Abstract
Feature-based attention is known to enhance the responses of cortical neurons tuned to the attended feature, but its control mechanisms remain poorly understood relative to those of spatial attention, which involve a network of cortical and subcortical structures. Subcortical structures are thought to serve as important control points in the flow of information, but while feature-based attention has been observed in the cortex, it is not known whether its effects also can be observed in the subcortex. We therefore functionally imaged the human subcortical visual nuclei while subjects detected changes in separate fields of moving or colored dots. We found that when the fields were disjoint, spatially attending to one field enhanced hemodynamic responses in the superior colliculus (SC), lateral geniculate nucleus (LGN) and two retinotopic pulvinar nuclei. When the two dot fields were spatially coincident, feature-based attention to the moving versus colored dots enhanced the responses of the pulvinar and voxels located along the ventromedial surface of the LGN, corresponding to the location of the magnocellular layers, while voxels along the dorsolateral surface of the LGN, corresponding to the location of the parvocellular layers, showed the opposite effect; the SC was inconsistently modulated among subjects. These feature-based attentional modulations could not be explained by differential allocations of spatial attention. All of the subcortical nuclei demonstrated enhancement in hemodynamic activity preceding the attentional switches between the features; however, suppression was observed in voxels along the lateral edge of the LGN, perhaps corresponding to the thalamic reticular nucleus, and voxels in what appeared to be the superficial layers of the SC. We conclude that feature-based attention operates throughout the visual system via modulation of activity in neurons that encode the attended feature.