Data-driven filtration and segmentation of mesoscale neural dynamics
The neocortex contains a constellation of sensory-motor regions whose functional interactions provide the basis of cognition and behavior. Simultaneously recording neuronal group activity across the cortical hemispheres is essential for understanding the nature of information flow across cerebral networks. Moreover, unbiased and robust methods for measuring functional interactions across the cortex within individual living subjects is critical for substantive tests of genetic and environmental factors that influence brain development and function. To this end, we image pan-neuronally expressed genetically encoded calcium indicators transcranially across the neocortex of unanesthetized, behaving mice throughout development. Recording from behaving mice produces a unique set of challenges, including optical and blood artifacts associated with movement. In addition, areal patterning of the cortex can vary among individuals, ages, and genotypes thus an unbiased, flexible workflow for video acquisition and analysis is necessary for producing high quality segmentations of functional structure across neocortex. To address these challenges, we have developed an eigendecomposition-based workflow that isolates hemodynamic and optical artifacts to recover underlying calcium activity patterns, and segments independent regions of the brain to create maps of functional units across the developing cortex. These unique, data-driven maps provide a reference for understanding developmental, genetic, as well as individual variation between functional units of the brain, and provide a method for extracting optimized time courses from the cortical surface without the need for stimulation-based mapping or anatomical post-processing and alignment. In addition, we quantify the quality of separation of independent sources and use the resulting metrics as feedback to optimize our video acquisition parameters. The open source methods developed here are flexible enough to be utilized with various subject ages and genotypes, as well as numerous experimental configurations and computer architectures. Here we additionally present applications of these methods to anesthesia and development datasets. Overcoming these challenges opens the possibility of using these techniques to help address a number of key objectives in neuroscience, including the quantification of robust inter-areal dynamics between functional motifs across cerebral networks.