Insights from other pathologies
Whilst it is widely acknowledged that FOG is not unique to PD and is frequently observed in atypical parkinsonian syndromes, high-level gait disorders, normal pressure hydrocephalus, vascular diseases, and other neurodegenerative diseases [
30,
31], it is unclear as to whether they share common pathophysiology. Furthermore, due to a lack of well conducted, large-scale observational studies, many basic elements of FOG, such as the frequency/duration of episodes, gait pattern generation, imbalance and the impact of cues are not well characterised across these other freezing conditions. Clearly, not all of these disorders have a profound loss of dopamine or response to treatment, but most are described as parkinsonian, potentially reflecting disturbances in motor networks that may be structural or functional.
These non-PD groups are clearly challenging to recruit and study. Therefore, future prospective studies with harmonised data collection across multiple centres to achieve sufficient statistical power should be constructed with open de-identified data sharing. Simple clinical data (e.g., motor, cognitive, and psychiatric assessments) collected across all subjects may highlight hitherto unrecognised relationships. A more rigorous examination, in a smaller number of patients at dedicated centres, should include detailed gait kinematics and neuromechanics with simultaneous measurements of axial motion (dynamic posturography) and EMG [
32], along with standardised neurophysiological (e.g., ambulatory and seated EEG) and neuroimaging (e.g., structural/resting state MRI, dopamine/dopamine transporter imaging) data collection. Comparison of these data with the FOG and balance disturbances observed in PD would allow modellers to build more accurate network models that capitalise on real-world perturbations across clinical, neurophysiological, and neuroimaging data to help identify the relevance of contributing pathways.
Insights from disease progression and treatment
Given the progressive nature of FOG, any complex modeling approach will require high-quality longitudinal data that record motor and non-motor features, as well as medication use. Clearly, the increased prevalence of FOG with disease duration might suggest a dopaminergic aetiology [
33], but it must be appreciated that with disease progression there will also be increasing pathology across multiple neurotransmitter systems and a breakdown in the functional/structural connectivity across disseminated brain networks [
34,
35]. Furthermore, it has also been suggested that increasing dopaminergic therapy in the advanced stages might have a causal role in the development of FOG, particularly in relation to the rare phenomenon of ON FOG [
36].
To address the respective roles of disease progression and medication use, large collaborative studies with standardized data collection and recording are required. A variety of observational and interventional studies could be considered, each of which would have significant issues regarding feasibility. Perhaps most simply, one could envisage leveraging from planned prospective natural history studies like the Parkinson Progression Marker Initiative 2.0 (PPMI 2.0—NCT04477785), which will establish a deeply phenotyped cohort assessing the progression of clinical features, digital outcomes, as well as imaging, biologic, and genetic markers in study participants with de novo PD, prodromal PD, and healthy controls. The addition of more extensive gait assessments to a dedicated study like PPMI 2.0 with additional cognitive, affective, autonomic, sleep, and daily physical activity measures, combined with regular, standardised assessments at home and in the office as discussed above, would improve our understanding of the protective and provocative factors for developing FOG. Such a study could be conducted in newly diagnosed patients or potentially in a more enriched cohort of patients, five years from initial diagnosis who have an ‘at-risk’ phenotype (e.g. anxiety, non-tremor dominant, impaired repetitive motor task performance, executive impairments) with a higher likelihood of transition from being non-freezers [
37,
38]. One further approach could be an interventional study where newly diagnosed (drug naïve) patients would be randomised to different treatment arms to explore the role of L-dopa, dopamine agonists, and monoamine oxidase-B inhibitors. This approach could be utilised to determine the differential impacts of delayed initiation of L-dopa and dopamine agonists versus those starting therapy at diagnosis, and the effect of the dosing level. Indeed, it has been hypothesised that levodopa may induce maladaptive plasticity in the striatum, which disproportionally increases the mismatch between motor and non-motor (cognitive and limbic) loops, leading to gait freezing (the levodopa paradox) [
6]. Previous lines of evidence from the pre-levodopa era and observations in third-world countries would seem to refute this assertion [
39]. Constructing the necessary prospective study to address this issue would prove prohibitively expensive and would obviously pose significant challenges for recruitment and retention. Therefore, retrospective chart review may offer a more pragmatic approach (see below). It would seem unlikely that even such an interventional trial would help our understanding of ON FOG, which is a rare phenomenon where there is a worsening of FOG following L-dopa [
40]. Previously, ON FOG has been addressed by kinematic studies in the ON and OFF states during an appropriately rigorous levodopa challenge, including serum levodopa levels [
41], but longitudinal assessments are now required to determine whether OFF FOG evolves into ON FOG (where freezing is seemingly caused by levodopa), ON–OFF FOG (where FOG persists in the ON state) or if they develop and evolve separately. These insights would have specific consequences for any modeling approach, as well as our definition of the phenomenon and its treatment [
42].
Insights from non-gait freezing
One critical aspect that could be exploited in our understanding of FOG is whether freezing is restricted to gait or represents a more universal phenomenon. The concept of ‘motor blocks’, where sudden episodes of motor breakdown are provoked by repetitive upper and lower limb tasks, as well as by speech sequences has long been recognised [
43,
44]. These freezing episodes in other effectors also typically present with faulty initiation-termination responses, particularly when progressing towards the end of an automated motor sequence. When patients who experience FOG are required to perform declining movement amplitudes at fast speed within self-initiated sequencing tasks, there seems to be a consistent degradation of the neural coding of movement cycles, triggering episodes of motor output breakdown. Regardless of the functional activity, impairments in the accurate coding of the motor network appear to disable the normal motor re-initiation. For example, one recent study that required patients to perform accelerated weight-shifting sequences without stepping, demonstrated greater disturbances in freezers compared to non-freezers, which were exacerbated in OFF [
45].
These behavioural observations regarding non-gait freezing have prompted efforts to identify any common neural correlates. Previously, neurophysiological studies have described beta oscillations as the ‘idling state’ of the brain and that voluntary movement requires a desynchronization of this activity. Significantly, volitional movements in PD are associated with impaired desynchronisation in these beta oscillations [
46], and pathological beta activity has been identified within the subthalamic nucleus of PD patients with FOG [
47]. Indeed, suppression of this beta activity either by open- [
48] or closed-loop [
49] DBS has been shown to ameliorate freezing episodes either during normal walking or when stepping in place. Furthermore, recent work recording subthalamic activity from chronically-implanted DBS electrodes has identified that FOG is characterised by a low-frequency cortical-subthalamic decoupling, which is lateralized to the hemisphere with less striatal dopaminergic innervation [
50]. Analogous to these findings in gait, a study utilising cortical EEG to investigate a finger sequencing task has shown that compared to non-freezers, PD patients with FOG have increased beta oscillations (i.e., reduced desynchronisation) in the supplementary motor area prior to volitional movements [
51]. In addition, a separate EEG study examining the effect of a dual-task on finger tapping has highlighted that increases in prefrontal beta-band synchronization are predictive of upper limb freezing [
52]. Thus, as well as highlighting the neurophysiological similarities between upper limb and gait freezing, this study also underscores the potential contribution of prefrontal executive dysfunction, which has been described in FOG [
53,
54]. The role of dopaminergic pathways in non-gait freezing has been less well explored to date. One study reported that dopamine replacement did not influence the frequency of events during wrist flexion/extension [
55], whilst a virtual reality (VR) gait paradigm where patients utilized foot pedals has shown amelioration of freezing-related phenomena [
56]. Further work utilizing more automatic finger movement or handwriting paradigms [
57] is required to confirm these observations.
Despite these overlapping neurophysiological features, freezing in the upper limbs has been observed in a substantial proportion of non-gait freezers [
57], suggesting that this phenomenon may capture a substrate of FOG but not the full picture. However, it should be highlighted that in a recent prospective cohort of 60 non-freezers, assessed prospectively for two years (12 convertors), repetitive finger tapping was found to identify those patients at risk of developing FOG [
38] and is therefore worthy of further consideration.
Paradigms do exist to assess non-gait freezing that could specifically explore trembling in place, the sequence effect, and the role of treatment (e.g., medication and DBS). Force sensors, keyboards, and smartphone apps can all be used to quantify motor blocks during foot pedalling [
58‐
60] and alternating hand/finger tapping [
61,
62]. Indeed, many studies have been conducted utilising a VR gait paradigm where freezing episodes recorded from foot pedal movement have been correlated with actual FOG [
58]. To date, combining this VR gait paradigm with fMRI has identified the neural correlates of freezing [
63,
64] and related triggers including turning [
65], doorways [
56,
66], and dual-tasking [
67]. Furthermore, this technique has been utilized to record multi-unit activity in the subthalamic nucleus (STN) during DBS surgery and demonstrated a pathological surge of beta activity prior to the onset of a freeze that differed from the recordings associated with volitional stopping [
68]. In addition, this beta-band activity was unidirectionally and selectively linked with STN theta activity, which in turn was unidirectionally and selectively linked with the 3–8 Hz trembling in EMG activity seen in the lower limb muscles that activated the foot pedal. Similar to manipulating lower limb conditions for eliciting FOG, a simple test whereby subjects have to vary the size of writing strokes to fit in a funnel figure at fast speed, has also been able to elicit finger freezing episodes that can be correlated with self-reported FOG [
57].
Therefore, these non-gait freezing paradigms are potentially valid models for studying FOG with multi-modal techniques. Additionally, repetitive tasks can be remotely employed via telemedicine platforms, which are relatively inexpensive and could serve as safe proxy markers or predictors of FOG, providing significant data for future modeling work. Thus, these non-gait approaches appear to be capable of providing information about the circuit mechanisms that account for the manifestations of FOG, which could be implemented by systems biologists to construct testable models.
Insights from reductionist observations
As highlighted above, many reductionist studies comparing patients with and without FOG across clinical features and a range of biomarkers (e.g., MRI, DBS, EEG, PET) have provided insights into the role of many different physiological processes and anatomical regions. For example, a lesion analysis performed in a series of 14 patients who developed FOG, demonstrated discrete disturbances in the cerebellar locomotor region (CLR), an area functionally connected to the dorsal medial cerebellum [
69]. Work from a recent meta-analysis of neuroimaging studies in PD has also suggested that CLR activation may play a compensatory role in locomotion [
70]. As outlined above, other studies have suggested a more generalized pathophysiology, which has allowed a common final pathway to be postulated [
71,
72]. However, the question must be raised as to whether all the relevant anatomical regions work together to produce a common input to this pathway or whether they speak separately to this single common pathway or multiple pathways.
Whilst not anatomically connected, disseminated regions of the brain are functionally connected to each other and can therefore exert influence. For example, it might be proposed that the common final pathway for FOG is associated with impaired corticothalamic and corticostriatal networks that lead to an increase in pallidal inhibitory outflow (globus pallidus internus—GPi), which is often accentuated by glutamatergic input from the STN in the presence of increased response conflict, leading to the emergence of 5–7 Hz oscillations between the two nuclei (STN-GPi). The STN activity also leads to impaired cerebellar output [
71]. Ultimately, the increased pallidal outflow manifests as impaired coordination of flexor–extensor pairs in the lower limbs, leading to gait arrest. Conceptually, if there was a significant burden of pathology directly affecting this common final pathway, there would be a more pervasive gait disturbance manifesting with more constant FOG or other gait disturbances. In addition, there are many nodes that feed into this locomotor network (e.g., cortical regions dealing with conflict resolution) that may have varying degrees of input depending on circumstances. At times, these input nodes will fail and trigger the common final pathway, whereas there may be strategies to compensate for this demand such as focusing attention (e.g., cueing). To accurately model these connections, the input of observations obtained from our current reductionist datasets, such as those from neuroimaging and neurophysiology, will be needed, and then systematic perturbations, both inhibitory and facilitatory, should be applied to determine whether the prediction matches the observation. For example, to explore the neural underpinnings of cueing, patients could undertake multi-modal experiments such as simultaneous EEG and fMRI, where a VR environment has sections with and without lines presented on the floor. Such approaches would be able to probe the neural networks (imaging) and dynamic power spectra (neurophysiology) in patients with and without FOG. Behavioural data could then be fed back into computer models manipulating these neural parameters to make predictions about how cueing may ameliorate FOG.
Going forward, studies that can collect multi-modal data from the same patients to inform the modeling approaches are required, along with constant validation approaches. Patients undergoing DBS represent a unique opportunity to record from within the brain and with the advent of sensing-stimulating devices used in many centers worldwide, it will become easier to repeat longitudinal assessments over time. However, it should be noted that most patients undergoing DBS are not usually severe freezers, and rare cases of FOG following DBS have been reported [
73]. In addition, MRI in such patients is challenging post-implantation and DBS signals can create significant artefact with concurrent EEG. Therefore, other prospective patient cohorts, as described above, might offer greater utility, especially if there was an opportunity to implement novel neurophysiological and neuroimaging approaches. For example, newer MRI methodologies to accurately image brainstem nuclei by detecting iron and neuromelanin content may be very useful and appear highly reproducible [
74]. In addition, PET studies exploring the role of dopaminergic and cholinergic systems, as well as amyloid burden, have been published [
75‐
78]. Other PET work has measured cerebral glucose metabolism with gait tasks performed during the uptake time of the radioligand
18F-fluorodeoxyglucose to explore the corticobasal-thalamocortical circuitry implicated in FOG [
79]. Meanwhile, work on noradrenergic and serotonergic neurotransmitters is lacking and represents a real gap for modeling FOG. Finally, whilst ambulatory EEG has been generating useful data for modeling FOG, there probably needs to be a greater emphasis on developing other dynamic imaging techniques. Some work in FOG has been conducted using functional near-infrared spectroscopy (fNIRS) to examine changes in oxygenated-haemoglobin levels that occur in the gait assessment of patients with FOG, but obviously the region of interest is often limited to the forehead, offering little insight into the rest of the brain’s activity [
80,
81]. Novel multi-optode fNIRS systems are now increasingly being used covering wider brain areas and guidelines to reduce artefacts are also available [
82]. Finally, it is not clear yet whether micro-dose ambulatory PET will prove useful, but clearly, being able to measure ligand activity during gait and balance tasks may prove incredibly helpful for understanding systems biology [
83].