Human striatal recordings reveal abnormal discharge of projection neurons in Parkinson’s disease

Increased Firing Frequency of SPNs in PD.

SPNs of patients with PD were spontaneously overactive, firing at much higher frequencies in comparison with other patient groups (examples in Fig. 1 A–D). In the baseline parkinsonian state in absence of antiparkinsonian drugs and in the resting condition, the mean firing frequency of SPNs was 30.2 ± 1.2 Hz (between 13.5 and 47.9 Hz). Similar increases of activity were found in both putamen and caudate areas, but putamen neurons fired at slightly higher frequency than those in the caudate nucleus (P < 0.01; Fig. 1E). The SPN activity in patients with PD was increased by more than 10-fold compared with the very low firing frequencies (2.1 ± 0.1 Hz) found in patients with ET (P < 0.001). In addition, the SPN activity in patients with PD was 3-fold higher than in patients with ID (9.3 ± 0.6 Hz), thus establishing a distinct level of hyperactivity in PD (Fig. 2A).

The very low SPN firing found in patients with ET is congruent with the classic description of normal activity in animals (24), which was also shown here in the normal NHP using the same procedures and data analyses as in human recordings (SI Materials and Methods). These correlative findings suggest that the striatal function is spared in ET and support the use of this comparison as a substitute for normal subjects. The comparison between patients with ET and ID showed SPN firing frequencies moderately increased in ID (Fig. 2A). The escalation of frequency changes from ET to ID and again to PD reveals two pathological states of the striatum with SPN activity changing in the same direction but reaching an exaggerated level in PD. The strict comparison of only putamenal neurons increases the difference between PD and ID because SPNs of patients with PD had higher firing rates in the putamen than the caudate nucleus.

The SPN firing frequency in the parkinsonian macaques was markedly increased by comparison with the normal animal (23.3 ± 1.6 Hz and 1.6 ± 0.1 Hz, respectively; P < 0.001; Figs. 1F and 2B). In NHPs [one normal macaque and two advanced parkinsonian macaques modeled with chronic 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) administration], a total of 52 SPNs were recorded (control NHP, n = 18; parkinsonian NHPs, n = 34). The SPN firing frequencies in these animals were in the range previously reported for the parkinsonian and normal NHP conditions (18). The increase in SPN activity from normal to advanced parkinsonian NHPs was similar to that from patients with ET to PD. Furthermore, the MPTP-induced change in SPN firing frequency was computed to determine the mean predicted probability of PD. This analysis that included all 155 SPNs from 26 patients with ET, ID, or PD showed that the increase of SPN firing frequency to a minimum of 13 Hz (the minimum observed in the parkinsonian macaque) significantly predicted PD in patients (logit model, χ²: 151.6, P < 0.001, 93% accuracy; Fig. 2C).

Increased Burst Firing of SPNs in PD.

SPNs of patients with PD at rest fired not only faster, but frequently with brief spike bursts composed of 10.5 ± 1 spikes, 7 ± 0.3 ms intraburst interspike intervals (ISIs), and 72 ± 9 ms duration (Fig. 3A). Firing of bursts was found in 50% of the recorded SPNs in patients with PD, which were designated as “bursty” SPNs at a minimum of 1 detected burst. The rate of burst activity per bursty SPN in patients with PD was 9.5 ± 1.2 bursts per 10 s. Burst activity was present to a similar extent in putamen and caudate SPNs of patients with PD (P = 0.3). This activity pattern was not present in SPNs of patients with ET in whom the characteristic occasional isolated SPN spikes were followed by prolonged silences compatible with the low firing rate of these neurons in the normal animal. Firing of bursts was also detected in SPNs recorded in patients with ID (Fig. 3B). However, compared with patients with PD, the SPN burst activity was lower in patients with ID (firing of bursts was found in 29% of the recorded SPNs in patients with ID at a rate of 4.8 ± 0.8 bursts per 10 s, P < 0.001; Fig. 3 C and D). The parameters for burst detection using the “surprise” method (25) were selected with strict limits for unequivocal classification of short duration spike bursts in all groups of patients. This method avoided the detection of longer spiking epochs as bursts in the typically irregular but overactive spike train of SPNs in PD that would have resulted in overestimation of burst activity (SI Materials and Methods). Nevertheless, we applied additional methods of burst analysis, including some that are less influenced by varying firing frequencies (26–28). The percentage of neurons that fired with bursts was similar, but the burst rate differed across detection methods, revealing the influence of frequency. However, significant differences in SPN burst activity between PD, ID, and ET were consistently found with all detection methods (Fig. S3). Therefore, there is a clear increase in the SPN burst activity in PD. Notably, the progression of burst firing from ET to ID and further to PD was similar to changes in the frequency domain, suggesting that common mechanisms probably generate the higher firing rates and bursting.

SPNs of the studied parkinsonian macaques also fired with bursts, as detected with the same burst parameters and algorithms as used for analysis in SPNs of patients. Burst firing in SPNs of these animals was as frequently detected as previously reported (19) (41% of the recorded SPNs in the two MPTP-treated NHPs fired with bursts at the rate of 4.4 ± 0.6 bursts per 10 s). Also in agreement with previous data, spontaneous firing of bursts in SPNs was not observed in the normal animal during recordings at rest (SPN burst activity has been described in normal primates in association with active movement) (29). This difference in the SPN activity pattern between the normal and parkinsonian state in NHP reproduces the changes observed between patients with ET and PD (Fig. 3 C and D).

Patients.

A significant number of surgical cases were excluded from the study because of the electrode trajectory, particularly with thalamic DBS targets in the ET group. Also all procedures under anesthesia, which involved the majority of patients with ID, were excluded. Nevertheless, with time a significant number of striatal units could be classified and analyzed in patients with PD, ID, and ET. Sex, age, disease duration, clinical score, and surgical target in each case are provided in Table 1. All patients were assessed by movement disorder specialists 1–5 d before surgery. Patients with PD were assessed off and on medication with the Unified Parkinson’s Disease Rating Scale III (UPDRS III) (55). Patients with ID were evaluated with the Fahn-Marsden Scale (FMS) (56). The severity of tremor in patients with ET was rated using the Tremor Rating Scale (TRS) (57). The ET diagnosis of patients included in the study was based on strict exclusion criteria that involved the presence of any dystonic features or PD-like symptoms. All regular medications in patients with PD (levodopa, ropinirole, apomorphine, entacapone, pramipexole, rasagiline, and antidepressants), ID (trihexyphenidyl, baclofen, pyridostigmine, and benzodiazepines), and ET (gabapentin, nadolol, metoprolol, primidone, and alprazolam) were withheld for more than 12 h before surgery.

Electrophysiology in Patients.

Standard methods of microelectrode-guided stereotactic surgery for the implantation of DBS electrodes into the target basal ganglia nuclei were used. The target selection for DBS (GPi, STN, or VIM) was decided on the basis of clinical features and predicted effects of DBS on different regions to achieve the best clinical improvement (see the target distribution in Table 1). Preoperative MRI was performed to identify the target and plan the trajectory of DBS electrodes. The theoretical coordinates to target were taken from MRI-guided targeting system (Framelink software, Medtronic) and copied into MRI-based proprietary algorithm where they were adjusted if needed (Onetrack software developed at Emory University). The surgical procedure was performed under local anesthesia, and the complete neurophysiological mapping was done without any sedative agents. Extracellular recordings were performed with tungsten microelectrodes (impedance adjusted to 0.1–0.3 MΩ at 1.0 kHz), and signals (50 kHz sampling frequency) were amplified, filtered (0.3–2 kHz), digitized, and stored using a data acquisition system (Axon Guideline System 3000, FHC). The tungsten electrodes were conditioned with a metallic coating that facilitated isolation of SPNs for several minutes as in standard NHP recordings; however, typical platinum-iridium electrodes were occasionally used to verify data compatibility with either electrode in both human and NHP recordings. Neuronal activity was collected in the dorsomedial and dorsolateral putamen or the dorsolateral caudate depending on the DBS target, but neurons near the limits of these areas according to final mapping analysis were excluded to avoid inclusion of border cells. The number of collected cells in each patient depended on electrode penetrations and time constrictions in each individual surgery.

Electrophysiology in Nonhuman Primates.

Three adult NHPs, species Macaca (Ga and Ad: Macaca fascicularis, males, 5–7.5 kg; Gu: Macaca mulatta, female, 5 kg) were include for comparative striatal recordings in two conditions, normal and parkinsonian state. Two animals (Ad and Gu selected to include one of each subspecies and each sex) were rendered parkinsonian by systemic administration of MPTP until producing a model of advanced parkinsonism following procedures previously described (18). These animals had a moderate to severe motor disability score (58) and were maintained under levodopa treatment (oral Sinemet, 100–200 mg/d). The third animal (Ga) was kept as a normal control for recordings. All animals were trained to sit quietly in the primate chair and had standard surgical implants of recording chambers. Microelectrode-guided mapping of basal ganglia was used before collection of data for analysis. All recordings proceeded in the morning, and on those days the antiparkinsonian medication was withdrawn. Tungsten microelectrodes (same impedance as above) were used for striatal single-cell activity recordings, and signals (20 kHz sampling frequency) were amplified and band-pass filtered with 100–8,000 Hz (Multichannel Acquisition Processor, Plexon). All electrode tracks were postcommisural and mostly placed in the posterolateral areas of the putamen (motor territories). The middle portion of the caudate nucleus was also recorded in some electrode tracks. The sets of data collected in each animal were similar. The same methods for driving the electrode and recording spiking in the striatum as described in the human recordings were applied to NHP recordings. Data with acceptable signal to noise ratio were saved for 3–4 min for off-line analysis. At the end of the study, animals were euthanized and their brains examined histologically for confirmation of recording areas based on electrode tracks (Fig. 1F), as previously reported (19).

Analysis of Neuronal Activity.

Human and NHP data were analyzed off-line with the same methods. Spike sorting (Offline Sorter, Plexon) separated units with spontaneous activity. Threshold-crossing spike detection followed by cluster separation using the principal component analysis was applied. The analysis can distinguish the waveforms of SPNs and interneurons (TANs and FSIs). Fig. 1 B–D shows examples of cluster separation. After spike sorting, the established criteria to classify single neuronal activity as pertaining to SPN were applied (59). These criteria are: (i) waveform, (ii) firing frequency, (iii) ISI histogram, and (iv) autocorrelogram. Firing frequency was not used for classification of SPNs, but the following criteria and numerical cutoffs were used: waveform of short duration (0.9–1.4 ms), variable/irregular spiking with unimodal ISI histogram (peak close to 0 ms and large variation of ISIs including long intervals of ≥1 s), and a central peak or “narrow” central valley in autocorrelograms. TAN classification: bi- or multiphasic waveform of long duration (>1.8 ms), 2–15 Hz firing frequency, tonic activity with unimodal or “bimodal” ISI histogram, and a “wide” central valley in autocorrelagrams (Fig. S2) (53, 60, 61). FSI classification was as follows: small and short wavefrom (<0.9 ms), higher than 18 Hz firing frequency with or without bursts, unimodal ISI histogram without intervals >1 s, and a central peak in autocorrelograms (62, 63). All units classified as TANs and FSIs or unclassified units (uncertain location and ambiguous parameters) were excluded from further analysis.

The total duration of the spike train was included in postprocessing analysis. Firing frequency (rate meters with 1 s/bin width) and ISIs versus time were computed using Matlab. Burst activity was analyzed with the surprise (probability)-based algorithm using preselected parameters of burst detection and a surprise threshold s ≥ 3 (NeuroExplorer, Nex Technologies) (19, 25). Parameters for burst detection were based on typical short epochs of spiking that could be unequivocally classified as burst activity. Limits were as follows: maximal interval to start (10 ms), maximal interval to end bursts (10 ms), minimal interburst interval (20 ms), and minimal number of spikes or burst duration (3 spikes and 20 ms). These parameters set a definition of bursts that avoided the detection of longer spiking epochs as bursts. After running the analysis with the preselected parameters, neurons were separated in two groups depending on positive or negative burst detection. After burst detection, the surprise value was examined for acceptance (≥3). Neurons with bursts detected at s ≥ 3 were classified as bursty with the obtained results (burst number, rate, etc.). Neurons with bursts detected at s < 3 were subjected to repeated analysis setting a surprise threshold ≥3, which limited detection to those with a higher surprise value, and the characteristics of detected bursts were verified to be similar to the selected short epochs (this resulted in a lower burst number in few neurons). SPNs were thus grouped as bursty or nonbursty, and burst parameters (spike number in burst, intraburst ISI, burst duration and rate per 10 s) were analyzed. As we recognized the effect of firing frequency on burst detection using the surprise method (25), we applied additional burst detection methods including some that are less sensitive to firing rate and compared all results. We used the robust Gaussian surprise (RGS) (28), rank surprise (RS) (27), and ISI-threshold (ISI_N) methods (26). All methods were computed using Matlab. The RGS method detects bursts with a robust adaptability to higher firing frequency. We used a P value of 0.05 to calculate the central location of all ISI distributions, and selected three as the minimal number of spikes to be considered a burst. The RS method applies a nonparametric approach for burst detection, which is based on the ranks of ISIs in the entire spike train. The minimum surprise value to accept the detected burst with this algorithm was 4. The ISI_N method is also an efficient burst detection method, which does not depend on ad hoc or post hoc detection criteria and was set to detect similar bursts. In each analyses, the proportion of neurons with burst activity and the burst rate per 10 s were calculated for each patient group (PD, ID, and ET) and NHP group (normal and parkinsonian), and results were compared with ANOVAs followed by Bonferroni post hoc tests in patients or unpaired t tests in NHP. Results of burst detection with each method are shown in Fig. S3.

Statistics.

The SPN firing frequency was compared among patient groups using ANOVAs (one-way analysis of variance), and followed by Bonferroni post hoc test after the ANOVA F indicated statistical significance, and between caudate and putamen in patients with PD using unpaired t tests. The SPN firing frequency was compared between the control and parkinsonian animals using unpaired t tests. To determine the predictability of PD based on SPN hyperactivity, binary logistic regression was used for a yes/no model constructed with the minimum firing rate in the parkinsonian NHP and including all patients with ET, ID, and PD (the goodness of fit was assessed with the Hosmer and Lameshow test). Burst rate in SPNs between patients with PD and ID or between caudate and putamen in patients with PD was compared using unpaired t tests. All values are expressed as mean ± SEM (significance: P < 0.05).

Patients.

The Institutional Review Board of Emory University reviewed and approved the study, and all patients gave their informed consent. Patient inclusion was based on the following conditions: (i) an established, unequivocal diagnosis made by a movement disorder specialist, (ii) recordings in DBS surgery without general anesthesia or sedative drugs, and (iii) electrode trajectories passing through the striatum as permitted by the individual surgical approach and the intended mapping of the DBS target area. Because selection of patients for surgery depended solely on medical indication for DBS treatment, their clinical characteristics could not be matched across groups (Table 1). Patients were asked to relax and not to move during the recordings to obtain data of spontaneous neuronal firing at resting conditions and avoid movement-related activity. Tremor at rest was not observed during the striatal recordings in the majority of patients, including many patients with PD. The recording method was the same not only across the groups of PD, ID, and ET patients, but also in nonhuman primates. As SPN recording is difficult with rapid movement of the electrode, the electrode was lowered very slowly with a hydraulic microdrive in patients to simulate the small steps (10–20 µm; NAN Instruments) used in animal recordings. Typically, spiking could be found variably in the striatal tracks as the electrode advanced at a slow speed. Neurons with very low activity or pausing as the electrode was advanced could have been undetected similarly in patients with PD, ID, and ET. After listening to or seeing a spike, the movement stopped to record the activity independent of how fast or slow the spiking was. Then, the position of the electrode was slightly adjusted for best unit isolation, and after firing stability was verified, data were saved for 1–3 min. We established this minimum duration of the spike train to be accepted for analysis to avoid very short recordings that are inadequate to determine the firing frequency of neurons with variable activity. Single spike traces where the unit could be classified directly on-line as TAN interneurons by the typically large and long waveform (≥1.8 ms) were not saved (53). All other spike trains were stored for off-line analysis.

NHP.

Studies in NHPs (three adult macaques) were conducted in accordance with the NIH Guide for the Care and Use of Laboratory Animals (54).

Details of methods and data analyses are provided in SI Materials and Methods.