In 2012, Tadashi Miyamoto and his colleagues identified a single receptor in the Drosophila brain that functions as an internal fuel gauge. The receptor, Gr43a, senses fructose circulating in the hemolymph. When the fly has eaten, hemolymph fructose rises, Gr43a detects it, and the brain registers a full tank. When the fly is hungry, fructose drops, and Gr43a reports empty. The receptor is both necessary and sufficient for this measurement: remove Gr43a, and the fly loses the ability to tell whether it has fed.
For fourteen years, that was the complete story. A fructose sensor. A fuel gauge. Known, characterized, contained.
A sensor that starts lying
Raquel Francés and her colleagues in the Brain Plasticity Unit at CNRS and ESPCI Paris published a finding in Nature in 2026 that broke the containment. They trained well-fed Drosophila to associate an odor with mild electric shocks, using the standard aversive olfactory learning protocol. The training was spaced: multiple rounds separated by rest intervals, the same structure known to produce durable memory across species.
After spaced training, the Gr43a neurons started behaving strangely. In well-fed flies, the fructose sensors responded to hemolymph sugar as though the animal were fasting. The sensor had not broken. It had been reprogrammed. Each spaced training session progressively disinhibited the fructose-sensing neurons, moving them incrementally toward a fasting-state reading despite satiation.
The fly was full. Its brain reported hunger.
The chain
What follows is a four-step cascade that crosses from metabolic sensing to memory storage with no direct neural connection between the two.
The disinhibited Gr43a neurons, responding to fructose as though the fly were starving, release thyrostimulin. Thyrostimulin is a heterodimeric glycoprotein hormone composed of two subunits, Gpa2 and Gpb5, belonging to one of the oldest hormone families known. The Gpa2/Gpb5 pairing predates the split between invertebrates and vertebrates, conserved for over half a billion years.
Gr43a neurons have no projections to the mushroom body, the insect brain's center for learned associations. Thyrostimulin reaches it through the hemolymph, a long-range hormonal signal. It binds Lgr1 receptors on the α/β Kenyon cells, the specific mushroom body subpopulation that stores associative memory. The Kenyon cells respond by shifting their metabolism: enhanced mitochondrial pyruvate uptake in the axons, detectable within two hours of training. This metabolic reprogramming is what consolidates the memory.
Break any link and long-term memory vanishes while short-term memory remains untouched. Silence Gr43a neurons: long-term memory gone. Knock down Gpb5, the thyrostimulin subunit: gone. Knock down Lgr1, the mushroom body receptor: gone. Three independent interruptions, the same result. The pathway is necessary and specific to consolidation.
One more test: feed the fly coconut oil instead of sugar after training. No memory. The pathway requires sugar, not calories. The phantom hunger signal works only when there is real sugar for the forged sensor reading to detect.
Why spacing works
Massed training, the same number of shock-odor pairings crammed into a single session, fails to produce long-term memory in Drosophila. This has been documented for decades. Standard explanations invoke consolidation windows, synaptic saturation, or the need for rest between encoding events.
Francés's data point to a different mechanism. The Gr43a disinhibition accumulates across spaced sessions. A single session nudges the sensor partway toward the fasting reading. A second session pushes further. By the end of spaced training, the accumulated disinhibition crosses a threshold: the sensor reports phantom fasting in a fed fly.
Massed training fails because the disinhibition has no time to build. The sessions sit too close for the incremental shift to accumulate. Spacing provides the construction schedule for a phantom signal.
Borrowed infrastructure
The metabolic consolidation pathway that Gr43a and thyrostimulin activate was never built for memory. Pierre-Yves Plaçais and colleagues in the same ESPCI group had previously shown that long-term memory formation in Drosophila requires elevated energy metabolism in the mushroom body. The mitochondrial pyruvate uptake, the glucose metabolism shift: this is metabolic machinery that the brain uses during actual fasting to manage energy under scarcity.
The problem evolution faced was how to run this pathway in a well-fed fly. Building a dedicated memory consolidation system would require new genes, new receptors, new regulatory architecture. Evolution found a cheaper solution: forge the activation signal. After aversive training, Gr43a neurons learn to report hunger on demand. The existing metabolic pathway activates as though fasting were real. Consolidation runs on borrowed infrastructure.
The fasting is fictional. The consolidation is real.
The most interesting question is where else this is happening. Forging a phantom version of a physiological state to commandeer an existing pathway may be a strategy evolution uses broadly. We would miss it, because the standard approach to studying a pathway is to look for the real signal that activates it. Nobody looked at a fructose sensor and asked what happens when it starts counterfeiting.