Attention changes connectivity strength in a hierarchical manner across the visual network

Attention improves behavioral performance by selectively enhancing neural responses to attended task-relevant stimuli. Several models have been suggested that explain modulation of neural responses by attention. These models are typically limited to effects of attention on single regions, and do not take into account the hierarchical organization of regions involved in processing. Here, we introduce a model that quantifies the connectivity changes across the visual hierarchy encompassing regions V1, V2, V3, up to hV4 and V5/MT+. We measured fMRI activity in humans performing a demanding visual attention task during ultra-long blocks lasting 2 minutes, alternating with passive viewing blocks involving the same stimuli. This paradigm allowed for high-quality functional connectivity measurements free of confounds related to on- and offset effects of stimulus blocks. Functional connectivity was measured between regions of the dorsal attention network (DAN) and visual regions, as well as between default mode network (DMN) regions and visual regions. We then quantified the slope and baseline of connectivity strength of a given DAN or DMN region with the visual hierarchy, as a function of attention. The results revealed that each of the DAN regions had a gradient in its connectivity strength along the visual processing hierarchy: the DAN regions showed stronger connectivity with high-level areas that decreased towards low-level areas, revealing a descending gradient from V5/MT+ and V4 towards V1. Attention enhanced this baseline connectivity pattern in additive manner for IPS, whereas right FEF additionally increased the slope, thus showing also multiplicative effects along the visual hierarchy. DMN regions had an inversed gradient of connectivity with the visual hierarchy. Attention tended to have multiplicative effects for all regions, but (negative) additive effects only in left lateral parietal cortex. The current study provides a first quantitative model of attention induced changes in connectivity between attention and default mode networks and the hierarchy of early visual areas