Long-Term Potentiation

Long-term potentiation is an experience-dependent, persistent increase in the strength of a chemical synapse, typically induced by brief, high-frequency ("tetanic") presynaptic stimulation and expressed as a larger postsynaptic response to each subsequent presynaptic spike. It is measured electrophysiologically as an increase in the slope of the field excitatory postsynaptic potential (fEPSP), or in the amplitude of excitatory postsynaptic potentials/currents (EPSPs/EPSCs) recorded from single neurons, and the increase can last from hours (in a slice) to many weeks (in an intact, behaving animal). LTP is not one phenomenon but a family. The best-studied form, at the Schaffer collateral–CA1 synapse in the hippocampus, is NMDA-receptor-dependent, associative, and input-specific: it requires a coincidence of presynaptic glutamate release and postsynaptic depolarization (associativity, the cellular embodiment of Hebb's postulate), and it strengthens only the synapses that were active during induction, leaving neighboring, inactive synapses on the same cell untouched (input specificity, also called synapse specificity). A related property, cooperativity, describes the fact that a minimum number of afferent fibers must fire together to reach the depolarization threshold for induction — weak single-fiber stimulation fails, but the same fibers succeed when recruited alongside others. Other hippocampal and cortical synapses show LTP with different rules — the mossy fiber–CA3 synapse, for example, expresses a presynaptic, NMDA-receptor-independent form triggered by calcium entry through voltage-gated channels and cAMP-PKA signaling rather than by coincidence detection. This article focuses on the canonical NMDA-dependent CA1 form, because it is both the best mechanistically dissected and the form most directly linked, by pharmacological and genetic evidence, to learning and memory. Its natural counterpart in the bidirectional plasticity toolkit is Long-Term Depression and Spike-Timing Plasticity, which uses the same NMDA receptor but different calcium dynamics to weaken rather than strengthen synapses.

Step 1 — Baseline transmission. At rest, a presynaptic action potential in a Schaffer collateral axon releases glutamate, which acts on two postsynaptic ionotropic receptor types co-localized in the dendritic spine: NMDA and AMPA receptors. AMPA receptors are fast, ligand-gated cation channels that mediate the bulk of baseline excitatory transmission — they open promptly upon glutamate binding and close within a few milliseconds. NMDA receptors also bind glutamate, but at resting membrane potential their channel pore is physically occluded by an extracellular Mg²⁺ ion lodged inside the pore; glutamate binding alone is not sufficient to open them.

Step 2 — Coincidence detection. Tetanic stimulation (classically, a 100 Hz train for 1 second, or 4 pulses at 100 Hz repeated as "theta-burst" patterns closer to endogenous hippocampal rhythms) causes rapid, repeated AMPA-receptor-mediated depolarization of the postsynaptic spine. This depolarization electrostatically expels the Mg²⁺ blocking ion from the NMDA receptor pore. Only when depolarization (from strong or summed presynaptic drive) coincides with glutamate binding (from presynaptic release at that same synapse) does the NMDA receptor conduct current. This dual requirement — ligand AND voltage — is precisely why the NMDA receptor is a molecular AND-gate, the physical substrate of Hebb's postulate.

Step 3 — Calcium influx as the trigger. Once unblocked, the open NMDA receptor conducts both Na⁺ and, critically, Ca²⁺. The resulting rise in postsynaptic spine calcium concentration is the necessary and sufficient trigger for LTP induction: injecting calcium chelators (e.g., BAPTA) into the postsynaptic cell blocks LTP even when NMDA receptors are intact, while directly elevating postsynaptic calcium can induce potentiation without any NMDA receptor activation at all. The amplitude and time course of the calcium transient also determines direction of plasticity — large, fast calcium rises favor potentiation (LTP), while smaller, prolonged rises favor depression (LTD), a dichotomy explored further in Long-Term Depression and Spike-Timing Plasticity.

Step 4 — Kinase cascade. Elevated calcium, together with calmodulin, activates CaMKII (Ca²⁺/calmodulin-dependent protein kinase II), the single most important enzyme in LTP induction. CaMKII autophosphorylates at Thr286, which locks it into a persistently active state even after calcium falls back to baseline — a molecular memory switch. Active CaMKII, along with PKC and PKA acting in parallel, phosphorylates AMPA receptor subunits (notably GluA1) to increase their single-channel conductance, and phosphorylates other synaptic proteins that promote receptor trafficking.

Step 5 — AMPA receptor trafficking and expression. The dominant mechanism of early LTP expression is the rapid insertion of additional AMPA receptors into the postsynaptic density, driven by CaMKII-dependent exocytosis of receptor-containing vesicles and by anchoring proteins (such as PSD-95 family scaffolds) that stabilize newly delivered receptors at the synapse. More AMPA receptors per synapse means a larger postsynaptic current for the same amount of released glutamate — this is what a physiologist actually measures as "potentiation."

Step 6 — Early vs. late phase. Early-phase LTP (E-LTP), lasting roughly 1–3 hours in vitro, depends entirely on covalent modification and trafficking of existing proteins — it survives protein-synthesis inhibitors. Late-phase LTP (L-LTP), lasting many hours to weeks, requires new gene transcription and translation. Persistent kinase activity (PKA, and MAPK/ERK) phosphorylates the transcription factor CREB (cAMP response element-binding protein), driving expression of plasticity genes and structural proteins. This late phase is accompanied by structural remodeling — enlargement of existing dendritic spines and, over longer time, growth of entirely new spines and synapses — converting a transient biochemical change into a durable anatomical one.

Step 7 — Retrograde signaling (a caveat). Some early studies suggested that LTP also has a presynaptic component — an increase in glutamate release probability — communicated by a retrograde messenger (nitric oxide has been proposed) traveling from the potentiated postsynaptic spine back to the presynaptic terminal. This remains more contested than the postsynaptic AMPA-trafficking story and is best presented as a partially resolved side mechanism rather than settled fact.

Bliss and Lømo (1973). The founding experiment was performed by Timothy Bliss and Terje Lømo in the anesthetized rabbit hippocampus. They stimulated the perforant path (input from entorhinal cortex to the dentate gyrus) with brief high-frequency trains and recorded the evoked field potential in dentate granule cells. A single tetanus (e.g., 10–20 pulses at 100 Hz) increased the slope of the population EPSP and the amplitude of the population spike, and this enhancement persisted for hours in the anesthetized preparation and — in a companion chronic study in awake, freely moving rabbits — for days to weeks. This was the first direct demonstration that a mammalian central synapse could undergo a use-dependent, long-lasting increase in efficacy, exactly the property theorists had predicted but never observed.

Collingridge, Kehl, and McLennan (1983) — the NMDA receptor requirement. A decade later, Graham Collingridge and colleagues showed that bath application of AP5 (D-2-amino-5-phosphonovaleric acid), a selective NMDA receptor antagonist, blocked the induction of LTP at the Schaffer collateral–CA1 synapse in hippocampal slices without affecting baseline synaptic transmission. This was the critical pharmacological dissection: normal, moment-to-moment synaptic transmission is carried mostly by AMPA receptors and survives AP5 intact, but induction of the lasting change specifically requires NMDA receptor activation. It directly established the receptor-level coincidence-detection mechanism, because AP5 selectively removes the receptor whose voltage-dependence gives the synapse its Hebbian, associative property.

Malenka, Kauer, Zucker, and Nicoll (1988) — postsynaptic calcium is necessary and sufficient. Using intracellular injection of the calcium chelator BAPTA into the postsynaptic CA1 neuron, this group showed that LTP could be completely prevented even with normal presynaptic stimulation and intact NMDA receptors — proving calcium's necessity postsynaptically, not merely correlating with it. Complementary experiments showed that directly elevating postsynaptic calcium (e.g., via photolysis of caged calcium) could produce potentiation without any afferent tetanus at all, establishing sufficiency. Together these experiments pinned down the causal chain to a single, specific postsynaptic ion.

Morris, Anderson, Lynch, and Baudry (1986) — LTP and behavior. Richard Morris and colleagues asked whether blocking LTP pharmacologically would also block hippocampus-dependent spatial learning in behaving animals — not just in a dish. Rats received chronic intraventricular infusion of AP5 and were trained on the Morris water maze, a task requiring the animal to find a submerged, invisible platform in an opaque pool using distal spatial cues, a task known to depend on an intact hippocampus. AP5-treated rats showed substantially impaired spatial learning (longer latencies and more circuitous swim paths to the platform, and reduced preference for the correct quadrant on probe trials with the platform removed) compared with saline-treated controls, while the same drug dose left visually-cued, non-spatial versions of the task (platform visible) essentially unaffected — a crucial control showing the deficit was specific to spatial/hippocampal memory rather than to general sensorimotor or motivational impairment. Because the same AP5 concentration that impaired learning also blocked LTP induction in slices from treated animals, this experiment provided the strongest single piece of evidence linking the cellular phenomenon to the behavioral one, though it does not by itself prove LTP is the memory trace — correlation of drug effects across two assays is compelling but not identical to a mechanistic proof of identity.

Later genetic and optogenetic refinements. Subsequent work using genetically modified mice — for example, mice lacking specific NMDA receptor subunits selectively in the hippocampus, or CaMKII autophosphorylation-dead mutants (T286A) — reproduced both the LTP deficit and the spatial learning deficit with molecular specificity unavailable to pharmacology alone. More recent optogenetic experiments (e.g., Nabavi et al., 2014) have shown that experimentally inducing LTP-like potentiation, or its reversal via LTD-like depression, at defined synapses can respectively strengthen and abolish a specific associative fear memory, moving the field from "LTP correlates with memory" toward "LTP-like changes can causally write and erase a specific memory trace."

Example 1 — Reading an LTP dose-response and specificity experiment. A researcher records fEPSP slope from CA1 in a hippocampal slice. Baseline slope is stable at 1.0 mV/ms for 20 minutes. She delivers a weak tetanus to pathway A (a single 1-second, 100 Hz train to a small bundle of Schaffer collaterals) while pathway B (an independent, non-overlapping bundle of fibers onto the same postsynaptic cells) receives only low-frequency test pulses throughout. Results: pathway A's fEPSP slope jumps to 1.6 mV/ms immediately after the tetanus and stabilizes near 1.5 mV/ms (50% potentiation) for the remaining 3 hours of recording. Pathway B remains at 1.0 mV/ms throughout. Reasoning: because only the tetanized pathway (A) potentiates while the untetanized pathway (B), sharing the same postsynaptic cell, does not change, this demonstrates input specificity — LTP is confined to synapses that were themselves active during induction, not spread indiscriminately across the whole cell. If she now delivers a second, much weaker tetanus to a third, minimal-fiber pathway (C) alone, and it fails to potentiate, but succeeds when fired simultaneously with pathway A's tetanus (even though C's own fiber activity is unchanged), this demonstrates cooperativity/associativity: weak input can be potentiated if it is paired with strong, coincident depolarization from another convergent input, exactly the AND-gate logic of the NMDA receptor. Quantitatively, if she then applies AP5 (50 μM) before the tetanus on a fresh slice, pathway A's slope after tetanus stays at 1.05 mV/ms — essentially no potentiation — while a control slice given vehicle shows the usual 1.5 mV/ms. The percentage potentiation blocked is: (1.5 − 1.05) / (1.5 − 1.0) × 100% ≈ 90% of the LTP is AP5-sensitive, consistent with the interpretation that the great majority of Schaffer collateral–CA1 LTP is NMDA-receptor-dependent, while a small NMDA-independent residual may reflect other mechanisms (e.g., voltage-gated calcium channel contributions).

Example 2 — A clinical/pharmacological reasoning case. A new investigational drug, "Compound X," is found in radioligand-binding assays to bind the NMDA receptor's glycine co-agonist site with high affinity and to act as a partial agonist there (glycine/D-serine binding at this site is required, together with glutamate at the main site, for full NMDA receptor gating). In hippocampal slices, low doses of Compound X given alone slightly enhance LTP magnitude compared with vehicle, but high doses profoundly impair LTP. In freely-moving rats, low-dose Compound X mildly improves water-maze acquisition, while high-dose Compound X produces severe spatial-learning deficits indistinguishable from AP5. How should a student reason through this? Compound X is a partial agonist at an obligatory co-agonist site. At low doses, it may occupy some previously-unoccupied glycine sites on receptors that were glycine-limited, boosting NMDA receptor function modestly (mimicking a "shortage" being partially corrected) — consistent with the mild LTP and learning enhancement. At high doses, however, a partial agonist increasingly out-competes the natural full agonist (endogenous glycine/D-serine) for the same site while itself producing only sub-maximal channel activation upon occupancy — net effect is a reduction in NMDA receptor current despite receptor occupancy, phenocopying an NMDA antagonist like AP5. This predicts, and the data confirm, that high-dose Compound X should mimic AP5 in both slice electrophysiology and water-maze behavior, again reinforcing (via a second, independent pharmacological manipulation of the same receptor) the causal link between NMDA-receptor function, LTP, and spatial learning — while also illustrating why partial agonists can have biphasic, dose-dependent, seemingly paradoxical effects.

Timothy Bliss trained as a neurophysiologist and, together with the Norwegian physiologist Terje Lømo, carried out the founding experiments as part of Lømo's doctoral work under Per Andersen in Oslo in the late 1960s, publishing the definitive results in 1973 after Bliss joined the effort at the National Institute for Medical Research in London. Lømo had originally been investigating basic synaptic transmission properties in the perforant path when he noticed that test stimuli delivered shortly after a tetanizing train produced anomalously large responses — an observation that could have been dismissed as an artifact of electrode drift or anesthesia, but which the pair pursued rigorously with careful controls (input specificity across independent pathways, dependence on stimulation pattern, persistence over many hours) that transformed a curious blip into a repeatable phenomenon worth naming. Graham Collingridge, working in Britain in the early 1980s, supplied the pharmacological dissection that turned LTP from a described phenomenon into a mechanistically tractable one, by identifying the specific receptor gate (NMDA) that separates ordinary transmission from lasting plasticity. Richard Morris, developing the water maze task specifically to test hippocampal spatial memory, then closed the loop between molecule and behavior.

LTP-related mechanisms sit at the center of research on learning disabilities and intellectual disability syndromes: Fragile X syndrome and several forms of autism spectrum disorder involve mutations affecting proteins that regulate AMPA receptor trafficking or mGluR-dependent plasticity, and mouse models of these conditions show altered (often exaggerated LTD or deficient LTP) synaptic plasticity that correlates with their behavioral phenotypes, making plasticity assays a standard preclinical screen for candidate therapies. In Alzheimer's disease research, soluble amyloid-beta oligomers are shown to directly impair LTP induction in hippocampal slices at concentrations found in patient brain tissue, well before frank neuronal loss — a leading hypothesis for why memory symptoms precede detectable cell death, and a major rationale for therapeutic strategies aimed at restoring synaptic function rather than only clearing plaques. In pharmacology and anesthesiology, understanding NMDA receptor pharmacology born from LTP research directly explains the mechanism of ketamine (an NMDA antagonist used as anesthetic and, at sub-anesthetic doses, as a rapid-acting antidepressant) and informs the cognitive side-effect profile of drugs that modulate glutamatergic signaling. In rehabilitation and stroke neuroscience, principles of activity-dependent synaptic strengthening underlie the rationale for intensive, repetitive task-practice therapies intended to drive plasticity in surviving cortical circuits, and inform brain-stimulation protocols (e.g., theta-burst transcranial magnetic stimulation, whose very naming and pulse pattern derive directly from LTP-inducing stimulation protocols in animal slices) used experimentally to enhance motor recovery or treat depression. Finally, in basic and computational neuroscience, LTP and its counterpart LTD supply the empirically-grounded learning rules used in artificial neural network models of hippocampal function, and continue to anchor ongoing work in Molecular Mechanisms of Memory Consolidation and the broader study of the Hippocampus and Spatial Memory system.

Ask the tutor: How does the direction of postsynaptic calcium signaling determine whether a synapse undergoes LTP or LTD?