Hippocampus and Spatial Memory

A place cell is a pyramidal neuron in the hippocampus, most heavily studied in areas CA1 and CA3, that increases its firing rate dramatically — often from near-zero baseline to tens of spikes per second — whenever an animal is located within a restricted region of its environment, called the neuron's place field. Any given cell in a given environment has one or a few place fields, typically covering roughly 10-30% of a small open arena for dorsal CA1 cells, and different cells have fields in different places, so that across the whole population, the entire environment is tiled. Critically, the same cell will have a place field in one environment and no reliable field, or a field in a completely different location, in another environment. This context-dependence is called remapping, and it is the key evidence that place cells encode an abstract, learned representation of space rather than reacting directly to any single sensory cue. A grid cell, discovered three decades later in the medial entorhinal cortex (MEC), which is the principal cortical input to the hippocampus, fires not at one location but at multiple locations that form a strikingly regular hexagonal lattice tiling the entire environment the animal explores. The spacing between firing fields and the size of each field increase systematically along the dorsal-to-ventral axis of the MEC, so that dorsal grid cells have small, closely spaced fields (tens of centimeters apart) and ventral grid cells have large, widely spaced fields (meters apart). This provides a multi-scale metric — a ruler — for space, analogous to how different spatial frequencies build up a complex image. Together with head-direction cells (which fire only when the animal's head points in a particular compass direction, independent of location), border cells (which fire specifically near environmental boundaries), and speed cells (whose firing rate scales with the animal's running speed), grid cells supply the raw metric and self-motion signals that the hippocampus combines with sensory landmark information to construct the cognitive map — a term coined by Edward Tolman in 1948 and given a cellular mechanism by O'Keefe and Nadel's 1978 book The Hippocampus as a Cognitive Map. This map is not a static picture but a dynamically updated, self-localizing coordinate system, continuously corrected by both external landmarks and internally generated movement signals via path integration (dead reckoning).

Step 1 — Sensory and self-motion inputs converge on entorhinal cortex. Visual landmark information from parietal and visual association cortices, vestibular and proprioceptive self-motion signals (via the head-direction circuit, which originates in brainstem vestibular nuclei and passes through the lateral mammillary nuclei and anterodorsal thalamic nucleus), and olfactory/contextual information all converge on the entorhinal cortex, the gateway to the hippocampus.

Step 2 — Grid cells compute a path-integrated metric. Within MEC layer II/III, grid cells appear to combine head-direction input and running-speed input to perform path integration: continuously updating an internal estimate of position by integrating velocity over time, the neural equivalent of dead reckoning on a ship. Because path integration alone accumulates error, grid cell firing is periodically reset and corrected using visual landmark information reaching entorhinal cortex from postrhinal and perirhinal cortex. Computational models (attractor network models) propose that grid firing emerges from a two-dimensional neural sheet of recurrently connected cells whose activity forms a bump (or several bumps) that moves across the sheet in a direction and speed set by the animal's actual movement — an idea supported by the discovery that grid cells with the same spacing and orientation form functional "modules" that shift coherently.

Step 3 — Entorhinal grid, border, and head-direction signals project to hippocampus. MEC layer II projects to the dentate gyrus and CA3 via the perforant path; MEC layer III projects directly to CA1. This is the anatomical substrate for the classic trisynaptic circuit: entorhinal cortex → dentate gyrus → CA3 → CA1 → back to entorhinal cortex (deep layers), which then projects back out to neocortex. CA3 also has extensive recurrent collaterals connecting CA3 pyramidal cells to one another, forming an autoassociative network capable of pattern completion — retrieving a whole stored spatial pattern from a partial or degraded cue.

Step 4 — Hippocampal place fields emerge from combinatorial convergence. A CA1 or CA3 place cell's firing field arises from the convergence of many overlapping, differently-spaced, differently-phased grid cell inputs, plus direct sensory/contextual input via the perirhinal and postrhinal-to-entorhinal-to-hippocampal route, plus recurrent CA3 dynamics. Because many different combinations of grid phases can, in principle, sum to produce a single localized bump of excitation at one place and only one place, this system can in theory generate an enormous number of distinguishable place codes — one of the proposed computational advantages of using periodic grid codes as a basis set, analogous to a Fourier decomposition of position.

Step 5 — Synaptic plasticity stores and stabilizes the map. When an animal first enters a novel environment, place fields are somewhat unstable and can drift or update over the first ten to twenty minutes of exploration. NMDA-receptor-dependent long-term potentiation (LTP) at the Schaffer collateral synapses (CA3→CA1) and at perforant path synapses (entorhinal cortex→dentate gyrus/CA3) is thought to stabilize which afferents drive which cell, converting an initially labile pattern of activity into a stable, reliable place field that can persist for weeks. Pharmacological or genetic blockade of NMDA receptors or of the CaMKII pathway central to LTP (see Long-Term Potentiation) degrades both place field stability and spatial learning in the water maze, directly linking the cellular plasticity mechanism to the behavioral phenomenon.

Step 6 — Remapping reconfigures the code, not the geography. When the environment, the task, or even the animal's internal state changes sufficiently, CA3 and CA1 populations can undergo global remapping: a largely new, uncorrelated set of cells becomes active with new place fields, as if the population had switched to an entirely different "map" for a different context. Smaller changes (like moving a single wall) can produce rate remapping: the same cells keep firing in the same locations but change their firing rates. This dual coding scheme lets the hippocampus represent both "this is a different place" (global remapping) and "this is the same place but something changed" (rate remapping).

O'Keefe and Dostrovsky, 1971. John O'Keefe and Jonathan Dostrovsky recorded single hippocampal units in freely moving rats using chronically implanted microelectrodes, a technique that itself required years of development to allow stable single-unit recording in an unrestrained animal. They found that certain hippocampal neurons fired selectively when the rat was in a particular part of the testing platform, regardless of the rat's orientation and largely independent of any single sensory cue they tested. O'Keefe called these "place cells" and proposed, radically for the time, that the hippocampus was not merely a memory register but the neural instantiation of Tolman's cognitive map. This was published as a short report in Brain Research in 1971, followed by the fuller theoretical treatment with Lynn Nadel in the 1978 book The Hippocampus as a Cognitive Map.

Key control experiments establishing that place fields are not simple sensory responses. Subsequent work showed that place fields persist when the lights are turned off, that they remain stable when a prominent cue card is moved unless the animal's frame of reference is fully disrupted, and that rotating a single landmark rotates the place field correspondingly — showing the cell uses landmarks to anchor an internally generated map rather than firing in direct response to the landmark itself. Muller and Kubie (1987) systematically manipulated cue cards and environment boundaries and demonstrated that place fields depend on the geometric relationship of the animal to the whole configuration of cues, not any one cue in isolation.

Remapping experiments. Bostock, Muller, and Kubie (1991), and later Leutgeb and colleagues (2004, 2005) working in the Moser laboratory, showed that changing the shape of the environment, the color of the walls, or the behavioral context could cause an abrupt, large-scale reorganization of which cells fire where — global remapping — while smaller manipulations produced graded rate changes without relocation of fields — rate remapping. This distinguished the hippocampal code as a flexible, combinatorial system rather than a fixed topographic map, analogous to a telephone switchboard that can be rewired for each new context rather than a photograph that is simply redrawn.

Fyhn, Molden, Witter, Moser, and Moser, 2004; Hafting, Fyhn, Molden, Moser, and Moser, 2005. Recording from the medial entorhinal cortex, the Moser laboratory in Trondheim discovered cells whose firing fields, when plotted over the whole area an animal explored, formed a near-perfect hexagonal lattice — the grid cell. Critically, this periodic structure appeared even in a completely empty, cue-poor arena, showing that the pattern was internally generated rather than externally imposed by the room's geometry. Grid spacing and field size increased systematically along the dorsoventral axis of MEC, matching a long-standing observation that place field size in hippocampus also increases along the corresponding dorsoventral (septal-to-temporal) axis.

Discovery of head-direction cells and border cells rounding out the system. Head-direction cells were first described by Ranck (1984) and characterized in detail by Taube, Muller, and Ranck (1990) in postsubiculum; border cells were described by Solstad, Boccara, Kropff, Moser, and Moser (2008) in MEC, firing specifically along one or more walls of an enclosure regardless of its size or shape. Together with grid cells, these give the entorhinal-hippocampal system a full instrument panel: compass heading, distance-to-boundary, and metric position.

Human evidence. Patient H.M., who underwent bilateral medial temporal lobe resection in 1953 for epilepsy and was studied extensively by Brenda Milner, could not form new long-term declarative memories, including new spatial memories of places he had never seen before his surgery, directly implicating the hippocampus in human spatial-episodic memory. Eleanor Maguire's structural MRI studies of London taxi drivers (2000) found significantly enlarged posterior hippocampal gray matter volume correlating with years of navigation experience, and later single-cell recordings during epilepsy surgery (Ekstrom, Kahana, Fried and colleagues, 2003) directly recorded place-cell-like and view-cell-like activity in humans navigating virtual environments, extending the rodent findings to our own species. In 2014, the Nobel Prize in Physiology or Medicine was awarded jointly to John O'Keefe and to May-Britt Moser and Edvard Moser "for their discoveries of cells that constitute a positioning system in the brain."

Example 1 — Reading a grid cell's spatial autocorrelogram. A researcher records a putative grid cell in dorsal MEC while a rat forages in a 1.5 m × 1.5 m square box. The raw firing-rate map shows six-to-seven discrete firing fields arranged so that each field has six roughly equidistant neighbors. The spatial autocorrelogram of this rate map shows a clear central peak surrounded by six peaks at 60° intervals, and a gridness score (a standard measure comparing the autocorrelogram's rotational symmetry) of 0.82, well above the typical threshold of ~0.3-0.4 used to classify a cell as a grid cell in most laboratories. The distance between neighboring fields (grid spacing) measures 42 cm. Question: what would you predict for a grid cell recorded simultaneously in a ventral MEC electrode in the same animal, and why? Reasoning: Given the well-established dorsoventral gradient in which grid spacing and field size increase moving ventrally, the ventral cell should show a larger spacing — empirically, spacing increases from roughly 30 cm near the dorsal pole to several meters near the ventral pole of MEC, changing in a discrete, stepwise fashion across "modules" rather than continuously. So the ventral recording should show a coarser hexagonal lattice, larger individual fields, and fewer repeats visible within the same 1.5 m box (possibly appearing as only one or two fields since the whole box may fit inside a single grid period). This illustrates the multi-scale ruler idea: dorsal modules give fine spatial resolution over short distances, ventral modules give coarse resolution needed for large-scale environments, and combining several modules of different scale allows a very large, non-repeating range of unambiguous positions to be encoded, much as combining base-2 and base-10 digits of different place value lets a small number of digits encode a huge range of numbers.

Example 2 — A remapping experiment, predicting the outcome. An experimenter records 60 place cells simultaneously from CA1 while a rat explores a gray circular arena (Environment A), then places the same rat in a black square arena in a different room (Environment B), then returns it to the gray circular arena (A again). In Environment A, 35 of the 60 cells show a clear, stable place field; in Environment B, a substantially different, largely non-overlapping subset of about 38 cells shows fields, and correlations between the A rate maps and the B rate maps for individual cells average near zero. When the rat is returned to A, the original A place fields reappear at high spatial correlation (~0.85) with the first A session. Question: is this global remapping or rate remapping, and what does the return-to-A result rule out? Reasoning: This is global remapping: a largely different population of cells becomes active, not merely a change in firing rate of the same cells at the same locations (which would instead be rate remapping, seen typically with subtler changes like altering wall color alone while keeping shape and room constant). The high correlation upon return to A (~0.85, comparable to test-retest reliability within a single stable session) rules out the possibility that the CA1 network was simply drifting randomly or degrading due to fatigue, electrode movement, or nonspecific novelty; instead it shows that A and B are each represented by a distinct, stored and retrievable map that the network can reinstate on demand, consistent with the CA3 autoassociative network performing pattern completion for a well-learned map while orthogonalizing genuinely different contexts, a computation attributed to the pattern-separation function of the dentate gyrus upstream.

John O'Keefe was born in New York City in 1939 to Irish immigrant parents and trained initially in psychology, receiving his PhD from McGill University in 1967, where the physiological psychology tradition of Donald Hebb (whose theoretical work underlies much of what became Long-Term Potentiation research) was strongly influential. O'Keefe moved to University College London, where he began recording single hippocampal neurons in freely behaving rats at a time when almost all electrophysiology was done in anesthetized or restrained preparations, a technically demanding choice that made the 1971 discovery possible. He spent essentially his entire career at UCL, patiently building up the case for the cognitive map theory over four decades against considerable early skepticism, since the dominant view at the time treated the hippocampus mainly through the lens of the H.M. amnesia case as a generic "memory structure" rather than a spatial one. May-Britt Moser and Edvard Moser, both born in Norway in 1963, met as psychology students, married, and pursued postdoctoral training partly in O'Keefe's own laboratory before founding the Kavli Institute for Systems Neuroscience in Trondheim. Their systematic mapping of the entorhinal cortex, moving the recording electrode step by step across the dorsoventral axis, led to the serendipitous discovery of the hexagonal grid pattern, illustrating how a hypothesis-neutral, careful anatomical survey can reveal structure nobody had predicted from theory alone.

Alzheimer's disease and early diagnosis. The entorhinal cortex, particularly layer II, is among the earliest sites of neurofibrillary tangle pathology in Alzheimer's disease, and grid-cell-like signals measured with fMRI or MEG in humans performing virtual navigation tasks are being explored as sensitive early biomarkers, potentially detecting circuit dysfunction before clinical memory symptoms or measurable hippocampal atrophy appear.

Traumatic brain injury and epilepsy surgery. Because medial temporal lobe structures are common surgical targets and common sites of injury, understanding the trisynaptic circuit informs both surgical planning (trying to spare tissue critical for spatial and episodic memory) and rehabilitation strategies for patients with post-traumatic spatial disorientation.

Rodent models of neuropsychiatric and neurodegenerative disease. Place-cell and grid-cell recordings in mouse models of Alzheimer's disease, schizophrenia, and stress-related disorders are a standard tool for linking molecular pathology (amyloid plaques, altered synaptic plasticity, disrupted GABAergic interneurons) to concrete, quantifiable deficits in spatial coding, giving a translational bridge between cellular mechanism and behavioral impairment.

Brain-computer interfaces and neuroprosthetics. Decoding algorithms originally built to reconstruct an animal's position from place-cell population activity are now foundational methods in the broader field of neural decoding, feeding directly into the population-level analyses used throughout Neural Circuits and Network Dynamics and into experimental hippocampal memory prostheses aimed at restoring memory function after injury.

Understanding memory beyond space. Human fMRI studies find grid-cell-like hexagonal signals during purely conceptual "navigation" through abstract feature spaces (for example, mentally comparing objects varying along two continuous dimensions), suggesting the entorhinal-hippocampal metric system may be repurposed generally for organizing knowledge, a finding actively shaping current theories of how the brain represents structured knowledge and schemas, and connecting spatial cognition research directly back to general theories in Learning and Memory Systems.

Ask the tutor: How would place cell and grid cell firing be expected to change in an animal with early entorhinal cortex tau pathology, before any detectable hippocampal atrophy?