Reproducible patterns of neural activity without attractors in cortical networks

Cortical activity often consists of recurring population events of correlated neuronal firing and highly reproducible firing patterns. Because of their resemblance with attractor dynamics, the attractor model prevails today, although it has not been firmly demonstrated. Here, we used a unique dataset, with co-registered two-photon calcium imaging and electron microscopy of the same cortical tissue, to test the central assumption of attractor networks: recurrently active "core" neurons should be strongly interconnected. We report that, contrary to the attractor paradigm, core neurons have fewer weaker connections compared to other neurons. Instead, core neurons funnel the flow of multiple connectivity pathways. Computational models give a mechanistic account of these features showing that distance-dependent connectivity forms converging-diverging motifs and, at their funneling centers, core neurons are found. Thus, reproducible cortical activity and connectivity can be explained without postulating underlying attractor networks but rather by the existence of overlapping information flows.