Pedunculopontine Nucleus Microstructure Predicts Postural and Gait Symptoms in Parkinson's Disease

There is an urgent need to identify individuals at risk of postural instability and gait difficulties, and the resulting propensity for falls, in Parkinson's disease.

A BS TRACT: Background: There is an urgent need to identify individuals at risk of postural instability and gait difficulties, and the resulting propensity for falls, in Parkinson's disease. Objectives: Given known relationships between posture and gait and degeneration of the cholinergic pedunculopontine nucleus, we investigated whether metrics of pedunculopontine nucleus microstructural integrity hold independent utility for predicting future postural instability and gait difficulties and whether they could be combined with other candidate biomarkers to improve prognostication of these symptoms. Methods: We used stereotactic mapping of the pedunculopontine nucleus and diffusion tensor imaging to extract baseline pedunculopontine nucleus diffusivity metrics in 147 participants with Parkinson's disease and 65 controls enrolled in the Parkinson's Progression Markers Initiative. We also recorded known candidate markers of posture and gait changes: loss of caudate dopamine and CSF β-amyloid 1-42 levels at baseline; as well as longitudinal progression motor symptoms over 72-months. Results: Survival analyses revealed that reduced dopamine in the caudate and increased axial diffusivity in the pedunculopontine nucleus incurred independent risk of postural instability and gait difficulties. Binary logistic regression and receiver operating characteristics analysis in 117 participants with complete follow-up data at 60 months revealed that only pedunculopontine nucleus microstructure provided more accurate discriminative ability for predicting future postural instability and gait difficulties than clinical and demographic variables alone. Conclusion: Dopaminergic and cholinergic loss incur independent risk for future postural instability and gait difficulties, and pedunculopontine nucleus microstructure can be used to prognosticate these symptoms from early Parkinson's disease stages. © 2020 The Authors. Postural instability and gait disturbance (PIGD) and the resulting increase in falls is associated with a more pernicious disease course in Parkinson's disease (PD), with rapid development of dementia, 1 greater disability, 2 and early mortality. 3 PIGD and falls tend not to respond well to dopamine-replacement medications, but virtualreality training and nondopaminergic pharmacological interventions have shown early promise. 4 However, their clinical impact will depend on timely and tailored interventions. As such, there is an urgent need to understand the neural mechanisms that underpin these symptoms and to develop the means to identify individuals at risk.
Longitudinal observational studies have revealed that future postural and gait impairments in PD can be predicted at very early disease stages using dopamine transporter (DAT) imaging of the caudate and measurement of cerebrospinal fluid (CSF) β-amyloid 1-42. 5,6 Identification of these candidate biomarkers raises the potential that individuals at risk of PIGD can be identified even at very early disease stages, thereby enabling the development of tailored, preventive interventions to mitigate these symptoms. However, taking lessons from cognitive biomarker research, 7 predicting PIGD with the level of accuracy necessary for clinical trials will depend on our ability to provide a multimodal combinatorial biomarker that reflects the multiple underlying pathophysiological pathways of these symptoms.
There is now strong evidence that loss of brainstem cholinergic neurons, originating in the pedunculopontine nucleus (PPN), plays a significant role in the development of PIGD symptoms. Studies consistently report that cholinergic PPN lesions in primates cause significant PIGD. 8,9 In humans, in vivo cholinergic positron emission tomography (PET) imaging reveals that people with PD with a history of falls-which are more prevalent in people with PIGD 10 -have significantly decreased thalamic cholinergic innervation (a major component of which originates in the PPN). 11 In addition, the now well-recognized link between PIGD and cognitive decline in PD 12 is suggested to reflect a common underlying pathology of these symptoms originating in the cholinergic neurons of the nucleus basalis of Meynert (nbM). 13 Thus, the cholinergic contribution to PIGD progression is likely to be driven directly by motor pathways that include the PPN and may also be mediated by cognitive changes following nbM loss. However, to our knowledge, no studies have been conducted in humans to determine the specific contributions of these regions to PIGD in PD.
In vivo structural image analysis of the nbM is now possible via precision, subregional stereotactic mapping developed from postmortem histology and magnetic resonance. 14 These methods have revealed the spatiotemporal patterns of nbM atrophy in Alzheimer's disease 14 and were recently shown to predict the onset of cognitive decline in early PD. 15,16 Currently, however, no established imaging method to measure the structure of the PPN exists. Yet such a measure would provide a critical investigative tool to understand the causes of PIGD symptoms and to intervene in their development.
Recently, a stereotactic map of the PPN was developed to improve targeting of the nucleus for deep brain stimulation surgery for dopamine-unresponsive PIGD symptoms. 17 Although automated segmentation of gray and white matter in the brainstem using standard procedures is not accurate, making volumetry of the PPN technically difficult because of its brainstem location, it remains possible that metrics of microstructural integrity provided by diffusion tensor imaging (DTI) would be informative to further understand the neuropathology of PIGD symptoms. Indeed, DTI of the substantia nigra and nbM have revealed microstructural changes in these regions that reflect progression of motor and nonmotor PD symptoms. 16,18 In the current study, we combine precision stereotactic mapping of the PPN with DTI to test the hypothesis that microstructural integrity in this region can predict future PIGD symptoms in PD. Given previous suggestions that the nbM may be an important structure for PIGD, 13 we will also explore whether the nbM microstructure can predict PIGD. Because the PPN microstructure, caudate DAT, and CSF β-amyloid 1-42 concentrations reflect different underlying pathologies, we expect that the PPN microstructure will provide additive prediction of PIGD symptoms over that provided by caudate DAT and/or CSF β-amyloid 1-42 concentrations. As such, we additionally test the prediction that there will be utility in combining these candidate biomarkers to improve prognostication of PIGD in PD.

Study Design
Participants Participants selected from the PPMI for the current study were all those with coregistered T 1 -weighted and diffusionweighted magnetic resonance imaging (MRI) available at baseline, along with at least baseline Movement Disorder Society-UPDRS (MDS-UPDRS) scores. There were 147 participants with PD and 65 control participants.
Clinical, Dopamine Imaging, and CSF Assessments volume of interest template for extraction of regional count densities (http://www.indd.org/).

PPN
The PPN is the primary area of interest in this study, and diffusion metrics in this region were accessed via a mask created using the methods described in ref. 17. Briefly, postmortem MRI was performed on the brain of a 66-year-old woman without parkinsonism or cognitive decline. Following autopsy, the brain was fixed, dehydrated, serially sectioned, and Nissl stained. 23 Light and darkfield microscopy was used to enhance contrast and perform the segmentation of the nuclear boundaries of the PPN, creating a mask of the entire PPN region. Following digitization, the images were 3-dimensionally registered with the postmortem MRI, and the PPN mask transformed to MNI space via transforms generated following normalisation of the postmortem MRI to MNI space. 24 NbM Given the previous clinical and animal lesion studies showing a relationship between the nbM and PIGD deficits, we also extracted diffusion metrics from a basal forebrain region of interest (ROI), created as described in ref. 14. In brief, the map was derived from combined postmortem MRI and subsequent histological preparation of a brain specimen of a 56-year-old male who had died without any evidence of cognitive decline or psychiatric illnesses. Cerebral MRI scans were performed in situ. Cholinergic nuclei were identified and delineated on digital pictures of the stained brain slices, then manually transferred from the digital pictures into the corresponding magnetic resonance slices. Transformation of the delineations from the space of the dehydrated brain into the space of the in situ brain scan was performed using a high-dimensional nonlinear registration between the 2 brain scans 25 before final transformation from in situ space into MNI space using diffeomorphic anatomical registration through exponentiated lie algebra (DARTEL). The stereotactic cholinergic basal forebrain map distinguishes different cholinergic subdivisions. In the current article, we retained the ROI corresponding to the nbM.
Two additional ROIs were used as controls: a whole gray matter (GM) mask created from binarized images of GM segmentations produced as part of the DARTEL procedure, used to control for more widespread changes to brain diffusivity. In addition, to control for the potential contamination of the PPN ROI by nearby white matter, we drew a circular ROI in the corticospinal tract (CST), adjacent to the PPN mask, for use as a nuisance variable in statistical tests.

MRI Processing
a smaller selection of 20 images. Figure 2 illustrates this check, showing a representative image from a single participant with the PPN mask overlaid along with the atlas-based coordinates of the rostral pole of the PPN.
Finally, mean eigenvalues from the ROIs were extracted from each of the 3 eigenvalue images, and mean, axial (along the principal axis), and radial (perpendicular to the principal axis) diffusivity (mD, aD, rD, respectively) were calculated according to the following: mD = (λ 1 + λ 2 + λ 3 )/ 3; aD = λ 1 ; rD = (λ 2 + λ 3 )/ 2. Please note that we have not utilized the fractional anisotropy maps available in the PPMI for the current analysis. We avoided this because changes in fractional anisotropy are extremely difficult to interpret in GM (degeneration of Localization of the pedunculopontine nucleus using the current automatic method (voxels are labeled in white following procedure outlined in Fig. 1). The voxels identified as belonging to the pedunculopontine nucleus were then confirmed using the manual atlas-based method described in Zrinzo and colleagues 22  neurons oriented in different directions would cause a paradoxical increase in fractional anisotropy, whereas degeneration of neurons oriented in the same direction would have the opposite effect), and this may be particularly true in a region such as the PPN that is expected to be heavily myelinated.

Baseline Demographics, Clinical Scores, Diffusivity, DAT, and CSF Measures
Analyses of variance and post hoc Tukey tests were used to compare demographics, clinical measures, diffusivity metrics, mean caudate DAT, and CSF β-amyloid 1-42 metrics between the controls and PD participants who did/did not develop more severe PIGD. Subsequent univariate tests were conducted for diffusivity metrics that showed between-group differences, controlling for age, UPDRS part III scores and whole GM and CST diffusivity. Note that diffusivity metrics were matched in each analysis, for example, comparisons/correlations involving PPN aD controlled for whole brain aD and CST aD.

Survival Analysis
Cox proportional hazards regression analyses were used to test the hypothesis that development of more severe PIGD can be predicted by PPN microstructure measured at the time of PD diagnosis. None of the variables included violated collinearity assumptions. Time to severe PIGD was calculated as the number of months since study enrolment until the PIGD threshold (PIGD >5) was reached. One participant already meeting this threshold at baseline was excluded from this analysis. The prediction of severe tremor, rigidity, and bradykinesia were also evaluated in separate Cox regressions.
Control variables considered for inclusion in the analysis were age, Montreal Cognitive Assessment, UPDRS part III, baseline PIGD, and sex. This clinical and demographics only model was run as a first step, and all predictors surviving P < 0.2 were retained for the full model, with sex and Montreal Cognitive Assessment not meeting this criterion. Diffusivity metrics from each ROI were then added to the model separately (with whole GM and CST diffusivity added as additional control variables).
Caudate DAT and CSF β-amyloid 1-42 concentrations were subsequently added to the clinical, demographics, and diffusivity models to determine their shared and/or unique predictive relationships with diffusivity metrics for PIGD.

Binary Logistic Regression and ROC Analysis
In a smaller sample of participants with full data available at the 60-month follow-up visit (N = 117), binary logistic regression was used to model the predictive relationships between PIGD and candidate markers along with clinical and demographic variables. Using the pROC 26 package in R, ROC curves were constructed from the predicted values from a demographics and clinical characteristics-only model and from models including candidate markers together or in isolation. Areas under the curves (AUC) were compared with the DeLong method, 27 with the prediction that inclusion of biomarkers would provide more accurate discriminative ability for predicting PIGD than the clinical and demographic-only model.

Correction for Multiple Comparisons
All analyses are presented with uncorrected P values, and those surviving false discovery rate correction for number of diffusivity metrics and number of ROIs included in the analysis are noted in the text.

Results
Clinical, Demographic, and Diffusivity Characteristics at Baseline in Controls, People with PD Who Develop PIGD During Follow-Up Visits, and People with PD Who Do Not Develop PIGD at Follow-Up A total of 27 (18.4%) participants with PD developed severe PIGD during the 72-month follow-up period. One-way analysis of variance and post hoc Tukey tests revealed that the PD participants who developed PIGD during the follow-up period were older and had worse UPDRS part III scores at baseline than those who did not develop PIGD. This group was also found to have greater aD in the PPN when compared with both controls and participants with PD who did not develop PIGD (statistics reported in Table 1).
The latter finding survived false discovery rate correction for 2 (ROI: nbM and PPN) × 3 (diffusivity metrics: aD, mD, and rD) multiple comparisons, and univariate analyses revealed that PPN aD was significantly greater in the PD participants who did develop PIGD (compared with those who did not) even after controlling for age, sex, UPDRS part III scores, Montreal Cognitive Assessment scores, and CST aD and whole GM aD (F = 10.57, P = 0.001).
More severe tremor, rigidity, and bradykinesia developed in 28, 26, and 31 patients with PD, respectively. There was some overlap between these groups (8,14, and 15 patients with more severe tremor, rigidity, and bradykinesia also had more severe PIGD). However, diffusivity metrics were not different in those developing these symptoms versus those who did not (T < 1.62, P > 0.10).

Do Diffusivity Characteristics Incur Risk for PIGD Independently from Clinical and Demographic Markers for PIGD: Survival Analysis
Clinical and demographic variables (age, baseline PIGD, and total UPDRS part III scores) were entered into Cox regression models along with diffusivity metrics in the nbM and PPN. Whole GM and CST diffusivity were also entered as control regions. NB, the 3 diffusivity metrics (aD, mD, and rD) extracted from the 2 ROIs (nbM and PPN) were entered into separate models, and therefore results are corrected for 3 × 2 multiple comparisons.
PPN aD was the only metric to be associated with significant independent risk of developing PIGD (PPN aD: Wald = 9.32, P = 0.002 [survives false discovery rate correction for multiple comparisons]). For the nbM and all other diffusivity metrics, Wald <0.90 and P > 0. 35. To determine the specificity of PPN aD for predicting PIGD, we repeated the survival analysis to predict tremor, rigidity, and bradykinesia. None of these were found to be predicted by PPN aD (Wald <1.81, P > 0.18).

Do Diffusivity Characteristics Incur Risk for PIGD Independently from Clinical, Demographic, and Other Candidate Markers for PIGD: Survival Analysis
Given the previous results, the analyses that follow were not performed for diffusivity metrics in the nbM.

Does PPN Diffusivity and/or Other Candidate Markers of PIGD Improve Discriminative Ability for Predicting PIGD over Demographic and Clinical Variables: Binary Logistic Regression
In a sample of participants with follow-up data available at 60 months (N = 117), a binary logistic

Discussion
Our study shows that microstructural changes in the PPN are associated with an increased risk for the development of PIGD symptoms in PD and have discriminative ability for predicting PIGD within 5 years, even at very early disease stages. This finding is specific to the symptoms of PIGD (development of significant tremor, rigidity, and bradykinesia symptoms were unrelated to PPN diffusivity) and was specific to the PPN (microstructural changes in the nbM did not incur risk for PIGD symptoms). The latter finding implies that degeneration in the PPN and its relationship with future PIGD does not reflect a general PD-related decline in cholinergic subcortical nuclei and supports the suggestion that pathology of the PPN can cause PIGD in PD. This raises the potential that measures of PPN microstructure could be used to stratify patients to enable tailored and preventive interventions for gait and balance deficits and for future clinical trials.
The anatomical and functional specificity of this finding for the PPN and PIGD, respectively, is consistent with our current understanding of this region's role in normal walking and balance. In animal models, PPN lesions are sufficient to induce gait deficits, and stimulation of the PPN results in locomotion. 8,9 Depth recordings from the PPN in humans reveal modulated single-unit firing during imagined walking 28 and enhanced α-band oscillations during stepping in place. 29 Lau and colleagues 28 provide evidence that the PPN influences gait in ways that extend beyond "pattern generation," including motor planning and adaptation. The methods reported in the current article will enable further investigation of the role of the PPN in these skills.
The PPN is a neurochemically heterogeneous structure, containing gamma aminobutyric acid (GABA)ergic, glutamatergic, and cholinergic populations. 30 The cholinergic neurons of the PPN seem to be particularly relevant for PIGD and falls as demonstrated by the gait deficits produced by targeted cholinergic lesions of the PPN in primates and rodents. 8 Positron emission tomography with cholinergic ligands and pathological data has shown relationships between PPN-thalamic cholinergic innervation and PIGD and falls in PD. 11 Although the current findings make no distinction between neurochemical subpopulations of PPN neurons, taken with the previous findings, we can speculate that PPN aD reflects, at least in part, degeneration of PPN-thalamic cholinergic projections.
As expected, given previous literature, 5 we found that the risk of PIGD was increased by low caudate DAT at baseline. This relationship was independent from the relationship between PPN aD and future PIGD, implying there are distinct dopaminergic and cholinergic contributions to the risk of developing these symptoms. There is a known link between PIGDs-specifically freezing of gait-and frontal executive functions, 31 and the dependence of these on caudate dopamine is well documented. It is reasonable therefore to speculate that the link between caudate dopamine depletion and PIGD is mediated by impaired executive function. Similarly, a previous study using DTI with tractography revealed that freezing of gait was associated with reduced connectivity between the PPN and the frontal cortex, which was in turn associated with impaired executive function. 32 As such, combining the methods reported here to study these links could reveal the multiple pathologies that underpin the failure of the motor and/or nonmotor networks to support posture and gait in PD.
In follow-up analyses, only PPN aD significantly increased accuracy of a predictive model for PIGD beyond clinical and demographic variables alone. These results indicate that although there is a clear relationship between dopamine deficiency and PIGD, 5 it may not discriminate between people who do/do not develop these symptoms. Conversely, cholinergic loss develops more heterogeneously in PD, and the current findings indicate that this variability provides greater predictive utility for the symptoms that are underpinned by this loss. The corticopetal cholinergic system originating in the nbM has been linked to gait speed and falls, 11,33 but the risk of PIGD was not predicted by nbM diffusivity in the current report. It has been suggested that gait changes and increased falls risk follow cognitive deterioration, particularly in executive function and attention, 34 and that degeneration in the nbM is the principal underlying factor. 13 We are currently investigating whether datasets with more fine-grained cognitive, gait, and postural measures (http://bam-ncl.co.uk/ our-work/studies/icicle-gait/) than provided by the PPMI are informative for the role of the nbM in these abilities.
Finally, we did not observe a relationship between CSF β-amyloid 1-42 concentrations at baseline and risk of PIGD. Yet, previous data has consistently reported a relationship between PIGD severity and CSF β-amyloid levels, 5,6 as well as neocortical β-amyloid deposition measured with PET. 35 It is possible that the relationship between CSF β-amyloid 1-42 concentrations and PIGD symptoms is more readily detected using methods that characterize PIGD severity using more fine-grained, continuous measures rather than when PIGD is expressed as a dichotomous variable as required by survival and binary regression analyses. As such, the lack of a predictive relationship reported here between CSF β-amyloid 1-42 concentrations and UPDRS PIGD >5 should be viewed as a limitation of these methods rather than evidence that a relationship does not exist.
Some caveats to the current study must be acknowledged. First, volumetry of the PPN is difficult because of the requirement to accurately segment white matter and GM in this region. As such, we have not corrected for partial volume effects. However, our aim was to provide a measure of structural change in the region that may have utility for understanding and predicting PIGD in PD. Therefore, whether our prognostic marker reflects micro-and/or macro-structural changes is not a focus of the current article. That said, the application of more accurate segmentation procedures for brainstem structures are warranted. Similarly, the seeming specificity of the marker for PPN aD, rather than mD or rD, is currently open to interpretation. The current analysis is not able to determine the specific neurodegenerative process that leads to increased aD. However, we are continuing to investigate the relationship between the PPN and PIGD using alternative DTI metrics 36 that will enable us to address these additional questions.
In addition, we defined PIGD >5 as the end point for our predictive analyses as it allowed us to distinguish individuals who are beginning to experience functional limitation attributable to at least 1 PIGD marker. However, it is not clear from the current study whether this is the most sensitive end point to use. It is also not clear whether there is a linear relationship between PPN degeneration and the emergence of PIGD symptoms. Future work could aim to explore these relationships to determine the point at which PPN degeneration becomes risky for developing PIGD as well as whether longitudinal progression of PPN degeneration is more informative than cross-sectional baseline data for predicting PIGD. Finally, it must also be noted that PIGDs are overrepresented in atypical parkinsonian disorders, 37 and it is therefore possible that patients developing PIGD in the current sample may later be diagnosed with atypical parkinsonian disorders.
In conclusion, stereotactic mapping alongside DTI of the PPN can be used to prognosticate PIGD in PD. The relationship between baseline PPN microstructure and future PIGD was independent from the relationship between caudate dopamine deficiency and future PIGD, implying that there are distinct dopaminergic and cholinergic contributions to the risk of developing these symptoms. It is now necessary to further investigate how measures of PPN and dopaminergic integrity could be combined in algorithms to predict the onset of particular components, or complications, of PIGD, such as freezing of gait or falls.