ACORN: Adaptive Consensus of Repeated Clusterings Identifies Pyramidal Neurons and Interneurons in the Primate Motor Cortex Based on Extracellular Spike Waveforms.

Abstract

Understanding how neural circuits generate behavior requires knowledge of both neural activity patterns and the cell types they originate from. While the scale of neuronal recordings has dramatically increased since the advent of dense multicontact probes, accurate identification of cell types remains elusive. Traditional approaches have used spike width to attempt a classification, with narrow spikes assumed to be inhibitory interneurons. However, fast pyramidal tract neurons (PTNs; known to be excitatory) with thin spikes and interneurons with wide spikes challenge these assumptions. Here we accomplish three related objectives. First, we produce a ground truth dataset of extracellular spike shapes from electrophysiologically identified pyramidal neurons from primary (M1) and supplementary motor areas (SMA) and optogenetically identified interneurons in M1 in female macaque monkeys. Next, we show that spike width correlates with antidromic latency not only in M1 but also in the SMA, suggesting that it is a physiologically meaningful measure across the motor cortex. Finally, we introduce our unsupervised ensemble clustering method, "ACORN," and characterize its performance on our neocortical macaque dataset, the C4 cerebellar database, and a juxtacellularly recorded mouse dataset, demonstrating strong out-of-the-box performance across species, brain areas, and recording techniques. ACORN can cluster neuronal waveforms without training or tuning and provides data-derived confidence values for its predictions. Our ground truth data can be used as a reference to help identify or verify clusters in novel datasets. Together, our method and database provide a streamlined foundation for electrophysiological cell classification without necessitating changes to recording paradigms.