Artificial intelligence-enabled clinical phenotyping of human brown fat.

Abstract

Adipose tissue has a central role in systemic energy balance and the pathogenesis of obesity and its sequelae^1,2. In contrast to white adipose tissue, brown adipose tissue (BAT) uses energy to produce heat³. Detectable BAT in humans is associated with decreased prevalence of cardiometabolic diseases⁴. However, the complete spectrum of health conditions associated with BAT remains unknown, and high-throughput methods to quantify BAT activity in humans are lacking^5,6. Here we leveraged big data and artificial intelligence to systematically define the clinical impact of BAT in a cohort of 101,243 patients from a large cancer centre. Our unbiased analyses revealed that, across all diagnoses, cardiovascular and psychiatric diseases were less prevalent in people with BAT. Using a machine learning approach, we semi-automatically quantified BAT activity in approximately 8,700 patients and demonstrated that higher BAT activity, especially in the axillary depot, is associated with lower odds of cardiometabolic diseases and healthier white fat distribution. People with BAT also exhibited more favourable glycaemic parameters, lipid profiles, inflammatory markers and echocardiographic indices. Analysis of medication usage demonstrated that angiotensin receptor blocker use was positively associated with BAT presence. As proof of concept, we showed that the angiotensin receptor blocker losartan increased brown adipocyte activity and whole-body energy expenditure. Altogether, our studies harnessing artificial intelligence and big data identify new clinical phenotypes linked to BAT and serve as a platform for drug repurposing to activate this tissue in humans.