Long-Term Depression: How Weak Signals Prune Synapses

What LTD Is and Where It Happens

Long-term depression is a long-lasting, use-dependent reduction in the efficacy of synaptic transmission. Where LTP makes a synapse louder, LTD makes it quieter — and crucially, the change outlasts the stimulus that caused it, persisting for hours or days rather than seconds.

LTD is not a single phenomenon but a family of related processes found across the brain:

• Hippocampal CA1 and neocortex — an NMDA-receptor-dependent form studied since Dudek and Bear's landmark 1992 report; a template for how experience prunes cortical circuits.
• Cerebellum — LTD at parallel-fiber-to-Purkinje-cell synapses, discovered by Masao Ito around 1982, and long proposed as the cellular substrate of motor learning and the vestibulo-ocular reflex.
• Striatum — an endocannabinoid- and mGluR-dependent form important for habit formation.

All share one endpoint: fewer functional AMPA-type glutamate receptors at the postsynaptic density, so each glutamate release produces a smaller excitatory postsynaptic current (EPSC).

The Mechanism, Step by Step

Consider the canonical hippocampal case. LTD hinges on the amount and duration of postsynaptic calcium — the same second messenger that drives LTP, but at a different level.

• 1. Weak, prolonged input. Low-frequency stimulation (~1 Hz) causes modest glutamate release and only partial relief of the Mg²⁺ block on NMDA receptors.
• 2. A small, sustained Ca²⁺ rise. Ca²⁺ trickles in through NMDA receptors to a modest ~0.2–0.5 µM — high enough to activate phosphatases, but too low to fully engage the kinases that drive LTP.
• 3. Phosphatase cascade. Ca²⁺/calmodulin activates calcineurin (PP2B), which dephosphorylates inhibitor-1, unleashing protein phosphatase 1 (PP1).
• 4. AMPAR dephosphorylation. PP1 removes the phosphate from GluA1 Serine-845, destabilizing the receptor at the synapse.
• 5. Endocytosis. Dephosphorylated AMPARs are internalized via clathrin/dynamin-dependent endocytosis, shrinking the postsynaptic response.

The Bienenstock–Cooper–Munro (BCM) rule captures this quantitatively: below a sliding modification threshold θ_m, activity produces depression; above it, potentiation.

Key Molecules and Characteristic Numbers

The molecular cast differs by brain region, and the contrast is instructive:

• Hippocampal LTD (phosphatase-driven): calcineurin (PP2B) and PP1 dephosphorylate GluA1 at Ser845 (the PKA site). A modest Ca²⁺ signal (~0.2–0.5 µM sustained over minutes) tips the balance toward removal.
• Cerebellar LTD (kinase-driven): here the logic inverts. A large Ca²⁺ transient — from voltage-gated Ca²⁺ channels plus mGluR1-triggered IP₃-mediated release — activates PKCα, which phosphorylates GluA2 at Ser880. That phosphorylation ejects GluA2 from its anchor GRIP and hands it to PICK1, driving endocytosis. Nitric oxide and the MAPK–PKC positive-feedback loop make the switch nearly all-or-none.

Concrete example: the standard hippocampal LTD protocol is 900 pulses at 1 Hz (15 minutes), reliably depressing the field EPSP by roughly 20–40% in young rodent slices. Cerebellar LTD is classically induced by pairing parallel-fiber and climbing-fiber input at ~1 Hz, and its magnitude is read out as a lasting drop in Purkinje-cell EPSC amplitude.

How LTD Is Studied and Regulated

LTD is measured electrophysiologically. In acute brain slices, researchers stimulate an afferent pathway and record the field EPSP slope or whole-cell EPSC amplitude before and after a conditioning protocol; a persistent drop that lasts beyond ~30–60 minutes qualifies as LTD.

Key experimental levers include:

• Pharmacology: the NMDAR blocker AP5 (D-APV) abolishes hippocampal LTD; phosphatase inhibitors (okadaic acid, FK506/cyclosporin A against calcineurin) block it too, cementing the phosphatase requirement.
• Genetics: knock-in mice with mutated GluA2 C-terminal phosphorylation sites, or PICK1 knockouts, blunt cerebellar LTD.
• Optical readouts: two-photon imaging of dendritic spines shows LTD accompanied by spine shrinkage or elimination, physically pruning the connection.

LTD is metaplastically regulated: prior activity shifts the modification threshold θ_m, so a synapse's recent history determines whether the same stimulus depresses or potentiates it. This is the sliding threshold at the heart of the BCM theory.

How LTD Compares to Its Cousins

LTD is best understood by contrast with related plasticity phenomena:

• vs. LTP: same synapse, same Ca²⁺ trigger, opposite direction. LTP follows a large, brief Ca²⁺ rise that favors kinases (CaMKII) and AMPAR insertion; LTD follows a modest, prolonged rise that favors phosphatases and AMPAR removal. This Ca²⁺-amplitude dependence is the classic explanation for their bidirectionality.
• vs. depotentiation: depotentiation reverses a previously established LTP back toward baseline, whereas LTD drives a naive synapse below its starting strength. They can share molecular machinery but are distinct phenomena.
• vs. synaptic scaling: scaling is a homeostatic, cell-wide, multiplicative adjustment of all a neuron's synapses to keep firing in range; LTD is input-specific (Hebbian), altering only the active synapses.
• vs. short-term depression: a presynaptic, vesicle-depletion effect lasting milliseconds to seconds — not the persistent, postsynaptic AMPAR-trafficking change of LTD.

Significance, Disease Relevance, and Open Questions

LTD matters because a memory system that could only strengthen would quickly saturate. LTD provides the erasing, sharpening, and pruning that keeps circuits from clogging — enabling reversal learning, forgetting of the irrelevant, and the fine-tuning of motor programs. Cerebellar LTD is a leading candidate mechanism for motor learning and adaptation of the vestibulo-ocular reflex, though its exact necessity remains debated.

Disease connections are significant:

• Fragile X syndrome: loss of FMRP exaggerates mGluR-dependent LTD, a core feature of the mGluR theory of the disorder and a drug target.
• Alzheimer's disease: soluble amyloid-β oligomers facilitate LTD and impair LTP, biasing synapses toward weakening and loss.
• Addiction and depression: aberrant LTD in striatal and cortical circuits is implicated in maladaptive plasticity.

Open questions: How precisely does the same Ca²⁺ ion decode into opposite outcomes? Is cerebellar LTD strictly required for motor learning, or is it redundant with other sites? And can LTD-targeting drugs correct plasticity imbalances in Fragile X or Alzheimer's without disrupting normal memory?

Hippocampal (NMDAR-dependent) LTD vs. cerebellar (PKC-dependent) LTD vs. LTP

Feature	Hippocampal LTD	Cerebellar LTD	LTP (for contrast)
Typical induction	Low-frequency stimulation, ~1 Hz × 900 pulses	Paired parallel-fiber + climbing-fiber activity	High-frequency tetanus, ~100 Hz
Ca²⁺ signal	Modest, prolonged rise (~0.2–0.5 µM)	Large Ca²⁺ transient via VGCC + IP₃ release	Large, brief rise (>1 µM)
Dominant enzyme	Phosphatases: calcineurin (PP2B) → PP1	Kinase: PKCα	Kinase: CaMKII
AMPAR outcome	GluA1 Ser845 dephosphorylated → endocytosis	GluA2 Ser880 phosphorylated → PICK1 swap, endocytosis	GluA1 insertion into membrane
Net synaptic effect	Fewer surface AMPARs → weaker EPSC	Fewer surface AMPARs → weaker EPSC	More surface AMPARs → stronger EPSC