Deep cerebellar stimulation reduces ataxic motor symptoms in the shaker rat

Abstract

Objective:

Degenerative cerebellar ataxias (DCAs) affect up to 1 in 5,000 people worldwide, leading to incoordination, tremor, and falls. Loss of Purkinje cells, nearly universal across DCAs, dysregulates the dentatothalamocortical network. To address the paucity of treatment strategies, we developed an electrical stimulation-based therapy for DCAs targeting the dorsal dentate nucleus.

Methods:

We tested this therapeutic strategy in the Wistar Furth shaker rat model of Purkinje cell loss resulting in tremor and ataxia. We implanted shaker rats with stimulating electrodes targeted to the dorsal dentate nucleus and tested a spectrum of frequencies ranging from 4 to 180 Hz.

Results:

Stimulation at 30 Hz most effectively reduced motor symptoms. Stimulation frequencies over 100 Hz, commonly used for Parkinsonism and essential tremor, worsened incoordination, while frequencies within the tremor physiologic range may worsen tremor.

Interpretations:

Low-frequency deep cerebellar stimulation may provide a novel strategy for treating motor symptoms of degenerative cerebellar ataxias.

Introduction

Degenerative cerebellar ataxias (DCAs) affect 1 in 5000 individuals¹ and their motor symptoms frequently include gait incoordination, tremor, and falls. DCAs have numerous etiologies, Mendelian and sporadic. Despite twenty years since Spinocerebellar ataxia-causing genes were first identified^2,3, treatment strategies remain limited. Therapeutic approaches vary⁴, but they typically fail to sufficiently alleviate symptoms^5–9. Most degenerative ataxias feature Purkinje cell loss^10,11, resulting in modified inputs to the deep cerebellar nuclei; thus, neuromodulation targeting these deep cerebellar nuclei may succeed in numerous DCAs.

Deep brain stimulation in the deep cerebellar nuclei, downstream from degenerating Purkinje cells, may ameliorate the effects of Purkinje cell loss. While previous work assessed thalamic and basal ganglia stimulation to treat DCAs^12,13, we targeted stimulation to the dorsal dentate nucleus because the dentate is immediately downstream from Purkinje cells; further as the primary cerebellar motor output, it processes the majority of deep cerebellar motor control information. While deep cerebellar stimulation is novel for DCAs, it can yield therapeutic changes in the context of spasticity¹⁴ and recovery from stroke^15–17.

We tested whether dorsal dentate nucleus electrical stimulation reduces degenerative ataxia motor symptoms using the Wistar Furth shaker rat, an x-linked recessive DCA model. Shaker rats progressively lose Purkinje cells but retain an intact overall cerebellar architecture and dentatothalamocortical pathway^18,19. Symptoms begin at ~9 weeks with a full-body tremor, progressing to a severe, shaking ataxia, with frequent falling. The shaker rat’s recapitulation of the salient features of DCAs, its large cerebellar targets, and the high-throughput nature of rodent experiments made it ideal for these studies. We adapted a center-of-mass tracking design²⁰ and wrote custom software to directly measure tremor, gait coordination, and falls in an automated, operator-independent fashion. We tracked symptom presentation in shaker and wild type rats from which to compare symptom progression and determined baselines and appropriate ages for symptom analysis.

Therapeutic electrical stimulation frequencies vary across neurological diseases. Parkinsonian DBS is usually 100+ Hz, though 60 Hz can effectively treat bradykinesia²¹. 10 kHz stimulation is effective for spinal cord stimulation for chronic leg and back pain²², while sub-50 Hz can be effective for stroke recovery^15,16, chronic pain²³, and catalepsy²⁴. Certain parameters improve some symptoms while worsening others: high-frequency stimulation for primary Parkinsonism symptoms exacerbates vocalization deficits^25,26. We hypothesized that low-frequency stimulation would be most effective for ataxia because low-frequency stimulation enhances network throughput^16,27, while high-frequency stimulation functions as an informational lesion^28–30.

Thus, we tested frequencies from 4–180 Hz. We found 30 Hz to be most effective, while substantially lower or higher frequencies may worsen specific symptoms. Given that DBS effects can change over long periods, such as in dystonia treatment³¹, we tested the effects of stimulation over several hours and found them to hold.

Deep cerebellar stimulation may offer the rare chance to improve gait ataxia. Beyond presenting a new therapeutic option and new tools for ataxia study, our work is relevant to ongoing discussions of DBS mechanism and should encourage researchers to attempt low frequencies in novel DBS uses when symptoms may emerge from a neurological loss of information.

Methods

All surgical and experimental procedures were performed in the Department of Neurology at the University of Utah, were approved by the institutional animal care and use committee of the University of Utah, and complied with U.S. Public Health Service policy on the care and use of laboratory animals.

Experimental Protocols

We collected center of mass data in several cohorts of awake, unrestrained Wistar Furth rats (Fig. 1 C). First, we recorded six shaker rats and four wild type rats weekly from 7–35 weeks of age. Second, we surgically implanted seven shaker rats with chronic, bilateral, stimulating electrodes targeted to the dorsal dentate nucleus at 18–19 weeks of age and recorded repeatedly under various conditions from 20–28 weeks of age (Fig. 1 A,B). Center of mass data were collected from this cohort both on and off stimulation. After recordings, we perfused rats with phosphate buffer solution through the cardiovascular system and fixed with 4% paraformeldahyde before taking cerebella for histological analysis. Numbers of rats included in stimulating experiments were based on power analysis from preliminary data based on α=.05, power of 0.9, and a prediction of ~25% surgical or electrode failure.

Fig 1. Network model and experimental setup.

A. The dentate nucleus receives inhibitory input from Purkinje cells, progressively lost in the shaker rat model, and excitatory input from mossy and climbing fiber collaterals from pons and inferior olive. B. Short-term Purkinje cell disinhibition facilitates movement dentate bursting and consequent movement. We hypothesized that, in the absence of Purkinje cells, the dentate cannot respond properly to excitatory collateral inputs. While Purkinje cell millisecond-scale timing is likely necessary for fine motor control, we proposed that enhancing the dentate’s ability to respond to excitatory collateral inputs may assist with less precise coordination, such as with gait. C. Cartoon of data collection setup. Load cells interface through a DAQ card to LabVIEW for power and to provide data to quantify center of mass in real time at 1000 Hz. The pulse generator provides electrical stimulation to the rat through a tether.

Motor Analysis.

We created custom LabVIEW software to track center of mass at high frequency, based on previous work quantifying rodent center of mass by Fowler et al.²⁰ We attached a 1 cubic foot Plexiglas enclosure to the recording counter and centered four Honeywell model 31 miniature load cells under the corners of the Plexiglas enclosure, each attached to the corner of a 35-pound granite ballast (Fig. 1A). We placed a rigid, sub-100 gram detachable platform on the load cells below the bottom of the Plexiglas enclosure. We made a 2 cm hole on top of the chamber to allow tether connection. We soldered load cell inputs and outputs to a National Instruments USB-6008 DAQ interface for powering cells and collecting voltages.

We sampled load cell voltages at 1000 Hz in LabVIEW and computed position based on the weighted averages of forces applied at each of the load cells and their respective positions.²⁰ We exported positional data to MATLAB and wrote MATLAB code to analyze tremor, straightness of gait, and fall rates.

We analyzed tremor through position data Fourier analysis. Tremor frequencies observed during preliminary data collection were typically 2.5–8 Hz. Thus, we conservatively analyzed tremor from 2 to 10 Hz. We computed Fourier Transforms of position over successive 5-second time intervals, averaged the Fourier Transforms of age-matched WT rats, and integrated the area between the affected rat’s Fourier Transforms and the average of those from age-matched WT rats from 2 to 10 Hz. We normalized this value on a scale with 0 as WT average and 1 as the highest within-group weekly average for shaker rats.

We computed straightness of gait as an incoordination surrogate. Distances traveled over a threshold of 8 mm in 1 s were auto-detected and analyzed for straightness of gait. Importantly, the specific threshold was unimportant: we repeated analyses with a number of thresholds and found results to be similar across thresholds. We chose 8 mm as it was sufficiently large to eliminate shifts in weight but not too great to eliminate movements in which a turn was being made. For each movement, we computed the net displacement of the movement and the total distance traveled, sampled at 20 Hz. We computed the ratio of total movement distance to net displacement, thus giving an index for straightness of gait. A perfectly straight ambulation would give a ratio of exactly 1 — not achievable, as a minor amount of sway will be present in even healthy animals as they alternate feet — while a particularly uncoordinated movement with a great deal of truncal sway might yield a value of ~3.

We automated the detection of falls by identifying large millisecond-scale changes in force applied to load cells, only present during falls. During falls, the abrupt impulse applied to the non-affixed platform would de-weight at least one load cell for a duration of several milliseconds, providing a brief change in the force to that cell much more rapidly than could be caused by the most rapid movements made by rats. We tested various force thresholds and selected one that maximized agreement with qualitative fall determination, with 95+% agreement between automated detection and observed falls.

For all quantified measures, we measured the effects of DCS in a normalized fashion. As the data to be collected were numerous enough to require a minimum of several weeks of data collection time, and the motor symptoms in the shaker rat are constantly evolving, we alternated recordings on- and off-stimulation, with the first condition randomly chosen, and normalized on-stimulation symptom severity to off-stimulation averages made on the same day. For each symptom, we normalized such that a value of zero matched the average value collected from age-matched wild types, and a value of one matched the average value collected during sham stimulation, i.e. with stimulation turned off and symptoms untreated.

Surgical Procedures.

We subcutaneously administered rats 0.01 mg/kg Buprenex. After a two-hour wait, we anesthetized with 1–2% isoflurane. We shaved and disinfected the surgical site and placed rats on a heating pad in a stereotactic frame. We administered 0.1 mL Bupivicaine to the incision site, and then opened to the scalp and dried. We marked craniotomy sites with respect to bregma³². We used a burr to drill 7 holes anterior to the craniotomy sites and placed seven titanium bone screws to anchor eventual implants. We bilaterally targeted microelectrodes to the dorsal dentate nuclei. Given that craniotomies more than ~10 mm posterior from bregma induced bleeding in preliminary studies, we targeted through burr holes 9.5 mm and 3.5 mm lateral from bregma and then angled electrodes through 6.1 mm at 14.3° to reach their target (3.5 mm lateral, 11.0 posterior, 5.9 mm ventral from bregma). We implanted two-channel microstimulating arrays — 75 μm platinum-iridium electrodes with 200-μm center-to-center spacing from Microprobes for Life Science — through the holes and cemented to the bone screws with dental acrylic. After placing all elements, we smoothed over the headcap with additional acrylic. We sutured the incision and provided 0.1 mg/kg carprofen subcutaneously three times daily following implantation. Rats were given a two-week recovery period.

DCS.

For all implanted animal trials, whether stimulation was provided or not, we attached a two-channel tether to each of the array connectors on the headcap. We fed the tether through the top of the behavioral chamber to a commutator and connected the commutator to a pulse generator (Fig. 1B). We determined amplitude thresholds by gradually increasing current until side effects occurred, ranging from whisker twitching and sudden movements to full-body muscular contractions. We set current to 90% of the minimum that induced side effects, with a conservative maximum of 250 μA, based on computing a safe current ceiling of 300 μA³³. We delivered bipolar, biphasic stimulation with 100 μs pulse width. We applied stimulation for 5 minutes prior to on-DCS recordings and turned it off for 5 minutes prior to off-DCS recordings.

Motor Recordings.

We made weekly 30-minute recordings in six shaker rats and four WT rats 7–35 weeks old, demonstrating that the quantified symptoms were present in all shakers by 19 weeks. Thus, we collected motor data from array-implanted shaker rats between 20– 29 weeks old, ensuring the presence of all symptoms. Several factors necessitated within-day normalization when quantifying symptom alleviation. First, symptoms evolve in the shaker rat model and recordings lasted a number of weeks, sufficiently long for substantial evolution. Second, repeated recordings reduced environment novelty, potentially reducing exploratory behavior. Finally, we observed adaptive behaviors by shaker rats — using walls for support, wider stance, etc. — and could not discount progressive adaptation in the chamber over successive recordings.

While testing DCS, we made recordings during 40-minute intervals, alternating on- and off-stimulation, randomly selecting whether to first record on or off DCS, preventing bias from on vs. off order. We normalized symptoms recorded with stimulation to values recorded without stimulation during the same interval. In each scenario, we allowed 5 minutes in the stimulation setting — on or off — before collecting motor data for 5 minutes, avoiding wash-on and wash-off effects. Following the collection of motor data at all tested DCS frequencies — 4, 10, 20, 30, 40, 60, 80, 100, 130, and 180 Hz — we tested the effects of the most effective frequency, 30 Hz, over a longer duration to ensure that the effects of stimulation would still be present, at least over a moderate time scale. On two successive days, starting within 10 minutes of the same time of day, rats were placed in the recording box for two hours with stimulation turned on or off — with random selection as to which recording was made first — and then recorded for 15 minutes in the condition. Thus rats were recorded during the final 15 minutes of a 2-hour, 15-minute session in the motor recording chamber either with stimulation the full time or no stimulation provided at any point.

Histology.

Following data collection, we anesthetized rats, perfused through the cardiovascular system with PBS, fixed with 4% PFA, and removed cerebella. We soaked cerebella in 4% PFA, cryoprotected, and flash froze each in OCT compound before cutting into 25-micron coronal slices with a Leica 5100s cryostat. Finally, we located electrode tracts and estimate electrode tip locations.

Statistical Analysis.

When data were distributed in an approximately Gaussian fashion, summary values consist of mean and standard error. When data were distributed in a non-Gaussian fashion, we found summary values from 100,000 bootstrapped populations resampled with replacement and present means and inner quartiles. Statistical test abbreviations consist of t-test as student t-test and bca as bootstrapped confidence assessment. When multiple frequencies were compared to sham stimulation (Fig. 3) and when multiple ages of animals were, significance was determined from α=.05 following Holm-Bonferroni multiple comparison correction; resulting nominal and corrected p-values are listed.

Fig 3. Effects of deep cerebellar stimulation on motor symptoms; side effect thresholds vs frequency.

Error bars in A-C and grey in D represent bootstrapped middle quartiles. A. 10, 20, and 30 Hz stimulation resulted in tremor reduction (bca, p<.005 nominal, p<.05 corrected), while 40 Hz stimulation trended towards improvement (bca, p=.0116 nominal, not significant corrected). 4 Hz stimulation trended towards worsened tremor (bca, p=.0304 nominal, not significant corrected), and 60+ Hz stimulation had no significant effect. B. 20 and 30 Hz stimulation improved straightness of gait (bca, p<.005 nominal, p<.05 corrected). 130 and 180 Hz stimulation, individually, trended towards worsened incoordination, but had a combined significant effect of increased incoordination (bca, p=.00447 nominal, p<.05 corrected). C. 30 Hz stimulation resulted in reduced fall rates (bca, p=.00021 nominal, p<.05 corrected). D. Higher stimulation frequencies resulted in motor side effects with lower amplitudes. To avoid damage to neural tissue, we did not test amplitudes higher than 250 μA. While no rats were excluded from A-C, one rat was excluded from D due to both exhibiting no side effects at any frequency-amplitude combination and having both electrodes well off target (see Fig. 5).

Results

Ten rats, six shaker and four wild type, made motor recordings weekly from 7–35 weeks of age. Eight 17–18 week shaker rats were bilaterally implanted with stimulating arrays targeted to the dorsal dentate nucleus, and seven survived surgery. The electrode headcaps dislodged from two rats during the course of experimentation. All rats were tested at both 30 Hz — the most effective frequency — prior to headcap loss.

Model Characterization.

Tremor emerges strongly with symptom onset.

We compared tremor, straightness of gait, and fall rates between shaker and wild type rats at 7–35 weeks of age (Fig. 2). In all shaker rats, tremor presented either first or alongside other symptoms. Four rats exhibited tremor at 9–11 weeks of age, the last two having symptom onset at 14 and 19 weeks. These were consistent with previous qualitative determinations that shaker rats first display symptoms between 9 and 20 weeks of age.¹⁸ While qualitative observations noted onset no earlier than 9 weeks, we quantitatively found that shaker rats demonstrated significantly more tremor than wild type animals (bca, p<.00001) at all ages from 8 weeks on. Power spectra of representative 10-week shaker and wild type rats are shown (Fig. 2A), along with tremor vs. age (Fig. 2B).

Fig 2. Quantification of motor symptoms in the Wistar Furth shaker rat.

A. Power spectral densities of center of mass tracking data for 10-week-old wild type (blue) and shaker (red) Wistar Furth rats. The ~5 Hz peak in the shaker rat indicates ~5 Hz tremor. B. Tremor, measured across 2–10 Hz, was greater in affected animals than in wild type animals at 8 weeks and beyond (bca, p<.00001). Lighter color indicates bootstrapped inner quartiles in B, D, E. C. Age-matched, 26-week-old wild type (blue) and shaker (red) rat center of mass traces during rapid movements. Color changes from dark to light indicate 1 s change in time, while dashed arrows denote displacement. Wild type animals made straighter movements. D. Comparing total distance traveled during rapid movements vs. net movement displacement, affected rats demonstrated progressive incoordination and were significantly less coordinated than control animals at all time points after 9 weeks (bca, all p<.0002). E. Affected rats fall significantly more than wild type rats at 9+ weeks (bca, all p<.03).

Incoordination of gait progresses more slowly.

Representative movements made by 24-week shakers and wild types depict shaker rat truncal sway as they fail to travel in a straight line (Fig. 2C). Shaker rats walked straighter at 7 weeks (bca, p=.0033), but gait trended less straight for several months before stabilizing, with less straight gait than wild type rats at 10+ weeks (Fig. 2D, bca, p<.0002 at all 10+ week ages).

Fall rate progression largely follows incoordination of gait.

Finally, while shaker rats fell less than wild type animals at 7 weeks of age (bca, p=.0050), they fell more frequently at 9+ weeks of age (bca, p=.043 at 9 weeks of age, p<.03 at all ages beyond 9 weeks). Similar to straightness of gait, fall rates trended worse for several months before stabilizing (Fig. 2E).

Deep Cerebellar Stimulation.

30 Hz deep cerebellar stimulation reduces symptom severity on an acute timescale.

We acutely tested deep cerebellar stimulation (DCS) in 20–28 week shaker rats. We normalized on-condition recordings to off-condition recordings from the same day. In general, 30 Hz DCS was most effective. Tremor was significantly reduced by 10, 20, and 30 Hz stimulation (Fig. 3A, bca, p<.0022 nominal for each, p<.018 corrected). 40 Hz trended towards reduced tremor (bca, p=.012 nominal, not significant corrected), 4 Hz stimulation trended towards worsened tremor (bca, p=.030 nominal, not significant corrected), and 60+ Hz stimulation had little-to-no effect. Gait was significantly straightened by 20 and 30 Hz stimulation (Fig. 3B, bca, p<.0049 nominal for each, p<.044 corrected), while other frequencies up to 100 Hz had little-to-no effect. 130 Hz and 180 Hz stimulation, used for Parkinsonism and Essential Tremor, trended towards worsened gait (bca, p=.013 and p=.092 nominal, respectively, not significant corrected). Analyzing 130 and 180 together based on their likely similar neural effect yielded worsening of gait (bca, p=.0048 nominal, p=.043 corrected). 30 Hz stimulation reduced falling (Fig. 3C, bca, p=.00021 nominal, p=.0021 corrected), while 40 Hz stimulation trended towards improvement (bca, p=.038 nominal, not significant corrected).

With each rat, at each frequency, we determined minimum amplitudes that induced side effects, testing stimulation at 90% of this threshold. Interestingly, we found the relationship between current threshold and frequency to approximately resemble that of a strength-duration curve, only with frequency on the x-axis rather than pulse width (Fig. 3D). One rat did not exhibit side effects at any amplitude and frequency combination (see Fig 5), but it has not been excluded from analyses, except from the side effect threshold plot.

Fig 5. Electrode tip location estimates.

Figure adapted from Paxinos and Watson³², based on coronal slicing, and 11.3 mm posterior from bregma, with dentate nucleus situated approximately 10.6 to 11.7 mm posterior from bregma. Electrode tracts were identified histologically, and tip locations were estimated and diagrammed. Numbers correspond to approximate distance of tip posterior from bregma in millimeters. Note that “LatC” refers to the lateral cerebellar nucleus, synonymous with dentate nucleus, while int refers to interposed nucleus. For six animals in which stimulation produced side effects, 10 of the 12 electrode tracts were identified, and tip estimates are shown in black. For one animal in which stimulation did not generate side effects, both electrode tracts were identified, and tip estimates are shown in grey.

30 Hz stimulation is still effective on a moderate timescale.

After testing all frequencies, we tested 30 Hz DCS over a moderate timescale. We compared on-stimulation vs. off-stimulation conditions over back-to-back days at the same time of day, making 15-minute recordings after rats spent 2 hours tethered and within the specified condition, order of presentation randomized. In terms of tremor (Fig. 4A, bca, p=.0117) and straightness of gait (Fig. 4B, bca, p=.0404), 30 Hz stimulation significantly improved symptoms after being applied for 2 hours. In terms of falls, 30 Hz stimulation trended towards improved symptoms, but changes were insignificant (Fig. 4C), with limited data points and large variability.

Fig 4. 30 Hz stimulation maintains efficacy over a longer timeframe.

30 Hz and sham stimulation were presented for a period of 2.25 hours on back-to-back days, starting at the same time of day, order of presentation randomized. Motor recordings were made during the last 15 minutes and compared between groups. Error bars refer to bootstrapped middle quartiles. A. Tremor is significantly lessened by 30 Hz stimulation after 2 hours stimulation duration (bca, p=.0117). B. Gait is significantly straightened by 30 Hz stimulation after 2 hours stimulation duration (bca, p=.0404). Note that a value of 1.0 refers to perfectly straight gait, while wild type rats average a value of approximately 1.5. C. Fall rates trended positively with the application of 30 Hz stimulation after 2 hours duration, but were not significantly improved.

Histological Analysis.

Rats with electrodes implanted in or near the dentate nucleus received treatment benefit.

We cut cerebella into 20 μm slices and localized 12 of 14 electrode tracts (Fig 5). We found that all but three electrode tips were estimated to be within 500 microns of the dorsal dentate nucleus, at which point the dorsal dentate nucleus would be largely activated by 250 μA stimulation from our electrodes. One animal demonstrated no side effects at any combination with both electrodes located an estimated 1+ mm from the target, and one animal had one electrode ~1 mm from the target but another within 200 μm, demonstrating side effect thresholds consistent with other animals and therapeutic benefit, indicating that unilateral stimulation may be useful. Finally, half of electrodes localized would likely have activated at least a portion of the dorsolateral interposed nucleus.

Discussion

DBS is clinically established to treat medically refractory forms of several neurological diseases, but DBS for cerebellar disorders has not been studied extensively. This work demonstrates that deep cerebellar stimulation may provide a novel therapeutic strategy in ataxia treatment. We used a unique genetic rodent model with progressive PC degeneration, allowing precise disease process timing and obvious symptoms. Using a rat rather than mouse model made targeting more reliable, allowing us to observe improvement in motor phenotypes. Further, using a rodent model gave us the opportunity to test many frequencies, with multiple trials per frequency per animal, to more precisely determine the effects of stimulation.

Comparison with pharmaceutical strategies.

It is possible that therapeutic effects could be stronger in humans, with better electrode targeting programming. Deep cerebellar stimulation, if translated, may prove a more widely useful therapeutic strategy than pharmaceutical approaches. Arguably the most promising pharmaceutical approach proposed for ataxias is anti-sense oligonucleotides (ASOs)^34,35. While ASOs will likely prove more beneficial than DCS for patients with specific ataxias for which ASOs exist, patients with sporadic ataxias, spinocerebellar ataxias without an ASO, or hereditary ataxias for which the gene is unknown cannot benefit from ASOs. In many of these cases, patients could be DCS candidates. While it is unreasonable to hypothesize that DCS will slow disease progression or resolve very fine motor control symptoms, it may provide substantial gait ataxia relief.

Low-frequency stimulation may function through network enhancement.

Parkinsonism and Essential Tremor are typically treated with high frequency stimulation³⁶. However, our work indicates that lower frequencies would be more effective in the treatment of ataxia. High frequency DBS functions as an informational lesion^28,29 or functional deafferentation³⁷, while it has been proposed that low-frequency (<50 Hz) stimulation enhances network output^16,27. Purkinje cells regulate movement timing and strength through connectivity with the deep cerebellum, and complete suppression of Purkinje cells suppresses motor activity³⁸. Thus, while Parkinsonism is treated by informational lesion²⁹, degenerative ataxias may require network enhancement. In this model, DCS enhances the passage of remaining coordination-relevant information through the dentate, perhaps primarily through pontine mossy fiber and olivary climbing fiber collaterals. This could make sense given that low frequency DBS increases firing variability and informational capacity³⁹.

Very low- and high-frequency stimulation may worsen symptoms.

While low frequency stimulation was most effective, very low frequencies in the range of tremor frequency may worsen tremor. Local field recordings in macaques have implicated deep cerebellar oscillations in tremor generation, with the frequency of oscillations corresponding to the frequency of tremor, and coherence between local fields potentials and finger acceleration⁴⁰. Thus, introducing further tremor-frequency-range oscillations to the region may increase drive of tremor. On the other end of the frequency range, high frequency stimulation did not significantly modify tremor, but frequencies above 100 Hz may worsen incoordination. It is possible that high frequency stimulation further blocks collateral signals while ineffectively reducing low-frequency oscillatory power, perhaps due to low side effect thresholds at high frequencies. This could explain the contrast to Parkinsonian DBS, in which high frequency stimulation reduces low frequency oscillations⁴¹.

Stimulation of deep cerebellar nuclei vs other targets for ataxia.

Previous studies in ataxia patients showed that thalamic and basal ganglia stimulation improved cerebellar tremor, without consistently improving gait ataxia^12,13. We propose that it may be possible to treat tremor through a wide set of stimulation locations and parameters, while improving coordination is more difficult. Inspired by the ideas that Essential Tremor depends on interplay across many oscillatory regions and may represent an intermediate state of Purkinje cell degeneration⁴², we propose that Purkinje cell misfiring and loss dysregulate the cerebellothalamocortical network, allowing oscillatory propagation through the network. It stands to reason that tremor generation could be disrupted at numerous nodes. However, movement coordination depends on the deep cerebellar nuclei. Purkinje cell loss directly reduces not only informational input, but also the dentate’s ability to respond to collateral input through lost short-term disinhibition. As mossy fibers are too few⁴³ and climbing fibers are too slow⁴⁴ to generate the dentate bursting that precedes movement, short-term disinhibition from Purkinje cells facilitates bursting activity⁴⁵ and movement⁴⁶. The absence of Purkinje cells may prevent short-term disinhibition and replace it with functionally dissimilar chronic disinhibition. Thus, low-frequency stimulation may aid in the propagation of coordination-relevant signaling from collaterals in the absence of short-term disinhibition. In fact, a single climbing fiber collateral arborization spreads several hundred microns within the rat dentate⁴⁷, appropriate for generating coordination. In this interpretation, stimulation downstream from the deep cerebellar nuclei would not improve signal propagation through the deep cerebellum, hence the lack of gait ataxia improvement.

Finally, in the discussion of deep brain stimulation of other targets in the context of ataxia, it is important to touch on pulse width modulation. Recent work indicates the great potential of pulse width shortening to avoid ataxic side effects in essential tremor patients with thalamic high-frequency stimulation, downstream from the deep cerebellum⁴⁸. In this work, we chose to perform studies with a relatively long pulse width to increase the volume of tissue activated; as high-frequency and low-frequency DBS having substantially different effects on neural circuits and information processing, it is unlikely that low-frequency long-pulse stimulation will generate motor side effects, as confirmed by other rodent deep cerebellar stimulation studies^16,17. However, given the recent work by Choe et al., it will be important in the future to optimize stimulation in terms of not just frequency and amplitude, but also pulse width.

Stimulation frequency vs side effect amplitude threshold and mechanisms.

The relationship between stimulation frequency and side effect threshold hints at the importance of pacing of inputs to the deep cerebellar nuclei, especially given that Purkinje cell firing rate changes parallel motor symptom evolution prior to Purkinje cell loss.⁴⁹ It is possible that high frequency stimulation may easily induce something akin to the bursting that assists with movement generation. Indeed, the side effect profile typically observed with overstimulation comprised simultaneous contractions of multiple muscle groups, consistent with overexciting motor circuits. With lower frequencies, the amplitude required to produce enough excitatory input to generate unintended muscular contraction raises due to a reduction in stimulation rate. While we did not determine stimulation thresholds at frequencies above 1 kHz, it would be interesting to determine whether there is a horizontal asymptote or whether the relationship resembles a higher-order polynomial, as in the case of ultra-high frequency thalamic stimulation losing efficacy in regularizing firing and suppressing tremor⁵⁰. If the latter, it could be that the low- vs. high-frequency stimulation mechanisms are similar in the deep cerebellar nuclei to those in the basal ganglia, thalamus, etc. The correct frequency choice for DBS treatment for a particular disease would simply depend on whether symptoms are generated primarily by gain of function, requiring high frequency stimulation, or loss of function, requiring low frequency stimulation.

Potential caveats.

We have only demonstrated improvements in ataxia specific to gait. It seems unlikely that the complex millisecond-timescale computations performed by Purkinje cells required for very fine motor control — enabling, say, piano performance — can be recovered through electrical stimulation. However, gait ataxia improvement, alone, would represent a substantial improvement in patient treatment. Finally, based on relatively poor precision of bregma-based electrode targeting in rodent models, it is possible that the dentate nucleus was not the sole activated nucleus in all rats, and the interposed nucleus was likely activated in some. Given similar downstream regions activated by interposed and dentate and related functions, it is reasonable to propose the interposed nuclei as a target. Future studies must be performed to determine whether it is ideal to stimulate in both or only the dentate. Finally, we must note that it is, of course, difficult to fully differentiate tremor from ataxia, as previous groups studying potential DBS therapies have noted¹³. A full-body tremor could seemingly result in an increase in our ataxia measure. However, comparison of the tremor vs ataxia plots in both figures 2 and 3 clearly indicates that the tremor and ataxia measures are independent, as they evolve quite differently with age and frequency of stimulation.

It is clear that more study of low-frequency deep cerebellar stimulation to treat ataxias is warranted. Additional rodent work must be completed, but given the poor current treatment options, this therapeutic strategy may hold promise for numerous patients in the future.

Supplementary Material

Supplemental Video 1

Video 1. Two wild type Wistar Furth rats are compared to two similar age Wistar Furth shaker rats with typical symptom severity. Note that while the wild type rats walk in a coordinated fashion, the shaker rats exhibit tremor, severe ataxia of gait, truncal sway, and a large amount of stumbling in their movement.

Supplemental Video 2

Video 2. A chronically implanted Wistar Furth shaker rat with tremor-dominant symptoms is shown with repeated wash-on stimulation and wash-off from stimulation. Note immediate tremor reduction accompanied with washing on stimulation and immediate tremor reemergence with washing off sitmulation. Wash-on and wash-off effects are typically stronger than more chronic effects – typical tremor reduction is over 50% with 30 Hz stimulation, but this would quantitatively represent an even stronger effect -- but give the best ability to compare in short segments without bias, as tremor strength fluctuates under both conditions.

Supplemental Video 3

Video 3. A chronically implanted Wistar Furth shaker rat with ataxia-dominant symptoms is shown off and on stimulation, with representative, typical changes in ataxia. Note the severe truncal sway, postural instability, and stumbling that occur with regular frequency without stimulation, all reduced by stimulation. While stimulation does not fully reduce symptoms and the rat still frequently uses its surrounding environment for support, 30 Hz stimulation brings about a 30% reduction in ataxia of gait in the average rate.