A small, computationally flexible network produces the phenotypic diversity of song recognition in crickets

Abstract How neural networks evolve to recognize species-specific communication signals is unknown. One hypothesis is that novel recognition phenotypes are produced by parameter variation in a computationally flexible “mother network”. We test this hypothesis in crickets, where males produce and females recognize mating songs with a species-specific pulse pattern. Whether the song recognition network in crickets is computationally flexible to recognize the diversity of pulse patterns and what network properties support and constrain this flexibility is unknown. Using electrophysiological recordings from the cricket Gryllus bimaculatus, we built a model of the song recognition network that reproduces the network dynamics as well as the neuronal and behavioral tuning for that species. An analysis of the model’s parameter space reveals that the network can produce all recognition phenotypes known in crickets and even other insects. Biases in phenotypic diversity produced by the model are consistent with the existing behavioral diversity in crickets, and arise from computations that likely evolved to increase energy efficiency and robustness of song recognition. The model’s parameter to phenotype mapping is degenerate – different network parameters can create similar changes in the phenotype – which is thought to support evolutionary plasticity. Our study suggest that a computationally flexible mother network could underlie the diversity of song recognition phenotypes in crickets and we reveal network properties that constrain and support behavioral diversity.