Place Cells and Grid Cells Diverge on the Same Open Field

Put a rat in an open box and record from its brain, and two very different pictures of space appear. Place cells in the hippocampus fire at one location. Grid cells in the entorhinal cortex fire at many, arranged in a hexagonal lattice. The divergence is real, reproducible, and still not fully explained by any single circuit model.

The Open Field Paradox

The same arena produces two firing patterns that look nothing alike. A place cell is sparse: it fires when the animal crosses one small region, the place field, and stays quiet elsewhere. A grid cell is periodic: it fires whenever the animal passes through any vertex of a repeating triangular grid that tiles the whole floor.

Both cell types sit in the same medial temporal lobe system, connected by dense reciprocal projections. The hippocampus receives input from the entorhinal cortex, and the entorhinal cortex receives return projections from the hippocampus. Yet the spatial codes they carry differ in scale, symmetry, and stability.

That divergence matters because it rules out the simplest circuit models. If grid cells were just place cells summed together, the two should share a common spatial signature. They do not. Grid periodicity is largely independent of place field location, and place fields remap when the environment changes while grid patterns often hold their spacing.

How Place Cells Were Found

John O'Keefe recorded single units from the hippocampus of freely moving rats in the early 1970s. The animals foraged for food pellets scattered across a small open platform while a movable microdrive advanced fine wires into the brain. O'Keefe noticed that some pyramidal neurons fired only when the rat was in a particular part of the enclosure.

The finding was not immediate. Earlier recordings in restrained animals had shown responses to various stimuli, and the spatial correlate only became obvious once the animal could move freely. The key procedural choice was behavioral: let the rat walk, and record for long enough to sample the whole arena.

Place fields turned out to be stable across sessions. A cell that fired in one corner on Monday often fired in the same corner on Tuesday, as long as the room cues stayed put. Rotate the cues, and the field rotates with them. That stability made the cognitive map idea concrete: the hippocampus holds a usable representation of location, not just a response to sensory input.

The recordings also showed that place cells are not the only spatial signal in the hippocampus. Interneurons fire differently, and some pyramidal cells have multiple fields. The sparse, location-locked pattern is a common case, not a universal rule.

Even within the hippocampus, the proportion of active place cells varies with behavioral state. During sleep or quiet rest, the same cells that fired during exploration replay their sequences in compressed time, a phenomenon that has been studied as a potential mechanism for memory consolidation. That replay complicates the simple view of place cells as purely sensory-driven.

The Grid Cell Discovery

Edvard and May-Britt Moser's lab reported grid cells in the dorsocaudal medial entorhinal cortex in 2005. The cells fired at regular spatial intervals as rats explored an open field, and the firing locations formed a hexagonal lattice. The pattern was not obvious in single passes; it emerged from plotting many spikes across a long recording session.

The procedural details mattered. The arena was large enough to contain several grid vertices, the recording lasted long enough for the animal to cover the floor repeatedly, and the spike sorting separated the target cell from nearby units. Without any one of those, the periodicity could have been missed or misread.

Independent confirmations followed in bats, monkeys, and eventually humans, though the human work relies on different methods and smaller samples. The core finding, a periodic spatial code in entorhinal cortex, has held up across labs and species. The exact lattice spacing varies along the dorsoventral axis, from roughly 30 centimeters in dorsal regions to larger spacing in ventral regions of the rat entorhinal cortex.

The Moser lab also found other spatial cell types in the same region: head-direction cells, border cells, and speed cells. The entorhinal cortex is not a single-purpose grid machine. It carries a mixed population of signals that collectively support navigation.

Procedural Choices That Shape Results

Arena size and wall geometry change what you see. Small enclosures can truncate grid fields or compress place fields against the walls. Square, circular, and rectangular arenas produce different boundary effects, and a lab that switches shapes mid-study can introduce apparent remapping that is really a geometry artifact.

Tetrode versus silicon probe yield changes the sample. Tetrodes give good isolation for a small number of cells; high-density silicon probes give hundreds of units at once but require more careful spike sorting. A study that reports grid cells from tetrodes and place cells from silicon probes is comparing two different sampling regimes, not just two cell types.

Spike sorting thresholds decide which units make the cut. An isolation distance that is too loose lets noise in; one that is too strict throws away real cells. Because grid cells are a minority population, small changes in sorting can shift the reported proportion substantially. This site has covered a related problem in hippocampal replay research, where larger samples changed the effect size.

Behavioral sampling rate sets the resolution of the spatial map. If position is sampled at 30 hertz and spikes are binned into 2-centimeter pixels, the map looks clean. Coarser tracking or larger bins smear the lattice and can make a grid cell look like a place cell with multiple fields.

What the Divergence Tells Us

Grid cells look like a path integrator. Their periodic firing depends on self-motion cues, and they maintain spacing when external landmarks are removed. Place cells look more like sensory anchors: they lock onto landmarks and remap when those landmarks move. The two codes could serve different jobs in the same navigation system.

The trade-off is that neither code is sufficient alone. A pure path integrator accumulates error over time. A pure landmark map fails when landmarks are absent. The system likely combines both, with grid cells providing a metric and place cells providing a stable reference. How the combination works at the circuit level is still an open question.

Model-organism caveats apply. Rat and mouse data dominate the literature, and the human evidence comes from intracranial recordings in patients with electrodes implanted for clinical reasons. Those samples are small and not representative. Inferences about human spatial cognition from rodent grids should be stated as hypotheses, not results.

A related piece on this site argues that lab notebooks and code repositories diverge on reproducibility. The same logic applies here: the published figure is not the same as the recorded session, and the gap between them is where replication fails.

Practical Steps for Replication

Match arena dimensions across labs. If you are comparing grid spacing to a published value, use the same enclosure size and shape, or report the difference explicitly. A 1-meter square arena and a 1-meter circular arena are not interchangeable for grid analysis.

Report spike sorting parameters in full. Include the isolation distance, the refractory period violation rate, and the software version. A reader who cannot reproduce your unit selection cannot reproduce your cell counts.

Use open-source tracking tools and publish the raw position data. DeepLabCut and similar packages have made pose tracking accessible, but the tracking output still needs to be shared alongside the electrophysiology. Without it, the spatial map cannot be re-derived.

Share raw electrophysiology data. Several archives now accept high-density probe recordings, and the storage cost is falling. A grid cell result that ships with its raw traces is far more useful than one that ships with only the rate map. This site has made a similar case for publishing the controls alongside the result.

Finally, report the proportion of cells that showed the effect, not just the examples. Grid cells are a minority in entorhinal recordings, and place cells are a minority in hippocampal recordings. The proportion is part of the result.

Open Questions and Future Directions

One unresolved issue is how grid cells and place cells interact during learning. Some models propose that grid cells provide a stable metric that place cells use to anchor their fields, while others suggest that place cells teach grid cells about the environment. Disentangling these possibilities requires simultaneous recordings from both regions in the same animal, which is technically demanding but increasingly feasible with high-density probes.

Another open question concerns the role of theta oscillations. Both place cells and grid cells show phase precession relative to the theta rhythm, but the relationship between the two phase codes is not fully understood. Some studies suggest that theta provides a temporal reference that helps bind the two spatial codes, but the evidence is still correlational.

Finally, the developmental trajectory of these cells remains an active area. Grid cells appear later in development than place cells, and their emergence coincides with the maturation of entorhinal-hippocampal circuits. Understanding how these codes are built during development could inform models of how they might degrade in aging or disease.