Spiral Brain Waves Carry Memory Beyond Fixed Wiring

In April 2026, recordings from 24 epilepsy patients showed memory tasks riding planar, spiral, source, and sink waves across the cortex. Those traveling brain waves shifted shape with the task, and a decoder read the behavior from the pattern about 70% of the time.

The wiring between neurons still changes on a scale of days and months. Behavior changes in seconds. The new maps treat the moving electric field as the layer that can retask the same tissue that fast, which is a problem for tools that still listen mainly to spikes.

Seizure Grids Caught the Storm’s Eye

The data did not come from a healthy volunteer in a scanner. They came from people already in hospital with grids and strips on the cortex so surgeons could find the tissue that starts seizures. Anup Das, a postdoctoral fellow in Joshua Jacobs’s lab at the University of Chicago, then asked those patients to hunt treasure in a virtual world and to hold lists of English letters in a Sternberg working-memory task.

Jacobs, a professor of neurology, has long specialized in these intracranial recordings. Each contact covers about 1 square centimeter of brain surface and can pick up activity in as many as 1 million cells. Put many contacts a few centimeters apart, and a wave that looked like a simple sweep from a single probe starts to show a center, a rotation, or a collapse.

Bard Ermentrout, a Distinguished University Professor of Mathematics at the University of Pittsburgh and a co-author of the April paper, had been seeing related shapes in mice. He argues that older “planar” maps may have been the outer arms of a larger storm. You feel wind in one direction until you stand at the eye.

THE 2026 HUMAN RECORDINGS

• The cohort: 24 neurosurgical patients, 9 on a spatial treasure-hunt task and 15 on a verbal Sternberg task.
• The readout: Wave shape alone classified behavior about 70% of the time, against 50% chance.
• The tempo: A given pattern held for a few tens of milliseconds, then broke and reformed.
• The scale: Contacts about 1 square centimeter across, each summing as many as 1 million cells.

That geometry is still a medical accident. Electrodes sit where seizures sit. Jacobs has said the team may be missing how common the patterns are in tissue no surgeon ever opens.

Treasure Hunts Favored Sources, Word Lists Sinks

Das, Ermentrout, Erfan Zabeh of Columbia University, and Jacobs reported the human results on April 11, 2026, in Nature Communications under the title that names the menagerie: planar, spiral, and concentric traveling waves. They filtered narrowband peaks from a 3 to 40 Hz sweep, tracked phase across the grid, and kept only epochs stable enough to name a shape.

The shapes were not a curiosity on the side of the task. They lined up with what the person was doing, and even with individual remembered items. In one case Ermentrout said Das could tell a subject had been looking at the English letter “H” from the wave pattern alone. Patterns differed across people. Remembering a location might throw a spiral in one patient and an outward source in another. Within a person, the match was tight enough to decode the behavior from the wave shape.

WAVE TYPES TIED TO THE TWO MEMORY TASKS

Jacobs told colleagues the rotating wave turned up a little more often in the spatial task, which asks for more than a letter list. He reads that as a hint that spirals suit harder memory work. The paper is careful not to claim a universal code. It claims a decoder can use the spatial pattern, and that the pattern is not a single plane.

What Rotating Waves Do in Mouse Cortex

Human grids see a patch. A mouse can be imaged across the whole sheet. On June 18, 2026, Zhiwen Ye, then at the University of Washington in Nicholas Steinmetz’s lab, reported rotating waves centered on somatosensory cortex in Science. The team combined widefield calcium imaging with Neuropixels probes in 15 mice and kept rotating events whose radius was at least 0.69 mm.

Steinmetz, an associate professor of neurobiology and biophysics, described a wave that rotates over space and time, depends on a circular circuit in sensory cortex, and pulls on activity across the brain. Ye, now at the Shenzhen Medical Academy of Research and Translation, put the anatomical question in one line: if the waves did not matter, why wire cells that way?

WHAT THE MOUSE CIRCUIT ACTUALLY DID

• The track: Axons in sensory cortex ran in a ring that matched the wave’s path.
• The mirror: Rotations on the left and right hemispheres, and between sensory and motor cortex, lined up.
• The cut: Severing the circular circuit weakened the rotating waves.
• The whisker: A puff on the left face launched clockwise rotation in right sensory cortex, with a motor-cortex partner.
• The game: Well-formed rotating waves spread when mice got a visuomotor trial right, and failed to form when they got it wrong.

Cortical and deep structures were not separate processors on those trials. They were a shared pattern. That is the closest thing in this year’s papers to a causal test: break the ring, and the hurricane fades.

Visual Cortex May Use Waves to Guess Ahead

A review in Neuron, posted July 21, 2026, and carried in the September 2 issue, tried to say what a wave is for in vision. Lyle Muller of the University of Texas at Dallas and John Reynolds of the Salk Institute, writing with Alexandra Busch and Zachary Davis, treat neural traveling waves as a way one cortical map can hold what just happened beside what is happening now, and so guess what comes next.

Reynolds first reported traveling waves in the visual system of awake animals in 2020. Those waves lined up with whether an animal noticed an object in front of it, a laboratory version of missing keys on the table. The 2026 review gives that circuit four jobs: it can change perception from moment to moment, turn recent input into an internal copy, issue a short-term prediction, and store then replay sequences that unfold in time.

Reynolds said the paper was the first single frame for what cortex can compute by having recurrent, wave-generating wiring. He compared that statistical trick to how large language models learn structure and then generate from it. The analogy is a claim about function, not a proof. It does say why a lab that once treated waves as anesthesia noise now treats them as a generative engine on a sensory map.

A Century of Planar Maps Hid the Hurricane

Hans Berger’s scalp EEG in the 1920s made alpha, beta, theta, and gamma household names in clinics. Those labels are frequencies. They do not say whether the oscillation is standing still, sliding, spinning, or collapsing. For decades, many arrays were too coarse, or too sparse, to tell.

Jacobs and Uma Mohan, now a neuroengineer in the Surgical Neurology Branch at the National Institutes of Health, had already shown that direction alone can switch with memory. In Nature Human Behaviour on March 8, 2024, they used theta- and alpha-band (2 to 13 Hz) waves in 160 epilepsy patients, 93 of whom showed traveling waves. In episodic memory, those waves tended to travel in a posterior-to-anterior direction during successful memory encoding and the reverse, front to back, during recall. In one cluster the recall direction was strongest 865 ms before the word came out.

Mohan’s reading was practical. Visual cortex sits at the back, prefrontal memory systems at the front. A back-to-front wave is a decent signature of packing a sight into memory. A front-to-back wave is a decent signature of pulling it out. The 2026 paper does not throw that axis away. It says the axis was the visible edge of a larger set of shapes.

HOW THE WAVE MAP GOT SHARPER

1. 1920s: Scalp EEG shows frequency bands that rise and fall with sleep, attention, and memory, without a clear spatial form.
2. 2020: Reynolds’s group reports traveling waves in awake visual cortex that track whether an animal detects a stimulus.
3. March 8, 2024: Mohan, Jacobs, and colleagues show 2 to 13 Hz waves reverse direction between encoding and recall in 93 of 160 patients.
4. April 11, 2026: Das, Zabeh, Ermentrout, and Jacobs add spirals, sources, and sinks in 24 patients and decode behavior at about 70%.
5. June 18, 2026: Ye, Steinmetz, and colleagues find rotating waves on a circular axon layout in 15 mice and weaken them by cutting that circuit.
6. July 21, 2026: Muller, Busch, Davis, and Reynolds argue traveling waves can carry a short-term prediction across a visual map.

The brain still spends about half its energy holding electrochemical gradients so neurons can fire on demand. The new work says some of that budget is spent on patterns that move, not only on spikes that jump a synapse.

Miller’s Herders Meet a Synapse-Only Rebuttal

Earl K. Miller, a cognitive neuroscientist at MIT’s Picower Institute, has been the loudest translator of these maps into a theory of thought. He says the literature is leaving the question of whether waves matter and treating them as a major motif of how cortex processes information. In monkey prefrontal cortex, his group found rotating waves after a distraction, with fuller rotations on trials the animals got right. He has called those rotations herders that steer cortex back onto the right path.

Miller’s picture is physical. Neurons sit close to spike threshold. A passing field can tip the voltage at synapses and make a cell more or less likely to fire. Anatomy is the long-term store. Waves are how that store gets expressed as a thought on the timescale of a decision. He has also argued, in later theoretical work from the Picower group, that interfering waves can run analog computations in parallel in a way a step-by-step digital circuit cannot.

György Buzsáki, a systems neuroscientist at New York University, does not buy the field as a computer. In his account, current is made at synapses. Cells compute, spike, and talk to each other. You do not need anything extra happening in the extracellular space. The stronger and more synchronous the population, the bigger the field. The pattern is a useful diagnostic, like a heartbeat, not a second code.

WHERE EXPERTS DISAGREE

• Miller: Traveling waves are a flexible control layer that can excite or inhibit tissue faster than anatomy can rewire, and they recover a computation after a distraction.
• Buzsáki: The field is a byproduct of synaptic current. Read it to learn about the circuit, but do not give the water around the cells a job the cells already do.
• Ye and Steinmetz: A circular axon layout, a cut that weakens the rotation, and better-formed waves on correct trials are awkward facts for a pure byproduct story.

Both sides can live with the 70% decoder. A byproduct that tracks a letter you are holding is still a signal. The fight is whether you should stimulate the wave itself, or only the synapses under it.

Stimulation Plans Aim at Wave Shape

That fight is no longer academic for people building memory aids. Jacobs has already sketched the next experiment in public: if a traveling wave helps a given kind of encoding, maybe you can push that wave with stimulation and help someone remember. Ermentrout’s metaphor is blunter. Neural activity can ride a wave like a surfboard. If the wave is too weak to travel, the word never arrives.

If a traveling wave is relevant for a certain kind of memory encoding, then if you want to help a person remember things better, maybe you can apply brain stimulation that strengthens that kind of traveling wave.

Joshua Jacobs, professor of neurology, University of Chicago

The University of Chicago summary of the work is explicit about the devices in mind. Characterizing the patterns is a step toward mathematical models of each wave, then toward brain stimulation that strengthens that kind of traveling wave, including brain-computer interfaces and transcranial magnetic stimulation for cognitive decline. Mohan made the same bet in 2024 from the opposite direction: if someone’s waves are moving the wrong way at recall, a well-timed pulse might turn them around.

Most implants still decode spikes, local power, or a smoothed potential. They do not decode a spiral versus a sink. If Das’s decoder is even partly right, that is wasted structure. If Buzsáki is right, stimulating the field is a roundabout way of stimulating synapses, and the pretty geometry is a map, not a lever. The mouse cut, the 70% human readout, and the reverse-direction memory waves are already on the table. The lever has not been pulled in a patient.

Until it is, the practical fact is smaller than the theory and sharper than the old engine-noise story. In 24 people, the shape of a traveling wave, held for tens of milliseconds on a seizure grid, was enough to say what the person was doing about 70% of the time, including the letter in front of their eyes.