A young zebrafish takes a sniff, and suddenly everything changes. After a few days of training, this tiny fish—barely bigger than a grain of rice—can tell two smells apart with startling precision: one that leads to food, and one that doesn't. But the real discovery isn't in the fish. It's in the squishy, crackling network of brain cells that makes that decision possible.
Scientists have long wondered how learning reshapes the way a brain handles sensory information. One old idea was that the brain stores each familiar smell like a photograph in a drawer—a single, fixed pattern of neural activity that gets pulled out whenever the odor appears. But a team led by Rainer Friedrich at Harvard University suspected something more flexible was going on.
To find out, the researchers trained juvenile and adult zebrafish to link one odor with food and another with no reward. After training, the fish swam eagerly toward the feeding area when they caught the scent of the food-linked odor—clear proof they had learned the connection. Their results were published in the journal Nature Neuroscience.
Then came the surprising part. When the team measured activity in a brain region that forms internal "smell maps," they found that the response to the same odor was different every single time the fish encountered it. And the activity faded quickly once the odor disappeared. There was no fixed pattern to be found—no single mental photograph.
Instead, the key turned out to be geometry. Neuroscientists describe the range of activity patterns a brain can produce using something called neural manifolds—think of them like clusters on a map, where each point represents the combined firing of many brain cells at one moment. Nearby points mean similar patterns, and together they trace out the shape of how a smell is represented. These shapes can stretch across many dimensions, one for every brain cell involved, and they can twist into wildly complex forms.
To decode this geometry, Friedrich's team teamed up with SueYeon Chung, a theoretical neuroscientist at the Flatiron Institute and New York University, now at Harvard, who built a mathematical framework just for studying these manifolds. Together, they discovered something beautiful: after training, the clusters linked to important odors simply became easier to tell apart. Computer models reproduced the same effect, and fish whose brain responses were more distinct were better at telling the two smells apart.
"What is going on when the brain learns" is finally coming into view, Friedrich says. He notes that researchers don't yet know exactly which changes in connections between brain cells cause this reorganization. But the big picture is clear: learning a smell isn't about stashing it in a fixed mental drawer. It's about reshaping an entire map so that the important things stand out. For a hungry zebrafish—and for scientists trying to understand memory itself—that's a discovery worth sniffing out.