Integrative Role of Attention Networks in Multi-Frequency Modular Organization of Human Brain

1570 

Abstract Submission 

Hüden Neşe¹, Ahmet Ademoğlu¹, Tamer Demiralp²

¹Boğaziçi University, Istanbul, Uskudar, ²Istanbul University, Istanbul, Fatih

Hüden Neşe  
Boğaziçi University
Istanbul, Uskudar

Ahmet Ademoğlu  
Boğaziçi University
Istanbul, Uskudar

Tamer Demiralp  
Istanbul University
Istanbul, Fatih

Even though there is converging evidence of hierarchical organization in the cerebral cortex, with sensory-motor and association regions at opposite ends, the mechanism of such hierarchical interactions remains elusive (Mesulam, 1998). This organization was primarily investigated in terms of the spatiotemporal dynamics of intrinsic connectivity networks (ICNs); however, more effort is needed to investigate network dynamics in the frequency domain. Network science has a powerful tool, multilayer modularity analysis, to examine frequency-dependent changing patterns in brain connectivity (Betzel & Basset, 2017). It also gives us the opportunity to determine the flexibility and the topological roles (hub/connector) of individual brain regions. In this study, we used the help of network theory to investigate the integrative role of brain regions.

We used resting-state fMRI data (96 participants) from the HCP dataset with 400-parcel 7-network parcellation (Van Essen et al., 2013; Schaefer et al., 2018). We calculated connectivity matrices in three frequency bands (0.011 - 0.038Hz, 0.043 - 0.071Hz, and 0.076 - 0.103Hz) via a phase-based connectivity estimation approach. We performed multilayer modularity analysis, considering the connectivity matrices at each frequency band as a layer (Mucha et al., 2010). At the subject level, we calculated the flexibility of each brain region. Then, we checked the significance of estimated flexibility values with a Wilcoxon rank sum test by comparing them to the ones obtained from 100 randomized networks. Moreover, two graph metrics, Normalized Participation Coefficient (PCnorm) and Within Module Degree z-score (WMz), were calculated for each brain region to investigate parcels' node roles in the network (Guimera & Amaral, 2005). We identified the hubs and connectors with high WMz and PCnorm, respectively. At the group level, we applied consensus clustering algorithm to find group-level modular organization robust to intersubject variability (Lancichinetti & Fortunato, 2012). We identified main modules and regions with frequency-domain flexibility.

At the group level, three large modules were consistent over frequency bands; one dominated by VN, one dominated by SMN, and one cognitive module covering LN, CN, and DMN. Attention networks were distributed to these three main modules. Our results indicated that most of the group-level flexible parcels, which we called integrative regions, belonged to attention networks, especially the salience ventral attention network. At subject level, a Wilcoxon rank sum test indicated that most of the flexibility values of real connectivity networks, except some parcels belonging to limbic, were significantly different from the randomized networks (p-fdr<0.05). When we compared ICNs in terms of their flexibility, we observed that ICNs were clustered into two groups: somatosensory networks with lower flexibility and attention and cognitive networks with higher flexibility. When we investigated connectors that were common in three frequency bands, 16 out of 26 belonged to SVAN and 7 out of 26 belonged to DAN. Therefore, we concluded that attention networks play an important role in integration not only by simultaneously participating in different modules via different frequency bands but also by connecting various modules at each frequency band.

We proposed that the decomposition of the conventional BOLD frequency band into sub-bands may allow for the observation of simultaneous, parallel processes, which can provide a mechanistic perspective on the integration among modular structures. Given the proposed hierarchical organization among ICNs, it was consistent with our expectations to find those integrative regions predominantly in attention networks that connect the two ends of the hierarchy. Our multi-frequency modularity analysis results emphasize the integrative role of attention networks and the importance of multiband frequency analysis of brain networks.

Connectivity (eg. functional, effective, structural) ¹

fMRI Connectivity and Network Modeling ²

Other - Resting-state; fMRI; Multilayer Network Analysis; Flexibility; Phase-based Connectivity Estimation; Integration; Salience Ventral Attention Network