Does stimulus perception depend on large populations of neurons, or on the activity of single neurons? Neither, according to a new study — in some situations, perceptual decisions are based on population activity in the most useful subset of neurons.

The middle temporal (MT) area of the primate brain is important for the perception of moving visual stimuli, and contains many neurons that respond to objects that move in a preferred direction. When monkeys are asked to discriminate between two widely separated directions of motion, it is thought that information from many MT neurons is pooled. However, it now seems that finer discriminations rely on more selective information pooling.

Purushothaman and Bradley trained monkeys to compare the directions of motion of two moving stimuli. The differences in direction were small — less than 3° — and the authors recorded the activity of neurons in area MT while the monkeys did the task. They then analysed the responses of the neurons to the different stimulus directions, and compared the neuronal activity with the psychophysical performance of the monkeys.

The 'choice probablility' of a neuron is a measure of how well its firing rate correlates with perception — for example, if a neuron usually shows higher activity when the monkey perceives the second stimulus as moving anticlockwise relative to the first, regardless of the actual direction of motion, it has a high choice probability and is likely to be involved in perception. In this study, neurons whose preferred directions of motion were about 70° away from that of the reference stimulus showed the highest choice probabilities. As MT neurons are broadly tuned, the tuning curves of these neurons are probably steepest around the directions of the test stimuli, which would allow them to show large differences in activity for relatively small changes in direction.

Do monkeys selectively use information from these neurons to make perceptual decisions, or do they pool neuronal activity indiscriminately? On the fine discrimination task used in this study, indiscriminate pooling of neuronal activity was much less accurate than the monkeys' performance, indicating that some other strategy is used. When the authors analysed the covariance of neuronal activity and perceptual performance, they found that perceptual decisions correlated more closely with the activity of 'high-precision' MT neurons than with the activity of low-precision neurons. According to the authors, to achieve the observed performance levels, the monkeys must use selective pooling of activity from those neurons whose responses are most informative for the task.

In an accompanying News and Views article, Naler and DeAngelis discuss the implication that areas such as MT can use different coding strategies for different tasks, such as broad versus fine discrimination. As they conclude, studies that use multielectrode arrays to record simultaneously from many neurons will provide greater insight into population coding.