Coordinated hippocampal-thalamic-cortical communication crucial for engram dynamics underneath systems consolidation

Douglas Feitosa Tomé, Sadra Sadeh, Claudia ClopathNature Communications 2022 · DOI

TL;DR

This paper presents a biologically plausible computational model showing that systems consolidation—the transfer of memories from the hippocampus to the cortex—requires a coordinated three-way dialogue involving the thalamus. By tuning synaptic plasticity timescales across a hippocampus-thalamus-cortex circuit, the researchers successfully replicated the "maturation" of cortical memories and provided a mechanistic explanation for varying patterns of amnesia.

What Problem Does This Paper Address?

Neuroscientists have long known that memories undergo "systems consolidation," a process where they initially depend on the hippocampus (HPC) but eventually become stored in the neocortex (CTX). However, the exact "how" remains a mystery.

Recent experimental breakthroughs have identified "engram cells"—the physical traces of memory. These studies revealed a puzzling dynamic: hippocampal engrams are active early but fade over time ("de-maturation"), while cortical engrams start "silent" (unresponsive to cues) and gradually become functional ("maturation"). Existing computational models often rely on abstract architectures that fail to capture these specific cell-level dynamics or ignore the role of subcortical intermediaries like the thalamus. This paper seeks to bridge that gap with a model that respects both biological circuitry and the diverse timescales of synaptic change.

Methods

The authors developed a multi-region spiking recurrent neural network (RNN) consisting of leaky integrate-and-fire neurons. The core of their approach is the implementation of diverse synaptic plasticity timescales.

  • The Circuit: They compared a standard two-region model (HPC-CTX) against a novel three-region model (HPC-THL-CTX) that includes the mediodorsal thalamus.
  • Plasticity Mechanisms: The model uses a sophisticated cocktail of plasticity rules:
  • Hebbian: Triplet spike-timing-dependent plasticity (STDP).
  • Non-Hebbian: Heterosynaptic plasticity (to prevent runaway activity) and transmitter-induced plasticity (linked to consolidation).
  • Inhibitory Plasticity: To regulate firing rates and maintain stimulus specificity.
  • The Protocol: The simulation follows three phases: Training (stimulus presentation), Consolidation (spontaneous activity/reactivations), and Testing (presentation of partial cues to check recall).

Datasets

As a computational modeling paper, it does not use external "big data" sets like ImageNet. Instead, the authors used:

  • Synthetic Stimuli: Four non-overlapping random patterns of Poisson spikes.
  • Spatial Stimuli: Horizontal bars used to test the three-region model's circular receptive fields.
  • Overlapping Stimuli: Used to evaluate the model's ability to handle task similarity and interference.

Key Insights and Ideas

  • The Thalamus is Essential: The authors found that direct monosynaptic connections from the HPC to the CTX were insufficient to replicate experimental data. The thalamus acts as a crucial relay that couples "reactivations" (spontaneous replays of memory) across the brain.
  • Divergent Plasticity Timescales: The "aha moment" is that consolidation emerges naturally if subcortical regions (HPC, THL) learn faster than the cortex, but the hippocampus has a slower "consolidation" constant, making its memories less stable over long periods.
  • Coupled Reactivations: During simulated "sleep" (the consolidation phase), the regions spontaneously fire together. If you decouple this activity—by blocking the thalamus or hippocampus—the cortical memory never "matures."
  • Inhibitory Engrams: The model predicts that inhibitory neurons also form engrams. In the cortex, these inhibitory engrams become more stimulus-specific over time, helping to "sharpen" the memory trace.
  • Explaining Amnesia: The model provides a unifying theory for why hippocampal damage sometimes causes "temporally graded" amnesia (losing only recent memories) and sometimes "flat" amnesia (losing all memories). It suggests the outcome depends on the strength of the thalamocortical coupling established during encoding.

Key Results

  • Successful Replication: The model successfully showed CTX engrams transitioning from silent to active (maturation) while HPC engrams did the opposite (de-maturation) over a simulated 24-hour period.
  • Recall Accuracy: In the three-region model, cortical recall accuracy rose from ~40% immediately after training to nearly 100% after consolidation.
  • Interdependence: Blocking HPC output during consolidation caused CTX recall accuracy to plummet to near zero, matching biological observations that the cortex needs hippocampal "input" to mature its own memory traces.
  • Thalamic Role: Blocking the thalamus similarly disrupted cortical maturation, confirming its role as a necessary bridge in the circuit.

Limitations and Future Work

  • Timescales: The simulation covers 24 hours, whereas biological systems consolidation can take weeks, months, or even years. The authors suggest their timescales are "subjective" and depend on the learning rates used.
  • LFP Rhythms: While the model captures the coupling of activity, it does not explicitly model the specific oscillatory waveforms (like 200Hz ripples or 12Hz spindles) seen in real brains.
  • Semanticization: The model focuses on the stability of memory but does not address how memories might change from specific episodes (episodic) to general facts (semantic) over time.
  • Future Directions: The authors suggest future work should investigate "catastrophic forgetting" (how new memories affect old ones) and use chemogenetic tools to test their specific predictions about inhibitory engrams in the thalamus.

Summary updated

Abstract

Systems consolidation refers to the time-dependent reorganization of memory representations or engrams across brain regions. Despite recent advancements in unravelling this process, the exact mechanisms behind engram dynamics and the role of associated pathways remain largely unknown. Here we propose a biologically-plausible computational model to address this knowledge gap. By coordinating synaptic plasticity timescales and incorporating a hippocampus-thalamus-cortex circuit, our model is able to couple engram reactivations across these regions and thereby reproduce key dynamics of cortical and hippocampal engram cells along with their interdependencies. Decoupling hippocampal-thalamic-cortical activity disrupts systems consolidation. Critically, our model yields testable predictions regarding hippocampal and thalamic engram cells, inhibitory engrams, thalamic inhibitory input, and the effect of thalamocortical synaptic coupling on retrograde amnesia induced by hippocampal lesions. Overall, our results suggest that systems consolidation emerges from coupled reactivations of engram cells in distributed brain regions enabled by coordinated synaptic plasticity timescales in multisynaptic subcortical-cortical circuits.
BibTeX
@inproceedings{tome2022coordinated,
  author = {Douglas Feitosa Tomé and Sadra Sadeh and Claudia Clopath},
  title = {{Coordinated hippocampal-thalamic-cortical communication crucial for engram dynamics underneath systems consolidation}},
  booktitle = {Nature Communications},
  year = {2022},
  doi = {10.1038/s41467-022-28339-z},
}
Figure