Rapid cortical dynamics associated with auditory spatial attention gradients

Behavioral and EEG studies suggest spatial attention is allocated as a gradient in which processing benefits decrease away from an attended location. Yet the spatiotemporal dynamics of cortical processes that contribute to attentional gradients are unclear. We measured EEG while participants (n = 35) performed an auditory spatial attention task that required a button press to sounds at one target location on either the left or right. Distractor sounds were randomly presented at four non-target locations evenly spaced up to 180° from the target location. Attentional gradients were quantified by regressing ERP amplitudes elicited by distractors against their spatial location relative to the target. Independent component analysis was applied to each subject's scalp channel data, allowing isolation of distinct cortical sources. Results from scalp ERPs showed a tri-phasic response with gradient slope peaks at ~300 ms (frontal, positive), ~430 ms (posterior, negative), and a plateau starting at ~550 ms (frontal, positive). Corresponding to the first slope peak, a positive gradient was found within a central component when attending to both target locations and for two lateral frontal components when contralateral to the target location. Similarly, a central posterior component had a negative gradient that corresponded to the second slope peak regardless of target location. A right posterior component had both an ipsilateral followed by a contralateral gradient. Lateral posterior clusters also had decreases in α and β oscillatory power with a negative slope and contralateral tuning. Only the left posterior component (120–200 ms) corresponded to absolute sound location. The findings indicate a rapid, temporally-organized sequence of gradients thought to reflect interplay between frontal and parietal regions. We conclude these gradients support a target-based saliency map exhibiting aspects of both right-hemisphere dominance and opponent process models.