Structural and molecular mechanisms of the spacing effect

Jump to navigation Jump to search

This article by Dr Piotr Wozniak is part of SuperMemo Guru series on memory, learning, creativity, and problem solving.


New memories form as new synaptic connections. When memories are in use, they may undergo stabilization by adding AMPA receptors. When memories are not in use, new dendritic filopodia may scout their surroundings for new synaptic targets. Frequent use of memory relies on fast AMPA transmission. Infrequent use may allow of filopodial growth, and allow of the activation of NMDA receptors that would (1) stabilize the synapse (adding new AMPA receptors), and (2) stabilize the dendritic arbor (incl. retraction of filopodia). Memory disuse may result in a failure of postsynaptic activation and forgetting by interference due to a takeover of the activation target by newly stabilized dendritic spines, or the reuse of NMDA-based silent synapses.

Concept network

  • the brain is a concept network
  • spacing effect is necessary for the efficient optimization of memory that underlies intelligence
  • a concept neuron corresponds with a single meaningful idea
  • a neural link between two concepts represents a single memory
  • a concept neuron is activated by a set of dendritic inputs forming the activation pattern
  • in terms of a spaced repetition learning procedure, a combination of inputs corresponds with the decoded question/stimulus, while the activation corresponds with the answer/response
  • the neuronal activity is suppressed by inhibitory input that serves activation selectivity
  • a depolarization signal sent out via the axon corresponds with the activation of a concept in memory

Dendritic filopodia

New memories

Fast transmission

  • filopodial growth would be kept in check by fast AMPA transmission (and halted or reversed by nearby NMDA currents in stabilization)
  • fast transmission keeps a dendritic arbor relatively inert on the assumption that good use needs little change

Post-synaptic inactivity

  • filopodial growth should be accelerated by low postsynaptic activity (see: Dendritic arbors undergo branching followed by stabilization)
  • long periods of postsynaptic inactivity result in the growth of new dendritic filopodia in the vicinity of an active axon
  • glutamate release from active axons stimulates filapodial growth via mGluR receptor
  • filopodia scout the neural environment in search of attractive axonal targets. Conceptually this means that concepts seek relevant activations by other concepts
  • rich filopodial growth may result in a drop in memory retrievability
  • new filopodia may contribute to forgetting via interference by establishing new synaptic connections



New memories

  • newly formed synapses and silent synapses are primarily populated by NMDA receptors
  • synapse stabilization adds AMPA receptors to the PSD

Fast transmission

  • fast AMPA transmission leads to hyperpolarization that prevents the activation of NMDA receptors
  • without NMDA calcium currents, stabilization is negligible, impact on inactive filopodia is less pronounced

Post-synaptic inactivity

  • sprouting of nearby filopodia builds up a reservoir of AMPA receptors that can be remobilized for the purpose of stabilization
  • phosphorylation of AMPA receptor may contribute to memory retrievability, fast transmission, and potential activation of NMDA receptors
  • NMDA-induced currents (in a stabilized synapse), in conjunction with the activation of mGluR receptor would result in retracting unused filopodia
  • retraction of filopodia could allow of a transfer mobilized AMPARs to the stabilized synapse, which would enhance the spacing effect


  • co-activation of the axon and the postsynaptic neuron may result in NMDA calcium currents that stabilize the synapse
  • activation of the NMDA receptors would be more likely in case of interference from proximal filopodia
  • NMDA-induced currents would favor incorporation of new AMPAR units (adding to molecular stabilization)
  • increase in stability means facilitated AMPA transmission and slower growth of filopodia (see: AMPA receptors stabilize the dendritic branch)


  • forgetting would occur primarily via interference
  • post-synaptic inactivity may result in a takeover of a concept neuron by new patterns based on the maturation of new dendritic spines
  • prolonged disuse of the synapse can also lead to memory decay by molecular and cellular changes (e.g. phosphorylation status, subunit structure, AMPA endocytosis, cytoskeletal stability, etc.)
  • prolonged disuse may result in the ultimate loss of the dendritic spine

For more see: Mechanism of forgetting


The above reasoning is derived from the statistical properties of memory as described in the Neurostatistical Model of Memory. The data has largely been obtained by employing a spaced repetition algorithm. Statistical properties of memory make it possible to hypothesize about the interaction of the individual molecular, cellular and neural phenomena that occur while wiring the concept network of the brain in the course of development and learning. The molecular and structural interpretation is largely mine, and is based on well-documented facts of neuroscience. Some of the hypothetical assumptions are pretty bold. My boldness is justified by the idea that any falsifiable model is better than an absence of models.

  • I proposed the two component model of long-term memory in 1988 (for details see: Two components of memory)
  • in Optimization of learning in 1990, I hypothesized for the first time about molecular correlates of the memory model. At that time, structural changes seemed unlikely as I was not aware of the speed of the growth of filopodia. However, glutamate, increase in the number of receptors and the calcium signal were already included in the hypothetical model
  • in Economics of learning in 1995, in co-operation with Dr Edward Gorzelanczyk, we refined the model by hypothesizing about the role of NMDA receptors, mGluR, protein phosphorylation, and the possibility of exposing new receptors on the post-synaptic membrane
  • in the new millennium the evidence for dendritic dynamics kept mounting and it became obvious that structural changes are necessary when forming new memories. In retrospect, the whole idea of a dynamic concept network seems obvious and necessary. The reliance on solely molecular changes might be energetically prudent, but computationally infeasible
  • in 2017, mounting structural evidence resulted in my proposition of the two-component model of memory stability
  • in 2019, I concluded that the sprouting of new filopodia might be the main cause of forgetting via interference. This would make them a good candidate for memory retrievability. All we need is a good link between the activation of NMDA receptors and the retraction of nearby filopodia. That mechanism would be a solid candidate for the neural explanation of the spacing effect.
  • the presented model was compiled in March 2020, and should be seen as the best fit between biology and statistics as seen via the inspirational lens of spaced repetition.


Caution! The presented model of the spacing effect is hypothetical (there are many alternatives). It can still include errors, or need a major revision. However, its evolution over the last 30 years seems to indicate a good convergence with data. If you are aware of biological data that contradicts the model or is hard to explain in its light, please let me know.

For more texts on memory, learning, sleep, creativity, and problem solving, see Super Memory Guru