High channel count electrophysiology and data processing for freely-moving animals

PROJECT SUMMARY Simultaneous recording and stimulation of larger populations of neurons distributed throughout the brain is needed to rigorously evaluate theories of neural computation at the cellular level in mammals. Previously, we introduced a direct-to-disk data acquisition architecture (Kinney et al., 2015) designed to work with close-packed silicon probes (Scholvin et al., 2016) to enable 1000-channel neural recording in head-fixed animals (Preliminary Data). Through pilot studies we demonstrated the successful recording of terabytes of neural spiking activity (Preliminary Data), but also discovered some shortcomings of the architecture. Two design elements in particular were limiting. First, our headstages were too bulky for freely-moving experiments. Second, our acquisition hardware did not have the ability to quickly analyze all 1000 channels of data. As a result, it took days to weeks to understand the neural activity content of the terabyte-size recordings. For ultra-high-channel count neural recordings to become routine, the acquisition architecture must allow and facilitate rapid online and offline analysis of large amounts of data. A computer architecture with local data storage and analysis is favored, since a 1000-channel recording (e.g. 1000 channels sampled with 16 bits at 30 kHz) generates neural data at a sustained rate that exceed typical (gigabit ethernet) network connection speed to compute clusters or the cloud. Accordingly, we propose a 1024-channel silicon probe for freely-moving electrophysiology experiments in combination with a data acquisition system optimized for easy data analysis. The novel silicon probe will record and stimulate 1024 closed-packed sites, be compact enough for freely-moving rodent experiments, and reduce headstage cost by a factor of 5 (lt$1 per channel). Furthermore, the redesigned acquisition hardware will not only capture 1024 channels of neural data and store to solid-state drive over a high-speed bus, but will now also copy the data to a GPU for spike sorting and RAM for visualization both online and offline. To test the system, we will perform 1024-channel freely-moving neural recordings in rodents, in collaboration with (at least) 3 labs with expertise (see Letters of Support).