New AI Imaging Detects ‘Striking’ Obesity

GLP-1 receptor agonists are being prescribed to tens of millions of patients with obesity. What no one can yet answer is whether those drugs are protecting their peripheral nerves or damaging them further. Small clinical studies suggest GLP-1 receptor agonists may improve nerve structure in patients with diabetic neuropathy, yet other studies document cases where patients develop peripheral neuropathy because of rapid weight loss due to GLP-1 drugs. That question gained new urgency with a recently published Nature study. A team at Helmholtz Munich in Oberschleißheim, Germany, used an AI system called MouseMapper to scan every nerve, immune cell, and 31 organs in a transparent mouse body — all at once. What they found suggests obesity-associated nerve damage extends far beyond the extremities where clinicians typically measure it. First, it flagged significant damage in the infraorbital nerve, the branch of the trigeminal nerve that carries sensation from the whisker pad. The fine, branching endings of the nerve were depleted by roughly 60%. But the thick main trunk of the nerve was virtually untouched. Consequently, the mice had lost much of their ability to feel through their whiskers. “The trigeminal nerve finding was one of the striking findings, but the broader surprise was that the algorithm could reveal many structural changes across the body that would be very difficult to detect manually,” said Ali Ertürk, PhD, director of the Institute for Intelligent Biotechnologies at Helmholtz Munich and senior author of the study. But a mouse study means nothing for a patient unless the same biology shows up in human tissue. So the team obtained postmortem trigeminal ganglia from people who had been overweight and ran the same molecular analysis. The same pathways were failing, in the same structure but in a different species. The molecular fingerprint of nerve distress that MouseMapper had flagged in a mouse was already present in human tissue. What’s still currently unknown is how much nerve damage has been accumulating in the tens of millions of patients now being treated with GLP-1 drugs. A Problem of Shape The reason no one had attempted this before is not for want of curiosity but because they couldn’t solve a geometrical puzzle. Researchers have been able to make entire mouse bodies transparent through tissue clearing and to image them in three dimensions using light-sheet microscopy. But a single high-resolution scan produces tens of terabytes of data, representing billions of cells across an entire mammalian body. And nerves, unlike tumors or immune cells, are not compact objects an algorithm can count. They are filaments that run for centimeters through muscle, bone, and fat, branching into ever-finer threads that disappear against the tissue background. “That is several orders of magnitude more complex than detecting round cell bodies or tumor nodules,” Ertürk said. MouseMapper solved this by building on vesselFM, an AI pre-trained on a vast dataset of blood vessel images and then fine-tuned for nerves. Vessels and nerves are remarkably alike in structure, as both branch and contain tubular networks. That shared geometry let the AI transfer what it knew about one to the other. The system screens the whole mouse, flags where something has changed, and only then does the team zoom in. Following the Signal When MouseMapper compared lean and overweight mice, nerve density had dropped across the body. But in the head, the signal sharpened into something specific. The infraorbital nerve had lost much of its branching architecture. The team extracted a mathematical graph of the nerve’s structure and counted its endpoints, its branch points, and the segments connecting them. Nerve endings, edges, and vertices were all reduced by 58% to 61%. But the main trunk was intact, meaning the damage they had graphed was confined to the periphery. The terminal branches had retracted while the cable they branched from held — the signature of dying-back degeneration, in which an axon withers from its tip inward. It is the same process that takes sensation from the feet of a patient with diabetes. “You lose those distal nerve endings — those are the most susceptible to die-back and degeneration — but you can still maintain the main nerve bundle where they branched off from,” said Kristy L. Townsend, PhD, professor of neurological surgery at the Ohio State University, Columbus, Ohio, who studies nerve loss in metabolic disease. She was not involved in the study. Townsend noted that the study’s power lies in treating neuropathy as a whole-body problem rather than an organ-by-organ one. Examining the study’s extended data, she pointed to evidence of denervation across the lymph nodes, thymus, heart, and subcutaneous fat — what she called “broad neuropathy” beyond the trigeminal finding that made the headline. But she also flagged what the tool cannot see. “The marker that they’re using, the resolution of the imaging, they’re really seeing large fiber damage,” she said, “and we know that a lot of the damage around the body is small fiber axons that are very difficult to resolve.” The study also used only male mice of a single inbred strain, which she compared to “cloning one human.” The Molecular Fingerprint Imaging can reveal a damaged structure but not the cause of that damage. To learn why the nerve was failing, Doris Kaltenecker, PhD, lead scientist at the Institute for Diabetes and Cancer at Helmholtz Munich and the MouseMapper study’s first author, dissected the trigeminal ganglia — the clusters of nerve cell bodies that give rise to the infraorbital nerve — from the same mice and ran spatial proteomics, a mass-spectrometry method that catalogs the proteins present in a small, defined piece of tissue. Of more than 6000 proteins identified, 230 were significantly altered in overweight animals. Of those, 67 were upregulated — present in greater abundance, meaning the cells were producing more of them — while 163 were downregulated, meaning production had fallen. The ganglion began producing many more proteins involved in cytoskeletal remodeling — the machinery cells use to reorganize their internal structure — and in inflammatory signaling cascades, suggesting the nerve was actively restructuring itself and ramping up its inflammatory response. What the ganglion was making less of was arguably more alarming. The downregulated proteins included multiple members of the SERPIN-A family. These proteins normally act as shields to block inflammatory enzymes from breaking down the structural tissue around nerves. In obesity, the nerve was producing fewer of them, effectively lowering its defenses at the same time the inflammatory assault was intensifying. Zooming out, three groups of pathways stood out: axon guidance, which directs a growing nerve branch where to extend; regulation of the actin cytoskeleton, which maintains the internal scaffolding that gives a nerve its shape; and the complement cascade, an arm of the innate immune system that can destroy tissue when overactive. No single upstream trigger could explain the pattern. “Obesity creates a complex environment characterized by chronic low-grade inflammation, altered lipid metabolism, and metabolic stress,” Kaltenecker said. The pathways appear to fail in parallel, not from one master switch. The human comparison was small but pointed. Postmortem trigeminal ganglia from five lean and four overweight individuals, obtained through the University of Leipzig, Leipzig, Germany, showed pathway-level convergence with the mouse data — axon guidance, neurodegeneration, and actin cytoskeleton regulation were all disrupted. The convergence is meaningful, but “similar biological processes were altered, not the same individual proteins,” Kaltenecker said. Whether people with obesity experience trigeminal sensory deficits remains unstudied. The Canary in the Coalmine, the Tip of the Iceberg To Townsend, the trigeminal finding almost certainly understates the real damage. The imaging method resolves large nerve bundles, but much of obesity’s damage occurs in small fibers too fine for this resolution to capture. “It’s a canary in the coal mine,” she said. “There’s much more nerve damage that couldn’t be seen.” That assessment reframes the study for clinicians. In a 2024 review in Diabetes, Townsend laid out evidence that peripheral neuropathy has been documented across the heart, skin, muscle, gut, pancreas, liver, and fat in metabolic disease — essentially every metabolically active organ examined. She described the central and peripheral nervous systems as “one continuous circuit” and called peripheral neuropathy “the number one manifestation of neurodegenerative disease in humans,” affecting an estimated 30 million Americans. The burning and numbness patients report in their feet, she said, is “probably just the tip of the iceberg.” Beneath it may lie a broader neuropathy silently degrading the nerve supply of organs throughout the body — contributing to dysfunction that has never been connected to nerve loss, because until recently there was no way to see it. What GLP-1 Drugs Might Do — and What They Might Not The reframing lands squarely on the drugs now reshaping obesity medicine. GLP-1 receptor agonists are among the most widely prescribed new medications in the world, and their receptors sit not only in the brain but also on peripheral nerves. Could these drugs protect or regenerate nerves beyond what weight loss alone achieves? Ertürk sees MouseMapper as built for exactly that question, to screen a drug’s effects across the whole body at once, asking which tissues benefit and which might be harmed. Kaltenecker said her group has already begun testing whether GLP-1 receptor agonists reverse the nerve and proteomic changes observed in this study. But Townsend offered a caution that should give prescribers pause. Rapid weight loss itself — regardless of the drug — can stress peripheral nerves. A recognized clinical condition called treatment-induced neuropathy of diabetes, in which sudden improvement in blood sugar paradoxically triggers acute nerve damage, has now been documented in patients taking semaglutide. “Figuring out who’s at risk, who might have underlying neuropathy, before they go on these drugs,” Townsend said — that may prove clinically important “so they lose weight a little bit more slowly and don’t cause any toxicity to their nerves.” Townsend's lab is testing whether catching nerve damage early — while the supporting Schwann cells remain in a repair-ready state and the nerve retains some plasticity — makes reversal possible before small-fiber damage progresses to the loss of larger bundles. “We do think there’s a critical window,” she said. Whether screening should now extend to the face, she is not ready to say. Metabolic neuropathy in patients still begins in the feet and hands, and that remains the right place to start. If one unbiased scan of a single mouse revealed damage to a nerve nobody was watching, how many other canaries are sitting in the human body — in organs and tissues no one has thought to check — simply because, until now, there was no way to look? Ertürk is the co-founder of Deep Piction, a company applying AI to biomedical imaging; full disclosure information for study authors is available in the original study publications. Townsend is co-founder of Neuright, Inc., a company developing diagnostics and gene therapy approaches for peripheral neuropathy, and holds patents related to neuropathy treatments, devices, and biomarkers.