Cortico-cerebellar coordination facilitates neuroprosthetic control

Abstract

Temporally coordinated neural activity is central to nervous system function and purposeful behavior. Still, there is a paucity of evidence demonstrating how this coordinated activity within cortical and subcortical regions governs behavior. We investigated this between the primary motor (M1) and contralateral cerebellar cortex as rats learned a neuroprosthetic/brain-machine interface (BMI) task. In neuroprosthetic task, actuator movements are causally linked to M1 "direct" neurons that drive the decoder for successful task execution. However, it is unknown how task-related M1 activity interacts with the cerebellum. We observed a notable 3 to 6 hertz coherence that emerged between these regions' local field potentials (LFPs) with learning that also modulated task-related spiking. We identified robust task-related indirect modulation in the cerebellum, which developed a preferential relationship with M1 task-related activity. Inhibiting cerebellar cortical and deep nuclei activity through optogenetics led to performance impairments in M1-driven neuroprosthetic control. Together, these results demonstrate that cerebellar influence is necessary for M1-driven neuroprosthetic control.

Fig. 1.. Direct and indirect modulation of M1 and cerebellar activity with neuroprosthetic learning.

(A) Schematic of neuroprosthetic task box where direct neural control of a feeding tube (θ = angular position) was exerted. Each trial started with the tube at P₁. (B) Trial structure is shown depicting when audio tone cue and door movements occur. A successful trial required movement of the tube to P₂ within 15 s. (C) Change in task completion time as a function of trial number from a representative session. Line shows moving average of 20 trials. Dots show individual trial task completion times. Inset: Illustration of the recording scheme in M1 and cerebellum from a frontal-side view. (D) Change in time to task completion (left) and reduction in the percentage of unsuccessful trials (right) from early to late trials across all sessions. Bars indicate the means and error bar is SEM. (E) Position of the feeding tube from P₁ to P₂ is shown from a single session (mean ± 1 × SD). (F) Peri-event histogram (PETH) and rasters from early and late trials from a single M1 TR_d unit is shown in left and right, respectively. (G) Same as (F) but for a M1 TR_i unit. (H) Same as (F) but for a M1 TU unit. (I) Same as (F) but for a cerebellum TR_i unit. ***P < 0.001.

Fig. 2.. Coordinated neuroprosthetic task-related oscillations emerge in M1 and cerebellar LFPs.

(A) Raw and filtered LFP trace from an example session showing increase in 3- to 6-Hz oscillations after task start during late trials in both M1 and cerebellum LFPs. Raw trace shows mean in bold overlaid on individual trial traces, and filtered trace shows mean in bold and SEM in shaded band. Right: Illustration of the recording scheme in M1 (pink) and cerebellum (orange) from a frontal-side view. (B) Spectrogram from a representative M1 channel showing increase in 3- to 6-Hz power during late trials. (C) Increase in 3- to 6-Hz power in M1 LFP from early to late trials across sessions. (D) Same as (B) but from a representative cerebellum LFP channel. (E) Same as (C) for 3- to 6-Hz cerebellum LFP power across sessions. (F) Coherogram from a representative pair of M1-cerebellum LFP channel pair showing increase in 3- to 6-Hz coherence during late trials. (G) Change in 3- to 6-Hz coherence from early to late trials across sessions. ***P < 0.001.

Fig. 3.. M1 and cerebellum spike-LFP locking increases with learning.

(A) The mean M1 LFP (top row) or cerebellum LFP (bottom row) time locked to occurrences of spikes from M1 TR_d during task period from a representative session. (B) Same as (A) for a M1 TR_i unit. (C) Same as (A) for a M1 TU unit. (D) Same as (A) for a cerebellum TR_i unit. (E) Box plot of percentage change in STA amplitude for M1 LFP in each of the categories of units (bottom and top box boundaries are 25th and 75th percentiles, respectively, line inside the box is the median, bottom and top error lines are 10th and 90th percentiles, respectively, “+” indicates outliers outside these bounds). (F) Same as (E) for changes in STA amplitude with cerebellum LFP. ***P < 0.001; nonsignificant (n.s.), P > 0.05.

Fig. 4.. Increase in neural subspace correlation between task-related units of M1 and cerebellum.

(A) Description of CCA. CCA finds a linear combination of binned spike counts from M1 units (x₁, x₂, … x_n) and cerebellum units (y₁, y₂, … y_n) that maximizes the correlation between M1 and cerebellum. (B) Example identification of significant canonical variables (CVs; green lines) relative to trial-shuffled data (gray distribution, 10⁴ shuffles). Significant threshold at 95th percentile of the distribution is shown in dotted gray line. Two CVs crossed this threshold in this example session. (C) Single-trial M1 task–related (TR) and cerebellar TR_i spiking activity along with CV1 activation from M1 TR–cerebellar TR_i CCA aligned to the task start. (D) M1 TR and cerebellum TR_i subspace activity (from the CV1) around task start (−2 to 2 s) for an example session. Each dot represents one time bin of early or late trials from the session. Canonical correlation score is given by r. (E) Change in the canonical correlation score from early to late trials across all sessions for M1 TR and cerebellar TR_i units. (F) Same as (E) for M1 TU and cerebellar TR_i units. ***P < 0.001; n.s., P > 0.05.

Fig. 5.. Cerebellum TR_i neural activity predicts M1 BMI-potent neural activity.

(A) Description of GLM model; GLM-Cd: GLM model predicting M1 TR_d activity from cerebellum TR_i’s; GLM-Ci: GLM model predicting M1 TR_i activity from cerebellum TR_i’s; GLM-CU: GLM model predicting M1 TU activity from cerebellum TR_i’s. (B) Regression was used to identify a cerebellum neural population space that predicted BMI-potent M1 activity. GLMs were fit to predict the M1 BMI task–related direct (TR_d)/task-related indirect (TR_i)/task unrelated (TU) neural state from cerebellum TR_i activity; multiple time lagged copies of each cerebellum TR_i unit were used as predictors. (C) Distribution of regression weight magnitude in one example session for the GLM-Cd model (fitted to neural data binned at 10 ms). Top: For each cerebellum TR_i unit, regression weights were assigned for a variety of time lags. To emphasize the time of the maximum absolute weight of each neuron, values here are normalized to each neuron’s maximum value. Units are sorted according to the time of the largest magnitude weight. Tick marks on the right edge indicate the units shown in (D). Bottom: Histogram of the τ values with the largest magnitude weight for this dataset. Abs, antibodies. (D) Example nonnormalized weights for two cerebellum TR_i neurons from one example session (neural data binned at 10 ms). Height of bars indicates weights, for example, neurons at different time lags (τ) relative to the M1 BMI-potent activity, with negative τ values meaning that cerebellum TR_i leads. (E) Box plot comparing R² values for three different GLM models (fitted to neural data binned at 50 ms), GLM-Cd, GLM-Ci, and GLM-CU; left to right (box plot conventions are same as Fig. 3C). *P < 0.05; n.s., P > 0.05.

Fig. 6.. BMI performance gets impaired with cerebellar cortical inhibition.

(A) Fluorescence image of a coronal brain section showing neurons expressing JAWS (green) in the cerebellar cortex (Simplex and Crus I). (B) Cerebellar cortical inhibition increases time to task completion. (C) PETH of example M1 TR_d units from different sessions, during late trials, with (laser on) and without (laser off) cerebellar inhibition. (D) Box plot showing change in modulation depth (MD_Δ) of M1 TR_d units from early to late trials, with and without cerebellar inhibition. Box plot conventions are the same as Fig. 3C. (E) Same as (C) for M1 TR_i units. (F) Same as (D) for M1 TR_i units. (G) Spectrograms of an example M1 LFP channel showing an absence of 3- to 6-Hz power during cerebellar inhibition (right) in late trials. The 3- to 6-Hz power emerges during late trials in the same day session where cerebellar cortical inhibition was suspended. (H) Three- to 6-Hz power emerges during late trials on the same day session where cerebellar inhibition was not done. (I) Same as (A), showing neurons expressing JAWS (green) in the DCN. (J) DCN inhibition increases time to task completion. (K) Same as (C) for DCN inhibition. (L) Same as (D) for DCN inhibition. (M) Same as (C) for M1 TR_i units. (N) Same as (D) for M1 TR_i units. (O) Same as (G) for DCN inhibition. (P) Same as (H) for DCN inhibition. ***P < 0.001 and *P < 0.05.