Abstract
Rest tremor is one of the most prominent clinical features of Parkinson's disease (PD). Here, we hypothesized that cortico-basal ganglia neurons tend to fire in a pattern that matches PD tremor frequency, suggesting a resonance phenomenon. We recorded spiking activity in the primary motor cortex (M1) and globus pallidus external segment of 2 female nonhuman primates, before and after parkinsonian state induction with 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine. The arm of nonhuman primates was passively rotated at seven different frequencies surrounding and overlapping PD tremor frequency. We found entrainment of the spiking activity to arm rotation and a significant sharpening of the tuning curves in the 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine state, with a peak response at frequencies that matched the frequency of PD tremor. These results reveal increased sensitivity of the cortico-basal ganglia network to tremor frequency and could indicate that this network acts not only as a tremor switch but is involved in setting its frequency.
SIGNIFICANCE STATEMENT Tremor is a prominent clinical feature of Parkinson's disease; however, its underlying pathophysiology is still poorly understood. Using electrophysiological recordings of single cortico-basal ganglia neurons before and after the induction of a parkinsonian state, and in response to passive arm rotation, this study reports increased sensitivity to tremor frequency in Parkinson's disease. We found sharpening of the population tuning to the midrange of the tested frequencies (1-13.3 Hz) in the healthy state that further increased in the parkinsonian state. These results hint at the increased frequency-tuned sensitivity of cortico-basal ganglia neurons and suggest that they tend to resonate with the tremor.
Introduction
Parkinson's disease (PD) is characterized by motor, autonomic, sleep, and neuropsychiatric symptoms (Kalia and Lang, 2015; Jost and Reichmann, 2017; Armstrong and Okun, 2020) that correlate with anatomic and physiological changes in the basal ganglia (Poewe et al., 2017). Tremor, an involuntary repetitive movement at rest, which generally presents a frequency of 4-6 Hz, is one of the primary motor clinical symptoms of PD (Timmermann et al., 2003; Jankovic, 2008; Dirkx et al., 2016). Many PD patients are categorized as tremor-dominant, and most display tremor at some stage of their disease (Jankovic et al., 1990). However, the fundamental question of the pathobiology and physiological mechanisms underlying PD rest tremor is still poorly understood.
One of the chief physiological hallmarks of PD patients and animal PD models is abnormal β and tremor-frequency oscillatory activity in several nodes of the cortico-basal ganglia circuit, including the external segment of the globus pallidus (GPe) and the primary motor cortex (M1) (Volkmann et al., 1996; Levy et al., 2000; Raz et al., 2000; Brown, 2003; Ivica et al., 2018). Several studies have shown that synchronous oscillatory activity (Nambu and Llinas, 1994; Nini et al., 1995; Raz et al., 2000; Brittain and Brown, 2014) in the cortico-basal ganglia loop plays a key role in PD pathophysiology. However, this abnormal β activity and tremor do not seem to be governed by the same mechanism (Levy et al., 2002; Moran et al., 2008; Hirschmann et al., 2013; Asch et al., 2020): studies on PD patients who underwent deep brain stimulation indicate that the reduction in neuronal activity in the beta band of the subthalamic nucleus (STN) is correlated with improvement in akinesia and rigidity, but not tremor (Kühn et al., 2006; Beudel et al., 2015; He et al., 2020). A growing body of evidence suggests that tremor severity is not correlated with other motor symptoms, thus hinting at a pathophysiological distinction between tremor and akinesia/bradykinesia and rigidity (Wilson et al., 2000; Lees, 2007; Koh et al., 2010).
The physiological mechanism of rest tremor in PD remains enigmatic. Correlations between tremor and rhythmic neuronal firing have been observed in several basal ganglia structures. These correlations have been documented in both the 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) PD model of nonhuman primates (NHPs) (Bergman et al., 1994; Rodriguez-Oroz et al., 2009) and in human patients (Rodriguez et al., 1998). Other studies have argued for a higher-level association between tremor and the ventralis intermedius (Vim) nucleus of the thalamus (Hirai et al., 1983), or suggested that tremor is related to the activity of the lateral motor thalamus (Magnin et al., 2000). The recent “dimmer-switch” hypothesis posits that parkinsonian tremor involves both the basal ganglia and the cerebello-thalamo-cortical circuit, where the basal ganglia is presumed to operate as the “switch” that activates the tremor, whereas the cerebello-thalamo-cortical circuit acts like a dimmer by modulating the amplitude of the tremor (Helmich, 2018).
Here, we explored the relationship between the entrainment of neural activity in the NHP cortico-basal ganglia network by passive rhythmic arm movements. This paradigm served to explore (1) whether neurons in the cortico-basal ganglia network show a tendency to fire at a pattern that matches these rhythmic movements; (2) whether this entrainment peaks at the tremor frequency, thus implying a resonance-like mechanism; and (3) whether the entrainment pattern is affected by dopamine depletion and the emergence of PD symptoms.
Materials and Methods
Animals
Two healthy African green (Cercopithecus aethiops) Vervet monkeys (Monkeys F and S; females, weighing 3.6 and 4.1 kg, respectively) were used in this study. Both animals underwent 8 weeks of training to familiarize them with application of forced rotation of their left arm. The animals' behavior in the experimental setup was monitored by video surveillance (GV-650, Geovision). The animals had access to standard primate chow and tap water ad libitum. Their weight, behavior, and physical condition were monitored daily by the researchers and qualified veterinarians. All procedures were conducted in accordance with the Hebrew University Guidelines for animal care and the National Institutes of Health's Guide for the care and use of laboratory animals. The procedures were approved and supervised by the Institutional Animal Care and Use Committee (Ethics Committee research number: MD-07-10 233-5).
Surgical procedure and MRI
The animals were implanted with a Cylux MRI-compatible recording chamber (Alpha-Omega) tilted at a 40° angle from the mid-sagittal plane on the right hemisphere, and a head-holder (Crist Instruments). The chamber was attached to the skull by titanium screws (Crist Instruments) and wires (Fort Wayne Metals). The surgical procedure was conducted under general anesthesia, induced by intramuscular ketamine (Fort Dodge Animal Health, Wyeth) and Domitor (Orion Corporation Animal Health) and maintained by inhaled isoflurane and N2O. Antibiotic (Cephtriaxone, Vitamed) and analgesic (Rymadil, Pfizer) treatment was given perioperatively. After a recovery period of 5 d, the animals underwent an additional week of training and an MRI scan (Siemens Medical Solutions). During the MRI examination, the monkeys were sedated with intramuscular ketamine and Domitor.
MPTP treatment
The monkeys were treated with the neurotoxin MPTP (MPTP-hydrochloride, Sigma) to induce a parkinsonian state. Five injections of 0.4 mg/kg (i.m.) were given over the course of 4 d (two injections on the first day) under mild (10 mg/kg) ketamine-induced sedation. Parkinsonian symptoms were developed gradually during the week of the injections. Both primates exhibited marked rigidity, rest tremor, considerably flexed posture, and a reduction in blinking frequency. The level of parkinsonism was assessed by the Imbert primate PD scale (Imbert et al., 2000) and compared with the levels measured in the healthy state.
Mechanical oscillator/driver
The left arm holder of the monkey was attached though a metal shaft to a DC motor (Oriental Motors). The moving arm holder's central axis was aligned with a magnetic angular location measurement device (a goniometer, Honeywell Sensotec) to determine the location of the animal's arm. The motor's rotation frequency was set via a custom-made controller (Physiology Department, Electronics Laboratory, Hebrew University) operated by dedicated computer software (developed in Microsoft Visual C++, Visual Studio 6, Microsoft) through a digital I/O card (National Instruments).
Experimental course
Rhythmic passive arm rotation was generated by the mechanical oscillator. Rotations were made in a pseudo-random sequence at the following frequencies: 1, 2, 3.5, 5, 8, 10, and 13.3 Hz. Each trial was a 90 s epoch of discrete frequency movements, followed by a 60 s intertrial interval (ITI). Recordings were made of extracellular spiking activity from the GPe and the M1 (see below) during the rotation trials and the ITIs. After the first recording phase in the healthy animals, a parkinsonian state was induced by MPTP intoxication, and a course of passive rotation of the arm and neuronal activity recordings was made in this state.
Electrophysiological recording
The animals' movements were recorded by the magnetic angular location measuring goniometer and 3D accelerometers (Honeywell Sensotec) fastened to the limbs proximal to the wrist and ankle joints.
In each recording session, two separate arrays of four tungsten glass-coated microelectrodes (150 μm diameter) were positioned in the arm-related area of the M1 and GPe, respectively. The microelectrodes were advanced separately (EPS, Alpha-Omega Engineering), with each array confined within a cylindrical metal guide (1.36/1.65 mm inner/outer diameter, Double MT, Alpha-Omega Engineering). The electrode signals were amplified (gain 5000), bandpass filtered (range 1-6000 Hz) using a four-pole Butterworth analog filter, and digitized using a 12-bit A/D converter filter (MCP, Alpha-Omega Engineering). All signals recorded on the system were stored for offline analysis (Alpha-Lab, Alpha-Omega Engineering).
Structure identification (to verify the recording location) was performed differently for each structure. Identification of the arm-related area of M1 was performed by elicitation of movement in the contralateral arm following micro-stimulation. The parameters for M1 identification were the threshold values of ≤20 μA for short duration (50 ms) trains (200 Hz) of short (0.2 ms) cathodic-anodic balanced pulse stimulation (Lemon et al., 1986) and elicitation of neuronal activity in response to passive limb manipulation, as judged by the experimenters' auditory impression. Identification of the GPe was performed according to its electrophysiological properties and the neuronal discharge of neighboring structures (DeLong, 1971). Thus, entry into the GPe was marked by an increase in background noise, as well as the appearance of high-frequency discharge (HFD) units exhibiting long (>500 ms) pauses of activity.
Data analysis
Tremor quantification
PD rest tremor was evaluated using four 3D accelerometers that were fastened to the NHPs' limbs, proximal to the wrist and ankle joints, during the ITIs (no passive arm movement). For each axis of each accelerometer, in both experimental states (healthy and PD model), we bandpass filtered the signal between 3 and 9 Hz, using a very sharp transition filter. We calculated the average root mean square of the absolute normalized signal in the healthy state and used this mean ± 2.5 SDs as the detection threshold for tremorous epochs. For each PD model filtered recording, we identified the detection threshold crossings and defined the epoch beginning and end as the points at which the signal went below the marking threshold, which was set as the mean ± 1.5 SDs. We calculated the power spectral density (PSD) of the tremor epochs and normalized (by dividing each power data point by the total power). We then averaged across tremor epochs to get the mean normalized PSD of the tremor epochs in each axis for the four 3D accelerometers.
Spiking signal
The inclusion criteria for single cells were an isolation score ≥0.85 and a stable firing rate (FR). Isolation scores range between 0 and 1 and represent the extent to which the recordings exclusively captured all of a given cell's spikes (Joshua et al., 2007). A “stable” cell had a relatively constant FR, within the time frame and experimental condition.
Cell type division and selection
In the GPe, only HFDs with a mean FR exceeding 20 spikes/s were selected. In M1, we divided the units into “narrow” and “wide” based on the shape of the spike's waveform (Mountcastle et al., 1969; Contreras, 2004). The division thresholds were determined based on the histogram of spike widths.
FR
For stability testing, the FR was calculated in 2 s windows. In each, the number of spikes was counted, and the sum was divided by the length of the window in seconds to obtain the spikes per second. For FR analysis, stable (well-isolated) segments were divided into 1 s windows, and the number of spikes counted in each. The FR was calculated by averaging these sums.
Autocorrelation function
We calculated the autocorrelation function of the single spike signals based on the method described in Abeles (1982). For each spike in the stable, well-isolated single spike signal, we calculated the times of the spikes in the following 1 s, relative to that spike, at a time resolution of 1 ms. For each unit, we calculated its autocorrelation function by taking the mean over all 1s binary signals. Finally, we smoothed the autocorrelation by convolution with a Gaussian kernel with a width of 10 ms. For the population autocorrelation function, we took the mean over all units in each brain region (M1 and GPe), cell type (narrow and wide units in M1), and experimental state (healthy vs PD model).
Cross-correlation function
The binary single spike signal was convolved with a Gaussian kernel with a width of 10 ms to transform it into an FR signal. We then divided the signal into 1 s segments based on the goniometer troughs and the mean of these segments was the cross-correlation of the unit with the arm rotation.
PSD was calculated using Welch's method, with a Hamming window of 3 s and 50% overlap. For each window, the signal mean was removed to avoid large contributions of the DC component. The frequency resolution was ⅓ Hz. A normalized PSD was then calculated by dividing each value (for the frequencies of interest at 0-15⅓ Hz) by the sum of the power (up to 15⅓ Hz). The normalized PSD was then averaged to obtain the population response for all brain regions, cell types and two experimental conditions.
Magnitude squared coherence (MSC) between the goniometer signal, a measure of the arm movement and the neural activity, was calculated using MATLAB's function “mscohere” with a Hamming window of 3 s and 50% overlap. The frequency resolution was ⅓ Hz.
Responsive units are units which, for a given arm rotation frequency, showed a significant PSD peak at that frequency. To reduce experimenter bias in the classification, we used a MATLAB implementation of a linear support vector machine classifier. We manually labeled 650 (200 responsive, 450 nonresponsive) samples of 1386 samples. From the labeled data, 200 responsive samples and 200 nonresponsive (chosen randomly of the 650) were used to train, cross-validate, and test the results of the linear support vector machine. We used 160 samples from each group (responsive and nonresponsive) and trained a model with sevenfold cross-validation. We then evaluated the resulting models' performance (accuracy and F1 score) by taking the majority vote of the seven resulting models as the predicted label and compared it with the true label. We used the majority vote method to label the rest of the data as well.
Statistical analysis
Changes in measures of the neural signal (FR, PSD, MSC), and their association with passive arm rotation and experimental state (healthy vs PD model), were evaluated with a linear mixed-effects model using IBM SPSS statistics program. Linear mixed-effects models are extensions of simple linear models (e.g., linear regression) that allow both fixed and random effects to be detected. This approach is better suited to our data than the classical ANOVA because it can deal with data containing missing values. In our case, many cells were recorded, but most were not recorded with all arm movement frequencies.
Changes in the number of responsive neurons was estimated using a χ2 test. We also used Fisher's exact test, since the χ2 test only provides an approximation of the real distribution and can be biased when the sample size is small, or the data are unequally distributed across cells in the table. Fisher's exact test performs better for small, sparse, and unbalanced data.
Results
Two NPHs (Cercopithecus aethiops, Monkeys F and S; females, 3.6-4.1 kg) were evaluated in the healthy state and the MPTP PD model state. Tremor was developed and was assessed clinically, together with other PD symptoms. We also performed a tremor quantification analysis of the data from 3D accelerometers that were fastened to the NHPs' limbs, proximal to the wrist and ankle joints. We recorded spiking activity from GPe and M1 at rest and while the NPHs' arm was rotated at seven different frequencies (ranging from 1 to 13.3 Hz). Figure 1 illustrates the experimental setup. In M1, cells were classified into “narrow” and “wide” units based on the spike width (Mountcastle et al., 1969; Contreras, 2004). Eighty-four narrow and 61 wide units met the criteria for further analysis in the healthy state. In the PD model state, 83 narrow and 40 wide cells met these criteria. In the GPe, we chose only HFD cells. Eighty-one and 80 cells met the criteria for further analysis in the healthy and PD model states, respectively. The number of analyzed samples in each experimental state, brain region, and frequency of arm rotation are displayed in Table 1. The results were similar for the two NHPs and therefore were pooled in what follows.
Number of samples analyzed per experimental state, brain region, and arm rotation frequency
Experimental setup. During the recordings, the primates sat in an experimental chair, with both hands restrained. The left arm holder was attached to a mechanical oscillator that generated a passive arm rotation. The moving arm holder's central axis was aligned with a magnetic angular location measuring device (a goniometer). The movements were also recorded by 3D accelerometers fastened to the limbs proximal to the wrist and ankle joints. A recording chamber that was surgically implanted to the primate's right skull enabled online neural recordings.
Parkinsonian rest tremor frequency peaks close to 5 Hz
The normalized PSD of the different axes of each accelerometer revealed a dominant frequency close to 5 Hz, which was more prominent in the NHPs' left limbs (Fig. 2a). We found tremorous epochs in 61% of the ITIs (no arm rotation) recorded in the PD model state. The durations of these epochs ranged from 1.12 to 23.6 s with a mean duration of 2.7 s and an SD of 2 s (Fig. 2b).
PD rest tremor frequency peaked at ∼5 Hz. a, Normalized PSD (normalized by dividing each power data point by the total power) of the four 3D accelerometers. b, Histogram of tremor epoch durations. Inset, Percent recorded ITIs in the PD state that contained tremorous epochs. c, Normalized PSD of the multiunit activity in M1 (top subplot) and GPe (bottom sublot), in response to PD rest tremor.
To assess the neural response to the tremorous epochs, we extracted single-unit activity during tremor epochs, in both M1 and GPe. Analysis at the single-unit level did not reveal tuning to the PD rest tremor. We suspected that the absence of tuning did not reflect the neurons' characteristics but rather a spectral distortion that led to reduced identification of low-frequency oscillations (Rivlin-Etzion et al., 2006). Thus, we repeated the analysis at the multiunit level and found tuning of M1 multiunit activity to tremor frequency (Fig. 2c). These results are consistent with M1's stronger response to the passive arm rotation (presented below).
M1 and GPe neural FRs are modulated by MPTP
The FRs of both narrow and wide units in M1 were significantly increased (F(675) = 5.44, p = 0.02 and F(368) = 32.73, p < 0.001, respectively) after the onset of the parkinsonian state (Fig. 3a). In the GPe, there was a significant interaction (F(732) = 6.77, p = 0.009) between arm rotation frequency and experimental state (healthy vs PD models) (Fig. 3b). The main effect of experimental state was not statistically significant but showed a trend toward a drop in FRs after the onset of the parkinsonian state, consistent with previous findings (Filion et al., 1991; Wichmann and DeLong, 2003).
FR in response to passive arm movements was associated with an increase in M1 and a decreasing trend in the GPe after the induction of the parkinsonian state. Mean FRs of (a) narrow and wide M1 units and (b) GPe neurons as a function of the arm rotation frequency. For each region, the two experimental states are presented. Error bars indicate standard error of the mean (SEM). Inset, All conditions on the same y axis.
Single neurons in M1 and GPe are tuned to arm rotation frequency
Figure 4 presents an example of the association between arm rotation frequency and the frequency of the electrophysiological activity. In this example, the neural discharge rate in both M1 (wide unit) and GPe (HFD) was tuned to the arm movement frequency. This time domain association was evident in the frequency domain as well, as shown in Figure 5, which shows the peaks of the neural PSD and the MSC at the arm rotation frequency.
An example of the time domain response of a wide M1 unit and GPe cell to arm rotation, illustrating the tuning of neuron firing to arm movement frequency. Example of M1 and GPe tuning to arm oscillations, as measured by a magnetic angular location measuring device (goniometer) attached to a moving arm holder. Columns correspond to the different experimental arm rotation frequencies. Data are from the PD model state. a, Passive arm rotation as measured by the goniometer. b, Recorded neuronal signal filtered using a Butterworth bandpass filter between 300 and 6000 Hz. c, Raster plots aligned with the period beginning with the goniometer. d, Mean FR calculated from the raster data.
An example of the frequency domain response of a wide M1 unit and GPe cell to arm rotation, illustrating the tuning of neuron firing to arm movement frequency. Frequency analysis of the response to arm rotation at different frequencies of the cortical and pallidal reactive cells from Figure 4. Columns correspond to the different experimental arm rotation frequencies. a, PSD of the passive arm rotation (as measured by a goniometer attached to an arm holder). b, PSD of the neural FR and (c) MSC between the goniometer and spikes.
M1 and GPe population activities are tuned to arm rotation frequency
The time-series dynamics were evaluated as the autocorrelation of population activity and the cross-correlation between population activity and arm rotation (measured with a goniometer). In the naive NHPs, the M1 neuronal population exhibited oscillatory behavior, which was more dominant for wide units than narrow units. The periodicity of the oscillations matched that of the arm rotation and showed increased amplitude subsequent to the onset of the parkinsonian state (Fig. 6, top two panels).
Neuronal population activity oscillates with arm rotation, at a similar frequency, and more strongly in the PD model state. Autocorrelation of spiking population activity, and its cross-correlation with the arm rotation signal (measured with a goniometer). Data are mean ± standard error of the mean (SEM).
In the GPe, oscillatory behavior was only detected in the NHP PD model, most notably in response to arm rotation at 5 Hz and adjacent frequencies (3.5 and 8 Hz, Fig. 6, bottom).
M1 and GPe population spectral measures peak at the arm rotation frequency
Strong tuning to the arm movement frequency was observed in the PSD of both types of M1 neural populations in the naive NHPs (Fig. 7, top two rows). However, only wide units showed a significant main effect of the arm rotation frequency (F(227) = 4.63, p = 0.032) and more prominent tuning in the PD than in the healthy model (approaching statistical significance, F(227) = 3.68, p = 0.056). The tuning was only apparent in the GPe in the PD model, with both main effects (arm rotation frequency and experimental state) reaching statistical significance (F(394) = 77.47, p < 0.001 and F(394) = 4.82, p = 0.029, respectively; Fig. 7, bottom row).
Neuronal population tuning to the frequency of the imposed arm rotations increases after induction of the parkinsonian state. a, Normalized PSD (normalized by dividing each power data point by the total power of frequencies ≤ 15⅓ Hz) and MSC between the spiking activity and the arm movement signal (measured with a goniometer). Data are mean ± standard error of the mean (SEM). b, Mean peak prominence of the MSC functions. Error bars indicate the standard error of the mean (SEM). a, b, Results are presented for the two conditions, two regions, and two cell types within M1.
There was a significant difference between the prominence of the peaks of the MSC of PD and healthy NHPs in the GPe and wide M1 units (F(496) = 9.45, p = 0.002 and F(260) = 8.8, p = 0.003, respectively). For both M1 cell types (narrow and wide), the main effect of the motor frequency was highly significant (F(479) = 12.18, p < 0.001 and F(260) = 5.74, p = 0.017, respectively).
To test whether there was stronger tuning of M1 and GPe neurons to the arm rotation in the range of tremor frequency, we repeated the analysis. This time, instead of calculating statistics for each arm rotation frequency, we divided them into two groups: tremor-range frequencies (3.5-8Hz) and no-tremor frequencies (1, 2, 10, 13.3 Hz). This grouping strengthened the results and highlighted the tuning to the tremor frequency range (between 3.5 and 8 Hz). A highly significant difference between states (healthy vs PD model) was found in the GPe (F(394) = 7.381, p = 0.007) and wide M1units (F(227) = 15.23, p < 0.001). The main effect of tremor frequency (arm rotation in the tremor frequency range vs out of tremor range) was also statistically significant (GPe: F(394) = 18.97, p < 0.001; wide M1 units: F(227) = 4.51, p = 0.035).
The MSC analysis showed similar results, but here the statistical significance was found for the narrow M1 units as well. A significant main effect was found for both experimental state (GPe: F(496) = 26.81, p < 0.001; wide M1 units: F(260) = 33.98, p = 0.000; narrow M1 units: F(479) = 11.6, p = 0.001) and tremor frequency (GPe: F(496) = 5.32, p = 0.021; wide M1 units: F(260) = 15.16, p < 0.001 and narrow M1 units: F(479) = 7.861, p = 0.005). Phase-locking analysis examining the spiking activity, and the arm rotation signal resulted in relatively low phase locking indices, with unclear tunning in the GPe.
Overall, both the time and frequency response analyses revealed that the spiking activity of neurons in both brain regions was highly tuned to arm rotation frequency. This tuning was significantly stronger for the midrange of the frequencies assessed here, compared with the extremes, and increased in the Parkinsonian state.
Increased fraction of responsive M1 and GPe neurons to arm rotation frequency after MPTP
Responsive neurons exhibited a significant PSD peak in their spiking activity at different arm rotation frequencies (Fig. 5). Neurons were classified as “Responsive” or “Nonresponsive” by a linear support vector machine. The ratio of responsive neurons was estimated before and after the administration of MPTP and the onset of the parkinsonian state. We found an increased number of responsive neurons after MPTP treatment in all the recorded brain regions and cell types. The increase was larger for the GPe and wide M1 units. Next, we tested whether this change would be larger for frequencies within the range of PD rest tremor frequency. We compared the number of responsive neurons in the range of 3.5-8 Hz to the number responsive neurons for the nontremor frequencies (1, 2, 10, 13.3 Hz). We found a significant effect in the GPe and for wide M1 units. Finally, we tested for a change in responsive units as a function of arm rotation frequency. In M1, we only found a significant increase in wide units and only for 5 and 8 Hz. A more widespread effect was detected in the GPe and was higher for frequencies that paralleled the tremor (Fig. 8).
Increased sensitivity to arm rotation frequency, mostly around tremor frequencies, in the PD model state. a, Percentage of responsive neurons that showed a significant oscillatory peak in PSD at the arm rotation frequency out of all recorded cells. A statistically significant effect for condition was found in both M1 (narrow units: χ2(1, N = 517) = 3.15, p = 0.076, p = 0.046 with Fisher's exact test; wide units: χ2 (1, N = 273) = 17.84, p = 2.4 × 10−5, p = 1.83 × 10−5 with Fisher's exact test) and GPe (χ2 (1, N = 576) = 44.43, p = 2.64 × 10−11, p = 7.08 × 10−12 with Fisher's exact test). Comparison between the change in number of responsive neurons (from the healthy to the PD model state) in the range of 3.5-8 Hz and the change in number of responsive neurons for the nontremor frequencies (1, 2, 10, 13.3 Hz) was statistically significant in wide M1 units (χ2 (1, N = 273) = 31.04, p = 3 × 10−6) and in the GPe (χ2 (1, N = 576) = 47.47, p = 1.22 × 10−9). b, Percentage of responsive neurons as a function of the arm rotation frequency. M1 narrow: χ2 (2, N = 517) = 37.32, p = 0.007; M1 wide: χ2 (2, N = 273) = 59.44, p = 4.75 × 10−6; GPe: χ2 (2, N = 576) = 62.92, p = 1.32 × 10−6. Post hoc pairwise comparisons (with Bonferroni correction) for each frequency were only statistically significant for some frequencies, most notably in the range of the tremor frequency. In M1 wide units: 5 Hz (χ2 (1, N = 44) = 5.15, p = 0.023, p = 0.024 with Fisher's exact test) and 8 Hz (χ2 (1, N = 39) = 11.8, p = 5.93 × 10−4, p = 0.024 with Fisher's exact test). In GPe: 2 Hz (χ2 (1, N = 85) = 10.91, p = 9.56 × 10−4, p = 0.0011 with Fisher's exact test), 3.5 Hz (χ2 (1, N = 87) = 4.09, p = 0.043, p = 0.047 with Fisher's exact test), 5 Hz (χ2 (1, N = 94) = 8.17, p = 0.0043, p = 0.0049 with Fisher's exact test), 8 Hz (χ2 (1, N = 81) = 15.17, p = 9.82 × 10−5, p = 5.53 × 10−5 with Fisher's exact test), and 13.3 Hz (χ2 (1, N = 78) = 7.96, p = 0.0048, p = 0.051 with Fisher's exact test). ***p < 0.001. **p < 0.01. *p < 0.05.
Discussion
Following MPTP administration, two NHPs developed a parkinsonian low-frequency rest tremor. The tremorous epochs were dominant, with a frequency close to 5 Hz. Multiunit analysis revealed M1 activity that was tuned to the PD rest tremor (neural peak and phase entrainment could not be estimated in this population level study). Passive rotation of the arm of the NHPs at seven different frequencies (1-13.3 Hz) led to stronger tuning of the spiking activity of MI and GPe neurons to the midrange frequencies. This range is similar to the tremor frequency of the African green monkey (Bergman et al., 1990; Heimer et al., 2006), and contains the tremor frequency measured in this study. In the motor cortex (M1), the tuning effect was prominent in both the healthy and parkinsonian states, whereas for the GPe, the effect was only apparent in the PD model state. In the GPe and wide M1 units (pyramidal), the tuning was stronger for the tremor range and was enhanced in the MPTP PD model. The results of neural activity-arm movement coherence analysis were consistent with the results of the power-spectrum analysis, with addition of significance to the narrow M1 units (interneurons). Finally, analysis of the responsive neuron counts revealed increased proportion of responsive GPe and wide M1 neurons in the PD model state, most notably in the range of the tremor frequency.
More than 200 years after James Parkinson's description in An essay on the shaking palsy (Jost and Reichmann, 2017; Obeso et al., 2017), vastly more is known about the pathophysiology and mechanism of PD. Recordings from patients undergoing deep brain stimulation have established the existence of abnormal synchronization and excessive β oscillations within the basal ganglia (Brown and Williams, 2005; Foffani et al., 2005; Weinberger et al., 2006). In the healthy brain, β oscillations probably function as a “status-quo signal” where a reduction in beta power occurs before and during movement (Kühn et al., 2004). Similarly, a reduction in abnormal beta activity by levodopa treatment was shown to be strongly correlated with improvement in akinesia and rigidity, but not tremor (Kühn et al., 2006). Tremor is probably an independent symptom, with a different pathophysiology from the akinetic/rigid symptoms of PD (Helmich et al., 2012), but its exact mechanism remains unclear.
Although our results do not explain the onset of the rest tremor, they suggest that a positive feedback loop between the neural discharge and the tremor itself may account for its characteristic frequency and persistence. Specifically, tremor may be caused by abnormal transient oscillatory activity in areas related to motor activity, a known characteristic of the disease (Deffains et al., 2016; Anastasopoulos, 2020). Because of the match between tremor frequency and the natural frequency of the cortico-basal ganglia cells, activity is increased, reinforcing the tremor phenomenon.
Several studies have suggested that the characteristic STN pathologic activity pattern in PD reflects alterations in both the cortex and GPe (Magill et al., 2001; Bevan et al., 2006). Evidence of increased sensitivity in basal ganglia nuclei to low-frequency rhythms emerging from the cortex after dopamine depletion further support this notion. For example, using a computational model of the STN-GPe network, Terman et al. (2002) showed that overactivation of the striatal-GPe pathway could increase the tendency of STN and GPe neurons to support low-frequency activity (Terman et al., 2002). Other observations suggest that the excitatory STN and inhibitory GPe form a feedback loop that engages in synchronized bursting. These authors proposed that the STN and GPe constitute a central pacemaker modulated by striatal inhibition of GPe neurons that could be responsible for synchronized oscillatory activity in the normal and pathologic basal ganglia (Plenz and Kital, 1999). Finally, Helmich's “dimmer-switch” model (2012) presents a similar idea where neuronal activity related to parkinsonian tremor first arise in the basal ganglia and then propagate to the cerebello-thalamo-cortical circuit, where the tremor rhythm is maintained and amplified.
More research is needed to fully account for the mechanism that causes tremor and the typical low frequency (4-6 Hz) of rest tremor in PD. We suggest that a resonance mechanism may play a role, in which positive feedback and/or a network loop may favor the 4-6 Hz domain and can contribute to the persistence of the tremor. The role of resonance could lie in determining the tremor frequency or in propagating and maintaining it. Even if such processes do take place, other features need to be accounted for. Since increased sensitivity can account for the persistence of tremor but not its onset, a possible trigger should be sought. Further, it remains unclear why sensitivity should increase following striatal dopamine depletion. This study thus constitutes a first step toward answering these questions and gaining a better understanding of the mechanism's underlying rest tremor in PD.
Footnotes
This work was supported in part by Israel Science Foundation, Deutsche Forschungsgemeinschaft (Retune, TRR 275), and Israel-China Bin-National Science Foundation to H.B.; and Israel Science Foundation Grant ISF 2128/19 to R.E. We thank Tuvia Kurz for the primate illustration in Figure 1.
The authors declare no competing financial interests.
- Correspondence should be addressed to Noa Rahamim at noa.rahamim{at}mail.huji.ac.il
