I did not expect the most striking molecule in a blood test for aging to be the one that makes feces smell like feces. Researchers in Shanghai measured 56 age-linked metabolites in the plasma of 18,476 people, and of all of them, 3-methylindole had the most significant positive association with death. You probably know it by its common name, skatole.
I laughed out loud at that. Then I got stuck on a question I couldn’t let go of: what is a bacterial waste product doing inside a measure of how fast a person ages?
A smell made by microbes
Your own cells don’t make skatole in any meaningful amount. Bacteria in your colon, mainly Clostridium and Bacteroides species, build it out of tryptophan, an amino acid from your food, by way of an intermediate called indole-3-acetic acid. Indoles like it slip across the gut lining by diffusing through cell membranes, and once skatole reaches the liver, a set of CYP450 enzymes (CYP1A, CYP2A and CYP2E1) breaks it down. The level in your blood therefore depends on what your microbes are churning out, how much gets through the gut wall, and how quickly your liver clears it.
That last detail is what made me sit up. The research team sorted their metabolite panel into biological domains, and one of them is host-microbiome co-metabolism: chemistry that your body and your gut bacteria do together. Skatole rose with age across the cohort. Whether the microbes are driving decline or just reporting on it, this study can’t say. A different microbial community, a leakier gut and a slower liver could each push skatole up, and the authors themselves say their proposed pathways still need experimental confirmation. Even so, I find it a little thrilling that a blood measure of aging carries a fingerprint of the bacteria living in you, and not only of you.
Two clocks with two jobs
The paper in Science Translational Medicine, from Shanghai Jiao Tong University, drew fasting plasma from participants aged 1 to 96 at six centers in China and built two tools from it, because healthy and sick people need different questions asked of them.
For healthy people the question is how old your blood looks. A stacked ensemble of machine-learning models estimates a “metabolomic age” and subtracts your calendar age. The gap is called metabolomic age acceleration, or mAA, and a positive number means your chemistry is running older than your birthday.
For people already living with chronic disease, the team built a separate “age deviation” measure tied to mortality risk. Illness scrambles exactly the chemistry a standard clock reads, so a clock trained on healthy blood tends to misfire on sick blood. Building a second tool for them is a smart design choice, and it’s the one I’d want if I were the patient.
The public dataset lays out the split: 5,252 healthy people to train the age model, 5,456 healthy people to test it, and 7,768 people with chronic disease for the age-deviation work. The 56 metabolites were narrowed down from 232 features measured by nano-assisted laser desorption/ionization mass spectrometry. Baseline blood was drawn in 2019, disease status was rechecked in 2023, deaths were tracked through January 2025, and the follow-up ran 5.5 years.
Opposite skatole sat 4-hydroxyproline, a collagen-related molecule that fell with age and had the most significant inverse association with mortality. Creatinine, a muscle breakdown product the kidneys filter out, climbed with age, and leucine rose after midlife and went with a higher mAA. Among the pathways that stood out were threonine, serine and glycine metabolism, cysteine and methionine metabolism, and butanoate metabolism. Butanoate is butyrate, the short-chain fatty acid gut bacteria ferment from fiber, though a pathway lighting up in a statistical analysis isn’t the same as measuring how much butyrate anyone’s microbes actually make.
The chronic-disease group gave the same kind of picture. Higher creatinine, uric acid and urea nitrogen went with greater age deviation, while higher hemoglobin, hematocrit and red-cell counts went with less. Chronic kidney disease, gout and stroke were each tied to higher age deviation, the measure rose with multimorbidity, and men scored higher than women. I’d treat those as markers, not verdicts: creatinine reflects muscle mass as well as kidney filtering, and a low plasma hydroxyproline doesn’t by itself show that anyone’s collagen is failing. Still, kidney chemistry turns up in this panel again and again, and kidneys are the body’s filter.
The lifestyle findings read like a public-health poster with a couple of oddities. Higher mAA tracked with active smoking, secondhand smoke, shift work, hair dye use and a parental history of premature cardiovascular disease. In the diet data it went with more sugar, processed meat, fried food, organ meat and seafood, while mixed grains and exercise went with a younger reading. Those exploratory associations weren’t adjusted for multiple comparisons, so I’d hold the seafood and organ-meat results loosely. Test enough foods and some will light up by chance, and nothing here separates the fish from how it was cooked.
The change-over-time data is the most interesting to me, but it comes from a smaller group, since only 1,500 people had their plasma profiled a second time. In that data, people who became more active or cut back on fried food saw their mAA shift in a favorable direction. New secondhand-smoke exposure, more preserved food, worsening depressive symptoms and a move into overweight or obesity shifted it the other way. Among people with hypertension, those who got their blood pressure below 140/90 by 2023 showed better changes in age deviation. What this shows is a biomarker moving alongside a behavior. It doesn’t prove that a walk buys you years, but it’s the right direction, and it’s measured in people rather than in mice.
On prediction, people in the top quartile of mAA had significantly higher risk of fatty liver disease, hypertension, diabetes and death than people in the bottom quartile, and the dataset abstract adds incident cancer. Both measures were more strongly associated with all-cause mortality, and separated high-risk from low-risk people better, than a frailty index, the American Heart Association’s Life’s Essential 8 score, or routine blood biomarkers. The improvement was modest. When the team retrained its method on UK Biobank and NHANES data, it predicted age well, which shows the modeling approach carries over to other populations. That isn’t independent validation of this plasma clock, and with a predominantly Chinese cohort, the authors say the mortality predictions still need testing elsewhere.
What the public files can’t check
Posting the data on Dryad deserves credit. The three files contain a de-identified ID, cohort group, chronological age and 56 normalized metabolite intensities, labeled by mass-to-charge value. There are no disease outcomes, no deaths and no clinical measures, and the record doesn’t list any code. With those files an outside researcher can probe the age model. Rerunning the mortality and disease analyses, which carry the paper’s heaviest claims, would take data the group hasn’t posted.
The record lists 21 grant awards under China’s National Natural Science Foundation, and no company appears in it. The authors have made their age framework available through a public web interface, and they say the metrics may help identify individuals at elevated risk “before overt clinical disease” and stratify risk in people who are already sick.
That promise has been made before. In 2017, Duke’s Daniel Belsky compared 11 aging measures in nearly 1,000 members of New Zealand’s Dunedin cohort and found that they didn’t agree on how fast any given person was aging. “Based on these results, I’d say it’s premature to market aging tests to the public,” he said. The Shanghai clock is a sharper instrument than anything in that comparison. It’s also still a research tool, built in one population, and nobody outside the group has checked its death predictions.
I came in expecting another longevity gimmick with machine learning painted on, and I’m leaving fascinated by a smell molecule my gut bacteria make out of dinner. I would not buy a biological-age test on the strength of this paper, and I won’t treat skatole as a verdict on anyone’s gut until outside researchers get the outcome data and run the death predictions themselves.
Sources
- Science Translational Medicine (2026): Metabolomic aging and mortality clocks in 18,476 individuals
- Dryad dataset record: Metabolomic aging and mortality clocks based on ensemble learning
- News-Medical (2026): Your age may not tell the whole story about your health, but your blood could
- Inside Precision Medicine (2026): Metabolomic aging clock predicts chronic disease, mortality risk
- Frontiers in Pharmacology (2021): New insights into gut-bacteria-derived indole and its derivatives in intestinal and liver diseases
- Duke Today (2017): Aging tests yield varying results