Hebbian plasticity in parallel synaptic pathways: A circuit mechanism for systems memory consolidation

Michiel W. H. Remme, Urs Bergmann, Denis Alevi, Susanne Schreiber, Henning Sprekeler, Richard Kempter

TL;DR

This paper proposes a novel computational model, the Parallel Pathway Theory (PPT), to explain how memories are consolidated and transferred across brain regions. By combining Hebbian plasticity with parallel synaptic pathways, the model demonstrates how memories can be gradually transferred and retained over long periods, offering a mechanistic explanation for phenomena like power-law forgetting.

What Problem Does This Paper Address?

The brain's ability to consolidate memories—transforming transient experiences into stable long-term representations—is fundamental to learning and cognition. Declarative memories, for instance, initially depend on the hippocampus but are eventually stored in the neocortex. While processes like neuronal replay are known to be involved, the precise cellular and network mechanisms underlying this "systems memory consolidation" remain poorly understood. Key questions include how memories are transferred, why their representations might change, and why some memories persist while others fade. This paper aims to provide a mechanistic explanation for these processes, bridging the gap between cellular plasticity and large-scale memory dynamics.

Methods

The authors propose the Parallel Pathway Theory (PPT), which posits that systems memory consolidation arises from the interplay of two ubiquitous neural features: parallel synaptic pathways and Hebbian plasticity.

The core mechanism involves:

  • Parallel Pathways: A memory is initially stored in an "indirect" pathway (e.g., trisynaptic pathway in the hippocampus) which has a longer transmission delay. A "direct" pathway, with a shorter delay, runs in parallel.
  • Hebbian Plasticity (STDP): Spike-Timing-Dependent Plasticity (STDP) is the learning rule. When presynaptic spikes consistently precede postsynaptic spikes with a short delay, the synapse is strengthened.
  • Consolidation Process: The indirect pathway acts as a "teacher" for the direct pathway. During consolidation (e.g., sleep), the delayed signals from the indirect pathway trigger postsynaptic spikes. The direct pathway's inputs, arriving slightly earlier, are then potentiated by STDP. This process effectively "copies" the memory representation from the indirect to the direct pathway.

The authors use:

  • Mathematical Analysis: They develop a theoretical framework using rate-based neuron models to derive the learning dynamics and identify conditions for stable consolidation.
  • Computational Modeling: They simulate this mechanism using integrate-and-fire neurons and multi-compartmental neuron models to test its robustness to neuronal nonlinearities and different representations.
  • Hierarchical Iteration: They extend the model to a hierarchy of pathways to explain consolidation from the hippocampus to the neocortex and power-law forgetting.

Datasets

Not discussed in the paper. The authors use simulated data generated by their models.

Key Insights and Ideas

  • Indirect pathway as a teacher: The crucial insight is that the longer delay in the indirect pathway is not a hindrance but a prerequisite for STDP-driven consolidation into the faster direct pathway.
  • Consolidation as linear regression: Mathematically, the process of consolidation is shown to be equivalent to performing a linear regression, where the direct pathway learns to approximate the input-output mapping of the indirect pathway.
  • Robustness to representation differences: The model demonstrates that consolidation can occur even if the neural representations (e.g., place cells vs. grid cells) in the parallel pathways differ, as long as there's some correlation.
  • Hierarchical consolidation and power-law forgetting: Iterating the PPT across multiple stages explains how memories can be transferred to distant cortical areas and why forgetting often follows a power-law, a phenomenon observed in human psychophysics.
  • Speed-up of recall: As memories consolidate into shorter, more direct pathways, retrieval time decreases, explaining faster responses for consolidated memories.

Key Results

  • Successful memory transfer: Simulations show that synaptic weights in the direct pathway accurately copy the pattern from the indirect pathway, demonstrating successful consolidation.
  • Quantitative agreement with lesion studies: The model reproduces experimental findings from rodent lesion studies on the role of the perforant path in memory consolidation.
  • Power-law forgetting: The hierarchical model successfully predicts power-law forgetting curves, matching human behavioral data.
  • Robustness: The mechanism is shown to be robust to neuronal nonlinearities and differing cue representations in parallel pathways.
  • Extended memory lifetime: The hierarchical structure, with decreasing plasticity at higher levels, allows for memory retention on the order of years.

Limitations and Future Work

  • Focus on feedforward networks: The current model primarily uses feedforward networks and local learning rules. Future work could explore the role of recurrent connections and backpropagating signals for more complex representation learning.
  • Reconsolidation: The model does not currently explain reconsolidation, the process by which consolidated memories can become labile again upon retrieval.
  • Anatomical constraints: The model assumes the existence of appropriate parallel pathways and representations. Future work could investigate cases where such pathways are absent or inadequate, potentially explaining why some memories remain hippocampus-dependent.
  • Semantization: The model provides a basis for understanding how memories might become more semantic and lose episodic detail through consolidation into simpler representations. Further exploration of this transformation is suggested.
  • Non-declarative memories: The authors suggest the PPT could also apply to non-declarative memories (e.g., motor skills, fear memories), which warrants further investigation.

Summary updated

Abstract

Abstract Systems memory consolidation involves the transfer of memories across brain regions and the transformation of memory content. For example, declarative memories that transiently depend on the hippocampal formation are transformed into long-term memory traces in neocortical networks, and procedural memories are transformed within cortico-striatal networks. These consolidation processes are thought to rely on replay and repetition of recently acquired memories, but the cellular and network mechanisms that mediate the changes of memories are poorly understood. Here, we suggest that systems memory consolidation could arise from Hebbian plasticity in networks with parallel synaptic pathways — two ubiquitous features of neural circuits in the brain. We explore this hypothesis in the context of hippocampus-dependent memories. Using computational models and mathematical analyses, we illustrate how memories are transferred across circuits and discuss why their representations could change. The analyses suggest that Hebbian plasticity mediates consolidation by transferring a linear approximation of a previously acquired memory into a parallel pathway. Our modelling results are further in quantitative agreement with lesion studies in rodents. Moreover, a hierarchical iteration of the mechanism yields power-law forgetting — as observed in psychophysical studies in humans. The predicted circuit mechanism thus bridges spatial scales from single cells to cortical areas and time scales from milliseconds to years. Author summary After new memories are acquired, they can be transferred over time into other brain areas — a process called systems memory consolidation. For example, new declarative memories, which refer to the conscious memory of facts and events, depend on the hippocampus. Older declarative memories, however, also rely on neocortical networks. The cellular mechanisms underlying such a transfer are poorly understood. In this work, we show that a simple and in the brain ubiquitous connectivity pattern, combined with a standard learning rule, leads to gradual memory transfer. We illustrate our proposed mechanism in numerical simulations and mathematical analyses. At the neurophysiological level, our theory explains experimental findings on memory storage in the hippocampal formation when specific pathways between neural populations are disrupted. At the psychophysical level, we can account for the power-law forgetting curves typically found in humans. A consequence of the proposed model is that consolidated memories can yield faster responses because they are stored in increasingly shorter synaptic pathways between sensory and motor areas. By giving a mechanistic explanation of the consolidation process, we contribute to the understanding of the transfer of memories and the reorganization of memories over time.
BibTeX
@article{remme2021hebbian,
  author = {Michiel W. H. Remme and Urs Bergmann and Denis Alevi and Susanne Schreiber and Henning Sprekeler and Richard Kempter},
  title = {{Hebbian plasticity in parallel synaptic pathways: A circuit mechanism for systems memory consolidation}},
  journal = {PLOS Computational Biology},
  year = {2021},
  doi = {10.1371/journal.pcbi.1009681},
}
Figure