A shared circuit architecture for multiscale signal recognition in crickets

Abstract

Communication signals often contain behaviorally relevant structure on multiple temporal scales. Short-timescale features are commonly thought to be recognized first and then integrated into longer-timescale representations. Whether a shared circuit can instead recognize multiple timescales remains unclear. Cricket calling songs provide a tractable test because millisecond-scale pulses are grouped into chirps lasting hundreds of milliseconds. We combined female phonotaxis with a mechanistic model of the identified pulse-recognition network to ask whether this circuitry can also generate chirp selectivity. In Gryllus bimaculatus, the model, originally parameterized from pulse-scale electrophysiology and behavior, recovered the main boundaries of chirp preference without fitting to chirp behavior. Coincidence detection generated the minimum chirp-duration requirement, whereas suppressive computations in the networks input neuron shaped the upper boundaries for chirp duty cycle and period. Across 14 species, varying these suppressive computations captured major differences in chirp tuning. These findings reveal a temporally interleaved serial circuit in which long-timescale selectivity begins upstream of completed short-timescale selectivity, offering an alternative to both conventional short-first hierarchies and parallel pathways.