Fig 1.
GPi DBS microelectrode recordings.
(A) Diagram of serial microelectrode recordings for DBS implantation surgery and microelectrode dimensions. (B) Raw voltage is recorded from the microelectrode tip, referenced to the cannula, and filtered from 300 Hz to 9 kHz to analyze single units. (C) Microelectrode recording position was determined by co-registration of the post-surgical CT and pre-surgical MRI, and resliced in plane with the DBS lead position. (D) DBS lead contacts were identified and used to infer the microelectrode recording track. Yellow arrow indicates target depth. (E) T2, T1, and FGATIR MRI series were overlaid to segment GPi and surrounding anatomical borders. GPi = Internal globus pallidus; GPe = External globus pallidus; Put = Putamen; IC = internal capsule; Cd = Caudate; Th = Thalamus; LV = Lateral ventricle.
Fig 2.
Typical ‘cell-types’ seen on microelectrode recordings targeting the GPi.
(A) Representative pseudo-coronal T2 MRI displaying intraoperative micro recording track and zoomed-in segmentation of the GPi and surrounding anatomy. (B) Typical cell-types seen on microelectrode recordings targeting the GPi include bursting and pausing cells in the GPe and border cells and irregular firing neurons in the GPi. GPi = Internal globus pallidus; GPe = External globus pallidus; Put = Putamen; IC = internal capsule; Cd = Caudate; Th = Thalamus; LV = Lateral ventricle.
Fig 3.
Parameterizing spike units using CRP.
(A) Example of spike units sorted from a sample microelectrode recording site shown on pseudo-coronal T1 MRI. (B) Sorted units are parameterized using canonical response parameterization (CRP). Forward CRP identifies the spike end time point () at the maximum projection magnitude (
). Reverse CRP starting from
identifies the spike start time
creating a significant spike voltage window. (C) Individual spikes are parameterized by sorted unit shape C(t) and scaling factor
. Removing scaled spike shapes from the raw voltage trace V(t) leaves the residual local field potential
(LFP). (D) Overlaid V(t) and
with
, for spike k of unit n, at spike times
over duration
:
.
Fig 4.
Serial parameterization and spike sorting.
(A) Example spike sorted trace from Fig 3. Multiple spikes occurring in the same spike window are missed by standard spike sorting. (B) The largest unit is parameterized to obtain for unit 1. (C)
is subtracted from the filtered voltage trace and re-sorted using the same voltage threshold. (D) The second unit is parameterized. (E) parameterization of both units shows resolved overlapping spikes. (F) For the sample voltage trace, serial parameterization and spike sorting increased the number of detected spikes by over 900.
Fig 5.
Example of unit shapes, durations, projections, and parameterizations.
(A) Four examples illustrate different unit shapes, their durations, and projection magnitude profiles. (B) Parameterization values for example units, including spike window duration, spike magnitude, signal-to-noise ratio, and variance explained, respectively.
Fig 6.
Spike similarity comparisons across recording sites.
Spike similarity is calculated by taking the dot product of two amplitude-normalized units over the outer limits of to
. (A) Sample comparison of highly similar unit shapes with different raw amplitudes. (B) Sample comparison of two units with opposite polarity and a negative similarity score. (C) Sample comparison of two dissimilar unit shapes. A similarity score of 1 indicates identical shapes, -1 for identical inverted shapes, and 0 for completely dissimilar shapes.
Fig 7.
Hierarchical clustering of spike unit shapes.
Hierarchical clustering was performed on the distance metric (1-similarity score) of all sorted units in the left GPi microelectrode recording track. (A) The hierarchical structure was visualized using a dendrogram. The number of unique cluster shapes was determined to be 6 based on the elbow point method. (B) Spike similarity matrix and resulting clusters. (C) Normalized unit waveforms and average cluster shape. Shape 3 (n = 22) and shape 2 (n = 17) were the most common shapes present, followed by shapes 5 (n = 5) and shape 1 (n = 4). Shapes 4 and 6 were both composed of one unit.
Fig 8.
Spike shape clustering on anatomy.
(A) The 6 clustered spike shapes plotted by depth from target in the GPi. (B) Spike shapes are plotted along the visualized microelectrode recording track on hand-segmented anatomy on psuedocoronal FGATIR MRI. Spike shapes showed regional clustering, consistent with white matter (lamina, IC), GPe, and GPi regions. (C) Mean (+/- SEM) inter-spike intervals and firing rates across clustered spike shapes. Shape 5 was present at the very top and very bottom of the track. Shape 3 was divided into 2 groups from +6 mm to +10 mm and -2 mm to 0 mm from target. Shape 2 was present primarily between +1 mm and +5 mm. Shape 1 was closely intermixed with the upper cluster of shape 3 units. GPi boundaries were closely associated with shape 2 and the bottom cluster of shape 3. Segmented GPe boundaries were associated with shape 1 and the top cluster of shape 3. Shape 5 occurred near the top and bottom of the MER track. Put = Putamen; GPe = External globus pallidus; GPi= Internal globus pallidus; IC = internal capsule.
Fig 9.
Capturing pulse-modulated firing using .
(A) Raw signal traces from the motor thalamus microelectrode recording and EKG. (B) a single cardiac cycle was isolated (-0.1 to 0.9 from R-wave) with two parameterized units. (C) The mean of each parameterized spike was plotted in reference to the aligned R-wave of the average EKG signal. An impulse-response function was fitted to the baselined distribution using a nonlinear least square fitting algorithm using the trust-region-reflective method in the form
. Unit 1 shows clear impulse-response-like pulse modulation (
; n = 2.33; m = 9.60;
), whereas unit 2 did not show cardiac-tuning (A = 24.79; n = 0.67; m = -2.95;
).
values show that unit 1 is pulse-modulated, while unit 2 is not.
Fig 10.
Local field potential isolation using spike parameterization and removal.
(A) Example raw voltage trace from a microelectrode recording site. (B) zoomed in raw voltage trace. (C) Two sorted units parameterized and plotted on the filtered voltage trace. (D) subtracting the parameterized spikes from the unfiltered trace isolates the local field potential. (E) Power spectral density was calculated from the raw and LFP signal with spikes removed. (F) Average 200–300 Hz band power at 6 different recording sites from the raw signal (black) and LFP (Green). (G) Spike removal at 6 recording sites from one MER track showed to a significant reduction in 200–300 Hz band power by approximately 25% (paired t-test, t(5) = 5.37, *p = 0.003).