Memories in a network with excitatory and inhibitory plasticity are encoded in the spiking irregularity

Júlia V. Gallinaro, Claudia Clopath

TL;DR

This paper proposes that memories in the brain might be encoded not just by the strength of connections between neurons, but also by the irregularity of their firing patterns. They demonstrate that even "silent" memories, where neurons aren't actively firing, leave a detectable trace in their spiking irregularity, which can be read out by other neurons and contributes to memory longevity.

What Problem Does This Paper Address?

The brain stores memories through changes in the connections between neurons, forming "cell assemblies." While we know how these assemblies are formed, how they encode information, especially when dormant or "silent," remains a puzzle. Traditional views focus on synaptic strength, but this paper explores an alternative: could the pattern of neural activity, specifically its irregularity, carry crucial information? This is important because silent memories are thought to be more energy-efficient, and understanding their encoding could reveal how the brain maintains information without constant high activity.

Methods

The researchers used computational models of spiking neurons (specifically, leaky integrate-and-fire neurons) to simulate neural networks. They incorporated two key types of plasticity:

  • Excitatory Plasticity: Modeled using a "triplet-based" spike-timing dependent plasticity (STDP) rule, which strengthens connections when neurons fire in specific temporal patterns.
  • Inhibitory Plasticity: Modeled using an "inhibitory STDP" (iSTDP) rule, which helps balance excitatory and inhibitory inputs to neurons.

They simulated the formation of cell assemblies by stimulating a subset of neurons, leading to changes in synaptic weights. They then analyzed the firing patterns (mean firing rate, coefficient of variation of inter-spike intervals - CV) of neurons within these assemblies and compared them to neurons outside. They also investigated how this irregularity could be read out by other neurons using short-term plasticity (STF) and how it affected the decay of synaptic weights over time.

Datasets

Not discussed in the paper. The study relies on simulated data generated by their computational models.

Key Insights and Ideas

  • Irregularity as a Memory Trace: The most striking finding is that even when cell assemblies are "silent" (meaning their average firing rate doesn't change significantly), their internal firing patterns become more irregular (higher CV). This irregularity acts as a signature of the embedded memory.
  • Readout with Short-Term Plasticity: This irregular firing pattern isn't just a passive byproduct; it can be actively detected and utilized by other neurons. The paper shows that short-term facilitation (STF) in the connections to a readout neuron allows it to distinguish between regular and irregular input, effectively decoding the memory trace.
  • Longevity of Memories: The increased spiking irregularity within assemblies also contributes to the long-term stability of the memory. Stronger assemblies, characterized by higher irregularity and correlation, show slower decay of their synaptic weights, suggesting they are more resistant to forgetting.
  • Silent but Informative: This work challenges the notion that "silent" memories are truly inactive. Instead, they are "activity-silent" in terms of mean rate but "pattern-active" due to their altered spiking statistics.

Key Results

  • Formation of EI assemblies, even without a significant change in mean firing rate, led to a statistically significant increase in the coefficient of variation (CV) of inter-spike intervals for neurons within the assembly.
  • This increased CV could be decoded by a readout neuron employing short-term facilitation (STF), leading to a modulated firing rate of the readout neuron that reflected the input's irregularity.
  • Stronger assemblies (formed with higher excitatory weights) exhibited slower decay of their excitatory synaptic weights, a phenomenon attributed to both increased spiking irregularity (CV) and increased correlation between assembly neurons.

Limitations and Future Work

  • External Stimulation for Assembly Formation: The current model relies on external stimulation to initiate assembly formation. Future work could explore how assemblies might form spontaneously or through other mechanisms without transient external input.
  • Readout Neuron Connections: The paper hardwires the connections from the assembly to the readout neuron. Future research could investigate how these readout connections themselves might be learned or shaped by plasticity.
  • Slow Signal Encoding: The CV-based encoding is described as a "slow signal." The paper acknowledges that faster signals might be needed for certain cognitive functions and suggests exploring how to combine different coding schemes.
  • In Vivo Relevance: While the models are based on known plasticity rules, direct experimental validation of this specific encoding mechanism in biological systems remains an open area.

Summary updated

Abstract

Abstract Cell assemblies are thought to be the substrate of memory in the brain. Theoretical studies have previously shown that assemblies can be formed in networks with multiple types of plasticity. But how exactly they are formed and how they encode information is yet to be fully understood. One possibility is that memories are stored in silent assemblies. Here we used a computational model to study the formation of silent assemblies in a network of spiking neurons with excitatory and inhibitory plasticity. We found that even though the formed assemblies were silent in terms of mean firing rate, they had an increased coefficient of variation of inter-spike intervals. We also found that this spiking irregularity could be readout with support of short-term plasticity, and that it could contribute to the longevity of memories.
BibTeX
@article{gallinaro2021memories,
  author = {Júlia V. Gallinaro and Claudia Clopath},
  title = {{Memories in a network with excitatory and inhibitory plasticity are encoded in the spiking irregularity}},
  journal = {PLOS Computational Biology},
  year = {2021},
  doi = {10.1371/journal.pcbi.1009593},
}
Figure