Edited by: Sidarta Ribeiro, Edmond and Lily Safra International Institute of Neuroscience of Natal, Brazil
Reviewed by: Antonio Pereira, Federal University of Rio Grande do Norte, Brazil; Erich T. Fonoff, Hospital das Clinicas of University of São Paulo, Brazil
*Correspondence: Sabato Santaniello, Department of Biomedical Engineering, Institute for Computational Medicine, Johns Hopkins University, Baltimore, MD 21218-2686, USA. e-mail:
†These authors contributed equally to this work.
This is an open-access article distributed under the terms of the
Deep brain stimulation (DBS) of the subthalamic nucleus (STN) directly modulates the basal ganglia (BG), but how such stimulation impacts the cortex upstream is largely unknown. There is evidence of cortical activation in 6-hydroxydopamine (OHDA)-lesioned rodents and facilitation of motor evoked potentials in Parkinson's disease (PD) patients, but the impact of the DBS settings on the cortical activity in normal vs. Parkinsonian conditions is still debated. We use point process models to analyze non-stationary activation patterns and inter-neuronal dependencies in the motor and sensory cortices of two non-human primates during STN DBS. These features are enhanced after treatment with 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP), which causes a consistent PD-like motor impairment, while high-frequency (HF) DBS (i.e., ≥100 Hz) strongly reduces the short-term patterns (period: 3–7 ms) both before and after MPTP treatment, and elicits a short-latency post-stimulus activation. Low-frequency DBS (i.e., ≤50 Hz), instead, has negligible effects on the non-stationary features. Finally, by using tools from the information theory [i.e., receiver operating characteristic (ROC) curve and information rate (IR)], we show that the predictive power of these models is dependent on the DBS settings, i.e., the probability of spiking of the cortical neurons (which is captured by the point process models) is significantly conditioned on the timely delivery of the DBS input. This dependency increases with the DBS frequency and is significantly larger for high- vs. low-frequency DBS. Overall, the selective suppression of non-stationary features and the increased modulation of the spike probability suggest that HF STN DBS enhances the neuronal activation in motor and sensory cortices, presumably because of reinforcement mechanisms, which perhaps involve the overlap between feedback antidromic and feed-forward orthodromic responses along the BG-thalamo-cortical loop.
High-frequency (HF) Deep Brain Stimulation (DBS) of the basal ganglia (BG) is a clinically recognized treatment for movement disorders in Parkinson's disease (PD), but its therapeutic mechanisms are still not fully understood (DeLong and Wichmann,
Studies in 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP)-treated non-human primates showed that, in primary motor cortex (M1), Parkinsonism evokes bursting, synchronous oscillations and decreased specificity to movements, and may cause excessive synchronization between the BG and the cortex (Doudet et al.,
Studies in both normal and 6-hydroxydopamine (OHDA)-lesioned rodents (Li et al.,
Studies on PD patients showed that a single STN DBS pulse produces cortical-evoked potentials (Baker et al.,
This study aims to (i) test the hypothesis in the non-human primate that the effects of STN DBS on cortex vary with the stimulation frequency and involve reinforcement phenomena (i.e., a combination of antidromic and orthodromic effects) above 100 Hz; (ii) to determine whether the cortical response to DBS varies in normal vs. MPTP conditions; and (iii) to determine whether DBS affects the non-stationary dependencies between spike trains of neurons in small cortical ensembles (i.e., recorded from one microelectrode).
We used single unit recordings collected in the primary motor (M1) and sensory (S1) cortex of two non-human primates during STN DBS at 50, 100, and 130 Hz. In one animal, recordings were collected both before and after treatment with MPTP. We developed point process models (Snyder and Miller,
We also investigate the prediction power of the proposed point process models by using tools from the information theory. In particular, the receiver operating characteristic (ROC) curve, the area under the ROC curve (AUC value), and the information rate (IR) are used to measure the impact of the spiking histories on the prediction of the cortical discharge patterns (Bamber,
Point process models have been recently applied to a wide range of neural systems (Frank et al.,
Two non-human primates (
The experiment is described in (Gale,
Once the recording chamber was in place, daily microelectrode recordings were made to identify the sensorimotor region of the STN. Position of the microelectrode was referenced to the stereotactically placed recording chamber and the microdrive coordinates were transferred to the stereotatic atlas of the
Once the STN was identified, a reduced scale model of the human DBS lead was placed with the deepest contact at the bottom of the STN (NuMed Inc., Hopkinton, NY). Anatomical placement of the lead was later confirmed by histological examination (Figure
The electrical stimulation consisted of constant-current symmetric biphasic square-wave pulses, which were delivered between the most distal and the most proximal contact (C0 and C3, respectively). For each pulse, the cathodic phase preceded the anodic phase at C0 (reverse for C3). Pulse width was 90 μs/phase and amplitude was 80% of the current producing tonic contraction (animal A: 0.55 mA; animal B: 0.33 mA), presumably from current spread to the internal capsule. The current amplitude producing tonic contraction was determined during 130 Hz stimulation. We considered three stimulation frequencies: 50, 100, and 130 Hz.
Microelectrode recordings were collected from separate sites of M1 and S1 cortex (pre- and post-central gyrus, respectively) on a semi-daily basis. Somatosensory testing was conducted to delineate motor from sensory cortex for each recording and included (i) passive movement of the animals' limbs, (ii) stimulation of the skin, and/or (iii) palpation of the muscle of the arm and leg. For each recording site, multiple sessions of STN stimulation were made at the frequencies described above and, for each session, continuous recordings were collected 30 s before and 8 to 30 s during DBS. Extracellular action potentials were acquired through platinum-iridium microelectrodes (tip exposure: 10–20 μm; impedance: 0.4–0.6 MΩ; FHC, Inc., Bowdoinham, ME). Electrophysiological signals were bandpass-filtered (0.5–50 kHz) and digitally converted to 25 samples/s for offline analysis. Validated offline software was used to isolate and remove the stimulus artifacts and to discriminate the action potentials from the background noise (Montgomery et al.,
Animal A received initial infusions of MPTP via the right intracarotid artery (0.04 mg/kg) followed by three systemic doses of 0.2 mg/kg, administered intravenously over the course of several weeks, until the animal demonstrated a consistent motor impairment. Effects of the treatment were assessed by observing the animal's spontaneous cage behavior. Because stimulation was applied only while the animal was restrained in the chair, no formal clinical assessment of the effects of DBS was conducted.
At the completion of the study, both animals were anesthetized with ketamine (15 mg/Kg, administered intramuscularly) and monitored until unresponsive to sensory stimulation. After that, animals were administered heparin (10,000 units, administered intravenously), which was allowed to circulate for one minute, and, subsequently, were administered a lethal dose of buthanasia-B. Animals were then transcardially perfused with saline followed by 4% phosphate-buffered paraformaldehyde. Brains were removed and immersion perfused for a minimum of two weeks. Brains were blocked and sectioned in the coronal plane using a cryostat (50 μm thickness). Recording sites in cortex were reconstructed by identification of gliosis along the microelectrode and electrophysiological landmarks (Gale,
Table
0 | 16 | 11 | 74 | 18 | 10 | 6 |
50 | 10 | 5 | 58 | 11 | 10 | 3 |
100 | 7 | – | 56 | – | 9 | 5 |
130 | 3 | 11 | 13 | 16 | 9 | 6 |
0 | 37 | 76 | 240 | 54 | 48 | 6 |
50 | 25 | 40 | 190 | 35 | 48 | 3 |
100 | 21 | – | 188 | – | 46 | 5 |
130 | 3 | 76 | 36 | 47 | 46 | 6 |
The point process models used for our analyzes are described in (Santaniello et al.,
For each neuron, we defined a model for the CIF conditioned on the neuron's own spiking history, the spiking history of any other neuron simultaneously recorded (same ensemble), and the DBS input. The model has the structure (Kass and Ventura,
For each neuron and stimulation frequency, an estimate of Θ along with 95% confidence bounds were provided separately before and during stimulation by maximizing the likelihood of observing the recorded spike trains (Brown et al.,
The goodness-of-fit of each point process model (2–5) was assessed on the validation data by testing the Kolmogorov-Smirnov plot after time rescaling of the spike trains (Barbieri et al.,
Since the spiking propensity λ in (2) is given by a Poisson factor (
In particular, we say that a generic neuron
Similarly, given a pair of neurons (
We evaluated the prediction performance of each model (i.e., how well that model predicts the spike trains of the correspondent neuron) by using the ROC curve (Fawcett,
For each neuron, we also estimated the impact of the spiking histories on the prediction of the spiking activity by computing the information rate (Cover and Thomas,
The testing conditions are denoted below with “###-DBS-normal” or “###-DBS-MPTP,” depending on whether normal or MPTP-treated animals are considered, with ### being the DBS frequency (50, 100, or 130 Hz). “0-DBS-normal” and “0-DBS-MPTP” denotes the no-DBS condition in normal and MPTP-treated state, respectively. Although no formal clinical assessment of the DBS effects was conducted, we refer to 130 Hz DBS and 50–100 Hz DBS as “therapeutic” and “non-therapeutic,” respectively, based on the evidence reported in (Hashimoto et al.,
A total of 135 neurons were recorded in the motor and sensory cortices of two non-human primates (119 in animal A and 16 in animal B). The anatomical sites of the microelectrode recordings containing these neurons and the location of the stimulation leads are shown in Figures
In both animals, the most ventral of the four electrical contacts of the DBS lead was at the ventral boarder of the STN. The most dorsal contact was either in the zona incerta or the ventral thalamus just dorsal to the STN. Thus, bipolar stimulation across the most ventral to the most dorsal contacts spanned the whole STN, with possible additional stimulation of the pallidofugal fibers in the zona incerta. However, this is not inconsistent with the clinical use of STN DBS. Indeed, it is argued that stimulation of the pallidofugal fibers may account for much of the clinical efficacy of STN DBS (Wichmann and DeLong,
Figures
In order to assess the prediction power of the different terms in model (2–5) for each neuron in the dataset, we first computed the probability of spiking of the neuron at every time
Figure
The AUC value provides a further assessment of the prediction power. The AUC depends on both true- and false-positive rates computed for all the possible thresholds on the spiking probability, and ranges from 0.5 (chance level predictor) to 1 (perfect predictor) (Bamber,
0 | 0.53 ± 0.06 | 0.51 ± 0.05 | 0.53 ± 0.04 |
50 | 0.70 ± 0.06 | 0.57 ± 0.07 | 0.59 ± 0.07 |
100 | 0.81 ± 0.09 | – | 0.62 ± 0.11 |
130 | 0.81 ± 0.08 | 0.74 ± 0.09 | 0.60 ± 0.09 |
0 | 0.52 ± 0.05 | 0.53 ± 0.06 | 0.50 ± 0.05 |
50 | 0.62 ± 0.07 | 0.54 ± 0.09 | 0.55 ± 0.04 |
100 | 0.72 ± 0.06 | – | 0.63 ± 0.14 |
130 | 0.65 ± 0.09 | 0.72 ± 0.12 | 0.67 ± 0.07 |
A further assessment of the prediction power of the spiking and DBS histories vs. a history-independent Poisson process was provided by the
0 | 22.3 ± 5.7 | 23.9 ± 0.8 | 24.3 ± 0.9 |
50 | 26.5 ± 1.4 | 26.2 ± 0.4 | 26.1 ± 1.0 |
100 | 28.6 ± 2.0 | – | 26.8 ± 2.1 |
130 | 29.0 ± 2.6 | 27.0 ± 1.0 | 27.4 ± 1.2 |
0 | 23.4 ± 2.2 | 23.1 ± 1.2 | 22.1 ± 1.5 |
50 | 27.5 ± 1.0 | 25.9 ± 1.0 | 25.0 ± 0.0 |
100 | 28.4 ± 1.1 | – | 26.6 ± 1.5 |
130 | 26.9 ± 1.2 | 26.9 ± 1.4 | 27.3 ± 1.0 |
We used normalized post-stimulus time histograms (PSTH, Montgomery,
Examples of PSTH are in Figure
In both cortices, the post-stimulus response was highly temporally consistent (small jitter, Figures
Figure
Interestingly, despite the shape of the PSTH was consistent across the neuronal populations and the animals (e.g., see Figures
The post-stimulus spiking propensity is further characterized by the model parameters {γν}8ν = 1 in (5) (Figures
The recurrent activation patterns were classified as short-term (i.e., period of the pattern between 3 and 7 ms) or long-term (i.e., period between 30 and 50 ms) in order to capture different simultaneous dynamics and compare their relative impact on the neuronal population. These time periods correspond to the parameters {β
With 0-DBS-normal, recurrent short-term patterns [“recurrent fast patterns” (RFPs)] occurred in approximately 30% of the M1 neurons in both animals (Figure
DBS impacted both the activation patterns and the Poisson factor
We found that the value of
In M1 cortex, DBS decreased the fraction of neurons with recurrent patterns (Figures
Figures
In S1 cortex, DBS induced mild modulation of the recurrent activation patterns (Figures
For every ordered pair of neurons (
With 0-DBS-normal, approximately 35% of M1 pairs from animal A and 20% from animal B showed recurrent dependencies (Figures
Figures
In both cortices, the PSTH with 130-DBS-MPTP showed an early temporally consistent response, which could be antidromic. The lag between the first post-stimulus spike and the DBS pulse was smaller with 130-DBS-MPTP than 130-DBS-normal both in M1 cortex (0.96 vs. 1.52 ms) and S1 cortex (0.88 vs. 1.28 ms), and the fraction of neurons with early response was higher with 130-DBS-MPTP than 130-DBS-normal in M1 (73 vs. 30%).
The point process model parameters show that the percentage of S1 neurons with RFPs and FEDs was higher (
In this study we found that (i) MPTP increases the activation patterns and the ensemble dependencies at rest in neurons from the S1 cortex of non-human primates, (ii) the effects of DBS depend on the stimulation frequency and the disease conditions, and (iii) therapeutic 130 Hz STN DBS reduces short-term patterns and dependencies and evokes a short-latency phase-locked increment of the spiking activity, while non-therapeutic 50 Hz DBS reduces the burstiness of the spike trains (Gale,
Our work exploited single unit recordings and point process models of the neuronal spike trains whose prediction power was measured with the ROC curve and the information rate. These measures reveal that the prediction power of the DBS input increases with the stimulation frequency while the spike variability (which is captured by the Poisson factor
The analysis of the point process models (2–5) indicate that, in the M1 cortex (normal conditions), 100–130 Hz DBS suppresses the history-dependent activation patterns and ensemble dependencies, decreases the Poisson factor, and increases the impact of the DBS history (higher AUC and
Pattern regularization and pattern overriding have been speculated as potential therapeutic mechanisms of HF DBS (Montgomery and Baker,
Our analysis shows that RFPs and FEDs are reduced more than the long-term patterns and dependencies with 100- and 130-DBS-normal. Also, the point process models for M1 neurons show higher sensitivity to short-term histories and dependencies during non-therapeutic DBS, and the incidence of RFPs vs. long-term patterns increases under MPTP conditions. These facts suggest that suppressing the RFPs and FEDs could have some relevance for the therapeutic mechanisms of HF DBS.
The lower sensitivity to long-term patterns vs. RFPs is counterintuitive because exaggerated cortical oscillations in the beta band (15–35 Hz, which is covered by the long-term patterns in our study) have been found both in PD patients (Marsden et al.,
The physiological origins of these sequences, however, remain unclear. They could be due to orthodromic stimuli from the cortico-BG-thalamo-cortical loop (Hashimoto et al.,
The abundance of RFPs, however, would be hardly conciliated with synfire chains or BG-thalamic projections, as the period (3–7 ms), which roughly indicates the most likely latency between two consecutive spikes, is close to the time of monosynaptic activation, whereas both mechanisms are polysynaptic. On the other hand, it could be possible that the polysynaptic activation is shortened because of a persistent depolarized state of the neurons, but our data indicates that, under HF DBS, the discharge rate is comparable or even higher than before stimulation, while the RFPs are suppressed. Finally, because we used regular DBS instead of bursts or random pulses, we cannot rule out the possibility that, under HF DBS, the latency between each DBS pulse and the correspondent effects is longer than the inter-pulse interval, i.e., we do not know whether the suppression of an activation pattern is due to the most recent DBS pulse or the pulse preceding the most recent one. In the latter case, the suppression of the RFPs under HF DBS could still be due to the block of polysynaptic mechanisms, although this would be hardly conciliated with the results under 50-DBS-normal (i.e., suppression of neuronal activity but mild change in RFPs).
A plausible, although purely speculative, interpretation could be that these short-term patterns are due to mechanisms intrinsic to the neuronal membranes (e.g., post-refractory period rebound), which are activated by random spikes elicited by multiple, phase-unlocked synaptic inputs (e.g., distal cortical neuron, thalamic projections, etc.). In this case, DBS would affect the neuronal membranes by shifting the rebound timing. For example, according to the PSTHs in Figure
Another important issue here is whether reinforcement mechanisms were evoked by DBS. We reported that 130 Hz DBS evoked highly temporally consistent short-latency responses in both cortices, which would rule out intervening synapses or BG-thalamo-cortical projections, and are most consistent with antidromic activation (Montgomery,
We argue that the antidromic activation elicits a significant response only in a few neurons, while the monosynaptic orthodromic activation has a stronger impact and may be facilitated by recurrent patterns from thalamus, which would determine a depolarized state (Montgomery,
Recordings after MPTP treatment were performed in animal A only and resulted in smaller sets of neurons. Nevertheless, our study indicates that, at rest, (i) there was a larger fraction of S1 neurons with recurrent patterns and dependencies, (ii) the history dependencies had a stronger impact on the spike propensity (Figure
These results suggest higher synchronization in S1 cortex and higher sensitivity to the ensemble activity in M1 cortex at rest under MPTP conditions. Also, the spiking propensity of the M1 neurons at rest is modulated similarly by short- (last 5–6 ms) and long-term spiking histories, and the fraction of neurons with long-term activation pattern reduce under MPTP, thus becoming similar to the fraction of neurons with short-term patterns. This lack of preferred patterns could be due to the suppression of long-term dependencies, which could originate from thalamo-cortical polysynaptic projections.
Finally, the average model parameters correspondent to short-term patterns and ensemble dependencies (10–15 ms lag) are larger under 130-DBS-MPTP than 130-DBS-normal (Figures
The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.
Dr. Sarma was supported by the Burroughs Wellcome Fund CASI Award 1007274, the National Science Foundation CAREER Award 1055560, and NIH R01NS073118-02. Dr. Gale was supported by the American Parkinson's Disease Association. Dr. Montgomery was supported by the Dr. Sigmund Rosen Fund of the University of Alabama at Birmingham.