Population Rate Dynamics and Multineuron Firing Patterns in Sensory Cortex

Michael S. Okun, Pierre Yger, Stephan Lawrence Marguet, Florian Gérard-Mercier, Andrea Benucci, Steffen Katzner, Laura Busse, Matteo Carandini, Kenneth D. HarrisJournal of Neuroscience 2012 · DOI

Abstract

Cortical circuits encode sensory stimuli through the firing of neuronal ensembles, and also produce spontaneous population patterns in the absence of sensory drive. This population activity is often characterized experimentally by the distribution of multineuron "words" (binary firing vectors), and a match between spontaneous and evoked word distributions has been suggested to reflect learning of a probabilistic model of the sensory world. We analyzed multineuron word distributions in sensory cortex of anesthetized rats and cats, and found that they are dominated by fluctuations in population firing rate rather than precise interactions between individual units. Furthermore, cortical word distributions change when brain state shifts, and similar behavior is seen in simulated networks with fixed, random connectivity. Our results suggest that similarity or dissimilarity in multineuron word distributions could primarily reflect similarity or dissimilarity in population firing rate dynamics, and not necessarily the precise interactions between neurons that would indicate learning of sensory features.

Introduction

Neuroscientists have long observed that the brain produces "spontaneous" activity even in the absence of external stimuli. In adult animals, this spontaneous activity often looks remarkably similar to the activity evoked by natural stimuli. The sampling hypothesis suggests this occurs because the brain has learned a probabilistic model of the world; spontaneous activity represents the brain "sampling" from its internal library of possible sensory experiences. This paper, published in The Journal of Neuroscience, investigates whether this similarity is a signature of high-level learning or a simpler consequence of how neural populations fluctuate together.

Methods and Models

The researchers recorded multineuron activity from the primary auditory cortex (A1) of rats and the primary visual cortex (V1) of cats. Activity was represented as binary "words" (vectors of 0s and 1s) in 2ms time bins. To quantify the dissimilarity between word distributions, they used symmetrized Kullback-Leibler divergence (KLdiv).

The core of the study is the Raster Marginals Model, a null model algorithm that generates synthetic spike trains. This model preserves the Mean Firing Rate (MFR) of each individual neuron and the Population Rate Distribution (PRd)—the frequency with which a specific number of neurons fire simultaneously. Crucially, it randomizes which specific neurons fire together, thereby destroying any precise, learned interactions.

Overview of multineuron word analysis, brain states, and population rate dynamics.

Key Insights

State Over Stimulus

A major finding was that the internal brain state (synchronized vs. desynchronized) has a much larger impact on firing patterns than the actual sensory stimulus. Word distributions from two different periods of silence in the same brain state were more similar than spontaneous vs. evoked activity across different states. For example, the KLdiv between spontaneous and evoked activity was measured at ~7 bits/s, while the KLdiv between synchronized and desynchronized spontaneous states was ~98 bits/s.

Comparison of evoked and spontaneous word frequencies showing high similarity (KLdiv = 7 bits/s).

Dominance of Population Rate Fluctuations

The Raster Marginals Model—which ignores specific neuronal partnerships—accurately predicted experimental word distributions. In Rat A1, the model predicted word distributions with a KLdiv of ~16 bits/s, significantly outperforming a model that only considered individual firing rates (~115 bits/s). This implies that "who" is firing together matters much less than "how many" are firing together. The authors proved mathematically that the KLdiv between population rate distributions acts as a strict lower bound for the KLdiv of the full multineuron word distributions.

Mimicking Learning in Random Networks

By adjusting global parameters like spike-frequency adaptation or background drive in a randomly connected simulation of 10,000 neurons, the authors replicated the developmental shift where spontaneous activity begins to match evoked activity. This suggests that what appears to be the learning of sensory features may actually be a change in global network excitability or "tone."

Discussion and Limitations

The results suggest that similarity in multineuron word distributions primarily reflects similarity in population firing rate dynamics. However, the study was conducted on anesthetized animals, and the authors acknowledge that awake, behaving animals might exhibit more precise, non-random interactions. Furthermore, the Raster Marginals model did not provide a perfect fit, suggesting that some residual pairwise correlations could still be relevant for neural coding. Future work is needed to determine if these global fluctuations carry functional information or are simply a constraint of cortical architecture.

Summary updated

Abstract

Cortical circuits encode sensory stimuli through the firing of neuronal ensembles, and also produce spontaneous population patterns in the absence of sensory drive. This population activity is often characterized experimentally by the distribution of multineuron "words" (binary firing vectors), and a match between spontaneous and evoked word distributions has been suggested to reflect learning of a probabilistic model of the sensory world. We analyzed multineuron word distributions in sensory cortex of anesthetized rats and cats, and found that they are dominated by fluctuations in population firing rate rather than precise interactions between individual units. Furthermore, cortical word distributions change when brain state shifts, and similar behavior is seen in simulated networks with fixed, random connectivity. Our results suggest that similarity or dissimilarity in multineuron word distributions could primarily reflect similarity or dissimilarity in population firing rate dynamics, and not necessarily the precise interactions between neurons that would indicate learning of sensory features.
BibTeX
@article{okun2012population,
  author = {Michael S. Okun and Pierre Yger and Stephan Lawrence Marguet and Florian Gérard-Mercier and Andrea Benucci and Steffen Katzner and Laura Busse and Matteo Carandini and Kenneth D. Harris},
  title = {{Population Rate Dynamics and Multineuron Firing Patterns in Sensory Cortex}},
  journal = {Journal of Neuroscience},
  year = {2012},
  doi = {10.1523/jneurosci.1831-12.2012},
}

Loading related content...

Figure