High-Resolution Laminar Identification in Macaque Primary Visual Cortex Using Neuropixels Probes

Laminar electrode arrays allow simultaneous recording of activity of many cortical neurons and assignment to layers using current source density (CSD) analyses. Electrode arrays with 100-micron contact spacing have been used to estimate borders between layer 4 versus superficial or deep layers, but in macaque primary visual cortex (V1) there are far more layers, such as 4A which is only 50–100 microns thick. Neuropixels electrode arrays have 20-micron spacing, and thus could potentially discern thinner layers and more precisely identify laminar borders. Here we show that laminar distributions of CSDs lack consistency and the spatial resolution required for thin layers and accurate layer boundaries. To take full advantage of high density Neuropixels arrays, we have developed approaches based on higher resolution electrical signals and analyses, including spike waveforms and spatial spread, unit density, high-frequency action potential (AP) power spectrum, temporal power change, and coherence spectrum, that afford far higher resolution of laminar distinctions, including the ability to precisely detect the borders of even the thinnest layers of V1.


Figure S1
. Comparing spatial averaging to downsampling for CSD profiles evoked by flipping the black screen to the white screen through the dominant eye.The same four penetrations in Figure 2E-H are shown here (A: A10, B: A14, C: B12, D: B13).In the downsampling process, LFPs from electrodes with increasing vertical spacing (d=20, 40, 60, 80, 100µm, gray) were used to calculate CSDs, then they were interpolated (Cubic Spline) to the vertical resolution of 20µm.In the spatial averaging process, CSD profiles of 20µm spacing were Gaussian filtered with increasing width (σ=20, 30, 40µm, red).The CSD profiles were independently color-mapped according to the maximum absolute deviation (MAD) in each profile.

Figure S2 .
Figure S2.Average Metrics in V1 template.(A) Normalized average power for each stimulus.Gray lines are for each penetration, and red lines and shaded ribbons show Mean±SEM.(B) Normalized average local coherence for each stimulus is shown similarly to (A).(C) Normalized average ΔP/P in time window: [30, 100]ms after stimulus onset.(D) Normalized average ΔCSD in the same time window of C. The colormapped lines and shaded gray ribbons in C and D show Mean±SEM.Square Markers are the same as in the figures of this article.

Figure S3 .
Figure S3.Coefficient of Variation.Variability of normalized depth profiles (gray lines in Figure S2) across penetrations is accessed by coefficient of variation (STD/Mean) for power (red), local coherence (blue), ΔP/P (orange), and ΔCSD (green) in response to different stimulus conditions.Square Markers are the same as in the figures of this article.

Figure S4 .
Figure S4.Depth profile of spike amplitude and firing rate.(A) Spike amplitude.(B) Firing rate during the whole recording session of a penetration.The stimulus set includes static/drifting, achromatic/chromatic gratings.(A-B) gray lines for each penetration, red line and shaded ribbon for Mean±SEM.(C) PSTH on the normalized layer template.Square Markers are the same as in the figures of this article.

Figure S7 .
Figure S7.Thickness of primary visual cortex.(A) The distance was measured from the location of penetration perpendicular to V1/V2 border.Here, all penetrations except A17 (penetration location information lost) were included.(B-D) Distributions of thickness across two monkeys, different ocular dominance columns, and COFD groups.significant differences were found in each category (Kruskal-Wallis Test, p>0.05).COFD-A: Achromatic ON/OFF domains; COFD-LM: L-and M-cone ON/OFF domains; COFD-S: S-cone ON/OFF domains.

Figure S8 .
Figure S8.Average metrics from gamma band LFP in V1 template.(A-B) Average gamma band (30-100Hz) spectrum profile of power (A) and local coherence (B) across penetrations.The square markers and color bars were the same as the figures in this article.(C-D) Normalized average power and local coherence in the gamma band.Gray lines are for each penetration, color-lines and corresponding shaded ribbons show Mean±SEM.

Figure S9 .
Figure S9.Average metrics from baseline window in V1 template.(A-B) Average baseline AP spectrum profile of power (A) and local coherence (B) across penetrations.The square markers and color bars were the same as the figures in this article.(C-D) Normalized average baseline AP power and local coherence.Gray lines are for each penetration, color-lines and corresponding shaded ribbons show Mean±SEM.

Figure S10 .
Figure S10.AP power and local coherence averaged across frequencies for each trial.(A-C) Powers for the same penetrations (A: A2, B: A3, C: B1) in Figure 5. (D-F) Local coherences for the same penetrations (D: B12, E: B10, F: B3) in Figure 6.Power and local coherence profiles were independently color-mapped according to the minimum and maximum in each profile.The green lines and corresponding shaded ribbons showed the Mean±SEM across trials.

Figure S11 .
Figure S11.Penetration sites on V1 surface of monkey A. White lines together with L/R texts indicate the of ocular dominance column.

Figure S12 .
Figure S12.Electrodes drift correction.(A) Probe drifting during the recording time for penetration B4 was visualized as average spike amplitudes shifting in consecutive batches (batch length: ~3.3 sec).(B) Probe drifting of penetration B1 same batch length as penetration B4.Drift estimated for penetration B1 using the default method in Kilosort 3 (drift range: [-300, 300] µm, number of blocks: 13, drift range for each block: [-300, 300] µm).(D) Drift estimated for penetration B1 using `imregdemons` function in MATLAB.(E) Expected correction result of penetration B1 after applying (`imwarp` function in MATLAB) the estimated drift in C. (F) Expected correction result of penetration B1 after applying (`imwarp` function in MATLAB) the estimated drift in D.