Receptive field formation by interacting excitatory and inhibitory synaptic plasticity

Claudia Clopath, Tim P. Vogels, Robert C. Froemke, Henning SprekelerbioRxiv 2016 · DOI

TL;DR

This paper proposes a computational model where the interplay between excitatory and inhibitory synaptic plasticity can explain how neurons in the brain develop specialized receptive fields, crucial for processing sensory information like sound. The model highlights that the tuning of inhibitory inputs plays a key role in shaping these receptive fields, leading to the observed co-tuning of excitation and inhibition in sensory cortices.

What Problem Does This Paper Address?

Our brains are incredibly adept at processing sensory information. For instance, neurons in the auditory cortex become tuned to specific sound frequencies, developing "receptive fields" that define what they respond to. Interestingly, this tuning isn't just for excitatory inputs (which excite the neuron) but also for inhibitory inputs (which suppress it), and these two tend to be "co-tuned." While we know that synaptic plasticity (the ability of connections between neurons to change strength) is the underlying mechanism for learning and development, the precise rules governing how this co-tuning emerges and is maintained, especially the interaction between excitatory and inhibitory plasticity, remain poorly understood. This paper aims to bridge that gap by proposing and analyzing a computational model that can reproduce these observed phenomena.

Methods

The authors employ a computational modeling approach using a single, rate-based neuron. The core of their model lies in simulating two types of synaptic plasticity:

  • Excitatory Synaptic Plasticity: This follows a Hebbian learning rule, meaning synapses strengthen when both the pre- and post-synaptic neuron are active. To prevent instability, this rule is combined with a weight-limiting mechanism, either subtractive or multiplicative normalization.
  • Inhibitory Synaptic Plasticity: This uses a "balancing" learning rule that adjusts inhibitory weights to maintain the neuron's firing rate close to a target value. This rule is crucial for achieving E/I balance.

They explore different scenarios by varying the "stimulus specificity" (how tuned the inputs are to particular stimuli) and "learning rates" of both excitatory and inhibitory plasticity. They also analyze the model's dynamics mathematically, deriving effective learning rules and examining the stability of different network states.

Datasets

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

Key Insights and Ideas

  • Inhibition as a "Sliding Threshold": Unspecific (broadly tuned) inhibitory inputs, combined with inhibitory plasticity, can act like a dynamic "sliding threshold" for neuronal activity. This mechanism, similar to BCM rules, allows for the development of stimulus selectivity in excitatory inputs.
  • The Importance of Inhibitory Tuning Width: The degree to which inhibitory inputs are tuned to specific stimuli is critical. Broadly tuned inhibition supports receptive field formation, while narrowly tuned inhibition leads to a detailed balance but hinders selectivity.
  • Competition and Asymmetry: Synaptic competition, amplified by normalization mechanisms, can also drive stimulus selectivity, especially when there are small asymmetries in the input signals.
  • Co-tuning is a Natural Outcome: The model demonstrates that the interaction between excitatory and inhibitory plasticity naturally leads to the co-tuning of their receptive fields, mirroring experimental observations.
  • Learning Rate Dynamics: The relative speeds of excitatory and inhibitory plasticity significantly influence the emergent receptive field properties, with imbalances leading to oscillations or instability.

Key Results

  • The model successfully reproduces the emergence of stimulus-selective receptive fields in the presence of both excitatory and inhibitory plasticity.
  • It shows that unspecific inhibition can lead to BCM-like receptive field formation.
  • It demonstrates that stimulus-specific inhibition, while promoting detailed E/I balance, can prevent receptive field emergence unless input asymmetries are amplified.
  • The study highlights that broader inhibitory tuning supports receptive field formation and co-tuning, aligning with experimental findings.
  • The model can reproduce complex phenomena like the plasticity of receptive fields observed in auditory cortex, including shifts and stabilization under different stimulation conditions.

Limitations and Future Work

The authors acknowledge several limitations:

  • Simplified Neuron Model: They use a relatively simple, linear, rate-based neuron model. They expect their findings to generalize to more complex, nonlinear neuron models but acknowledge this is an assumption.
  • Feedforward Network: The analysis is primarily focused on feedforward networks. While they suggest their findings might extend to recurrent networks, this is not explicitly modeled.
  • Ignoring Temporal Dynamics: Temporal aspects of neuronal and synaptic integration, as well as spike timing-dependent plasticity, were largely ignored to simplify the analysis. They note that these factors could influence the precision of E/I balance and introduce selectivity to temporal input features.
  • Experimental Characterization of Inhibitory Plasticity: The authors point out that the experimental understanding of inhibitory synaptic plasticity is less advanced than that of excitatory plasticity, making direct validation challenging.

Future work could explore the impact of recurrent connections, more complex neuron models, and the role of spike timing-dependent plasticity. Investigating the diversity of inhibitory neuron types and their specific roles in plasticity would also be valuable.

Summary updated

Abstract

Abstract The stimulus selectivity of synaptic currents in cortical neurons often shows a co-tuning of excitation and inhibition, but the mechanisms that underlie the emergence and plasticity of this co-tuning are not fully understood. Using a computational model, we show that an interaction of excitatory and inhibitory synaptic plasticity reproduces both the developmental and – when combined with a disinhibitory gate – the adult plasticity of excitatory and inhibitory receptive fields in auditory cortex. The co-tuning arises from inhibitory plasticity that balances excitation and inhibition, while excitatory stimulus selectivity can result from two different mechanisms. Inhibitory inputs with a broad stimulus tuning introduce a sliding threshold as in Bienenstock-Cooper-Munro rules, introducing an excitatory stimulus selectivity at the cost of a broader inhibitory receptive field. Alternatively, input asymmetries can be amplified by synaptic competition. The latter leaves any receptive field plasticity transient, a prediction we verify in recordings in auditory cortex.
BibTeX
@misc{clopath2016receptive,
  author = {Claudia Clopath and Tim P. Vogels and Robert C. Froemke and Henning Sprekeler},
  title = {{Receptive field formation by interacting excitatory and inhibitory synaptic plasticity}},
  year = {2016},
  doi = {10.1101/066589},
  howpublished = {bioRxiv},
}
Figure