Theta Rhythms Coordinate Hippocampal–Prefrontal Interactions in a Spatial Memory Task

A rat is running a figure-eight maze. Picture it: a narrow central arm with two reward points at the ends, and a barrier that forces one turn on the first pass while leaving the choice open on the second. Every time that animal pauses at the decision point and picks a direction, two brain regions that sit centimeters apart must briefly agree. The hippocampus holds the spatial map — where the rat just came from. The medial prefrontal cortex holds the rule — alternate. The question Matthew Jones and Matthew Wilson set out to answer is how those two structures coordinate. The answer turns out to be a rhythm. The core problem is one of architecture. The brain is a collection of specialists. The hippocampus builds a spatial map of the environment through place cells — neurons that fire when the animal occupies a particular location. The medial prefrontal cortex, or mPFC, handles rules, context, and decision-making. These are different jobs, handled by different tissue. But spatial working memory — holding a recently visited location in mind to guide the very next choice — demands both at once. The hippocampus must send its spatial information upstream to the prefrontal cortex, which needs it to apply the alternation rule correctly. Jones and Wilson's proposal is that this transfer doesn't happen continuously. It happens selectively, exactly when the task demands it, and theta oscillations in the four to twelve hertz range are the mechanism. To test this, Jones and Wilson trained six rats on a continuous alternation task using a figure-eight maze. Each trial paired a forced-turn sample run with a subsequent choice test run. The forced-turn epoch didn't require the rat to remember anything — the barrier made the decision for it. The choice epoch required memory: the animal had to recall which arm it had just visited and go to the other one. Correct choices were rewarded with chocolate. Across eight recording sessions, rats performed at eighty-three percent correct on average — well above chance but imperfect enough to generate informative error trials. Running speeds on the critical central arm were similar across choice and forced-turn conditions, around thirty-eight to forty centimeters per second, which ruled out movement differences as a confound. The technical achievement here matters. Jones and Wilson implanted adjustable tetrode arrays in both dorsal CA1 of the hippocampus and the ipsilateral medial prefrontal cortex simultaneously. Tetrodes are multi-wire electrodes that record from nearby neurons and separate their spikes by comparing signal amplitudes across the wires. This simultaneous bilateral recording — one hundred forty-nine CA1 neurons and one hundred sixty-five mPFC neurons isolated across all sessions — let the team ask not just what each region was doing, but whether they were doing it together. The first answer came from spike cross-correlations: a measure of how consistently neurons in one region fire close in time to neurons in the other. Jones and Wilson found that on forced-turn runs, when no working memory was needed, the mean peak cross-correlation between CA1 and mPFC neuron pairs was 0.009. During correct choice runs, it jumped to 0.024 — nearly three times higher, with a p-value below 0.01. During error trials, it fell back to 0.015, significantly reduced compared to correct choices. This selectivity is the key point. The two regions weren't simply correlated whenever the animal was running. They became correlated specifically when the task required hippocampal information to reach prefrontal decision circuits, and only when that transfer succeeded. The spike correlations are just the behavioral signature. The mechanistic story is in the local field potentials — the slow oscillations that reflect coordinated activity across thousands of neurons in each region. Jones and Wilson measured coherence between CA1 and mPFC local field potentials across the frequency spectrum. The result was clean: a significant peak appeared only in the theta band, four to twelve hertz, and only during correct choice epochs. Mean theta-band coherence was 0.32 during correct choice runs, compared to 0.19 during forced-turn runs and 0.20 during error trials — both significantly lower than the correct choice value. At delta frequencies, below four hertz, coherence hovered around 0.12 to 0.16 regardless of epoch. Nothing significant appeared above twelve hertz. The signal was theta, and only theta. Phase-locking adds the single-neuron layer to this picture. In CA1, seventy-one percent of pyramidal cells showed significantly nonuniform phase distributions relative to the local theta oscillation — they preferred to fire at particular moments in the theta cycle rather than randomly. In mPFC, forty-nine percent of neurons were significantly phase-locked to the CA1 theta rhythm. But here's where the behavioral dependence becomes vivid: mPFC neurons' phase-locking was not constant. The circular concentration coefficient — a measure of how tightly spikes cluster around a preferred phase, where zero means random and higher values mean more concentrated — was 0.19 during correct choice epochs and dropped to 0.10 during both forced-turn runs and error trials. CA1 neurons, by contrast, maintained similar phase-locking across all three conditions, around 0.43 to 0.53. The hippocampus kept its rhythmic beat steady. The prefrontal cortex only locked onto it when memory-guided decisions were demanded. One additional number points toward directionality. Shifting the CA1 local field potential forward by an average of thirty milliseconds maximized the mPFC phase concentration. CA1 appears to be leading mPFC — the spatial signal originates in the hippocampus and is received by the prefrontal cortex, not the other way around. Think of it this way: the theta rhythm works like a shared metronome. When CA1 and mPFC are both locked to the same beat, spikes from the two regions land in the same roughly fifty-millisecond windows repeatedly, cycle after cycle. That temporal alignment is what produces the elevated cross-correlations. The regions don't need a direct hardwired amplification of their connection — they just need to agree on the timing. And they do, but only when the task calls for it. The error trials make the argument most sharply. When the rat chooses wrong, all three measures of coordination collapse. Spike cross-correlations drop from 0.024 to 0.015. mPFC phase-locking drops from 0.19 to 0.10. Theta coherence drops from 0.32 to 0.20. The error values land almost exactly on top of the forced-turn baseline — the condition where no memory transfer was needed at all. Jones and Wilson interpret this as coordination being not just present during successful memory use, but specifically absent when memory fails. The two regions don't stay synchronized and still produce an error — they fail to synchronize, and that failure is the error. Jones and Wilson are careful about what they conclude from this. They're not claiming that theta coordination is the only thing happening during spatial working memory, or that it uniquely defines the hippocampal-prefrontal circuit. What they propose is that theta-frequency coupling may be a general mechanism — a way for any two distant brain regions to create temporary, task-specific connections without permanent anatomical rewiring. The coordination is dynamic: it rises when behavior demands interregional communication and relaxes when it doesn't. Specialized regions can still encode their own information independently while occasionally synchronizing to share it. The clinical shadow here is schizophrenia. Patients show spatial working memory deficits, and the disorder is associated with altered function of GABAergic interneurons — the cells that help generate and maintain oscillatory rhythms — in both the hippocampus and prefrontal cortex. If theta rhythms are the medium through which the hippocampus passes spatial information to prefrontal decision circuits, then disrupting those rhythms would break the transfer without necessarily damaging either region's independent function. The behavior would look like a memory failure. The mechanism would be a timing failure. The elegant core of what Jones and Wilson found is this: the brain doesn't hard-wire regions together. It uses rhythms to connect them when needed and release them when not. A rat running a figure-eight maze, pausing at the center, choosing correctly — that moment of correct choice is also a moment when two brain regions are briefly, precisely, rhythmically synchronized. When the choice is wrong, they aren't. The rhythm isn't decoration. It's the decision. This lecture was created by ennepō. Go to https://ennepo.ai to Discover, Create and Follow the latest research in your field. Read when you can. Listen when you want to.