The Biology of Aging
Aging has a dominant scientific framework — twelve interacting processes that go wrong over a lifetime — and it is far better at describing what breaks than at telling anyone what to do about it. This section walks through each process and keeps those two questions apart, because their answers come apart badly.
Read this section for why things work and as a filter for claims, not for things to take: no drug or supplement aimed at a single process of aging has yet been shown to slow aging in healthy people. The best-supported habits — training, a Mediterranean-style diet, sleep, not smoking — are covered in the other pillars and gathered in the healthspan long list; the exposures and markers that sit between biology and behaviour are under Foundations. Most of the evidence here comes from cells, animals and human observational studies; the human trials are few, mostly small, and mostly measured a blood marker rather than health.
The gap the field is trying to close
People now live far longer than they live well. World Health Organization data from 183 member states put the average gap between healthy life expectancy and total life expectancy at 9.6 years, with the United States widest at 12.4 years and women's gap 2.4 years larger than men's; across the two decades measured, the gap grew.[1]
Lifespan is how long you live; healthspan is the years lived free of chronic disease, cognitive decline and disability. Geroscience, the research programme behind this section, treats biological aging itself as the main risk factor for the diseases that fill the gap — cardiovascular, neurodegenerative, cancer, metabolic — rather than treating each separately.[2] If the underlying processes can be slowed, several diseases are slowed at once.
The stated goal is to compress morbidity: shorten the period of illness and dependence at the end. A diet or training programme that buys five more good years before the first chronic diagnosis does more than a drug that adds five years at the tail end. The pages below are mostly an account of how far the evidence gets toward that promise.
Twelve processes: a good map, a contested theory
The dominant framework is the hallmarks of aging — originally nine processes, twelve since 2023.[3] They sort into three tiers: primary hallmarks, the root damage; antagonistic hallmarks, responses that start out protective and turn harmful; and integrative hallmarks, the body-wide result once damage outruns repair. They feed each other: short telomeres can push cells into senescence, which drives inflammation.
As a description, most of it is well established across species and tissues: DNA damage accumulates, chromosome caps shorten, the marks that switch genes on and off drift, recycling slows, damaged cells linger instead of dying, regenerative reserve drains, and a low-grade inflammation rises. Three hallmarks — failing protein quality control, breakdown in signalling between cells, and drift in the gut microbes — are only moderately established even as descriptions.
Whether any of the twelve cause aging rather than accompany it is a further step, and which are drivers rather than passengers is openly contested among the framework's own authors. A survey of dozens of aging biologists found no majority view on what aging is, what causes it, or when it begins.[4] The clearest causal evidence is for senescent cells, in mice: injecting about half a million of them into young mice slowed their walking for months, and in older mice shortened their lives.[5]
Several of the framework's authors hold equity in companies built around single hallmarks — a telomere-measurement firm, senescence and reprogramming ventures — all disclosed in the 2023 paper. That is context for confident claims about particular hallmarks, not grounds to discount the map.
The same authors also describe hallmarks of health, what working biology looks like,[6] later extended to psychosocial adaptation.[7] Loneliness, chronic stress and lack of purpose go with immune dysregulation and higher mortality. Pooling 148 cohorts, people with strong social relationships had about 50% greater odds of surviving follow-up, comparable to quitting smoking;[8] how much of that is causal rather than a marker of underlying health is unsettled — see Purpose.
The practical use of the map is as a filter. When someone sells an intervention on the strength of a hallmark, ask which question they are answering: that the process is real, or that their product changes it. Those are almost never the same evidence.
Why the levers that work, work
Two ideas do most of the explaining, and both cut against the intuition that more of a good thing is better; the mechanisms are well worked out, but the step from them to slower aging is inference.
The first is hormesis: a brief, controlled dose of stress triggers an adaptive overcompensation, while the same stress sustained is just damage. A hard training session — and, it is thought, a fast, a sauna or a cold exposure — briefly strains the cell's mitochondria, its power plants, and the response is to clear the damaged ones and build better ones.[9] It also explains why high-dose antioxidant supplements do not slow aging, and why in small trials taken around training they blunted part of what the training does: the stress is the signal. Dose and recovery decide the outcome, which is why unrecovered overtraining pushes the same lever the wrong way.
The second is that the growth–repair switch has a floor as well as a ceiling. The nutrient-sensing pathways, the field's most targeted machinery, follow a U-shape: too much growth signalling feeds metabolic disease and cancer, too little brings frailty, muscle loss and weaker immunity.[10] The aim is to restore the swing between feeding and fasting, not to hold growth down. The right balance changes with age: a moderate, plant-forward intake suits midlife, while past about 65 defending muscle comes first — worked through under Protein.
Trials aimed at a single hallmark have mostly disappointed
Human trials exist for most of the twelve; most measured a blood marker rather than an outcome, and many with proper endpoints missed:
- Urolithin A, sold to help cells clear worn-out mitochondria, missed both primary endpoints in older adults.[11] A maker-sponsored trial in middle-aged adults found a modest strength gain but missed its own primary endpoint.[12]
- Spermidine, a wheat-germ compound that switches on cellular recycling, moved neither memory nor its biomarkers over twelve months in 100 older adults — though at about 0.9 mg a day, a dose its own authors called too low to settle the question.[13]
- A senolytic knee injection — a drug meant to clear worn-out cells — improved knee-osteoarthritis pain no more than placebo, and its follow-up study was stopped early; some reviews still report pain relief that the registry results do not show.[14]
- Nicotinamide riboside, a vitamin B3 relative sold to raise the cellular coenzyme NAD⁺, did not improve muscle stem-cell function in older adults recovering from a muscle injury; it raised NAD⁺ in blood but not in muscle.[15]
Results in worms, flies and mice have repeatedly failed to carry across. The hallmarks hub rates each hallmark on both questions.
Where trying to fix a hallmark can hurt
- Self-administered senolytics. In a 19-person early-phase study with no control group, six months of dasatinib plus quercetin — a leukaemia drug with a plant flavonoid — accelerated several epigenetic-aging clocks in healthy adults and shortened a methylation estimate of telomere length.[16] The two clocks best validated against health outcomes, GrimAge and DunedinPACE, did not move, and two of the accelerated clocks were back to baseline by six months, but two others (Hannum and PhenoAge) were not, and the telomere measure had not recovered. Not proof that senolytics age you; a concrete reason healthy people should not take them.
- Chasing longer telomeres. Genetically longer telomeres lower coronary, kidney and lung-scarring risk, and Alzheimer's risk only very slightly, while raising several cancers including glioma, as well as high blood pressure and multiple sclerosis.[17] The two directions roughly cancel: inherited telomere length showed no clear effect on lifespan.[18] "Longer is better" is false; that does not make short safe.
- Stem-cell clinics. A 2021 census counted more than 2,700 in the United States alone.[19] A Pew Charitable Trusts tally found at least 360 people harmed worldwide between 2004 and 2020 — 20 deaths, nine cases of partial or complete blindness, over 100 hospitalisations,[20] among them three women blinded after a clinic injected fat-derived cells into their eyes.[21]
- Boosting repair machinery. The protein-protection systems that keep healthy cells alive also keep tumours alive: deleting the master heat-shock regulator protects mice from cancers driven by a common growth-signalling mutation, because malignant cells lean on that system to survive their own chaos.[22] This is mechanism, not a documented human harm — which is the point: the boosters are sold before the risk is measured.
Measuring biological age: useful for populations, too noisy for you
Two 50-year-olds can be biologically a decade apart; biomarkers try to tell such people apart. They predict well across populations but are not reliable enough to act on one reading, and on average none yet beats the calendar.
Epigenetic clocks read methylation marks — chemical tags on DNA at thousands of fixed positions — whose pattern changes with age. The first clocks were trained on calendar age and are only loosely tied to health. GrimAge, trained against lifespan, is the strongest single methylation predictor of death from any cause, beating both the first clocks and telomere length.[23] DunedinPACE reports a rate of aging rather than an age — a speedometer, not an odometer — which makes it the most sensitive research endpoint for whether an intervention works.[24] Midlife labs treats clocks as interesting but not a decision tool: no standard yet says when one is accurate enough to guide one person's care.
As endpoints, clocks have barely moved. In the CALERIE trial of calorie restriction, the cleanest randomised test, participants achieved about 12% restriction against a 25% target, the pace-of-aging clock slowed by a few percent, and the other clocks did not consistently follow.[25]
Reliability is the larger problem. The same blood sample run twice returns noticeably different ages on the older clocks, enough to swamp the differences a consumer test claims to detect.[26][27] Saliva fed into blood-trained clocks errs badly, there are no inter-lab standards, and no regulator accepts a clock as a clinical endpoint. The same person sampled days apart varies too, DunedinPACE included. If you track one anyway, use the same vendor and tissue every time and read the trend.
Cheaper readouts hold their own. In the US Health and Retirement Study, a score built from seven inflammation blood markers, on its own, slightly out-predicted the six methylation clocks it was compared with for four-year mortality (only age itself did better; once other risk factors were added, score and clocks were a near tie), and still predicted death after adjusting for the clocks, so it reads something they miss.[28] Heart rate variability, the beat-to-beat fluctuation in heart rhythm, is a non-invasive, continuous read on how the nervous system is regulating the body.
Clonal haematopoiesis is the closest thing to a hallmark you can read off a blood test. With age, blood production narrows onto a few expanded clones; when one carries a growth-driver mutation, an ordinary blood sample shows it. Standard sequencing finds it in about one adult in ten past 65 and nearly one in five in their 90s.[29][30] Pooled studies put coronary risk about 75% higher,[31] and in the original data carriers had about four times the risk of an early heart attack after accounting for the usual risk factors, apparently because mutant immune cells stoke inflammation in the artery wall.[32] No treatment has been shown to change those outcomes — measurable is not the same as actionable.
Topics covered in depth
The twelve hallmarks of aging →
The hub: each hallmark rated twice — how well established the process is, and how well established that targeting it helps. Start here; the deep dives below assume its framing.
Primary hallmarks — the root damage. DNA corruption in Genomic instability; shortening chromosome caps in Telomere attrition; drifting gene switches in Epigenetic alterations; failing protein quality control in Loss of proteostasis, the widest gap between mechanism and lever and the thinnest human evidence; and slowing cellular recycling in Disabled macroautophagy.
Antagonistic hallmarks — compensations that turn harmful. Miscalibrated fuel gauges in Deregulated nutrient sensing, failing power plants in Mitochondrial dysfunction, and damaged-but-undying cells in Cellular senescence.
Integrative hallmarks — the systemic result. Drained regenerative reserve in Stem-cell exhaustion, signalling breakdown in Altered intercellular communication, body-wide simmer in Chronic inflammation, and gut-microbial drift in Dysbiosis.
Immunosenescence →
The immune system ages on its own schedule: the thymus gives way to fat from the teenage years, and immune cells protect less while inflaming more. They are renewed by pulses of an immune growth signal and worn out by continuous exposure — which may be why exercise helps, and why drugs built on that signal are cancer therapies, not longevity tools.
Bile, microbiome, and metabolic health →
Bile acids are signalling molecules that gut bacteria modify, at the junction of liver, gut and metabolism — though nothing bile-specific is needed to support the system.
What's overrated
- The hallmarks as a to-do list. The map shows what goes wrong; it does not follow that each item has a lever, and for several it demonstrably does not.
- Clock readings as a personal scoreboard. Real predictive value across populations; weak evidence for changing any individual's decisions; a test-retest problem the marketing rarely mentions.
- Mouse results, generally. Several of the most-cited findings here worked in mice and did nothing measurable in people. Laboratory mice are also housed below their comfortable temperature, so at 22 °C a mouse spends more energy making heat than its whole body uses at rest,[33] which distorts exactly the metabolic and mitochondrial readouts most often reported.[34] That argues for discounting a striking metabolic result, not dismissing mouse longevity work: in one long-lived mutant strain, comfortable housing did not shorten their lives.[35]
- "More is better", in both directions. More autophagy, more proteasome activity and longer telomeres each have documented downsides; so does maximal suppression of growth signalling. The goal is restored regulation, not a maximised dial.
- The drug frontier as imminent. No drug has been shown to extend human life. TAME, Targeting Aging with Metformin, the trial usually cited as proof the field is moving, has never enrolled anyone and has only a fraction of the money it needs — a proposal, not a study in progress.