Your organs age at different rates: what organ-specific aging clocks show

TL;DR

• Organ clocks read organ-enriched plasma proteins - Stanford's 2023 Nature paper measured 4,979 proteins in 5,676 adults and found 893 of them tissue-specific enough to model 11 organs.
• About 20% of people are strongly accelerated in at least one organ, and 1.7% are multi-organ agers; accelerated organ ageing carried 20–50% higher mortality risk.
• It has replicated at scale - UK Biobank, 44,498 people followed up to 17 years, put an especially aged brain at an Alzheimer's hazard ratio near 3.1, comparable to one copy of APOE4.
• Organ ages are not independent dials - Whitehall II found faster ageing of one organ raised susceptibility to morbidity across several.
• An age gap is not a diagnosis or an imaging finding - nothing was visualised, and plasma protein levels shift with inflammation, hydration, medication and recent exercise.
• Cohort association is not individual accuracy - none of these measures is validated for tracking one person's response to an intervention.

---

Educational content, not medical advice. This article contains no doses, schedules or protocols.

Why two biological-age tests can give you two different numbers

Someone arriving at this topic has usually just had the experience: two tests, one body, two answers, several years apart. The instinct is that one of them must be wrong.

That instinct assumes the two tests are measuring the same thing under one name. They are not, and the ways they differ are systematic rather than random.

They use different inputs. One test reads DNA methylation at CpG sites in white blood cells. Another reads the concentration of thousands of proteins in plasma. A third combines routine clinical chemistry - albumin, creatinine, glucose, inflammatory markers - into a single index. These are three different molecular layers of the same person, changing on different timescales in response to different things.

They were trained against different targets, which is the deeper split. The first generation of epigenetic clocks was trained to predict chronological age - the model was rewarded for guessing your birthday correctly from your methylation. The second generation was trained against mortality and phenotypic health measures, so it was rewarded for ranking people by how likely they were to die or fall ill, not by how old they were. A third generation was trained against the rate of multi-system decline measured longitudinally. These are three genuinely different questions. A model optimised to reproduce your birth certificate and a model optimised to rank mortality risk will disagree about you, and both can be behaving exactly as designed.

They were trained in different cohorts. A model fitted in a middle-aged British civil-service cohort and a model fitted in a multi-ethnic American cohort inherit the age structure, disease prevalence and measurement platform of the population they were built in. Applied to a person who resembles neither, they extrapolate.

And they answer with different statistical objects. "Your biological age is 39" and "your pace of aging is 0.94 years per year" are not the same kind of statement, even when both are printed next to a photograph of a mitochondrion.

So the disagreement is not a malfunction. It is what you should expect from two different measurements wearing one name. The useful framing is not which number is right but what was this particular model rewarded for getting right, and is that the thing I care about.

Epigenetic clocks, briefly - what they measure and what they miss

DNA methylation clocks remain the reference point for this entire field, and they are covered properly in epigenetic clocks. The short version is that specific cytosines across the genome gain or lose methyl groups in patterns that track age with striking regularity, and a regression model over a few hundred of those sites reproduces chronological age to within a few years.

Two things they miss matter here.

The first is anatomy. A standard epigenetic clock is usually run on whole blood. Whatever it reports is a single, body-wide summary - and much of it reflects the cells actually being sampled, which are immune cells. If your kidneys are aging faster than the rest of you, a blood methylation clock has no dedicated channel to say so.

The second is reliability, which is less discussed than it should be. Individual CpG measurements carry real technical noise. Work published in Nature Aging in 2022 showed that running the same sample twice through six widely used clocks could produce deviations of up to roughly three to nine years between technical replicates. That is not a small artefact when the entire clinical claim rests on a five-year gap. The same work introduced principal-component versions of those clocks, which brought most technical replicates to within about a year of each other - a genuine methodological advance, and one worth knowing exists, because it means "this clock" and "the reliable version of this clock" are not always the same product.

A 2025 comparison in Nature Communications evaluated fourteen epigenetic clocks against ten-year onset of 174 disease outcomes in 18,859 individuals, and found that the second-generation clocks - those trained against mortality and phenotype - substantially outperformed the first-generation ones for disease prediction. Which is exactly what the training-target argument above predicts.

Proteomic clocks: reading organ age from plasma proteins

The organ-specific approach starts from a straightforward observation. Most proteins in blood come from everywhere. But a subset is organ-enriched: expressed at far higher levels in one tissue than in any other, and detectable in plasma because tissues shed protein into circulation. If you can measure enough of those, you have an indirect window on the state of the tissue they came from.

The anchor paper is Oh and colleagues from Tony Wyss-Coray's group at Stanford, published in Nature in December 2023. The team measured 4,979 plasma proteins in 5,676 adults spanning the human lifespan, across five independent cohorts. Of those proteins, 893 - about 18% - met their definition of organ-enriched, with the brain contributing the largest share. They then trained machine-learning models on the organ-enriched protein subsets to estimate an age for each of 11 major organs, and defined an age gap: the difference between the model's estimate for an organ and the person's chronological age.

The population-level findings were substantial. Roughly 20% of people showed strongly accelerated age in at least one organ; about 1.7% were "multi-organ agers." Accelerated aging of an organ was associated with a 20–50% higher mortality risk in these cohorts. Organ-specific disease tracked organ-specific aging - accelerated heart aging was associated with markedly elevated heart-failure risk, and accelerated brain and vascular aging predicted Alzheimer's disease progression about as strongly as plasma pTau-181, the best-established blood biomarker for that condition.

It has since replicated at much larger scale and on a different measurement platform, which matters more than any single paper.

• **UK Biobank, Nature Medicine, 2025.** Organ ages for 11 organs estimated from 2,916 plasma proteins in 44,498 individuals, followed for up to 17 years. Organ age estimates were sensitive to lifestyle factors and medications and were associated with future onset of heart failure, COPD, type 2 diabetes and Alzheimer's disease. An especially aged brain carried an Alzheimer's hazard ratio of about 3.1 - comparable in magnitude to carrying one copy of APOE4 - while a youthful brain was associated with protection comparable to two copies of APOE2, independent of APOE genotype.
• ***Nature Aging*, 2025.** Organismal and ten organ-specific clocks trained in UK Biobank (n = 43,616) and validated in cohorts from China (n = 3,977) and the United States (n = 800), with cross-cohort correlations of 0.98 and 0.93. Brain aging was the organ most strongly linked to mortality.
• **Whitehall II, The Lancet Digital Health, 2025.** Age gaps for nine organs derived from SomaScan plasma proteomics in 6,235 British civil servants aged 45–69 at sampling in 1997–99, followed for 20 years through national health records. Faster aging of a specific organ generally increased susceptibility to morbidity affecting multiple organs - not only its own.

That last point is worth sitting with, because it complicates the intuitive picture. Organ ages are not independent dials.

What "an older brain than heart" is actually claiming

This is where careful reading earns its keep, because the phrase does far less work than it sounds like it does.

An organ age gap is a statement about where your measured plasma protein profile sits relative to a reference distribution. The model learned what the organ-enriched protein pattern of a 45-year-old typically looks like in the training cohort; your pattern resembles the pattern more typical of a 52-year-old; the output is "+7 years."

That is the entire claim. Specifically, it is not:

• A diagnosis. No organ age gap names a disease, and none of these models were built or approved as diagnostic tests.
• An imaging finding. Nothing was visualised. No structure was examined. A "brain age" from plasma has not looked at your brain.
• A statement that the organ is diseased or damaged. Protein concentrations in blood shift with inflammation, infection, kidney clearance, recent exercise, medication, hydration and time of day. Some of that is signal about tissue state; some of it is the ordinary noise of being a body on a particular Tuesday.
• A measurement of that organ directly. It is a plasma inference about tissue, mediated by whatever governs protein release into and clearance from circulation.

The "brain age vs heart age" comparison people find most compelling - one organ apparently running ahead of another - is also the comparison most vulnerable to this. Two organ models have different protein counts, different training accuracy and different error bars. A three-year gap between them may be a real difference in tissue trajectory, or it may be that one model is simply noisier than the other. Consumer reports rarely show you which.

How well organ ages predict anything - the honest state of the validation

The literature above is good work. It is also answering a narrower question than most readers assume, and the gap between those two things is the single most important idea on this page.

Cohort association is not individual accuracy. A hazard ratio of 3.1 describes how the average future differed between two groups defined by a measurement. It does not tell you the probability that any particular person in the high group develops the disease, and it cannot be converted into one without absolute risks, competing-risk structure and a population that actually resembles the reader. Strong group-level separation is entirely compatible with wide, overlapping distributions at the individual level - and in these datasets, that is the usual situation. This is the same discipline that applies to reading any lab value against a population range, which is why optimal versus normal lab ranges is worth reading alongside this.

Test-retest reliability is the question vendors answer least often. For any longitudinal claim - "your kidney age fell two years" - the relevant number is how much the result moves when nothing about you has changed. For epigenetic clocks that number has been measured and published, and it was uncomfortable enough to motivate an entire methodological fix. For proteomic organ clocks, platform-level reproducibility has been examined and cross-cohort correlations are high, but a published, per-organ, within-person test-retest figure is not something a consumer should assume exists for the specific assay they purchased. Ask for it. Its absence is informative.

Effect sizes at the individual level are modest. These models were built to rank populations, and they rank populations well. Point estimates for one person carry uncertainty that is usually neither displayed nor propagated into the report - a criticism made directly in the 2025 npj Aging discussion of whether the field needs aging clocks at all, which identified ignored prediction uncertainty as a core structural weakness.

Almost none of these have been validated as a way to track an individual's response to an intervention. This is the claim consumers most want and the evidence least supports. Demonstrating that a measure moves in the right direction and that the movement means something requires interventional trials with the clock as a pre-specified endpoint, adequate power, and evidence that clock change predicts outcome change. That work is largely not done. Regulators have not accepted these measures as surrogate endpoints, and researchers in the field say plainly that the surrogate-endpoint question remains open. A number that goes down after three months of effort is, on current evidence, at least as likely to reflect measurement variation, an acute inflammatory state, or regression to the mean as it is to reflect changed trajectory.

None of that makes the biology wrong. It makes the consumer inference premature.

Where this is research and where it is a consumer product

The line is cleaner than the marketing around it suggests, and it is worth drawing without reference to any particular company.

The underlying models from the Nature work are publicly available as an open-source package, which is a real contribution to transparency. But a published model and a validated clinical test are different objects. Running an open model on a plasma sample produces a number; it does not confer clinical validity on that number for the person it came from.

This site does not recommend, rank or discourage any specific vendor. The questions that actually separate a defensible offering from a weak one are boring and answerable: which platform, which model, which reference population, what is the reported test-retest variability, and is the result delivered with clinical interpretation or as a PDF. A provider who can answer all five is treating you as a patient. A provider who answers none is selling a number.

What to do with a result you did not expect

Suppose a report says your kidney is eight years older than you are. Here is the honest sequence.

Do not treat it as a diagnosis, and do not dismiss it either. It is one measurement from one model with unstated error bars, and it is also not nothing.

Take it to a clinician, and take the actual report. The useful move is almost always conventional: a well-chosen panel of established tests for the organ in question, interpreted in the context of your history, symptoms, medications and family history. Established organ-function testing is far better validated than any aging clock, and it is what a result like this should prompt rather than replace. If the conventional workup is clean, that is meaningful information about the aging-clock result - not the other way round.

Consider what was going on when the sample was drawn. Acute illness, recent intense exercise, an inflammatory flare, dehydration, a new medication, poor sleep, or a change in kidney clearance can all move plasma protein concentrations. A single draw during any of those is a poor basis for a conclusion about a decade.

Do not start anything on the strength of it. No supplement, compound or protocol has been shown to move an organ age gap in a way that has been demonstrated to matter, and there is no validated basis for using these numbers to select or titrate an intervention. Interest in the underlying biology - including areas like mitochondria and aging - is legitimate and worth understanding. Treating a proteomic age gap as an instruction is not.

Let the clinician own the actionable inference. Every genuinely useful thing that can come out of an unexpected organ age result - further testing, a change in surveillance, addressing a modifiable risk factor already known to matter - is a clinical decision made with information a report does not contain.

The reasonable posture, given where the evidence actually stands, is curiosity without urgency. These are among the most interesting measurements in aging biology. They are not yet instruments you should steer by alone.

Frequently asked questions

Why do my two biological age tests disagree? Almost certainly because they are different measurements sharing a name. Check what each was built from - methylation, plasma proteins, clinical chemistry - and what each was trained to predict: chronological age, mortality, phenotypic health, or rate of decline. Two models optimised for different targets in different cohorts should be expected to disagree about one person. Then check whether either report gives a test-retest figure; part of the difference between any two results is measurement noise, and for epigenetic clocks that component has been shown to be non-trivial. Neither test is necessarily "wrong." They are answering different questions. What biological age tests measure covers the families in more detail.

Can I get an organ-specific aging test? Assays of this kind exist commercially in several markets, and the underlying models from the published research are openly available. Availability is not the same as validation, and this article does not point at any particular provider. If you are considering one, the useful questions are which proteomics platform is used, which published model is applied, what reference population your result is compared against, what the reported within-person reproducibility is, and whether a clinician interprets the result with you. Regulatory status differs by jurisdiction and most such tests are not diagnostics.

If my brain age is high, does that mean I will get dementia? No. The cohort findings are real - an especially aged brain-protein profile was associated with an Alzheimer's hazard ratio of roughly 3.1 in UK Biobank, comparable in magnitude to one copy of APOE4 - but a hazard ratio compares groups over years of follow-up. It is not a personal probability, and this article will not convert it into one, because the arithmetic to do that validly for an individual does not exist in the published literature. Most people in a higher-risk group do not develop the outcome. Cognitive concerns are assessed clinically, not by a plasma proteomic index.

Can I lower an organ age with lifestyle changes or supplements? Organ age estimates have been shown to vary with lifestyle factors and medications across populations, which is biologically encouraging. What has not been shown is that deliberately moving the number in one person changes that person's trajectory, or that any specific compound reliably moves it at all. No aging clock is currently accepted as a validated surrogate endpoint, and none of these measures has been established as a way to track an individual's response to an intervention. The things with real evidence behind them for cardiovascular, metabolic and cognitive aging are worth doing on their own merits - not as a strategy for improving a score.

Related reading

• Epigenetic clocks - how DNA methylation clocks work, which ones matter, and why one result is a trend line rather than a verdict
• What biological age tests measure - the four families of test and what each was trained against
• Optimal versus normal lab ranges - the same population-versus-individual discipline applied to routine bloodwork
• Mitochondria and aging - the cellular biology underneath the organ-level picture
• Telomere length testing - what a consumer telomere test can and cannot tell you
• Klotho: the longevity protein - what it is, whether it can be measured, and what early human trials can tell us
• Why two biological age tests disagree - training targets, reference populations and assay noise

Part of Testing & Biological Age: The Measurement Guide.

Educational content, not medical advice. This article contains no doses, schedules or protocols, and nothing in it recommends, ranks or discourages any specific test, provider or product. Organ-specific aging estimates are research-derived measures, not diagnostic tests, and no result described here can be converted into an individual risk figure. Any unexpected test result should be discussed with your own clinician alongside conventional assessment. Claims for the Malaysian market are subject to MAB requirements.

Explore it with verified supply. Every Vivre Labs lot is independently tested and the certificate of analysis is published, so you know exactly what is in the vial.

References

• Oh HS, Sung YJ, Wyss-Coray T, et al. "Organ aging signatures in the plasma proteome track health and disease." Nature, 2023;624(7990):164–172. (4,979 plasma proteins measured in 5,676 adults across five independent cohorts; 893 proteins, ~18%, classified as organ-enriched, most from brain; models for 11 major organs; ~20% of the population strongly accelerated in one organ, 1.7% multi-organ agers; accelerated organ aging associated with 20–50% higher mortality risk; accelerated brain and vascular aging predicted Alzheimer's disease progression comparably to plasma pTau-181.)
• Oh HS, et al. "Plasma proteomics links brain and immune system aging with healthspan and longevity." Nature Medicine, 2025. (Organ ages for 11 organs from 2,916 plasma proteins in 44,498 UK Biobank participants, up to 17 years of follow-up; associations with incident heart failure, COPD, type 2 diabetes and Alzheimer's disease; aged brain associated with Alzheimer's hazard ratio ≈3.1, comparable to one APOE4 copy, independent of APOE genotype.)
• "Organ-specific proteomic aging clocks predict disease and longevity across diverse populations." Nature Aging, 2025. (Organismal and ten organ-specific clocks trained in UK Biobank, n = 43,616; validation cohorts in China, n = 3,977, and the USA, n = 800; cross-cohort correlations r = 0.98 and 0.93; brain aging most strongly linked to mortality.)
• "Proteomic organ-specific ageing signatures and 20-year risk of age-related diseases: the Whitehall II observational cohort study." The Lancet Digital Health, 2025;7(3):e195–e204. (6,235 participants aged 45–69 sampled 1997–99; age gaps for nine organs derived via SomaScan and the open-source organage package; 20-year follow-up through national health records; faster aging of one organ generally associated with morbidity affecting multiple organs.)
• Higgins-Chen AT, et al. "A computational solution for bolstering reliability of epigenetic clocks: implications for clinical trials and longitudinal tracking." Nature Aging, 2022. (Technical replicate deviations of up to ~3–9 years across six major epigenetic clocks; principal-component versions brought most technical replicates to within ~0–1.5 years.)
• "An unbiased comparison of 14 epigenetic clocks in relation to 174 incident disease outcomes." Nature Communications, 2025. (18,859 individuals; ten-year onset of 174 disease outcomes; second-generation clocks outperformed first-generation clocks for disease prediction.)
• "Do we actually need aging clocks?" npj Aging, 2025. (Critique of abstract definitions, inconsistent clinical validation, and unreported prediction uncertainty; surrogate-endpoint status not established.)
• organage - open-source Python package implementing the published organ-age models for SomaScan plasma proteomics data (GitHub / PyPI).