Patterns of Bidirectional Communication between Cortex and Basal Ganglia during Movement in Patients with Parkinson Disease

Patients and electrode implantation.

All patients gave their informed written consent to participate in this study, which was approved by the ethics committee of the Hospital for Neurology Pierre Wertheimer (Lyon, France), in accordance with The Code of Ethics of the World Medical Association (Declaration of Helsinki, 1967). We studied 12 patients (eight men and four women; mean age, 59 ± 6 years; range, 51–65) with PD for 6–25 years (all idiopathic, except one with the Parkin gene mutation). Handedness was assessed using the Edinburgh, Oldfield, Handedness Inventory (Oldfield, 1971), and all patients, except one, were right handed.

Patients underwent bilateral implantation of deep-brain electrodes in the subthalamic nucleus (STN) for the treatment of severe PD (Table 1). All patients had good levodopa responsiveness [mean reduction in the United Parkinson's Disease Rating Scale (UPDRS) motor score by levodopa, 6 months preoperatively: 76 ± 8%] and severe motor complications. The exclusion criteria were as follows: age ≥70 years old, cognitive decline (score <130 on the Mattis Dementia Rating Scale), major depression on the Beck Depression Inventory (score >20), abnormal brain magnetic resonance imaging (MRI), and concomitant severe medical pathologies.

The surgical procedure has been detailed in a previous publication (Thobois et al., 2002). Briefly, STN was located by MRI and ventriculography. Implantation was done under local anesthesia. Microrecordings (Guideline System 3000a; Molecular Devices, Foster City, CA) and macrostimulation were performed to determine the best trajectory among the three tested for each side (Radionics, Burlington, MA). Then, definitive electrodes were implanted bilaterally (model 3389-28; Medtronic, Minneapolis, MN). Brain MRI was performed postoperatively to check the final electrode placement and exclude surgical complications. At least one contact was considered to lie in the dorsal STN after postoperative MRI. Six days later, a subcutaneous programmable pulse generator (Itrell II or Kinetra; Medtronic) was implanted under general anesthesia and connected to the implanted electrodes. Seven of the 12 patients have so far reached their 1 year follow-up and were assessed OFF medication/ON and OFF stimulation. In these, UPDRS motor scores exhibited a 54 ± 16% decrease from OFF medication/OFF stimulation (33.3 ± 10) to OFF medication/ON stimulation (16 ± 9). Across all 12 patients, levodopa equivalent medication fell from 1125 ± 135 mg (SEM) preoperatively to 350 ± 84 mg during postoperative follow-up (paired t test: t₍₁₁₎ = 5.10; p < 0.01). These observations (Table 1) are consistent with adequate targeting.

Recordings and paradigm.

Data were collected 5 d postoperatively in the interval between macroelectrode implantation and subsequent connection to a subcutaneous stimulator. Subjects were comfortably seated in front of a table during a 100 s rest period and while they made externally paced single taps (alternating extension and flexion at the proximal metocarpophalangeal joint) with the dominant index finger on a custom-made tapping device: taps had to be performed approximately every second after a regular tone given by a metronome. Two sessions of seven series of 20 taps separated by 20 s rest periods were recorded. Note that the metronome was kept on during the recording performed at rest to avoid any difference in auditory processing between the two conditions. Each recording was performed after overnight withdrawal of antiparkinsonian medication (OFF medication) and 1 h after administration of the patient's usual morning levodopa equivalent dose plus 25% (ON medication).

Scalp electroencephalogram was recorded according to the 10:20 international system from an electrode pair over C3 and FC3 in right-handed patients or C4 and FC4 in the left-handed patient. In addition, we recorded from Cz and FCz. Note that C3/4 and Cz approximately overlie the hand area of the sensorimotor cortex, including the primary motor cortex (M1) and the SMA, respectively, although electrodes here will not solely pick-up activities from these areas. EEG electrodes were referenced to linked ears, but a bipolar extraction of C3/4-FC3/4 and Cz-FCz was performed off-line to limit volume conduction between candidate areas. The sparseness of EEG recordings was dictated by surgical dressings.

STN LFPs were recorded from the four contacts of each DBS macroelectrode, referenced to linked ears. Thereafter, we performed an off-line extraction of bipolar signals from adjacent contact pairs 01, 12, and 23, in which contact pair 01 was the deepest. This was also performed to limit volume conduction. Signals were amplified, bandpass filtered between 0.5 and 90 Hz (6 dB per octave; Biopotential Analyser Diana, St. Petersburg, Russia), and recorded onto a personal computer using custom-written software (A. Pogosyan). Sampling rates were 184 Hz for five patients, 1000 Hz for one patient, and 1200 Hz for six patients. In the latter two instances, data were interpolated off-line to a common sampling frequency of 184 Hz. Overall, the risk of aliasing was very small, given that sampling rates were just above twice or more the steep low-pass filter.

Data analysis.

Data were analyzed using Spike2 software (Cambridge Electronic Design, Cambridge, UK) and Matlab (Mathworks, Natick, MA). EEG signals and simultaneously recorded LFP signals were divided into 10 s segments either at rest or during tapping, when the period between 2 and 12 s after the onset of each tapping series was selected. We analyzed STN LFPs from the contact pair contralateral to the dominant hand that showed the highest β band power at rest OFF medication, based on evidence that this activity is generated in dorsolateral “motor” STN (Kühn et al., 2005; Chen et al., 2006; Weinberger et al., 2006; Trottenberg et al., 2007).

Using a multiple autoregressive model (MAR), we calculated signal power at three sites [the lateral sensorimotor cortical area (C3/4-FC3/4, hereafter termed “M1” solely for convenience), the central cortical motor area (Cz-FCz, hereafter termed “SMA” for convenience), and STN] and evaluated the possible DTF based asymmetries of flows between cortical–subcortical sites pairs (M1 to STN, STN to M1, SMA to STN, STN to SMA) for both motor states (rest and during movement) and both dopaminergic states (ON and OFF medication). A detailed description of the methodology and principles of the DTF can be found in the studies by Korzeniewska et al. (2003) and Cassidy and Brown (2003). Briefly, the DTF investigates any possible asymmetry in the flow of coherent activity between sites (Kaminski and Blinowska, 1991). To this end, the MAR model that best described the signals coming from the three sites of interest (M1, SMA, STN) was determined. Unlike power measures, the MAR methodology is essential for calculating the DTF, because the DTF is built directly from the MAR coefficients. Power, however, can also be calculated from the model for comparison with the DTF. Following the procedure detailed by Cassidy and Brown (2003), a Bayesian methodology was applied to estimate the parameters of the autoregressive model. This approach is desirable in that it provides full probabilistic distributions for all of the model parameters. The autoregressive coefficients can be used to construct a bounded, normalized measure (the DTF) that provides information on the effective direction of coupling between each site. In a previous study (Sharott et al., 2005), a Bayesian methodology was also used to calculate the model order. Although model order selection is important per se, it has become clear that within a sensible range it alters the spectra relatively little (Schlögl and Supp, 2006). Given the computational burden of computing the model order for each data epoch, we therefore selected a constant model order of 10, which is ample for the number of frequency components being investigated (Schlögl and Supp, 2006). Multivariate analysis was performed, which, at any given frequency, consists of directed coherence (dCoh) values giving the coherence in each direction between the three sites. In cases in which multiple channels are analyzed multivariate, compared with bivariate, analysis gives a potentially more accurate result, because all these channels are incorporated in the same MAR model. Indeed, in these cases, bivariate analysis can give confusing results (Kus et al., 2004). For any pair of sites, if the dCoh is significantly greater in one direction, the DTF can be considered “asymmetric” in that direction. Conversely, if the dCoh estimates are not significantly different, the DTF can be considered “symmetric” (Sharott et al., 2005).

Statistics.

Power data were logged, and dCoh data were Fisher transformed for subsequent statistical analysis. All statistical analyses were conducted using Statistica version 6 (StatSoft, Tulsa, OK). Log power and transformed dCoh data were averaged across three different frequency bands: sub-β (3–13 Hz), β (14–35 Hz), and γ (65–90 Hz). These discontinuous frequency bands were selected because several previous studies have shown that spectral peaks and physiologically reactive LFP power in the STN are seen below 35 Hz and again from ∼65–90 Hz (Brown et al., 2001; Williams et al., 2002; Priori et al., 2004; Alegre et al., 2005; Fogelson et al., 2005; Alonso-Frech et al., 2006; Devos and Defebvre, 2006; Kühn et al., 2006a,b; Androulidakis et al., 2007). The precise frequency range of the β band was chosen to match (as far as possible, given differences in sampling rates) the band determined to be focally generated in the STN in recent intraoperative (Kühn et al., 2005; Chen et al., 2006) and postoperative (Kühn et al., 2006a) recordings. This frequency range (14–35 Hz) extends a little higher than in previous publications, but the extension of the band is further justified by two studies that include the extent of spectral peaks in individual nuclei. In these, STN LFP spectra in 5 of 16 (Kühn et al., 2004) and 13 of 26 (Doyle et al., 2005) nuclei had peaks involving or extending into 31–35 Hz. Similarly, although previous publications defined the γ band as extending to 80 Hz (Alonso-Frech et al., 2006) or 85 Hz (Williams et al., 2002; Silberstein et al., 2003), a more recent report demonstrated that spectral peaks in the LFP of PD patients treated with levodopa in the range of 62.3–93 Hz (Trottenberg et al., 2006). The definition of a broad “sub-β band” (3–13 Hz) that includes θ and α bands was motivated by the need to preserve statistical power.

For each brain site, we separately analyzed log power data using a 3 × 2 × 2 ANOVA with repeated measures on all factors: frequency band (sub-β, β, γ) × dopaminergic state (ON vs OFF) × motor state (rest vs movement).

With respect to the DTF, we first checked whether experimental dCoh in each direction was significantly different to that derived from shuffled data. These control data sets (in which no significant flow is expected between all three brain regions) resulted from the randomization of each series of STN and EEG values recorded for each patient, in both dopaminergic states at rest and during movement. dCoh data were first examined with six 2 × 2 × 2 × 2 ANOVAS with repeated measures on all factors: dopaminergic state (ON vs OFF) × motor state (rest vs movement) × flow direction (cortex:STN vs STN:cortex) × shuffled state (experimental or unshuffled data vs control or shuffled data). All ANOVAs were separately applied for each of the three frequency bands (sub-β, β, γ) and cortico–subcortical pairs of sites (M1/STN, SMA/STN). These ANOVAs were principally performed to establish whether experimental dCoh exceeded control (shuffled) dCoh but also demonstrated whether there were differences between conditions.

Second, we performed two 3 × 2 × 2 ANOVAs with repeated measures on all factors: frequency band (sub-β vs β vs γ) × flow direction (cortex:STN vs STN:cortex) × pair of brain sites (M1/STN, SMA/STN) at rest for each dopaminergic state (ON and OFF). These ANOVAs allowed a comparison across bands and pairs of sites. Note that we did not explore dCoh between mesial and lateral electroencephalography from the multivariate analysis, because this was likely to be biased by volume conduction effects and so as not to weaken statistical analyses through the inclusion of too many factors.

Data entered in ANOVAs were nonspherical (as determined by Mauchly's tests), so we applied Greenhouse-Geisser corrections to the resulting p values. We used Fisher least significant difference post hoc tests where appropriate. Corrected p values ≤0.05 were considered significant, although those near 0.05 should be interpreted with caution.