Abstract
In the sensory systems, most neurons’ firing rates are tuned to at least one aspect of the stimulus. Other neurons are appear to be untuned, meaning that their firing rates do not depend on the stimulus. Previous work on information coding in neural populations has ignored untuned neurons, based on the tacit assumption that they are unimportant. Recent experimental work has questioned this assumption, showing that in some circumstances, neurons with no apparent stimulus tuning can contribute to sensory information coding. These findings are intriguing, because they suggest that – by virtue of our ignoring putatively untuned neurons – our understanding of neural population coding might be incomplete. At the same time, several key questions remain unanswered: Are the impacts of putatively untuned neurons on population coding due to weak tuning that is nevertheless below the threshold the experimenters set for calling neurons tuned (vs untuned)? And why do there appear to be untuned neurons in the brain? Do mixed populations of tuned and untuned neurons have a functional advantage over populations containing only tuned neurons? Using theoretical calculations and analyses of in vivo neural data, I answer those questions by: a) showing how untuned neurons can enhance sensory information coding; b) demonstrating that this effect does not rely on weak tuning; and c) identifying conditions under which the neural code can be made more informative by replacing some of the tuned neurons with untuned ones. These conditions specify when there is a functional benefit to having untuned neurons in a circuit, and thus suggest a reason why the brain might contain untuned neurons. Overall, this work shows that, even in the extreme case, where some neurons have no tuning, those neurons can still contribute to sensory information coding, and thus should not be ignored.