Neural network architecture of a mammalian brain

Abstract

Connectomics research is making rapid advances, although models revealing general principles of connectional architecture are far from complete. Our analysis of 10 6 published connection reports indicates that the adult rat brain interregional connectome has about 76,940 of a possible 623,310 axonal connections between its 790 gray matter regions mapped in a reference atlas, equating to a network density of 12.3%. We examined the sexually dimorphic network using multiresolution consensus clustering that generated a nested hierarchy of interconnected modules/subsystems with three first-order modules and 157 terminal modules in females. Top–down hierarchy analysis suggests a mirror-image primary module pair in the central nervous system’s rostral sector (forebrain–midbrain) associated with behavior control, and a single primary module in the intermediate sector (rhombicbrain) associated with behavior execution; the implications of these results are considered in relation to brain development and evolution. Bottom–up hierarchy analysis reveals known and unfamiliar modules suggesting strong experimentally testable hypotheses. Global network analyses indicate that all hubs are in the rostral module pair, a rich club extends through all three primary modules, and the network exhibits small-world attributes. Simulated lesions of all regions individually enabled ranking their impact on global network organization, and the visual path from the retina was used as a specific example, including the effects of cyclic connection weight changes from the endogenous circadian rhythm generator, suprachiasmatic nucleus. This study elucidates principles of interregional neuronal network architecture for a mammalian brain and suggests a strategy for modeling dynamic structural connectivity.