A second signal in the brain: how a 2018 mechanism could redraw the map of animal navigation
Eight years on from a quiet lab finding about how animals learn to find their way, the mechanism is being reframed as a candidate 'second signal' that couples reward to direction — with consequences well beyond neuroscience.
On 10 July 2026, the neuroscience aggregator @_TheTransmitter republished a long-form recap of a finding that has spent almost a decade threading through conferences without quite breaking out: a mechanism by which animals appear to learn how to navigate. The recap, surfaced via researcher Cremieux Recueil, credits work first sketched in 2018 and now associated with Yvette Fisher's lab at UC Berkeley — a group that has spent years pulling apart the dopaminergic circuitry that lets a fruit fly walk across a featureless box and still know where it is.
The interesting wrinkle is not the phenomenon itself. It is the architecture. Rather than relying on a single neuromodulator — dopamine, the long-celebrated "reward" signal — to stamp in a useful direction, the system seems to require two: one carrying information about whether the animal's behaviour worked, and another carrying something closer to which direction the animal was heading when the reward arrived. Separated, neither fully explains learning. Combined, they appear to draw the internal map that a navigating animal relies on the next time it sets out.
The news peg is modest. No journal retraction, no paradigm collapse, no Nobel-watch angle. What it does is reframe a finding that had been sitting in a niche corner of fly neuroscience into a more general story about how the brain binds two streams of information into a single memory.
The move that looks like counting
Navigation, in the lab setups Fisher and her peers use, is stripped down to something almost embarrassing to watch. A tethered fruit fly sits on a small ball surrounded by a panoramic LED display. The LEDs project a single bright stripe that the fly can rotate by walking. The stripe moves; the fly walks to keep it stable. Then the experimenters flip the world — rotate the visual cue a fixed number of degrees, without the fly doing anything to cause it — and reward the animal only if it walks back to the original orientation. A fly, in other words, is being asked to hold a heading in working memory, then update it against a reward.
What Fisher's group reported, beginning in 2018, is that the dopaminergic neurons that deliver the reward signal do not by themselves make the fly learn to hold the right heading. The animal's internal "compass" — a population of neurons in the ellipsoid body, called compass cells, that tile 360 degrees of direction — only locks in if a second signal is paired with the dopamine at the moment of reward. That second signal appears to mark which compass cells were firing when the animal was rewarded. Without it, the brain has no way of knowing which direction the reward belonged to.
The implication is bracing in its simplicity. To learn "this heading is good," the brain needs two facts at once: that a reward happened, and where the animal was heading when it did. That is a hypothesis the field is now taking seriously across vertebrates as well. The write-up circulated on 10 July frames the mechanism as a candidate answer to one of the oldest puzzles in associative learning — why a generic reward signal does not, by itself, write the right memory to the right place.
A counter-frame older than the field
The dominant story in this corner of neuroscience has, for decades, been elegantly parsimonious: dopamine is the brain's broadcast reward signal, and learning happens wherever dopamine lands. Reinforcement-learning theory, in particular, treats a single positive prediction error as sufficient to teach a network anything it is in a position to learn. That framing has produced an enormous amount of working AI.
It is also, the navigation literature has been quietly insisting for years, slightly too clean. Real animals — and reinforcement-learning systems that imitate them — struggle to learn in conditions where the same reward could plausibly belong to many different actions. The classic problem is called "credit assignment," and the textbook answer until recently has been to engineer the network so that credit is forced onto one channel at a time. Fisher's mechanism proposes a biological equivalent: the brain solves credit assignment by tagging the reward with a second, simultaneous signal indicating where the animal was when the good thing happened. The recap that surfaced this week reads it as the field's first real candidate for a cellular solution to the problem.
This is a story about two signals, not one. And once a second signal is on the table, the obvious next question is whether it is one thing — a single "tagging" neuromodulator — or several.
The structural picture, without the scaffolding
Strip the neuroscience jargon out and what the lab has been building toward is an architecture for memory that is genuinely different from the textbook version. The textbook says: a positive outcome strengthens whatever behaviours preceded it. The newer picture says: a positive outcome strengthens whichever behaviour happened to be ongoing in a specific internal coordinate — here, heading — at the moment the reward was delivered. The distinction is the difference between a system that learns what and a system that learns where, and what, together.
That is not a subtle refinement. It changes what kinds of questions a memory system can answer. A system that learns only whether a behaviour was rewarded can adapt: do more of that. A system that also learns where the animal was when the reward landed can plan: return to this orientation the next time, because that is where the reward is. The first is operant conditioning; the second is the precondition for anything one would call navigation.
A research programme that runs in this direction matters beyond flies. Mammalian place cells, grid cells and head-direction cells are all variations on the same architectural idea: a self-organising map of where the animal is at each moment, against which reward signals can later be written. If a second neuromodulatory signal is needed to bind reward to heading even in fruit flies, the likely — though still unproven — inference is that something analogous operates in rodent and human brains as well. That would re-frame a generation of place-cell results.
What to watch next
Three things move the story. First, replication in another lab, with another species, in another navigation task — the field is unlikely to treat the second-signal model as canonical until a second group independently shows the dissociation. Second, identification of the second signal at the molecular level. The recap published this week is intentionally agnostic about what carries the directional tag. Resolving that — whether it is acetylcholine, octopamine, or something else — is the obvious next paper. Third, translation into mammalian navigation: rat head-direction cells, mouse place cells, and ultimately whether humans carry a comparable dual-tagging system for episodic memory.
The mechanism is not yet established science in the heavyweight sense. There is no public dispute, but there is no replicated, cross-species consensus either. What this week's recap really does is upgrade a 2018 finding that had been gathering dust at fly conferences into a question the rest of neuroscience will be expected to take a position on.
The Monexus culture desk reads the 10 July write-up against the laboratory work it derives from rather than against the popular frame of "dopamine = reward," which has begun to look, in this corner of the field, like a useful simplification rather than a full description.
Wire provenance
This editorial synthesis draws on the following public wire/social posts:
- https://x.com/cremieuxrecueil/status/2074501871335817217