Using DeepLabCut-Live to probe state dependent neural circuits of behavior with closed-loop optogenetic stimulation

Background Closed-loop behavior paradigms enable us to dissect the state-dependent neural circuits underlying behavior in real-time. However, studying context-dependent locomotor perturbations has been challenging due to limitations in molecular tools and techniques for real-time manipulation of spinal cord circuits. New Method We developed a novel closed-loop optogenetic stimulation paradigm that utilizes DeepLabCut-Live pose estimation to manipulate primary sensory afferent activity at specific phases of the locomotor cycle in mice. A compact DeepLabCut model was trained to track hindlimb kinematics in real-time and integrated into the Bonsai visual programming framework. This allowed an LED to be triggered to photo-stimulate sensory neurons expressing channelrhodopsin at user-defined pose-based criteria, such as during the stance or swing phase. Results Optogenetic activation of nociceptive TRPV1 + sensory neurons during treadmill locomotion reliably evoked paw withdrawal responses. Photoactivation during stance generated a brief withdrawal, while stimulation during swing elicited a prolonged response likely engaging stumbling corrective reflexes. Comparison with Existing Methods: This new method allows for high spatiotemporal precision in manipulating spinal circuits based on the phase of the locomotor cycle. Unlike previous approaches, this closed-loop system can control for the state-dependent nature of sensorimotor responses during locomotion. Conclusions Integrating DeepLabCut-Live with optogenetics provides a powerful new approach to dissect the context-dependent role of sensory feedback and spinal interneurons in modulating locomotion. This technique opens new avenues for uncovering the neural substrates of state-dependent behaviors and has broad applicability for studies of real-time closed-loop manipulation based on pose estimation. Manuscript Highlights Closed-loop system probes state-dependent behaviors at pose-modulated instances Bonsai integrates DeepLabCut models for real-time pose estimation during locomotion Phase-dependent TRPV1 + sensory afferent photostimulation elicits context-specific withdrawal responses