Cortical circuits are thought to contain a large number of cell types that coordinate to produce behavior. Current in vivo methods rely on clustering of specified features of extracellular waveforms to identify putative cell types, but these capture only a small amount of variation. Here, we develop a new method (WaveMAP) that combines non-linear dimensionality reduction with graph clustering to identify putative cell types. We apply WaveMAP to extracellular waveforms recorded from dorsal premotor cortex of macaque monkeys performing a decision-making task. Using WaveMAP, we robustly establish eight waveform clusters and show that these clusters recapitulate previously identified narrow- and broad-spiking types while revealing previously unknown diversity within these subtypes. The eight clusters exhibited distinct laminar distributions, characteristic firing rate patterns, and decision-related dynamics. Such insights were weaker when using feature-based approaches. WaveMAP therefore provides a more nuanced understanding of the dynamics of cell types in cortical circuits.

The electrophysiology data was collected from two rhesus macaques (Macaca mulatta) performing a red-green discrimination task. The trained animals performed reaches towards red or green targets to the left or right of a central static checkerboard showing different combinations of red or green squares. Either single tungsten electrodes or linear multi-contact 16-channel U-probes were inserted into dorsal premotor cortex to obtain neural data described below.

We recorded 996 units (778 from U-probes) from monkey dorsal premotor cortex. Neurons were isolated through standard procedures for extracelluar recordings and verified, where needed, through offline spike sorting (Plexon Offline Sorter). A snippet of 1.6 ms (30 kHz sampling rate thus 48 time points) of each spike with trough at 0.4 ms was captured and then filtered with a 4th-order 250 Hz high-pass Butterworth filter. These spikes were then upsampled to 480 points, fit with a cubic spline, normalized to be between -1 and +1, and averaged per unit. Waveforms with initial positive phase were excluded from analysis in the manuscript due to their putative association with axons.

The behavioral data is described in Chandrasekaran, C. et al. (2017).

The confocal imaging z-stack data was extracted from 30 micrometer coronal sections of rhesus macaque dorsal premotor cortex (n = 6; 2 samples each from 3 specimens).  For each sample, immunohistochemical stains were conducted for the non-overlapping inhibitory cell type markers of parvalbumin (PV), calretinin (CR), and calbindin (CB). Z-stacks were converted to maximum intensity projections (.tiff files) and positively-staining cells were counted by hand using the CellCounter plugin for Fiji (.xml files). Depth measurements for each cell was obtained and converted to laminar distributions in Python.

Instructions for how to use the data files used in the analysis and to generate manuscript figures (in Python) are contained in the Readme.md contained in the Zip file WaveMAP_Paper.zip. Manuscript figures are also included in the zipped folder Manuscript_figures.zip

Instructions for pre-processing raw averaged waveforms (in MATLAB) and associated data are contained in the README.md file in the zipped Preprocessing folder.

Raw confocal images with stereological cell count annotations are uploaded in the Zip file Image_data.zip.