Your driver's license records one age. A machine learning model trained on more than 7,000 blood proteins can now estimate more than 40 others, one for each cell type it can detect, and they do not agree with each other or with the license.
That is the finding from a Stanford-led team whose work appeared in Nature Medicine on June 15. Drawing on plasma from 60,542 individuals across three cohorts, the researchers built separate biological-age estimates for cell types spanning neuronal, immune, glial, endocrine, epithelial, and musculoskeletal origins.
The single most consequential result involves a brain cell most people have never heard of.
Aging Turned Out to Be Wildly Uneven Within One Body
The premise behind biological age research is that chronological age is a crude proxy. Two 60-year-olds can have very different tissue conditions, and these differences predict who gets sick.
Earlier work from the same lab scaled that idea to whole organs. Analyzing 44,498 UK Biobank participants aged 40 to 70 and roughly 3,000 blood proteins, Stanford researchers reported that an extremely aged brain carried 3.1 times the Alzheimer's risk of a normally aging one, while an extremely youthful brain carried roughly a quarter of that baseline risk. Brain age was also the single best predictor of overall mortality in that analysis, published in Nature Medicine in July 2025.
The new study pushes resolution down another level, from organs to the cells inside them, and the picture is patchy rather than uniform. Between 20% and 25% of individuals showed accelerated aging in a single cell type, while only 1% to 3% showed it across 10 or more. Those cellular signatures were associated with existing disease and predicted new disease and death over 15 years of follow-up.
Astrocytes were not the only standout. People with an aged skeletal-muscle-cell profile had 12.7 times the incidence of amyotrophic lateral sclerosis compared with those whose muscle cells read as youthful, a difference detectable more than three years before symptom-based diagnosis.
The APOE Finding That Cuts Both Ways
Astrocytes are star-shaped support cells that vastly outnumber neurons and maintain the chemical environment brain cells depend on.
People with the APOE4 genotype had older astrocytes than APOE3 carriers, but younger macrophages, a type of immune cell. APOE2 carriers, whose genotype is protective against Alzheimer's, showed the inverse pattern. The two alleles pushed different cell populations in opposite directions.
Then the part draws attention from Alzheimer's researchers. Among people carrying two copies of APOE4, the highest-risk genotype, extreme astrocyte aging roughly tripled the risk of developing Alzheimer's. Cumulative incidence over 15 years reached 38.3% in that group, compared with 12.6% in APOE4 homozygotes whose astrocytes aged normally. The same gradient held for lower-risk genotypes.
Stanford's own summary puts the flip side bluntly: for double-dose APOE4 carriers whose astrocytes lean youthful, the amplified risk is effectively neutralized.
This Is Not a Test You Can Ask For, Yet
Nothing here is available in a clinic, and nothing here should change what anyone does today.
The estimates come from research-grade proteomics platforms measuring thousands of proteins, not from a standard blood panel. There is no regulatory clearance, no established reference range, and no evidence that acting on a cell-type age score improves any outcome. Direct-to-consumer biological age tests already on the market are not this technology.
That said, commercialization is underway, and readers deserve to know it. Senior author Tony Wyss-Coray is a co-founder of Teal Rise and Vero Bioscience, two companies Stanford has licensed the technology to, and he has said a test could reach the market in the next two to three years. That is a disclosure, not a disqualification, but it belongs alongside the science.
The associations are also observational. Youthful astrocytes tracking with lower Alzheimer's risk does not establish that the astrocytes are doing the protecting, nor does it identify anything a person can do to keep them young. The mechanism is the open question, not the answer.
Where Brain Age Research Actually Stands
Several groups are now estimating brain age from different signals, and they do not agree on a standard.
An independent team built organ-specific proteomic clocks in Nature Aging, training on 43,616 UK Biobank participants and validating in cohorts from China and the United States, and again found brain aging most strongly linked to mortality. Structural MRI has generated a decade of work on the brain age gap. Sleep is another route entirely: a pooled analysis of 7,105 dementia-free adults across five cohorts, published in JAMA Network Open in March, computed a brain age index from overnight EEG microstructure and found each 10-year increase associated with a 39% higher rate of incident dementia, holding after adjustment for demographics, lifestyle, and comorbidities.
Blood proteomics, MRI, and EEG measure different things and are not interchangeable. What they share is a direction of travel: away from one number for the whole person, toward many numbers for many tissues. Anyone worried about memory or cognition should raise it with a physician, where validated tools, including p-tau217 blood tests, already exist.
Key Questions Answered
What did the study measure?
Machine learning models estimated the biological age of more than 40 human cell types using over 7,000 plasma proteins measured in 60,542 people across three cohorts.
What was the astrocyte finding?
Among people with two copies of APOE4, extreme astrocyte aging roughly tripled Alzheimer's risk, with a 15-year cumulative incidence of 38.3% versus 12.6% for those aging normally. Youthful astrocytes were associated with reduced risk.
Does everyone age at the same rate internally?
No. Between 20% and 25% of participants showed accelerated aging confined to a single cell type, while only 1% to 3% showed it across 10 or more cell types.
Can I get this test?
Not yet. It uses research-grade proteomics platforms that lack regulatory clearance and established reference ranges. The senior author has co-founded companies to commercialize the approach and estimates a timeline of two to three years.
Does this mean APOE4 carriers can escape their genetic risk?
The analysis is observational and identifies an association, not a cause or an intervention. Nothing in it says how to keep astrocytes young.
How does this relate to brain age from MRI or EEG?
They are separate technologies that measure different biological aspects. All links accelerated brain aging to worse cognitive outcomes, but none are standardized for clinical use.
