Brain Lunch: Neurotechnology Edition

The goal of systems neuroscience is to understand how neural activity generates behavior. A traditional experimental approach is to record from neural populations at times when subjects perform designated tasks. While such intermittent recordings provide brief ‘snapshots’ of task-related neural dynamics, they fail to address how neural activity is modulated outside of task context, or how it changes across behavioral states and time. Addressing these questions requires tracking the activity of neuronal populations continuously over weeks and months in behaving animals. Such experiments face significant technical challenges, including processing vast amounts of neural and behavioral data.

We present a low-cost, fully automated experimental platform that allows neural activity and behavior to be recorded continuously over several months. The large datasets we generate are analyzed using a novel processing pipeline, where the key step is an unsupervised spike-sorting algorithm that allows for automatic identification and tracking of single units in terabyte-sized datasets even when units have non-stationary spike-waveforms. We used our system to record activity in large populations of single neurons in motor cortex and striatum, often holding units for several weeks. In conjunction with the neural recordings, high-resolution behavioral data was acquired using high-speed cameras and head-mounted 3-axis accelerometers, which, together with local field potentials, were used to identify epochs of sleep, rest, grooming, feeding, and to track and quantify movement kinematics during execution of a skilled motor task. 

We found that average firing rates and correlation structure in neuronal populations were stable across many days, even as they varied substantially across different behavioral states in a single day. Additionally, we found the motor representations of skilled behaviors to be remarkably stable at the single unit level, even over month-long timescales. These results demonstrate that neural circuits can maintain distinct task representations with long-term stability at the level of single neurons.