Neural populations revealed

308 How are actions reDresented in the brain? Information processing-ePresentation, as long as only one eye novement is being represented at any iven time. Thus, far from being a 'sloppy' Jay to organize information, coarse pop-ilation coding is efficient, accurate and elatively insensitive to noise in the res-lonses of single neurons. Will this lesson learned in the oculo-iotor system generalize to other action ystems, such as the control of the limbs, Microstimulation of a discrete point in the deeper layers of the monkey superior col-liculus produces a very fast saccade of the eyes with a particular direction and amplitude that depends only on the location of the microelectrode'.2. The latency is short (20-30 ms) and the response is machine-like. This is reminiscent of an invertebrate 'command' neuron whose firing can cause a complex coordinated behaviour to occur (see page 278 of ref. 3). The pattern of activity in the superior colliculus preceding a saccade is, however, quite wide-spread4 and it has been generally assumed that the summation of the responses in the active population determines the eye movement. This hypothesis predicts that the neurons with peak activity in the motor map code the location to which the eyes move. Lee, Rohrer and Sparks, in their elegant experiment published on page 357 of this issue', provide conclusive proof to the contrary, that the information coding for eye movement in the superior colliculus is represented by the distributed firing pattern of all the neurons that are activated. Such distributed representations of sensory information and motor commands may be quite common in the brain. A region in the deeper layers of the superior colliculus about 2 mm in diameter containing about 100,000 neurons is activated just before a typical saccade. Imagine that each of these neurons has attached to it an arrow representing the direction and magnitude of the eye movement that occurs when the stimulating electrode is centred on that neuron. Imagine further that the 100,000 vectors attached to these neurons are added vectorially, each arrow weighted by the activity level of its neuron. The resultant vector accurately predicts how the eyes will move. Lee et al. show' that when a subset of these neurons is deactivated by lidocaine, and can no longer contribute to the vector average, the eye movement will miss the target precisely as predicted by the resultant vector. This suggests a very large population of neurons contributes to …