Resting-state functional connectivity associated with gait characteristics in people with Parkinson’s disease

Abstract

Introduction:

Parkinson’s disease (PD) is a movement disorder caused by dysfunction in the basal ganglia (BG). Clinically relevant gait deficits, such as decreased velocity and increased variability, may be caused by underlying neural dysfunction. Reductions in resting-state functional connectivity (rs-FC) between networks have been identified in PD compared to controls; however, the association between gait characteristics and rs-FC of brain networks in people with PD has not yet been explored. The present study aimed to investigate these associations.

Methods:

Gait characteristics and rs-FC MRI data were collected for participants with PD (N=50). Brain networks were identified from a set of seeds representing cortical, subcortical, and cerebellar regions. Gait outcomes were correlated with the strength of rs-FC within and between networks of interest. A stepwise regression analysis was also conducted to determine whether the rs-FC strength of brain networks, along with clinical motor scores, were predictive of gait characteristics.

Results:

Gait velocity was associated with rs-FC within the visual network and between motor and cognitive networks, most notably BG-thalamus internetwork rs-FC. The stepwise regression analysis showed strength of BG-thalamus internetwork rs-FC and clinical motor scores were predictive of gait velocity.

Conclusion:

The results of the present study demonstrate gait characteristics are associated with functional organization of the brain at the network level, providing insight into the neural mechanisms of clinically relevant gait characteristics. This knowledge could be used to optimize the design of gait rehabilitation interventions for people with neurological conditions.

1. Introduction:

Parkinson’s disease (PD) is a movement disorder characterized by dysfunction in the basal ganglia (BG) [1]. Many of the motor deficits in PD, including tremor, bradykinesia, postural instability, and rigidity, are attributed to this underlying neural dysfunction. Gait deficits, including decreased velocity and increased spatial and temporal variability, are among the most debilitating symptoms in PD and are clinically relevant as they are indicative of increased risk of falls and worsening disease severity [2,3]. Common dopamine replacement medications do not adequately target all gait symptoms [4,5]. It is therefore of clinical importance to better understand the neural mechanisms of gait deficits in PD in order to design more effective therapeutic approaches to address gait disorders. However, the neuropathophysiology of gait is difficult to study given the movement restrictions of most imaging approaches.

Resting-state functional connectivity (rs-FC) MRI measures the temporal coherence of low-frequency blood-oxygen-level-dependent (BOLD) signals throughout the whole brain while participants lay still and alert [6]. Signals from different brain regions with greater temporal correlations with one another indicate areas of the brain that are functionally connected and can be associated with behavioral measures collected outside of the scanner. This is an ideal method for studying neural correlates of gait, as walking cannot be performed during an MRI scan.

Prior studies report associations between the strength of rs-FC and clinically relevant gait characteristics in older adults. For example, gait velocity has been associated with the strength of rs-FC within sensorimotor, frontoparietal, and visual networks [7–9]. Decreased gait performance while dual task walking, which can be an indicator of gait impairment and has been shown to affect gait velocity and variability [10], has been associated with decreased rs-FC in sensorimotor and prefrontal regions related to attention in healthy older adults [8]. rs-FC has also been studied in people with PD, showing differences in rs-FC within and between sensorimotor, thalamic, and cerebellar networks compared to controls [11]. However, network-level rs-FC in people with PD has not yet been investigated in relation to clinically relevant gait characteristics, such as velocity and spatiotemporal variability.

Previous rs-FC studies of gait in PD focused primarily on cortical networks, without adequately examining subcortical or cerebellar networks. In order to take a comprehensive, network-level approach to studying gait in PD, it is important to include regions and networks that are vital in motor performance, such as the BG, thalamus, and cerebellum [12]. Therefore, this study aimed to investigate associations between the strength of rs-FC within and between cortical, subcortical, and cerebellar brain networks with spatiotemporal gait characteristics in people with PD. We hypothesized stronger rs-FC within and between motor-related networks, including cortical, subcortical, and cerebellar networks, would be associated with better forward walking gait characteristics, and stronger rs-FC within and between attention-related networks would be associated with better dual-task gait characteristics. Finally, we also aimed to compare the relationship between network-level rs-FC and established clinical measures with clinically relevant gait characteristics.

2. Material and Methods

2.1. Participants

Data from the baseline evaluation of an exercise intervention study for individuals with PD (ClinicalTrials.gov ID: NCT01768832) [13] were used for the present study. All participants had a diagnosis of idiopathic PD and were tested in the OFF state of their medication (withdrawn for at least 12 hours). Participant inclusion criteria included: at least 50 years of age, at least a high school education, and no dementia (evidenced by an MMSE score ≥ 27). Motor performance was assessed by trained research staff using the Movement Disorders Society Unified Parkinson Disease Rating Scale Part 3 (MDS-UPDRS-III) [14]. This study and its ethical considerations were approved by the university’s Institutional Review Board and all participants gave written informed consent.

2.2. Gait Measures

Spatiotemporal gait parameters were measured using a five-meter GAITRite walkway (CIR Systems, NJ, USA). Primary gait outcome measures included velocity (m/sec) and the coefficient of variation (CV) (%) of step time (steps/min). Velocity was normalized to each participant’s average leg length (LL), in order to control for the effects of stature on gait velocity. Coefficient of variation was calculated as (standard deviation/mean) × 100.

For the gait assessments two conditions were collected: comfortable, normal paced forward walking and dual-task walking. For the dual-task condition, participants performed a verbal fluency task in which they listed words starting with a given letter of the alphabet while walking across the GAITRite. Five trials were collected and averaged for each condition. This provided on average 23.8±10.9 (mean±std) steps per condition, which can be used to reliably calculate gait variability measures [15]. To quantify the effect of the dual-task on gait, the dual-task change was calculated for velocity and step time CV as (dual-task – forward). A negative dual-task change for velocity and a positive dual-task change for step time CV indicates worse performance during the dual-task condition.

2.3. MRI data acquisition

MRI data were collected using a Siemens Trio 3.0T scanner and a standard 12-channel head coil. First, two structural scans were collected: a T1-weighted (T1W) sagittal, magnetization prepared rapid acquisition with gradient echo (MP-RAGE, TR=2400 ms, TI=1000 ms, TE=3.16 ms, FA=15°, 0.9 mm³ voxels, 8:09 min) and a T2-weighted (T2W) fast spin echo (TR=3200 ms, TE=455 ms, 1.0 mm³ voxels, 4:43 min). Two resting-state scans were acquired from BOLD sensitized fMRI (TR=2200 ms, TE=61 ms, 4.0 mm³ voxels, two 7:26 -minute runs of 200 frames each) while participants remained alert but relaxed with their eyes closed. Participants with sustained tremor observed during scans were removed prior to further analysis.

2.4. Preprocessing

For anatomical preprocessing, Freesurfer 5.0 [16] was used to segment the T1-weighted images and create grey matter, white matter, and cerebrospinal fluid masks for each participant. Standard BOLD preprocessing steps were employed, as previously described [17,18]. First, 14 frames were removed from each scan to account for magnetization equilibrium. Functional data were aligned using the following steps: 1) slice time correction to temporally align each slice to the start of each volume, 2) rigid body transformation to correct for head motion within and across runs, and 3) whole-brain mode 1000 normalization to normalize the data within each run. Anatomical data were resampled to 3×3×3 mm voxels and functional data were aligned to the T1-weighted structural image and then to a Talairach atlas using affine registration in a single step. To control for motion-related confounds, we used a frame censoring approach [18]. First, motion parameters were low-pass filtered at 0.1 Hz to identify high-frequency motion that may represent respiratory artifacts [17]. Next, frame-wise displacement was calculated on the filtered motion parameters [18] and frames with an FD greater than 0.1 mm were removed [17]. Segments with fewer than 3 contiguous frames were also removed (mean±sd frames retained across scans = 276.8±75.2). Participants without at least five minutes of BOLD data in total were removed from further analysis [19].

2.5. Functional Connectivity Processing

Standard functional connectivity processing followed the methods of Power et al. [18]. Specifically, the data were demeaned and detrended to take away any temporal trends or drifts caused by the scanner during data acquisition. This zeroed the mean of the data, removing potential measurement bias. Nuisance regression included variables for 1) the global signal, 2) white matter and CSF signal from FreeSurfer generated masks, and 3) six motion parameters, and their derivatives to remove their effects from the BOLD signal. A bandpass filter (0.009–0.08 Hz) was applied because low frequencies are most prevalent in the resting-state. Spatial smoothing, with a 6 mm full width at half maximum (FWHM) Gaussian blur, was used to average the intensities of neighboring signals to remove additional noise in the signal.

2.6. Functional Connectivity Analyses

A set of 238 8–10 mm spherical seeds covering the cortex [20], subcortex, and cerebellum [12] were used to define functional networks. rs-FC values were calculated as Fisher z-transformed temporal correlations between pairs of BOLD signals from each seed. Using previously defined network assignments for each seed [12], a correlation matrix was created for each participant. Within network correlation composite scores were calculated as the average correlation between seeds within the network, excluding the on diagonal correlations. Between network (i.e., internetwork) correlation composite scores were calculated as the average correlation scores between seeds from two different networks (e.g. basal ganglia network seeds and thalamus network seeds).

2.7. Networks of Interest

Networks including cortical and subcortical seeds were chosen based on previously defined associations [11,12]. Networks were identified as motor (somatomotor dorsal (SMd), basal ganglia (BG), cerebellum, thalamus), cognitive, including default mode network (DMN), attention (ventral attention (VAN) and dorsal attention (DAN)), and executive control (cinguloopercular (CO) and frontoparietal (FP)), or sensory (visual) networks (Figure 1A). Correlation matrices of the networks were visually inspected to confirm blocked structure of correlations similar to what has been observed previously in PD [11] (Figure 1B).

Figure 1. Networks of Interest.

A) A set of cortical and subcortical regions were used representing 10 distinct networks. Seeds are shown in the axial (x), coronal (y), and sagittal (z) views and color coded by their functional networks. B) Large-scale networks in people with PD, represented in an averaged correlation matrix of all participants. Strong network organization is shown, with high correlations within networks, represented in the diagonal, and lower correlations between networks, represented in the off-diagonals. Abbreviations: Cinguloopercular, CO; Dorsal attention network, DAN; Default mode network, DMN; Frontoparietal, FP; Ventral attention network, VAN.

2.8. Statistical Analysis

All statistical analyses were conducted in the R statistical computing environment [21]. Paired samples t-tests were performed to determine differences between forward and dual-task gait measures. Pearson’s correlations were run to determine associations between the strength of rs-FC scores and behavioral measures. A stepwise regression was performed using the MASS package in R [22], for dual-task velocity, including the significant rs-FC scores and MDS-UPDRS-III scores as predictors, covarying for age and sex. Data were winsorized when appropriate to account for outliers [23]. The networks of interest for the Pearson’s correlations were selected a priori based on hypotheses related to the underlying functions associated with the basal ganglia and other networks. A bonferoni correction was used to account for multiple comparisons of the 10 networks of interest, so statistical significance was set to α = .005.

3. Results

3.1. Participant Characteristics

Initially, 85 participants were identified from a previously collected dataset from an exercise intervention study. After study exclusions (two excluded for age under 50 years old, four excluded for MMSE score ≤ 26, and two for less than 12 years of education) and MRI processing (20 excluded for not enough usable frames, and 7 excluded for errors in FreeSurfer segmentation), data from 50 participants were retained for further analysis. Participant characteristics are summarized in Table 1. Two-sample t-tests, unless otherwise noted, were used to compare the demographic characteristics between the included and excluded participants after MRI processing. There was a small, but statistically significant, difference in MMSE score (p=.048), however the average scores for included participants (mean±sd = 28.4±1.3) and excluded participants (mean±sd = 29.0±1.3) were similar. There were no significant differences in age (p=.825), years of education (p=.260), MDS-UPDRS-III score (p=.586), Hoehn and Yahr score (Chi-square test, p=.979), or years since diagnosis (p=.157).

Table 1.

Participant Characteristics.

	PD
N (female)	50 (23)
Age	66.1 ± 7.4
MMSE, median (range)	29 (27,30)
Years of Education	15.7 ± 2.0
MDS-UPDRS-III (OFF)	36.9 ± 10.9
H&Y (OFF), median (range)	2 (1,3)
Years since diagnosis	5.5 ± 4.6

Values are mean ± SD, unless otherwise specified. Abbreviations: Hoehn and Yahr, H&Y; Mini Mental State Examination, MMSE; Movement Disorders Society Unified Parkinson Disease Rating Scale Part 3, MDS-UPDRS-III.

3.2. Gait Characteristics

Velocity was significantly higher in the forward (mean±sd = 1.36±0.24) compared to the dual-task (mean±sd = 1.08±0.28) condition (p<.001). Step time CV was significantly lower in the forward (mean±sd = 3.92±1.06) compared to the dual-task (mean±sd = 6.30±2.80) condition (p<.001). The mean±sd for velocity dual-task change was −0.21±0.14 and step time CV dual-task change was 0.58±0.61.

3.3. Within-Network Functional Connectivity

Strength of rs-FC within the visual network correlated with velocity for forward (r=0.32, p=.024), dual-task (r=0.41, p=.003), and dual-task change (r=0.31, p=.027) (Figure 2). There were no other moderate or significant correlations between gait characteristics and strength of rs-FC within the other networks of interest.

Figure 2. Moderate within and between network correlations with gait characteristics.

Velocity for forward, dual-task, and dual-task change correlated with strength of functional connectivity within the visual network (A, B, C). Velocity for dual-task and dual-task change correlated with BG-thalamus internetwork rs-FC, and step time CV for dual-task change was negatively correlated with BG-thalamus internetwork rs-FC (D, E, F). Other correlations between velocity and functional connectivity between cognitive, attention, sensory, and motor networks are also represented (G, H, I).

3.4. Between-Network Functional Connectivity

Strength of rs-FC between BG and thalamus networks correlated with velocity for dual-task (r=0.45, p=.001) and dual-task change (r=0.41, p=.003), with a similar trend for forward (r=0.27, p=.061), such that stronger BG-thalamus internetwork rs-FC corresponded with better (i.e. faster) gait. Similarly, stronger BG-thalamus internetwork rs-FC was associated with lower (i.e., better) step time CV for dual-task (r=−0.33, p=.019), with a similar trend for forward (r=−0.27, p=.054). In addition, BG-DAN internetwork rs-FC and DAN-visual internetwork rs-FC were associated with higher velocity for dual-task (r=−0.30, p=.032; r=0.32, p=.024, respectively). Finally, strength of DMN-visual internetwork rs-FC correlated with velocity for forward (r=−.30, p=.034) (Figure 2). There were no other moderate or significant associations between velocity or step time CV with strength of internetwork rs-FC.

3.5. Stepwise Regression Analysis

We ran a stepwise regression model for dual-task velocity, which had the greatest number of moderate or significant associations with rs-FC scores, to compare the relationship between network-level rs-FC and clinical measures with gait characteristics. The full model included MDS-UPDRS-III, within-network visual rs-FC, and internetwork rs-FC for BG-thalamus, visual-DAN, and DAN-BG as predictors for dual-task velocity, and covaried for age and sex. The model was optimized using a stepwise regression method that used the AIC score to determine the model with the lowest prediction error, while covarying for age and sex. MDS-UPDRS-III and BG-thalamus internetwork rs-FC were significant and the best two predictors of dual-task gait velocity (Table 2).

Table 2.

Summary of stepwise regression analysis predicting dual-task velocity. After stepwise regression was performed, age and gender were included as covariates in the models. Model 1 is the full model with all variables and Model 2 is the optimized model based on AIC score.

	Model 1				Model 2
Variable	Estimate	Std. Error	t	p	Estimate	Std. Error	t	p
(Constant)	1.298	0.399	3.255	.002	1.327	0.340	3.909	< .001
MDS-UPDRS-III	−0.007	0.004	−2.029	.049	−0.007	0.003	−2.231	.031
BG-Thalamus	1.144	0.616	1.858	.070	1.159	0.542	2.141	.038
Visual-DAN	0.837	0.817	1.025	.311	0.849	0.519	1.637	0.109
BG-DAN	0.512	1.052	0.487	.629
Visual	0.283	0.484	0.586	.561
Adjusted R2	0.228				0.252
F-statistic	3.072				4.305
Model p	.011				.003
AIC	10.706				7.464

Abbreviations: Akaike Information Criterion, AIC; Basal ganglia, BG; Dorsal attention network, DAN; Movement Disorders Society Unified Parkinson Disease Rating Scale Part 3, MDS-UPDRS-III.

4. Discussion

This study aimed to examine associations between spatiotemporal gait characteristics and rs-FC strength within and between brain networks in people with PD. The results of the study showed that dual-task velocity was associated with strength of rs-FC within the visual network, and between motor and attention networks, most notably between the BG and thalamus networks. The results also demonstrate BG-thalamus internetwork rs-FC and motor severity were both predictive of gait velocity in people with PD.

4.1. Dual-Task Walking Involves Cognitive Networks

The dual-task condition decreased velocity and increased step time CV compared to the forward condition. This is consistent with previous research, demonstrating that dual-task conditions decrease gait speed and increase gait variability in people with PD, due to the increased cognitive demand [10]. In the present study dual-task velocity was associated with visual-DAN and BG-DAN internetwork rs-FC, implicating the involvement of attention networks in gait performance during dual-task. This is consistent with previous research showing dual-task is associated with motor prefrontal networks that are related to attention [8]. We also found associations between dual-task change velocity and strength of rs-FC within the visual network and between the BG and thalamus networks. This may further implicate these networks as being involved in not just the motor component of dual-task performance.

4.2. Role of the Visual Network in Gait

In the present study velocity was associated with rs-FC strength within the visual network and is consistent with previous findings [8]. This association may be due to the visual network’s involvement in visuospatial processing, which has previously been associated with visual-cerebellum internetwork rs-FC [11], and visuospatial performance being linked to gait performance in people with PD [24]. Studies have also shown rs-FC within the visual network to be altered in people with PD who experience freezing of gait, further implicating the visual network in gait performance in people with PD [25,26]. A previous study also reports that gait velocity is associated with rs-FC of vestibular, sensorimotor, and frontoparietal networks [8], which we did not find in the present study. This could be due to methodological differences in the rs-FC analysis and differences in seed set selection between studies.

4.3. Gait and internetwork rs-FC

The most notable associations in the present study were internetwork associations with spatiotemporal gait measures, including rs-FC between BG-thalamus, BG-DAN, visual-DMN, and visual-DAN networks. Considering the involvement of both the BG and thalamus in communicating with cortical areas for motor control, and the complex interplay of physiological and neurophysiological control required to execute and control gait, this result is not surprising [27]. For example, previous work demonstrated that rs-FC between BG-thalamus is affected in PD relative to controls [11], which indicates there is a specific neuropathology that could be related to the gait deficits seen in PD. Collectively, these results also suggest that the neurophysiological consequences of PD are not necessarily restricted to local neuropathology or dopamine depletion [28].

4.4. Gait Variability and rs-FC

Increased gait variability was associated with reduced BG-thalamus internetwork rs-FC in the dual-task condition only. This is predictable considering the involvement of the BG and thalamus in motor performance. And, this demonstrates the importance of internetwork communication in the regulation of gait stability. In the present study we did not find associations between gait variability measures and cognitive networks (including attention), which differs from previous research in older adults showing gait variability associated with DAN-DMN internetwork rs-FC [7]. However, dual-task gait velocity was associated with attention network rs-FC. Thus, it is possible that different aspects of gait performance are associated with different networks. Considering the importance of gait variability as an indicator of fall risk in older adults [3], it is important to further investigate the neural mechanisms of gait variability in neurological conditions.

4.5. No Associations Between Network-Level Cerebellar rs-FC and Gait

We did not find associations between gait characteristics and rs-FC within or between networks involving the cerebellum network. The cerebellum is important in motor processing, including dual-task performance [29], and previous research has implicated its hyperactivity as a potential mechanism to compensate for striatal dysfunction in people with PD [30]. However, a limitation in the present study was participants were tested in the OFF state of their medication, and were therefore in a dopamine-depleted state. This has been shown to reduce rs-FC in the cerebellum in people with PD, possibly due to decreased connectivity between the cerebellum and basal ganglia-thalamus networks [31], however the effects of dopamine on rs-FC are not fully understood. Another explanation for the lack of associations between gait characteristics and the cerebellum network could be that the cerebellum is a complex structure that involves more than just motor processing. Thus, there may be more regional specificity within the cerebellum for certain behaviors, such as gait.

4.6. rs-FC May Aid Prognosis

The results of the stepwise regression model showed that MDS-UPDRS-III scores and BG-thalamus internetwork rs-FC together provided the strongest prediction of dual-task gait velocity. In other words, the addition of BG-thalamus internetwork rs-FC information improves upon standard clinical ratings for predicting gait performance. These results have important implications for future rehabilitation studies. rs-FC has the potential to be a useful biomarker in clinical trials to help with prognosis and patient selection for who may respond best to various rehabilitation approaches. Future studies with larger sample sizes and longitudinal assessments should investigate the robustness of rs-FC as a potential prognostic biomarker for gait dysfunction in people with PD.

4.7. Limitations and Further Research

We applied stringent motion censoring criteria in the preprocessing of the data. This was done to optimize clean data for the analysis to reduce spurious results. However, this presents a limitation as participants with severe tremor may have been excluded from the analysis, making the results less generalizable. Particularly, the results related to the cerebellar networks may not be generalizable to patients with tremor, which has been linked to cerebellar dysfunction [32]. Another limitation of rs-FC is that data are acquired while the participants are at rest, not performing a task. This resting state methodology was used in the present study because walking tasks are not feasible to collect in an MRI scanner. Another limitation of this study was the lack of a control group. Future research should compare people with PD and healthy controls which could provide greater insights into the specificity of these results as they pertain to people with PD, however gait declines similar to those observed herein are rarely seen in healthly aging. Future research could also compare patients with different symptom pathologies (e.g., freezing of gait, severe tremor, etc.) to draw more generalizable conclusions.

4.8. Conclusions

The present study demonstrates that gait characteristics are associated with strength of rs-FC within and between brain networks in people with PD. This study suggests that networks related to motor control, visuospatial performance, and attention are associated with gait characteristics in people with PD. In particular, internetwork rs-FC was most frequently associated with gait characteristics in PD, emphasizing both the deficits for integrating across sensory, motor, and cognitive information with PD and the widespread neurophysiological consequences of PD neuropathology.

Highlights.

• Visual network rs-FC was related to gait velocity in people with PD

• rs-FC between motor and cognitive networks was related to gait velocity

• Gait velocity and variability were associated with BG-thalamus rs-FC

• BG-thalamus rs-FC and MDS-UPDRS-III scores were predictive of gait velocity