Convergent information flows explain recurring firing patterns in cerebral cortex

Abstract

Cortical population events, short-lived patterns of neuronal activity that recur with some consistency, are central to sensorimotor coordination. These reproducible firing patterns are often attributed to attractor dynamics, supported by strong mutual connectivity. However, using multi-modal datasets — including 2-photon imaging, electrophysiology, and electron microscopy — we show that these reproducible patterns do not involve strongly interconnected neurons. Instead, we show that cortical networks exhibit hierarchical modularity, with core neurons acting as high-information-flow nodes positioned at module interfaces. These cores funnel activity but lack structural signatures of pattern completion units expected in an attractor network. Using computational models, we find that distance-dependent connectivity is necessary and sufficient to generate the modularity and transient reproducible events observed in cortex. Our findings suggest that cortical networks are instead pre-configured to support sensorimotor coordination. This work redefines the structural and dynamical basis of cortical activity, highlighting the link between modular structure and function.