Alpha oscillations in the pedunculopontine nucleus correlate with gait performance in parkinsonism

Subjects and clinical assessments

Seven patients with Parkinson's disease implanted with deep brain stimulation electrodes in the PPN region were assessed (Table 1). Following their initial implantation, electrodes were attached to special extension cables ‘externalized’ through the scalp to permit connection with recording equipment. Final implantation of deep brain stimulation hardware was therefore deferred for the period of ‘externalization’ (3–7 days). Eleven PPN electrodes were recorded from the seven patients. Patients were recruited from hospitals in Oxford and London, UK. Local ethics committee approval was obtained from both centres and participants gave written informed consent. Reaction time results from one of the patients have been reported previously (Thevathasan et al., 2010, 2011b).

The indication for PPN stimulation was severe gait freezing and postural instability, persisting even ‘ON medication’ and either causing frequent falls or precluding walking. In Parkinson's disease, gait freezing and postural instability become more frequent and less responsive to medication with disease progression (Giladi et al., 2001; Bloem et al., 2004). The overall prevalence of gait freezing and postural instability in Parkinson's disease is ∼50% (Macht et al., 2007). However, severe medication resistant gait freezing and postural instability as the predominant issue is unusual in Parkinson's disease and raises the question of atypical pathologies (Factor, 2008; Jankovic, 2008). As there is no definitive test in life, we stress that the diagnosis of Parkinson's disease in our series is presumptive.

Preoperative assessments included the motor subsection (part III) of the Unified Parkinson's Disease Rating Scale (UPDRS; score/108). OFF medication assessments occurred after overnight withdrawal (>12 h) of dopaminergic therapy. UPDRS was segmented into items 27–30 (IT27/30, score/16) assessing posture, gait and balance and residual items 1–26 (R-UPDRS, score/92) assessing bradykinesia, rigidity and tremor. Patients treated in Oxford also prospectively completed the Gait and Falls Questionnaire (score/64) which assesses Parkinsonian gait disturbance including gait freezing, festination and falls (Giladi et al., 2000). The Freezing of Gait Questionnaire (score/24) and Falls Question (score/4) are components of the Gait and Falls questionnaire (Giladi et al., 2000, 2009). For all motor scales, higher scores indicate worse function. The London patient was assessed with UPDRS (part II) items assessing gait, freezing and falls (combined score/16).

Techniques to target and implant electrodes in the PPN have been described previously (Pereira et al., 2008; Zrinzo et al., 2008; Foltynie and Hariz, 2010). In two early studies of stimulation in this vicinity, electrodes appeared to be placed lateral to the lemniscal system—in the peripeduncular nucleus rather than the PPN (Mazzone et al., 2005; Stefani et al., 2007; Yelnik, 2007; Zrinzo and Hariz, 2007; Zrinzo et al., 2007). In the current study, the PPN was targeted medial to the lemniscal system and lateral to the superior cerebellar peduncle and its decussation. Electrodes (Medtronic) were configured with four active contacts each spanning 1.5 mm. Electrodes in six patients were model 3389 (0.5 mm spacing between contacts) and in one patient were model 3387 (1.5 mm spacing between contacts). Contact coordinates were assessed on postoperative MRI or postoperative CT fused with preoperative MRI and transformed onto Montreal Neurological Institute (MNI) space using the fMRIB Software Library (Smith et al., 2004). Coordinates were calculated in millimetres from midline (laterality), ventrodorsal distance (d) from floor of the fourth ventricle and rostrocaudal distance (h) from a pontomesencephalic line connecting the pontomesencephalic junction to the caudal end of the inferior colliculi, as described previously (Ferraye et al., 2009). Electrode locations are summarized in Fig. 1A.

Experiments and recordings

Experiments took place 2–6 days after electrode implantation. Assessments were performed ‘OFF medication’, after overnight withdrawal of dopaminergic medication, to limit variance from fluctuating dopaminergic state and to maximize gait disturbance.

There were two conditions: rest and gait. For rest recordings, patients sat comfortably with eyes open for 2–3 min. For gait recordings, patients walked at their preferred speed along an unobstructed path 10 m long, 10–30 times (depending on speed and fatigue).

Local field potentials were recorded in bipolar configuration from consecutive contacts (01, 12 and 23) of each electrode (Fig. 2A). Local field potentials were band-pass filtered between 0.5 and 500 Hz. Single channel EEG was recorded from FzCz (frontal zero central zero). In Patients 1–6, a triaxial accelerometer (TMSI) was firmly fixed with tape over the spinous processes at the upper thoracic level to record trunk acceleration. A portable battery powered amplifier (Porti amplifier, TMSI) was attached with a waist belt. Data were sampled at 2048 Hz and recorded on a laptop (Porti32 software) via a long fibre-optic cable.

Data analysis and parameters

Data were analysed in Spike 2 (Cambridge electronic design) using routines therein. Unless stated otherwise, all frequency spectral analyses employed non-overlapping blocks of 4 s duration that afforded a resolution of 0.25 Hz (fast Fourier transform, Hanning window). The mean (±SEM) duration of recordings at rest was 222 ± 38 s and for gait was 253 ± 68 s.

Statistics

The Kolmogorov–Smirnov Test demonstrated that the distribution of normalized peak local field potential power and relative gait speed was not different from the normal. Normalized local field potential power in specific frequency bands was compared between conditions and/or recording sites using paired t-tests. Differences in normalized local field potential power across a range of frequencies were assessed with serial t-tests performed for each frequency. Correlations (Pearson's) were sought between peak local field potential frequencies across conditions and between normalized local field potential power and relative gait speed. Level of significance was P < 0.05.

Gait assessment

Following electrode implantation and without stimulation, all patients experienced a reduction in gait freezing, which persisted up to 6 weeks—a time course consistent with ‘stun effect’ (Koop et al., 2006). Preoperatively, when ‘OFF medication’, every patient experienced frequent, long duration freezing. However, postoperatively when ‘OFF medication’ during experiments, freezing in five patients was infrequent and then typically brief (<3 s). In two patients (who were both implanted unilaterally), freezing had improved but remained clinically severe. In one of these patients (Patient 6) discrete freezing episodes were identifiable while in the other (Patient 2) freezing was almost continuous. Festination was not observed during experiments. The mean cadence (±SD) across patients for every 4 s block of walking was 1.81 Hz (±0.28). Relative gait speed coefficients of variation ranged from 40 to 308% between patients, consistent with some degree of gait disturbance, despite the stun effect. Cadence was less variable, with coefficients of variation ranging from 6 to 21% between patients.

Local field potentials and their relationship to electroencephalography and gait

Therapeutic impact of chronic stimulation on gait

Therapeutic outcomes from chronic PPN region stimulation are presented in Table 2. Pre- and postoperative Gait and Falls questionnaire scores were available in all six cases operated in Oxford, affording six estimates of therapeutic impact. Gait and Falls questionnaire scores in these six patients improved with PPN stimulation [mean 42.0–29.8, t(5) = 4.425, P = 0.007]. The location where stimulation was applied along the rostrocaudal axis was calculated relative to the pontomesencephalic line. Where stimulation was bilateral, stimulation depth was the mean from both sides. The overall improvement of Gait and Falls questionnaire scores with PPN stimulation was only moderate. However, closer examination within this small sample suggests a possible relationship between stimulation depth and percentage improvement in the Gait and Falls questionnaire with the best outcome being achieved when stimulation was applied at the level of maximal alpha activity (Fig. 4B). Thus, a relationship was sought between power in the alpha band where stimulation was applied and therapeutic outcomes. With monopolar stimulation, alpha power was taken as the mean power recorded from the two flanking bipolar pairs. A correlation between mean log alpha power at rest and per cent postoperative Gait and Falls questionnaire improvement confirmed a correspondence between these measures (r = 0.85, P = 0.033). The picture was similar if the per cent improvement (20%) in UPDRS II items scoring freezing, falls and gait was included from the London patient in whom these measures were used to assess outcome (r = 0.81, P = 0.027).

Topographical distribution of oscillations

A rostrocaudal topographical organization of the PPN is supported by extensive data from animal research (Martinez-Gonzalez et al., 2011). Rostral PPN neurons predominantly express γ-aminobutyric acid and are strongly interconnected with the basal ganglia including subthalamic nucleus and internal pallidum (Mena-Segovia et al., 2004; Ros et al., 2010; Martinez-Gonzalez et al., 2011), which exhibit beta activity in untreated Parkinson’s disease (Hammond et al., 2007). Caudal PPN neurons express predominantly acetylcholine and glutamate (Martinez-Gonzalez et al., 2011). Cholinergic neurons arborize widely, including to cortex and locomotor centres (Skinner et al., 1990; Mena-Segovia et al., 2008). It is therefore congruent that we found beta oscillations in the rostral PPN region and a distinct oscillatory activity in the caudal PPN region, alpha activity, coherent with cortex and associated with locomotion. This interpretation is also supported by the one previous study in which PPN local field potentials were assessed in parkinsonian patients according to recording site depth (Weinberger et al., 2008). In that study, beta oscillations were recorded in an area corresponding to what is here defined as the rostral PPN region. In an example local field potential recording, a spectral peak can also be identified in the alpha band in what is defined here as the caudal PPN region (1–4 mm below the inferior colliculus). Of further interest, neurons considered by the authors as likely cholinergic (having long duration action potentials and low firing rates) were found predominantly in the caudal PPN region (Weinberger et al., 2008). Such topography could explain why different centres have reported different dominant oscillatory patterns, in the alpha or beta frequency bands, from PPN recordings—as this would be determined by the depth of the implanted electrodes (Androulidakis et al., 2008a, b; Weinberger et al., 2008; Tsang et al., 2010)

The origin of the focal alpha activity in the PPN region needs consideration. The distribution of electrode contacts in this study was broad—and this facilitated the relationship between oscillatory power and location to be recognized. Some contacts lie more caudal than any described location of the PPN but these were not the ones expressing alpha at highest power across patients. Across patients, un-normalized peak alpha power at rest, which correlated strongly in frequency and location with alpha power during gait, was maximal between 2 mm and 6 mm beneath the pontomesencephalic junction. The spatial accuracy of this finding is limited by the bipolar nature of the local field potential recordings and the averaging of data across patients. However, with this limitation in mind, a key question is whether this region corresponds to the PPN. The PPN comprises a well-defined ‘pars compacta’ subregion but also a reticular ‘pars dissipata’ with indistinct boundaries (Olszewski and Baxter, 1954). The atlas of Olszewski and Baxter (1982) identifies the PPN based on cytoarchitectural methods. Later, cholinergic PPN neurons were identified in humans by immunohistochemical labelling of choline acetyltransferase (Mesulam et al., 1989; Manaye et al., 1999). Although these latter studies do not provide stereotactic coordinates, their relevance has been heightened by the involvement of cholinergic PPN neurons in gait and its dysfunction (Karachi et al., 2010) and several observations are of localizing value. While cholinergic PPN neurons were clustered most densely in the pars compacta, the pars dissipata accounts for the greater proportion of cholinergic neurons (Manaye et al., 1999). Cholinergic PPN neurons span at least 7 mm in the rostrocaudal plane being at highest density 2.5 mm under the rostral pole (as defined by choline acetyltransferase staining, so potentially missing any rostral component not expressing choline acetyltransferase; Manaye et al., 1999). Cholinergic PPN neurons were noted to lie where the ‘superior cerebellar peduncle ascends towards the dorsolateral pons towards its decussation’ with the pars compacta located at the level of the decussation (Mesulam et al., 1989; Manaye et al., 1999). Identifying these anatomical landmarks on the axial images of Patient 3 suggests that the levels 2–4 mm below the pontomesencephalic line would therefore correspond to the caudal PPN region. Other gait related entities also exist in the PPN region including the cuneiform and subcuneiform nuclei, which together with the PPN comprise the mesencephalic locomotor region (Alam et al., 2011). Both the PPN and cuneiform nucleus are activated during fast imagined gait in functional MRI studies in healthy subjects (Karachi et al., 2010). Microelectrode recordings have suggested that neurons that modulate firing in response to imagined gait actually tend to be located in the subcuneiform region dorsal to the PPN (Piallat et al., 2009). However, boundaries between these various nuclei are indistinct—potentially confounding precise determination of the source of neuronal recordings and of the structures responsible for clinical effects of stimulation in this region. On the other hand neurons of the PPN, but not the cuneiform or subcuneiform, are reported to degenerate in association with falls in Parkinson's disease (Karachi et al., 2010). Furthermore, cytoarchitectural atlases suggest that the cuneiform and subcuneiform nuclei do not extend as caudally as the zone where we recorded maximal alpha band activity (Olszewski and Baxter, 1982).

The distribution of PPN oscillatory activity concurs with previous work in Parkinson's disease suggesting that alpha and beta oscillatory networks are segregated (Litvak et al., 2010). The functional nature of alpha and beta oscillations in Parkinson's disease also appears distinct. Beta band activity is pathologically increased in Parkinson's disease, is suppressed by l-dopa and high-frequency subthalamic stimulation and correlates with deficits of bradykinesia and rigidity (Silberstein et al., 2005; Kuhn et al., 2006; Eusebio et al., 2010). Alpha power in the PPN tends to increase with l-dopa, suggesting that it could be pathologically attenuated in Parkinson's disease (Androulidakis et al., 2008a, b). It now also appears that attenuation of PPN alpha activity is associated with gait freezing whilst increases in PPN alpha power in Parkinson's disease correlated with improved gait.

Pedunculopontine nucleus alpha oscillations and locomotor function

A potential confound needs consideration; that the correlation between alpha power and gait speed could be due to movement artefact. Several factors militate against this. First, our local field potential recordings were bipolar, so a signal common across contacts should have been subtracted out. Second, the relative gait speed correlated maximally with alpha oscillations (7–10 Hz), which differed in frequency from the accelerometer frequencies used to derive gait speed (1–3 Hz). Third, correlation with the relative gait speed was relatively specific to alpha local field potential activity, evidenced by the peak in correlation spectra at alpha frequencies and the lack of any correlation with beta power. Fourth, correlations were present even when local field potential power was normalized to broad band activity, which effectively eliminated any common artefact from movement.

An important qualification is that the correlation between gait speed and alpha power does not necessarily imply causation and the relationship could be epiphenomenal. For example, alpha power may be permissive for higher gait speeds or reactive to it. However, correlations were found within a brain area that is functionally relevant to gait disturbance, as evidenced by the potential for improvement upon PPN stimulation. Furthermore, the positive nature of the correlation between alpha synchronization and increasing gait speed, concurs with the synchronization of alpha power observed with l-dopa (Androulidakis et al., 2008a, b), which also tends to improve gait. Yet, even if PPN alpha oscillations are mechanistically important in gait disturbance, the relationship does not appear obligatory or exclusive. The mean correlation was 0.43, so that fluctuations in alpha power only predicted ∼20% (r²= 0.19) of the variance in relative gait speed. A further question is whether this alpha activity also relates specifically to gait freezing. Due to the reduction in freezing postoperatively due to a possible stun effect, we could only compare freezing episodes with alpha power in a single patient, who was implanted unilaterally. This demonstrated an attenuation of alpha activity related to the onset of freezing. This suggests that higher gait speeds could have, at least partly, resulted from relief of freezing related deficits even if discrete freezing episodes did not frequently occur. For example, patients with gait freezing are reported to have continuous background deficits in gait, including reduced gait velocity, step length and increased spatiotemporal variability (Hausdorff et al., 2003b; Chee et al., 2009; Snijders et al., 2011). It is notable that attenuation of alpha local field potential power preceded freezing onset. However, this latter finding should be interpreted cautiously given the limited accuracy in determining when exactly freezing begins.

Thus, a particular limitation of the current study was the limited characterization of gait performance in terms of relative gait speed and the analysis of freezing in only one patient. Gait dysfunction in Parkinson's disease is complex, with varying involvement of features such as gait initiation, turning, postural instability and gait velocity (Bloem, 1992; Morris et al., 1994; Hausdorff et al., 2003b; Sofuwa et al., 2005). The failure to capture these features might also help explain why fluctuations in alpha power only predicted ∼20% of the variance in relative gait speed.

By what mechanism could PPN alpha activity relate to gait performance, as indexed by relative gait speed? Gait speed reduces in healthy subjects, elderly fallers and in Parkinson's disease during the performance of a second, unrelated task (‘dual tasking’) (Hausdorff et al., 2003a; Springer et al., 2006; Lamoth et al., 2011). In elderly subjects, an inability to ‘walk whilst talking’ predicts falls (Lundin-Olsson et al., 1997). Such findings implicate ‘attention’ and the effective allocation of processing resources that flows from it, as a potentially important factor influencing gait speed. In Parkinson's disease, attentional deficits are common and there is also impaired automaticity of movement so that processing demands are higher (Wu and Hallett, 2005, 2008). Parkinsonian patients with gait freezing are reported to have even more attentional deficits than those without gait freezing (Amboni et al., 2008; Yogev-Seligmann et al., 2008). Dual tasking can worsen gait freezing, as can other precipitants that are thought to ‘distract’ attention away from gait (Giladi and Hausdorff, 2006).

There is increasing evidence that alpha activity has an important role in attention and the allocation of processing resources. The synchronization of occipital alpha with eye closure was once interpreted to reflect passive ‘idling’ (Berger, 1929; Pfurtscheller et al., 1996). However, alpha activity is now considered to support active suppression of task irrelevant processes (Jensen and Mazaheri, 2010). For example, during working memory tasks, cortical alpha power in visual and motor–sensory areas increases and the degree of synchronization correlates with the number of items recalled (Jensen et al., 2002; Haegens et al., 2010). Correlation of performance with oscillatory power in task irrelevant regions suggests suppression rather than mere idling. It has been proposed that within the motor system, suppression of competing processes with alpha could aid the smooth execution of motor programmes (Pfurtscheller and Neuper, 1994; Suffczynski et al., 2001). Regions such as the subthalamic nucleus and caudal PPN that have distributed functional connectivity, including with cerebral cortical areas, would be placed to operationalize this putative role. Thus PPN deep brain stimulation is able to increase blood flow in an extensive network of subcortical and cortical areas involved in balance and motor control (Strafella et al., 2008; Ballanger et al., 2009).

Clinical relevance to pedunculopontine nucleus stimulation

Gait freezing improved following electrode implantation in the absence of any stimulation in all patients, although this was less pronounced in patients implanted unilaterally. Gait freezing is notorious for improving during medical assessments, perhaps due to attentional or placebo effects. However, we observed that the improvement in gait freezing persisted for up to 6 weeks post-implantation—a time course arguably more consistent with a stun effect. In the subthalamic nucleus, the ‘stun’ or ‘microlesion’ effect of surgery is usually attributed to suppression of beta activity in the target nucleus by acute tissue disruption from electrode implantation (Chen et al., 2006; Koop et al., 2006). However, in the caudal PPN, alpha activity persisted despite the apparent stun effect and correlated positively with gait performance. Accordingly, we speculate that stun effects in PPN surgery might predominately arise from microlesioning inhibitory supraspinal influences along the electrode trajectory. Excessive inhibition of the PPN (for example from the internal pallidum) has been considered a pathophysiological factor causing gait and postural disturbance in Parkinson's disease, in addition to PPN neuronal degeneration (Aziz et al., 1998). Consistent with this reasoning, in the 1-methyl-4-phenyl-1,2,3,6-tetrahydropyridine (MPTP) primate model of Parkinson's disease, the microinjection of a γ-aminobutyric acid antagonist into the PPN improved mobility (Nandi et al., 2002). Microlesioning inhibitory afferents to the PPN would be expected to have analogous effects. If this reasoning is correct, then the occurrence of a possible stun effect would not necessarily indicate accurate electrode placement within the PPN, nor necessarily portend benefit from stimulation.

The topographic arrangement of oscillations in the PPN region along the rostrocaudal axis raises the possibility that caudal and rostral PPN stimulation could have different therapeutic effects. Alpha and beta oscillatory activity could provide functional biomarkers for these different subregions. Alpha and beta activity in the PPN region were focal, evidenced by un-normalized oscillatory power at rest dropping by an average of 42% for alpha activity and 45% for beta activity across bipolar recording sites. This is similar to the gradients of beta activity reported across the dorsolateral subthalamic nucleus that can be detected in intraoperative and postoperative recordings and used to guide deep brain stimulation electrode implantation (Chen et al., 2006).

Only the alpha activity recorded in the caudal PPN region was demonstrated to be functionally related to gait and gait freezing. Therapeutic outcomes from PPN region stimulation were variable between patients as were the locations along the rostrocaudal axis where stimulation was applied. Interestingly, the best outcome was achieved when stimulation was applied at the level of maximal alpha activity, 2–6 mm below the pontomesencephalic junction, so that per cent improvement in Gait and Falls questionnaire score at different depths closely correlated with the alpha power recorded at the same depths. We take this to be an encouraging preliminary finding, noting that patient numbers were small, unilateral and bilateral stimulation were mixed, postoperative periods were relatively short and our outcome measure was not objective. Accordingly, it is important that future studies examine whether the efficacy of PPN stimulation relates to stimulation location ‘within’ patients, preferably using objective gait measures. Still, the current study is important in providing physiological evidence, supported by consistent clinical findings, that the site of stimulation along the rostrocaudal axis of the PPN region might be a factor determining therapeutic outcomes.