaut experiment (Scott Kelly vs Mark Kelly, one year in space) showed methylation clocks are sensitive to environmental stress but mostly revert — suggesting they are not measuring stable biological age. The proteomics finding that the two most significantly changed proteins (at p=10^-80) extend lifespan when expressed in animals provides a causal anchor that clocks lack. The breakdown signature is also actionable: stop the breakdown = biomarker improvement = treatment signal within months.