Inferring neural population codes for Drosophila acoustic communication

Social communication between animals is often mediated by sequences of acoustic signals, sometimes spanning long timescales. How auditory neural circuits respond to extended input sequences to guide behavior is not understood. We address this problem using Drosophila acoustic communication, a behavior involving the male’s production of and female’s response to long, highly variable courtship songs. Here we ask whether female neural and behavioral responses to song are better described by a linear– nonlinear feature detection model vs. a nonlinear accumulation model. Comparing both models against head-fixed neural recordings and pure-behavioral recordings of unrestrained courtship, we found that while both models could explain the neural data, the accumulation model better predicted female locomotion during courtship, outperforming several alternative predictors. To understand how the accumulation model encoded song to predict locomotion, we analyzed the relationship between neural activity simulated by the model and female locomotion during courtship— this revealed the model’s reliance on heterogeneous nonlinear adaptation and slow integration. Finally, we asked how adaptation and integration processes could cooperate across the model neural population to encode temporal patterns in song. Simulations revealed how adaptation can transform song inputs prior to integration, allowing finescale song information to be retained in the population code for long periods. Thus, modeling fly auditory responses as a nonlinearly adaptive, accumulating population code accounts for female locomotor responses to song during courtship and suggests a biologically plausible mechanism for the online encoding of extended communication sequences.