Temporal compression mediated by short-term synaptic plasticity

Christian Leibold, Anja Gundlfinger, Robert Schmidt, Kay Thurley, Dietmar Schmitz, Richard KempterProceedings of the National Academy of Sciences 2008 · DOI

TL;DR

This paper proposes a novel mechanism for how the brain bridges the vast temporal gap between slow behavioral timescales (seconds) and fast neuronal processing (milliseconds). They demonstrate that short-term synaptic plasticity, combined with subthreshold neuronal oscillations, can "compress" slow temporal patterns into rapid sequences of neural spikes, enabling learning and memory formation.

What Problem Does This Paper Address?

The brain needs to represent and process information across vastly different timescales. While neuronal firing and synaptic events happen on the millisecond scale, behaviors like navigation or memory recall unfold over seconds. How can these slow, behaviorally relevant events be translated into the fast, precise timing of neuronal activity that drives learning and memory? This disconnect poses a significant challenge in understanding neural computation. The authors hypothesize that short-term synaptic plasticity, which has memory timescales in the hundreds of milliseconds to seconds, can act as a crucial intermediary.

Methods

The researchers employed a multi-pronged approach:

  • In Vitro Electrophysiology: They recorded from hippocampal CA3 pyramidal cells in brain slices. They simulated synaptic inputs (EPSCs) with varying amplitudes and timing, combined with subthreshold membrane potential oscillations (mimicking theta rhythms). This allowed them to measure how EPSC amplitude influenced the timing of action potentials (APs).
  • Computational Modeling: They developed a conductance-based two-compartment model of a CA3 pyramidal cell. This model was used to reproduce the experimental findings and to perform more detailed simulations, allowing for finer control over parameters and the calculation of information transmission.
  • Information Theory: They used mutual information to quantify how much information about the EPSC amplitude was encoded in the timing of the output spikes.
  • Learning Simulations: They trained a "tempotron" neuron, a model capable of learning from temporal spike patterns, to discriminate between compressed temporal sequences generated by their model. This demonstrated the functional relevance of their proposed mechanism for downstream processing.

Datasets

Not discussed in the paper. The study primarily relies on simulated data generated by their computational model and experimental data from ex vivo recordings.

Key Insights and Ideas

  • Synaptic Plasticity as a Temporal Buffer: Short-term synaptic plasticity, particularly facilitation with its long decay time constants (seconds), acts as a memory buffer for slow temporal patterns.
  • Temporal Compression via Amplitude-to-Timing Conversion: The amplitude of a synaptic input, modulated by short-term plasticity, is translated into the precise timing of a neuron's output spike. This conversion is significantly enhanced by subthreshold oscillations.
  • Subthreshold Oscillations as a "Clock": Theta-frequency oscillations provide a rhythmic framework, allowing the neuron to convert a continuous range of synaptic amplitudes into discrete spike timings within a compressed temporal window (milliseconds).
  • Input Phase Matters: The relative timing (phase) of the synaptic input with respect to the subthreshold oscillation is critical for efficient information transfer and successful learning.
  • Downstream Neuron Properties are Key for Learning: While information transmission is important, the ability of a downstream neuron to learn from these compressed temporal patterns is highly dependent on its own properties, such as the decay time constant of its postsynaptic potentials.

Key Results

  • Enhanced Temporal Coding Range: Subthreshold oscillations significantly increased the range of delays between synaptic input and action potential firing, effectively compressing temporal information. This range was approximately 51 ± 2 ms, roughly half a theta period.
  • Information Transmission: The researchers quantified the mutual information between EPSC amplitude and spike timing. They found that optimal information transmission occurred for specific input phases (around 180°-270°) and was significantly boosted by subthreshold oscillations, achieving up to 4.2 bits/spike.
  • Successful Temporal Pattern Discrimination: A tempotron neuron successfully learned to discriminate between temporally compressed spike patterns generated by their model. The success rate was highly dependent on the input phase and the downstream neuron's EPSP decay time constant, highlighting the interplay between encoding and decoding mechanisms.
  • Reverse Replay Mechanism: The model naturally produces a "reverse replay" of the original slow temporal pattern, a phenomenon observed in hippocampal place cell sequences.

Limitations and Future Work

  • Model Simplifications: While the two-compartment model captures key dynamics, it is a simplification of a real neuron. The authors acknowledge that the optimal input phases for learning are specific to the chosen learning paradigm, oscillation properties, and downstream neuron kinetics.
  • In Vitro vs. In Vivo: The experiments were conducted in brain slices, and while efforts were made to mimic in vivo conditions, further validation in freely behaving animals would be beneficial.
  • Generalizability: The study focused on hippocampal CA3 neurons and theta oscillations. Future work could explore whether this mechanism applies to other brain regions and different types of oscillations and synaptic plasticity.
  • Role of Inhibition: The paper primarily focuses on excitatory synaptic plasticity. Investigating the role of short-term depression in inhibitory synapses could provide a more complete picture.
  • Learning Rule Specificity: The tempotron learning rule was used as a proof of concept. Exploring how other learning rules interact with this temporal compression mechanism would be valuable.

Summary updated

Abstract

Time scales of cortical neuronal dynamics range from few milliseconds to hundreds of milliseconds. In contrast, behavior occurs on the time scale of seconds or longer. How can behavioral time then be neuronally represented in cortical networks? Here, using electrophysiology and modeling, we offer a hypothesis on how to bridge the gap between behavioral and cellular time scales. The core idea is to use a long time constant of decay of synaptic facilitation to translate slow behaviorally induced temporal correlations into a distribution of synaptic response amplitudes. These amplitudes can then be transferred to a sequence of action potentials in a population of neurons. These sequences provide temporal correlations on a millisecond time scale that are able to induce persistent synaptic changes. As a proof of concept, we provide simulations of a neuron that learns to discriminate temporal patterns on a time scale of seconds by synaptic learning rules with a millisecond memory buffer. We find that the conversion from synaptic amplitudes to millisecond correlations can be strongly facilitated by subthreshold oscillations both in terms of information transmission and success of learning.
BibTeX
@inproceedings{leibold2008temporal,
  author = {Christian Leibold and Anja Gundlfinger and Robert Schmidt and Kay Thurley and Dietmar Schmitz and Richard Kempter},
  title = {{Temporal compression mediated by short-term synaptic plasticity}},
  booktitle = {Proceedings of the National Academy of Sciences},
  year = {2008},
  doi = {10.1073/pnas.0708711105},
}
Figure