A model for oscillatory gating of information flow between neural circuits as a function of local recurrence

Physical connections between neurons are hardwired on the timescale of seconds, but infor-
mation and signals must be routed flexibly for computation on the same timescales. Many species exhibit circuit-level oscillations with systematic phase relationships between spikes and the regional oscillation. We present a mechanistic neural model of communication, regulated by oscillations. If one or more networks (the “transmitters”) drives a second (the “receiver”) and both network states are periodically modulated, what relationships in their oscillations will allow the transmitter to optimally drive the receiver? As the receiver network moved parametrically from a ‘sensory’ regime to a ‘memory’ regime, we show that the optimal phase offset depends on its operating mode. A sensory network responds optimally when cells are maximally depolarized, or when the transmitter-receiver pair have zero phase offset, while a network with strong recurrent weights, external inputs are drowned out by local contributions when cells are maximally depolarized, and a phase offset of 90 degrees is optimal. Our results generalize to networks with global inhibition, asynchronous inputs and to networks with discrete fixed points in the strong coupling regime.