Abstract
The encoding of sensory information by populations of cortical neurons forms the basis for perception but remains poorly understood. To understand the constraints of cortical population coding we analyzed neural responses to natural sounds recorded in auditory cortex of primates (Macaca mulatta). We estimated stimulus information while varying the composition and size of the considered population. Consistent with previous reports we found that when choosing subpopulations randomly from the recorded ensemble, the average population information increases steadily with population size. This scaling was explained by a model assuming that each neuron carried equal amounts of information, and that any overlap between the information carried by each neuron arises purely from random sampling within the stimulus space. However, when studying subpopulations selected to optimize information for each given population size, the scaling of information was strikingly different: a small fraction of temporally precise cells carried the vast majority of information. This scaling could be explained by an extended model, assuming that the amount of information carried by individual neurons was highly nonuniform, with few neurons carrying large amounts of information. Importantly, these optimal populations can be determined by a single biophysical marker—the neuron's encoding time scale—allowing their detection and readout within biologically realistic circuits. These results show that extrapolations of population information based on random ensembles may overestimate the population size required for stimulus encoding, and that sensory cortical circuits may process information using small but highly informative ensembles.
Introduction
A central debate in neuroscience is how the brain encodes information: is it "distributed" across many neurons, or "sparse," relying on just a few? Previous research often suggested a distributed model because, when researchers randomly sampled groups of neurons, the total information grew steadily with the number of cells. This implied that to recover complex sensory data (like natural sounds), the brain must monitor massive populations.
However, the authors of this paper suspected a "sampling bias" in these earlier conclusions. They hypothesized that by averaging the performance of randomly selected groups, researchers were washing out the contributions of highly specialized neurons. The challenge was to determine if "optimized" subpopulations—the best of the best—could perform the same task with far fewer members, and if so, what biological traits make those neurons special.
Methods
The researchers recorded single-unit activity from the auditory cortex of three Macaque monkeys (Macaca mulatta) while they listened to a 52-second loop of natural sounds (animal calls, speech, and environmental noise). The study utilized a dataset of 49 single units recorded across 8 sessions from the caudal auditory cortex (areas A1, CM, and CL).
To analyze the data, they used information theory to calculate mutual information and employed a diagonal linear classifier to predict sound snippets based on neural spike counts. Crucially, they compared two coding models: a "labelled line" code that preserves unit identity and a "pooled" code that sums activity across the population. They also used a "forward selection" algorithm to build optimized populations by iteratively adding the most informative neurons.
Key Insights
The "Averaging Trap"
The paper demonstrates that random sampling creates a "distorted view" of neural coding. Because most neurons carry little information, random groups look like a homogenous, mediocre mass. This makes it seem like a large population is required to achieve high information levels. However, when the "best" neurons are selected, information saturates incredibly fast. About 30% of the recorded neurons were sufficient to capture 95% of the total information available in the entire ensemble.
Temporal Precision as a Biomarker
The most informative neurons were characterized by high temporal precision (short encoding time scales) rather than high firing rates. These neurons respond to rapid changes in sound with millisecond accuracy. Crucially, these "expert" neurons can be identified by a simple biophysical marker—their autocorrelation function. This suggests that downstream brain areas could potentially "tune in" to these high-precision channels using standard synaptic mechanisms.
Key Results
- Scaling Differences: In random populations, information grew slowly and linearly. In optimized populations, information followed an exponentially saturating curve, reaching near-maximum levels with only ~12–15 neurons.
- Model Fit: A "heterogeneous" model (assuming highly unequal information across neurons) perfectly matched the experimental data ($R^2 > 0.99$), while the standard homogenized model failed to explain the optimized population's performance.
- Robustness: Optimized populations performed consistently better than random groups even when tested on entirely different sets of sound snippets, indicating they are not merely overfit to specific stimuli.
Discussion and Limitations
While this study suggests that sensory information is concentrated in small, temporally precise ensembles, several questions remain. The study used "pseudo-populations" (shuffling trials to remove correlations); while controls suggested noise correlations were small, simultaneous recordings would provide a more definitive picture. Additionally, the role of the "silent majority"—the 70% of less informative neurons—is still unknown. They may provide redundancy or encode features not tested here. Finally, proving that biological circuits actually implement the selection of these high-precision neurons remains a task for future experimental work.
Summary updated