The Hippocampus

Shape, name, and location

The hippocampus is a paired structure lying deep in the medial temporal lobe, curling along the floor of the temporal horn of the lateral ventricle, one on each side. In coronal section it has a distinctive curved, in-rolled form, and it was that appearance that gave it its name: sixteenth-century anatomists thought it looked like a seahorse, and hippocampus is the Greek for exactly that, a sea-horse or sea-monster.

The hippocampus does not stand alone. It sits at the core of a larger complex, the medial temporal lobe memory system, which also includes the entorhinal, perirhinal, and parahippocampal cortices wrapped around it, and the amygdala just in front. Anatomists sometimes use the term "hippocampal formation" to mean the hippocampus proper together with the dentate gyrus and the subiculum. It matters to be aware of this, because the classic amnesia cases involve damage to the whole complex, not to the hippocampus alone, and careless writing that credits everything to "the hippocampus" glosses over a genuine ambiguity.

Unlike the six-layered neocortex, the hippocampus is allocortex, an older kind of cortex with only three layers. This simpler, more regular architecture is one reason it became the workhorse of cellular neuroscience: its circuit is laid out neatly enough to be traced, sliced, and recorded from, and a very large fraction of what is known about synaptic plasticity was learned in hippocampal slices.

The trisynaptic circuit

The hippocampal circuit is famous for being one of the few in the brain that can be drawn as a simple line. Information flows largely in one direction through three successive synapses, hence the name, and this orderly arrangement is what made the hippocampus the classic preparation for studying how synapses change with experience.

1. Entorhinal cortex: the front door

   Almost all cortical input to the hippocampus is funnelled through the entorhinal cortex, which gathers highly processed information from association areas across the brain. Its output fibres, called the perforant path, penetrate into the hippocampus.

2. Dentate gyrus (first synapse)

   The perforant path synapses onto the granule cells of the dentate gyrus. Because there are far more granule cells than entorhinal inputs, and because they fire sparsely, the dentate is thought to perform pattern separation: pulling similar inputs apart so that similar experiences do not become confused.

3. CA3 (second synapse)

   Granule cells send powerful mossy fibres to the pyramidal cells of CA3. CA3 is unusual in being densely connected back onto itself, a recurrent network, which makes it well suited to pattern completion: reconstructing a whole memory from a partial cue, which is what happens when a fragment of a song brings back an entire afternoon.

4. CA1 (third synapse)

   CA3 projects to CA1 by the Schaffer collaterals. This is the single most studied synapse in neuroscience: it is where long-term potentiation was first demonstrated, the lasting strengthening of a synapse that follows intense use and that remains the leading cellular model of memory.

5. Subiculum and out

   CA1 passes to the subiculum, the main output stage, which projects back to the entorhinal cortex, closing the loop, and onward through the fornix to the mammillary bodies and the anterior thalamus.

Two features of this circuit are worth holding on to. The first is that it is a loop: cortex to entorhinal to hippocampus to entorhinal to cortex. The hippocampus does not sit at the end of a pipeline; it sits in the middle of a circular one, which is exactly what a structure that must bind cortical activity together and hand it back would need. The second is that its two computational hallmarks are opposites. The dentate gyrus separates similar things; CA3 completes partial things. Memory needs both: you must not confuse today's parking spot with yesterday's, but you must be able to recall today's from a single glimpse of the car park.

The case of H.M.

Henry Molaison was twenty-seven years old and had suffered from intractable epilepsy for years when, in 1953, the surgeon William Scoville removed a large portion of his medial temporal lobes bilaterally, including most of the hippocampus, the amygdala, and the surrounding cortex. It was an experimental procedure, undertaken because nothing else had worked.

The seizures did improve. But Molaison emerged with something no one had predicted. His intelligence was intact, and remained so: he could hold a conversation, reason, and score normally on IQ tests. His perception was intact. His memory for his childhood and for the years before the operation was largely preserved. What he could not do, from the day of the operation until his death in 2008, was form a new lasting conscious memory. He would read the same magazine repeatedly without recognising it. He met his doctors thousands of times and never remembered any of them. He could hold information in mind for as long as he rehearsed it, and it was gone the moment he was distracted.

Scoville and the psychologist Brenda Milner published the case in 1957, and it changed neuroscience. Before it, memory was widely assumed to be a diffuse property of the brain as a whole, not something with an address. H.M. proved otherwise. Three conclusions followed, and all three still stand.

That third finding is the one that most reshaped psychology. Skill learning of the kind H.M. retained depends on the basal ganglia and cerebellum, not the medial temporal lobe. The modern taxonomy of memory types, declarative versus procedural, explicit versus implicit, descends directly from what one patient could and could not learn.

Forming memory is not storing memory

It is tempting to conclude from H.M. that memories live in the hippocampus. They do not, or at least not for long, and the same case proves it. H.M. could remember his childhood. Those memories had been formed decades earlier, and they survived the removal of the tissue that had formed them. If the hippocampus were the store, they would have gone with it.

The prevailing account is systems consolidation. When an experience occurs, the cortex represents its various elements, the sight, the sound, the place, the feeling, in the areas that handle each. The hippocampus binds these scattered fragments into an index: a pointer that can reactivate the whole ensemble. Over subsequent weeks, months, and years, particularly during sleep, the hippocampus repeatedly reactivates these cortical patterns, and the cortex gradually builds direct connections between them. Eventually the memory can be retrieved without the hippocampal index at all. See memory consolidation for the fuller treatment.

This predicts, and clinical observation confirms, a temporally graded retrograde amnesia: hippocampal damage tends to spare distant memories while impairing those from the years immediately before the injury, which had not yet finished consolidating. It is a strange and specific prediction, and the fact that it holds is one of the strongest arguments for the model.

The hippocampus is therefore not a warehouse. It is closer to a librarian who catalogues new acquisitions and, over the years, hands the volumes over to the permanent stacks. Take the librarian away and the existing collection stands, but nothing new is ever shelved.

Place cells and grid cells

In 1971 John O'Keefe and Jonathan Dostrovsky reported something unexpected from electrodes in the rat hippocampus. Individual pyramidal neurons fired vigorously when the animal was in one particular part of its enclosure, and were silent everywhere else. Move the rat to another spot and a different cell took over. They called them place cells, and they proposed that the hippocampus was, among other things, a map.

The finding was met with resistance. The hippocampus was the memory structure; what was it doing representing space? But the result held, and it was extended. In 2005 Torkel Hafting, Marianne Fyhn, and Edvard and May-Britt Moser reported cells in the entorhinal cortex, one synapse upstream, whose firing fields were not single locations but a repeating triangular lattice covering the entire environment. These grid cells tile space like graph paper, providing a metric, a coordinate system, that the hippocampus can use. Related cells were found nearby: head-direction cells signalling which way the animal faces, border cells signalling proximity to a boundary.

O'Keefe and the Mosers shared the 2014 Nobel Prize in Physiology or Medicine for this work, which established that the brain contains an explicit, cellular representation of space, and that the entorhinal-hippocampal system is where it lives.

Memory and space, one system

For a while the memory account and the map account of the hippocampus looked like rivals. They are now generally seen as two faces of one function. An episodic memory is a memory of something that happened somewhere, at some time; the "where" is not an optional extra but part of the structure of the memory itself. A system that binds elements into a coherent scene, and that can reconstruct that scene from a partial cue, is exactly what both remembering an event and navigating an environment require.

The evidence supports the unification. Human hippocampal neurons show place-like responses during virtual navigation. Patients with hippocampal amnesia are impaired not only at remembering the past but at imagining the future or an unfamiliar scene, producing accounts that are fragmentary and lacking in spatial coherence. Whatever the hippocampus does, it seems to be needed for constructing a scene, whether that scene is remembered, imagined, or currently being walked through.

A widely discussed finding concerns London taxi drivers, who must memorise a vast street network to qualify. Structural imaging has repeatedly found their posterior hippocampi to be larger than those of controls, with the difference related to years of experience, and studies following trainees over time have found the change emerging with training. It is a striking demonstration of neuroplasticity in an adult human, though it should be read carefully: it concerns a specific, extreme, spatially demanding expertise, and it is not evidence that memory training generally enlarges the brain.

A fragile structure

For all its importance, the hippocampus is among the most vulnerable tissue in the brain, and it fails in several distinct ways.

Hypoxia and ischaemia. Hippocampal pyramidal cells, particularly in CA1, are exceptionally sensitive to a shortage of oxygen; the region is sometimes called the vulnerable sector for this reason. After cardiac arrest or a period of severe oxygen deprivation, CA1 neurons can be lost while much of the rest of the brain survives, producing an amnesic syndrome in a patient who is otherwise cognitively intact.

Epilepsy. The hippocampus is a common origin of temporal lobe seizures, and repeated seizures can in turn damage it, producing the scarring known as hippocampal sclerosis. This was the reason for H.M.'s surgery, and it remains a reason for surgery today, though modern operations are unilateral precisely because of what his case taught.

Chronic stress. The hippocampus is densely populated with receptors for cortisol, the principal human stress hormone, and it is part of the feedback loop that shuts the stress response down. Sustained high cortisol is associated with dendritic atrophy in hippocampal neurons and with reduced hippocampal volume in conditions involving prolonged stress. A structure that regulates the stress axis is itself damaged by that axis running unchecked, which is an unhappy piece of design.

Alzheimer's disease. The pathology of Alzheimer's disease appears early in the entorhinal cortex and hippocampus, before it spreads widely through the cortex. This is the anatomical reason the earliest and most characteristic symptom is difficulty forming new memories, while older memories and other faculties are relatively preserved at first. Hippocampal atrophy is visible on MRI and is one of the structural markers used in assessing the disease.

Finally, the hippocampus is one of the very few regions where new neurons may be generated in the adult brain, in the dentate gyrus. The extent to which this occurs in adult humans remains genuinely contested, with careful studies reaching opposite conclusions, and it should not be presented as settled. What is not in doubt is that it happens robustly in rodents and that it matters there for pattern separation.

Related topics