Decomposition of task-specific responses in the multiple demand network

Neuroimaging evidence suggests that a distributed network of brain areas, known as the multiple demand (MD) network, is consistently recruited across a wide range of cognitive tasks. The MD network includes specific areas in prefrontal and parietal cortices. However, the fine-grained organization of this network is poorly understood. Here we aim to comprehensively characterize the functional subdivisions within the MD network using a naturalistic fMRI paradigm. 20 subjects were instructed to perform 14 different tasks on a set of movie stimuli. These tasks were designed to target a wide range of cognitive domains including visual, spatial, categorical, emotional, auditory, linguistic, social, and semantic processes. fMRI data were also collected while subjects passively watched the movies. The MD network was first delineated by localizing cortical areas that were significantly more active in all 14 tasks compared to the passive-viewing condition. The principal component analysis was then applied on task-specific responses of cortical points within the MD network. The first component was correlated with activities for all tasks, and its spatial map revealed the core, highly multimodal subregions of the MD network. The other components showed preferences for a subset of tasks. In particular, the second component revealed a sharp distinction between MD regions that were preferentially active in visual/spatial versus linguistic/semantic tasks. A graph analysis on the entire cortex also showed a large-scale distinction between visual/spatial and linguistic/semantic areas, with MD regions linking the two communities of cortical areas. Overall, our results provide new insights into how the MD network and its fine-grained architecture contribute to the human intelligent behavior.