Generative network models identify biological mechanisms of altered structural brain connectivity in schizophrenia

Abstract

Short title: Generative models of brain networks in schizophrenia Word Count (Abstract): 220/250 Word Count (Article Body): 3999/4000 Number of Figures: 3 Number of Tables: 1 Number of References: 62 This paper contains Supplementary Materials. AbstractBackground: Alterations in the structural connectome of schizophrenia patients have been widely characterized, but the mechanisms leading to those alterations remain largely unknown.Generative network models have recently been introduced as a tool to test the biological underpinnings of the formation of altered structural brain networks. Methods:We evaluated different generative network models to investigate the formation of structural brain networks in healthy controls (n=152), schizophrenia patients (n=66) and their unaffected first-degree relatives (n=32), and we identified spatial and topological factors contributing to network formation. We further investigated the association of these factors to cognition and to polygenic risk for schizophrenia.Results: Structural brain networks can be best accounted for by a two-factor model combining spatial constraints and topological neighborhood structure. The same wiring model explained brain network formation for all groups analyzed. However, relatives and schizophrenia patients exhibited significantly lower spatial constraints and lower topological facilitation compared to healthy controls. The model parameter for spatial constraint was correlated with the polygenic risk for schizophrenia and predicted reduced cognitive performance. Conclusions:Our results identify spatial constraints and local topological structure as two interrelated mechanisms contributing to normal brain development as well as altered connectomes in schizophrenia. Spatial constraints were linked to the genetic risk for schizophrenia and general cognitive functioning, thereby providing insights into their biological basis and behavioral relevance.

A generative network model of neurodevelopment

“…2a-2d). Mirroring previous findings in adult samples 24,25,29 , we found that models driven by geometry and topology outperform the pure geometric spatial model and homophily-based models achieve the lowest energy for our pediatric sample ( Fig. 2e).…”

“…The enrichment analysis that accompanied our GNM takes a very different approach. As far as we are aware, this is the first study aiming to bridge models of whole brain organizational emergence and genetics in this way (for work using generative models, see refs 24,25,29,43,44 and work that integrates Allen Brain Atlas gene data with functional and structural brain imaging, see refs [45][46][47][48] ). Nodal costs covaried with genes enriched for highly costly metabolic processes, including catabolic processes, protein transport and cellular components centered around the ribosome and endoplasmic membranes.…”

“…K can be thought of as a topological value (or 'rule') driving the intention for node n i to connect with node n j . K is parameterized by a distinct scalar g. K can take a range of different forms and can, in principle, be selected from any non-geometric growth rule used to model social and economic networks [26][27][28][29] . One simple example is the 'matching' rule 24 : nodes form connections with other nodes on the basis of their normalized overlap in neighborhood -i.e.…”

Modeling spatial, developmental, physiological, and topological constraints on human brain connectivity

“…Recent modeling of human connectome data suggests that the EDR offers an incomplete account of connectome architecture. Specifically, this work indicates that EDR-based models are less accurate in reproducing several topological features of empirical connectomes when compared to models that combine a distance penalty with a preference to form topologically favorable connections, thus more closely capturing the cost-value trade-off implicit in Cajal's laws (5,(28)(29)(30)(31)(32). In particular, these studies suggest that models combining a distance penalty with a homophilic attachment rule, in which connections are more likely to form between nodes that connect to other similar nodes (28)(29)(30)33), offer better accounts of the empirical data.…”

“…Recent evidence indicates that existing-cost-topology models cannot capture the topography of certain properties, such as the network degree sequence and, by extension, location of connectome hubs, even when model parameters are optimized for this objective (5,31,32) (although see also (30)).…”