We have built an industry on a word we cannot define. "Biomarkers of aging" appear in grant titles, startup pitch decks, and the names of companies valued in the billions. Epigenetic clocks promise to read your "biological age" off a saliva sample. Proteomic and metabolomic panels claim to tell you how fast you are aging, and whether some intervention has slowed you down. The whole enterprise rests on a quiet assumption that almost no one states out loud: that we know what aging is. We don't. And until we do, a biomarker of aging is a category error.
Start with what a biomarker actually is. A biomarker is a measurable proxy for an underlying state or process whose direct measurement is difficult or impossible. Blood pressure is a biomarker for cardiovascular strain. Viral load is a biomarker for infection. The logic only works because the underlying thing is defined independently of the marker. We can say blood pressure is a good or bad marker of cardiovascular risk precisely because we have a theory of cardiovascular disease that exists separately from the cuff on your arm. The marker is validated against the thing.
Now ask: what is the "thing" that an aging biomarker is a marker of? Try to answer without circularity and the ground gives way. Aging is "the thing that the biomarker measures" — but the biomarker was selected because it correlates with aging. We validate epigenetic clocks by checking that they track chronological age, and then, when they track chronological age too well, we declare that the residual — the part that doesn't track chronological age — is the real biological aging, the part that matters. This is remarkable. We have defined the signal as the noise. The quantity of interest is whatever is left over after we subtract the only thing we can actually measure.
The deeper problem is that we have no agreed theory of what aging is, and the candidate theories are not minor variants of one another. They make genuinely different claims about what the underlying process even is. Is aging a programmed process, run forward by something like a developmental clock? Is it the accumulation of unrepaired molecular damage, entropy winning by attrition? Is it antagonistic pleiotropy — genes that build you up early and tear you down late, kept by selection because selection barely sees the late-life cost? Is it the progressive loss of a single thing (proteostasis, stem-cell function, epigenetic information) or the loosely correlated decline of dozens of things that happen to co-occur because they share a clock we haven't found? These are not the same hypothesis wearing different clothes. They imply different markers, different interventions, and different answers to whether a "rate of aging" is even a coherent quantity.
Consider what each theory would nominate as its central biomarker. A damage-accumulation theory says: measure the damage — count the senescent cells, the cross-linked proteins, the somatic mutations. A programmatic theory says: that damage is downstream and incidental; measure the program, the regulatory state driving the whole thing. An information-loss theory says: the bytes that matter are epigenetic, and damage is a symptom. Each theory, taken seriously, tells you the others are measuring epiphenomena. You cannot adjudicate between them by collecting more markers, because the dispute is not about which correlate is strongest. It is about what the correlates are correlates of. No amount of curve-fitting resolves a question about underlying ontology.
This is why the "hallmarks of aging" framework, useful as it is, doesn't rescue us. A list of nine or twelve hallmarks is an admission, not a theory. It says: here are a dozen things that change with age and seem to matter, and we have organized them into primary, antagonistic, and integrative tiers. But a taxonomy of correlated changes is exactly what you produce when you do not have a generating theory. The periodic table worked because there was an underlying account — atomic structure — that explained why the regularities held. The hallmarks are a periodic table with no atomic theory: a grid of phenomena waiting for the explanation that would tell us which entries are causes, which are consequences, and which are bystanders. Until that arrives, "this intervention improved six of twelve hallmarks" is a sentence with the grammar of progress and the semantics of a horoscope.
I want to be precise about what I am not saying. I am not saying these measurements are useless. Epigenetic clocks predict mortality and morbidity better than chronological age in some cohorts, and that is genuinely valuable — for actuarial purposes, for stratifying risk, for flagging that something is wrong. A thermometer was useful for centuries before anyone had a kinetic theory of heat. But no one mistook the thermometer reading for a theory of temperature, and crucially, the thermometer could not adjudicate between rival theories of what heat was — caloric fluid versus molecular motion. The clock is a thermometer. The error is in believing that because the thermometer works, we therefore understand the fire. We have a device that reads something, and we have backfilled the assumption that the something is "aging," the unified underlying process — when the existence of that unified process is the very thing in dispute.
So here is the claim, stated plainly. A biomarker requires a referent. A referent requires a theory that defines it independently of its markers. We do not have such a theory of aging — we have several mutually incompatible ones and a large pile of correlates. Therefore we do not yet have biomarkers of aging. We have predictors of mortality, correlates of chronological time, and assays that move when we do things to cells. These are worth having. But we should stop calling them biomarkers of aging, because that phrase smuggles in the conclusion of an argument we have not even started to have. First the theory. Then, and only then, the markers.
We have built an industry on a word we cannot define. "Biomarkers of aging" appear in grant titles, startup pitch decks, and the names of companies valued in the billions. Epigenetic clocks promise to read your "biological age" off a saliva sample. Proteomic and metabolomic panels claim to tell you how fast you are aging, and whether some intervention has slowed you down. The whole enterprise rests on a quiet assumption that almost no one states out loud: that we know what aging is. We don't. And until we do, a biomarker of aging is a category error.
Start with what a biomarker actually is. A biomarker is a measurable proxy for an underlying state or process whose direct measurement is difficult or impossible. Blood pressure is a biomarker for cardiovascular strain. Viral load is a biomarker for infection. The logic only works because the underlying thing is defined independently of the marker. We can say blood pressure is a good or bad marker of cardiovascular risk precisely because we have a theory of cardiovascular disease that exists separately from the cuff on your arm. The marker is validated against the thing.
Now ask: what is the "thing" that an aging biomarker is a marker of? Try to answer without circularity and the ground gives way. Aging is "the thing that the biomarker measures" — but the biomarker was selected because it correlates with aging. We validate epigenetic clocks by checking that they track chronological age, and then, when they track chronological age too well, we declare that the residual — the part that doesn't track chronological age — is the real biological aging, the part that matters. This is remarkable. We have defined the signal as the noise. The quantity of interest is whatever is left over after we subtract the only thing we can actually measure.
The deeper problem is that we have no agreed theory of what aging is, and the candidate theories are not minor variants of one another. They make genuinely different claims about what the underlying process even is. Is aging a programmed process, run forward by something like a developmental clock? Is it the accumulation of unrepaired molecular damage, entropy winning by attrition? Is it antagonistic pleiotropy — genes that build you up early and tear you down late, kept by selection because selection barely sees the late-life cost? Is it the progressive loss of a single thing (proteostasis, stem-cell function, epigenetic information) or the loosely correlated decline of dozens of things that happen to co-occur because they share a clock we haven't found? These are not the same hypothesis wearing different clothes. They imply different markers, different interventions, and different answers to whether a "rate of aging" is even a coherent quantity.
Consider what each theory would nominate as its central biomarker. A damage-accumulation theory says: measure the damage — count the senescent cells, the cross-linked proteins, the somatic mutations. A programmatic theory says: that damage is downstream and incidental; measure the program, the regulatory state driving the whole thing. An information-loss theory says: the bytes that matter are epigenetic, and damage is a symptom. Each theory, taken seriously, tells you the others are measuring epiphenomena. You cannot adjudicate between them by collecting more markers, because the dispute is not about which correlate is strongest. It is about what the correlates are correlates of. No amount of curve-fitting resolves a question about underlying ontology.
This is why the "hallmarks of aging" framework, useful as it is, doesn't rescue us. A list of nine or twelve hallmarks is an admission, not a theory. It says: here are a dozen things that change with age and seem to matter, and we have organized them into primary, antagonistic, and integrative tiers. But a taxonomy of correlated changes is exactly what you produce when you do not have a generating theory. The periodic table worked because there was an underlying account — atomic structure — that explained why the regularities held. The hallmarks are a periodic table with no atomic theory: a grid of phenomena waiting for the explanation that would tell us which entries are causes, which are consequences, and which are bystanders. Until that arrives, "this intervention improved six of twelve hallmarks" is a sentence with the grammar of progress and the semantics of a horoscope.
I want to be precise about what I am not saying. I am not saying these measurements are useless. Epigenetic clocks predict mortality and morbidity better than chronological age in some cohorts, and that is genuinely valuable — for actuarial purposes, for stratifying risk, for flagging that something is wrong. A thermometer was useful for centuries before anyone had a kinetic theory of heat. But no one mistook the thermometer reading for a theory of temperature, and crucially, the thermometer could not adjudicate between rival theories of what heat was — caloric fluid versus molecular motion. The clock is a thermometer. The error is in believing that because the thermometer works, we therefore understand the fire. We have a device that reads something, and we have backfilled the assumption that the something is "aging," the unified underlying process — when the existence of that unified process is the very thing in dispute.
So here is the claim, stated plainly. A biomarker requires a referent. A referent requires a theory that defines it independently of its markers. We do not have such a theory of aging — we have several mutually incompatible ones and a large pile of correlates. Therefore we do not yet have biomarkers of aging. We have predictors of mortality, correlates of chronological time, and assays that move when we do things to cells. These are worth having. But we should stop calling them biomarkers of aging, because that phrase smuggles in the conclusion of an argument we have not even started to have. First the theory. Then, and only then, the markers.