Brain network communication models

© 2020 Caio Pimentel SeguinCommunication between neural elements underpins all aspects of brain functioning. Large-scale neural signalling unfolds atop the human connectome, the complex network that describes how gray matter regions are interconnected by white matter projections. The mechanisms governing the propagation and communication of signals across the connectome remain unknown. The main focus of this thesis is the investigation of network communication models aimed at elucidating how the anatomical substrate of nervous systems facilitates and constrains functional interactions between gray matter regions. To date, the vast majority of network neuroscience studies have assumed neural signalling occurs via topological shortest paths. This is reflected by the widespread use of graph measures such as global efficiency, betweenness centrality and the small world coefficient. In recent years, researchers have begun to question this assumption on the basis that communication via shortest paths is contingent on centralized knowledge of connectome topology, and thus may not be a biologically realistic signalling model. This has led to the exploration of decentralized strategies of network propagation. Most efforts in this direction are focused on diffusive communication, which typically models neural signalling from the perspective of random walk processes. While these approaches do not mandate knowledge assumptions about network organization, they fail to promote efficient and energetically frugal neural information transfer. Therefore, the literature on brain network communication models is currently concentrated on the opposing strategies of shortest path routing and diffusive communication. This thesis aims to reconsider this dichotomous state by investigating alternative brain network communication models. In Chapter 3, we explore the concept of navigation in mammalian connectomes. Navigation is a greedy routing strategy in which information is propagated based on the spatial positioning of brain regions. Using human, macaque and mouse brain networks, we provide evidence that connectome organization is conducive to decentralized efficient communication under navigation. Specifically, the combination of empirical connectome topology and geometry was necessary for successful network navigation, with disruptions to either attribute resulting in marked decreases of navigation efficiency. These findings suggest that brain network architecture may have evolved to facilitate efficient decentralized information transfer, and indicate a three-way relationship between topology, geometry and communication in nervous systems. Decentralized network communication models can be asymmetric. This means that the efficiency of signalling paths may vary depending on the direction of information flow. Importantly, this behaviour occurs even in undirected networks. This unexplored facet of network communication provides the opportunity to study directional patterns of signalling in the human structural connectome, in which all connections are considered bidirectional due to the inability of diffusion imaging to resolve axonal directionality. In Chapter 4, we develop the statistical framework of send-receive asymmetry and demonstrate that it contributes novel insight into large-scale neural signalling directionality. Crucially, this chapter provides cross-modal evidence for the utility of decentralized communication models by demonstrating a statistical association between send-receive asymmetry and the directionality of effective connectivity. Lastly, in Chapter 5 we perform a systematic evaluation of the main neural signalling models proposed in the network neuroscience literature. We evaluate models in terms of their (i) predictive utility of interindividual variation in human behaviour and (ii) structure-function coupling strength. We hypothesize that communication models performing better in these criteria may provide more parsimonious characterizations of information transfer mechanisms in the human brain. Importantly, we benchmark communication models against structural connectivity, and provide evidence that accounting for polysynaptic communication improves the behavioural and functional predictions derived from direct anatomical connections alone. Combining behavioral and functional results into a single ranking of communication models positioned navigation as the top model, suggesting that it may more faithfully recapitulate biological neural signalling patterns. The results in this chapter contribute to elucidating the relationship between human behaviour, functional connectivity and connectome communication. Collectively, the findings reported in this thesis further our knowledge of large-scale neural signalling, promoting a unified understanding of brain structure, function and communication