Abstract
While ipsilesional cortical electroencephalography has been associated with poststroke recovery mechanisms and outcomes, the role of the cerebellum and its interaction with the ipsilesional cortex is still largely unknown. We have previously shown that poststroke motor control relies on increased corticocerebellar coherence (CCC) in the low beta band to maintain motor task accuracy and to compensate for decreased excitability of the ipsilesional cortex. We now extend our work to investigate corticocerebellar network changes associated with chronic stimulation of the dentato-thalamo-cortical pathway aimed at promoting poststroke motor rehabilitation. We investigated the excitability of the ipsilesional cortex, the dentate (DN), and their interaction as a function of treatment outcome measures. Relative to baseline, 10 human participants (two women) at the end of 4–8 months of DN deep brain stimulation (DBS) showed (1) significantly improved motor control indexed by computerized motor tasks; (2) significant increase in ipsilesional premotor cortex event-related desynchronization that correlated with improvements in motor function; and (3) significant decrease in CCC, including causal interactions between the DN and ipsilesional cortex, which also correlated with motor function improvements. Furthermore, we show that the functional state of the DN in the poststroke state and its connectivity with the ipsilesional cortex were predictive of motor outcomes associated with DN-DBS. The findings suggest that as participants recovered, the ipsilesional cortex became more involved in motor control, with less demand on the cerebellum to support task planning and execution. Our data provide unique mechanistic insights into the functional state of corticocerebellar-cortical network after stroke and its modulation by DN-DBS.
Significance Statement
The study aims to understand the brain mechanisms underlying the effects of cerebellar dentate deep brain stimulation (DN-DBS), a novel upcoming therapy for chronic stroke. We provide evidence that functional improvements as a result of DN-DBS therapy were accompanied by significant improvements in task behavior and ipsilesional cortex excitability. More importantly, we show that the role of the cerebellum in motor control, which increases in the poststroke state, decreased as a function of DN-DBS–mediated ipsilesional cortex facilitation.
Introduction
Stroke is a leading cause of serious and long-term disability in the industrialized world (Virani et al., 2021), emphasizing the need for novel therapies to promote poststroke rehabilitation. Despite great progress in acute interventions (Prabhakaran et al., 2015; Mendelson and Prabhakaran, 2021), there remains an enormous need for technologies aimed at enhancing chronic poststroke rehabilitation. Our group has recently developed and translated a novel approach for poststroke motor recovery (Wathen et al., 2018) that is based on DBS of the dentato-thalamo-cortical (DTC) pathway, targeting its origin at the cerebellar DN. Our recent first-in-human, Phase I study (Baker et al., 2023) provided a unique opportunity to examine the physiology of the DN and corticocerebellar-cortical connectivity during motor recovery. Specifically, we examined the movement-related electrophysiology of DN and corticocerebellar-cortical networks, before and after combined DN-DBS therapy and rehabilitation, in chronic poststroke survivors using scalp EEG and invasive DN local field potential (LFP) recordings.
EEG has been extensively used to characterize poststroke recovery mechanisms and predict treatment outcomes to novel interventions (Keser et al., 2022). Recovery of motor function has been related to an increase in movement-related desynchronization power of alpha–beta band over the ipsilesional cortex (Platz et al., 2002; Kaiser et al., 2012; Tangwiriyasakul et al., 2014; Ray et al., 2020). In addition to event-related desynchronization (ERD) power, increased resting state connectivity within both ipsilesional (Wu et al., 2015; Cassidy et al., 2021) and contralesional cortex (Sun et al., 2021) in the alpha–beta band has been associated with better treatment outcomes. Despite a plethora of EEG-based cortical markers of stroke motor recovery, the role of the cerebellum, a known contributor to cortical function in both health (Manto et al., 2012) and disease (Machado and Baker, 2012), and its interaction with the ipsilesional cortex during motor recovery and rehabilitation is still largely unknown.
The motor cortex and the contralateral cerebellum are tightly interconnected through the DTC and cortico-ponto-cerebellar (CPC) pathways (Palesi et al., 2017). This reciprocal connectivity is critical for predictive timing and control during motor tasks (Soteropoulos and Baker, 2006; Manto et al., 2012). Stroke leads to disruption of the CPC tract and reduced metabolism in the contralesional cerebellum. This phenomenon, termed as crossed cerebellar diaschisis (Pantano et al., 1986; Machado and Baker, 2012), is a major impediment to motor rehabilitative efforts as the hypoactive cerebellum provides less excitatory support to the ipsilesional cortex (Takasawa et al., 2002; Kim et al., 2019). Examining the cerebellum and its connectivity dynamics with the ipsilesional cortex is key to understanding recovery mechanisms and will aid the development of novel interventions.
We have previously characterized poststroke sensorimotor oscillatory power and connectivity between the ipsilesional cortex and cerebellum using EEG and DN LFPs (Gopalakrishnan et al., 2022). We found that the ipsilesional cortex and contralateral DN were strongly coupled in the low beta band during both phasic movements and steady-state isometric contractions. This coupling was critical for maintaining motor task accuracy and points to the relevance of low beta band corticocerebellar coherence (CCC) in poststroke motor control. We now report, for the first time, changes in oscillatory activity between the ipsilesional cortex and contralateral DN in response to DN-DBS combined with physical rehabilitation. We show that the role of the cerebellum decreases with successful rehabilitation, likely because errors between planned and executed movement diminish with improvements in motor impairment and function.
Materials and Methods
Participant demographics, stroke extent, and data collection procedures have been previously published in detail (Gopalakrishnan et al., 2022) and summarized in Table 1. In summary, data were derived from a single-site, Phase I clinical trial (Clinicaltrials.gov NCT02835443) investigating the safety and feasibility of DN-DBS in chronic, poststroke motor rehabilitation. All research participants provided written informed consent. The study was monitored by the Cleveland Clinic Internal Review Board and the U.S. Food and Drug Administration under an Investigational Device Exemption. All participants enrolled in the trial underwent surgical implantation of a DBS lead in the DN contralateral to the affected cerebral hemisphere. The study was conducted using the Vercise DBS System (Boston Scientific).
Participant demographics and clinical characteristics
Clinical trial
The clinical trial design, detailed study protocols, inclusion/exclusion criteria, and outcomes of the trial have been published recently in greater detail (Baker et al., 2023). In summary, an open-label, nonrandomized, single-arm design was used. Participation for each participant spanned 20–24 months. Participants underwent physical therapy (PT) for 1 month followed by implantation of an eight-channel DBS lead in the DN contralateral to the ipsilesional cortex. Following a 1 month, postoperative recovery period, participants resumed PT for an additional 2 months to establish baseline function prior to DN-DBS therapy. Participants then entered the programming phase (4–10 weeks combined with PT) to determine the optimal DBS parameters. This was followed by a combined DBS and rehabilitation phase, during which DBS was delivered continuously for 4–8 months while PT continued. Subsequently, participants entered the rehabilitation carryover phase to wean off DBS over 1 month with continued PT followed by surgical explant of the DBS hardware. PT administered by a licensed therapist involved both adaptive tasks, where the difficulty of the task was segmentally graded based on individual performance, and repetitive tasks, where a functional task was repeated continuously to reinforce successful performance. Each session lasted 1–1.5 h, twice weekly and was supplemented with a targeted home exercise program of 3–5 exercises on other days.
Data collection
All electrophysiological data collection (refer to Gopalakrishnan et al., 2022 for more details) occurred during a 5–7 d period immediately following surgical implantation (implant phase) and again immediately preceding surgical removal of the DBS lead (explant phase). The average time elapsed between implant and explant phase was 14 ± 3 months. LFPs from the DN, 36-channel scalp EEG, and electromyography (EMG) from the flexor carpi radialis were acquired using the BrainAmp DC with BrainVision Recorder software (Brain Products). Force exerted during the grip motor task (described below) was recorded from a strain gauge-based isometric hand dynamometer (Model HD-BTA, Vernier). All electrophysiological data were collected at 5,000 samples/s with an online low pass of 1 kHz and high-pass filter of 0.1 Hz.
Task
All participants performed a visuomotor block tracking task (Gopalakrishnan et al., 2022) during both implant and explant phase using the paretic hand. Force applied to the dynamometer controlled the vertical displacement of a 2-D ball on an LCD display as the participant tracked a scrolling, square wave trace as accurately as possible. The force profile displayed four distinct phases in each trial (Fig. 1): (1) onset, (2) hold, (3) offset, and (4) relaxation. The “onset” phase is a phasic period where the participants applied force to the dynamometer to reach 30% of their maximal voluntary contraction (MVC). The “hold” phase was defined as the steady-state contraction period following “onset”. The “offset” period is the phasic period following “hold” when the participants withdrew force from the dynamometer to bring the 2-D ball back to the baseline level. The “relaxation” phase was defined as the period following movement “offset” that showed a minimal force production across participants. The participants either tracked a square wave with a “hold” period of 5 or 2.5 s plateau, depending on their ability, but in general they performed three trials of tracking that lasted 100 s each, with a 3–5 min rest period between each trial. The square wave duration was matched between implant and explant phase.
Comparison of motor skill between implant and explant. A, Exemplar force profiles from a moderately impaired (MI) participant (left) and severely impaired (SI) participant (right) that performed a 5 s block tracking from implant and explant phases. Shaded rectangular area highlights the variance in force during implant compared with explant during the isometric hold phase. The SI participant shown here is the only one that performed a 5 s block tracking task, while others performed a 2.5 s block tracking task. The different phases of the force profiles are shown on the bottom. On, onset; H, hold; Off, offset; and R, relaxation. B, AI comparison between implant and explant phase across all participants (ALL), moderately impaired (MI), and severely impaired (SI). C, RMSSD comparison between implant and explant phase across all participants (ALL), moderately impaired (MI), and severely impaired (SI). ††p < 0.0001.
Motor impairment and function
Motor impairment and function of the upper limb was assessed at implant and explant surgery by an occupational therapist. The Upper Extremity Fugl-Meyer (UEFM) is a common, reliable, and valid instrument for assessment of upper limb motor impairment following stroke (Fugl-Meyer et al., 1975). We used UEFM total score [i.e., the sum of upper extremity, wrist, and hand functions (parts A–D; maximum score = 66)] and hand score [i.e., total score across wrist and hand function, including coordination and speed (parts B–D; maximum score = 30)]. Lower scores indicate higher impairment. Motor function of the upper limb was assessed using Arm Motor Ability Test (AMAT) that has two subscales that measure Functional Ability (AMAT-FA) and Quality of Movement (AMAT-QOM) based on 28 different assessments scored on a 0–5 integer basis. We computed an average of all the different assessments to create a total score for each of the subscales. Higher score indicates higher function.
Motor skill
Task performance was quantified during the hold phases of the visuomotor block tracking motor control task. During the hold phase, we measured force variability using the root mean square of the successive difference (RMSSD) as well as performance of the hand maintaining the designated 30% MVC using accuracy index (AI) given by the following:
EEG and LFP data preprocessing
All data analysis was performed using the Matlab-based FieldTrip toolbox (Oostenveld et al., 2011) and custom-written scripts. In summary, EEG data were (1) visually inspected to eliminate temporal channels with artifacts arising from compensatory movements using a manual artifact rejection method in FieldTrip; (2) parsed into trials with interval −3.5 to 8.5 s relative to movement onset (for 5 s block tracking) or −2.25 to 4.75 s relative to movement onset (for 2.5 s block tracking); (3) filtered using a discrete Fourier transform filter or spectral interpolation method to remove line noise; (4) downsampled to 400 samples/s; (5) cleaned of muscle, EKG, and ocular artifacts in EEG using an independent component analysis; (6) transformed to scalp current densities using a surface Laplacian approach; and (7) left–right flipped so that data from all the participants were aligned to a common (right) side reflecting the affected hemisphere. LFP data were subjected to steps 2, 3, and 4, since the other steps were applicable to EEG only. The ipsilesional cortex was defined by seven electrodes: F4, FC2, FC6, C4, CP2, CP6, and P4. To estimate activity in the DN LFP, a bipolar derivative of the contacts of the DBS lead was performed. Participants with an eight-contact linear lead had seven bipolar channels (viz., LFP1-2, LFP2-3, LFP3-4, LFP4-5, LFP5-6, LFP6-7, and LFP7-8). In participants with eight-contact directional leads, the two levels of three radially arranged directional contacts were pooled to mimic an omnidirectional contact, before computing a bipolar derivative resulting in three bipolar channels (viz., LFP1-234, LFP234-567, and LFP567-8).
EEG and LFP data analysis
To investigate the excitability of the ipsilesional cortex and DN, we performed an event-related desynchronization/synchronization (ERD/S) analysis (Pfurtscheller and Lopes da Silva, 1999) during phasic movement onset and offset periods. To investigate the interaction between the ipsilesional cortex and DN, also referred to here as CCC, we performed a coherence and Granger causality (GC) analysis (Bastos and Schoffelen, 2015) during steady-state hold and relaxation periods. Our earlier work (Gopalakrishnan et al., 2022) identified the low beta (13–20 Hz) as the primary frequency band that characterizes both excitability and communication between the ipsilesional cortex and DN during the implant phase. Though our focus is on the low beta band, quantitative analysis for theta (5–7 Hz), alpha (8–12 Hz), and high beta (21–30 Hz) are also presented.
ERD/S analysis
Artifact-free, preprocessed EEG and LFP time series from each trial were decomposed using a 7-cycle Morlet wavelet into their time-frequency representations in the 5–45 Hz range with frequency steps of 1 Hz. The raw power spectra were then log transformed, and event-related power change in the low beta band was calculated by averaging the normalized power of all time points in each frequency bin relative to the average power of that frequency across the whole experimental session. ERD/S power at movement onset and offset was defined as the normalized average relative power over a 1 s window centered at movement onset and offset. Power less than zero are referred to as ERD, while power greater than zero are referred to as ERS.
CCC/GC analysis
CCC (5–45 Hz in steps of 0.5 Hz, multitaper method with spectral smoothing set to 3) was computed between all DN LFP channels and scalp EEG electrodes (Rosenberg et al., 1998) after partializing for the contribution of the EMG. The CCC were further Z-transformed (Maris and Oostenveld, 2007) to correct for the variable number of degrees of freedom (number of trials) across participants. In each participant, a nonparametric cluster-based permutation approach (Gopalakrishnan et al., 2022) was applied to the implant phase data to identify one DN LFP channel that exhibited the largest coherence cluster in the low beta band on the ipsilesional cortex during the “hold” and “relaxation” phase combined. As a result, in each participant, we identified one DN LFP channel and three neighboring EEG channels that clustered over the ipsilesional cortex with largest coherence. At explant phase, we directly compared the changes in CCC after DN-DBS therapy using the same LFP–EEG channel pairs. To further investigate causal interactions, we performed a nonparametric spectral GC analysis (Dhamala et al., 2008) between the DN and ipsilesional cortex using the above identified channels.
Statistical analysis using generalized linear mixed effects models
ERD/S, CCC, GC, AI, and RMSSD were compared between implant and explant for the affected extremity using generalized linear mixed effects models (GLMMs; Fromer et al., 2018). GLMMs are flexible extensions of linear regressions which can accommodate repeated measurements among participants, non-Gaussian response variables, and unbalanced data. In these models, high-dimensional data do not have to be averaged to a single data point per participant. Instead, all data for all participants are retained, and the clustering inherent in the data is accounted for in the model (e.g., ERD/S clustered within three neighboring EEG electrodes, clustered within block of testing, clustered within participant). Each model for ERD/S, CCC, and GC was fit using random effects for participant, electrode, and block to account for the multilevel clustered structure of the data. AI and RMSSD models included participant as a random effect. Time (implant vs explant) was a fixed effect in all models. A compound symmetry covariance matrix was applied. ERD/S and CCC were multivariate normal, while GC, AI, and RMSSD metrics were lognormal. Fit was evaluated using Bayes information criterion and visual examination of residuals. GLMMs were fit using SAS Studio v. 3.81 with significance indicated by p < 0.05 for the model parameter for time (implant vs explant). Models were built for the total sample and for the sample stratified by impairment group. Each model parameter for time (β), with its standard error (SE), is presented in the Results. This number represents the adjusted mean difference or adjusted log mean difference between implant and explant, when accounting for all sources of clustering in the data.
Correlation analysis
To determine whether electrophysiological measures from the ipsilesional cortex and DN can predict motor recovery and outcome, we performed a regression analysis to correlate change in the clinical outcome measures of motor impairment (ΔUEFM and ΔUEFM hand score), motor function (ΔAMAT-FA and ΔAMAT-QOM), and motor skill (ΔAI and ΔRMSSD) with the following:
Change/progression in ERD/S, CCC, and GC (denoted as ΔERD/S, ΔCCC, and ΔGC) from implant to explant phase in theta, alpha, low beta, and high beta bands.
Baseline value of ERD/S, CCC, and GC (denoted as ▾ERD/S, ▾CCC, and ▾GC) at implant phase in theta, alpha, low beta, and high beta bands.
A linear regression model was applied to the data comparing the electrophysiological measure (independent variable) and motor functional/behavioral measure (dependent variable) to determine the gradient slope. Null hypothesis was defined as zero gradient (i.e., no linear relationship). An empirical null distribution was formed by shuffling the order of the dependent variable, while retaining the order of the independent variable. The actual gradient was then compared with this null distribution generated from 2,000 permutations, and a p value measured by integrating the null distribution between the actual gradient and infinity divided by the total integral. Significance level was set at α = 0.05 (two-tailed). Time selection: For ΔERD/S and ▾ERD/S, average activity during onset and offset periods were used. For ΔCCC/GC and ▾CCC/GC, average activity during hold and relaxation periods were used. Channel selection: In the case of ΔERD/S of the ipsilesional cortex, a linear regression was fit for all seven channels on the ipsilesional cortex to identify the spatial extent of correlation. Hence, α = 0.05 was further divided by 7 to account for multiple comparisons, which meant that a p < 0.0036 in any one channel would indicate statistical significance. In the case of ▾ERD/S of the ipsilesional cortex, the fit was performed only for channel “CP2” because our prior work (Gopalakrishnan et al., 2022) showed that ERD/S from this channel was correlated with DN ERD/S and task accuracy. Hence, we wanted to test the hypothesis if ▾ERD/S from CP2 would be predictive of outcome. In the case of ΔERD/S and ▾ERD/S of the DN, a linear regression was fit for average activity of all DN channels. In case of ΔCCC/GC and ▾CCC/GC, fit was performed for average CCC computed between the chosen LFP channel that exhibited largest coherence cluster on the ipsilesional cortex and corresponding EEG channels on the ipsilesional cortex.
Data availability
Patient data with identifiers cannot be made available to the public. Deidentified data can be obtained from the corresponding author upon request and after review to ensure compliance with sponsor and institutional guidelines.
Results
Demographics
A total of 15 participants were screened for eligibility, and only 12 participants were enrolled in the clinical trial. However, data presented in this study were not acquired from the first two participants as the investigators were beginning to establish the safety and feasibility of DN-DBS and did not externalize the leads. The dataset included 10 participants with severe or moderate-to-severe upper extremity hemiparesis after middle cerebral artery ischemic stroke, and their demographics have been previously published (Gopalakrishnan et al., 2022). The mean age was 57.3 ± 7.21 years (48–70 years of age). Two participants were female, and six participants had stroke on the right hemisphere affecting their left extremity. Participant demographics and clinical characteristics are provided in Table 1.
Motor impairment and function
Both UEFM total and UEFM hand score showed significant improvement at the explant phase compared with those at the implant phase. The median (Q1, Q3) UEFM total score of the participant sample at implant and explant were 23.50 (19, 26) and 41 (23, 48; p = 0.0020 using Wilcoxon signed rank test), respectively. The median UEFM hand score of the participant sample at implant and explant were 7 (4, 10) and 16.5 (6, 20; p = 0.0059 using Wilcoxon signed rank test), respectively.
Both the AMAT-FA and AMAT-QOM scores showed significant improvement at the explant phase compared with those at the implant phase. The median (Q1, Q3) AMAT-FA of the participant sample at implant and explant were 2.767 (1.678, 3.142) and 4.303 (2.250, 4.464; p = 0.0020 using Wilcoxon signed rank test), respectively. The median AMAT-QOM score of the participant sample at implant and explant were 2.821 (1.571, 3.035) and 3.982 (2.142, 4.214; p = 0.0020 using Wilcoxon signed rank test), respectively.
Participant stratification
Four of the 10 participants had an AMAT-FA and AMAT-QOM score <2 at implant and hence were categorized as severely impaired (SI), while the remaining participants were categorized as moderately impaired (MI). The same four participants also had a UEFM hand score <6 at implant. Whenever significant differences were noted in the ERD/S, CCC, and GC analysis at a group level, we stratified the data into SI and MI subgroups to evaluate the contribution of impairment level on these metrics.
The median (Q1, Q3) UEFM total score of SI participant sample at implant and explant were 18.5 (15.5, 19.5) and 22 (18, 23.5; no significant difference using Wilcoxon signed rank test), respectively. The median UEFM hand score of the SI participant sample at implant and explant were 4 (3, 4.5) and 5 (3.5, 6; no significant difference using Wilcoxon signed rank test), respectively. The median UEFM total score of the MI participant sample at implant and explant were 25.5 (24, 28) and 46.5 (44, 48; p = 0.0312; Wilcoxon signed rank test), respectively. The median UEFM hand score of the MI participant sample at implant and explant were 9.5 (7, 12) and 19.5 (17, 20; p = 0.0312; Wilcoxon signed rank test), respectively.
The median AMAT-FA score of SI participant sample at implant and explant were 1.660 (1.375, 1.767) and 2.142 (1.910, 2.392; no significant difference using Wilcoxon signed rank test), respectively. The median AMAT-QOM score of SI participant sample at implant and explant were 1.482 (1.250, 1.714) and 1.964 (1.660, 2.339; no significant difference using Wilcoxon signed rank test), respectively. The median AMAT-FA score of MI participant sample at implant and explant were 3.035 (2.821, 3.678) and 4.410 (4.321, 4.607; p = 0.0312; Wilcoxon signed rank test), respectively. The median AMAT-QOM score of MI participant sample at implant and explant were 3.017 (2.892, 3.607) and 4.196 (4, 4.25; p = 0.0312, Wilcoxon signed rank test), respectively.
Task
All 10 participants completed the block tracking task successfully. At implant, four participants (3 SI) completed 2.5 s block tracking, while six participants (5 MI) completed 5 s block tracking. The AMAT-FA score of participants that performed 2.5 s block tracking was significantly lower (p = 0.033; Wilcoxon rank sum test) than participants that performed 5 s block tracking, corroborating task preference was influenced by motor function. To maintain consistency, at explant, participants performed block tracking with the same duration that they performed at implant. At implant, participants who completed 5 s block tracking tracked an average 63 ± 18 blocks, while participants who completed 2.5 s block tracking tracked an average 97 ± 49 blocks. Participants always tracked a greater number of blocks during the explant phase. We subsampled the first n blocks from explant phase in order to match the number of blocks tracked in the implant phase.
Motor skill
At a group level, AI (measure of accuracy of task; Fig. 1A) was significantly higher at explant, compared with implant for the full sample (β = 0.0242; SE = 0.0028; p < 0.0001), for MI participants (β = 0.0262; SE = 0.0028; p < 0.0001) and for SI participants (β = 0.0228; SE = 0.0057; p < 0.0001). RMSSD (measure of force variability; Fig. 1B) was significantly lower at explant, indicating improved force stability, compared with implant for the full sample (β = −0.7828; SE = 0.0322; p < 0.0001), for both MI (β = −1.0821; SE = 0.0443; p < 0.0001) and SI participants (β = −0.2532; SE = 0.0401; p < 0.0001).
EEG ERD/S comparison between implant and explant
Movement onset
ERD/S power in the low beta band (Fig. 2A) was significantly higher (synchronization) in the explant phase compared with implant phase in all participants taken together (β = 0.0066; SE = 0.0033; p = 0.044). When participants were stratified based on impairment, the SI subgroup exhibited ERS and did not show a significant change from implant to explant phase (β = 0.0062; SE = 0.0072; p = 0.390), whereas the MI subgroup showed an increase (β = 0.0068; SE = 0.0033; p = 0.039), while still exhibiting ERD. ERD/S measures in other frequency bands did not show a significant difference between implant and explant phase at a group level (Fig. 2B).
EEG ERD/S comparison between implant and explant phase during movement onset and offset. A, Top, Time stratified by impairment shows average ERD/S power in the low beta band across all participants (ALL), moderately impaired (MI), and severely impaired participants (SI). Middle, Topoplot shows the spatial distribution of ERD/S power over the ipsilesional cortex in MI participants. Bottom, Average time-frequency plot derived from ipsilesional channels shows the temporal and spectral extent of ipsilesional ERD/S in MI participants. Time 0 represents movement onset. B, Time stratified by impairment shows comparison of average ERD/S power in ALL, MI, and SI between implant and explant phase in theta (top), alpha (middle), and high beta (bottom) frequency bands. Panels C and D are analogous to panels A and B for the movement offset phase. Time 0 represents movement offset. Five out of six MI participants did a 5 s block tracking task, while one participant did 2.5 s block tracking. ††p < 0.0001, †p < 0.001, **p < 0.01, *p < 0.05.
Movement offset
ERD/S power in the low beta band (Fig. 2C) was significantly lower (desynchronization) in the explant phase compared with that in the implant phase in all participants taken together (β = −0.0137; SE = 0.0028; p < 0.0001). When participants were stratified based on impairment, the SI subgroup did not show a significant change from the implant to explant phase (β = −0.0061; SE = 0.0045; p = 0.185), whereas the MI subgroup showed a significant decrease (β = −0.0177; SE = 0.0035; p < 0.0001). ERD/S power in the high beta band was significantly lower (desynchronization) in the explant phase compared with that in the implant phase at a group level and MI subgroup (Fig. 2D).
DN ERD/S comparison between implant and explant
Movement onset
ERD/S power in the low beta band (Fig. 3A) was significantly lower in the explant phase compared with that in the implant phase in all participants taken together (β = −0.0198; SE = 0.0045; p < 0.0001). When participants were stratified based on impairment, both SI (β = −0.0570; SE = 0.0098; p < 0.0001) and MI (β = −0.0065; SE = 0.0031; p = 0.037) showed a significant decrease. However, SI participants still displayed ERS at explant. ERD/S power in the theta, alpha, and high beta bands were significantly lower in the explant phase compared with that in the implant phase at a group level, MI, and SI subgroups (Fig. 3B).
DN ERD/S comparison between implant and explant phase during movement onset and offset. A, Top, Time stratified by impairment shows average ERD/S power in the low beta band across all (ALL), moderately impaired (MI), and severely impaired participants (SI). Bottom, Average time-frequency plot derived from DN contacts show the temporal and spectral extent of DN ERD/S in MI participants. Time 0 represents movement onset. B, Time stratified by impairment shows comparison of average ERD/S power in ALL, MI, and SI between implant and explant phase in theta (top), alpha (middle), and high beta (bottom) frequency bands. Panels C and D are analogous to panels A and B for the movement offset phase. Time 0 represents movement offset. Five out of six MI participants did a 5 s block tracking task, while one participant did 2.5 s block tracking. ††p < 0.0001, †p < 0.001, **p < 0.01, *p < 0.05.
Movement offset
ERD/S power in the low beta band (Fig. 3C) was significantly lower (desynchronization) in the explant phase compared with that in the implant phase in all participants taken together (β = −0.0040; SE = 0.0019; p = 0.035). When participants were stratified based on impairment, SI did not show a change from implant to explant phase (β = −0.0011; SE = 0.0041; p = 0.788), while MI showed a significant decrease (β = −0.0051; SE = 0.0021; p = 0.017). ERD/S power in the alpha band was significantly lower in the explant phase compared with that in the implant phase at a group level and MI subgroup (Fig. 3D).
CCC/GC comparison between implant and explant
Hold phase
CCC in the low beta band (Fig. 4A) was significantly lower in the explant phase compared with that in the implant phase in all participants taken together (β = −0.9483; SE = 0.1791; p < 0.0001) and participants stratified by impairment (SI: β = −0.5324; SE = 0.2035; p = 0.017; MI: β = −1.2256; SE = 0.2561; p < 0.0001). GC in the low beta band directed from the ipsilesional cortex to DN (Fig. 5A) was significantly lower in the explant phase compared with that in the implant phase in all participants taken together (β = −1.2339; SE = 0.2541; p < 0.0001) and only in MI (β = −1.5105; SE = 0.3220; p < 0.0001) when stratified by impairment. GC in the low beta band directed from the DN to ipsilesional cortex (Fig. 5C) was significantly lower in the explant phase compared with that in the implant phase in all participants taken together (β = −1.5618; SE = 0.2636; p < 0.0001) or stratified (SI: β = −1.5790; SE = 0.3711; p = 0.0002; MI: β = −1.5359; SE = 0.3641; p = 0.0005). CCC and GC (both directions) in the high beta band (Figs. 4B, 5B,D) and GC directed from the DN to ipsilesional cortex in the theta and alpha band (Fig. 5D) were significantly lower in the explant phase compared with those in the implant phase at a group level.
CCC comparison between implant and explant phase during hold and relaxation. A, Left, Grand averaged topoplot from all participants (ALL) in the low beta band showing clusters of significant CCC over the ipsilesional cortex in the implant phase and its reduction during explant phase. The topoplots are thresholded at a coherence value of 1.6. Middle, Grand averaged line plot (mean and SEM) of CCC across all participants. Solid lines indicate the CCC between the DN and ipsilesional cortex, while dotted lines show the CCC between DN and a remote location (F3, AF3, FP1 highlighted in red dotes over the topoplot) on the cortex that is not related to motor activity. Shaded box indicate the frequencies over which CCC was significantly different between implant and explant phases. Right, Time by impairment interaction of CCC across ALL, moderately impaired (MI), and severely impaired participants (SI). B, Time stratified by impairment shows comparison of CCC in ALL, MI, and SI between implant and explant phase in theta (left), alpha (middle), and high beta (right) frequency bands. Panels C and D are analogous to panels A and B for the relaxation phase. ††p < 0.0001, †p < 0.001, **p < 0.01, *p < 0.05.
GC comparison between implant and explant phase during hold. A, Top, Grand averaged line plot (mean and SEM) of GC (directed from the ipsilesional cortex to DN) across all participants. Solid lines indicate the GC between the DN and ipsilesional cortex, while dotted lines show the GC between DN and a remote location (F3, AF3, FP1 highlighted in red dotes in Fig. 4A topoplot) on the cortex that is not related to motor activity. Shaded box indicate the frequencies over which GC was significantly different between implant and explant. Bottom, Time by impairment of GC (directed from the ipsilesional cortex to DN) in the low beta band across all participants (ALL), moderately impaired (MI), and severely impaired (SI). B, Time stratified by impairment shows comparison of GC (directed from the ipsilesional cortex to DN) in ALL, MI, and SI between implant and explant phase in theta (top), alpha (middle), and high beta (bottom) frequency bands. Panels C and D are analogous to panels A and B for GC (directed from the DN to ipsilesional cortex). ††p < 0.0001, †p < 0.001, **p < 0.01, *p < 0.05.
Relaxation phase
CCC in the low beta band (Fig. 4C) was significantly lower in the explant phase compared with that in the implant phase in all participants taken together (β = −0.4793; SE = 0.2242; p < 0.0375). When participants were stratified based on impairment, the SI subgroup did not show a change from implant to explant phase, while the MI subgroup showed a significant decrease (β = −0.9678; SE = 0.3217; p < 0.0054). It is important to note that the SI group did not show substantial CCC in the implant phase, suggesting lower levels of compensatory mechanisms in this poststroke subpopulation, as discussed below. GC in the low beta band directed from the ipsilesional cortex to DN (Fig. 6A) was significantly lower in the explant phase compared with that in the implant phase in all participants taken together (β = −0.5672; SE = 0.2569; p = 0.0320) and only in MI (β = −0.8442; SE = 0.3167; p = 0.0124) when stratified based on impairment status. GC in the low beta band directed from the DN to ipsilesional cortex (Fig. 6C) was significantly lower in the explant phase compared with that in the implant phase in all participants taken together (β = −0.7015; SE = 0.2044; p = 0.0012) and only in SI (β = −0.8895; SE = 0.2374; p = 0.0014) when stratified based on impairment status. CCC and GC (both directions) in the high beta band (Figs. 4D, 6B,D) and CCC and GC (directed from the DN to ipsilesional cortex) in the alpha band (Figs. 4D, 6D) were significantly lower in the explant phase compared with those in the implant phase at a group level.
GC comparison between implant and explant phase during relaxation. A, Top, Grand averaged line plot (mean and SEM) of GC (directed from the ipsilesional cortex to DN) across all participants. Solid lines indicate the GC between the DN and ipsilesional cortex, while dotted lines show the GC between DN and a remote location (F3, AF3, FP1 highlighted in red dotes in Fig. 4A topoplot) on the cortex that is not related to motor activity. Shaded box indicate the frequencies over which GC was significantly different between implant and explant. Bottom, Time by impairment of GC (directed from the ipsilesional cortex to DN) in the low beta band across all participants (ALL), moderately impaired (MI), and severely impaired (SI). B, Time stratified by impairment shows comparison of GC (directed from the ipsilesional cortex to DN) in ALL, MI, and SI between implant and explant phase in theta (top), alpha (middle), and high beta (bottom) frequency bands. Panels C and D are analogous to panels A and B for GC (directed from the DN to ipsilesional cortex). ††p < 0.0001, †p < 0.001, **p < 0.01, *p < 0.05.
Correlation analysis
Baseline electrophysiology as a predictor of motor function, impairment, and skill
Low beta band ▾ERD/S of ipsilesional CP2 during movement onset showed a significant negative correlation with ΔUEFM (p = 0.01) and ΔUEFM hand (p = 0.02) outcome measure. That is, participants that showed greater ERD in the CP2 at baseline showed greater improvements of upper extremity motor impairment during the clinical trial (Fig. 7A). Similar results were observed for alpha and high beta bands in the ipsilesional CP2 (Fig. 7A). Low beta band ▾ERD/S of DN during movement onset showed a significant negative correlation with ΔUEFM (p = 0.015) and ΔUEFM hand (p = 0.02) outcome measure. That is, participants that showed greater ERD in the DN at baseline also had greater impairment improvements during the clinical trial (Fig. 7B). Similar results were observed for theta, alpha, and high beta bands in the DN (Fig. 7B,C). In addition, low beta band ▾CCC during relaxation phase showed a significant positive correlation with ΔAMAT-FA (p = 0.0005), ΔAMAT-QOM (p = 0.002), ΔUEFM (p = 0.02), and ΔUEFM hand (p = 0.02) outcome measures (Fig. 8A). That is, participants that showed greater CCC at baseline also had greater increase in functional gains and impairment improvements. Similar results were observed for alpha and high beta band ▾CCC (Fig. 8B). No significant correlations were found with motor skill (ΔAI and ΔRMSSD).
Baseline ERD/S as a predictor of motor function and impairment. A, Linear model showing relationship between ▾ERD/S in the alpha, low beta, and high beta bands in the ipsilesional CP2 during movement onset and ΔUEFM/ΔUEFM hand score. Significant negative correlation was observed. B, Linear model showing relationship between ▾ERD/S in the theta, alpha, low beta, and high beta band in the DN during movement onset and ΔUEFM/ΔUEFM hand score. Significant negative correlation was observed. C, Linear model showing relationship between ▾ERD/S in the high beta band in the DN during movement offset and ΔUEFM score. Significant negative correlation was observed. p values are indicated in the plots.
Baseline CCC/GC as a predictor of motor function and impairment. A, Linear model showing relationship between ▾CCC in the low beta band during relaxation phase and ΔAMAT-FA/ΔAMAT-QOM/ΔUEFM/ΔUEFM hand score. Significant positive correlation was observed. B, Linear model showing relationship between ▾CCC/▾GC in the alpha, ▾CCC in the high beta band during relaxation phase and ΔAMAT-FA/ΔAMAT-QOM/ΔUEFM/ΔUEFM hand score. Significant positive correlation was observed. p values are indicated in the plots.
Change in electrophysiology between implant and explant in association with motor function, impairment, and skill
Ipsilesional ΔERD/S during movement offset in the low beta band at EEG channels F4 (p = 0.0035 (UEFM), p = 0.003 (UEFM hand), and FC2 (p = 0.003, p = 0.002) showed a significant, negative correlation with the ΔUEFM and ΔUEFM hand outcome measures (Fig. 9A). That is, participants that showed the greater ERD in frontal and frontocentral areas at explant compared with implant also showed greater improvements from impairment. This effect was not observed in any other frequency band. Likewise, ΔCCC during relaxation phase in the low beta band was significantly negatively correlated with ΔAMAT-FA (p = 0.001), ΔAMAT-QOM (p = 0.006), ΔUEFM (p = 0.014), and ΔUEFM hand (p = 0.001) outcome measures (Fig. 9B). That is, participants that showed greater reduction in the CCC also displayed greater functional gains and improvements from impairment. ΔCCC during relaxation phase in the alpha band was significantly negatively correlated with ΔAI (Fig. 9C). That is, participants that showed greater reduction in the alpha CCC also displayed greater improvement in task accuracy. Lastly, ΔGC (ipsilesional cortex to DN) during relaxation phase in the alpha band was significantly negatively correlated with ΔAMAT-FA and ΔAMAT-QOM (Fig. 9C). That is, participants that showed greater reduction in the GC also displayed greater functional gains.
Change in electrophysiology between implant and explant in association with motor function and impairment. A, Linear model showing relationship between ΔERD/S in the low beta band over the ipsilesional cortex during movement offset and ΔUEFM/ΔUEFM hand score. Topoplots display the gradient (slope) observed over the ipsilesional cortex when ΔERD/S in each channel was fit against ΔUEFM and ΔUEFM hand score. Significant negative correlation was observed over the frontocentral channels. The regression plot shows the correlation between ΔERD/S in the FC2 channel and ΔUEFM/ΔUEFM hand score. B, Linear model showing relationship between ΔCCC (normalized to implant) in the low beta band during relaxation and ΔUEFM/ΔUEFM hand score. Significant negative correlation was observed. C, Linear model showing relationship between alpha band ΔCCC (normalized to implant)/ΔGC during relaxation and ΔAI/ΔAMAT-FA/ΔAMAT-QOM. p values are indicated in the plots.
Discussion
DN-DBS is an emerging application for poststroke rehabilitation, pioneered by our group (Machado and Baker, 2012). Results from our recent first-in-human clinical trial were promising, with robust improvements in impairment and motor function, particularly among participants with moderate-to-severe impairment who enrolled with at least minimal residual distal upper extremity motor function (Baker et al., 2023). Here, we investigated how corticocerebellar electrophysiological activity predicted response to treatment and changed with DN-DBS therapy combined with rehabilitation. We studied two distinct phenomena using a repetitive block tracking task as follows: (1) ERD/S in the ipsilesional cortex and contralesional DN that signifies the excitability of these areas during phasic movement execution and (2) CCC/GC that signifies the connectivity between the ipsilesional cortex and contralesional DN during static motor control. While ERD/S looks at the oscillatory power within those areas, CCC measures the phase synchronization of the oscillations between those areas. Our data provide comprehensive evidence that supports the presence of poststroke functional connectivity between the contralesional cerebellar DN and the ipsilesional cortex. Furthermore, it demonstrates the modulation of this connectivity by the rehabilitative effects of DN-DBS.
Our previous studies showed the effects of DN-DBS combined with rehabilitation on motor outcomes indexed by the UEFM and AMAT (Baker et al., 2023) and characterized the poststroke, pretreatment, and cortico-cerebello-cortical electrophysiology (Gopalakrishnan et al., 2022). Specifically, we showed that CCC in the low beta band was positively correlated with task accuracy, which we attributed to compensatory activity. In the present work, we extend our findings related to DN-DBS treatment effects on motor function, showing significant gains in motor skill and control, indexed by AI and RMSSD. We also demonstrate the effects of DN-DBS on cortico-cerebello-cortical physiology, showing significant gains in ipsilesional cortical ERD at movement offset (i.e., finger extension) that correlated with the extent of improvements in motor impairment and function. This effect was particularly noteworthy in the premotor areas and driven by MI participants, which showed relatively strong ERD and CCC at the implant (baseline) phase. Furthermore, we found significant decrease in CCC during hold and relaxation phases, including significant decrease in causal interactions between the DN and ipsilesional cortex, and that these decrements in the low beta band also correlated with the extent of improvements in impairment and motor function achieved during combined DN-DBS and rehabilitation treatment. Altogether, our findings support the increased role of the ipsilesional cortex and decreased role of the DN in association with improved motor outcomes, corroborating a lower demand on the DN to support motor task planning and execution.
ERD/S during phasic movement initiation and execution
In the healthy condition, movement-related ERD, particularly in the beta band, is associated with increased corticospinal and cortical excitability (Pfurtscheller and Lopes da Silva, 1999; Takemi et al., 2013). After stroke, movement-related ERD in the ipsilesional cortex has been found to be significantly reduced, and this reduction has been associated with impairment level (Rossiter et al., 2014). Likewise, increments in the sensorimotor ERD have been shown to parallel recovery (Ray et al., 2020). Our baseline data are concordant, with SI participants showing decreased corticospinal excitability evidenced by ERS (Figs. 2A, 3A). Here, we have shown an increase in ipsilesional cortical and contralesional DN ERD after combined DN-DBS and rehabilitation treatment. We also showed that the increased ERD significantly correlated with improvements in motor impairment and function. Importantly, our task required active finger extension at offset to release the grip, a function that is commonly impaired after stroke and an important index for functional recovery during rehabilitation (Smania et al., 2007; Lang et al., 2009).
CCC/GC during isometric contraction and relaxation
While ERD/S quantifies cortical excitability, CCC quantifies phase synchrony, which is pronounced during isometric effort periods and abolished during phasic movement (Kristeva et al., 2007). Previous studies conducted by our group and others (Soteropoulos and Baker, 2006; Gopalakrishnan et al., 2022) have confirmed the presence of CCC in the beta frequency band in the poststroke state, corroborating cerebellar processing during motor learning (Mehrkanoon et al., 2016) and its role in compensatory efforts (Belardinelli et al., 2007). Furthermore, we showed that low beta band CCC is directly correlated with task performance after stroke (Gopalakrishnan et al., 2022), suggesting that cerebello-cortical activity increases in order to achieve motor execution in moderately and severely impaired individuals. Perhaps the most striking findings of the present study are the significant decreases in CCC and GC-indexed bidirectional communication between the cerebellar DN and ipsilesional cortex after treatment that were correlated with improvements in motor impairment and function. Taken together, the results suggest that corticocerebellar information flow increases in the poststroke state, supplementing the ipsilesional cortex to achieve motor control, and decreases with functional improvements mediated by DN-DBS.
ERD/S and CCC as prognostic biomarkers
The prognostic and predictive utility of EEG electrophysiology in stroke has been well studied (Vatinno et al., 2022). Specifically, increments in the alpha and beta band ERD over the central and centroparietal sites have been associated with improved motor outcomes (Rossiter et al., 2014; Ray et al., 2020; Leonardi et al., 2022). This measure at baseline (implant phase) was significantly predictive of treatment response in our cohort and correlated with greater treatment response. In addition to ERD, we provide evidence for the first time that baseline CCC is a key neurophysiological mechanism that predicted functional recovery, perhaps driving the ERD changes in areas that receive cerebellar motor inputs and mediate motor preparation, control, and learning, including the sensorimotor and premotor cortices.
Limitations
The study has several limitations, inherent to the nature of the work conducted during a first-in-human, Phase I clinical trial. The sample size is limited, possibly compensated by the novelty of the data from invasive DN electrophysiology in humans and the robustness of the findings. The sample size was dictated by the power required for an early Phase I clinical trial to assess the safety of the DN-DBS procedure. We also recognize the relative heterogeneity of poststroke motor function across the participants. While this is inevitable, we limited the variability to the extent possible and delineated the dataset by separately presenting SI and MI participants. The motor task itself had to be customized to the needs of the individual participant. The majority of the MI participants performed a 5 s block tracking task, while majority of the SI participants preferred a 2.5 s task. Though the task was dictated by impairment level, we overcame this limitation by choosing specific regions within each trial (onset, hold, offset, and relaxation phases) that were parsed to make comparisons meaningful. Given that the clinical trial was open label, significant efforts were made to reduce the confounds related to the effects of rehabilitation alone on observed motor recovery. Participants were enrolled in the chronic phase of disability after at least 1 year since stroke and completion of standard of care rehabilitative efforts. Furthermore, all participants underwent a total of 3 months of renewed structured rehabilitation as part of the clinical trial before DN-DBS was activated. While these efforts diminish the confounds between DN-DBS and rehabilitation, we will not completely distinguish the independent contributions of DN-DBS and rehabilitation on motor outcomes until the next phase randomized controlled trial is completed.
Conclusions
DN-DBS therapy combined with rehabilitation was associated with significant benefits in motor skill and control that corroborate the previously reported improvements of impairment and motor function. These motor benefits were correlated with significant changes in cortico-cerebello-cortical physiology characterized by increments in ipsilesional cortical ERD and reductions in causal interactions between the DN and ipsilesional cerebral cortex. Taken together, the results indicate that DN-DBS–mediated motor improvements result in reduced demand for poststroke DN involvement in motor control.
Footnotes
This project was funded through National Institutes of Health (NIH) BRAIN initiative award 5UH3NS100543-05.
↵*R.G. and D.A.C. contributed equally to this work.
A.G.M. and K.B.B. are consultants and have intellectual property licensed to Enspire DBS. A.G.M. is the chief medical/scientific officer of Enspire DBS. Enspire DBS funded part of the DN-DBS clinical trial for stroke rehabilitation, but not the present research. This research was funded by the NIH BRAIN initiative.
- Correspondence should be addressed to Andre G. Machado at machada{at}ccf.org.
