We included 16 patients with idiopathic PD and DBS and 16 age- and gender-matched HC. From these, we selected 11 PD patients with clinically confirmed FoG episodes (Snijders et al., 2012). We excluded two further PD patients from analysis, one owing to technical problems during the recording and another owing to the inability to walk during the experimental session. Finally, we analyzed data from 9 PD freezers (3 female, age 66.4 ± 7.2 years) and 16 HC (6 female, age: 58.5 ± 4.6 years). Detailed patient characteristics are given in Table 1.

Inclusion criteria were idiopathic PD with akinesia-rigidity type and time since DBS implantation more than 3 months. Exclusion criteria were Mini Mental Status Examination < 22, Beck’s Depression Inventory > 13, and other neurological or neuromuscular disease except PD. The local Ethics committee of the University of Tübingen approved the study (application no. 732/2012BO2) and all subjects provided written consent to participate in the study.

Descriptive are given as mean ± standard deviation, except for the time percentage frozen given as median value, due to a non-parametric data distribution.

In the first part of our analysis we aimed to compare frequency spectra from 1 to 45 Hz. We decided to explore a broad frequency range of interest based on the fact that (i) frequency domain analysis in ambulatory EMG has only sparsely been studied before, and (ii) in cortical and basal ganglia showed diverse abnormalities in this broad frequency range with regard to motor symptoms, freezing or cognitive processes including the theta, alpha, and also broad beta band (de Hemptinne et al., 2015; Tinkhauser et al., 2017; Anidi et al., 2018; Fischer et al., 2018; Hell et al., 2018; Pozzi et al., 2019). First we compared the frequency spectrum of PD freezers in ‘StimOff’ and during ‘regular gait’ with HC in TA and GA. Furthermore, we compared in PD freezers in ‘StimOff’ the frequency spectra (1–45 Hz) of ‘regular gait’ and ‘freezing’ in TA and GA. Then, we analyzed the effect of stimulation of either target (STN_only, SNr_only, respectively) on muscle activity, comparing the frequency domain spectra of PD freezers in ‘regular gait‘ ‘StimOff’ with ‘STN,’ and ‘StimOff’ with ‘SNr’ for TA and GA.

For statistical comparison of the frequency-domain spectra, we used a cluster-based permutation test as implemented in the Fieldtrip toolbox to address for multiple comparisons (Maris and Oostenveld, 2007). This test is based on the Monte Carlo permutation principle and identifies significant changes between conditions using clusters of adjacent frequencies. We performed 5000 randomizations and considered an adjusted two-sided alpha level of p < 0.05 significant.

On average, HC walked for 198 ± 55 s (mean ± standard deviation) with a cadence of 105 ± 10 steps per minute and a step length of 0.46 ± 0.05m. PD patients in ‘StimOff’ walked on average 126 ± 53 s with a cadence of 106 ± 16 steps per minute and a step length of 0.23 ± 0.09 m. In ‘STN,’ PD patients walked on average 157 ± 36 s with a cadence of 101 ± 15 steps per minute and a step length of 0.28 ± 0.08 m. In ‘SNr,’ PD patients walked 151 ± 57 s, with a cadence of 108 ± 18 steps per minute, and a step length of 0.21 ± 0.08 m.

We performed a statistical comparison that revealed significant differences in stride length between HC and PD in ‘StimOff’ (p = 0.00006, Mann–Whitney-U-Test). Since gait speed is known to have an impact on additional kinematic parameters (Fukuchi et al., 2019), we additionally calculated the over ground walking speed in HC: 0.7 ± 0.1 m/s, and in PD in ‘StimOff’: 0.4 ± 0.1 m/s (p = 0.00007, Mann–Whitney-U-Test), which also differed significantly between the two groups. These kinematic differences might also affect underlying electrophysiological differences between HC and PD in ‘StimOff.’ Therefore, we correlated individual gait speed to (i) the frequency of the individual alpha peak (ρ = –0.51, p = 0.16) and (ii) to the relative power of the individual alpha peak (both muscles: ρ = –0.42, p = 0.26; TA ρ = –0.07, p = 0.87; GA ρ: –0.45, p = 0.13). All correlations were negative and do not suggest that lower EMG power relates to gait speed.

In healthy people EMG activity of hand movement and fine motor task is mostly located in the beta band (15–30 Hz) or Piper rhythm (35–60 Hz) (Brown, 2000), in contrast to PD patients, who showed predominant activation around or below 10 Hz in EMG in dopaminergic off state. Effective L-Dopa or STN-DBS therapy lead to suppression of this pathological activity (Salenius et al., 2002; Weiss et al., 2012). However, the relation of these activation abnormalities to PD gait remained unknown. We found that PD freezers show elevated activity from 6 to 45 Hz in the antagonistic leg muscles during effective regular gait, and this was centered in the low frequency range of the alpha and low beta bands. This was present during both regular gait and during the freezes to similar extent. Since the activity was not specific to the freeze itself but existed already during preserved gait, muscular activation at lower frequency around and below 10 Hz may be a general feature of the PD off state mirroring pathological supraspinal activation patterns on the level of the spinal cord (Salenius et al., 2002; Weiss et al., 2012; Flood et al., 2019).

Nevertheless, the finding raises considerations on whether activation at lower frequency contributes to freezing, such that it could represent a more unstable motor system which yields susceptibility to encounter freezing episodes (Scholten et al., 2016b). Activation abnormalities around and below 10 Hz were identified as a pathological feature of the multistage locomotor network comprising cortex (Marsden et al., 2001; Scholten et al., 2016a), basal ganglia (Hammond et al., 2007; de Solages et al., 2010; Chen et al., 2019), brainstem (Thevathasan et al., 2012), and spinal pattern generator (Marsden et al., 2001; Weiss et al., 2012; Scholten et al., 2016a). From these data and the present study it is plausible to reason that these activation abnormalities relate to freezing, since freezers show premonitory activation abnormalities already outside or immediately preceding a freeze in contrast to non-freezers (Singh et al., 2013; Toledo et al., 2014; Storzer et al., 2017). Instability of the motor system may be a pre-requisite to freezing there mirroring the susceptibility of a PD patient to loose effective spinal motor output (Scholten et al., 2016b,2020). During a freeze itself, the pathological rhythm can either stay unchanged or even increase (Scholten et al., 2016a; Anidi et al., 2018; Chen et al., 2019; Pozzi et al., 2019).

In this sense, we speculate that low-frequency activation of the spinal pattern generator of PD freezers could be interpreted as an abnormally slow and prominent rhythm. This might oppose the rapid adjustment of antagonistic muscle activity as is needed for the recurrent cycling of activation and deactivation cascades throughout the cyclic gait phases. Mathematically spoken, motor output in the high beta and gamma range > 20 Hz would allow for much faster reprogramming and adaptation of the spinal pattern generator (Schoffelen et al., 2005) as opposed to the activation abnormalities at slower rhythms observed in this freezer group. This would lead – depending on the individual cadence – to a much slower adaptation and also performance of the alternating activation – deactivation changes of the antagonistic leg muscles throughout the full gait cycle, which can only interact in the slow alpha and low-beta rhythms in PD. In healthy subjects activation of the TA will take for about 10% of the gait cycle, from initiation to full recruitment (Hart et al., 2006). Considering a cadence of around 105 steps per minute in our cohort of healthy controls indicates a duration of the gait cycle of 0.57 s and a necessity of adaptions within 57 ms. The prominent slow alpha and low-beta rhythms in PD patients will not allow for more than a temporal resolution of around 50–100 ms, which is inherently slow to allow rapid adjustments of the gait cycles and leads to abnormal temporal activation patterns in PD freezers (Nieuwboer et al., 2004). Moreover, the ability to adapt gait to the external or internal requirements is limited at such slow frequencies, and this may comply with the clinical observation that gait performance can be disrupted with cognitive interference (Nutt et al., 2011; Nonnekes et al., 2015). Instead, it was stabilized with rhythmic cueing, as was the oscillatory beta activity time-locked to the gait cycle (Fischer et al., 2018).

Besides the enhanced alpha and low beta activity our data also reveal higher power spectra > 20 Hz in GA, which could be due to (i) either postural differences in PD patients, caused by a more flexed posture (Nieuwboer et al., 2004) and a tendency to shift the center of gravity forward when walking, resulting in toe walking or (ii) reflect compensatory mechanisms caused by cortical drives to the spinal cord which are located in a higher frequency range (Schoffelen et al., 2005). To this point it is rather difficult to disentangle whether this high frequency activation in GA is a causal or compensatory mechanism in the first place.

Subthalamic nucleus -stimulation in PD is a potent treatment for L-Dopa sensitive FoG. Clinical studies reported on reduced occurrence of FOG and sever falls in PD freezers after undergoing STN-DBS (Schlenstedt et al., 2017; Barbe et al., 2020; Cebi et al., 2020). However, to prevent adverse outcomes of STN-DBS it is important to avoid co stimulation of the pallido-thalamic tract crossing on the level of the zona incerta as was for example found in antero-medially displaced electrodes or by delivering large amounts of energy and increasing the electrical field (Moreau et al., 2008; Fleury et al., 2016). When applying effective STN-stimulation along with clinical improvement also pathological changes of gait parameters in PD freezers, such as stride length and stride amplitude improve (Pötter-Nerger and Volkmann, 2013; Scholten et al., 2017). Notably, high frequency stimulation of the STN also led to a reduction of beta-band activity in STN-LFPs. In our experiment the kinematic parameters do not reveal significant differences between ‘STN’ and ‘StimOff.’ Scholten et al. (2017) report results of a larger group of PD freezers in which gait parameters, temporal and spatial, showed improvement with STN- or SNr-stimulation. Especially the stride length improved when applying STN-DBS. In the subgroup we choose for our data analysis also the stride length improves in STN-condition, but the changes are not statistically significant, probably due to the smaller size of the cohort. Additionally also the absolute number of freezing episodes across subjects and the absolute time frozen improved when we applied STN-DBS in contrast to the ‘StimOff’-condition, and this clinical improvement was associated with a reduction of pathological activity at low frequency in TA and GA. The modulation effect on oscillatory patterns of STN-DBS is not locally limited to the STN, but may impact on the functionally connected areas, in particular the subthalamo-cortical circuits (de Hemptinne et al., 2015; Weiss et al., 2015). On the one hand STN has an excitatory net effect on SNr and GPi, resulting in more inhibitory control on PPN/MLR which was brought in context to FoG (Shine et al., 2013; Weiss et al., 2019). STN-DBS is attenuating the exaggerated glutamatergic output (Benabid et al., 2003) and the pathological beta oscillations (Kühn et al., 2008; Eusebio et al., 2012). This means that STN-DBS could potentially act on releasing the pallidothalamic inhibition of the primary motor cortex, or by modulating the cortex more directly via the hyperdirect connections. However, toning down pathological STN activity could also change SNr activity through the monosynaptic connections (Milosevic et al., 2018; Weiss et al., 2019). Thus, observing a change of muscular activity with STN-DBS would not necessarily mean that the effect is transmitted via the ‘ascending’ cortical and then corticospinal pathway, but would still allow for contributions of the ‘descending’ nigropontine route. However, our second finding that SNr_mono stimulation did not affect both muscular activity and clinical FoG measures argues against this alternative interpretation. Therefore, we propose that the primary effect of high-frequency STN-DBS was delivered via the subthalamo-cortical circuits of either the indirect or the hyperdirect pathway – or maybe both (Gradinaru et al., 2009; de Hemptinne et al., 2015). STN holds projections via the subthalamo-pallido-cortical ‘hyperdirect’ and the striato-external pallido-subthalamo-internal pallido-cortical ‘indirect’ pathway, both executing inhibitory control on thalamo-cortical activation patterns and preventing effective motor output (Delwaide et al., 2000). STN-DBS can effectively modulate the oscillatory patterns of cortical areas via these pathways and electrophysiological changes are accompanied by improved clinical outcome parameters (Kuriakose et al., 2010; de Hemptinne et al., 2015; Weiss et al., 2015). Considering the fact that in our experiment clinical as well as oscillatory improvements were achieved via STN-DBS, but not SNr-DBS, it seems reasonable to consider the underlying pathological enhance alpha and low beta activity also being primary transmitted via the subthalamo-cortical pathway and the descending projections of the pyramidal tract to spinal motor neurons. Further support for this hypothesis comes from studies on the pathology of ULF, when freezing episodes were associated with increased cortical activity (7–11 Hz), muscular activity (6–9 Hz), and increased intermuscular coherence, the latter of which was proposed as marker for cortical control of muscular activity (Scholten et al., 2016a). Nevertheless, cortical contributions do not exclude subcortical contributions to muscular activation at low frequencies. Future research using combined EMG-LFP-EEG measurement in gait paradigms will help to further investigate the underlying network interactions of pathological activation pattern in FoG (Kühn et al., 2009; Lewis and Barker, 2009; Anidi et al., 2018; Pozzi et al., 2019). Especially high resolution time-frequency analyses of LFP-EEG data could help differentiate a primary cortical or subcortical source of the muscular activation abnormalities of the alpha and beta band.

In this study, we investigated the role of muscular activation abnormalities at alpha and beta frequencies for FoG taking advantage from ambulatory EMG recordings, and studied their supraspinal modulation with STN- and SNr-DBS. We were able to analyze data from a very homogenous group of finally 9 PD freezers with an acinetic-rigid subtype, susceptibility of FoG and electrode localization of the most caudal contact being located in SNr, a group size comparable to those of previous electrophysiological studies (Anidi et al., 2018; Pozzi et al., 2019). Although observing a large quantity of freezes under laboratory conditions has been recognized as a challenge in FoG research (Lewis and Shine, 2016; Weiss et al., 2019) we were able to gather an absolute time of 144 s frozen in ‘StimOff,’ which can be considered sufficient data material to stabilize the frequency domain spectra from a statistical viewpoint (Mima and Hallett, 1999). A limitation is that the data of freezing episodes stem from only 5 of the 9 subjects which limits the interpretation and generalizability of comparing the spectra of the freezing state with regular gait. The same applies for comparisons of clinical features, i.e., number of freezing episodes, absolute time frozen, etc. The descriptive indicate a marked reduction of freezing, considering the number of patients expressing FoG (4 in ‘StimOff,’ 1 in ‘STN’), the number of freezing episodes (15 in ‘StimOff,’ 2 in ‘STN’) and the absolute time frozen (144 s in ‘StimOff,’ 50 s in ‘STN’). Still the subgroup is quite small and performing a Wilcoxon signed rank test would not show statistical significant improvement. But we were therefore careful in interpreting our finding, and suggest to leave it to future studies to test, whether muscular activity shows a further increase during freezes as compared to regular gait. Nevertheless, the above summarized evidence across the distributed freezing-network levels draws a coherent picture that activation around and below 10 Hz in basal ganglia, brainstem, cortex, and spinal motor neurons is critical to freezing susceptibility, i.e., representing a general failure of neuronal gait integration in PD freezers that yield a risk for expressing freezing episodes on this grounds.

We performed several correlation analyses of clinical measurements and electrophysiological parameters. There was no correction for multiple comparisons, since we performed the correlation analysis with exploratory intent and did not interpret them in a confirmatory way. Instead, we suggest to reproduce these findings in independent and larger cohorts.