Fig 1.
Comparing GPCSD to other methods on simulated CSD dipole.
(A) From left to right, ground-truth noiseless LFP, ground-truth CSD pattern, traditional CSD (tCSD) prediction, GPCSD prediction, and kernel CSD (kCSD) prediction. (B) Same as top row, but with noise added to the ground-truth LFP before CSD prediction. GPCSD accurately recovers the pattern compared to tCSD, and appears more robust to LFP noise than either tCSD or kCSD.
Fig 2.
Spatial, spectral, and temporal properties of LFP and GPCSD.
(A) Periodograms (averaged across trials) showing power spectra at six cortical depths along the lateral probe for CSD (left) and LFP (right), computed for estimated fast-timescale processes (solid) and slow-timescale processes (dashed). Horizontal grey dotted lines indicate approximate boundaries between superficial, medium, and deep cortical layers. (B) Relative 10Hz power for CSD (red) and LFP (blue) as a function of cortical depth. (C) Time courses for a single trial for an electrode in the middle cortical layers, broken into slow, fast, and total, for CSD (left) and LFP (right). While the temporal components were similar in LFP and CSD, they did exhibit differences due to the different spatial properties of the CSD.
Fig 3.
Phase coupling graphs in auditory CSD and LFP.
(A) Results of edgewise torus graph phase coupling inference both within- and across-probe for CSD (left) and LFP (right); depth along each probe is increasing left to right and top to bottom, and dashed lines indicate approximate boundaries between superficial, medium, and deep cortical layers. Colored entries correspond to edges, with color representing the log10 of the p-value. Within-probe, the LFP had edges primarily along the diagonal, while the CSD contained more edges, including some connections across superficial, medium, and deep layers. Between probes, the CSD torus graph contained a noticeable set of edges, primarily along the diagonal, while the LFP torus graph had very few edges between probes. (B) Graph showing significant between-probe CSD torus graph edges, with lateral probe nodes ordered by depth in the left column and medial probe nodes ordered by depth in the right column; dashed lines indicate approximate boundaries between superficial, medium, and deep cortical layers. Many of the cross-probe connections occurred near the same depth on both probes, though there appear to be some edges connecting lateral probe deep layers to medial probe superficial layers. (C) Simplified graph between superficial (S), medium (M), and deep (D) cortical layers. Edge color corresponds to a 95% bootstrap confidence interval lower bound for the partial PLV value (reflecting coupling strength, which falls between 0 and 1). The strongest cross-probe connection was between the deep layers of the lateral probe and the superficial layers of the medial probe.
Fig 4.
Estimated CSD evoked response components.
Left to right: Trial-averaged multi-unit activity (MUA) relative to baseline, estimated CSD evoked response, components returned by the image segmentation (colors correspond to arbitrary cluster number). Top row corresponds to the lateral probe and bottom row to the medial probe. The evoked responses for both probes have similar features but slightly different spatial and temporal properties. In particular, both the MUA and CSD evoked responses indicate that the evoked response begins earlier in the medial probe than the lateral probe.
Fig 5.
Per-trial shifts and correlations in CSD evoked response components.
(A) Spatiotemporal plot of evoked response components for the lateral probe (left) and the medial probe (right), colored to show separate components. Black circles represent centers of mass of each component, and an edge between two black circles indicates significant correlation in the per-trial shifts of the two components (darker, thicker edges indicate larger correlation values). Within each probe, the pattern of connections suggests that many components of the early evoked response (occurring before 60 ms) have related time shifts on a trial-to-trial basis; the lateral probe also has correlated time shifts in the later evoked components. Between probes, there are connections between early evoked components at similar depths, with some evidence of shift correlations between lateral probe early components and medial probe later components. (B) Kernel density estimates of the peak times, across trials, of the evoked components in each probe with dashed lines marking putative cortical depth boundaries (separating superficial, medium, and deep layers). The medial probe responses generally precede the lateral probe responses across depths (confirmed by pairwise testing on the difference in means, p < 0.001 corrected). In addition, the ordering of responses across depths appears similar in each probe, with the earliest responses occurring in the superficial and deep layers, followed by the medium depth layers.
Fig 6.
Phase coupling in Neuropixels data.
(A) Neuropixels probe LFP electrode locations (circles) for V1 (left) and LM (right), colored by putative region (VIS: visual cortex, CA: Cornu Ammonis, DG: dentate gyrus, TH: thalamus, N-L: no label). Putative cortical layer boundaries are overlaid on the red visual area electrodes with layer numbers indicated along the right side. Along the center of the probe are the locations we chose for CSD estimation (yellow diamonds), with one location centered in each cortical layer. (B) Torus graph phase coupling graphs based on theta oscillations at two time points relative to the stimulus. Edges shown for torus graph p < 0.0001. Edge color indicates the lower bound of a 95% bootstrap confidence interval on the partial PLV value. It appears the strongest edges were between V1 and LM at similar cortical depths, with some evidence that edges were stronger at 70ms than at 0ms. (C) Similar to B, but for beta oscillations; dashed edges indicate edges with weaker evidence (torus graph p < 0.001). Similar to theta band, the connection patterns were mostly across similar layers but appeared slightly stronger at 70ms compared to 0ms.