Mechanisms of Deep Brain Stimulation in Movement Disorders as Revealed by Changes in Stimulus Frequency

Increased efferent activity from stimulated nucleus is propagated to downstream nuclei 

A question that arises from these results is whether the increased activity in the efferent axons of the stimulated nucleus is propagated via synaptic transmission to downstream nuclei. One hypothesis is that during HFS, changes take place in the somas of neurons in the stimulated nucleus that render these neurons incapable of releasing neurotransmitter, despite action potential propagation down the axons.33 Results that support this hypothesis include reduced secretions of prolactin, dopamine and norepinephrine in cells after 24 hours of HFS, suggesting that HFS suppresses the cellular machinery necessary to transmit activity to downstream neurons.34 However, this study was conducted in non-neuron cell types, and the application of these results to electrically active neurons is debatable. An alternative hypothesis is that repeated high-frequency activation leads to transmitter depletion and cessation of transmission from presynaptic to postsynaptic neurons.35

On the other hand, multiple studies confirm that the increased activity in efferent axons of the stimulated nucleus is propagated to downstream nuclei, although the propagation of DBS-induced action potentials may not be 1:1.36 During HFS in the STN, firing rates increased in the GPe and GPi (excitatory targets of the STN), which suggests that the efferent activation of the STN axons was propagated to downstream nuclei, and these effects were maintained for at least 5 minutes.27 Similarly, firing rates in the SNr of rats with neuroleptic-induced catalepsy, which receives excitatory inputs from the STN, increased from ∼29 Hz to ∼42 Hz during HFS.37 Functional magnetic resonance imaging revealed that primary areas of activation during STN DBS were the ipsilateral thalamus and GP,38 and more specifically the GPe39, which are all downstream targets of the STN. Furthermore, glutamate levels in healthy rats increased in GP and SNr as a function of STN DBS frequency up to a plateau at 130 Hz.40 The increase in glutamate levels was maintained during DBS at frequencies as high as 350 Hz, but was slightly lower than at 130 Hz DBS, indicating that STN neurons released additional glutamate during high-frequency DBS. These data indicate that stimulated activity in the efferent axons of the stimulated nucleus is indeed propagated to downstream nuclei.

Basal ganglia and thalamic pathophysiology in movement disorders 

Complex changes in the rate, patterns, and information flow in neurons and networks of the basal ganglia and thalamus accompany movement disorders.41 There is an increase in the presence of bursts, low frequency oscillations, and synchronization in neurons in the basal ganglia and thalamus in movement disorders.22, 42, 43, 44, 45, 46, 47, 48 Such changes in discharge patterns and synchronization play an important role in the pathophysiology of movement disorders. Motor symptoms including tremor, bradykinesia, and rigidity are associated with increased neuronal synchronization and low-frequency rhythmic oscillation,49, 50, 51 and bursts and oscillations parallel the appearance of motor symptoms in PD.52, 53, 54

A comparison of single unit GPi activity in subjects with PD versus Huntington’s disease (HD) highlights the importance of the patterns of neuronal activity in addition to the rate of neuronal activity in movement disorders. The classical rate model of the basal ganglia divides the basal ganglia into the direct and indirect pathways (i.e., activity in the direct pathway initiates movements while activity in the indirect pathway inhibits movements). The clinical manifestations of PD and HD are polar opposites; PD is a hypokinetic disorder, whereas HD is a hyperkinetic disorder. Thus, the classical rate model predicts increases in the rates of GPi neuronal firing in PD (due to over-activity in the indirect pathway) and predicts decreases in the rates of GPi neuronal firing in HD.55, 56 However, although the firing rates in the GPi are not significantly different in these two diseases, the firing patterns are different.47 The number of GPi neurons that participate in bursts is lower in HD subjects than in PD subjects.47 The burst indices and coefficients of variation of neuronal firing are also significantly lower in HD subjects than PD subjects.47 Finally, there is an increase in oscillatory activity in tremulous PD subjects as compared to HD subjects.47 Hence, there is an increase in bursting, irregular activity, and oscillatory firing patterns in the GPi of PD subjects as compared to HD subjects.47

In addition to these findings in humans, in vivo recordings in nonhuman primates revealed that GPi firing rates are not significantly different between healthy and Parkinsonian (1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine [MPTP]-treated) monkeys.52 These findings indicate that the symptoms of movement disorders are paralleled not only by changes in the rate of activity in the basal ganglia and thalamus, but perhaps more importantly by changes in the firing patterns of neurons.57

Changes in neuronal firing during DBS 

DBS can cause changes in the patterns of activity of neurons surrounding the electrode.5 Changes in neuronal activity in the basal ganglia and thalamus during DBS include changes in rates and patterns of firing (i.e., pathological oscillations, bursting, and regularity of neuronal interspike intervals), as well as changes in metabolic rates, all of which are strongly dependent on the frequency of stimulation. The methods of assessing these changes in activity include recording single unit activity, recording local field potentials, simulating neuronal activity in computational models, measuring changes in metabolic activity, and functional imaging. Both single unit and local field potential recordings during DBS share the problem of contamination by the stimulation artifact. Although methods exist for subtracting or rejecting stimulus artifacts, these methods undoubtedly neglect some amount of activity and we do not know how much information is lost or the relative importance of the lost information. The technical difficulty of stimulating and recording from the same DBS lead, in conjunction with simultaneous artifact removal, previously precluded the recording of local field potentials in the stimulated nucleus during DBS in humans. However, methods to overcome the technical challenges of recording local field potentials during DBS were recently introduced,58 and studies using these methods are likely to be forthcoming. Computational models provide detailed information regarding the activity of neurons during simulated DBS, but are limited in their application by the assumptions made in generating the models and the uncertainty of model parameters. Finally, indirect methods of assessing neuronal activity (e.g., functional imaging and measurements of metabolic activity) are compatible with simultaneous DBS, but the link of these signals to neuronal activity is unclear, and they lack the ability to detect the fine structure of rates or patterns of neuronal activity. Despite the weaknesses of the individual techniques for assessing neuronal activity during DBS, evidence from multiple modalities indicates that DBS generates important changes in neuronal firing rates and patterns.

First, high-frequency stimulation reduces the pathological oscillations that are otherwise elevated in PD and other neurological disorders.59 High-frequency (>70 Hz) STN DBS reduced pallidal beta oscillations (11–30 Hz) in PD subjects in a manner similar to the administration of dopamine precursors, whereas 25 Hz DBS increased the power of pallidal beta oscillations.60 In addition, high-frequency STN DBS at 130 Hz or 185 Hz attenuated STN beta oscillations in subjects with PD during the period just after DBS was turned off.20, 61 Similar trends were observed in MPTP monkeys, where single unit activity can be recorded during HFS of the STN. The number of individual oscillatory STN neurons decreased from 37% just before HFS to 12% during HFS, and then returned to 35% within 3 minutes after HFS.22

High-frequency DBS also generally decreases pathological synchronized burst activity in the basal ganglia and thalamus. The percent of SNr cells contributing to burst activity increased from 60% to 90% after neuroleptic injections in rats, and then decreased back to 62% during high-frequency STN stimulation.37 Similarly, the frequency and duration of thalamic bursts were reduced in response to HFS of the GPi in monkeys.26 Simulations of model thalamocortical relay neurons during 65 Hz and 130 Hz stimulation also indicate that intrinsic bursts were disrupted more by 130 Hz DBS than by 65 Hz DBS.62 However, single-unit recordings in the STN of PD subjects suggest that the cell bodies in the STN may respond with more bursts during STN DBS.19 Four out of 21 neurons switched from an irregular firing pattern before DBS to a burst firing pattern during 140 Hz DBS, and one neuron switched from a burst pattern to an irregular pattern. Rat STN slice experiments demonstrated that STN HFS produced stable bursting activity in some STN neurons with intraburst interspike intervals that were phase-locked to the stimulation frequency or its subharmonics (e.g., one spike every 2 or 3 stimuli).63 In this case, it appears that STN neurons that burst during HFS do so in a regular fashion. However, the dichotomous nature of somatic and axonal firing may prevent the propagation of somatic bursts to the axon terminals. In fact, computational results demonstrated that antidromic invasion of action potentials created by phase-locked high frequency activation of axons can induce somatic bursts.64

Computational and experimental results suggest that neuronal firing becomes more regular during high-frequency DBS as compared with low frequency DBS or stimulation off. The coefficient of variation (CV) of model neuron instantaneous firing rates increased during stimulation at frequencies <50 Hz, but decreased during stimulation at frequencies >90 Hz (FIG. 1B).16 During HFS of the STN in rats with neuroleptic-induced catalepsy, the spontaneous firing rates of a portion of SNr neurons increased, whereas the rates of others decreased. However, “in all of the cases, the SNr firing pattern was regularized.”37 In addition, the entropy of interspike intervals (i.e., a measure of the irregularity or disorder of neuronal spiking) is decreased in GPi and GPe during HFS of the STN in MPTP monkeys, but increased by low-frequency stimulation.65 During low-frequency stimulation in these monkeys, interspike interval distributions returned to the pre-stimulation state around 10 ms after stimulus pulses, which demonstrated a failure to override pathological activity with low-frequency DBS.27 However, high-frequency STN DBS produced a four-phased orderly pattern of inhibition-excitation-inhibition-excitation in the pallidum and thalamus that occurred in the ∼9 ms following stimulus pulses in the STN.27 Similar experiments performed during HFS in the GPi of MPTP monkeys also demonstrated that the intrinsic firing pattern in these neurons was replaced by “predicted, orderly discharge that was locked to the stimulus.”66 These results suggest that effective high-frequency DBS overrides pathological bursts and oscillations in the basal ganglia and thalamus, and replaces these deleterious patterns of activity with regularized firing that is phase-locked to the stimulus.

Additional support for the hypothesis that changes in the pattern of neuronal activity during DBS are more important than changes in the rate of activity is found in the opposite effects of GPe lesions and GPe DBS. Lesions in the GPe of MPTP monkeys worsened motor symptoms; retrieval times in a feeding task were longer than retrieval times prior to the GPe lesion, and total cage movement times decreased.67 Conversely, DBS in the GPe of subjects with PD ameliorated motor symptoms.68 These findings demonstrated that DBS does not act with the same mechanism as lesions (i.e., silencing neurons in the stimulated nucleus). Rather, they suggest that GPe DBS (as with DBS in other targets) overrides pathological activity in the stimulated nucleus.

Changes in metabolic rates of neurons in the basal ganglia during DBS also suggest that DBS alters neuronal firing patterns. Meissner et al.69 measured 2-deoxyglucose (2-DG) accumulation in healthy monkeys, MPTP monkeys, and MPTP monkeys after 10 days of STN HFS (FIG. 3A). All nuclei except the STN showed increased 2-DG accumulation in MPTP monkeys as compared with healthy monkeys. After 10 days of STN HFS, 2-DG accumulation was much closer to that of healthy monkeys than that of MPTP monkeys, suggesting some degree of normalization of activity in the basal ganglia and thalamus during STN HFS. Because 2-DG accumulation is a measure of global synaptic activity, it is intuitive to assume that the spike rates in nuclei with excitatory inputs from the STN would parallel 2-DG accumulation. However, this was not the case (see FIG. 3B); the mean firing rates of GPe and GPi neurons in MPTP monkeys were higher during STN HFS than before STN HFS,27 and the mean firing rates of SNr neurons increased during STN HFS in rats with neuroleptic-induced catalepsy.37 These results suggest that HFS drives seemingly opposite changes in synaptic activity and firing rates in the basal ganglia, which is a difference that may be exacerbated by differences in the duration of HFS in the previously described experiments. The divergence in 2-DG accumulation and firing rate trends with and without high frequency DBS may indicate that 2-DG accumulation depends not only on the firing rate of neurons and synapses in the nucleus of interest, but also on complex changes in the patterns of activity that take place in the same nucleus. Hence, changes in the patterns of activity with DBS may be more important than changes in the rate of activity with DBS.

View Large Image Download to PowerPoint

FIG 3. 2-deoxyglucose (2-DG) accumulation and firing rates in healthy monkeys, 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) monkeys, and MPTP monkeys after chronic high-frequency stimulation (HFS). A: Mean 2-DG accumulation in the basal ganglia and thalamus of healthy monkeys, MPTP monkeys, and MPTP plus subthalamic nucleus (STN) HFS monkeys, expressed as percent of control.69 Measurements provide an estimate of global synaptic activity in the relevant nuclei. After MPTP treatment, 2-DG accumulation increased in all nuclei except the STN where accumulation decreased. During STN HFS, glucose uptake levels returned close to normal. VA, ventral anterior thalamus; VL, ventral lateral thalamus. B: Mean spike rates in the external segment of globus pallidus (GPe) and internal segment of globus pallidus (GPi) in MPTP monkeys before and during HFS, as well as spike rates in substantia nigra pars reticulate (SNr) of healthy rats, and rats with catalepsy before and during HFS. Legend in (A) applies to (B) as well. *Data from an individual monkey in Hashimoto, et al.27 †Data from Degos et al.37

Hypotheses for ineffectiveness of low-frequency DBS 

Here, we present two possible explanations for why low-frequency DBS fails to alleviate motor symptoms and in some cases exacerbates symptoms. The long interpulse intervals present in low-frequency DBS trains may be too long to mask intrinsic activity in the stimulated nucleus, and even in the presence of stimulation, the intrinsic pathological activity persists. Under this hypothesis, low-frequency DBS fails to remove the deleterious patterns of activity in the stimulated nucleus, but instead generates additional spikes on top of the intrinsic activity patterns. These additional spikes further disrupt the patterns of activity in the basal ganglia and thalamus, and through a yet to be defined mechanism, they lead to exacerbation of symptoms. It should be noted that such a change in the pattern of neuronal firing also increases the rate of firing in the stimulated nucleus, and the change in rate may affect symptom exacerbation in conjunction with the changes in pattern.

A second hypothesis is that the lack of effectiveness of low-frequency DBS results from the long interpulse intervals enabling and even promoting rebound bursts in the stimulated nucleus or in downstream nuclei. In other words, low-frequency DBS not only fails to override pathological activity, but actually produces and promotes synchronized pathological activity. Recent computational71 and experimental72 results support this hypothesis. Person and Perkel72 measured the responses of songbird thalamic neurons to stimulation of inhibitory area X inputs (analogous to pallido-thalamic connections in humans). Area X fibers were stimulated with trains containing Poisson-distributed interpulse intervals with an average rate of 140 Hz, and thalamic neurons responded with intermittent rebound spikes and bursts (FIG. 5A). The authors examined the timing of stimulus pulses preceding rebound events and found that in almost all cases, rebound spikes occurred after a pause between stimulus pulses of 20 ms or longer (FIG. 5B). During low-frequency DBS, pulses excite synchronously the synaptic inputs to the stimulated nucleus resulting in inhibition of the post-synaptic cells.73, 74 Pauses in the input (i.e., the interpulse intervals) then enable rebound bursting in the stimulated nucleus, particularly in nuclei with robust inhibitory input. This sequence has also been observed in human thalamus just after microstimulation through the recording electrode.12 These bursts, enabled only when the interpulse interval is sufficiently long, can corrupt the fidelity of thalamic throughput75 and thereby exacerbate motor symptoms. Rebound bursts are also generated in STN slice neurons after pauses in high-frequency activation of GABAA76 or GABAB77 receptors, and these bursts may be propagated to basal ganglia outputs that in turn disrupt thalamic throughout.

View Large Image Download to PowerPoint

FIG 5. Inhibitory postsynaptic potentials drive rebound events in thalamic neurons during stimulation with high frequency Poisson trains. A: Perforated-patch recording of a thalamic neuron during delivery of Poisson distributed stimulation of area X afferents (mean rate of 140 Hz). Tonic hyperpolarization is interrupted by occasional rebound events (inset; scale = 50 ms). IPSPs, inhibitory postsynaptic potentials. B: Raster plot of stimulus pulses preceding rebound events. Each row represents the time that stimulus pulses were delivered during the 200 ms preceding a rebound event; there are 34 rebound events represented and each begins at the black arrow. A pause (i.e., interpulse interval) of 20 to 25 ms (beginning at the white arrow) immediately preceded burst responses in almost all cases. Reproduced with permission from Person and Perkel72 (Copyright, Elsevier Inc. [2005]).

DBS mechanisms in diseases and targets where low-frequency DBS is effective 

The parallels between changes in neuronal firing patterns and motor symptoms as a function of frequency suggest that there is a minimum critical DBS frequency at which symptom alleviation and neuronal regularization occurs.70 However, DBS is effective in some nuclei and diseases at frequencies less than 90 Hz. Early findings suggest that DBS at 25 Hz in the pedunculopontine nucleus (PPN) improves gait in PD.78 Case reports evaluating bilateral GPi DBS for HD yielded conflicting results during DBS at 40 Hz; chorea was improved in one subject,79 but it was poorly controlled in another.80 Finally, studies examining the effects of the frequency of GPi DBS for subjects with dystonia yielded heterogeneous results. Stimulation at 50 Hz81 or 80 Hz82 improved Burke-Fahn-Marsden scale scores better in some subjects than stimulation at frequencies higher than 90 Hz.

A simple, yet plausible, explanation for the effectiveness of low-frequency DBS in some targets and nuclei is that the pathological activity in these nuclei/diseases has a low intrinsic frequency, and thus can be overridden with DBS at lower stimulation frequencies (FIG. 6).16 That is not to say that high-frequency DBS will not work as well in these targets, but that the minimum effective frequency will be lower. First, while the effects of PPN DBS on PPN activity are unknown, firing rates in the PPN of a single PD subject (range, 8–29 Hz in 27 multiunit recordings) suggest the presence of low-rate intrinsic activity of the PPN in subjects with PD.83 Furthermore, cytochrome oxidase subunit I, choline acetyl transferase, and substance P mRNA expression decreased significantly in the PPN of MPTP monkeys as compared with healthy monkeys,84 suggesting that the pathological activity in the PPN has a low intrinsic frequency. Second, although the average rate of GPi neuronal firing is approximately the same in PD and HD, (mean rate ± standard error of the mean = 89.9 ± 3 Hz in PD and 81.8 ± 4.3 Hz in HD), intrinsic neuronal activity in PD includes more bursting and irregular spiking.47 This difference in firing patterns may alter the response of the GPi to DBS and make it possible for GPi DBS to be effective in HD at a lower frequency than in PD. Synchronous burst activity includes many high-frequency spikes over short intervals during bursts (intraburst frequencies can be over 200 Hz). To drive the cell out of bursting, a high-frequency stimulus may be necessary. However, results from 40 Hz GPi DBS in HD were conflicting, suggesting that the difference between the minimal effective frequencies for GPi DBS in PD and HD may not be very large. Third, the intrinsic firing rates of GPi neurons in a population of subjects with dystonia (i.e., heterogeneous etiologies) was much lower than in subjects with PD (mean rates ± standard error of the mean = 95.2 ± 2.3 Hz in PD and 55.3 ± 1.3 Hz in dystonia).85 Similar findings were observed in subjects with cervical dystonia (mean rates ± standard error of the mean = 91.7 ± 3 Hz in PD and 71.4 ± 2.2 Hz in cervical dystonia).86 Finally, simulations of the response of a model neuron to DBS demonstrate that the critical frequency (i.e., the minimum frequency required to mask intrinsic activity) increases as the frequency of intrinsic activity increases (FIG. 6).16 These results suggest that in disease states and nuclei where lower rates of activity accompany the pathology, low-frequency stimulation may provide effective relief of symptoms. Hence, the minimum frequency necessary for maximal symptom relief may vary across DBS targets and disease states.

View Large Image Download to PowerPoint

FIG 6. The minimum (critical) frequency needed to override intrinsic neuronal activity depends on the frequency of the intrinsic activity. A: An integrate-and-fire neuron was stimulated extracellularly, and the output of the axon was recorded. Intrinsic activity (fintrinsic = mean ± standard deviation Hz) was generated by an intracellular somatic current injection. B: The response of the integrate-and-fire neuron with intrinsic activity of 20 ± 5 Hz to no stimulation (0 Hz) and to stimulation at 20 Hz and 100 Hz. Stimulation at 20 Hz superimposes on top of the intrinsic activity, whereas stimulation at 100 Hz completely masks intrinsic activity and replaces it with activity locked to the stimulus train. C: Coefficient of variation of the instantaneous firing rate as a function of stimulus frequency for intrinsic frequencies of 20 ± 5 Hz and 100 ± 11 Hz. Intrinsic activity at 20 ± 5 Hz is masked (coefficient of variation = 0) at a lower frequency than intrinsic activity at 100 ± 11 Hz. Adapted with permission from Grill et al.16