Researchers used artificial intelligence to quantify brown adipose tissue activity in 8,709 people and then worked backwards from a clinical medication pattern to a possible drug mechanism. In the 101,243-patient retrospective cohort, people with detectable brown fat had lower prevalence of several cardiovascular and psychiatric diagnoses. Use of the blood-pressure drug losartan was positively associated with brown-fat presence.[1][2]
The study does not show that prescribing losartan activates brown fat in patients. It combines three evidence stages: an observational clinical association, experiments in cultured adipocytes and treatment in mice. That sequence creates a plausible repurposing hypothesis; the human therapeutic claim remains untested.[1]
The AI turned a yes-or-no scan finding into an activity measure
Brown fat consumes energy to generate heat. It can appear on clinical FDG-PET/CT scans because metabolically active tissue takes up the labelled glucose tracer. Earlier work often classified a person as brown-fat positive or negative. The new pipeline semi-automatically measured tissue volume and tracer uptake, allowing the researchers to compare total activity and specific anatomical depots.[1][2]
From the cohort to the quantified sample
| Measure | Value | Context |
|---|---|---|
| Clinical cohort | 101,243 patients | The data came from a large cancer centre, not a population screening programme.Source 1 |
| Detectable brown fat | 8,897 patients | The remaining 92,346 were classified as brown-fat negative.Source 1 |
| Propensity-matched comparison | 8,897 positive and 17,405 negative patients | Matching included age, sex, body-mass index and ambient temperature.Source 1 |
| AI-quantified activity | 8,709 patients | 97.9% of the brown-fat-positive group; 188 could not be quantified because of scan field-of-view differences.Source 1 |
| Activity measure | Mean FDG uptake multiplied by tissue volume | This is a metabolic imaging proxy, not a direct thermometer reading.Source 1 |
The medication search produced an association, then a laboratory target
Among medications positively associated with brown-fat status, the team focused on angiotensin receptor blockers. Losartan had an odds ratio of 1.36 for brown-fat presence, with a 95% confidence interval of 1.13 to 1.63 and P=0.006. Valsartan’s reported odds ratio was 1.52. These values describe association after the study’s modelling; they do not prove the medicines created the tissue.[1]
Three layers of evidence
| Layer | What the study found | What it cannot establish alone |
|---|---|---|
| Patient records | Losartan and some other medicines were associated with detectable brown fat. | Causation, dose response or a treatment benefit.Source 1 |
| Cultured adipocytes | Higher losartan doses increased UCP1 and thermogenic activity; an inactive metabolite implicated PPARγ. | The concentration, metabolism and whole-body effect in patients.Source 1 |
| Mice | Two weeks of treatment increased thermogenic markers and energy expenditure under tested conditions. | Safety and efficacy as a brown-fat therapy in humans.Source 1Source 2 |
The mechanistic result was more specific than a generic receptor-blocking story. EXP3179, a losartan metabolite without affinity for the drug’s usual AT1 receptor target, increased Ucp1 expression. Blocking PPARγ reduced the response. The authors therefore propose a PPARγ-dependent route that is at least partly independent of classical AT1 signalling.[1]
The cohort is large, but it is not neutral
The patients were scanned at a cancer centre for clinical reasons and without standardized cold stimulation. Cancer type, treatment and the circumstances that led to imaging can shape both medication use and metabolism. The researchers matched and adjusted extensively, but no retrospective design can eliminate every unmeasured difference between people taking and not taking a drug.[1]
The central interpretation traps
| Tempting conclusion | What the evidence supports instead |
|---|---|
| “Brown fat prevents disease.” | Brown-fat presence and activity were associated with lower prevalence of several conditions in this cohort.Source 1 |
| “AI diagnosed metabolic disease.” | The model quantified an imaging phenotype that was compared with clinical records.Source 1 |
| “Losartan activates brown fat in people.” | Human use was associated with brown-fat presence; activation was demonstrated in cells and mice.Source 1Source 2 |
| “A common blood-pressure drug is now an obesity treatment.” | The study identifies a repurposing candidate that requires prospective human testing.Source 1Source 2 |
The platform may matter even if losartan does not
The broader research method is a reverse-translational loop. Start with a very large clinical dataset, use AI to quantify a phenotype that was previously difficult to measure at scale, search records for associated medicines, and then test plausible mechanisms in the laboratory. The loop can generate candidates from human evidence before committing to costly trials.
Its weakness is also visible: every stage can amplify selection and measurement bias. A tracer-uptake model, a cancer-centre cohort and medication records are not interchangeable with a randomized intervention. The achievement is not a proven brown-fat treatment. It is a disciplined route from millions of clinical observations to one testable human question: does losartan, at an appropriate dose and in an appropriate population, safely activate brown fat and improve a meaningful outcome?
Sources & further reading
- Artificial intelligence-enabled clinical phenotyping of human brown fat Nature · 14 Oct 2026 · Accessed 16 Oct 2026
- A blood-pressure drug may activate health-promoting brown fat National Institutes of Health · 14 Oct 2026 · Accessed 16 Oct 2026