Associating Resting-State Functional Connectivity and Improvements in Reactive Balance in Parkinson’s Disease

Abstract

Reactive stepping is impaired in people with Parkinson’s disease (PwPD) and can be trained through practice. However, these improvements are often variable across individuals. Identifying neurological predictors and potential mechanisms of this variability can improve the efficiency of rehabilitation. This study investigated the association between improvements in backward reactive stepping performance through training and resting-state functional magnetic resonance imaging (rs-fMRI) in PwPD. 15 PwPD underwent rs-fMRI and an eighteen-week multiple-baseline study, which included baseline assessments (B1 and B2, 2-weeks apart), a 2-week training protocol, and post-training assessments immediately after protocol (P1) and 2-months later (P2), in which we assessed anterior-posterior margin of stability (MOS_AP) at first foot contact. Linear regression analyses assessed the relationship between functional connectivity, using a region of interest approach, and immediate (P1-B2) and retained (P2-B2) improvements in MOS_AP during reactive stepping. Results showed that higher right thalamus-right amygdala connectivity was related to immediate MOS_AP improvements (p_FDR<0.05). Right thalamus-right amygdala, left caudate-left hippocampus, and left thalamus-left hippocampus connectivity were associated with better long-term retention of MOS_AP (p_FDR’s<0.05). These findings suggest thalamo-limbic coupling may contribute to immediate and retained improvements in reactive balance in PwPD and could aid in identifying individuals who would benefit most from balance rehabilitation.

Introduction

PD is a progressive neurodegenerative disease characterized by tremor, rigidity, bradykinesia, and postural instability [1]. Falling is a major concern for people with PD (PwPD). Over 50% of PwPD fall annually, resulting in injury, reduced quality of life, and death [2].

One critical aspect of balance is reactive stepping, defined as the quick step taken to regain balance after a slip or trip [3]. Reactive stepping is impaired in PwPD [4], as PwPD exhibit smaller reactive steps with reduced margin of stability (MOS) than non-PD peers [5], [6], [7], [8]. PwPD can improve reactive steps through training, and these improvements relate to fewer falls [7], however there is considerable variability in the responsiveness of reactive stepping in training across individuals [7], [9].

Reactive balance relies on cortical and subcortical brain networks including the frontal and parietal lobes, thalamus, basal ganglia, limbic cortices, brainstem, and cerebellum [10]. However, literature investigating neural control of reactive stepping in PwPD remains relatively sparce, and has used varying imaging modalities such as electroencephalography (EEG) [11], functional near-infrared spectroscopy (fNIRS) [12], diffusion tensor imaging (DTI) [13], and functional magnetic resonance imaging (fMRI) [14]. Although results have varied, they generally support broad spinal and supraspinal involvement for stepping in PwPD. In particular, Monaghan and colleagues found that structural brain connectivity (measured via DTI) related to improvements in reactive step training [13]. Studies integrating neuroimaging markers and rehabilitative interventions can identify which individuals are most likely to benefit from reactive stepping training and can facilitate progress towards individualized fall-prevention neurorehabilitation.

Resting-state functional connectivity (rsFC) is a non-invasive imaging approach that has been widely used to understand co-activation patterns, and hence presumed connectivity, across brain regions [15]. The aim of this study was to identify brain rsFC patterns that predict immediate and retained learning effects of a 2-week reactive stepping intervention during backward reactive stepping in PwPD.

Methods

Data presented here are part of a larger clinical trial and elements of these data have been previously published [7], [8], [9], [13]. The current analysis associating rsFC and changes in backward reactive stepping is novel.

Participants

15 participants completed all trial assessment visits and had complete rsFC data. Participants were recruited through physician referral, local support groups, and from fliers placed in the community. Participants were considered “at risk for fall”, consistent with previous criteria [7]. All participants were tested in the “ON” medication state, had no deep brain stimulation, and met the full inclusion and exclusion criteria outlined previously [7]. Procedures were approved by the IRB at Arizona State University and written informed consent was obtained from all participants in accordance with the Declaration of Helsinki. The study was registered at ClincialTrials.gov (NCT03895814).

Clinical Assessment and Behavioral Protocol

Participants completed the following clinical assessments: part 3 of the Movement Disorders Society – Unified Parkinson’s Disease Rating Scale (MDS-UPDRS-III), Montreal Cognitive Assessment, Mini-BESTest, New Freezing of Gait Questionnaire, and Activities-Specific Balance Confidence (ABC) Scale. Demographic information is summarized in Table 1.

Table 1.

Participant Demographics

Demographics	PD n=16
Age (years ± SD)	71.69 ± 6.150
Sex (Male/Female)	10 / 6
MDS-UPDRS-III (mean ± SD)	33.69 ± 9.370
Hoehn and Yahr Scale
Stage 1	1
Stage 2	9
Stage 3	6
Disease Duration (years ± SD)	4.375 ± 3.830
LEDD (mg ± SD)	552.3 ± 235.4
MoCA (mean ± SD)	26.56 ± 2.800
Mini-BESTest (mean ± SD)	21.43 ± 2.930
NFOG (mean ± SD)	4.312 ± 5.470
ABC (mean ± SD)	71.70 ± 10.16
DGI (mean ± SD)	18.81 ± 2.510

Abbreviations: ABC = Activities of Balance Confidence Scale; DGI = Dynamic Gait Index; LEDD = Levodopa Equivalent Daily Dose; MDS-UPDRS-III = Movement Disorder’s Society-Unified Parkinson’s Disease Rating Scale, Part III; MoCA = Montreal Cognitive Assessment; NFOG = New Freeze of Gait Questionnaire

Reactive stepping was conducted at four assessment timepoints. Participants first completed two baseline assessments (B1 and B2), separated by two-weeks, followed by a two-week training period. Then, post-intervention assessments were performed immediately after training (P1) and after two-months (P2) to evaluate immediate learning (P1-B2) and long-term retention (P2-B2). Reactive stepping training was performed using support-surface translations generated by a Bertec split-belt instrumented treadmill (Bertec Corporation, Columbus, Ohio). Participants completed 3 reactive steps per direction (forward, backward, leftward, rightward) at each timepoint. Training consisted of six training sessions over two-weeks, each consisting of 8 perturbations in each direction.

Behavior Data Analysis and Outcomes

As described previously [7], a 14-camera 3D motion capture system (100 Hz; Motion Analysis Corporation, Santa Rosa, California) recorded trunk and lower limb kinematics, while force plate data was collected from the treadmill.

The current study’s reactive stepping outcome was anterior-posterior MOS (MOS_AP) during backward losses of balance measured. Backward stepping MOS_AP was chosen as it was the variable with the most pronounced improvement over time through training, and backward stepping is particularly impacted in PwPD [16]. MOS_AP was analyzed using custom software in The MotionMonitor xGen, and was averaged across trials within each visit [7].

rs-fMRI Preprocessing

fMRI scans underwent automated preprocessing using the afni_proc.py tool within the AFNI toolbox. Scans were skull-stripped prior to using the afni_proc.py tool. The following preprocessing steps were performed: MRI data were de-spiked, slice-time corrected, co-registered to the T1-weighted anatomical scan, and warped to standard space. Non-linear warping of the functional MRI data was performed to the MNI_avg152 template. Head motion (above 1mm scan-to-scan) and motion outliers (over 0.05mm) were regressed out of analysis utilizing the ANATICOR function in AFNI. A bandpass filter was applied with a cut off frequency between 0.01 and 0.1. Data were spatially smoothed with a 0.4mm² full width half maximum (FWHM) Gaussian kernel.

rs-fMRI Acquisition

MRI data were collected on a 3T Philips Achieva MRI scanner equipped with an eight-channel receiver head coil (Philips Medical Systems, Andover, Massachusetts). The protocol included a 4-minute resting state functional and a T1-weighted structural scans. T1-weighted structural image data was acquired using magnetized-prepared rapid gradient echo sequence (MPRAGE) (repetition time [TR]=13ms, echo time [TE]=4.5ms, field of view=256 x 256mm, slice thickness=1mm, number of slices=180). Resting-state data was acquired using echoplanar imaging sequence (TR=3000ms, TE=25ms, field of view=64x64mm, slice thickness=4mm, number of slices=50)

rs-fMRI Analysis

Resting-state functional connectivity analysis was preformed using pre-determined regions of interest (ROIs) derived from the FreeSurfer atlas, including the left and right (L/R) cerebellum-cortex, L/R thalamus, L/R caudate, L/R putamen, L/R hippocampus, L/R amygdala, and L/R basal forebrain [17]. These 14 ROIs were selected based on previous work assessing brain regions potentially related to reactive stepping in PwPD [18]. For each participant, correlation and Fisher z-transformed matrices were obtained for each participant using 3dNetCorr function in AFNI [19]. ROI to ROI connections were extracted from the upper triangle of each matrix (Fisher Z) for subsequent analysis. There were 91 total ROI-to-ROI connections.

Statistical Analysis

All statistical tests were performed in R Version 4.5.1 (2025-06-13) and R Studio Version 2025.09.0+387 (released 2025-10-13). Linear mixed-effects models were used to assess the effect of visit (time) on MOS_AP, with visit modeled as affixed effect and participants as a random intercept. Age and MDS-UPDRS-III score were included as covariates. Pairwise comparisons were performed between immediate learning (P1-B2) and long-term retention (P2-B2).

Separate linear regression models were performed for MOS_AP with each ROI connection (i.e., right thalamus-right amygdala), covarying for age and MDS-UPDRS-III score. Standardized beta (β) coefficients were calculated, and Benjamini-Hochberg false-discovery rate (FDR) correction was applied to correct for multiple comparisons across all ROI connections. Analyses were conducted separately for immediate learning (P1-B2) and long-term retention (P2-B2).

Sensitivity analyses were conducted by including ABC and Mini-BESTest scores as additional covariates to the models to assess the robustness of the primary findings. Statistical significance for all models, primary and sensitivity analyzes, were defined as p<0.05 (two-tailed) and p_FDR<0.05.

Results

Backwards Reactive Stepping

Consistent with previous analyses of this dataset [7], the subset of participants included in the current cohort showed a statistically significant increase in MOS_AP between P1 and B2 (immediate learning) [(t(15.0) = 3.085, p-value = 0.008] and between P2 and B2 (long-term retention) [(t(14.3) = 3.615, p-value = 0.003].

Association Between Resting-State ROI Connections and AP Margin of Stability

Immediate learning (β=0.619, 95% CI [0.031, 0.257], p_FDR=0.031) and long-term retention (β=0.599, 95% CI [0.016, 0.193], p_FDR=0.033) of MOS_AP was associated with higher right thalamus–right amygdala connectivity (Figure 1; Table 2). Additionally, long-term retention of MOS_AP was associated with higher left caudate–left hippocampus (β=0.748, 95% CI [0.047, 0.196], p_FDR=0.017) and left thalamus–left hippocampus connectivity (β=0.644, 95% CI [0.026, 0.162], p_FDR=0.044) (Table 2). All other ROI connections were non-significant.

Figure 1: Association Between AP Margin of Stability (MOS_AP) and Learning.

A) ROI-to-ROI connection between the right thalamus and right amygdala. B) Scatterplots and linear regression results representing MOS_AP immediate learning (P1-B2) and long-term retention (P2-B2) for the right thalamus-right amygdala ROI connection.

Table 2. Significant Associations Between Resting-State ROI Connections and MOS_AP Learning.

Separate linear regression models were used to assess the association between the MOS_AP and the ROI connection. Age and MDS-UPDRS-III score were included as covariates. Unstandardized beta and standard error (SE), standardized beta, t-value, p-value, FDR-corrected p-value, and 95% confidence intervals (CI) are shown for each significant model. pFDR values are reported for ROI predictor only; covariates were not included in multiple-comparison correction.

Model	Unstandardized		STD Beta	t-value	p-value	p-value (FDR-corrected)	95% CI
	Beta	SE
Immediate Learning (Post Intervention 1 – Baseline 2)
Right Thalamus – Right Amygdala	0.144	0.052	0.619	2.787	0.016	0.031*	[0.031, 0.257]
Age	−0.007	0.003	−0.592	−2.593	0.024		[−0.012, −0.001]
MDS-UPDRS, Part III	0.000	0.002	−0.057	−0.264	0.796		[−0.004, 0.003]
Long-Term Retention (Post Intervention 2 – Baseline 2)
Right Thalamus – Right Amygdala	0.105	0.040	0.599	2.593	0.025	0.033*	[0.016, 0.193]
Age	−0.006	0.002	−0.675	−2.774	0.018		[−0.011, −0.001]
MDS-UPDRS, Part III	0.000	0.001	0.040	0.175	0.864		[−0.003, 0.003]
Left Caudate – Left Hippocampus	0.122	0.034	0.748	3.595	0.004	0.017*	[0.047, 0.196]
Age	−0.004	0.002	−0.496	−2.537	0.028		[−0.008, −0.001]
MDS-UPDRS, Part III	0.002	0.001	0.391	1.780	0.103		[−0.001, 0.005]
Left Thalamus – Left Hippocampus	0.094	0.031	0.644	3.056	0.011	0.044*	[0.026, 0.162]
Age	−0.003	0.002	−0.312	−1.443	0.177		[−0.007, 0.001]
MDS-UPDRS, Part III	0.001	0.001	0.148	0.688	0.506		[−0.002, 0.004]

Abbreviations: FDR = false-discovery rate; MDS-UPDRS = Movement Disorders Society – Unified Parkinson’s Disease Rating Scale; MOS_AP = anterior-posterior margin of stability; ROI = region of interest; STD = standardized

Significance indicted by * p< 0.05 (FDR-corrected).

Sensitivity Analyses

When ABC score was added as a covariate, results were largely consistent with the primary analysis. In contrast, when Mini-BESTest was added as a covariate, significant models became non-significant (data available upon request).

Discussion

This study evaluated the association between rsFC and changes in backward reactive stepping following a two-week training intervention in PwPD. The primary finding was that greater rsFC between right thalamus-right amygdala was associated with larger immediate learning and long-term retention of MOS_AP during backward reactive stepping. This is the first study to employ rsFC to predict changes in motor learning during reactive balance in PwPD.

All participants were “at risk for falls”. Fear of falling in PwPD is a common concern, which can elicit acute emotional responses such as fear and anxiety. Previous work utilizing rs-FC has shown an association between fear of falling and greater amygdala-hippocampal FC in multiple sclerosis [20] and fear of falling was negatively associated FC with the amygdala connectivity in PwPD and freezing of gait [21]. Additionally, prior work suggests that perceived or real penalties (i.e. falls) may reinforce learning during reactive balance, promoting rapid adaptions [22]. In this study, the current association between thalamic-amygdala FC and immediate learning and long-term retention may reflect a heightened fear-anxiety response that drives accelerated motor learning and adaptation to unpredictable perturbations.

The hippocampus, caudate, and thalamus have been linked to the limbic network which regulates cognition, arousal, mood and gait in PwPD. Previous work has shown that long-term memory may rely on caudate-hippocampus connectivity [23]. For example, increased activation of the parahippocampal gyrus was observed during an imaginary slip compared with imagined walking, likely reflecting a heightened limbic and emotional response after a slip [24]. In the current study, the hippocampus-caudate and hippocampus-thalamus FC connections were associated with long-term retention, not immediate learning, suggesting that hippocampal activity may be involved in retrieving memories and promotes retention of learning. Additionally, sensitivity analysis showed that addition of ABC as a covariate did not substantially influence the observed learning-brain structure relationships. This suggests that balance confidence (measured via ABC) is unlikely to influence the observed associations. However, the reduction in significance with addition of balance performance (MiniBESTest), objective balance performance may play a mediating role in this relationship.

This study extends earlier work on a similar PD cohort which also assessed the relationship between brain connectivity and reactive step improvements through training [13]. This previous work observed white-matter microstructural changes in the corona radiata, thalamic radiations, superior longitudinal fasciculi, and corpus callosum using DTI [13]. The variance between the previous study and current findings may be due to several reasons. First, the current study used rs-fMRI and conducted an ROI-to-ROI analysis of the cerebellar-subcortical network. This network was chosen as it has been shown to be the most predictive of reactive stepping performance in PwPD [14]. Further, rs-fMRI assess how brain regions communicate with each other (FC), while DTI imaging looks at structural integrity of the brain utilizing diffusion of water which can only be seen in white matter structures, thereby providing complementary insights into the neuromechanisms that underlying reactive stepping changes through training in PwPD.

Several limitations should be noted. First, the modest sample size (n=15) limits statistical power, reduces generalizability of results, and precluded our ability to perform mediation analyses. Second, while testing occurred “ON” levodopa to reflect typical daily function and quality of life, testing in the “OFF” medication state is needed to determine whether findings are independent of medication effects. Finally, participants were at risk for falls, potentially limiting generalization of findings.

The present study demonstrates that specific patterns of thalamo-limbic functional connectivity are associated with both immediate and retained improvements in MOS_AP during backward reactive step training in PwPD. These findings deepen our understanding of the neural mechanisms that may underlie balance rehabilitation in PwPD. Further studies should examine resting-state functional connectivity in a larger cohort to determine the robustness and clinical relevance of current findings.

Glossary

B1

baseline 1

B2

baseline 2

DTI

diffusion tensor imaging

EPI

echoplanar imaging

FDR

false discovery rate

fMRI

functional magnetic resonance imaging

fNIRS

functional near-infrared spectroscopy

MDS-UPDRS-III

Movement Disorders Society-Unified Parkinson’s Disease Rating Scale, Part III

MOS

margin of stability

MOS_AP

anterior-posterior margin of stability

MPRAGE

magnetized-prepared rapid gradient echo

P1

post intervention 1

P2

post-intervention 2

P1-B2

immediate learning

P2-B2

long-term retention

PD

Parkinson’s disease

PwPD

people with Parkinson’s Disease

ROI

region of interest

rsFC

resting-state functional connectivity

rs-fMRI

resting-state functional magnetic resonance imaging

TE

echo time

TR

repetition time