Automated detection of extracellular action potentials from single neurons

Abstract

Multi-electrode arrays (MEAs) non-invasively record extracellular action potentials (eAPs, also known as spikes) from hundreds of neurons simultaneously. However, because extracellular electrodes sample from the local electrical field, each electrode can detect eAPs from multiple nearby neurons. Interpreting spike trains at individual electrodes of high-density arrays requires spike sorting, a computational process which groups eAPs from single units based on assumptions of how spike waveforms correlate with different neuronal sources. Additionally, when experimental conditions result in changes to eAP waveforms, spike sorting routines may have difficulty correlating eAPs from multiple neurons at single electrodes before and after such waveform changes. We present here a novel, empirical method for unambiguously isolating eAPs from individual, uniquely identifiable neurons, based on automated multi- point detection of action potential propagation. This method is insensitive to changes in eAP waveform morphology because it makes no assumptions about the relationship between spike waveform and neuronal source. Our algorithm for automated detection of action potential propagation produces a fingerprint that uniquely identifies those spikes from each neuron. By unambiguously isolating eAPs from multiple neurons in each recording, on a range of platforms and experimental preparations, our method now enables high-content screening with contemporary MEAs. We outline the limitations and strengths of propagation-based isolation of eAPs from single neurons and propose how our automated method complements spike sorting and could be adapted to in vivo use.