Less is more: Importance of long-range exceptions in brain architecture

1826 

Abstract Submission 

Jakub Vohryzek¹, Morten Kringelbach², Gustavo Deco³

¹UNIVERSITAT POMPEU FABRA, Barcelona, Barcelona, ²University of Oxford, Oxford, Oxfordshire, ³Pompeu Fabra University, Barcelona, Catalonia

Jakub Vohryzek  
UNIVERSITAT POMPEU FABRA
Barcelona, Barcelona

Morten Kringelbach  
University of Oxford
Oxford, Oxfordshire

Gustavo Deco  
Pompeu Fabra University
Barcelona, Catalonia

How brain architecture shapes function is a deep question which has occupied many researchers, from the perspective of network neuroscience (Bullmore and Sporns 2009), brain modelling (Breakspear 2017) and spectral graph theory (Atasoy, Donnelly, and Pearson 2016). Some have even suggested that geometry plays a particularly relevant role in shaping functional activity (Pang et al. 2023a), although see this ongoing discussion (Faskowitz et al. 2023; Pang et al. 2023b). Here we focus on probing the importance of the rare long-range exceptions to the exponential distance rule of brain wiring (Markov et al. 2013). New evidence using turbulence has demonstrated the fundamental role of long-range connectivity in shaping optimal brain information processing (Deco et al. 2021). Here we used Laplacian decomposition of four different graph representations of the underlying anatomy to derive anatomical brain modes: exponential-distance rule (EDR) (Ercsey-Ravasz et al. 2013) and long-range exceptions (EDR+LR), geometry-based modes (geometry) and EDR modes (EDR binary and EDR continuous) (Figure 1 A). Our results show that EDR+LR achieves significantly better reconstruction of long-range functional connectivity (FC) compared to the other mode representations. Furthermore, pertinent to time-critical information processing, we show that a small subset of modes achieves a disproportionately high reconstruction of task MRI activity. When this subset of modes is considered, EDR+LR achieves better reconstruction for the 47 HCP tasks compared to the other mode representations, suggesting that less is more for information processing in the brain.

We used publicly available data by the Human Connectome Project (HCP) of resting-state and task fMRI of 255 participants. The various modes are derived from applying the laplace decomposition on the graph representation (Figure 1 B, i). These modes are used to reconstruct the fMRI activity by a linear combination of their contributions (Figure 1 B, ii). This serves to reconstruct the resting-state fMRI activity and in particular the long-range connectivity exceptions derived as high-correlation values (<0.5 correlation) and over a long euclidean distance (<40mm) as well as the task fMRI activation maps (Figure 1 B, iii). Four different graph representations were constructed and decomposed into their associated modes (Figure 1 C). Namely, the fitted weighted and binary Euclidean Distance Rule with lambda of 0.18, EDR with long-range connectivity and the geometric modes (Pang et al. 2023a).

An important feature of cortex dynamics are long-range functional connections, defined by high correlation values (<0.5 correlation) and euclidean distance (<40mm) (Figure 2 A). We reconstructed the FC long-range connections for an increasing number of modes (1-200) showing the EDR+LR modes to have the highest reconstruction correlation (Figure 2 B). EDR+LR is statistically higher compared to the other modes (paired t-test, pval<0.05) (Figure 2 C) . For the reconstruction of the 7 activation task fMRI maps lower frequency modes contribute disproportionately more toward the reconstruction error as seen by the elbow around 20 modes (Figure 2 D). We therefore reconstructed the error for the 47 HCP tasks benchmarked against the geometrical modes for the first 20 modes (blue - better reconstruction correlation of the chosen modes, red better for the geometric modes). On average EDR+LR showed the most accurate reconstruction across tasks and number of reconstructed modes 1-20 (Figure 2 E).

Here we show the importance of long-range connectivity as a key feature of shaping brain functional activity both for the spontaneous and task-based fMRI. Moreover, functional brain activity can be shown to be on a lower-dimensional manifold span by a subset of these fundamental modes with the most appropriate representation from the EDR+LR graph, suggesting that less is more for efficient information processing in the brain.

Connectivity (eg. functional, effective, structural) ²

fMRI Connectivity and Network Modeling ¹

Modeling