Kaiser Arndt (Dan English's Lab)

Department of Neuroscience at Virginia Tech, Blacksburg, VA, USA

We use in vivo electrophysiology in awake head-fixed mice to study granular retrosplenial cortex high frequency oscillations during hippocampal sharp wave-ripples and theta oscillations.

The ability to simultaneously record more neurons in electrical local field potential recordings is rapidly expanding as new devices are developed and technological advances are made (tetrodes to silicon probes to Neuropixels). Spike sorting algorithms responsible for extracting action potential wave-forms from extracellular electrical signals have made similar strides to those in the devices field. However, the rigor and reproducibility of the quality and accuracy of neurons extracted by these sorting algorithms are negatively impacted by the following features. (1) Accuracy of spike detection and clustering spanning multiple behavioral states coinciding with key LFP oscillations (EMG, ripples, theta) is poorly understood. (2) Action potential wave-forms markedly change over the course of single unit and population bursts allowing for potential commission and omission errors in cluster assigned spikes. (3) the first two issues are altered by the cell types and brain regions of interest (cortex vs thalamus vs hippocampus) which these algorithms do not take into account. These issues can be attributed, in some part, to the lack of sufficiently long ‘ground truth’ data that meets statistical power requirements. Ground truth data for spike sorting consist of extracellular recordings in which a subset of neurons is also recorded with a second, unambiguous technique, such as a glass electrode: a significant technical challenge. Ground truth data are essential for quantifying spike sorting performance and developing automatic spike sorters to replace manual methods which are time-consuming and error prone. Here we use silicon-juxtacellular hybrid microelectrode probes to obtain ground truth data in awake head-fixed mice. We found spike waveforms significantly change over the course of a burst in a non-linear fashion. This resulted in a significant depreciation of the signal to noise over the course of bursts, corresponding in an increase in omission errors. These effects were compounded by large amplitude local field potential oscillations, specifically EMG noise and sharp-wave ripples in our hippocampus recordings. Our data support the need for a growing ground truth data set across brain regions in all brain states, to accurately tune spike sorting algorithms to eventually be fully autonomous.

Current Publication: https://www.biorxiv.org/content/10.1101/2023.07.10.547981v1.