A behaviorally-relevant multiple-demand operating system underlying diverse cognitive tasks

929 

Abstract Submission 

Weidong Cai¹, Jalil Taghia², Vinod Menon¹

¹Stanford University, Palo Alto, CA, ²Uppsala University, Uppsala, Sweden

Weidong Cai  
Stanford University
Palo Alto, CA

Jalil Taghia  
Uppsala University
Uppsala, Sweden

Vinod Menon  
Stanford University
Palo Alto, CA

The human brain is a flexible, yet stable, system that allows rapid and adaptive allocation of cognitive resources to meet moment-by-moment changes in task demands (Braun et al., 2015; Shine et al., 2016; Taghia, et al., 2018). A converging body of evidence now points to a core set of distributed brain areas that are consistently engaged during diverse cognitive tasks (Cai et al., 2019; Dosenbach et al., 2006; Duncan, 2010). This commonality naturally raises the critical and challenging question of how the same brain areas might underlie cognition across multiple task domains. Addressing this question has the potential to uncover mechanisms underlying a multiple-demand, domain-general, functional system underlying cognition and identify transdiagnostic features of cognitive dysfunction in psychiatric and neurological disorders. Here we use a state space hidden Markov model and novel computational analyses to address this challenge. We identify common brain states that are dynamically engaged across seven different cognitive paradigms, across multiple participant cohorts, and demonstrate their behavioral relevance.

We leveraged a total of seven different fMRI experiments across a wide range of cognitive domains, including n-back working memory, continuous performance, cued task switching, Sternberg working memory, Stroop, and Stop-signal tasks, and relational processing tasks (Braver et al.,, 2021; Van Essen et al., 2012). The Bayesian switching dynamical systems state space (BSDS) algorithm was used to identify brain states in each task and examine their correspondence with brain states in a canonical n-back working memory reference task (Taghia, et al., 2018). The latent brain state was determined by unique patterns of activity and functional connectivity between key nodes of the salience, central executive, and default mode networks. Using the n-back working memory as a reference task, we asked whether task-optimal latent brain states that occur during the high cognitive load condition in the n-back task are also engaged during each of the other seven cognitive tasks. Our choice of the working memory task was motivated both by the fact that it is widely used to probe cognitive function and dysfunction, and by our identification of optimal and non-optimal brain states associated with cognitive performance and decision-making (Taghia et al., 2018). State temporal closeness, which measures the similarity of two latent brain states' temporal profiles, and state space closeness which measures the similarity of two latent brain states' space feature profiles, were developed to match latent brain states between different studies. Whether a latent brain state in an independent task matches an optimal working memory task brain state was determined by how close they were in their space and temporal parameters.

BSDS uncovered a number of latent brain states in each study. State temporal closeness and state space closeness measures consistently identified a shared dynamic latent brain state engaged across diverse experiments and four data cohorts (Figure 1). Importantly, despite significant differences in experimental paradigms, data acquisition protocols, and participant cohorts, the temporal properties of brain states predicted cognitive task performance in each of the tasks (ps<0.05, Figure 2). Moreover, the occurrence rates of the shared latent state also predicted behavioral performance (ps<0.05). Furthermore, weak engagement of the shared brain state was related to inattention symptoms (r=0.38, p=0.01), suggesting that our generative model is also relevant for investigations of psychopathology.

Our findings uncover a general dynamic brain state that is preferentially engaged during cognition, and demonstrates that functional circuits associated with the multiple-demand system can adaptively contribute to a wide range of cognitive functions.

Executive Function, Cognitive Control and Decision Making ¹

Higher Cognitive Functions Other

fMRI Connectivity and Network Modeling ²

FUNCTIONAL MRI

Other - working memory, cognitive control, latent brain state, human cognition, salience network, frontoparietal network, default mode network, brain behavior association, inattention, ADHD