Rate, not selectivity, determines neuronal population coding accuracy in auditory cortex

The notion that neurons with higher selectivity carry more information about external sensory inputs is widely accepted in neuroscience. High-selectivity neurons respond to a narrow range of sensory inputs, and thus would be considered highly informative by rejecting a large proportion of possible inputs. In auditory cortex, neuronal responses are less selective immediately after the onset of a sound and then become highly selective in the following sustained response epoch. These 2 temporal response epochs have thus been interpreted to encode first the presence and then the content of a sound input. Contrary to predictions from that prevailing theory, however, we found that the neural population conveys similar information about sound input across the 2 epochs in spite of the neuronal selectivity differences. The amount of information encoded turns out to be almost completely dependent upon the total number of population spikes in the read-out window for this system. Moreover, inhomogeneous Poisson spiking behavior is sufficient to account for this property. These results imply a novel principle of sensory encoding that is potentially shared widely among multiple sensory systems.Author summary: Neurons act together to encode information such as the nature of a sensory stimulus. The number of neurons used for individual stimuli and the nature of the encoding used are not well understood. Higher sensory areas have been observed to respond sparsely to sensory stimuli, meaning that only a relatively small number of neurons fire action potentials (or “spikes”) when any particular stimulus is present. Sparse spiking activity is present in primary auditory cortex but only after a sound has been playing for some period of time. Dense spiking, however, is present at stimulus onset. We found that each action potential in primary auditory cortex contributed approximately the same amount of information about a tone stimulus, which resulted in more accurate encoding from the dense onset spiking. The later sparse spiking retained stimulus information but required a longer time to read it out. This arrangement allows sensory stimuli to be identified rapidly yet represented efficiently for extended periods, while neurons still retain sensitivity to novel stimuli. Dense and sparse coding may therefore work together dynamically in order to represent complex, temporally overlapping sensory content.