Ultra-high density electrodes improve detection, yield, and cell type identification in neuronal recordings

Abstract

To understand the neural basis of behavior, it is essential to sensitively and accurately measure neural activity at single neuron and single spike resolution. Extracellular electrophysiology delivers this, but it has biases in the neurons it detects and it imperfectly resolves their action potentials. To minimize these limitations, we developed a silicon probe with much smaller and denser recording sites than previous designs, called Neuropixels Ultra (NP Ultra). This device samples neuronal activity at ultra-high spatial density (~10 times higher than previous probes) with low noise levels, while trading off recording span. NP Ultra is effectively an implantable voltage-sensing camera that captures a planar image of a neuron's electrical field. We use a spike sorting algorithm optimized for these probes to demonstrate that the yield of visually-responsive neurons in recordings from mouse visual cortex improves up to ~3-fold. We show that NP Ultra can record from small neuronal structures including axons and dendrites. Recordings across multiple brain regions and four species revealed a subset of extracellular action potentials with unexpectedly small spatial spread and axon-like features. We share a large-scale dataset of these brain-wide recordings in mice as a resource for studies of neuronal biophysics. Finally, using ground-truth identification of three major inhibitory cortical cell types, we found that these cell types were discriminable with approximately 75% success, a significant improvement over lower-resolution recordings. NP Ultra improves spike sorting performance, detection of subcellular compartments, and cell type classification to enable more powerful dissection of neural circuit activity during behavior.

Figure 1.. Neuropixels Ultra probes have a considerably denser site layout over a smaller spatial range compared to NP 1.0 and 2.0.

A (left), Schematic of a complete NP Ultra probe, including headstage and flex cable. (right) Scanning electron microscope images of an NP Ultra probe tip (top) and individual contacts, each with two slits that increase electrode surface area (bottom, top inset). B, Layout of NP Ultra sites compared to previous probes. C, Comparison of an example waveform on all three probes. The NP Ultra panel (right) shows the mean waveform of a unit recorded in the mouse primary visual cortex; this waveform was spatially re-sampled to approximate the signal on the 12 × 12 μm grid for NP 1.0 (left) and NP 2.0 (middle) site configurations. Heatmaps represent the interpolated voltage as proportion of the trough-to-peak amplitude recorded on the peak amplitude channel in the NP Ultra configuration. D, Example reconstructed morphology of a mouse L5 pyramidal neuron from the Allen Institute Cell Types Database (ID: 488679042) compared at scale to a NP Ultra probe. E, Example raw traces from a column of vertically adjacent recording sites, plotted according to their depth along on an NP Ultra probe. Shading indicates the time and location of two detected spikes. F, Measures of noise in saline (root-mean-square) and during recording in vivo (median absolute deviation) for four NP Ultra and four NP 1.0 probes. G, Example spatial (top left), spatiotemporal (top right) and temporal (bottom) waveforms recorded by NP Ultra from a regular-spiking (left) and a fast-spiking (right) unit in the visual cortex of an awake mouse. Spatial plots show the contour map of channel-wise maximum amplitudes normalized to the peak channel. The spatio-temporal plots represent voltage as a colormap across all channels in the column of sites containing the peak amplitude site. The temporal plots show the peak amplitude channel (red) with 40 nearby channels (gray).

Figure 2.. Localizing, clustering, and tracking spikes from Neuropixels Ultra probes.

A, “Drift map” spatial spike raster of detected neurons in an example recording. Each point represents one spike; y-coordinate represents (unregistered) depth along the probe at which the spike was detected; x-coordinate detection time; color is log-scaled spike amplitude. Superimposed black line indicates the estimated motion of the probe relative to the brain. Green horizontal lines indicate top and bottom boundaries of the probe. B, After probe motion registration, detected spikes are localized in 2D across the face of the probe. Left panel, each point represents a spike in registered 2D space, colored by cluster identity, with ellipses representing the spatial spread of each cluster. Green markers indicate the recording site locations. Right, scatter plot of depth versus spike amplitude; colors as in left. C, The template waveform for cluster 12, with the super-resolution template estimates overlaid. Left, the non-registered positions of cluster 12 spikes. The colored rectangles overlaid on top of the non-registered spike positions demarcate the spikes used for estimating each super-resolution template. Middle-left, super-resolved templates for cluster 12 are overlaid, showing how these templates capture shape variability for this unit. They are shown on 8 channels corresponding to 8 rows of the middle-right column of the recording channels. Each template is colored according to the bin where it was estimated shown in the left panel. Middle, residual waveforms obtained after subtracting the super-resolution templates shifted according to the estimated motion. Middle-right, residual waveforms obtained after subtracting the mean template. Right, residual waveforms were obtained after subtracting the mean template, not shifted. D, Drift map as in A, post-motion registration. Spikes colored by their cluster identity. E, Three spatial resampling patterns (Large Dense, NP2.0-like, and NP 1.0-like, from left to right). Superimposed grids indicate the NP Ultra site locations (thin gray grid) and the channel size and positions for the resampling patterns (bold black grid and colored sites). F, Left, Grid indicating the NP Ultra recording channels (black) and the NP 1.0-like resampled sites (magenta squares). Colormap indicates amplitude as in Fig. 1C. The corresponding NP Ultra and resampled waveforms are shown on the right, with all channels overlaid on top of each other, and the main channel colored in green for the NP Ultra and magenta for NP 1.0-like resampled pattern. G, Cumulative distribution function of all spike amplitudes, comparing large-dense, 2.0-like, and 1.0-like site arrangements. H, Peak site amplitudes of neurons in the original dataset recorded with NP Ultra versus the spatially resampled versions. * indicates the statistical significance of the difference, obtained using a two-sample z-test on the unit amplitude distribution. I, Template SNR measured for each recording site pattern, from the original dataset and spatially resampled versions. J, Spatial spread, computed as the standard deviation of coordinates, in the horizontal dimension. K, Spatial spread in the depth dimension across the face of the probe.

Figure 3.. Tracking neurons with NP Ultra using visually-evoked fingerprint responses.

A, Spatial (top left), spatio-temporal (top right) and temporal (bottom) waveforms recorded by NP Ultra of an exemplar neuron shown in B and C. B, Autocorrelogram of exemplar neuron shown in A and C. C, Peri-stimulus spike rasters and average rate (smoothed with causal half-gaussian filter with standard deviation = 45 ms) for an example neuron with reliable and stable visual response, aligned to the onsets of 3 of the 118 visual stimuli (indicated in 3 different bluish hues). Top two plots show odd and even trials from the pre-motion period, revealing a close match. Bottom left plot shows shuffled image ID control, with a poor match. Bottom right shows the post-motion responses, which again match the pre-stimulus un-shuffled responses. D, Mean yield of neurons (across n=6 sessions) from NP Ultra and spatially resampled datasets progressively filtered by the following three metrics: ≤4x difference in pre vs. post-motion firing rate (FR) in spikes/second (“Stable FR”), reliable visual response (“vis. resp.”), and 90% confidence of ≤20% contamination in the sliding refractory period metric (“sliding RF pass”). E, Differences in neuron yield between NP Ultra and each spatially resampled pattern calculated per session, using the same quality metrics as in D. Bars and error bars are mean±SEM. Stars indicate significant p-values from the Wilcoxon signed-rank test. F, Amplitudes of stable neurons with reliable visual responses across all sessions. Black diamonds and error bars are median and n=10000 bootstrapped 95% confidence interval of the median. Stars indicate significant p-values from Mann-Whitney U test. G, Stability ratios (ratio of correlations between mean pre- vs. post-motion and mean pre- vs. pre-motion visual fingerprints; see Supp. Fig. S12B,C) for neurons with reliable visual responses across all sessions. Colored points indicate neurons with a stable firing rate from the pre- to post-motion period; gray dots indicate neurons with unstable firing rates. Black diamonds and error bars are mean±SEM only from neurons with stable firing rates. Stars indicate significant p-values from unpaired two-sample t-test; only stable firing rate neurons included in the test. For all panels, * indicates p<0.05; ** indicates p<0.01, *** indicates p<0.001.

Figure 4.. Recordings from subcellular compartments.

A, Example small-footprint unit recorded from primary visual cortex. B, Kernel density histogram of measured spatial footprint across a total of 9,496 units recorded in visual cortex. Dashed line denotes the boundary (20 μm) between what we define as large and small footprint units. C, Schematic of NP Ultra recordings in cortex with surface application of muscimol [5 μM]. Red lines indicate axonal segments of intra- and extra-cortical neurons. D, Raster plot showing the firing rate of 77 units during an exemplar NP Ultra recording. Black arrow = time of visual stimulation, green arrow = time of muscimol application. Units are sorted by ascending footprint size, the boundary between large and small units is denoted by the red dashed line. E, Scatter plot of firing rate across all units measured within a 100 ms window pre- and post-muscimol application, colored by where a given unit was resistant or sensitive to muscimol. Dashed line indicates identity between pre- and post-FR. Insets display the waveforms and heat maps of two example muscimol-resistant units. F, Relationship between the spatial footprint and firing rate modulation index for all recorded units. Units from E marked by black boxes. G, Full waveform plots of the two units in E, respectively. H (left), Schematic of optotagging experiment in Sim1-Cre;Ai32+ L5b pyramidal neurons. (Right) histology from a Sim1-Cre;Ai32 animal showing labeled L5 pyramidal neurons in S1. I, PSTH of an example Sim1+ unit in response to 10 ms 470 nm photostimulation. J, Two example Sim1+ units with visible potentials propagating along the NP Ultra probe. K, Schematic of a L5 cell recorded with a switchable NP Ultra probe in the linearized (192 × 2 channels) configuration. L, Example bAPs recorded with NP Ultra (left) and NP 1.0 (right). Traces show spike-triggered average waveforms (n = 2,000 spikes) recorded from the column of channels that includes the peak amplitude channel. Colored traces track the minimum voltage recorded on each channel with a recorded amplitude above 5% of the maximum. M, Dendritic channel template SNR measured as a function of vertical distance from the peak amplitude channel for NP Ultra (192×2 configuration, best column selected, green), and NP Ultra (192×2 configuration, 4-channel average, cerulean) or NP 1.0 probes (magenta). Channel configuration diagrams are shown on the top right. N, Template SNR plotted against channel waveform amplitude for each channel in every probe configuration from M.

Figure 5.. NP Ultra recordings reveal small electrical footprint units in many brain regions of the mouse.

A, NP Ultra insertion trajectories from different mice and sessions, registered to the CCF atlas coordinates. B, Percentage of units with small footprints less than 20 μm in regions with the most recorded single units. Number of small units per region indicated. Dashed line corresponds to 10% C, Two example waveforms with one large and one small spatial footprint from each of six different brain regions, and spatial footprint histogram of all units from these regions. MO: Somatomotor areas; ORB: orbital area; VIS: visual areas; PIR: Piriform area; CTXsp: cortical subplate; CA1: hippocampal CA1; SUB: Subiculum; CA3: hippocampal CA3; DG: Dentate gyrus; cc: corpus callosum; VPM: ventral posteromedial nucleus of the thalamus; VPL: Ventral posterolateral nucleus of the thalamus; PO: Posterior complex of the thalamus; CP: Caudoputamen; ACB: Nucleus accumbens; MB: midbrain.

Figure 6.. Small footprint units are detected across different species.

A, Two example waveforms from four species (mouse visual cortex, monkey visual cortex, electric fish cerebellum, and bearded dragon medial cortex), with one large and one small spatial footprint. Two waveforms with similar peak amplitude in the same species were selected for comparison. B, Spatial footprint histogram of all units from each species.

Figure 7. Optotagging three inhibitory neuron classes measured with NP Ultra.

A, Scatter plot of waveform duration and footprint size, colored by the density of units within a given region of the graph. Note the existence of two maxima, one with relatively long waveform durations and large footprint, and another with short-duration waveforms and small footprint. B, Distributions of footprint size for all units, segregated by waveform duration (RS = regular-spiking, FS = fast-spiking). FS units displayed a bimodal distribution split at 20 μm. Black dashed line indicates the demarcation between “large” and “small” units. C, Schematic of experimental setup in which neurons in visual cortex transgenically expressing channelrhodopsin were recorded with NP Ultra probes and optotagged using 470 nm light. D, Trial-wise optogenetically-driven responses of an example PV-Cre;Ai32-expressing unit during a 10 ms 470 nm light pulse. E, Two example spatial, spatio-temporal, and temporal waveforms from optotagged units in each of the examined transgenic mouse lines (PV-Cre;Ai32, SST-Cre;Ai32, VIP-Cre;Ai32). F, Distributions of spatial footprint for each optotagged group of units. G, Left, Density histograms displaying the distribution of footprint between PV and FS units. FS units are subcategorized into FS_L (fast-spiking, large) and FS_S (fast-spiking, small) based on a footprint size threshold (20 μm). Right, Accuracy of Linear Discriminant Analysis (LDA) in separating different types of FS and PV units using pre-peak-to-trough ratio (prePTR) and peak-to-trough ratio (PTR) as features. H, Left, Cross-correlograms displaying the firing rate of example units as a function of spiking in two FS_S units (top/bottom rows). Dashed line indicates the time of an FS_S unit spike. Right, Autocorrelograms of the example FS_S units.

Figure 8.. Enhanced classification of inhibitory neuron types with NP Ultra.

A, Schematic of features extracted from single-channel waveform (1-ch features). Amplitude is the absolute difference between trough and peak. Duration is the time between trough and peak. Peak-to-trough ratio (PTR) is the ratio between absolute amplitudes of peak and trough, prePTR is the ratio of the absolute amplitudes of the pre-peak and trough. Red lines show slopes for repolarization and recovery. B, Comparison of waveform features extracted from the peak amplitude channel for each unit within a transgenic mouse line (all metrics except duration are log-scaled). All untagged units were split into three groups based on spike duration (RS = regular-spiking, FS = fast-spiking) and footprint size. Dots indicate the median and black vertical lines indicate the interquartile range. Dashed horizontal lines indicate the population median for each metric. C, UMAP projection of an equally sampled subset of NP Ultra-recorded units from each cell class. D, Classifier performance (out-of-bag score, mean ± S.D.) on NP Ultra (green) and NP 1.0 (blue) data using single and multi-channel features. E, Confusion matrices for opto-tagged only (left) and all cell class (right) classification using that 1ch features + footprint dataset. F, Difference in True Positive (TP) rate between NP Ultra and NP 1.0-like data for each cell class. G, Classifier accuracy (mean ± 95% confidence interval) across all cell classes when one feature is iteratively excluded from the dataset. Asterisks indicate significant changes in accuracy as a result of feature removal. Dashed line is the mean classifier accuracy with all 1-ch features + footprint. Inset shows the accuracy of the classifier when no 1-ch features are included in the dataset.