Share
How insitro unlocked population-scale human genetics in a tissue that has resisted measurement for decades – and what it reveals about peripheral approaches to metabolic disease
The Furnace We Couldn’t See
The human body contains a type of fat tissue engineered to burn energy rather than store it. Brown adipose tissue – dense with mitochondria, rich in uncoupling proteins – converts glucose and lipids into heat through non-shivering thermogenesis. For years, researchers have documented this tissue’s role in energy expenditure, its associations with favorable metabolic profiles, its potential as a therapeutic target for obesity and cardiometabolic disease.
What they have not been able to do is map its genetic architecture. The genetic variation was there, distributed across the genomes of hundreds of thousands of biobank participants. The barrier was measurement itself.
Brown fat, sparse and anatomically diffuse, concentrated in small depots around the neck and upper chest, resists the standardized quantification that genetic discovery requires. The gold standard for detection – PET-CT scans with radioactive tracers – involves radiation exposure and cannot scale to the many thousands of participants required for genome-wide association. Without scalable measurement, the genetic programs governing this tissue remained inaccessible to systematic investigation.
Research presented this week by instiro at the Keystone Symposia on Obesity Therapeutics demonstrates that this constraint can be resolved – and that resolving it opens a therapeutic frontier that has been structurally closed to drug discovery.
Why Human Genetics Matters
Translating brown fat biology into therapies has proven difficult. Many targets derived from rodent studies have failed to produce meaningful weight loss in human trials, despite showing BAT activation in mice. FGF21 agonists, orexin antagonists, β-adrenergic receptor agonists – each effective in rodents, each disappointing in humans. This translation gap underscores the need for discovery anchored in human genetics from day one.
“To power human genetics, you need tens of thousands of people – but brown fat measurements have historically required expensive approaches such as PET scans, which is hard to acquire at scale,” said Daphne Koller, Ph.D., founder and CEO of insitro. “AI enables these measurements to be derived from MRIs, unlocking a first-ever analysis of human genetics for BAT.”
Generating Phenotypes at Scale
Using the UK Biobank imaging cohort (n=69,598), insitro developed an AI-derived BAT phenotype from Dixon MRI fat-signal fraction maps, using the delta between abdominal and supraclavicular adipose fat-signal fraction to estimate brown fat content. This approach – measuring fat-signal fraction in the supraclavicular region where BAT resides and comparing it to pure white fat in the abdomen – creates the measurement infrastructure that genetic discovery requires.
Before proceeding to genetic association, the team validated biological specificity through multiple independent tests. The phenotype showed seasonal variation consistent with BAT thermogenesis, with strongest activity observed during late-winter months (p<0.001) – a pattern absent in broader adiposity measures and not explicitly trained into the model.
Phenome-wide association analyses confirmed correlations in the hypothesized direction with 16 of 20 cardiometabolic phenotypes tested, including body composition, lipid profiles, glucose homeostasis, and vascular health. The phenotype that enabled the first human BAT GWAS did not exist before insitro created it. AI did not accelerate this work. It made it possible.
The First BAT GWAS
When genome-wide association studies were conducted using this AI-derived phenotype, the results revealed genetic architecture that had been invisible to previous approaches.
The study identified 81 independent genetic loci, fine-mapped to 45 distinct genes. Among them: PRDM16, a master regulator that directs cell fate toward brown adipose tissue; PGC-1α, a coactivator that promotes thermogenic gene expression; EBF2, a transcription factor that determines and maintains BAT identity. These established regulators served as positive controls, confirming the methodology was capturing genuine signal. A BAT polygenic risk score showed causal associations across multiple cardiometabolic trait categories consistent with BAT’s posited metabolic effects.
The critical finding is that 27 of the 45 BAT genes are unique to our BAT GWAS and do not appear in GWAS of other fat depots or waist-to-hip ratio, even in substantially larger studies.
This tissue specificity carries direct therapeutic implications. The genetics of brown fat do not substantially overlap with the genetics of general obesity. The biological programs governing thermogenic fat are distinct from those controlling appetite and energy storage.
“Our results provide strong evidence for the important role that BAT plays across multiple metabolic health outcomes,” said Koller, “and may reveal differentiated mechanisms of peripheral fat reduction that are orthogonal to the central appetite pathways dominating current obesity therapies.”
From Association to Mechanism
insitro’s CellML™ platform screened these genetically supported targets in primary human white-adipocytes, using high-content imaging, transcriptomics, and functional assays to assess beige/brown-like character and lipid mobilization.
Lead candidates induced a beiging signature, the conversion of energy-storing white adipocytes into thermogenically active cells. Knockdown produced upregulation of Ucp1, the protein responsible for dissipating energy, alongside altered lipid droplet morphology, suggesting enhanced metabolic capacity.
“This is the difference between discovery driven by AI and human genetics, and discovery driven by trial and error,” said David Lloyd, Ph.D., SVP of Metabolic Disease and Translational Pharmacology. “Starting with scalable human phenotypes and genetic support allows us to move into functional validation with far more confidence and conviction.”
In diet-induced obese mice, a fat-targeting siRNA designed to reduce BAT-01 expression produced a 15% reduction in body weight over four weeks, driven by fat mass loss while preserving lean mass – all without impact on caloric intake. In response to knockdown, multiple white fat depots showed increased Ucp1 and decreased Leptin expression, findings consistent with successful induction of a beige-like phenotype.
The weight loss did not result from appetite suppression. The mice consumed the same calories. The mechanism represents peripheral targeting of energy expenditure through metabolically active fat, rather than centrally-acting appetite control, offering a therapy which may avoid typical GI related AEs observed with standard of care.
“These preclinical results point to BAT-linked targets that promote fat loss and cardiometabolic health through selective peripheral targeting while avoiding appetite suppression,” said Lloyd, “and may open new paths for differentiated obesity therapies.”
This research follows a methodology insitro has demonstrated across therapeutic areas: using AI to generate phenotypes at scale, anchoring discovery in human genetics, interrogating and validating mechanisms in human cells, and confirming efficacy in vivo before advancing candidates toward the clinic.
The same approach that unlocked brown fat genetics has been applied to MASH, where AI-derived liver fat phenotypes enabled identification of hundreds of genetic loci and led to programs now advancing toward clinical development. The methodology generalizes because AI can be used to scale human data to unlock disease-relevant phenotypes that have resisted population-scale measurement.
What Comes Next
→ Advancing the Pipeline: insitro is evaluating additional BAT-linked genes from the GWAS using CellML and in vivo studies to advance the most promising targets for tackling the global obesity epidemic through deeper mechanistic characterization and preclinical validation.
Thank you to the many insitrocytes who contributed to this work.
______________________________________________________________________________________________________________
Presented by: David Lloyd, Ph.D., SVP, Metabolic Disease and Translational Pharmacology
Research team includes contributions from: Kaiwen Xu, Rami Jaafar, Dan Lin, Jonathan Choy, Eilon Sharon, Nick Eriksson, John Pham, Joy Chen, Lin Gan, Pablo Garcia Nieto, Anna Shcherbina, Nav Ranu, Angela Detweiler, Ravish Malhotra, Vaishaali Natarajan, Kate Lozada, Shivani Thombare, Albin Huang, Colm O’Dushlaine, and many more insitrocytes.