Dual Dopaminergic and Limbic-Cognitive Contributions to Gait Parameters in De Novo Parkinson Disease

Abstract

Background and Objectives

Parkinson disease (PD) is a progressive neurodegenerative disease in which gait impairment is a common and disabling clinical feature. We aimed to characterize the independent and mediated contributions of striatal dopamine transporter (DAT) availability and gray matter (GM) volume to quantitative gait impairment in de novo PD.

Methods

In this prospective study, we consecutively recruited patients with de novo PD fulfilling the UK Brain Bank clinical criteria and healthy controls (HCs) without focal neurologic symptoms and parkinsonism at Wonju Severance Christian Hospital. All participants underwent GAITRite-based gait analysis, and patients further underwent brain MRI for GM volumetry, ¹⁸F-FP-CIT PET for striatal DAT availability, and motor and cognitive assessments. After exploratory gait-related correlation analyses, causal mediation analyses tested whether the effects of neuroimaging biomarkers on gait parameters were mediated by cognition or motor symptom severity.

Results

A total of 122 patients with de novo PD (mean age, 69.66 years; 45.90% female) and 177 HCs (mean age, 72.07 years; 51.41% female) were enrolled. Compared with HCs, patients with PD showed slower velocity, shorter step/stride lengths, reduced swing and single-support time, increased double-support time, and greater dispersion across multiple gait domains. Limbic-related GM volumes were associated with step/stride lengths and temporal phase composition, and these associations were substantially mediated by global cognition (representative ACMEs [95% CIs] for stride length: posterior cingulate cortex, 1.9559 [0.5606–4.2238]; hippocampus, 3.6722 [1.0707–8.2118]; thalamus, 2.8794 [0.0408–7.2535]; amygdala, 4.9696 [1.8241–10.3405]; proportions mediated, 19%–50%). Lower putaminal DAT availability was associated with greater stance time and double-support time variability, whereas lower caudate DAT availability was associated with longer swing and single-support time and altered phase parameters. These associations were not significantly mediated by bradykinesia/rigidity scores; direct effects remained significant (ADEs [95% CIs]: caudate-swing time, −0.0237 [–0.0428 to −0.0086]; putamen-double-support time variability, −2.0249 [−3.1445 to −0.8835]).

Discussion

These findings support a dual-pathway model of gait regulation in patients with de novo PD, where striatal dopaminergic denervation directly affects gait rhythmicity, whereas limbic system atrophy affects gait through direct and cognition-mediated pathways. This medication-naïve cohort may limit generalizability to the broader PD spectrum.

Introduction

Parkinson disease (PD) is the most common movement disorder, characterized by the cardinal symptoms of bradykinesia, rigidity, and rest tremor, underpinned by dopaminergic denervation in the striatum.^1,2 As the disease progresses, patients develop marked postural instability and gait disturbances, affecting daily functioning and quality of life.³ Particularly for gait, classical parkinsonian gait presents with a stooped posture and slow, shuffling steps with shortened stride length, reduced arm swing, and a narrow base width.^4,5 Heterogeneity in gait has also been reported among patients.^6,7 These comprehensive gait characteristics differentially contribute to increased fall risk and social isolation in patients with PD.⁸

Recent technology has enabled clinicians to easily assess quantitative gait measures using pressure sensor-based or three-dimensional camera-based motion capture devices.^9,10 Using these techniques, previous studies have consistently demonstrated that patients with PD, compared with healthy controls (HCs), exhibit reduced walking speed, shortened stride length, prolonged double-support time, and altered gait variability.^11,12 However, given that gait abnormalities are not merely manifestations of motor symptoms,¹³ there remains limited information on how quantitative gait measures relate to other parkinsonian symptoms, cognitive function, and neuroimaging biomarkers in an integrated manner.

It is important to note that most quantitative gait studies in patients with PD have included participants who had already initiated treatment with antiparkinsonian medications, which can substantially alter specific gait parameters. Studies in such individuals may confound the interpretation of the relationships between gait dysfunction and underlying neural pathophysiology. Although several studies have examined gait abnormalities in patients with de novo PD compared with HCs,^14,15 a comprehensive understanding of how motor, cognitive, and structural brain alterations jointly contribute to specific gait parameters has yet to be established, particularly in the context of explaining the significant gait heterogeneity in PD.

The aim of this study was to characterize comprehensive gait characteristics by systematically examining the relationships among quantitative gait parameters, clinical symptoms, cognitive function, and neuroimaging biomarkers in a cohort of patients with de novo PD. Specifically, we compared quantitative gait measures in patients with de novo PD with those of HCs and assessed the heterogeneity of gait patterns within the de novo PD group. We further explored the associations of these gait parameters with motor symptom severity, cognitive performance, dopamine transporter (DAT) availability, and gray matter (GM) volume and performed causal mediation analyses to determine whether the statistical effects of neuroimaging biomarkers on gait parameters were accounted for by cognition or motor symptom severity. By focusing on a medication-naïve population, this study sought to delineate the putative pathways through which neuroimaging, cognitive, and motor factors contribute to quantitative gait characteristics in early-stage PD.

Methods

Participants

In this prospective study, patients with de novo PD were consecutively recruited from the Movement Disorder Clinic at Wonju Severance Christian Hospital between May 2022 and December 2024 if they fulfilled the UK Brain Bank clinical criteria for PD¹⁶ and had no previous exposure to levodopa, dopamine agonists, monoamine oxidase-B inhibitors, amantadine, or any other antiparkinsonian medications at the initial visit. All recruited patients underwent neurologic examination, quantitative gait analysis, and clinical assessment at the initial visit, and brain MRI and ¹⁸F-FP-CIT PET were performed at 1 or 2 subsequent visits within 6 months of the initial visit. Exclusion criteria were as follows: no evidence of (1) dopaminergic deficit on ¹⁸F-FP-CIT PET and (2) structural brain lesions, including acute ischemic or hemorrhagic stroke, brain tumors, or normal pressure hydrocephalus.

HCs were recruited on a voluntary basis from the Health Check-up Center and Neurology Clinic during the same period. Prespecified eligibility criteria for HCs were as follows: (1) the absence of focal neurologic symptoms, (2) no evidence of parkinsonism based on neurologic examination, and (3) no symptoms that could affect gait, including dizziness, tremor, bradykinesia, gait disturbance, postural instability, or acute pain in the lumbar spine, pelvis, or lower extremities during walking. Exclusion criteria included a history of cerebral infarction; cerebral hemorrhage; dementia; idiopathic REM sleep behavior disorder (RBD); psychosis; peripheral neuropathy; or recent knee, hip joint, or spine surgeries.^17-22 RBD was evaluated using a semistructured questionnaire based on the RBD Severity Scale,²³ completed by the participants and, when available, their caregivers.

All clinical diagnoses of the participants were established through mutual agreement between 2 neurologists (M.S.B. and J.Y.H.).

Quantified Gait Analysis

All participants completed more than 4 trials of walking (average total walking distance, 21.64 ± 9.48 m) at their usual speed on the GAITRite Electronic Walkway (CIR Systems Inc., Peekskill, NY, USA), a 6-m-long and 0.6-m-wide walkway equipped with pressure-activated sensors for quantifying temporal and spatial gait parameters. The GAITRite system provides 20 gait parameters categorized into 5 subclasses²⁴: spatiotemporal (e.g., velocity in cm/second and cadence in steps/min), spatial (e.g., step length, stride length, and base width in cm), temporal (e.g., step time, swing time, stance time, and single-support and double-support time in seconds), phase (e.g., percentage of gait cycle for swing time, stance time, and single-support and double-support time), and variability (e.g., coefficient of variation [CV, %], for step length, stride length, base width, swing time, stance time, and double-support time). The spatiotemporal, spatial, temporal, and phase gait parameters for both the left and right sides were averaged, and these average values were subsequently used to calculate the gait variability parameters.

Clinical Assessments

All participants completed the Korean version of the Montreal Cognitive Assessment (MoCA) to evaluate their cognitive performance. The MoCA provides subscores for visuospatial abilities, naming, attention, language, memory, abstraction, and orientation, and a total score is calculated as the sum of these subscores.

Participants with de novo PD were also evaluated using the Movement Disorder Society–Unified Parkinson’s Disease Rating Scale (UPDRS) to assess motor symptom severity. From part III (UPDRS-III), we obtained subscores for tremor (items 3.15–3.18), rigidity (item 3.3), bradykinesia (items 3.4–3.9 and 3.14), and gait/postural instability (items 3.10–3.12).²⁵

Image Acquisition

Whole-brain three-dimensional T1-weighted MRI data were obtained using magnetization-prepared rapid acquisition gradient-echo sequences with the following parameters: slice thickness, 0.9 mm; pixel size, 0.4492 × 0.4492 mm; repetition time, 2000 ms; echo time, 2.43 ms; flip angle, 9°; and matrix size, 512 × 512 pixels. Images were reconstructed to 512 × 512 matrix over a 256-mm field of view.

¹⁸F-FP-CIT PET data were acquired using a dedicated PET/CT scanner (Discovery MI with LightBurst Digital 4-Ring Detector, GE Medical Systems, Milwaukee, WI, USA), which obtains images with a three-dimensional resolution of 2.3 mm at full width at half maximum at the center of the field of view. The acquisition began 180 minutes after an IV bolus injection of ¹⁸F-FP-CIT (5 mCi, 185 MBq) and was performed for 20 minutes in static mode. A head holder minimized head movement during the scan, and brain CT images were obtained for attenuation correction. The PET data were reconstructed in a 256 × 256 × 256 matrix with a 1.17 × 1.17 × 2.78-mm voxel size using the ordered-subsets expectation maximization algorithm with 6 iterations and 16 subsets.

Image Preprocessing

The magnetic resonance images were processed using the standard cross-sectional preprocessing pipeline in FreeSurfer v.7.4.0.²⁶ The intracranial volume (ICV) was also estimated for each participant. Cortical GM was parcellated according to the Desikan-Killiany atlas, and the regions were grouped to form 14 bilateral composite regions of interest (ROIs).¹³ Six subcortical GM structures were segmented in each hemisphere. In total, 20 bilateral ROIs were defined in the cerebrum. For each ROI, bilateral volume was calculated as the sum of the left and right regional volumes. The cerebellum was segmented using CerebNet²⁷ and parcellated according to the Diedrichsen atlas,²⁸ which includes 13 bilateral ROIs and 8 vermal ROIs. The brainstem was segmented and parcellated into the midbrain, pons, medulla oblongata, and superior cerebellar peduncle, using the FreeSurfer pipeline.²⁹ All regional volumes were normalized to ICV to yield normalized GM volumes (nGMVs). The preprocessing results were visually inspected and manually corrected according to FreeSurfer guidelines.

DAT availability was quantified using ¹⁸F-FP-CIT PET. The PET data were co-registered with the corresponding T1-weighted MRI scans. Striatal regions were segmented into anterior and posterior parts based on the midpoint of the anterior commissure-posterior commissure (AC–PC) line, and the putamen was further subdivided into anterior dorsal, anterior ventral, posterior dorsal, and posterior ventral parts based on the standard axial plane passing through the AC-PC line, after realigning the MRI scans to the Talairach coordinate system. Mean intensities were calculated within the bilateral anterior and posterior caudate nuclei and each putaminal subregion. The specific binding ratio (SBR) for each region was computed as (mean intensity of the region—mean intensity of the reference)/(mean intensity of the reference), using the bilateral occipital cortex as the reference.

Statistical Analysis

All statistical analyses were performed using MATLAB R2023b and R v.4.3.2. The significance level was set at p < 0.05. Complete-case analyses were performed to handle missing values in the MoCA and UPDRS.

Demographic variables were compared between patients with de novo PD and HC participants using the Student or Welch t tests for continuous scales, preceded by the Levene test, and chi-square tests for categorical variables.

Gait parameters were compared between the 2 groups using permutation-based t and F tests with 10,000 resamples after adjusting for age, sex, height, and disease duration. For location tests, the Freedman-Lane procedure was used, and statistical significance was determined using a two-sided test. For dispersion tests, residual permutation on full models was performed, and significance was assessed using a one-sided test of the null hypothesis of equal variances between groups against the alternative hypothesis that the variance in the de novo PD group was greater than that in the HC group. These dispersion tests were performed to investigate whether participants with de novo PD exhibited greater heterogeneity in gait parameters, thereby reducing the risk of spurious correlations in the subsequent analyses.

Exploratory correlation analyses between gait parameters and neuroimaging or clinical variables were performed using Pearson partial correlation coefficients, with age, sex, height, and disease duration as the covariates. Multiple comparisons for each gait parameter were corrected using the Benjamini-Hochberg false discovery rate procedure.

Causal mediation analyses³⁰ were conducted to investigate whether nGMV influenced gait parameters through cognitive performance and whether striatal SBR influenced gait parameters through the severity of the motor symptoms. In these analyses, the total MoCA score represented cognitive performance, and the sum of the bradykinesia and rigidity subscores of the UPDRS-III represented motor symptom severity. Pairs of gait parameters and neuroimaging variables were selected if they showed significant bivariate correlations in the exploratory analysis and significant heterogeneity in the dispersion tests. The statistical significance of the average causal mediation effect (ACME), average direct effect (ADE), and total effect was estimated using a regression-based approach,³¹ with 95% bias-corrected and accelerated bootstrap confidence intervals (CIs) derived from 10,000 resamples. Effects were considered significant if the CI did not include zero.

Sensitivity Analysis

Partial Spearman correlation analyses between gait parameters and neuroimaging or clinical variables were conducted to examine whether the exploratory Pearson-based association results were robust to monotonic nonlinear relationships or the influence of outliers.

We also performed a sensitivity analysis to assess the influence of potential violations of sequential ignorability in the causal mediation models.³¹ Specifically, we varied the correlation between the residuals of the mediator and outcome regression models as a sensitivity parameter and computed the critical points at which the residual correlation, product of residual R² values, and product of total R² values caused ACME or ADE to become zero, respectively.

We further conducted causal mediation analyses with alternative mediator specifications, testing cognition as a mediator of striatal DAT-gait associations and motor symptom severity as a mediator of regional GM volume-gait associations.

Standard Protocol Approvals, Registrations, and Participant Consents

This study was approved by the Institutional Review Board of the Wonju Severance Christian Hospital (Ref# CR322029), and the research protocol was aligned with the principles of the Declaration of Helsinki and its subsequent revisions. Written informed consent was obtained from all the participants.

Data Availability

Anonymized data supporting the findings of this study are available on request from the corresponding author. The data are not publicly available because of privacy or ethical restrictions.

Results

Demographic and Clinical Characteristics

A total of 122 patients with de novo PD and 177 HCs were enrolled in this study. Table 1 summarizes the demographic and clinical characteristics of the study participants. Patients with de novo PD were slightly younger than the HCs. The mean UPDRS-III score in the PD group was 22.11, with an average disease duration of 1.75 years. No significant differences were observed between the groups regarding sex distribution, education level, height, or weight. In the PD group, the intervals for gait assessment-MRI, gait assessment-PET, and MRI-PET were 35.16 ± 23.49, 36.45 ± 26.73, and 8.12 ± 25.61 days, respectively (mean ± SD). Nine patients with de novo PD had missing values in clinical scores.

Table 1.

Demographic and Clinical Characteristics of Participants

	Healthy controls	Participants with de novo Parkinson disease	p Value
n	177	122
Age (y)	72.07 ± 8.06	69.66 ± 9.33	0.0177
Sex (female)	91	56	0.3489
Education (y)	10.11 ± 4.25	9.16 ± 4.39	0.0619
Disease duration (y)	N/A	1.75 ± 1.70	N/A
Height (cm)	159.99 ± 8.12	160.68 ± 8.68	0.4808
Weight (kg)	62.11 ± 9.54	63.31 ± 11.53	0.3440
Leg length (cm)	86.89 ± 4.43	86.09 ± 9.25	0.3762
UPDRS-III score	N/A	22.11 ± 11.13	N/A
Tremor	N/A	3.92 ± 3.41	N/A
Rigidity	N/A	4.46 ± 3.37	N/A
Bradykinesia	N/A	8.84 ± 5.39	N/A
Gait/postural instability	N/A	2.51 ± 2.35	N/A
MoCA	N/A	20.30 ± 6.16	N/A

Abbreviation: MoCA = Montreal Cognitive Assessment; N/A = not applicable; UPDRS = Unified Parkinson’s Disease Rating Scale.

Continuous data are presented as mean ± SD; categorical data are presented as frequency.

Gait Characteristics in De Novo PD

Table 2 summarizes the gait parameters of the HC and de novo PD groups. After adjusting for age, sex, height, and disease duration, gait velocity was significantly slower and cadence was significantly higher in the PD group. Patients with PD exhibited significantly shorter step and stride lengths than HCs. Regarding temporal parameters, patients with PD showed reduced swing and single-support time, with increased double-support time. The phase parameters revealed lower percentages of the swing and single-support phases, accompanied by higher percentages of the stance and double-support phases in the PD group. Regarding gait variability, patients with PD exhibited greater CVs of stride length, swing time, and double-support time, whereas the CVs of step length and base width was reduced compared with that in controls.

Table 2.

Gait Parameters for Participants

	Healthy controls	Participants with de novo Parkinson disease	t statistics	p Value
Spatiotemporal parameters
Velocity (cm/second)	92.13 ± 17.90	83.25 ± 20.64	−3.6479	0.0003
Cadence (steps/min)	105.31 ± 9.95	107.64 ± 9.18	2.2083	0.0292
Spatial parameters
Step length (cm)	52.44 ± 8.47	46.44 ± 10.54	−5.5061	<0.0001
Stride length (cm)	105.47 ± 16.97	93.53 ± 21.23	−5.4442	<0.0001
Base width (cm)	9.64 ± 2.88	10.05 ± 2.73	0.2196	0.8272
Temporal parameters
Step time (s)	0.57 ± 0.06	0.56 ± 0.05	−2.1830	0.0319
Swing time (s)	0.40 ± 0.04	0.37 ± 0.05	−5.4203	<0.0001
Stance time (s)	0.74 ± 0.09	0.74 ± 0.08	0.1122	0.9113
Single-support time (s)	0.40 ± 0.04	0.37 ± 0.05	−5.4203	<0.0001
Double-support time (s)	0.33 ± 0.08	0.36 ± 0.09	3.1101	0.0021
Gait phases
Swing phase percentage	35.28 ± 2.32	33.59 ± 3.58	−4.5895	<0.0001
Stance phase percentage	64.73 ± 2.32	66.41 ± 3.59	4.5907	<0.0001
Single-support percentage	35.28 ± 2.32	33.59 ± 3.58	−4.6060	<0.0001
Double-support percentage	28.72 ± 4.81	32.32 ± 7.43	4.9610	<0.0001
Variability
CV step length (%)	15.98 ± 3.00	9.26 ± 4.60	−12.2895	<0.0001
CV stride length (%)	5.13 ± 1.77	7.33 ± 4.19	5.5697	<0.0001
CV base width (%)	23.73 ± 10.63	16.90 ± 6.02	−4.5387	<0.0001
CV swing time (%)	5.87 ± 2.37	7.05 ± 3.45	3.0581	0.0024
CV stance time (%)	6.14 ± 1.79	6.77 ± 3.10	0.1533	0.8767
CV double-support time (%)	8.74 ± 2.57	10.33 ± 3.80	3.4577	0.0006

Abbreviation: CV = coefficient of variation.

Data are presented as mean ± SD; p values were calculated using permutation-based t tests adjusted for age, sex, height, and disease duration.

The dispersion of the gait parameters in each group, adjusted for the same covariates, is presented in eTable 1. Among the spatial parameters, step and stride lengths showed significantly greater dispersion in the PD group than in the HC group. Among the temporal parameters, swing time, single-support time, and double-support time also exhibited significantly higher variances in the PD group. All phase parameters showed significantly greater variances than those in the HC group. Similarly, the CVs of step length, stride length, swing time, stance time, and double-support time exhibited greater dispersion in the PD group than in the HC group.

Association Between Motor Symptom Severity and Gait Patterns in De Novo PD

UPDRS-III scores were strongly associated with most gait parameters, including velocity; step and stride lengths; step, stance, and double-support times; all phase parameters; and CV of step length, stride length, and stance time in the de novo PD group, after adjusting for age, sex, height, and disease duration (Figure 1A, left). Among the UPDRS-III subscores, the bradykinesia and gait/postural instability subscores were correlated with the gait parameters (Figure 1A, right).

Figure 1. Associations Between Gait Parameters and Clinical Variables in the De Novo Parkinson Disease Group After Adjusting for Age, Sex, Height, and Disease Duration.

(A) Pearson cross-correlation matrix between gait parameters and motor symptom severity, assessed using the Unified Parkinson’s Disease Rating Scale Part III and its domain subscores. (B) Pearson cross-correlation matrix between gait parameters and cognitive function, assessed using the Montreal Cognitive Assessment and its domain subscores. Circle sizes represent categorized uncorrected p values, and yellow asterisks indicate statistical significance after controlling for multiple comparisons using the false discovery rate procedure. CV = coefficient of variation; MoCA = Montreal Cognitive Assessment; UPDRS = Unified Parkinson’s Disease Rating Scale.

Association Between Cognition and Gait Patterns in De Novo PD

The total MoCA score was significantly correlated with multiple gait parameters in the de novo PD group after adjusting for age, sex, height, and disease duration (Figure 1B, left). Among the cognitive domains (Figure 1B, right), the visuospatial domain score was associated with the CVs of step length, stride length, and swing time. The attention domain score was associated with the CVs of the step and stride lengths. Overall cognitive performance was positively correlated with gait velocity and step and stride lengths.

Association Between Neuroimaging Biomarkers and Gait Patterns in De Novo PD

After adjusting for age, sex, height, and disease duration, notable associations were observed between neuroimaging biomarkers and gait characteristics in the de novo PD group.

Regarding DAT availability (Figure 2A), greater variability in stance time and double-support time was significantly associated with lower SBR values across the putaminal subregions. Greater swing and single-support times were significantly associated with reduced SBR in the caudate subregions, and gait phase parameters were specifically associated with SBR in the posterior caudate nucleus.

Figure 2. Associations Between Gait Parameters and Neuroimaging Biomarkers in the De Novo Parkinson Disease Group After Adjusting for Age, Sex, Height, and Disease Duration.

(A) Pearson cross-correlation matrix between gait parameters and striatal specific binding ratio derived from ¹⁸F-FP-CIT PET. (B) Pearson cross-correlation matrix between gait parameters and gray matter volume normalized to intracranial volume derived from MRI. Circle sizes represent categorized uncorrected p values, and yellow asterisks indicate statistical significance after controlling for multiple comparisons using the false discovery rate procedure. CV = coefficient of variation.

Regarding cerebral GM volumes (Figure 2B), higher gait velocity and longer step and stride lengths were significantly correlated with greater volumes in the limbic system regions, including the cingulate cortex, thalamus, hippocampus, and amygdala. Stance and double-support times, as well as gait phase parameters, were associated with volumes in the frontal, parietal, temporal, precuneus, cingulate, and hippocampal regions. Swing time variability was also significantly associated with the volumes in similar regions. No significant associations were observed between gait parameters and cerebellar GM volumes or brainstem volumes (eFigure 1).

Causal Mediation Analysis

Causal mediation analysis delineated the complex relationships between neuroimaging biomarkers and gait parameters, accounting for the influence of cognition and motor symptom severity in the de novo PD group. Among pairs showing significant bivariate correlations in the exploratory analysis, mediation models with significant heterogeneity are presented in Figure 3 for gait-striatal SBR pairs and in Figure 4 for gait-nGMV pairs with significant ACMEs. The regression coefficients and mediation effects of all causal mediation models are presented in eTables 2–5.

Figure 3. Causal Mediation Models of Striatal Specific Binding Ratio on Gait Parameters Through Motor Symptom Severity.

The left panels present schematic diagrams of the mediation models with caudate SBR as the independent variable and (A) swing time and (B) single-support and double-support time percentages as outcome variables and (C) with putaminal SBR as the independent variable and the coefficients of variation of stance time and double-support time as outcome variables. The right panels display the corresponding mediation effects with 95% bias-corrected and accelerated bootstrap CIs. All analyses included age, sex, height, and disease duration as covariates. Red indicates significant effects, defined as confidence intervals that did not include zero. CV = coefficient of variation; SBR = specific binding ratio; UPDRS = Unified Parkinson’s Disease Rating Scale.

Figure 4. Causal Mediation Models Showing Significant Average Causal Mediation Effects of Gray Matter Volume Normalized to Intracranial Volume on Gait Parameters Through Cognitive Performance.

The left panels present schematic diagrams for causal mediation models incorporating (A) the coefficient of variation of step length, (B) step and stride lengths, and (C) single-support and double-support time percentages as outcome variables. The right panels display the corresponding mediation effects with 95% bias-corrected and accelerated bootstrap confidence intervals. All analyses included age, sex, height, and disease duration as covariates. Red indicates significant effects, defined as confidence intervals that did not include zero. CV = coefficient of variation; MoCA = Montreal Cognitive Assessment; nGMV = normalized gray matter volume.

Regarding striatal DAT availability, the SBR of the caudate nucleus significantly influenced the swing time (Figure 3A) and single-support and double-support time percentages (Figure 3B) after controlling for the sum of the bradykinesia and rigidity subscores of the UPDRS-III. Similarly, putaminal SBR significantly and directly influenced the variability in stance and double-support times (Figure 3C). None of the causal mediation models incorporating striatal DAT exhibited any significant ACMEs.

Regarding cerebral GM volume, the volume of the inferior parietal lobule significantly influenced step length variability through the total MoCA score in the de novo PD group (Figure 4A). The nGMVs of the limbic structures, including the posterior cingulate cortex, thalamus, hippocampus, and amygdala, exerted substantial mediating effects on the step and stride lengths (Figure 4B). Among these regions, the posterior cingulate cortex and amygdala exhibited full mediation through cognition of these gait parameters. In addition, hippocampal volume showed full mediation, with a significant ACME on single-support and double-support time percentages through cognition (Figure 4C). The proportion mediated in models showing significant ACME ranged from approximately 19% to 50%. Most causal mediation models demonstrated significant direct and total effects (eTable 4).

Sensitivity Analysis

Most exploratory associations were robust to potential outliers and monotonic nonlinearity because Spearman correlations largely reproduced the Pearson-based findings (eFigures 2 and 3), except for associations between gait phase parameters or step/stride length variability and MoCA total or subdomain scores, which were therefore interpreted with caution. By contrast, associations between UPDRS rigidity and phase parameters, and between velocity or step/stride length and striatal SBR, became significant only in the Spearman analysis, suggesting monotonic but nonlinear relationships. Sensitivity analyses for the ACMEs (eFigures 4–7) indicated mild-to-moderate robustness to unmeasured mediator-outcome confounding, with critical residual correlations of approximately 0.25–0.4 and corresponding R²-product values of approximately 0.04–0.10. The ADEs were highly robust, with most critical residual correlations >0.6 and R²-product values > 0.4 (eTables 6 and 7). No significant ACMEs were observed with alternative mediators, either for cognition in striatal SBR-gait associations or for motor symptom severity in GM volume-gait associations (eTables 8 and 9).

Discussion

This study elucidated the comprehensive gait characteristics of de novo PD with respect to 2 primary aspects: gait abnormalities and gait heterogeneity within patients with de novo PD relative to HCs. We further identified that the heterogeneity in gait patterns was related to both regional GM volume and striatal SBR. GM volume in the posterior cingulate cortex, hippocampus, amygdala, and thalamus was associated with shorter step and stride lengths and altered phase composition. These associations were statistically accounted for by global cognitive performance in the mediation analysis, with a significant mediation effect through global cognition, while striatal SBR was associated with altered swing time (equivalent to single-support time) and altered phase composition. By contrast, the associations between striatal SBR and gait parameters were not statistically mediated by parkinsonian symptoms.

Our quantitative gait findings indicate that patients with de novo PD already exhibit the classical parkinsonian gait pattern compared with HCs, with slower walking speed, increased cadence, shortened stride length, prolonged double-support time, and altered gait variability, in agreement with previous studies on de novo^14,15 and established PD.^11,12 The presence of these abnormalities in the early, medication-naïve stage suggests that gait changes in PD emerge early in the disease course and may arise in the context of cognition, regional atrophy of the cerebral cortices, and dopamine availability, which is independent of parkinsonian motor symptoms.

Most gait parameters that differed between patients with de novo PD and HCs also showed notable interindividual differences, indicating that gait abnormalities are not uniformly expressed in patients with PD. Similarly, patients with poor cognitive performance, particularly in the visuospatial and attentional domains, showed comparable patterns of gait abnormalities. This is consistent with previous findings that gait is not merely an automated motor activity but also depends on visuospatial^32,33 and attentional resources,³⁴ supporting the concept of cognitive-motor coupling.³⁵ By contrast, velocity, cadence, step time, and the CV of base width did not show marked heterogeneity beyond the group difference from HCs and may represent gait features that become commonly and relatively uniformly abnormal once PD is present.

Associations between gait parameters and neuroimaging biomarkers suggest that circuits involving the nigrostriatal dopaminergic and limbic systems make distinct contributions to gait regulation in de novo PD,^36,37 with these systems differentially supporting separate domains of gait control. The nigrostriatal dopaminergic pathway seems to directly support gait automaticity through the basal ganglia.³⁸ In de novo PD, a reduced putaminal SBR was associated with the rhythmic component of gait automaticity, consistent with previous findings that dopaminergic denervation in this region correlates with impaired fine-tuning of rhythmic motor control in PD.^39,40 In addition, a lower caudate SBR was associated with the temporal-structuring component of gait automaticity. This may reflect the restricted recruitment of associative neostriatal loops in organizing locomotor programs, potentially triggering freezing of gait,^37,41 an extreme manifestation of disturbed temporal gait timing. These relationships were independent of the severity of bradykinesia and rigidity because mediation effects through these symptoms were not significant, whereas the direct effects remained significant, underscoring the importance of dopaminergic denervation in gait regulation, in addition to its effects on limb motor symptoms.

By contrast, limbic system regions, including the posterior cingulate cortex, thalamus, hippocampus, and amygdala, seem to modulate spatial and phase aspects of gait through the mediation of cognition, reflecting the recruitment of higher-order cognitive resources when automaticity is compromised.³⁵ These regions are anatomically connected with both prefrontal and basal ganglia circuits and are known to integrate internal cues, emotional salience, and visuospatial information to maintain step amplitude scaling and phase coordination during locomotion.⁴² In our mediation analysis, global cognitive performance statistically accounted for a substantial portion of the associations between these regions and gait parameters, suggesting that reductions in cognitive efficiency can amplify the behavioral consequences of limbic region atrophy on gait, consistent with previous evidence that cognitive-motor coupling emerges early in patients with PD. In particular, the posterior cingulate cortex, which is involved in anticipatory allocation of spatial attention and modulation of attentional networks,⁴³ showed both direct associations with gait and associations that were statistically accounted for by cognition, aligning with emerging concepts of cognitive-motor coupling in PD.^35,44

We also found a notable cognition-mediated association between the GM volume in the inferior parietal lobule and spatial gait variability. This integrates visual, vestibular, and somatosensory information⁴⁵ and has been implicated in higher-order processes relevant to motor-cognition coupling in Alzheimer disease¹³ or in older adults.⁴⁶ Taken together, these findings further support a nondopaminergic contribution to gait regulation in de novo PD, although its interactions with other neural circuits require further investigation.

Several methodological strengths enhance the validity of our findings. The focus on medication-naïve patients eliminates the confounding effects of dopaminergic therapy, providing insights into the natural history of gait dysfunction in de novo PD without pharmacologic influence. The comprehensive multimodal approach, combining quantitative gait analysis, advanced neuroimaging, and detailed clinical assessments, enables the examination of complex brain-behavior relationships using sophisticated analytical methods. Causal mediation analysis, the method we used to investigate the causal relationship between these measures, allowed us to decompose the associations among brain measures, cognitive performance, and gait parameters into formally defined indirect and direct effects within the potential outcomes framework, rather than relying on simple algebraic combinations of path coefficients. By making the identification assumptions explicit and incorporating formal sensitivity analyses for unmeasured mediator-outcome confounding, this approach strengthened the causal interpretation and robustness of our mediation results in this observational cohort study.

This study has some limitations. First, because our cohort was restricted to de novo PD and recruited at a single center, our findings may not be fully generalizable across the broader PD spectrum or to other de novo PD populations. The relatively early untreated stage may have reduced clinical severity, which may partly explain the lack or relative weakness of some mediation effects. Moreover, our de novo PD cohort showed relatively low mean MoCA scores; although low scores have been reported in other de novo PD cohorts^47,48 and may broaden cognitive variability, the cognition-mediated findings warrant interpretation with caution. Second, the causal mediation results should be interpreted in light of the cross-sectional design, which precludes causal inference. In addition, because significant findings were limited to selected gait components, the proposed dual-pathway model may be most appropriately interpreted within the scope of these specific gait domains. Third, the MoCA and its derived domain subscores analyzed in this study may not fully capture cognitive domains that are particularly relevant to gait regulation in PD, given that it is primarily a global cognitive screening instrument. Finally, the dual-pathway model should be interpreted within the scope of the current imaging framework, which assessed striatal DAT availability and regional GM volumes. Other neural pathways not addressed in this study may also contribute to gait regulation in PD, including the cholinergic pathway involving the pedunculopontine nucleus⁴⁹ and cerebellar motor circuits.⁵⁰ Future research directions should include longitudinal studies strengthening causal interpretation of gait-brain relationships as PD progresses, including associations between changes in cerebellar or brainstem volumes and gait. Future studies should also extend this framework across the full PD spectrum while accounting for medication status, thereby providing a more generalizable understanding of gait-brain relationships in PD, including alternative mediation pathways. More comprehensive neuropsychological assessment for domain-specific cognition, complementary imaging modalities (for example, diffusion/fMRI or quantitative susceptibility mapping),^50,51 and broader gait characterization using three-dimensional motion capture or video-based gait analysis systems^9,10 may further clarify cognition-mediated and nondopaminergic contributions to gait regulation.

In conclusion, this study delineates comprehensive gait characteristics in de novo PD, showing that gait patterns that are abnormal compared with HCs also vary across patients according to striatal dopaminergic depletion, regional GM volume, cognitive performance, and motor symptom severity. The predominantly direct effects of putaminal dopamine depletion on fundamental gait parameters, in contrast to the mixed direct and cognition-mediated effects of limbic system atrophy on gait variability, support the evidence for dual control of gait and provide a neural basis for the strong modulation of gait by attentional demand, dual-task interference, and cognitive decline in de novo PD. This framework extends the understanding of gait regulation beyond traditional motor circuits to include the cognitive and limbic systems, potentially providing a foundation for developing targeted therapeutic approaches and supporting the utility of quantitative gait analysis as a digital biomarker in PD research and clinical practice in the future.

Glossary

ACME

average causal mediation effect

AC-PC

anterior commissure-posterior commissure

ADE

average direct effect

CV

coefficient of variation

DAT

dopamine transporter

GM

gray matter

HC

healthy control

ICV

intracranial volume

MoCA

Montreal Cognitive Assessment

nGMV

normalized GM volume

PD

Parkinson disease

RBD

REM sleep behavior disorder

ROI

region of interest

SBR

specific binding ratio

UPDRS

Unified Parkinson's Disease Rating Scale

Author Contributions

S-W. Kim: drafting/revision of the manuscript for content, including medical writing for content; study concept or design; analysis or interpretation of data. M.J. Lee: drafting/revision of the manuscript for content, including medical writing for content. J.W. Kim: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. J. Yoo: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. H-K. Kim: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. J.Y. Hong: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data. M.Y. Liberman: drafting/revision of the manuscript for content, including medical writing for content. M.S. Baek: drafting/revision of the manuscript for content, including medical writing for content; major role in the acquisition of data; study concept or design; analysis or interpretation of data.

Study Funding

This study was supported by a National Research Foundation (NRF) grant provided by the Korea government (NRF2022R1C1C1012535), and a grant of the Korea Health Technology R&D Project through the Korea Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic of Korea (grant number : RS-2025-25459509) to M.S. Baek, and a NRF grant provided by the Korea government (RS-2024-00358576) to S.W. Kim.

Disclosure

The authors report no relevant disclosures. Go to Neurology.org/N for full disclosures.