Characterizing the Sensorimotor Transformations underlying Fly Courtship through Computational Ethology

Abstract

How organisms integrate information across multiple sensory modalities to influence their behavior is a pressing question for systems neuroscience. In naturalistic experimental settings, the diversity of salient stimuli and high-dimensionality of behavior provide challenges for conventional methods of behavioral quantification. Here, we propose several extensions of recently developed methods in computational ethology to study the behavioral relevance of acoustic signaling in fly courtship. First, we propose a novel method for fast pose estimation in unmarked, freely behaving animals that facilitates unsupervised behavioral analysis. Second, we present a program for the automatic segmentation of Drosophila song that generalizes to multiple recording conditions and outperforms previous work. Using these methods in conjunction with a newly designed behavioral chamber featuring sensitive microphones and high-resolution imaging, we uncover differences in behavior surrounding the diversity of song modes. We also study the role of auditory lateralization in courtship. Our results reveal a characteristic turning behavior in response to lateralized song that is modulated by mechanical disruption of the auditory organs of female flies, suggesting bilateral comparison in song processing.