The cortical evoked potential corresponds with deep brain stimulation efficacy in rats

Abstract

Deep brain stimulation (DBS) of the subthalamic nucleus (STN) antidromically activates the motor cortex (M1), and this cortical activation appears to play a role in the treatment of hypokinetic motor behaviors (Gradinaru V, Mogri M, Thompson KR, Henderson JM, Deisseroth K. Science 324: 354–359, 2009; Yu C, Cassar IR, Sambangi J, Grill WM. J Neurosci 40: 4323–4334, 2020). The synchronous antidromic activation takes the form of a short-latency cortical evoked potential (cEP) in electrocorticography (ECoG) recordings of M1. We assessed the utility of the cEP as a biomarker for STN DBS in unilateral 6-hydroxydopamine-lesioned female Sprague Dawley rats, with stimulating electrodes implanted in the STN and the ECoG recorded above M1. We quantified the correlations of the cEP magnitude and latency with changes in motor behavior from DBS and compared them to the correlation between motor behaviors and several commonly used spectral-based biomarkers. The cEP features correlated strongly with motor behaviors and were highly consistent across animals, whereas the spectral biomarkers correlated weakly with motor behaviors and were highly variable across animals. The cEP may thus be a useful biomarker for assessing the therapeutic efficacy of DBS parameters, as its features strongly correlate with motor behavior, it is consistent across time and subjects, it can be recorded under anesthesia, and it is simple to quantify with a large signal-to-noise ratio, enabling rapid, real-time evaluation. Additionally, our work provides further evidence that antidromic cortical activation mediates changes in motor behavior from STN DBS and that the dependence of DBS efficacy on stimulation frequency may be related to antidromic spike failure.

NEW & NOTEWORTHY We characterize a new potential biomarker for deep brain stimulation (DBS), the cortical evoked potential (cEP), and demonstrate that it exhibits a robust correlation with motor behaviors as a function of stimulation frequency. The cEP may thus be a useful clinical biomarker for changes in motor behavior. This work also provides insight into the cortical mechanisms of DBS, suggesting that motor behaviors are strongly affected by the rate of antidromic spike failure during DBS.

INTRODUCTION

Deep brain stimulation (DBS) is an effective treatment for advanced Parkinson’s disease (PD) (1, 2) and alleviates bradykinesia, rigidity, and tremor. Despite decades of research and clinical use, there has been little change in clinical DBS parameters and techniques, largely because of an inadequate understanding of the mechanism of action (3). Recent advances, including closed-loop stimulation (4–7), patient-specific parameter optimization (8–10), and nonregular temporal patterns of stimulation (11–13), may improve DBS efficacy, but implementing these approaches requires a reliable biomarker that corresponds to the effects of DBS on motor behaviors (14). Most proposed biomarkers quantify spectral features (14), and here we examine the utility of a new potential biomarker based on cortical activity evoked by DBS.

Evoked compound action potentials (ECAPs) following subthalamic nucleus (STN) DBS can be recorded in the cortex (15–21), within the STN (22–26), and elsewhere in the basal ganglia (27, 28) and are a potential alternative to spectral biomarkers. STN DBS activates axons in the hyperdirect pathway that project from motor cortex (M1) to STN, and this antidromic activation of the cortex ameliorates hypokinetic motor behaviors (29–31). The cortical evoked potential (cEP; Fig. 1A) is a direct indicator of cortical activation, as it represents the aggregate, synchronous firing of the antidromically activated M1 projection neurons (17–20, 32–34). Importantly, the magnitude and latency of the cEP are strongly dependent on stimulation parameters, and higher stimulation frequencies reduce the cEP magnitude and increase its latency (17, 19). We hypothesized that the reduction in cEP magnitude at different stimulation frequencies would correlate with improvements in motor behavior, making it a potential biomarker for DBS efficacy.

Figure 1.

Short-latency cortical evoked potentials (cEPs) at different frequencies of subthalamic nucleus (STN) deep brain stimulation (DBS) in 6-hydroxydopamine (6-OHDA)-lesioned rats. A: the average cEP during 5-min recordings at 7 different stimulation frequencies (13, 25, 50, 75, 100, 130, 200 Hz) from a representative animal with alternating-phase stimulation. Note the reduced magnitude and increased latency of the cEP with increasing stimulation frequency. B: representative example of alternating-phase stimulation at 13 Hz. In red is the averaged anode-first stimulation, in green is the averaged cathode-first stimulation, and in blue is the average cEP from alternating-phase stimulation. Note that the stimulation artifact from 0 to 0.5 ms is largely eliminated, but the cEP from 1 to 2 ms remains. C: representative raw electrocorticography (ECoG) trace during 13-Hz stimulation showing the relative magnitude of the cEP. In red is the stimulation artifact, and in blue is the first 5 ms following each stimulation pulse that contains the short-latency cEP.

We evaluated the utility of the cEP as a biomarker in the unilateral 6-hydroxydopamine (6-OHDA)-lesioned rat model of PD. We quantified the per-animal correlation of different cEP features with improvements in motor behavior from DBS and compared the cEP features to several commonly used spectral biomarkers. The cEP magnitude and latency exhibited strong correlations with DBS efficacy that were highly consistent across animals, whereas the spectral biomarkers exhibited weaker average correlations due to a large variability across animals. The cEP may thus have utility as a biomarker for stimulation parameter selection, as the signal has a strong correlation with DBS efficacy, is robust across time and subjects, has a high signal-to-noise ratio, only requires a surface electrocorticography (ECoG) electrode as opposed to an extrapenetrating electrode within the brain, and uses only the most rudimentary of calculations, thus enabling rapid, real-time quantification.

MATERIALS AND METHODS

To assess the effects of STN DBS on hypokinetic motor behaviors, we used the unilateral 6-OHDA-lesioned rat, which is a validated model of Parkinson’s disease (PD) (35, 36) that shows responses similar to human PD patients during regular (37, 38) and nonregular (39, 40) temporal patterns of stimulation. Thirty female Sprague Dawley rats (250–300 g) were implanted unilaterally with stimulating microelectrodes in the STN and a bone screw above M1 for electrocorticography (ECoG). The rats were rendered hemiparkinsonian via unilateral infusion of 6-OHDA into the medial forebrain bundle (MFB). In one set of experiments, the ECoG was recorded during an awake, freely moving state while stimulation was applied. We used the ECoG to quantify the change in cEP features to different conditions and parameters of stimulation. In the second set of experiments, the ECoG was recorded during electrical stimulation at seven different stimulation frequencies (13, 25, 50, 75, 100, 130, 200 Hz) while the rats performed behavioral tasks to assess hypokinetic motor behaviors. This ECoG signal was then used to calculate the correlation between different electrophysiological biomarkers and improvements in motor behavior across different stimulation frequencies.

Electrode Implantation and 6-OHDA Lesioning

The study was conducted in compliance with the ARRIVE guidelines (41). All animal care and experimental procedures were approved by the Duke University Institutional Animal Care and Use Committee. When not performing experiments, animals were housed under USDA- and AAALAC-compliant conditions, with free access to food and water and a 12:12-h light-dark cycle. Rats were single-housed after implantation but given extra environmental enrichment. We conducted stereotactic surgery under 3.0–3.5% sevoflurane anesthesia, using aseptic technique and coordinates from a rat brain atlas (42). Heart rate, oxygen saturation, and body temperature were monitored throughout the surgery, and the body temperature was maintained between ∼35°C and 37°C with a heated water blanket. For analgesia, meloxicam (2 mg/kg sc) and bupivacaine (<2 mg/kg) were administered ∼15 min before incision. To prevent infection, enrofloxacin (Baytril, 5–10 mg/kg sc) was administered both before incision and 24 h after and antibiotic ointment was applied to the incision site after implantation. The cannula and electrodes were implanted unilaterally and ipsilateral to each other, with the hemisphere randomized between the rats (no effect of hemisphere was observed). Implantations were performed with a stereotaxic electrode manipulator and inserted manually at a rate of ∼100 µm/s. One stimulating microelectrode array (2 × 2, platinum-iridium, 75-μm diameter, 0.3-mm interelectrode spacing, and 10-kΩ impedance; Microprobes, Gaithersburg, MD) was implanted in the STN [3.6 mm posterior (P), 2.6 mm mediolateral (ML) from bregma; 6.6–6.9 mm dorsoventral (DV) from surface of brain]. A cannula (23 gauge stainless steel needle, cut to 1.9-cm length) was placed in the MFB [2.0 mm P, 2.0 mm ML from bregma; 8.5 mm from surface of skull]. In addition, stainless steel bone screws (Fine Science Tools, item no. 19010-00) were used to secure the headcap, with one bone screw located above M1 [2.5 mm anterior, 2.5 mm ML from bregma] used to record the M1 ECoG. A bone screw located above the cerebellum was used as a reference. The stimulating microelectrode array and the ECoG contacts were connected to separate headstages (Omnetics, A79000) for subsequent connections to stimulating and recording cables. Rats were given an additional dose of meloxicam 24 h after the first dose and were monitored closely.

At 1 wk after implantation, rats were lesioned under 3.0–3.5% sevoflurane anesthesia to cause unilateral degeneration of dopaminergic neurons in the substantia nigra pars compacta (SNc). Sixty minutes before lesioning, the rats were pretreated with 50 mg/kg pargyline and 5 mg/kg desipramine injected intraperitoneally. The 6-OHDA (Sigma-Aldrich; 5 mg 6-OHDA/3 mL 0.9% NaCl) solution was prepared immediately before use, and 10 µL was infused through the cannula at a rate of 1 µL/min. Rats were left to recover for at least 1 wk before any additional measurements. Lesions were assessed via methamphetamine-induced circling (see Electrical Stimulation), and circling of at least 3 turns/min in the direction ipsilateral to the lesion indicated >90% loss of dopaminergic neurons in the SNc (43, 44). If this criterion was not met, rats were lesioned again, up to a total of three times, and secondary administrations were required in most rats.

Electrical Stimulation

Stimulation was conducted with an isolated voltage-to-current convertor (A-M Systems) controlled by MATLAB (MathWorks, RRID:SCR_001622, 2020b) or LabView (National Instruments, RRID:SCR_0143252017, SP1) software. Charge-balanced biphasic pulses (pulse width 90 µs/phase when not specified) were applied at set stimulation frequencies. Bipolar stimulation was applied between two of the four electrodes. The stimulation amplitude and electrode contacts were determined for each rat before the start of experiments with a contact testing protocol in which each contact pair and stimulation amplitude were assessed on the basis of their ability to produce sustained motor responses during 130-Hz stimulation, including increased contralateral turning, decreased ipsilateral turning, increased activity, and a lack of motor side effects including involuntary muscle contractions of the limbs and neck. Stimulation amplitudes typically varied between 50 and 100 µA, much smaller than in human clinical DBS but consistent with amplitudes typically used for DBS in rats because of the small size of the STN and electrodes (44). For all trials, unless otherwise specified, during the first 5 s of stimulation the amplitude was ramped up linearly to help reduce potential onset side effects. When possible, an alternating-phase method of stimulation was applied, wherein successive pulses alternated between cathodic phase-first and anodic phase-first stimulation (Fig. 1B). This inverted the stimulation artifact between successive pulses, and when the evoked potential was averaged over time, the artifact was removed and only the evoked potential remained. However, in some rats this method was less therapeutically effective than standard stimulation with nonalternating phase orders, in which case the more effective method was used.

Characterization of the cEP across Stimulation Parameters

Simultaneous STN stimulation and ECoG recording were performed while the rats were in an awake, freely moving state in their home cages to characterize changes in cEP features across stimulation conditions. In one set of experiments, four different stimulation frequencies (25, 75, 130, 200 Hz) were applied at four different amplitudes (0.4, 0.6, 0.8, 1.0 max amplitude) titrated to each rat’s therapeutically effective amplitude determined during the contact testing protocol. In a second set of experiments, the same four stimulation frequencies were applied with three different pulse widths (60, 90, 120 µs) using each rat’s therapeutically effective amplitude. In a third experiment, the same four stimulation frequencies were applied both in an awake state followed by an anesthetized state under 3.0–3.5% sevoflurane anesthesia, with the amplitude once again set to each rat’s therapeutically effective amplitude and with rats anesthetized for at least 20 min before recordings began. For all three of these experiments, the conditions were randomized, the recordings were 3 min long, and there was a 5-min washout period between trials. Because of the low variability of the cEP within animals, each of these experiments was performed only once for each animal. The person conducting the experiment was aware of the stimulation settings used, but all offline analysis of the data was automated and thus blinding was not applicable.

In a subset of rats, we performed a probe-pulse stimulation technique to measure cEPs. We applied two superimposed stimulation trains, a first functional stimulation train that could have varying stimulation frequencies and amplitudes and a second probe stimulation train at a lower frequency with fixed stimulation parameters. This probe-pulse train was set at a subharmonic of the first functional train so that the probe pulses simply replaced those of the functional train. In our offline analysis we then only quantified the cEPs evoked from the probe pulses, enabling us to determine any functional changes in cEP features due to the different stimulation conditions. We used this technique with four stimulation frequencies (13, 75, 130, 200 Hz) and three amplitudes (0.6, 0.8, 1.0 max amplitude) titrated to each rat’s therapeutically effective amplitude. We used a probe pulse frequency of 13 Hz, and the subharmonic closest to 13 Hz was selected for the probe pulses (e.g., for 75 Hz, every 6th pulse was a probe pulse), with the probe pulse set to 100% amplitude. This technique was used during methamphetamine-induced circling (see Behavioral Tests) to assess improvements in motor behavior.

In another set of experiments, we quantified the refractory period of the cEP. While rats were in an awake, freely moving state we delivered a paired-pulse stimulation protocol using interpulse intervals (IPIs) of 0.5, 0.75, 1, 1.25, 1.5, 1.75, 2, 3, 4, 5, 15, 50, 100, and 1,000 ms, with 1 s between the start of each pair of stimuli and a total of 120 stimuli pairs applied over 2 min. The order of each 2-min stimulation epoch was randomized, and a 5-min washout period elapsed between epochs. The cEP magnitudes of the second pulses in each pair, normalized to the cEP magnitude at 1 Hz, were then fit by a sigmoid, $y\mathrm{=}\frac{1}{1\mathrm{+}{e}^{\left(t_{50}\mathrm{-}t\right)\mathrm{\times slope}}}$, to determine the refractory period of the cEP. Each experiment was performed once in each animal with stimulation contacts and amplitude chosen during the contact test to be therapeutically effective. The person conducting the experiment was aware of the stimulation settings used, but all offline analysis of the data was automated and thus blinding was not applicable.

Behavioral Tests

Simultaneous STN stimulation and ECoG recordings were conducted during behavioral tests at different stimulation frequencies to determine the correlation between various biomarkers and changes in motor behaviors. Seven different stimulation frequencies were used (13, 25, 50, 75, 100, 130, 200 Hz), with individual frequencies selected to span a range of known behavioral efficacies (37, 38, 45–47) and not to fall on harmonics of 60 Hz. In one animal we used a different set of stimulation frequencies, 10, 20, 30, 50, 60, 75, 100, 115, and 130 Hz, and the results from this animal were included in the correlation analysis but were not used in the characterization of the cEP over time.

We used two different behavioral tests to assess motor behaviors: methamphetamine-induced circling and an adjusting steps task. For the circling task, rats were given a single injection of methamphetamine (1.875 mg/kg in 0.9% saline ip) 20 min before placement in a cylinder (44). The cylinder was placed in a dark chamber, and an infrared camera was used to capture the rat behavior. A rotating electrical commutator (PlasticsOne) was used to prevent cable twisting. The behavioral effects of DBS were quantified with randomized blocks of the seven different stimulation frequencies. Each stimulation frequency was repeated at least three different times, with 60-s stimulation epochs spaced 120 s apart. Circling behavior was video recorded and tracked with behavioral analysis software (Clever Systems, Reston, VA). The angular velocity and distance traveled per minute (linear speed) were calculated offline from the tracking data with MATLAB. The normalized angular velocity and linear speed for each trial were calculated by dividing the average angular velocity or linear speed during the stimulation-on period by the average angular velocities or linear speeds during the 60-s pre-stimulation-off and post-stimulation-off periods immediately before and after each stimulation-on period. The normalized angular velocities and linear speeds for each stimulation rate were then averaged across all randomized blocks. Both the experiment (e.g., applying randomized stimulation blocks) and data analysis were automated, so blinding was not applicable.

We quantified deficits in contralateral limb use via a forelimb adjusting steps task, which is a validated measure of parkinsonian akinesia in rats (48, 49). Rats were held with their hindlimbs elevated and pulled backward at a steady rate over a 1-m glass runway (Runway, CleverSys). Each epoch consisted of three to five trials across the runway. The movement was video recorded from below and analyzed offline by counting the number of adjusting steps taken with each forelimb. The person conducting the experiment was aware of the stimulation settings used, but all offline analysis of the video recordings was blinded to the stimulation conditions. Behavioral efficacy of DBS was quantified by taking the ratio of the number of contralateral steps to the number of ipsilateral steps. Each of the seven stimulation frequencies, as well as a no-stimulation control, were applied in random order before and during each adjusting steps session. Stimulation was delivered for 5 min before each epoch began and continued through the epoch. A 5-min no-stimulation washout period elapsed between stimulation epochs. The 5 min of recording data before each stepping epoch began, during which the rat was stimulated but left in an awake, freely moving state, was also used on its own to quantify changes in the cEP over time (see Cortical Response to Variations in STN DBS Parameters) and to quantify the change in spectral biomarkers between baseline and 130-Hz stimulation (see Properties of Cortical Spectral Biomarkers).

When possible, multiple (typically 1–3) methamphetamine-induced circling and adjusting steps experiments were performed in each rat, with the behavioral score and biomarker values first averaged across all replicates of each stimulation condition before quantifying their correlations. For the washout periods during each task, most behavioral effects of DBS are known to wash out within minutes of turning stimulation off (50), and as such our washout periods of ≤5 min are typical for DBS experiments both in rodents (11, 40, 47, 51, 52) and in humans (11, 53, 54). Additionally, we conducted preliminary tests and found that the cEP magnitude and latency recovered within a few seconds after stimulation was turned off (data not shown).

Electrophysiological Recording and Analysis

The ECoG signal was sampled at either 20 or 100 kHz with a multichannel acquisition processor system (Plexon Inc., Dallas, TX) with a 50× gain. Four biomarkers were quantified offline with custom scripts in MATLAB: beta band power, beta band bursting, phase-amplitude coupling (PAC), and the average cEP from 0 to 5 ms after stimulation. From the average cEP profiles, we measured the value and latency of the short-latency cEP peak with MATLAB’s findpeaks function within the range of 1–3 ms after stimulus. We then quantified the magnitude, defined as the difference between the maximum and minimum voltage during the time from the cEP peak to the point 5 ms after stimulus. This measure of magnitude, as opposed to simply the peak value, was more reliable, as it was less affected by noise and has also been used to quantify human cEP magnitude (55).

For each spectral measurement, the stimulation artifacts were blanked, data were downsampled to 1 kHz, and the baseline was subtracted. Artifact blanking was performed with a Gaussian weighting scheme implemented over the 100 nearest stimulation artifact templates. Spectral power (13–30 Hz) was calculated with the mtspectrumc function from the Chronux MATLAB package (http://chronux.org/, RRID:SCR_005547) (56). We quantified the power in six different bands: theta (4–8 Hz), alpha (3–13 Hz), low-beta (13–20 Hz), high-beta (20–30 Hz), beta (13–30 Hz), and gamma (30–50 Hz). Power was then normalized to the broadband power from 4 to 90 Hz. To quantify PAC, we calculated the modulation index (MI), using the method from Ref. 57, with phase frequencies of 1–30 Hz and amplitude frequencies from 10 to 200 Hz. The PAC was then subdivided into 10 distinct phase/amplitude regions (Table 1), with individual boundaries selected based on the observed regions of strongest coupling across animals. Within each region, the average value of the 10 highest MIs within that region was quantified. Beta band bursting was calculated using the method from Ref. 5, performing band-pass filtering within set frequency bands (theta 4–8 Hz; alpha 8–13 Hz; low-beta 13–20 Hz; high-beta 20–30 Hz), applying the Hilbert transform, and then using a 75% threshold to quantify bursting behavior, thus accounting for changes in overall signal power. With this method, we quantified both the mean and 90th percentiles of the beta burst amplitude and durations. We also tracked the value of the 75% threshold itself, as it serves as a metric of overall power in that band.

Table 1.

cEP characterization statistics

Condition	n	Normalized by	Variable	cEP Magnitude			cEP Latency
				B	t	P	B	t	P
Amplitude	6	Single condition	Frequency	−3.0E−03	−20.31	<0.0001*	0.002	16.27	<0.0001*
			Amplitude	0.611	15.06	<0.0001*	0.030	0.96	0.3423
			Freq × Amp	-4.4E-03	−6.85	<0.0001*	1.0E−04	0.21	0.8346
		Each condition	Frequency	−4.2E−03	−24.18	<0.0001*	1.8E−03	16.98	<0.0001*
			Amplitude	−3.1E−03	−0.06	0.9498	1.7E−03	0.06	0.955
			Freq × Amp	−6.4E−05	−0.08	0.934	3.4E−05	0.07	0.9418
Pulse width	5	Single condition	Frequency	−3.3E−03	−19.46	<0.0001*	1.3E−03	9.47	<0.0001*
			Pulse width	4.3E−03	9.48	<0.0001*	6.5E−04	1.77	0.0824
			Freq × PW	−2.8E−05	−4.02	0.0002*	7.6E−07	0.13	0.8938
		Each condition	Frequency	−4.1E−03	−20.31	<0.0001*	1.3E−03	9.35	<0.0001*
			PW	3.8E−04	0.72	0.4756	−1.4E−04	−0.37	0.7094
			Freq × PW	−4.9E−07	−0.06	0.9525	−3.5E−08	−0.01	0.9953
Sevoflurane	6	Single condition	Frequency	−3.2E−03	−14.38	<0.0001*	1.3E−03	5.03	<0.0001*
			Sevoflurane	0.140	9.69	<0.0001*	−9.2E-02	−5.67	<0.0001*
			Freq × Sevo	−1.1E−03	−5.01	<0.0001*	9.2E−05	0.37	0.7132
		Each condition	Frequency	−4.3E−03	−14.93	<0.0001*	1.2E−03	5.43	<0.0001*
			Sevoflurane	8.2E−03	0.44	0.6624	3.0E−02	2.16	0.0368*
			Freq × Sevo	−1.2E−05	−0.04	0.9668	1.9E−04	0.92	0.3640

cEP, cortical evoked potential. n, Number of animals. *P < 0.05.

Histology

After the completion of experiments (2–6 mo after implant), rats were deeply anesthetized with urethane (1.8 g/kg ip) and perfused transcardially with 0.1 M phosphate-buffered saline (PBS) followed by 4% paraformaldehyde in 0.1 M PBS. The brain was postfixed overnight in 4% paraformaldehyde and then transferred to 30% sucrose. The brains were frozen in optimal cutting temperature compound and cut into 50-μm sections in either the coronal or horizontal plane with a cryostat (CM3050S, Leica Microsystems) and processed for two sets of staining: tyrosine hydroxylase (TH) and cresyl violet. TH immunohistochemistry was used to determine the extent of degeneration of dopaminergic neurons in the SNc and to visualize electrode locations (38, 40, 44). Briefly, after three rinses in PBS, brain sections were first incubated for 10 min in 3% hydrogen peroxide. The sections were rinsed and blocked for 1 h at room temperature in blocking solution containing 10% goat serum. The sections were then incubated in anti-tyrosine hydroxylase antibody (Millipore catalog no. AB152, RRID:AB_390204; 1:1,000) overnight at 4°C in PBS with 10% goat serum and 0.25% Triton X-100. After three rinses in PBS, the sections were incubated with biotinylated goat anti-rabbit secondary antibody (Vector Laboratories catalog no. BA-1000, RRID:AB_2313606; 1:250) with 10% goat serum and 0.25% Triton X-100 in PBS for 1 h at room temperature. After rinsing, the sections were incubated in a VECTASTAIN Elite ABC kit (Vector Laboratories) solution for 1 h and then visualized with DAB solution. Cresyl violet counterstaining was used to help verify electrode locations compared with a rat brain atlas (42).

Exclusion Criteria

There were two primary exclusion criteria for the implanted rats, decided a priori. The first was if the rats were not adequately lesioned after three separate attempts. This was verified behaviorally with methamphetamine-induced circling as well as with postmortem TH staining (see Histology). TH immunoreactivity in the striatum and SNc were visually compared across lesioned and nonlesioned hemispheres to verify the unilateral loss of dopaminergic cells (Supplemental Fig. S1; all Supplemental Material is available at https://doi.org/10.7924/r47s7t64c). The second was if the stimulation electrodes were improperly placed. Improper placement was determined if either the contact testing (see Electrical Stimulation) revealed no effective stimulation contacts or the postmortem histology (see Histology) revealed that the electrode tips were outside the STN. Of the 30 rats, 15 were successful and used in this study, 4 were excluded because of failure to induce a 6-OHDA lesion, and 7 were excluded because of improper electrode placement; additionally, 4 were excluded because of reaching humane end points that required euthanasia (2 rats had skin lesions, 1 had infection around the headcap, and 1 exhibited seizures after the 6-OHDA lesion).

Experimental Design and Statistical Analysis

All statistical tests were performed in JMP Pro 15.1 (Statistical Analysis System, RRID:SCR_008567) and incorporated the animal number as a random effect to account for repeated measures when applicable. In the first set of experiments, we characterized the effects of three variables (stimulation amplitude, pulse width, and anesthesia) on the cEP magnitude and latency at different stimulation frequencies. For each variable we performed a repeated-measures multiple regression analysis, quantifying the effects of the variable, stimulation frequency, and the interaction of the variable with stimulation frequency. In a second set of experiments, we fit the cEP magnitude over time to linear and exponential models and compared the model fits with the Akaike information criterion-corrected (AICc) weights, which provide an estimate of the quality of each model’s fit by balancing the trade-off between goodness of fit and model simplicity (58). We then performed repeated-measures linear regression on each term of the exponential models for each animal at each stimulation frequency. In another set of experiments, we characterized changes in spectral features between 0 and 130 Hz DBS, which we quantified with paired t tests. Finally, we quantified the Pearson’s product-moment correlation (PPMC) of five different biomarkers with two measurements of motor behavior. All experiments had a minimum sample size of 5 animals, which is based on a power analysis of our preliminary data performed in GPower 3.1 (59), using a minimum power of 0.8. All results are presented as means ± standard error and were considered significant at P < 0.05. Study data are available at the Duke Digital Repositories at https://doi.org/10.7924/r47s7t64c.

RESULTS

Cortical Response to Variations in STN DBS Parameters

We quantified the magnitude and latency of the cEP during DBS with four stimulation frequencies (25, 75, 130, 200 Hz) at four different amplitudes (0.4, 0.6, 0.8, 1.0 max amplitude) (Fig. 2A) and performed a repeated-measures multiple regression (Table 1; n = 6). Higher stimulation frequencies decreased the cEP magnitude and increased the cEP latency. Additionally, higher stimulation amplitudes increased the cEP magnitude but had no effect on cEP latency. Finally, there was a significant interaction between stimulation frequency and amplitude on the cEP magnitude, but not the cEP latency, indicating that at higher stimulation frequencies there was a smaller absolute effect of stimulation amplitude. However, when the cEP magnitudes at each amplitude were normalized to their value at the lowest stimulation frequency (25 Hz), there was no longer an effect of stimulation amplitude (Table 1), indicating that although stimulation amplitude affects the absolute reduction in cEP magnitude at high stimulation frequencies, it does not affect its relative change with frequency.

Figure 2.

Changes in cortical evoked potential (cEP) magnitude and latency at 4 different stimulation frequencies (25, 75, 130, 200 Hz) during different stimulation conditions. The left column of each panel is normalized for each animal to the 25 Hz value of a single condition, and the right column of each panel is normalized to the 25 Hz value of each condition. For each group we performed a repeated-measures multiple regression analysis (Table 1, *P < 0.05). A: cEP magnitude and latency as a function of stimulation frequency and amplitude (as % of each rat’s therapeutically effective amplitude) (n = 6 animals). Data in the Single Condition group are normalized to the 1.0 Amp value. For the Single Condition group, there were significant effects of frequency (Freq), amplitude (Amp), and Freq × Amp on the cEP magnitude but only a significant effect of frequency on the cEP latency. However, for the Each Condition group there was only a significant effect of frequency on both the cEP magnitude and latency. B: cEP magnitude and latency as a function of stimulation frequency and pulse width (PW; 60, 90, 120 µs) (n = 5 animals). Data in the Single Condition group are normalized to the 120 µs PW value. For the Single Condition group, there were significant effects of frequency, PW, and Freq × PW on the cEP magnitude but only a significant effect of frequency on the cEP latency. However, for the Each Condition group there was only a significant effect of frequency on both the cEP magnitude and latency. C: cEP magnitude and latency as a function of stimulation frequency during an awake vs. sevoflurane (Sevo)-anesthetized state (n = 6 animals). Data in the Single Condition group are normalized to the Awake value. For the Single Condition group, there were significant effects of frequency, sevoflurane, and Freq × Sevo on the cEP magnitude and significant effects of frequency and sevoflurane on the cEP latency. However, for the Each Condition group there was only a significant effect of frequency on the cEP magnitude but significant effects of both frequency and sevoflurane on the cEP latency. Error bars indicate ±SD.

We quantified the cEP amplitude and latency during DBS with three different pulse widths (60, 90, 120 μs) at four different stimulation frequencies (13, 75, 130, 200 Hz) (Fig. 2B) and performed a multiple regression (repeated-measures multiple regression, n = 5; Table 1). Similar to the cEP amplitude, there were significant effects of frequency, pulse width, and frequency × pulse width on the cEP magnitude but only a significant effect of frequency on the cEP latency. Additionally, when the cEP magnitude at each pulse width was normalized to its value at 25 Hz there was no longer a significant effect of pulse width.

We quantified the effect of sevoflurane anesthesia on the cEP. Four different stimulation frequencies were applied (25, 75, 130, and 200 Hz) while the rats were in an awake, freely moving state, and after these recordings rats were anesthetized for at least 20 min under 3.0–3.5% sevoflurane before a second round of recordings (Fig. 2C). There were effects of stimulation frequency and sevoflurane, with sevoflurane reducing the cEP magnitude and increasing its latency (repeated-measures multiple regression, n = 6; Table 1). Additionally, there was a significant interaction term between stimulation frequency and sevoflurane on the cEP magnitude but not the cEP latency, indicating that sevoflurane caused a smaller absolute reduction in cEP magnitude at high stimulation frequencies. However, similar to the effects of stimulation amplitude and pulse width, when the cEP magnitude was normalized to its value at the lowest stimulation frequency (25 Hz), there was no longer a significant effect of sevoflurane or interaction between stimulation frequency and sevoflurane. Thus, sevoflurane led to a smaller absolute reduction in cEP magnitude at high frequencies but did not change the relative reduction in cEP magnitude.

The refractory period of the cEP (Fig. 3) was determined as t₅₀ = 0.985 ms (n = 5), which made clear that the change in cEP magnitude at higher stimulation frequencies was not due to neuronal refractoriness. In addition to the refractory period, there was also a hyperexcitable interval between ∼5 and 10 ms, which corresponds to stimulation frequencies from 100 to 200 Hz.

Figure 3.

Cortical evoked potential (cEP) magnitude as a function of paired pulse interval, which varied from 0.5 to 100 ms (n = 5 animals). A: the average cEP waveform across paired pulse intervals in a representative animal. Note that the waveform from approximately time 0 to 1.0 ms is the stimulation artifact. B: each cEP magnitude normalized to its value using 1-Hz stimulation. Colors represent individual values from each rat. The data were fit with a sigmoid with its maximum value capped at 1 to determine the relative refractory period, which was t₅₀ = 0.985 ms. Notably, there was also a hyperexcitable region from ∼5 to 10 ms. IPI, interpulse interval.

We analyzed 5-min recordings of the cEP at different stimulation frequencies while the rat was in an awake, freely moving state to quantify how the cEP magnitude changed over time (n = 7). It appeared that there was an exponential decay in the cEP magnitude for stimulation frequencies ≥ 50 Hz (Fig. 4A), and to test this, we quantified the average cEP magnitude at six different time points (5, 10, 30, 60, 120, 300 s) at each stimulation frequency (Fig. 5A), normalized to the peak cEP magnitude in each animal. We compared linear and exponential fits at each stimulation frequency, using the AICc weights. The 13- and 25-Hz stimulations were best fit linearly, whereas stimulation frequencies ≥ 50 Hz were best fit with an exponential decay (Table 2).

Figure 4.

Change in cortical evoked potential (cEP) magnitude over time. A: the cEP magnitude over 5-min trials at 7 different stimulation frequencies (13, 25, 50, 75, 100, 130, 200 Hz) from a representative animal. The cEP is averaged over 500-ms windows. Note that there is only an observable change over time for stimulation frequencies ≥ 50 Hz. B: example of the exponential fitting process using the 130 Hz trial in A.

Figure 5.

Statistical analysis of the change in cortical evoked potential (cEP) magnitude over time. A: the cEP magnitude binned at 6 different time points (5, 10, 30, 60, 120, 300 s) at 7 different stimulation frequencies (13, 25, 50, 75, 100, 130, 200 Hz), normalized to each animal’s peak cEP magnitude (n = 7 animals). Error bars indicate ± SD. Each curve was fit to both a linear and an exponential model, and the Akaike information criteria-corrected (AICc) weight revealed that an exponential fit was superior for stimulation frequencies ≥ 50 Hz (Table 2). For these higher frequencies we then fit the cEP magnitude for each animal at each stimulation frequency to 3-parameter exponential decay curves (n = 7). B: the exponential decay time constant. C: the steady-state value. D: the y-intercept. A repeated-measures linear regression of each term revealed a significant effect of stimulation frequency for the steady-state value and the y-intercept but not for the time constant (Table 3, *P < 0.05).

Table 2.

Exponential vs. linear fits of cEP magnitude

Frequency	n	Model Fit	AIC	AICc Weight
13 Hz	7	Linear	−125.83329	0.602
		Exponential	−125.00473	0.398
25 Hz	7	Linear	−78.701648	0.676
		Exponential	−77.232482	0.324
50 Hz	7	Linear	−75.853155	0.074
		Exponential	−80.906623	0.926
75 Hz	7	Linear	−62.473505	0.000
		Exponential	−81.514918	1.000
100 Hz	7	Linear	−58.251364	0.000
		Exponential	−99.14181	1.000
130 Hz	7	Linear	−50.651149	0.000
		Exponential	−93.276194	1.000
200 Hz	7	Linear	−65.235884	0.000
		Exponential	−101.51872	1.000

AIC, Akaike information criterion; AICc, AIC corrected; cEP, cortical evoked potential. n, Number of animals.

To characterize further the cEP magnitude during high-frequency DBS (≥50 Hz), we quantified the cEP magnitude for each animal and stimulation frequency as a function of time, averaged in 500-ms windows, and fit it with an exponential decay model of $y\mathrm{=}a\mathrm{\times }\exp \hspace{0.17em}\left(\mathrm{-}\frac{x}{\tau }\right)\mathrm{+}c$ (Fig. 4B). The first 5 s of each trial was excluded from the analysis because we applied a 5-s ramp-up of stimulation amplitude at the beginning of each trial. We analyzed three terms: the y-intercept (a + c), the steady-state value (c), and the time constant (τ) (Fig. 5, B–D). We quantified the effect of stimulation frequency on each term with repeated-measures linear regression (n = 7; Table 3). Higher stimulation frequencies reduced both a + c and c but not τ (28.1 ± 4.3 s). The change in c is unsurprising, as it corresponds to the steady-state changes in cEP magnitude observed previously. However, the change in a + c was unexpected, as it indicates that with higher stimulation frequencies the predicted y-intercept decreases. Because we excluded the first 5 s in the analysis, we cannot state with certainty what occurs during that time. However, our refractory period experiment indicates that the true y-intercepts for these stimulation frequencies should all be approximately the same, so this observation indicates that, in addition to the long-term changes in cEP magnitude that occur from 5 to 300 s, there is likely an additional short-term change that occurs over the first 5 s that is also affected by stimulation frequency. The difference in timing of these stages may indicate multiple mechanisms causing the reduction in cEP magnitude, such as conduction failure of action potentials at the branch point (60), submyelin accumulation of potassium along the main axon (61), or invasion failure at the soma (62). The cEP latency exhibits a similar long-duration change over time (Supplemental Fig. S2), although we did not perform curve-fitting because of the cEP latency generally having far greater variability over time than the cEP magnitude. This is primarily due to the comparatively low temporal resolution of our system, as the differences in cEP latencies between stimulation frequencies are on the order of ∼0.1 ms, and our 100-kHz sampling frequency only has a resolution of 0.01 ms.

Table 3.

Temporal analysis of cEP magnitude

Term	n	B	t	P
a + c	7	−3.61E−04	−3.49	0.0129*
c	7	−5.89E−04	−4.05	0.0067*
τ	7	−6.20E−02	−2.14	0.0761

a + c, y-intercept; c, steady-state value; cEP, cortical evoked potential; τ, time constant. n, Number of animals. *P<0.05.

When interpreting the variance or “noisiness” of the cEP signal in Figs. 4 and 5, it is important to note two points. First, in Fig. 4 we averaged the cEP over 500-ms windows to visualize the dynamic changes in the cEP over time. However, larger windows can be used, and this will decrease the observed “noise.” For example, in the cEP characterization results shown in Fig. 2, the entire recording period was averaged to produce a single very clear cEP waveform (see Fig. 1). Second, it is not correct to view each trace in Fig. 4 as its own signal (as in some difference from a baseline). Rather, the difference between traces is the “signal” (i.e., can the cEP biomarker differentiate between one stimulation frequency and another?), and by this view, the cEP magnitude and latency clearly exhibit high SNRs, as the differences between stimulation frequencies are highly distinguishable.

Properties of Cortical Spectral Biomarkers

We quantified three spectral-based biomarkers from our cortical recordings: the power spectrum, spectral bursting, and PAC. We compared no stimulation and 130-Hz DBS using 5-min recordings during an awake, freely moving state (n = 11). For the no-stimulation trials, we performed artifact subtraction using the artifact times from the 130-Hz trials to ensure that the artifact blanking method was not contributing to any changes. The power spectra (Fig. 6) exhibited a pronounced peak at ∼7 Hz, as well as a peak in the high-beta range. DBS at 130 Hz reduced the low-frequency power in the alpha (8–13 Hz) and low-beta (13–20 Hz) bands (paired t test, n = 11; Table 4). There was no significant change in the theta (4–8 Hz), high-beta (20–30 Hz), or gamma (30–50 Hz) bands across animals, although a few animals did have prominent high-beta peaks that were modulated by DBS (Supplemental Fig. S3).

Figure 6.

Average spectral power during 5-min recordings while rats were in an awake, freely moving state (n = 11 animals) either with no stimulation (red) or with 130-Hz stimulation (blue). The 95% confidence interval for the no-stimulation group is in black. Spectra were normalized to the broadband power from 4 to 90 Hz. The deflection at 65 Hz is due to the stimulus artifact removal. Performing a paired t test of the effect of 130-Hz deep brain stimulation (DBS) in individual frequency bands showed a significant reduction in power in the alpha (8–13 Hz), and low-beta (13–20 Hz) bands but not in the theta (4–8 Hz), high-beta (20–30 Hz), and gamma (30–50 Hz) bands (n = 11 animals; Table 4, *P < 0.05).

Table 4.

Power spectrum paired t-test statistics

Band	Frequency, Hz	DF	t	P
Theta	4–8	10	1.66	0.129
Alpha	8–13	10	2.82	0.018*
Low-beta	13–20	10	2.38	0.038*
High-beta	20–30	10	0.11	0.914
Gamma	30–50	10	−1.95	0.080

*P < 0.05.

Bursting in the beta band in the human STN is a potential biomarker for closed-loop DBS (5), and we quantified whether similar bursting was present in the rat M1 ECoG. We quantified bursting in the theta (4–8 Hz), alpha (8–13 Hz), low-beta (13–20 Hz), and high-beta (20–30 Hz) bands, measuring the average burst duration and amplitude, as well as the 75% threshold (Fig. 7). There was a reduction in all three features during 130 Hz DBS, but only in the alpha and low-beta bands (paired t test, n = 11; Table 5). We did not detect any change in any bursting feature in the theta and high-beta bands.

Figure 7.

Spectral bursting characteristics during 5-min recordings while rats were in an awake, freely moving state (n = 11 animals) either with no stimulation or with 130-Hz deep brain stimulation (DBS). Bursting characteristics were quantified for 4 bands: theta (4–8 Hz), alpha (8–13 Hz), low-beta (13–20 Hz), and high-beta (20–30 Hz). In each band we quantified the average burst duration (top), average burst amplitude (middle), and 75% burst threshold (bottom). A paired t test of the effect of stimulation on each metric in each band showed a significant effect of stimulation for all 3 metrics in the alpha and low-beta bands but not in the theta and high-beta bands (n = 11 animals, Table 5, *P < 0.05).

Table 5.

Spectral bursting paired t-test statistics

Frequency, Hz	DF	Burst Duration		Burst Amplitude		Burst Threshold
		t	P	t	P	t	P
4–8	10	0.68	0.512	0.900	0.392	1.150	0.276
8–13	10	2.73	0.021*	3.280	0.008*	3.500	0.006*
13–20	10	2.55	0.029*	2.910	0.016*	3.140	0.011*
20–30	10	1.36	0.204	0.570	0.581	0.810	0.438

*P < 0.05.

Finally, we quantified the cortical PAC, which is an informative cortical biomarker in humans (63, 64). The average PAC across the 5-min trials was quantified for phase frequencies 1–30 Hz and amplitude frequencies 10–200 Hz (Fig. 8). In the baseline case there were two common regions of increased coupling: a low-phase, low-amplitude region with phase < 10 Hz and amplitude < 50 Hz, and a low-phase, high-amplitude region with phase < 20 Hz and amplitude > 50 Hz. However, there was substantial variation across animals, with some exhibiting far less coupling than others (see Supplemental Fig. S4 for plots from all individual animals). One hundred thirty-hertz DBS caused changes in coupling in most animals, but the specific subregions that changed were inconsistent across animals. When looking at the average difference in PAC across animals (Fig. 8C), there was a general decrease in theta/low-amplitude coupling and increase in alpha/high-amplitude coupling, but this was primarily due to a few outlier animals and was not present across all. To determine whether there were any consistent changes, we averaged the 10 highest PAC MI values in 10 different subregions (Table 6) and found that the only subregion with a significant change from stimulation was the high-amplitude region (phase 1–30 Hz, amplitude 75–200 Hz), which had an increase in PAC with DBS (paired t test, n = 11; Table 6). However, this effect was largely driven by a few animals and was not present in all (Supplemental Fig. S4).

Figure 8.

A and B: phase-amplitude coupling (PAC) during 5-min recordings while rats were in an awake, freely moving state (n = 11 animals) either with no stimulation (A) or with 130-Hz deep brain stimulation (DBS) (B). The color bar represents the average modulation index for each phase-amplitude pair across all animals. C: the average difference in modulation index between no stimulation and 130-Hz stimulation across animals. Across animals, DBS appears to decrease coupling in the low-phase, low-amplitude region (phase 1–5 Hz, amplitude 10–50 Hz) and increase coupling in the alpha, high-amplitude region (phase 8–13 Hz, amplitude 75–120 Hz). Binning the PAC into 10 subregions and quantifying the effect of stimulation revealed a significant difference only in the high-amplitude region (phase 1–30 Hz, amplitude 75–200 Hz) (paired t test, n = 11 animals; Table 6).

Table 6.

PAC regions and paired t-test statistics

Region	Phase, Hz	Amplitude, Hz	DF	t	P
All	1–30	30–200	10	0.39	0.705
Theta	1–8	10–50	10	−0.10	0.190
Alpha/beta	8–20	50–200	10	−1.41	0.190
Beta	13–30	50–200	10	0.23	0.822
Low phase	1–20	20–200	10	0.03	0.973
High Amp	1–30	75–200	10	−2.44	0.035*
Low phase/low Amp	1–20	20–50	10	−0.61	0.554
Low phase/high Amp	1–20	50–200	10	0.55	0.594
High phase/low Amp	20–30	30–50	10	1.00	0.342
High phase/high Amp	20–30	50–200	10	0.65	0.528

Amp, amplitude; PAC, phase-amplitude coupling. *P < 0.05.

Correlation between Electrophysiological Biomarkers and Motor Behavior

In the second set of experiments, we determined the per-animal correlations between two quantitative measurements of motor behavior and five different cortical biomarkers: the cEP magnitude, cEP latency, spectral power, spectral bursting, and PAC (Fig. 9; raw motor behavior scores provided in Supplemental Fig. S5). Multiple metrics were quantified for each of the spectral features, (e.g., low-beta and high-beta power), but only the metric that had the highest average correlation with behavior was used for subsequent analysis (the correlations between behavior and all metrics are provided in Supplemental Fig. S6). For the spectral power, the feature with the strongest correlation with behavior was the beta band power (13–30 Hz), for spectral bursting it was the average amplitude of the low-beta (13–20 Hz) bursts, and for PAC it was the theta/low-amplitude (phase 1–8 Hz, amplitude 10–50 Hz) range (Fig. 10). Across both behavioral metrics, the two cEP biomarkers, and especially the cEP magnitude, consistently exhibited the strongest average correlation with motor behavior, with little variation between animals (Fig. 10). The lower average correlations for the other biomarkers were due to the large variation between animals, as some rats had their strongest correlations with beta band power and PAC whereas others exhibited very weak correlations. This is likely because only those rats with high baseline beta band power would exhibit strong correlations, similar to the observation that elevated beta band power is not always seen in PD patients (65).

Figure 9.

Behavioral scores and selected biomarker values across a range of stimulation frequencies during 2 behavioral tasks. In each panel, the score for the adjusting steps task is in blue (n = 8 animals) and that for the methamphetamine-induced circling task is in red (n = 10 animals). Plotted are the average normalized (z score) values across animals, with error bars indicating ±SD. A: the behavioral scores for both tasks. Note that a higher score in the adjusting steps task and a lower score in the methamphetamine-induced circling task indicate superior improvement in motor behaviors. B: the phase-amplitude coupling (PAC) modulation index in the theta/low-amplitude range (phase 1–8 Hz, amplitude 10–50 Hz). C: the beta band power (13–30 Hz), normalized to the broadband power (4–90 Hz). D: the average low-beta band (13–20 Hz) burst amplitude. E: the cortical evoked potential (cEP) latency. F: the cEP magnitude.

Figure 10.

Correlation of biomarkers with motor behavior for the adjusting steps task (n = 9 animals) and methamphetamine (Meth)-induced circling (n = 11 animals). PAC-Theta is the phase-amplitude coupling modulation index in the theta/low-amplitude range (phase 1–8 Hz, amplitude 10–50 Hz), Beta Band Power is the power in the beta range (13–30 Hz) normalized to the broadband power (4–90 Hz), and Beta Burst Amplitude is the average amplitude of the low-beta band (13–20 Hz) bursts. A: the Pearson’s product-moment correlation coefficient squared, with the colored dots indicating each individual animal, the bars indicating the average value across animals, and the error bars indicating the SD. B: the proportion of animals with significant correlation probabilities < 0.05. The numbers above each bar indicate the numbers of animals with significant correlations out of the total number of animals. Across behavioral tasks, both cortical evoked potential (cEP) biomarkers, and especially the cEP magnitude, consistently demonstrated the strongest correlations, largely due to the low variability between animals.

In a subset of rats (n = 4), we performed an additional experiment to confirm further that the correlation between motor behavior and the cEP biomarkers holds across stimulation amplitude as well as stimulation frequency. We used a probe-pulse technique (see materials and methods) to maintain consistent cEP measurements across stimulation amplitudes, testing four stimulation frequencies (13, 75, 130, 200 Hz) at three amplitudes (60%, 80%, 100%). All four animals showed strong correlations with motor behavior for both cEP magnitude and latency (Fig. 11), and these results confirm the robust correlation between cEP features and motor behavior across stimulation conditions.

Figure 11.

Correlation of cortical evoked potential (cEP) magnitude and latency with methamphetamine-induced circling across stimulation frequencies and amplitudes. A: probe-pulse technique, which enables consistent probing of the cEP while changing the stimulation frequency/amplitude. The probe pulses, delivered at 13 Hz in this case, are kept at the same amplitude while the other pulses are allowed to vary. During the offline cEP quantification only the cEPs from the probe pulses are included in the analysis. B: results from the methamphetamine-induced circling task using the probe-pulse technique with stimulation frequencies of 13, 75, 130, 200 Hz and amplitudes of 0.6, 0.8, 1.0 max amplitude, each titrated to the individual rat’s therapeutically effective amplitude (n = 4 animals): the behavioral score (top), with lower scores indicating superior improvement in motor behavior, and the cEP magnitude (middle) and the cEP latency (bottom), with each normalized to their values at 13 Hz. Symbols indicate individual scores from each animal. C: correlation of the cEP magnitude and latency with circling score for each animal. All correlations had a statistically significant Pearson’s product-moment correlation coefficient, with average R² values for the cEP magnitude and latency of 0.650 and 0.754, respectively.

DISCUSSION

The short-latency cEP in 6-OHDA-lesioned rats was highly consistent across animals and was strongly affected by stimulation conditions, especially the stimulation frequency. These changes were not due to the refractory period and occurred over a prolonged period of time. The cEP magnitude and latency exhibited strong correlations with motor behavior across animals, whereas the three spectral biomarkers quantified from the M1 ECoG did not. These results support a potential mechanistic link between the cEP and changes in motor behavior from DBS and suggest that the cEP may be a suitable biomarker for stimulation parameter selection or closed-loop control.

Changes in cEP Magnitude and Latency

The cEP magnitude declined and the cEP latency increased as a function of stimulation frequency, consistent with prior studies in rats and humans (17, 19). The short-latency cEP derives from stimulation-evoked action potentials in the hyperdirect pathway, which propagate antidromically and result in the simultaneous somatic firing of cortical projection neurons (17–20, 32–34). The reduced cEP magnitude at high stimulation frequencies is likely due to antidromic spike failure along the hyperdirect pathway that causes fewer action potentials to propagate back into the somas (33, 47). This same manner of frequency-dependent sporadic spike failure has been noted elsewhere in the cortex (60, 66–69), in the orthodromic projections from the STN in the basal ganglia (27, 28), and in the periphery (70). Concomitant with this spike failure, the latency of evoked activity also increases during high-frequency stimulation (28, 66, 70, 71). These changes in latency and spike failure may be mechanistically linked, as modeling suggests that spike failure during high-frequency stimulation may be due to altered ion channel dynamics as a result of ion accumulation (61, 72). Spike failure may thus be preceded by a decrease in conduction velocity.

Correlation of Spectral-Based Biomarkers with Behavior

Higher stimulation frequencies generally induced greater improvements in motor behavior, with efficacy leveling off at 130 Hz and possibly declining at 200 Hz, similar to previous studies in both rodents and humans (37, 38, 45–47). The two cEP biomarkers, and especially the cEP magnitude, consistently exhibited the strongest correlation with motor behavior, with little variation between animals. There was far greater variability in the correlations with the spectral biomarkers, with some rats exhibiting high correlations with specific spectral biomarkers but with most correlations generally not significant. High intersubject variability is a common challenge with spectral biomarkers. For example, although STN beta band power is generally regarded as a reliable biomarker for PD, elevated beta band power is not always present in patients with PD (65, 73, 74), effective DBS does not always reduce beta band power (65, 74), many studies note no significant correlation between beta and patients’ overall clinical state (73, 75), and stimulation that induces higher beta band power does not worsen motor behaviors (76). Such observations have led many to focus instead on patient-specific spectral biomarkers (8, 10, 77), which necessitates correlating individual spectral features with behavior. Our results support this direction, as some cortical spectral features correlated strongly with motor state in some animals but not in others. Our results also highlight the limitations of reporting biomarkers under only two conditions (e.g., off vs. 130 Hz), as many of the spectral biomarkers demonstrated significant changes from baseline during 130-Hz DBS, but those same biomarkers were then poorly correlated with changes in motor behavior across a range of intermediate frequencies. Thus, at least in the 6-OHDA-lesioned rat model, M1 ECoG spectral features were inconsistent biomarkers of the changes in motor behavior from DBS. However, these same spectral metrics from other areas of the brain, such as the STN, may exhibit stronger and more consistent correlations.

Correlation of cEP with Behavior

The robust and consistent correlations between behavior and the cEP features across both stimulation frequency and amplitude are notable because there is not an overt mechanistic reason why the cEP response should be so closely related to behavior. Because the cEP is nonsynaptic, the observed change in cEP features is likely not a direct indicator of behavioral state. This is in contrast to other spectral biomarkers, such as beta band power or beta burst duration, which are thought to be “stop” signals that indicate bradykinesia and rigidity within the motor network (75).

Instead, the relationship between the cEP features and behavior may be related to the mechanism by which DBS modulates the cortico-basal ganglia-thalamic network, regardless of behavioral state. The changes in cEP features are likely due to antidromic spike failure, and one theory, as put forward in Ref. 78, is that this spike failure itself drives the mechanism of DBS by providing a highly desynchronized input to M1. The degree to which the cEP magnitude declines may thus be an indicator of the relative “desynchronizing” effect of the stimulation. An alternative interpretation is that improvements in motor behavior may depend on the total cumulative activation of cortex during stimulation, quantified by the product of the cEP magnitude and the stimulation frequency (cortical activation per second, CAPS). The CAPS showed a near-equivalent correlation with motor behavior as the cEP magnitude and latency (Supplemental Fig. S7) and a metric similar to the CAPS, using the antidromic spike success rate instead of the cEP magnitude, correlated with motor behavior in rats (47). These two explanations are mutually exclusive, as they are oppositely related to the spike failure rate. To disambiguate between the CAPS and cEP magnitude, future work should test the effects of very high stimulation frequencies, as the CAPS and cEP magnitude trends appeared to diverge at 200 Hz (Supplemental Fig. S7), or the effects of random patterns of stimulation, as random patterns may cause different rates of antidromic spike failure with the same average stimulation frequency.

Limitations

This study is subject to the potential limitations inherent to translating animal studies to humans. Rats were implanted unilaterally with a 6-OHDA neurotoxin model of PD, which does not replicate the natural cause or progression of PD. Additionally, we assessed only hypokinetic motor behaviors and did not assess other motor or cognitive symptoms common in persons with PD (e.g., tremor, posture, mood, arousal). Despite these limitations, this disease model and behavioral tasks demonstrated strong parallels to the effects in human of both regular (37, 38) and nonregular (39, 40) temporal patterns of DBS. Additionally, the cEP has been observed in humans and exhibits the same frequency-dependent effects demonstrated here (17), which match well the known frequency-dependent behavioral effects of DBS in humans (37).

Conclusions

The features of the cEP are strongly correlated with changes in motor behavior from DBS, and this adds to the rapidly growing literature on applications of evoked potentials during DBS. EEG recordings of the cEP during STN DBS implant surgery showed promise in determining contacts that will not generate motor side effects (21), and contacts that generated the highest magnitude cEP were more likely to be therapeutically effective (20). The cEP was also recently used as the biomarker for therapeutic efficacy in a proof-of-concept framework for data-driven stimulation parameter optimization (79). The high signal-to-noise ratio of the cEP and its simple quantification add to its potential clinical utility. Monitoring of evoked potentials during DBS may thus be useful for electrode placement, stimulation parameter selection, and potentially even closed-loop control. However, because the cEP is likely not an indicator of the current behavioral state, it may not be a useful control signal for closed-loop systems designed to titrate stimulation based on instantaneous motor behavior. Instead, the cEP could be used as a control signal for long-term adjustments in stimulation parameters based on changes in the local environment around the electrode, such as impedance changes due to the foreign body response (80, 81), or in automatic stimulation parameter tuning (79). The strong correlation between the cEP and motor behavior provides compelling support that antidromic cortical activation mediates improvements in motor behavior from DBS. Additionally, the correlation between changes in cEP magnitude and motor behavior as a function of stimulation frequency indicates that antidromic spike failure may contribute to the strong dependence of DBS efficacy on stimulation frequency.

DATA AVAILABILITY

The data that support this study are available at https://doi.org/10.7924/r47s7t64c.

SUPPLEMENTAL DATA

Supplemental Figs S1–S7: https://doi.org/10.7924/r47s7t64c.

GRANTS

This work was funded by NIH R37 NS040894.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the author(s).

AUTHOR CONTRIBUTIONS

I.R.C. and W.M.G. conceived and designed research; I.R.C. performed experiments; I.R.C. analyzed data; I.R.C. and W.M.G. interpreted results of experiments; I.R.C. prepared figures; I.R.C. drafted manuscript; I.R.C. and W.M.G. edited and revised manuscript; I.R.C. and W.M.G. approved final version of manuscript.

ENDNOTE

At the request of the authors, readers are herein alerted to the fact that additional materials related to this manuscript may be found at https://doi.org/10.7924/r47s7t64c. These materials are not a part of this manuscript and have not undergone peer review by the American Physiological Society (APS). APS and the journal editors take no responsibility for these materials, for the website address, or for any links to or from it.