Information-Limiting Correlations in Large Neural Populations

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Understanding the neural code necessitates examining how neuronal populations encode information.
• Theoretical models suggest correlated noise in large neural populations may limit information processing.
• Previous analyses of small neuronal ensembles (<100 neurons) have not detected information-limiting correlations.

Purpose of the Study:

• To investigate the existence of information-limiting noise correlations in large neural populations.
• To determine if small correlations impact information coding in the prefrontal cortex (PFC).

Main Methods:

• Simultaneous recording of over 500 neurons from the PFC (area 46) of two macaque monkeys using eight Utah arrays.
• Estimation of information related to saccades as a function of neuronal ensemble size.
• Comparison of information coding with and without noise correlations (by shuffling trials).

Main Results:

• Noise correlations were found to be small (approximately 10^-3) on average.
• Information coding scaled sublinearly with increasing ensemble size, indicating saturation.
• Shuffling trials to eliminate noise correlations resulted in a linear scaling of information with ensemble size.

Conclusions:

• The study provides strong evidence for the existence of information-limiting noise correlations in large PFC neuronal populations.
• Despite small pairwise correlations, they significantly affect information coding in populations of hundreds of neurons.
• This finding supports theoretical predictions that even subtle correlations can limit information transmission in complex neural systems.