Functional connectivity of a brain-scale neural circuit for motion perception

点击翻译按钮获取中文摘要

Abstract The transformation of visual cues into appropriate behavior requires the collaboration of diverse neurons across distant brain areas. A fundamental gap in our knowledge about these visuomotor transformations is understanding how these neurons are functionally connected, shaping neural response dynamics that give rise to behavioral output. This gap is due to the inaccessibility of mammalian model systems, in which simultaneous in vivo observation and manipulations across the brain is impossible as well as a lack of real-time computational frameworks that can capture these dynamics. Here, we plan to investigate the brain-scale functional connectivity underlying the visually guided optomotor response (OMR) in the genetically and optically accessible larval zebrafish. Our previous computational brain-scale models generate concrete predictions for circuit composition and connectivity strength between functional cell classes and behavior but fail to capture the individual neural dynamics of this system. Therefore, to generate realistic dynamic models and test these predictions, we propose leveraging integrated methods combining streaming data analysis, volumetric two-photon microscopy, holographic optogenetic manipulation, and training of multi-regional recurrent neural networks (RNNs). Using patterned photostimulation of single and groups of functionally and molecularly identified neurons, while simultaneously recording activity from other hypothesized downstream neurons, we will infer excitability, sign, and synaptic strength from the network's response. In Aim 1, we will first define neurons both functionally and by their neurotransmitter type across the brain including the pretectum, a conserved visual processing area. In Aim 2, we will train biologically constrained RNNs to predict functional connectivity between these neurons, which we will iteratively test and validate by photostimulating automatically selected neural targets while recording resulting neural activity across the pretectum, orchestrated by our streaming analysis software (improv). Next, we will use these integrated methods to map and model the functional connectivity of pretectal neurons with specific, identifiable premotor spinal projection neurons hypothesized to orchestrate specific behavioral aspects. In Aim 3, we will develop online, gradient-based RNN training of recorded neurons to permit real-time testing and refinement of the predicted brain-wide connectivity leading to behavior in individual zebrafish. These computationally integrated experiments will generate predictive dynamic models of how signals from each eye are transformed into behavior. Together, this research will apply innovative computational and all-optical technologies to decode the temporal neural dynamics underlying complex sensorimotor processing, promising essential insights for the development of treatment strategies for neuropsychiatric disorders that are manifested in the neural connectivity across multiple brain areas.