When do microcircuits produce beyond-pairwise correlations?

Area of Science:

• Computational Neuroscience
• Systems Neuroscience
• Retinal Circuitry

Background:

• Describing collective neural population activity is challenging.
• Retinal ganglion cell (RGC) activity patterns can often be explained by pairwise interactions.
• Higher-order statistics are sometimes necessary, influenced by input and circuit mechanisms.

Purpose of the Study:

• To investigate the emergence of higher-order interactions in a model of RGC circuit activity.
• To quantify the impact of higher-order interactions by comparing mechanistic models with pairwise maximum entropy (PME) models.
• To explain the empirical success of pairwise models in describing RGC population activity.

Main Methods:

• Developed a model of the RGC circuit with correlations generated by common input.
• Quantified higher-order interactions by comparing model responses to PME models.
• Analyzed an analytically tractable simplification of the RGC model to understand deviations from pairwise predictions.

Main Results:

• RGC output spiking patterns were surprisingly well captured by pairwise models across a broad range of stimuli.
• Bimodal input signals led to larger deviations from pairwise predictions than unimodal inputs in a simplified model.
• Upstream RGC circuitry's light filtering properties suppress input signal bimodality, reducing higher-order interactions.

Conclusions:

• Pairwise interactions are sufficient to describe RGC population activity under many conditions.
• The suppression of input bimodality by retinal circuitry is a key factor explaining the prevalence of pairwise correlations.
• This provides a novel mechanistic explanation for the widespread empirical success of pairwise models in the retina.