A zebrafish swims against a current it cannot overcome. Neurons fire with each attempt. Astrocytes — the cells that were supposed to be passive structural support, the glue between neurons — accumulate calcium in proportion to the number of failed attempts. At some threshold, the astrocytes release ATP, which enzymes convert to adenosine, which activates inhibitory neurons in the hindbrain. The fish stops swimming.
Disable the astrocytes with a laser. The fish never stops. Activate them artificially. The fish stops immediately. The stop signal doesn't come from neurons. It comes from the cells that neuroscience spent decades treating as scaffolding.
This finding (Chen, Ahrens et al., Science, May 2025) is one of three papers published simultaneously in Science, from different labs studying different organisms, that all reach the same conclusion: astrocytes actively control brain state, not just support it. Freeman's group showed the same in fruit flies. Guttenplan and Papouin's team demonstrated it in mice — astrocytes as the circuit effector through which norepinephrine produces behavioral adaptations.
The convergence across species matters. If astrocytes were performing a derived, specialized function in one organism, you might explain it as an evolutionary quirk. Three organisms spanning vertebrates and invertebrates suggests the control function is ancient and fundamental. Neurons handle the fast signaling — the millisecond-scale computation at synapses. Astrocytes handle the slow integration — the seconds-to-minutes timescale of tracking whether actions are working.
The anatomical arrangement makes this plausible in retrospect. A single astrocyte can envelop hundreds of thousands of synapses with its tendrils. It is positioned to sense aggregate network activity in a way no individual neuron can. It does not need to participate in the rapid-fire signaling to influence the outcome. It monitors the sum and adjusts the system's operating state.
The caffeine connection sharpens the picture. Adenosine receptors — specifically A2 receptors — are the downstream target of the astrocyte stop signal. Caffeine is an adenosine receptor antagonist. It blocks the signal that says “this isn't working, stop trying.” The subjective experience of caffeine — persistence, alertness, reluctance to quit — may not be primarily about neurons at all. It may be about silencing the astrocyte-mediated futility detector.
Marc Freeman, who led the fruit fly work, said it directly: “We live in the age of connectomics, where everyone loves to say if you understand the connections between neurons, we can understand how the brain works. That's not true.” The connectome — the complete wiring diagram of neural connections — has become the prestige project of neuroscience. But the connectome doesn't include astrocytes. It maps the circuit and ignores the controller. If 99% of neuroscientists don't consider astrocytes in their studies (Freeman's estimate), then 99% of neuroscience is studying the instrument while ignoring the hand that plays it.
The deeper structural point is about what counts as the functional unit of a system. Neuroscience assumes the functional unit is the neuron. The connectome project assumes the functional unit is the synapse. Both are correct at their timescale — milliseconds for synaptic transmission, seconds for neural circuit dynamics. But the astrocyte operates at a third timescale — minutes — and its function is not to compute but to decide when computation is futile. It is the part of the brain that knows when to give up. And until three labs in three organisms proved otherwise, it was classified as glue.