Ground truth in ultra-dense neural recording

Abstract

While biologists routinely record neural activity with multi-electrodes, spike sorting-- the process of attributing spikes to particular neurons-- remains a challenge that typically requires human curation. Due to technical limitations, there have been very few multi-electrode recordings done in concert with techniques such as patch clamp, which report the "ground truth" voltage state of a single neuron in a population. Such recordings would allow for the direct evaluation of spike sorting, which in turn could lead to further development and refinement of spike sorting methods. We developed a technique to establish a whole-cell or cell-attached patch recording in a cortical neuron of an awake or lightly anesthetized head-fixed mouse, with simultaneous extracellular recording of the same neuron and its neighbors with arrays of close-packed, "ultra-dense," electrodes (64-256, 9 x 9[mu]m electrodes spaced 2[mu]m apart on a shank). Our recordings constitute ground truth for spike sorting evaluation, and allow for the direct evaluation and improvement of an algorithm for automatic spike sorting that benefits from high electrode density relative to neuron packing density. Using this technique we show the patch-triggered extracellular waveforms of neurons at a high level of granularity distributed across cortex, and give a glimpse into the spiking activity of the network surrounding a patched neuron in vivo. We explore the dataset generated with this technique and discover a spike-bursting trajectory exhibiting apparent spike-frequency adaptation. This bursting trajectory was readily apparent in deep but not shallow cortical neurons in patch recordings, but was somewhat obscured in extracellular recordings, where spikes from neighboring neurons may overlap in time to contribute "noise." We show how this trajectory can be easily seen in a high-amplitude extracellular recording, and propose how it may be accentuated in lower amplitude recording through the use of blind source separation.