Patterned low-frequency deep brain stimulation induces motor deficits and modulates cortex-basal ganglia neural activity in healthy rats

Abstract

Deep brain stimulation (DBS) is an effective therapy for movement disorders, including Parkinson’s disease (PD), although the mechanisms of action remain unclear. Abnormal oscillatory neural activity is correlated with motor symptoms, and pharmacological or DBS treatment that alleviates motor symptoms appears to suppress abnormal oscillations. However, whether such oscillatory activity is causal of motor deficits such as tremor remains unclear. Our goal was to generate abnormal oscillatory activity in the cortex-basal ganglia loop using patterned subthalamic nucleus DBS and to quantify motor behavior in awake healthy rats. Stimulation patterns were designed via model-based optimization to increase power in the low-frequency (7–11 Hz) band because these oscillations are associated with the emergence of motor symptoms in the 6-hydroxydopamine lesioned rat model of parkinsonism. We measured motor activity using a head-mounted accelerometer, as well as quantified neural activity in cortex and globus pallidus (GP), in response to 5 stimulation patterns that generated a range of 7- to 11-Hz spectral power. Stimulation patterns induced oscillatory activity in the low-frequency band in the cortex and GP and caused tremor, whereas control patterns and regular 50-Hz DBS did not generate any such effects. Neural and motor-evoked responses observed during stimulation were synchronous and time-locked to stimulation bursts within the patterns. These results identified elements of irregular patterns of stimulation that were correlated with tremor and tremor-related neural activity in the cortex and basal ganglia and may lead to the identification of the oscillatory activity and structures associated with the generation of tremor activity.

NEW & NOTEWORTHY Subthalamic nucleus deep brain stimulation is a promising therapy for movement disorders such as Parkinson’s disease. Several groups reported correlation between suppression of abnormal oscillatory activity in the cortex-basal ganglia and motor symptoms, but it remains unclear whether such oscillations play a causal role in the emergence of motor symptoms. We demonstrate generation of tremor and pathological oscillatory activity in otherwise healthy rats by stimulation with patterns that produced increases in low-frequency oscillatory activity.

INTRODUCTION

Increased synchronized oscillatory activity in the cortex-basal ganglia-thalamus network is associated with motor symptoms in movement disorders, including Parkinson’s disease (PD) and essential tremor (ET). In PD, the magnitude of exaggerated beta-band activity in the basal ganglia and cortex (Bronte-Stewart et al. 2009; Brown 2007; Eusebio et al. 2011; Kühn et al. 2006; Silberstein et al. 2005; Weinberger et al. 2006) is correlated with motor symptoms such as bradykinesia and akinesia (Little and Brown 2012; Ray et al. 2008). Furthermore, this oscillatory activity is suppressed by pharmacological treatment and deep brain stimulation (DBS) (Kühn et al. 2008; Wingeier et al. 2006), and this suppression is correlated with improvement in PD motor symptoms such as bradykinesia/rigidity (Ray et al. 2008). Although nonlinear causality analysis has provided some evidence that abnormal oscillations cause parkinsonian tremor (Tass et al. 2010), it remains unclear to what degree the emergence of increased oscillatory activity is causal of motor symptoms.

In the unilateral 6-hydroxydopamine (6-OHDA) lesioned rat model of PD, low-frequency oscillations are exacerbated in the cortex-basal ganglia network (range 7–10 Hz) (Ge et al. 2012). Like beta oscillations in patients with PD, these low-frequency oscillations in the parkinsonian rat are associated with the development of motor symptoms, are particularly synchronous with head and neck tremor (Buonamici et al. 1986), and are suppressed by high-frequency DBS that alleviates motor symptoms [methamphetamine-induced circling and haloperidol-induced dyskinesia (McConnell et al. 2012) and locomotor activity (Yang et al. 2015)]. In addition, random or low-frequency DBS that is ineffective at relieving symptoms also fails to suppress low-frequency oscillatory activity (McConnell et al. 2016). Finally, a study in 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine-treated nonhuman primates also reported a correlation between suppression of low-frequency oscillations in the theta (4–7 Hz) and alpha (9–15 Hz) bands and improvements in motor symptoms such as akinesia during 130-Hz DBS (Rosin et al. 2011). Although these studies show correlation between pathological oscillatory activity and motor symptoms, it remains unclear whether such oscillations play a causal role in the emergence of motor symptoms.

The goal of the present study was to use DBS to generate oscillatory neural activity within the cortex-basal ganglia pathway and determine the motor symptoms induced in healthy rats. We used model-based optimization to design temporal patterns of DBS that generated low-frequency (7–11 Hz) oscillatory activity in the basal ganglia. We measured tremor in healthy awake rats during subthalamic nucleus (STN) DBS with these patterns, because low-frequency oscillations are associated with head and neck tremor in the parkinsonian rat (Buonamici et al. 1986) as well as tremor in persons with PD (Contarino et al. 2012; Reck et al. 2010; Timmermann and Fink 2011). We recorded neural activity from the motor cortex, because motor cortical activity differs strongly between effective and ineffective DBS (Li et al. 2007, 2012). We also recorded local field potentials from the globus pallidus (GP), because there is evidence that dopamine depletion in the 6-OHDA rat increases the power and coherence of pathological low-frequency oscillations in the GP and motor cortex of freely moving rats (Ge et al. 2012; Yang et al. 2015), and prior work has shown similar effects of STN DBS on neuronal activity in the GP and substantia nigra (SNr) of 6-OHDA rats (McConnell et al. 2012). Patterned low-frequency stimulation generated tremor and evoked large-amplitude potentials in the motor cortex, time-locked to high-frequency bursts within the stimulation pattern. These evoked potentials resembled the spike wave discharges seen in the freely moving 6-OHDA parkinsonian rat (Dejean et al. 2007). We conclude that high-frequency bursts in low-frequency patterned stimulation generated motor and neural activity in normal rats reminiscent of those seen in a rat model of parkinsonism.

METHODS

We delivered STN DBS to generate oscillatory neural activity along the cortex-basal ganglia pathway in healthy rats and quantified the effects on tremor and on neural activity in the cortex and globus pallidus.

Model-Based Design of Temporal Pattern of Stimulation

We coupled a biophysical network model of the basal ganglia to a genetic algorithm to design the temporal patterns of stimulation (Brocker et al. 2017). DBS with frequencies <50 Hz is ineffective or worsens motor symptoms (Birdno and Grill 2008), and we therefore used stimulation patterns with an average frequency of 50 Hz. The network model of the healthy basal ganglia (Brocker et al. 2017; Rubin and Terman 2004; So et al. 2012a) included the STN and the globus pallidus externus (GPe) and internus (GPi), with 10 Hodgkin-Huxley-style single-compartment neurons in each nucleus, as well as input from striatal neurons. The model neurons exhibited asynchronous firing consistent with in vivo observations in healthy rats. To represent DBS, intracellular current injection of a monophasic rectangular pulse with a current density of 300 µA/cm² and pulse width of 0.3 ms was delivered to each model STN neuron. The representation of DBS is simplified to only consider direct activation of STN neurons and does not consider activation of other neural elements, including axons of passage or presynaptic terminating axons, both of which may contribute to the symptom-relieving effects of STN DBS in PD (Gradinaru et al. 2009; Grill et al. 2008). Neural signals (spike times) were recorded from model GPi neurons, and multitaper spectral estimates of the spike times in the 7- to 11-Hz band were obtained using the Chronux neural signal analysis package (http://chronux.org/) and MATLAB. Model simulations were implemented in MATLAB using the forward Euler method with a time step of 0.01 ms and a total simulated time of 10 s.

The optimization technique used to design the temporal pattern of DBS was similar to the genetic algorithm reported previously (Brocker et al. 2017), with the exception that we used a cost function to design temporal patterns that either maximized (“GA MAX”) or minimized (“GA MIN”) the spectral power in the 7- to 11-Hz band of GPi model neuron spike times. Briefly, patterns of stimulation were encoded using bit strings (1-ms bins) with a value of 1 or 0 based on whether a pulse was present or not in a particular bin. Bit strings comprised 600 elements, and therefore each pattern was 600 ms long. We started with a random initial population of patterns, and at each generation, patterns were delivered to the model STN neurons and the GPi spectral power in the range 7–11 Hz was computed. The cost function was calculated as

$$\text{Total\hspace{0.17em}cost}=\frac{\alpha }{\text{Mean\hspace{0.17em}Max\hspace{0.17em}HVS}}+\left(\frac{1}{\beta }\right)\cdot {(F-F_T)}^2,$$

where F_T = 50 Hz, F is the frequency of DBS pattern being tested, mean max HVS is the resulting power in the 7- to 11-Hz band, and α = 100 and β = 10 are weighting factors selected such that the magnitudes of the addends were similar. This function was then used to rank the fitness of each pattern in each generation, and repopulation was used to create the next generation of patterns. Of the 150 patterns in each subsequent generation, 130 patterns were offspring obtained by combining “parent” patterns from the prior generation (see below), 10 patterns were new randomly generated patterns (immigrants) incorporated into the population to add genetic diversity, and 10 were the most fit patterns from the previous generation (elites) carried forward to ensure preservation of highly fit patterns. Selective pressure toward more fit patterns was exerted using a roulette wheel parent selection process, and one-point crossover was used to exchange genetic material between the parents and generate two offspring patterns. To mimic genetic mutation, 0.1% of the binary string elements in the offspring patterns were randomly switched. The genetic algorithm was run for 100 generations, after which the cost function had converged and the most fit pattern was selected for in vivo testing.

Experimental Measurements

Experiments were conducted in 15 adult female rats, including 10 Long-Evans and 5 Sprague-Dawley strains. We collected data from one hemisphere in each rat, except in one Sprague-Dawley rat where both hemispheres were tested independently. All surgical and experimental procedures were approved by the Duke University Institutional Animal Care and Use Committee.

Surgical procedure.

Animals were implanted unilaterally with stimulating microelectrodes in the STN and recording microelectrodes in the ipsilateral GPe, and bilaterally with stainless steel screws over the motor cortex. In one rat, we implanted stimulating microelectrodes bilaterally and did not implant GPe electrodes. Surgical procedures were similar to those described previously in detail (Brocker et al. 2017; McConnell et al. 2016). Briefly, rats were anesthetized using 1–3% isoflurane or 3–4% sevoflurane, and stereotactic surgery was performed using aseptic technique. First, a craniotomy was made based on atlas-based coordinates referencing STN to bregma. Next, we positioned a single microelectrode (0.5-MΩ tungsten microelectrode; Microprobes, Gaithersburg, MD) in the stereotactic manipulator for intraoperative recordings where we used real-time display of the recorded activity to search for a pattern of high-frequency neural activity consistent with STN. We then removed the single electrode and positioned the 2 × 2 array in the manipulator. We aligned the array such that the long axis of the electrodes was perpendicular to the brain surface. The low-impedance (10 kΩ) platinum-iridium stimulation electrode arrays (2 × 2 75-µm electrodes with interelectrode spacing of 300 µm; MicroProbes) were implanted into the STN using stereotactic coordinates (anterior −3.6 mm, lateral 2.6 mm, ventral −6.8 to −7.2 mm, relative to bregma; Paxinos and Watson 2007). Next, similar microelectrodes (i.e., 10-kΩ platinum-iridium 2 × 2 stimulation electrode arrays) were implanted into the GPe (ipsilateral to the STN stimulating electrodes) using stereotactic coordinates (anterior −1.0 mm, lateral 3.0 mm, ventral −5.2 mm, relative to bregma). Stainless steel screws (1-mm diameter) were inserted through the skull to abut the dura on the ipsilateral and contralateral motor cortex (anterior 2.5 mm, lateral ±2.5 mm, relative to bregma) for electrocorticographic recording. Finally, a cannula was implanted into the medial forebrain bundle (MFB; anterior 2.0 mm, lateral 2.0 mm, ventral −7.5 mm, relative to bregma) for later administration of 6-OHDA, because these rats were part of a cohort also used in other experiments to investigate the effects of STN DBS in the parkinsonian condition. All implanted electrodes and the cannula were secured by using dental acrylic to additional stainless steel screws in the skull. Postoperatively, rats were monitored closely and recovered for at least 1 wk before any experimental testing began.

Experiment setup.

All behavioral and neural recording experiments were conducted in awake and freely moving rats placed inside a Faraday cage. Each 1 min of stimulation epoch was followed by a 1-min postbaseline epoch and preceded by a 1-min prebaseline epoch. Thus epochs of stimulation were 2 min apart. On the basis of our prior experience in preclinical DBS experiments (McConnellet al. 2012, 2016), we used 1-min stimulation trials because they provided sufficiently long recordings to assess variability in responses within and across trials. Five stimulation patterns (Fig. 1C) were applied in a random order, and trials were repeated for each rat across multiple days (typically 2–5 different sessions). Our stimulation analysis did not distinguish between rats moving vs. stationary, but we averaged across multiple trials (conducted across different days) to capture such variability.

Fig. 1.

Model-based design of temporal pattern of deep brain stimulation (DBS) using a genetic algorithm. A: computational model of the healthy basal ganglia included globus pallidus externus (GPe), globus pallidus internus (GPi), subthalamic nucleus (STN), and striatum. Applied current (I_app) indicates input to the GPe and GPi from striatal neurons expressing inhibitory D2-type receptors and excitatory D1-type receptors, respectively. B: power in the 7- to 11-Hz frequency band of the GPi spike times from the model generated by different patterns of STN DBS. The GA MAX pattern maximized power with an average stimulation frequency of 50 Hz. C: stimulation patterns designed by the genetic algorithm that maximized (GA MAX) or minimized (GA MIN) power in the 7- to 11-Hz frequency band. The GA MAX and GA MIN patterns were shuffled such that power in the 7- to 11-Hz band was significantly reduced (shuffled GA MAX) or increased (shuffled GA MIN), respectively. D: sum of spectral power in the 7- to 11-Hz band in the GPi in the computational model for each pattern.

Stimulation settings.

Stimulation to the STN comprised charge-balanced, symmetric biphasic pulses with a pulse width of 90 µs. Before any experimental testing began, all possible bipolar electrode stimulating pairs were tested across a range of stimulation amplitudes (typically 10–100 µA) during 130-Hz STN stimulation while rats were in a circular recording chamber. The amplitude and contact configuration that elicited sustained motor responses (typically contralateral turning and increased motor activity) and lack of side effects (typically involuntary muscle contractions of the limb and neck) were identified for each rat. This bipolar configuration and amplitude, unique for each rat, was then used for the entire study. Stimulation amplitudes ranged between 25 and 70 µA for all the rats used in the study, and each stimulation epoch included an amplitude ramp-up over 5 s. These amplitudes were similar to those used in the 6-OHDA rat model of PD to alleviate motor methamphetamine-induced circling (So et al. 2012b) and akinesia (Brocker et al. 2017). In a subset of rats (n = 5), a range of stimulation currents was used to test the effects of stimulation intensity on tremor and neural responses. Stimulation patterns were generated using custom scripts (MATLAB) and applied through an isolated voltage-to-current converter (model 2200 analog stimulus isolator; A-M Systems) and a custom alternating current coupler.

Behavioral and neural recording setup.

A triaxial accelerometer (model ADXL330; Analog Devices; sensitivity 300 mV/g) was attached to the head cap of the rat to quantify tremulous head motion. A similar set up (i.e., accelerometer connected to the head cap) has been used in other studies to assess motor activity in freely behaving rodents (Long and Carmena 2013; Venkatraman et al. 2010). Stimulating and recording cables were connected to the external connectors in the head cap, and rats were placed in the Faraday cage. Field potentials and accelerometer signals were recorded using a multichannel acquisition processor system (MAP; Plexon, Dallas, TX). Signals were bandpass filtered (0.7 Hz–2 kHz, 2 and 4 poles, respectively), and field potentials were amplified 5,000 times before sampling at 20 kHz. All recordings were referenced to titanium screws inserted through the skull over the cerebellum. All recordings were ac coupled, eliminating any offset voltage from the recorded signal. All experimental sessions were video recorded using a digital camera (Sony DCR-SR80) for offline analysis.

Data Analysis

Analysis of motor behavior.

Motor behavior was quantified using the triaxial accelerometer and the recorded video. The accelerometer signals from each of the three axes were bandpass filtered (0.5–45 Hz), and power spectra for all three axes were calculated using the multitaper approach in the Chronux analysis toolbox (http://chronux.org/) in MATLAB. Although the patterns of tremor were reflected in all three axes of the accelerometer, we used the spectral power for the z-axis in our analysis because it was most sensitive to stimulation-induced movement. We summed the total power in the 6- to 12-Hz range to capture the primary and first harmonic of the tremor and to exclude steady-state acceleration due to gravity, and power was expressed as a percentage of total power over the 0.5- to 45-Hz range. Several studies reported that low-frequency oscillations in the 5- to 13-Hz range are exacerbated in the cortex-basal ganglia network in the 6-OHDA lesioned rats (Dejean et al. 2007; Ge et al. 2012; Yang et al. 2015), with most power seen in the 6- to 12-Hz band. The bandwidth of oscillatory activity was larger in experimental recordings than in the model neuron firing activity, and therefore we used a larger band (6–12 Hz) for quantification of the experimental data. Furthermore, these oscillations are synchronized with head/neck tremor in the 6-OHDA rat in the same frequency band (Buonamici et al. 1986). Comparisons were made between baseline and stimulation epochs as well as between stimulation epochs with different patterns.

We conducted time domain analysis of the accelerometer data by averaging the z-axis signal across multiple repeats of the stimulation pattern (i.e., 600-ms epochs during each stimulation trial as well as trials repeated across days). We did not include the first 5 s of the trial because the stimulation amplitude ramped up over 5 s, and therefore this pattern-triggered averaging included ~55 s of data per trial. Responses during 50-Hz stimulation were also analyzed in a similar manner (i.e., the stimulation epoch was analyzed in 600-ms segments). To quantify further these responses, we rectified and integrated the time series data from each axis for every 600-ms epoch during stimulation (i.e., area under the curve, AUC), averaged across trials, and compared across different stimulation patterns to evaluate the magnitude of head acceleration during stimulation. We also examined rats’ motor behavior from the recorded video for any behaviors, including stimulation-induced circling, and increased rearing or involuntary contractions of the limbs, which were rarely seen with our choice of stimulation amplitudes (25–70 µA). Our focus was exclusively on measurements of tremor, and we did not quantify any motor behavior from the recorded video.

Analysis of neural data.

Our preamplifier settings (high cut 2 kHz and sampling frequency of 20 kHz) allowed neural recordings in the presence of stimulation artifact (see Fig. 3). To confirm our ability to distinguish neural responses from stimulation artifacts, we conducted three additional experiments in a subset of rats: 1) neural recordings with stimulation polarity reversed during awake conditions (n = 6), 2) neural recordings during stimulation while rats were under isoflurane (0.5–1%) or sevoflurane (2–3%) anesthesia (n = 6), and 3) neural recordings with stimulation polarity reversed during anesthesia (n = 2). We bandpass filtered the neural signal in the 0.5- to 45-Hz band for further analysis and computed the power spectrum (0.5–45 Hz, with the multitaper approach using the Chronux toolbox) during awake conditions and under anesthesia for ipsilateral and contralateral motor cortex and GP field potentials. Similarly to the accelerometer signal analysis, we rectified and integrated the field potentials for each 600-ms epoch during stimulation and averaged across trials for different stimulation patterns. Coherence between accelerometer and cortical field potentials was computed using the Chronux toolbox within the low-frequency band of interest and compared across stimulation patterns.

Fig. 3.

Evoked potentials from the ipsilateral motor cortex in response to different temporal patterns of subthalamic nucleus deep brain stimulation with different stimulation polarities in an awake healthy rat. A: pattern-triggered evoked responses across different stimulation patterns (polarity: +/−) inclusive of the stimulation artifact (inset shows the short- and long-latency evoked response along with stimulation artifact). B: pattern-triggered evoked responses across different stimulation patterns with reversed polarity (i.e., −/+) inclusive of the stimulation artifact. C: polarity-averaged evoked responses across different stimulation patterns clearly showing the preservation of stimulation-induced short- and long-latency evoked activity and abolition of stimulation artifact. GA MAX and GA MIN, stimulation pattern designed by the genetic algorithm that respectively maximized and minimized power in the 7- to 11-Hz frequency band.

Histology

Rats were deeply anesthetized with pentobarbital sodium (100 mg/kg ip) and euthanized via intracardiac perfusion with 10% formalin. Brains were extracted and postfixed overnight at 4°C. Electrodes were removed, and the brain was placed in 30% sucrose solution for 2 days and then sectioned coronally at 50-µm thickness. Sections were mounted and stained with cresyl violet and cytochrome oxidase to identify the STN and locate the electrode tracks. We confirmed correct electrode locations in 10 of 15 rats. Of the remaining 5 rats, 3 rats were excluded from subsequent analysis due to unavailable or inconclusive histological evidence of electrode positioning. Two of these 5 rats were included in all analyses, even though we lacked histological samples of electrode positions, because the data from these rats were consistent with the larger cohort of 10 rats with histological confirmation (i.e., these 2 rats exhibited high spectral power in the low-frequency band for tremor and qualitatively similar cortical evoked responses during different stimulation patterns, and their AUC analyses for cortex and accelerometer signals were within the range of data for the larger cohort). Thus our analysis included 12 rats and, since 1 rat had bilateral implants, n = 13 hemispheres (see Table 1). Two of these 12 rats were also used in the study by Brocker et al. (2017) following 6-OHDA lesion, which was performed after all recordings had been performed for the current study.

Table 1.

Summary of animals used for the study

Rat ID	Species	Included in Analysis	STN Histological Confirmation	CTX Recordings	GP Recordings	Testing with Reverse Stimulation Polarity	Testing with 10 Stimulation Amplitudes
1	LE	Yes	Confirmed	Yes	Yes		Yes
2	LE	Yes	Confirmed	Yes			Yes
3 (bilateral)	SD	Yes	Confirmed	Yes
4	SD	No	No histology available
5	SD	Yes	No histology available	Yes	Yes	Yes
6	SD	Yes	Confirmed	Yes	Yes	Yes
7	SD	Yes	Confirmed	Yes	Yes	Yes
8	LE	No	Not placed in STN
9	LE	Yes	Confirmed	Yes	Yes
10	LE	Yes	No histology available	Yes	Yes
11	LE	Yes	Confirmed	Yes	Yes	Yes
12	LE	Yes	Confirmed	Yes	Yes	Yes
13	LE	No	Not placed in STN
14	LE	Yes	Confirmed	Yes	Yes	Yes
15	LE	Yes	Confirmed	Yes	Yes

Experiments were conducted in 15 adult female rats, including 10 Long-Evans (LE) and 5 Sprague-Dawley (SD) rats, with data collected from 1 hemisphere in each rat except in 1 SD rat (rat 3) where both hemispheres were tested independently. Subthalamic nucleus (STN) electrode locations were confirmed in 10 of 15 rats. Of the remaining 5 rats, 3 rats were excluded due to unavailable or inconclusive histological evidence of electrode positioning.

Statistical Analysis

To examine statistical differences in motor and neural responses across five stimulation patterns, we used repeated-measures analysis of variance (RM-ANOVA) with Fisher’s protected least significant difference (FPLSD) post hoc comparisons between stimulation patterns. For all statistical tests, a P value <0.05 was considered significant, and all data are means ± SE. All data analysis was done using custom scripts in MATLAB R2013b or R2015a.

RESULTS

We used model-based optimization to design temporal patterns of STN DBS that either maximized or minimized low-frequency oscillatory activity in a biophysically based network model of the basal ganglia. We then applied these patterns, as well as control patterns, including 50-Hz regular stimulation and shuffled versions of the model-designed patterns where the individual interpulse intervals were randomly shuffled, to awake and behaving healthy rats (n = 13 hemispheres) and measured tremor and local field potentials in motor cortex and GP.

Model-Based Design of Temporal Patterns Of Stimulation Using a Genetic Algorithm

A computational model of the healthy basal ganglia was coupled with a genetic algorithm (GA) to design stimulation patterns. The GA used a cost function that maximized the spectral power in the 7- to 11-Hz band in the firing times of model globus pallidus internus (GPi) neurons while maintaining an average pulse repetition frequency of 50 Hz, and the cost of the best stimulation pattern declined monotonically over 100 generations. The resulting stimulation pattern (termed GA MAX) included three high-frequency bursts resulting in a high coefficient of variation of 1.64, had an average frequency of 50 Hz, and increased the GPi 7- to 11-Hz spectral power by ~12.5 times compared with regular 50-Hz stimulation (Fig. 1). Subsequently, we revised the cost function to minimize the GPi spectral power in the 7- to 11-Hz band while maintaining an average pulse repetition frequency of 50 Hz. The resulting stimulation pattern (GA MIN) generated 7- to 11-Hz power similar to that generated by regular 50 Hz and was only ~8.5% of the power generated by the GA MAX pattern and had a lower coefficient of variation of 0.46. Finally, we shuffled the interpulse intervals (IPIs) from the GA MAX to create a pattern that resulted in low power in the computational model (shuffled GA MAX) to gain a better understanding of whether the specific order of the IPIs in the stimulation patterns played a role in causing tremor. Similarly, we shuffled the IPIs from the GA MIN to create a pattern that resulted in high power in the computational model (shuffled GA MIN). The spectral power generated by the shuffled patterns was increased (shuffled GA MIN) or decreased (shuffled GA MAX) by a factor of ~5 compared with the spectral power generated by the unshuffled patterns.

Effect of STN DBS on Tremor

We observed that STN DBS with the GA MAX and shuffled GA MAX patterns induced tremor in awake healthy rats characterized by repetitive and periodic motion of the head that was time-locked to stimulation. We quantified tremor using an accelerometer connected to the head cap and analyzed the z-axis (approximately vertical) signal in the frequency and time domain across the five stimulation patterns.

We summed the power spectral density (PSD) of the accelerometer signal between 6 and 12 Hz and expressed it as a percentage of total power between 0.5 and 45 Hz to quantify the activity in the low-frequency tremor band (Fig. 2A). There was a significant effect of DBS pattern on tremor power, and tremor during DBS with the GA MAX and shuffled GA MAX patterns was significantly higher than during DBS with the GA MIN, shuffled GA MIN, and regular 50-Hz patterns (P = 3 × 10⁻⁴, RM-ANOVA with FPLSD post hoc comparisons between stimulation patterns; Fig. 2B). Importantly, we did not observe any significant differences between tremor during the DBS OFF periods preceding delivery of the different stimulation patterns (P = 0.9, RM-ANOVA). Tremor power during stimulation with GA MAX and shuffled GA MAX patterns was significantly different from tremor power during the DBS OFF period preceding delivery of these two different stimulation patterns (P < 0.006, paired t-test for each stimulation pattern). This was not the case with GA MIN, shuffled GA MIN, and 50-Hz stimulation, suggesting that stimulation did not alter normal rats’ motor behavior.

Fig. 2.

Acceleration measured with a head-mounted accelerometer in response to different temporal patterns of subthalamic nucleus deep brain stimulation in healthy rats. A: power spectrum of the accelerometer signal during stimulation. Spectral peaks were observed in the low-frequency tremor band during stimulation patterns designed by the genetic algorithm that maximized power in the 7- to 11-Hz frequency band (GA MAX) and by GA MAX patterns shuffled such that power was significantly reduced (Shuff. GA Max). Black line indicates the mean and gray band indicates ±SE (n = 13 hemispheres). B: summed accelerometer signal power between 6 and 12 Hz, expressed as a percentage of total accelerometer signal power between 0.5 and 45 Hz [P = 3 × 10⁻⁴, repeated-measures (RM) ANOVA]. Data are means ± SE; ^a,bDifferent letters indicate significant differences (P < 0.05) with Fisher’s protected least significant difference (PLSD) post hoc comparisons). C: area under the curve (AUC), calculated by rectifying and integrating the accelerometer signal for the duration of the stimulation pattern (i.e., 600 ms) and averaging across trials and rats (P = 6 × 10⁻⁶, RM-ANOVA). Data are means ± SE. ^a,bDifferent letters indicate significant differences (P < 0.05) with Fisher’s PLSD post hoc comparisons. GA MIN, stimulation pattern designed by the genetic algorithm that minimized power in the 7- to 11-Hz frequency band; Shuff. GA MIN, GA MIN patterns shuffled such that power was significantly increased.

In 5 of 13 hemispheres we delivered DBS at two additional stimulation amplitudes (i.e., I − 10 µA and I + 10 µA, where I is the unique stimulation amplitude used for each rat). For all stimulation patterns except GA MAX, there were no significant differences in tremor power during DBS with I − 10, I, and I + 10 µA (one-way RM-ANOVA on amplitude for each pattern; P = 0.86, 50 Hz; P = 0.58, GA MIN; P = 0.54, shuffled GA MAX; P = 0.73, shuffled GA MIN). For GA MAX, there was a significant difference between I − 10 and I + 10 µA stimulation conditions (P = 0.02, RM-ANOVA with FPLSD post hoc comparisons), suggesting an increase in tremor intensity with an increase in stimulation amplitude. To investigate this further, in two rats we measured tremor during GA MAX stimulation at 10 different amplitudes (i.e., from 10 to 100 µA in steps of 10 µA) and 6 different bipolar electrode configurations. All 6 bipolar contact configurations induced tremor with stimulation amplitudes of I ± 20 µA, and tremor intensity typically increased up to the highest stimulation amplitude tested (100 µA). We also tested other GA patterns at 100 µA to examine whether otherwise nontremulous patterns (GA MIN and shuffled GA MIN) caused tremor at higher stimulation amplitude. Stimulation at 100 µA with shuffled GA MAX induced tremor, but the same stimulation amplitude with GA MIN and shuffled GA MIN patterns did not induce tremor in the same two rats.

The accelerometer signal was pattern-triggered averaged (i.e., time series data across repetitions of the stimulation pattern were averaged) and the AUC was used to quantify behavioral responses time-locked to the stimulation pattern (Fig. 2C). The AUC was significantly larger in response to GA MAX and shuffled GA MAX than in response to GA MIN, Shuffled GA MIN and regular 50-Hz DBS (P = 6 × 10⁻⁶, RM-ANOVA with FPLSD post hoc comparisons). Tremulous head acceleration was time-locked with stimulation bursts in the GA MAX and shuffled GA MAX patterns, and this was also reflected in the multiple peaks in the low-frequency band in the spectra (Fig. 2A). In summary, GA MAX and shuffled GA MAX patterns induced tremor in healthy, awake rats, and tremor was time-locked to high-frequency bursts within the stimulation patterns.

Neural Activity Evoked in Motor Cortex by STN DBS

We analyzed pattern-triggered cortical evoked potentials as well as PSD in the low-frequency band in response to different patterns of STN DBS. To distinguish cortical evoked responses from stimulation artifact, in a subset of six rats we reversed the stimulation polarity and averaged the evoked responses. The pattern-triggered averages of evoked ipsilateral motor cortex field potentials in Fig. 3, A and B, show that reversing the stimulation polarity inverted the polarity of the artifact but that the neural responses were largely unaffected. Figure 3C shows the polarity-averaged responses that eliminated the stimulation artifacts while preserving the evoked neural responses in the ipsilateral motor cortex under awake conditions. Furthermore, we recorded evoked potentials while rats were temporarily under anesthesia, and the polarity-averaged responses under anesthesia (Fig. 4A) were compared with the polarity-averaged responses under awake conditions (Fig. 3C). During awake conditions (Fig. 3), 50-Hz STN DBS evoked a short-latency positive response (“P1” response) followed by a low-amplitude medium-latency negative response (“N1” response). However, under anesthesia, the amplitude of the short-latency P1 response was substantially reduced and the N1 response was absent (Fig. 4A). Each stimulation pulse within the GA patterns triggered a short-latency P1 response (Fig. 3C). The N1 responses observed during regular 50-Hz stimulation were not present during GA patterns if there was another stimulation pulse that followed within the N1 response latency. The GA MAX pattern had three bursts that evoked a large-amplitude N1 response that was approximately four times greater than that caused by a single pulse (Fig. 3C), and these responses were not observed while the rat was under anesthesia (Fig. 4A). Under awake conditions, these responses were time-locked to the tremulous head motion during GA MAX stimulation and are reminiscent of the spike-wave discharges observed during low-frequency, high-voltage spindle oscillations in the 6-OHDA lesioned rat and thought to be associated with parkinsonism (Dejean et al. 2007; Ge et al. 2012; Yang et al. 2015). Cortical evoked responses during GA MIN were qualitatively similar (i.e., a short-latency P1, followed by medium-latency N1 response in the absence of any other stimulation pulse and a large N1 response if stimulation pulses were close to each other, e.g., the large N1 peak at ~495 ms in Fig. 3C) under awake conditions (Fig. 3C), and these responses were not observed under anesthesia (Fig. 4A).

Fig. 4.

Evoked potentials from the motor cortex in response to different temporal patterns of subthalamic nucleus deep brain stimulation in awake and anesthetized healthy rats. A: pattern-triggered polarity averaged evoked responses across different stimulation patterns while the rat was under sevoflurane anesthesia. Responses show cancellation of stimulation artifact, reduction in amplitude of short-latency responses, and absence of long-latency evoked responses. B: pattern-triggered polarity-averaged evoked responses in the contralateral motor cortex under awake conditions. C: polarity-averaged and bandpass filtered (0.5–45 Hz) evoked responses in the ipsilateral and contralateral motor cortex under awake and anesthetized conditions (iCTX, ipsilateral cortex; cCTX, contralateral cortex; AW, awake; AN, anesthesia). GA MAX and GA MIN, stimulation pattern designed by the genetic algorithm that respectively maximized and minimized power in the 7- to 11-Hz frequency band.

Next, we simultaneously recorded evoked potentials in contralateral motor cortex in awake rats and show the pattern-triggered and polarity-averaged responses in Fig. 4B. The responses that we highlighted earlier in the ipsilateral motor cortex under awake conditions (Fig. 3C) were not observed in the contralateral motor cortex (Fig. 4B). The lack of prominent spectral peaks under anesthesia for the ipsilateral motor cortical response and under awake conditions for the contralateral motor cortical response further confirmed elimination of stimulation artifact and short-latency (antidromic) evoked responses (data not shown). In summary, comparison of ipsilateral motor cortex evoked responses under awake and anesthetized conditions (Figs. 3C and 4A), comparison of ipsilateral and contralateral cortex evoked responses under awake conditions (Figs. 3C and 4B), and ipsilateral cortex evoked responses with different stimulation polarities (Fig. 3) allowed us to rule out stimulation artifacts and motion artifacts as causes of the evoked potentials.

We used bandpass filtering (0.5–45 Hz) to eliminate the high-frequency stimulation artifacts from the signal. This process also removed the short-latency evoked responses, because they contained high-frequency components. Filtering thus enabled clear visualization of the longer latency responses in the low-frequency band of interest. Examples of the filtered pattern-triggered averaged responses are shown in Fig. 4C for ipsilateral and contralateral motor cortex under awake and anesthetized conditions. Figure 4C clearly shows the large, long-latency negative responses observed in the ipsilateral motor cortex during GA MAX stimulation under awake conditions. These evoked responses for ipsilateral motor cortex for all rats under awake conditions (n = 13 hemispheres) are shown in Fig. 5 and illustrate that the responses were remarkably consistent across animals. Similar to GA MAX, the bursts in the shuffled GA MAX pattern caused large-amplitude long-latency responses. We calculated the AUC of the filtered pattern-triggered responses for all five patterns of stimulation (Fig. 5D), and GA MAX and shuffled GA MAX produced significantly greater evoked responses than the other three patterns (P = 2 × 10⁻⁶, RM-ANOVA with FPLSD post hoc comparisons). The AUC results did not change if we calculated the AUC of the unfiltered pattern-triggered averaged responses. Furthermore, in the five rats where DBS was delivered at two additional stimulation amplitudes (i.e., I + 10 and I − 10 µA), there were no significant differences between AUC for responses to regular 50-Hz stimulation (P = 0.83, RM-ANOVA). For the GA patterns, the AUC of the evoked potentials were larger with I + 10 µA than with I − 10 µA for all four patterns (one-way RM-ANOVA with FPLSD post hoc comparison on amplitude for each pattern; P = 0.01, GA MAX; P = 0.03, GA MIN; P = 7 × 10⁻⁴, shuffled GA MAX; P = 9 × 10⁻⁴, shuffled GA MIN).

Fig. 5.

Evoked potentials from the ipsilateral motor cortex in response to different temporal patterns of subthalamic nucleus deep brain stimulation in awake healthy rats. A: pattern-triggered bandpass filtered (0.5–45 Hz) evoked responses across all rats. Each line represents the average response per hemisphere. B: power spectra of the evoked potentials. Black line indicates the mean and gray band indicates ±SE. C: summed signal power between 6 and 12 Hz, expressed as a percentage of total signal power between 0.5 and 45 Hz [P = 1.5 × 10⁻⁵, repeated-measures (RM) ANOVA]. Data are means ± SE. ^a,bDifferent letters indicate significant differences (P < 0.05) with Fisher’s protected least significant difference (PLSD) post hoc comparisons. D: area under the curve (AUC), calculated by rectifying and integrating the evoked potentials for the duration of the stimulation pattern (i.e., 600 ms) and averaged across all trials and rats (P = 2 × 10⁻⁶, RM-ANOVA). Data are means ± SE. ^a,b,cDifferent letters indicate significant differences (P < 0.05) with Fisher’s PLSD post hoc comparisons. GA MAX and GA MIN, stimulation pattern designed by the genetic algorithm that respectively maximized and minimized power in the 7- to 11-Hz frequency band; Shuff. GA MAX and GA MIN, GA MAX and GA MIN patterns shuffled such that power in the 7- to 11-Hz band was significantly reduced or increased, respectively.

As in the analysis of tremor, we computed the PSD of the evoked potential between 6 and 12 Hz as a percentage of total power between 0.5 and 45 Hz. We observed large spectral peaks in the responses to GA MAX and shuffled GA MAX that were associated with the burst-induced evoked potentials (Fig. 5B). The power in the low-frequency band was greater for the GA MAX and shuffled GA MAX patterns than for the GA MIN patterns and 50-Hz DBS (P = 1.5 × 10⁻⁵, RM-ANOVA with FPLSD post hoc comparisons; Fig. 5C). Importantly, there were no differences in the power in the low-frequency band between patterns during the DBS OFF epochs preceding each stimulation trial (P = 0.4, RM-ANOVA with FPLSD post hoc comparisons). The large cortical evoked potentials and tremulous head motion observed during GA MAX and shuffled GA MAX patterns were time-locked to the stimulation bursts within these patterns, and therefore there was high coherence between the evoked potential and accelerometer signals in the low-frequency band (Fig. 6). In summary, both the GA MAX and shuffled GA MAX stimulation patterns induced neural activity in the low-frequency band in the ipsilateral motor cortex in healthy rats, whereas in the computational model only the GA MAX pattern generated high spectral power in the low-frequency band in model GPi neuron firing times.

Fig. 6.

Coherence between evoked potentials in the ipsilateral motor cortex and tremor (accelerometer signal) during different temporal patterns of subthalamic nucleus deep brain stimulation. Black line indicates the mean and he gray band indicates ±SE. GA MAX and GA MIN, stimulation pattern designed by the genetic algorithm that respectively maximized and minimized power in the 7- to 11-Hz frequency band; shuffled GA MAX and GA MIN, GA MAX and GA MIN patterns shuffled such that power in the 7- to 11-Hz band was significantly reduced or increased, respectively.

Neural Activity Evoked in Globus Pallidus by STN DBS

We also recorded evoked potentials in the GP in 10 of 13 hemispheres. Similar to responses in the motor cortex, the power in the low-frequency band was greater for the GA MAX and shuffled GA MAX patterns than for the GA MIN patterns and 50-Hz DBS (P = 0.02, RM-ANOVA with FPLSD post hoc comparisons; Fig. 7).

Fig. 7.

Spectral analysis of evoked potentials in the globus pallidus during different temporal pattern of deep brain stimulation. Summed signal power between 6 and 12 Hz, expressed as a percentage of total signal power between 0.5 and 45 Hz. Stimulation patterns designed by the genetic algorithm that maximized power in the 7- to 11-Hz frequency band (GA MAX) and GA MAX patterns shuffled such that power was significantly reduced (Shuff. GA MAX) were significantly different from the other 3 patterns (P = 0.02, repeated-measures ANOVA with Fisher’s protected least significant difference post hoc comparisons). Data are means ± SE. ^a,bDifferent letters indicate significant differences (P < 0.05). GA MIN, stimulation pattern designed by the genetic algorithm that minimized power in the 7- to 11-Hz frequency band; Shuff. GA MIN, GA MIN patterns shuffled such that power was significantly increased.

DISCUSSION

We used model-based optimization with a genetic algorithm to design temporal patterns of stimulation intended to generate low-frequency oscillatory neural activity in the basal ganglia of healthy rats, as has been observed in the 6-OHDA lesioned rat model of PD. The model-derived GA MAX pattern included high-frequency bursts and was more irregular than the GA MIN pattern and regular 50-Hz stimulation. In the computational model we focused on spectral power of the model GPi neurons, which differed across patterns, and during experiments we measured head acceleration and neural activity in ipsilateral motor cortex and GP, which showed strong agreement across the different patterns of stimulation. STN DBS with the GA MAX and shuffled GA MAX pattern in healthy rats induced oscillatory activity in the low-frequency band in the cortex and GP and generated tremor in all rats tested, whereas neither GA MIN, shuffled GA MIN, nor regular 50-Hz stimulation generated any such effects. The evoked neural and motor responses to the GA MAX and shuffled GA MAX pattern were synchronous and time-locked to stimulation bursts.

Previous studies demonstrated that the temporal pattern of stimulation can determine the effectiveness of DBS for symptom suppression in patients with ET or PD (Birdno et al. 2008, 2012; Brocker et al. 2017; Dorval et al. 2010), and the present study shows the utility of specific patterns to interrogate the relationship between patterns of neural activity and the symptoms of movement disorders. The generation of motor symptoms by low-frequency stimulation is variable across location of stimulation electrodes and individuals (Barnikol et al. 2008; Constantoyannis et al. 2004). Low-frequency (<50 Hz) DBS of the ventral posteromedial nucleus of the thalamus in patients with complex regional facial pain induced de novo tremor that disappeared with higher stimulation frequencies (Constantoyannis et al. 2004), whereas low-frequency STN DBS produced only modest increases in the symptoms of PD (Chen et al. 2011; Eusebio et al. 2008; Timmermann et al. 2004). Our results are consistent with those of Barnikol et al. (2008), who observed tremor entrainment by patterned low-frequency stimulation in a patient with spinocerebellar ataxia type 2 that disappeared under general anesthesia. Neural synchronization may be near maximal in the PD state and not enhanced by regular low-frequency STN DBS but can potentially be further exacerbated using a “symptogenic pattern” designed by computational modeling.

There is evidence that low-frequency oscillations in cortex and basal ganglia are associated with tremulous head movement in the PD rat (Buonamici et al. 1986), and we generated tremor using optimized stimulation patterns in healthy rats. It is unclear why the GA MAX and shuffled GA MAX stimulation patterns induced primarily head tremor, as opposed to forelimb tremor. Although we did observe tremulous motion in the contralateral forelimb in some rats, we did not have a means to quantify this movement, because it was obscured during many of the frames of the video recordings. Similarly, cholinomimetic drugs induce tremulous jaw movements in healthy rats (Cousins et al. 1998; Long et al. 2016; Salamone et al. 1998), and tremor was synchronous with the local field potentials recorded from the motor cortex. The GA MAX and shuffled GA MAX patterns that caused tremor and induced oscillatory activity in the low-frequency band contained high-frequency bursts, suggesting that bursting may lead to abnormal neural activity. Indeed, cells in the basal ganglia (Rubin et al. 2012) and cortex (Li et al. 2012) exhibit burst firing in the parkinsonian state, and effective DBS reduces this abnormal burst activity. In addition, burst discharges in the STN are synchronized with tremor-related electromyographic activity (Rodriguez et al. 1998), and DBS at a frequency similar to neurons’ preferred burst frequency evoked tremor that was phase-locked to the stimulus (Barnikol et al. 2008). Similarly, we observed that the evoked cortical activity and tremor motion were time-locked to bursts in the stimulation patterns. Finally, the low-frequency oscillatory activity observed in the 7- to 11-Hz band in parkinsonian rats is also associated with burst discharges in the basal ganglia (Paz et al. 2005). However, it should be noted that intrinsically generated bursts in the cortex and basal ganglia in the parkinsonian state have a lower intraburst frequency than the bursts in our stimulation pattern. Finally, Ma and Wichmann 2004 showed that DBS with oscillatory burst patterns resulted in reduced performance in a delayed response elbow movement task, and stimulating with the same average frequency but with more regular interstimulus intervals eliminated the reduction in performance. This is consistent with our current results with GA MAX, GA MIN, and regular 50-Hz stimulation patterns, all of which had the same average frequency but produced significantly different motor responses. It remains to be determined which elements of STN burst stimulation (i.e., interburst frequency, intraburst frequency) are most critical to the effects that we observed in the present study.

STN DBS evokes responses in the ipsilateral motor cortex at a short latency followed by a medium-to-long-latency response depending on the frequency of stimulation (Johnson et al. 2015; Santaniello et al. 2010; Walker et al. 2012). The cortical evoked responses that we observed in response to 50-Hz DBS are consistent with these prior results (MacKinnon et al. 2005; Zwartjes et al. 2013). In addition, we observed that high-frequency bursts in the stimulation patterns also evoked a large, longer latency response in the ipsilateral cortex that was not observed in the contralateral motor cortex and disappeared under anesthesia. The short-latency response is attributed to antidromic activation of layer V pyramidal neurons projecting to STN via the hyperdirect pathway, whereas the longer latency response may be due to polysynaptic activation along the orthodromic pathway through the basal ganglia-thalamus-cortex but is more likely the result of intracortical and thalamocortical synaptic interactions following antidromic activation of the hyperdirect pathway (Kumaravelu et al. 2018).

The pathological stimulation patterns generated head tremor that was synchronous with neural activity and was time-locked to stimulation bursts. This suggests that the neural and tremor activity were caused by the stimulation, rather than simply epiphenomena. There is evidence for dense projections from the STN to the orofacial motor area of the cortex in the rat (Degos et al. 2008). However, it is unlikely that STN DBS caused tremor by activating this pathway given the short latencies observed here. Another potential mechanism for the induced tremor is activation of descending pyramidal tract axons, either directly as they pass by the STN or indirectly through re-orthodromic propagation of antidromic action potentials generated in hyperdirect axons (Grill et al. 2008). Several experimental, imaging, and modeling studies have shown that DBS with clinically effective stimulation parameters can indeed result in activation of large-diameter myelinated axons outside the STN (Ashby et al. 1999; Mahlknecht et al. 2017; McIntyre et al. 2004; Tommasi et al. 2008), and this activation may be preferential to the corticobulbar tract over the corticospinal tract (Tommasi et al. 2008). Even if the cortical responses and induced tremor were evoked by antidromic activation of the cortex, this does not negate that generating oscillatory neural activity in the motor cortex and basal ganglia can indeed lead to tremulous motor activity. Our study does not allow us to identify the pathways responsible for the motor and neural responses that we observed here. Nevertheless, the neural and motor responses during pathological stimulation were synchronous and showed high coherence in the low-frequency band.

GRANTS

This work was supported by National Institute of Neurological Disorders and Stroke Grants R37 NS40894 and R01 NS079312 and by the Duke Compute Cluster.

DISCLOSURES

No conflicts of interest, financial or otherwise, are declared by the authors.

AUTHOR CONTRIBUTIONS

C.S.O., D.T.B., and W.M.G. conceived and designed research; C.S.O., D.T.B., and C.E.B. performed experiments; C.S.O. analyzed data; C.S.O. and W.M.G. interpreted results of experiments; C.S.O. prepared figures; C.S.O. drafted manuscript; C.S.O., D.T.B., C.E.B., and W.M.G. edited and revised manuscript; C.S.O., D.T.B., C.E.B., and W.M.G. approved final version of manuscript.