The brain can anticipate stimuli that are behaviorally relevant, and attention can be deployed in advance of the stimuli’s appearance. The prevailing view is that attention works by acting as a sort of gain on neural activity leading to an amplification of a signal. This gain influence is thought to also apply to “spontaneous” activity before a stimulus appears in the neuron’s receptive field, putting the neuron into a more excited state. The problem, however, is that attention increasing spontaneous activity of a visual neuron might be interpreted by postsynaptic populations as the appearance of a weak visual stimulus when no stimulus is present. 

“Our alternative idea is that for every neuron that increases it’s firing rate with attention in anticipation of an imminent stimulus, there is another neuron in the population for which the firing rate actually decreases with anticipatory attention,” notes Adam Snyder (Rochester, Neuroscience), a former post-doc with Matt Smith (Ophthalmology and CNBC, Pitt) and Byron Yu (Biomedical Engineering, ECE and CNBC, CMU). “In this way, the total amount of activity across the population is unchanged, reducing the chance of sending the wrong message to downstream brain areas.”

In a recent paper published in Nature Communications, “Distinct population codes for attention in the absence and presence of visual stimulation”, the researchers’ experimental data provide support for this latter view, raising questions on the standard idea that attention is simply a gain on the neural population activity that is independent of the sensory context. An alternative is that attentional modulations are context dependent, differing between unstimulated and stimulated contexts.

Nine sample neurons that show suppression during the prestimulus interval (the red curve is below the blue curve) but then gain modulation once the stimulus is presented (the red curve is above the blue curve, stimulus presentation at time=0). The standard gain model predicts that the red curve is always above the blue curve.