When a fruit fly locks onto the smell of fermenting fruit, it does not simply follow its nose upwind. New research published in Nature reveals that Drosophila employs a sophisticated two-state navigational strategy, tracking the edge of an odor plume rather than its center, and using directional memory to guide returns after each exit.
The finding challenges a long-held assumption that vector-based navigation (the ability to store and recall a direction to a goal) requires a brain larger than what an insect possesses. A fly has roughly 200,000 neurons in total, compared to the 86 billion in a human brain. Yet its central complex, a structure only a few hundred microns wide, implements a computational strategy that researchers previously associated with mammals and birds.
Edge tracking, not center tracking
The team, led by researchers at the Howard Hughes Medical Institute’s Janelia Research Campus and the California Institute of Technology, developed a closed-loop virtual-reality system for head-fixed flies walking on an air-supported ball. A nozzle yoked wind direction to the fly’s heading, and apple cider vinegar odor was infused into the airstream based on the fly’s fictive position in a virtual corridor.
The results were unexpected. Flies did not stay inside the odor plume once they found it. Instead, they repeatedly exited after an average of 4.9 seconds inside, then executed directed, efficient returns to the plume’s boundary. The overall trajectory was not a meander, it was a systematic alternation between two states: “leaving” and “returning,” with the fly tracking a single edge of a 50-mm-wide odor corridor.
Most upwind progress occurred during the brief inside bouts. When the fly lost the odor, it did not wander randomly; it headed back along a vector that corresponded to the direction of the plume’s edge. The researchers quantified this by analyzing entry and exit angles: flies entering and leaving a 45-degree plume showed non-random angular distributions, with return trajectories often perpendicular to the edge.
Directional memory, not positional
To test whether the flies were storing a positional memory (where the plume was) or a directional memory (which way to go back), the researchers performed two key experiments. In the first, flies tracked a 45-degree plume for 10 minutes, then the odor source was switched off entirely. The flies continued heading in the same direction used for previous returns, toward where the plume had been, even though it no longer existed. In the second, the plume was shifted 20 mm laterally each time the fly exited. The flies adjusted their trajectories on the next return, confirming they were using directional rather than positional cues.
The directional memory updates rapidly. A single operant training session, in which odor was delivered only when the fly headed in a specific return direction, significantly improved tracking of a plume oriented in the opposite direction. The behavioral model that best fit the data was a two-state switching system: at each boundary crossing, the fly updated a memory vector as a weighted combination of its previous memory and its current heading direction.
The neural hardware
Within the fly’s central complex, a set of midline structures that functions as a navigation hub, the researchers identified specific neurons responsible for each component of the strategy. EPG neurons in the ellipsoid body maintained a stable representation of the fly’s heading relative to the wind. When silenced via optogenetic inhibition, flies lost their ability to make directed returns and instead wandered upwind aimlessly.
FC2 neurons in the fan-shaped body encoded the goal direction, the vector back to the plume boundary. These neurons became active specifically during return trajectories, and their activity pattern diverged from the EPG heading signal before each turn toward the boundary. Silencing FC2 neurons produced the same navigational deficit as silencing EPG neurons.
The proposed circuit model involves plastic synapses between odor-responsive tangential neurons (which carry sensory information about the plume) and columnar neurons (which encode the directional memory). Entry-angle memory is stored and updated at these synapses each time the fly crosses the plume boundary.
What this means for insect intelligence
The finding adds to a growing recognition that insect brains are not simple reflexive systems. The central complex, already known to encode heading direction and to support path integration in navigating ants and bees, now appears to also support vector-based goal memory for olfactory search. The same conserved neural toolkit, heading direction, goal direction, and state switching, serves different navigational tasks across different insect species.
The researchers note that in natural environments, the strategy works best for coherent odor plumes near the source, where sharp boundaries exist. In downwind turbulent plumes composed of disconnected filaments, the boundary information is less reliable and the memory advantage diminishes. Flies likely switch between reflexive anemotaxis (moving upwind in response to wind) and vector-based edge tracking depending on the structure of the plume they encounter.
For robotics and artificial olfaction, the strategy offers a computationally efficient template. A two-state system with directional memory can locate a chemical source without requiring a concentration gradient map, which is often unavailable in turbulent environments. The fly’s solution to the plume-tracking problem turns out to be not simpler than expected, but more elegant.
Reference: Siliciano, A.F., Minni, S., Morton, C. et al. Nature (2026). DOI: 10.1038/s41586-026-10827-7