The hippocampus is a curved structure situated deep within the medial temporal lobe of each cerebral hemisphere. It is a major component of the brain systems that form memories of events and represent relationships among places, objects, and experiences. Its name derives from the Greek words for “horse” and “sea monster,” reflecting its resemblance to a seahorse. Although sometimes described as the brain’s memory center, it neither stores every kind of memory nor acts independently of other regions.
Research on the hippocampus has connected the study of human memory with the cellular mechanisms of learning. Damage to both hippocampi can severely impair the formation of new memories while leaving immediate attention, language, and some forms of skill learning relatively intact. Recordings from individual nerve cells have also revealed representations of location, elapsed time, and sequences of experience. These findings have made the hippocampus central to neuroscience, computational accounts of memory, and the investigation of neurological disease.
Anatomical organization and regional distinctions
The hippocampus lies along the floor of the temporal horn of the lateral ventricle. In humans, its anterior end forms an enlarged head, followed by a body and a narrower posterior tail. The left and right structures occupy corresponding positions but participate in partly differentiated networks. Their functions overlap substantially; the left hippocampus is more often associated with verbal memory, whereas the right is frequently implicated in spatial memory, although neither distinction is absolute.
Terminology varies across anatomical traditions. The hippocampus proper, also called Ammon’s horn or cornu ammonis, contains the CA fields. The broader hippocampal formation usually includes the hippocampus proper, dentate gyrus, and subiculum. Some definitions extend further to adjacent transitional regions. The entorhinal cortex is a closely connected gateway between the hippocampal formation and widespread association cortex, but it is not included in every definition of the formation itself.
Unlike the six-layered neocortex, much of the hippocampal formation has a simpler, three-layered organization. In the hippocampus proper, the principal excitatory neurons are pyramidal cells, whose cell bodies form a prominent layer. Their branching dendrites extend into neighboring layers where incoming axons establish synapses. The dentate gyrus instead has a densely packed layer of granule cells, a molecular layer containing their dendrites, and an underlying polymorphic region called the hilus. These arrangements create the sharply defined bands visible in histological sections.
The CA fields differ in their cells and connections. CA1 lies next to the subiculum and is a major output stage. CA3 contains extensive recurrent connections among pyramidal neurons. CA2 is a relatively narrow region with distinctive molecular characteristics and connections; experiments in rodents implicate it in social recognition memory. The term CA4 has historically been applied to neurons within the dentate hilus, but its boundaries and use differ among classification systems. Subfield measurements therefore depend partly on the anatomical convention being used.
Organization also changes along the hippocampus’s long axis. In rodents, dorsal regions contribute strongly to precise spatial processing, while ventral regions have stronger links to emotional and stress-related networks. Human posterior and anterior regions show broadly comparable, but not identical, distinctions. Anterior portions tend to participate in more integrative and affectively connected representations, whereas posterior portions often support finer spatial detail. These are graded differences rather than sharply separated modules. Connections with the amygdala, prefrontal cortex, and other regions help explain why remembering an event can combine location, meaning, and emotional significance.
Circuitry, inputs, and outputs
A classic description of hippocampal circuitry is the trisynaptic pathway. Signals from layer II of the entorhinal cortex reach dentate granule cells through the perforant path. Granule-cell axons, called mossy fibers, contact CA3 pyramidal cells. CA3 neurons then project through Schaffer collateral fibers to CA1. This arrangement offers an experimentally accessible sequence in which information passes through distinct populations with different connectivity and physiological properties.
The trisynaptic pathway is not a complete wiring diagram. Entorhinal layer II also projects directly to CA3, while layer III sends major inputs to CA1 and the subiculum. CA1 therefore receives both processed information from CA3 and more direct cortical input. The timing and dendritic location of these inputs affect how they interact. Output returns through the subiculum and deep layers of the entorhinal cortex to cortical association networks, allowing hippocampal processing to influence regions involved in perception, knowledge, and behavioral control.
Other routes connect the hippocampus to subcortical structures. The fornix carries major projections toward septal regions and the mammillary bodies, among other destinations. Connections through the mammillary bodies, anterior thalamus, and cingulate regions contribute to a distributed memory network. Consequently, severe memory impairment can arise from damage outside the hippocampus itself. This network organization also explains why the consequences of a lesion depend on which connecting pathways remain intact, rather than only on the amount of hippocampal tissue lost.
Most principal hippocampal neurons release glutamate, an excitatory neurotransmitter. Their activity is regulated by diverse inhibitory interneurons that release gamma-aminobutyric acid, or GABA. Different interneuron classes target different parts of principal cells: some regulate the cell body and initiation of spikes, while others control dendritic integration. Inhibition limits excessive excitation, shapes the timing of cell assemblies, and helps relatively small neuronal populations represent particular experiences. The hippocampus also receives modulatory inputs using acetylcholine, dopamine, serotonin, and norepinephrine, linking its operation to attention, novelty, arousal, and behavioral state.
Circuit architecture has inspired specific computational interpretations. Sparse granule-cell activity and inhibitory control in the dentate gyrus can help distinguish similar inputs. Recurrent CA3 connections can support the retrieval of a more complete representation from a partial cue. CA1 can integrate retrieved information with incoming cortical signals. These assignments are useful hypotheses, not exclusive job descriptions: each region participates in multiple operations, and activity depends on the task, learning history, and state of the wider network. Communication is also bidirectional, with feedback loops operating within and beyond the hippocampal formation.
Synaptic plasticity and cellular mechanisms of learning
The hippocampus is a major experimental system for studying how experience changes the nervous system. A central phenomenon is long-term potentiation, or LTP: a persistent increase in the effectiveness of synaptic transmission following particular patterns of activity. In 1973, Timothy Bliss and Terje Lømo reported long-lasting enhancement of transmission in the rabbit dentate gyrus after stimulating its inputs. Related forms of potentiation have since been studied extensively in hippocampal slices and intact animals.
At many CA3-to-CA1 synapses, LTP depends on the NMDA receptor, a glutamate receptor whose ion channel is ordinarily blocked by magnesium at resting membrane potentials. Glutamate binding together with sufficient postsynaptic depolarization relieves this block, allowing calcium to enter the cell. Calcium activates signaling pathways, including calcium/calmodulin-dependent protein kinase II, that alter synaptic function. An important consequence is increased contribution of AMPA-type glutamate receptors to the postsynaptic response. This mechanism provides a way for coincident presynaptic and postsynaptic activity to strengthen a connection.
Not every form of hippocampal plasticity uses this mechanism. Potentiation at mossy-fiber synapses onto CA3 neurons is commonly expressed through presynaptic changes and can be independent of postsynaptic NMDA receptors. Synapses can also weaken through long-term depression. Whether activity produces strengthening, weakening, or little lasting change depends on factors including its frequency, timing, prior synaptic history, and neuromodulatory context. The same circuit therefore supports multiple rules for adjusting information flow rather than a single universal learning rule.
Shorter-lasting changes can arise through modification of existing proteins and receptor trafficking. More persistent forms may require new protein synthesis, altered gene expression, and structural changes in dendritic spines. A spine can enlarge, shrink, appear, or disappear as networks adapt. Synaptic tagging and capture describes an experimental framework in which recently activated synapses acquire the ability to use plasticity-related proteins produced elsewhere in the neuron. Such mechanisms help explain how a brief experience might produce a durable but selectively distributed cellular change.
LTP is not identical to memory, and demonstrating potentiation in a tissue preparation does not establish that a particular experience is stored by that change. Nevertheless, evidence connects synaptic plasticity with learning: manipulations that disrupt relevant signaling pathways can impair hippocampus-dependent tasks, and learning can alter synaptic responses. Modern studies also identify neuronal ensembles active during an experience and test whether manipulating those ensembles changes later behavior. Memories appear to involve coordinated changes across many cells and brain regions, with synaptic strength, intrinsic excitability, inhibition, and structural remodeling all contributing.
Episodic memory, relational binding, and consolidation
The hippocampus is especially important for episodic memory, the memory of personally experienced events situated in a particular context. Remembering a conversation, for example, may require linking a person’s identity, spoken words, a location, and their order in time. These elements are processed across different cortical systems. The hippocampus helps bind them into a representation that can later be accessed through only part of the original experience.
This role became clear through studies of Henry Molaison, formerly known as patient H.M. In 1953, he underwent bilateral medial temporal surgery for severe epilepsy. William Scoville and Brenda Milner described his profound memory impairment in 1957. He could maintain information briefly when attention was sustained but had great difficulty forming lasting memories of new events. He nevertheless improved on some motor-learning tasks without remembering the training sessions. The operation affected structures beyond the hippocampus, so later cases with more restricted damage were essential for refining anatomical conclusions.
The distinction between declarative and nondeclarative memory captures part of this pattern. Declarative memory includes events and facts that can be consciously reported; nondeclarative learning includes habits, skills, priming, and some forms of conditioning. The hippocampus is particularly important for rapidly acquiring flexible relations among elements of experience. A person may retain a practiced motor skill despite losing the ability to remember when it was learned. Conversely, hippocampal involvement is not determined solely by whether a task is conscious: task structure and the relationships that must be learned also matter.
Memory consolidation operates at several scales. Cellular consolidation concerns processes stabilizing changes within neurons and synapses over minutes to hours or longer. Systems consolidation concerns changing interactions among hippocampal and cortical networks over longer periods. In standard accounts, the hippocampus initially supports associations among distributed cortical representations, while repeated reactivation gradually strengthens cortical connections. Some established knowledge can consequently become less dependent on the hippocampus. This is not a literal transfer of a complete memory from one storage location to another.
The long-term role of the hippocampus in detailed autobiographical recollection remains an active theoretical question. Some accounts predict declining dependence as memories age; others propose that richly contextual recollection continues to require hippocampal processing, even when general knowledge or the gist of an event becomes cortically supported. Findings vary with lesion extent, the specificity of testing, and how richly a memory is recalled. Hippocampal activity also occurs when people imagine future events or construct novel scenes, consistent with a capacity to combine remembered elements into new relational arrangements rather than merely replay the past.
Spatial representation and the organization of experience
In 1971, John O’Keefe and Jonathan Dostrovsky described hippocampal neurons in rats that fired preferentially when an animal occupied particular locations. Such neurons became known as place cells, and their preferred locations as place fields. Across a population, their activity provides information about where an animal is within an environment. Place fields are not simple responses to a single visual feature: they can persist despite changes in individual cues and can incorporate self-motion, landmarks, goals, and behavioral context.
Place-cell activity supports the idea of a cognitive map, an internal representation of relationships that can guide flexible behavior. A familiar environment can be represented by a distributed population pattern rather than by one neuron assigned to each location. When context changes substantially, neurons may alter their preferred locations, become active or inactive, or change firing rates. These phenomena, broadly called remapping, help separate experiences occurring in environments that share some physical features.
Spatial coding is distributed beyond the hippocampus. In 2005, researchers working with May-Britt and Edvard Moser reported grid cells in the medial entorhinal cortex. These cells exhibit multiple firing fields arranged in approximately triangular lattices as an animal moves through open space. Other nearby populations encode head direction, boundaries, and movement-related variables. O’Keefe and the Mosers received the 2014 Nobel Prize in Physiology or Medicine for discoveries concerning cells constituting a positioning system in the brain. Grid-like and place-related signals have also been investigated in humans, although human tasks and recording conditions differ substantially from rodent navigation experiments.
Hippocampal representations are not restricted to physical space. Neurons can respond at particular moments during a delay or at positions within a learned sequence. These “time cells” provide a population code for temporal structure, although their firing may also depend on movement, task demands, and other changing features. Other experiments show sensitivity to routes, goals, social relationships, or abstract task structure. Such findings suggest that spatial navigation is a particularly clear example of a broader ability to represent relationships and transitions among states.
A related distinction concerns pattern separation and pattern completion. Pattern separation makes overlapping experiences more distinguishable, such as remembering where a bicycle was parked today rather than yesterday. Pattern completion retrieves a fuller event from a partial cue, such as recognizing a room from one familiar feature. Dentate and CA3 circuitry provide plausible mechanisms for these complementary operations, but neither process belongs exclusively to a single subfield. Together they address a central memory problem: preserving differences between similar episodes while still allowing reliable retrieval from incomplete information.
Network rhythms, sleep, and replay
Hippocampal activity is organized in time as well as across populations of cells. During active exploration in rodents, a prominent theta rhythm often occurs at approximately 4–12 hertz, with the exact range depending on species and behavioral conditions. Theta reflects interacting local circuits and inputs, including those from the medial septum. Human hippocampal low-frequency rhythms are more variable and can occur in slower or more intermittent forms, making direct equivalence with rodent theta an oversimplification.
The timing of spikes within these rhythms can carry information. As a rodent crosses a place cell’s firing field, the cell often fires at progressively earlier phases of the theta cycle, a phenomenon called phase precession. Across a population, this timing can organize short sequences representing locations behind, near, and ahead of the animal. Theta sequences compress behaviorally extended events into brief intervals compatible with synaptic plasticity. Their interpretation depends on task and recording conditions, but they demonstrate that average firing rate alone does not capture hippocampal coding.
Faster gamma-frequency activity accompanies local processing and communication between regions. Different gamma components can be associated with different inputs or laminar locations. Researchers investigate whether their timing helps coordinate information arriving from CA3 and entorhinal cortex. Frequency bands are descriptive tools rather than discrete channels with universally fixed functions: measured rhythms depend on electrode position, behavioral state, analysis method, and the mixture of underlying electrical sources.
During quiet wakefulness and non-rapid-eye-movement sleep, the hippocampus produces sharp-wave ripples. A sharp wave reflects coordinated synaptic activity, prominently involving CA3 input to CA1, while the associated ripple is a brief high-frequency oscillation. Rodent CA1 ripples are often described in a range around 100–250 hertz; frequencies recorded in humans are commonly lower and depend on recording conditions. Within these events, neuronal sequences related to previous experience can recur in compressed form. Replay can proceed in forward or reverse order and sometimes depicts trajectories that were not recently traversed in that exact sequence.
Replay is implicated in memory stabilization, retrieval, and planning, but these functions are not interchangeable. Experiments that selectively interrupt ripples have impaired performance on some memory tasks, strengthening the case for a causal contribution. During sleep, hippocampal events can coordinate with cortical slow oscillations and thalamocortical sleep spindles, providing windows for communication across memory networks. Physiological ripples must also be distinguished from pathological high-frequency activity in epilepsy. Similar-looking frequency components can arise from different mechanisms, so behavioral state, anatomical location, and accompanying neural activity are essential to interpretation.
Development, adult neurogenesis, and comparative organization
The hippocampus develops from the medial portion of the embryonic cerebral hemisphere. As the hemisphere grows and folds, the hippocampal formation assumes its characteristic curved position within the temporal region. Its principal neurons arise from developmental precursor populations, migrate into layers, and establish connections over an extended period. Dentate granule-cell production continues later than the generation of many other cortical neurons, and the maturation of hippocampal circuits extends well beyond birth.
Early development is accompanied by substantial changes in memory. Infants and young children can learn, but their memories differ from the detailed, enduring autobiographical recollections typical of later life. Childhood amnesia cannot be attributed to the hippocampus alone: language, self-representation, cortical maturation, and retrieval conditions also contribute. Developing hippocampal subfields and their connections with cortical networks nevertheless form an important part of the biological explanation. Immature systems are not simply smaller versions of adult circuits; they differ in excitability, connectivity, and the stability of representations.
Adult neurogenesis—the generation of new neurons in an adult brain—is well established in the dentate gyrus of laboratory rodents. Precursor cells in the subgranular zone can produce granule neurons that mature, receive inputs, and join existing circuits. The survival and integration of these cells depend on multiple conditions. Rodent studies associate adult-born neurons with aspects of learning, discrimination among similar contexts, and behavioral adaptation, but their contribution varies with the age of the cells, experimental task, and method used to manipulate them.
The extent and functional importance of adult human hippocampal neurogenesis remain unsettled. Studies using postmortem tissue, cellular markers, birth-dating approaches, and molecular methods have reached differing conclusions about how many immature or newly generated neurons persist at different ages. Tissue preservation, disease history, marker specificity, and the difficulty of distinguishing a newly born neuron from a long-lived immature one complicate interpretation. Findings from rodents therefore cannot be assumed to establish the rate or behavioral significance of the process in adult humans.
Comparative anatomy places these mechanisms in a broader evolutionary setting. Mammals share a recognizable hippocampal organization, while birds possess a homologous hippocampal formation with a different overall layout. Studies of food-storing birds have connected hippocampal organization with spatial memory demands, though comparisons must account for body size, ancestry, season, and ecology. In reptiles and other vertebrates, related medial pallial regions provide evidence about the evolutionary history of these circuits. Across species, similar problems—remembering locations, distinguishing contexts, and organizing experience—are solved by structures whose cellular layouts and behavioral specializations are not identical.
Disease, injury, and clinical significance
Bilateral hippocampal damage can cause severe anterograde amnesia, meaning impaired formation of new lasting memories. The exact pattern depends on lesion extent, neighboring tissue involvement, and the condition of the wider memory network. Unilateral injury generally produces less global deficits, sometimes with greater effects on verbal or spatial material depending on the side and task. Immediate memory span and previously learned skills may remain relatively preserved, illustrating that memory impairment is not a uniform loss of every learned capacity.
The hippocampus is especially important in temporal lobe epilepsy. A common pathological finding in surgically treated cases is hippocampal sclerosis, involving neuronal loss and gliosis with characteristic but variable subfield distributions. Seizures and tissue injury have a complex relationship: prolonged seizures can damage hippocampal circuits, while altered circuits can contribute to further seizures. Some cases show granule-cell dispersion and reorganization of mossy fibers. These findings do not occur identically in every patient and should not be treated as a single inevitable sequence.
Alzheimer’s disease commonly affects the entorhinal–hippocampal memory network early in its symptomatic course. Tau pathology, synaptic dysfunction, and neuronal loss contribute to difficulty learning and retaining new information. Hippocampal atrophy on magnetic resonance imaging can support clinical assessment, but it is not specific to Alzheimer’s disease and is not sufficient by itself to establish the diagnosis. Similar volume changes can accompany other neurodegenerative conditions, vascular injury, epilepsy, or normal aging.
Hippocampal neurons, particularly in CA1, are vulnerable to oxygen deprivation and some metabolic disturbances. Cardiac arrest, severe hypoglycemia, and certain forms of stroke can produce memory impairment when the relevant tissue is injured. Inflammation and infection can also affect medial temporal structures; herpes simplex encephalitis is a notable example. Autoimmune limbic encephalitis may involve memory disturbance and seizures, with the hippocampus affected as part of a broader inflammatory process.
Research has reported group-level hippocampal differences in conditions including depression, post-traumatic stress disorder, and schizophrenia. Such associations do not establish that hippocampal size causes a disorder, that every affected person has the same alteration, or that an individual scan can diagnose it. Medication exposure, illness duration, early experience, age, and other factors can influence results. Similarly, the hippocampus contains receptors responsive to stress-related hormones, but laboratory findings about stress and hippocampal plasticity cannot be reduced to a simple rule that all stress causes permanent damage. Clinical interpretation requires integration of symptoms, examination, cognitive testing, imaging, and other relevant evidence.
Research history, experimental methods, and interpretation
The anatomical name hippocampus is generally credited to the sixteenth-century anatomist Giulio Cesare Aranzi. For centuries, its location and shape were better understood than its function. Earlier theories connected it with smell or broad emotional circuitry. In the twentieth century, the development of cellular neuroanatomy and experimental psychology gradually shifted attention toward learning and memory. Donald Hebb’s 1949 account of activity-dependent cell assemblies provided an influential theoretical framework, although it was not a hippocampus-specific model.
The memory studies of Scoville and Milner, place-cell recordings by O’Keefe and colleagues, and physiological work on long-term potentiation established complementary lines of investigation. Each addressed a different level: consequences of human brain damage, representations carried by neuronal populations, and mechanisms that alter synaptic communication. None alone supplied a complete theory. Their convergence made the hippocampus a particularly productive system for linking behavior to circuits and cellular biology.
Human research combines neuropsychological testing with structural imaging, functional magnetic resonance imaging, and, in selected clinical circumstances, intracranial recordings. Structural scans estimate volume and tissue properties; functional imaging measures blood-oxygenation-related signals rather than neuronal firing directly. Intracranial electrodes can record local electrical activity with much greater temporal precision, but participants are generally undergoing evaluation for epilepsy, and electrode placement is determined by clinical needs. These factors constrain how broadly findings can be generalized.
Animal studies allow more direct tests of mechanism. Researchers use controlled lesions, electrophysiological recordings, calcium imaging, pharmacological interventions, and genetic techniques to identify or alter particular cells and pathways. Optogenetic and chemogenetic methods can manipulate defined populations, while behavioral tasks test navigation, contextual learning, recognition, and temporal associations. Each method has limitations: artificial stimulation may not reproduce natural firing, calcium signals are indirect measures of spiking, and changing one population can alter the state of a much larger circuit.
Interpretation requires separating measurement from explanation. A neuron’s activity may correlate with location without encoding location alone; an imaging difference may accompany learning without being its cause. High-resolution scans cannot always identify subfield boundaries with the certainty of histology, and automated segmentation systems can disagree. Statistical adjustment for total brain size, age, movement, and multiple comparisons can materially affect results. Computational models offer precise accounts of pattern separation, associative retrieval, and sequence learning, but distinct models can reproduce similar observations. Experiments that distinguish among their predictions—rather than merely showing another correlation—remain central to determining how hippocampal circuits support memory and behavior.