Altered static and dynamic functional network connectivity in Parkinson's disease: A multisite functional magnetic resonance imaging study

Abstract

Aberrant static functional network connectivity (sFNC) and dynamic functional network connectivity (dFNC) have been inconsistently reported in Parkinson's disease (PD), reflecting the limitations of single-site studies with restricted sample sizes. To address these limitations, we leveraged pooled resting-state functional magnetic resonance imaging (rs-fMRI) data from a multisite study to explore FNC differences in PD. Data from 160 patients with PD and 75 age matched and sex-matched healthy controls (HCs) were collected from four repositories (PPMI, OpenfMRI, FCP/INDI). Independent component analysis combined with sliding-window and k-means clustering approach was applied to examine the FNC among seven resting-state networks: sensorimotor network (SMN), auditory network (AN), visual network (VN), cognitive control network (CCN), default mode network (DMN), subcortical network (SC), and cerebellar network (CB). The FNC and dynamic indices (fraction time, mean dwell time, transition number) were calculated. Static and dynamic measures were compared between the two groups. For sFNC, the PD group exhibited predominantly decreased connectivity between the VN and CCN/AN/SMN, and between the AN and SMN; alongside increased connectivity between the AN and SC/CB. For dFNC, four distinct recurrent states were identified. Significant group differences emerged in states 1, 3, and 4, characterized by widespread abnormalities in both intra- and inter-network connectivity (e.g., DMN, CCN, SC, CB, and sensory networks). Temporally, PD patients demonstrated altered dynamics, including reduced mean dwell time and fractional time in state 4, while showing increased mean dwell time in state 1 and fractional time in states 1 and 3. These multisite rs-fMRI findings demonstrate that PD is characterized by complex patterns of connectivity disruption and inflexible temporal network reconfigurations. Notably, dFNC metrics revealed extensive connectivity reductions that were not apparent in static analysis, highlighting their superior sensitivity and potential utility as biomarkers for tracking disease progression and therapeutic response.

Highlights

• •Altered static functional network connectivity revealed in Parkinson's Disease.
• •Four distinct dynamic connectivity states were identified in this multisite analysis.
• •Decreased network interactions in a specific dynamic state characterize PD.
• •PD patients exhibit longer dwell times in a differentiated sensory state.

1. Introduction

Parkinson’s disease (PD) is a complex, progressive neurodegenerative disorder primarily characterized by the progressive loss of dopaminergic neurons in the substantia nigra pars compacta and the pathological accumulation of α-synuclein protein aggregates within the brain (Poewe et al., 2017). The clinical presentation of this condition is predominantly defined by cardinal motor symptoms including as tremor, rigidity, and bradykinesia. However, non-motor manifestations—such as cognitive impairment, neuropsychiatric disturbances, and autonomic dysfunction—also play a significant role in contributing to the overall disability experienced by patients and the subsequent reduction in their quality of life (Chaudhuri et al., 2006). With the prevalence of Parkinson’s disease projected to double by the year 2040, it is evident that this condition represents an increasingly pressing global health challenge that demands attention and resources (Dorsey, Bloem, 2018a). Currently, the diagnosis of PD remains largely clinical, often proving to be unreliable during the early stages of the disease, and currently available therapies do not effectively modify the progression of the disease itself (Postuma et al., 2015). Consequently, there is an urgent and pressing need for the development of objective biomarkers that can accurately track the underlying pathology, predict patient outcomes, and provide insights into the pathophysiology of neural networks beyond the nigrostriatal system—such as the cortico-basal ganglia-thalamic circuits—thereby enhancing our understanding and management of this debilitating disorder.

Resting-state functional magnetic resonance imaging (rs‑fMRI) has reframed PD as a disorder of large‑scale brain networks. By measuring spontaneous low‑frequency blood oxygen level-dependent fluctuations, rs‑fMRI maps the brain’s intrinsic functional architecture. Static functional network connectivity (sFNC) analysis, which averages temporal correlations over an entire scan, has consistently revealed network dysregulation in PD. Alterations in connectivity within the sensorimotor network frequently correlate with motor deficit severity (Helmich et al., 2010, Wu et al., 2011). Notably, pathological changes also affect networks supporting higher‑order cognition. Increased connectivity within the default mode network (DMN) is linked to early cognitive impairment (Tessitore et al., 2012, Disbrow et al., 2014). Additionally, disruptions in visual, dorsal attention, and frontoparietal networks are linked to hallucinations and executive dysfunction (Yao et al., 2014, Gorges et al., 2015). Nevertheless, sFNC assumes temporal stationarity and may obscure meaningful neural dynamics.

The brain’s functional connections are intrinsically time-varying, exhibiting a remarkable ability to adapt and change over time. Dynamic functional network connectivity (dFNC) captures this phenomenon by identifying recurring, quasi-stable connectivity states that reflect the brain’s shifting patterns of interaction (Hutchison et al., 2013). In the context of PD, dFNC reveals a reduced temporal flexibility, where patients demonstrate a significant decrease in their ability to switch between different connectivity states, often spending prolonged periods in segregated and hypo-connected states that hinder optimal brain function (Kim et al., 2017, Díez-Cirarda et al., 2018). Importantly, these dynamic metrics frequently correlate more strongly with the severity of both motor and non-motor symptoms than traditional static measures; thereby, positioning dFNC as a potentially more sensitive circuit-level biomarker for assessing disease progression (Fiorenzato et al., 2019). Recent research has shifted its focus from merely describing these dynamic connectivity changes to understanding the underlying mechanisms at play. Critically, similar alterations in network connectivity are detectable even in the prodromal phase of the disease, such as in individuals with idiopathic rapid eye movement sleep behavior disorder, indicating the presence of early network dysfunction that may herald the onset of the disease cascade (Rolinski et al., 2016). Building on this, studies have highlighted the existence of symptom-specific network phenotypes, revealing distinct connectivity patterns associated with various symptoms such as fatigue, apathy, and specific cognitive domains (Baggio et al., 2015, Hou et al., 2022). The integration of multimodal approaches, particularly with positron emission tomography, has begun to elucidate the links between neurochemical deficits and network dysfunction; for instance, dopaminergic denervation has been shown to modulate corticostriatal connectivity, while the integrity of cholinergic systems correlates with the functioning of attention networks (Albin et al., 2022).

Despite these advances, a key translational bottleneck remains. The literature reporting these aberrant sFNC and dFNC patterns in PD is marked by inconsistency. For instance, one study reported increased DMN connectivity (Shen et al., 2024a), another found decreases or mixed patterns (Shen et al., 2024b). However, most findings derive from single-center studies with limited sample sizes and diversity, making them vulnerable to site-specific biases that undermine generalizability and reproducibility—a major concern for biomarker development (Marek et al., 2022). To address this, multisite studies with harmonized protocols are essential to validate findings, establish normative benchmarks, and capture the full heterogeneity of PD (Marek et al., 2018). Although consortia such as the Parkinson’s Progression Markers Initiative have demonstrated feasibility, a comprehensive multisite examination encompassing both static and dynamic connectivity across a large and diverse PD cohort is still lacking.

The present study leverages a large, multisite rs-fMRI dataset to conduct an integrative analysis of altered sFNC and dFNC in PD. We hypothesize the following. PD patients will show dysconnectivity in both sFNC and dFNC compared to healthy controls (HCs). Moreover, dynamic metrics will provide complementary explanatory power and superior predictive accuracy for clinical variability relative to static measures. By delivering robust, generalizable evidence, this work aims to advance functional network metrics toward clinically actionable tools for PD management.

2. Material and methods

2.1. Participants

Data were obtained from three sources: the Parkinson’s Progression Markers Initiative (PPMI) database (http://www.ppmi-info.org/data), the OpenfMRI database (accession number ds000245; https://openfmri.org/dataset/ds000245), and the FCP/INDI database (http://fcon_1000.projects.nitrc.org/indi/retro/parkinsons.html), including the NEUROCON dataset and the Tao Wu dataset. Informed consent was obtained in writing from each participant; all experimental protocols were approved by the respective Institutional Review Boards of the involved institutions.

All rs-fMRI datasets were inspected for the number of time points and slices to ensure data quality. Participants were excluded based on the following criteria: (1) discrepancies in the number of time points or slices compared to the standard protocol; (2) being non-right-handed; or (3) insufficient data quality.

2.2. Data preprocessing

All fMRI data preprocessing was conducted using the Data Processing & Analysis for Brain Imaging toolkit (DPABI, http://rfmri.org/DPABI), which is based on Statistical Parametric Mapping (SPM12, https://www.fil.ion.ucl.ac.uk/spm) and runs within MATLAB 2021a. The preprocessing steps were as follows (Allen et al., 2014): (1) the initial 10 time points were discarded; (2) slice timing correction was applied; (3) head motion was corrected; (4) spatial normalization was performed to align images with the Montreal Neurological Institute (MNI) space, resampling voxel dimensions to 3 mm × 3 mm × 3 mm; (5) spatial smoothing was conducted using a Gaussian kernel with a full width at half maximum (FWHM) of 4 mm; (6) detrending and regression of nuisance signals were performed, including Friston’s 24 head motion parameters as well as white matter (WM) and cerebrospinal fluid (CSF) signals. Following preprocessing, head motion parameters for all participants remained below 2.0 mm in maximum displacement across the x, y, z axes and below 2.0° in angular rotation about each axis.

2.3. Spatial group independent component analysis

Group spatial independent component analysis (ICA) was conducted utilizing the Group ICA of fMRI Toolbox (GIFT, version 4.0b, https://trendscenter.org/software/gift/), following a validated pipeline established by previous neuroimaging investigations (Allen et al., 2014, Espinoza et al., 2019). First, an information maximization algorithm (informax) was employed to perform ICA on all preprocessed data. Given that the data were sourced from multi-site, we input site-specific TR to mitigate the impact of scanning parameter differences across sites, and subsequently used the minimum description length (MDL) criterion to estimate the number of independent components (ICs), resulting in the classification of the data into 26 ICs (Li et al., 2007). This informax process was executed 100 times using the ICASSO software package to ensure the reliability and precision of the estimates (Himberg et al., 2004). Subsequently, back-reconstruction of the ICs, including spatial maps and time series, was performed for each participant. Fisher z-transformations were then applied to the time series. Ultimately, 16 were deemed physiologically relevant intrinsic connectivity networks rather than noise, based on criteria applied in earlier studies (Allen et al., 2014, Espinoza et al., 2019). Specifically, the peak cluster locations were in gray matter with minimal overlap with white matter, brain edges, or ventricles, and these components exhibited an increased ratio of low-frequency to high-frequency activity. These ICs were then categorized into seven functional networks, namely: sensorimotor network (SMN), auditory network (AN), visual network (VN), DMN, cognitive control network (CCN), subcortical network (SC), and cerebellar network (CB). Importantly, these seven functional networks have been frequently reported in earlier rs-fMRI studies (Allen et al., 2014, Barkhof et al., 2014) and have been assessed concerning dFNC (Allen et al., 2014, Barkhof et al., 2014, Espinoza et al., 2019).

2.4. Static functional network connectivity

We performed the analysis of sFNC using the MANCOVAN toolbox within the GIFT framework. We derived the sFNC by computing the Pearson correlation coefficient between the time courses of distinct resting-state networks (RSNs). In this study, we identified a total of 16 RSNs, resulting in the formation of a 16 by 16 correlation matrix for each participant. We then applied Fisher’s z-transformation to these correlation coefficients. To assess the pairwise differences in functional network connectivity among the 16 RSNs between the HCs and PD groups, we conducted a two-sample t-test. Age, sex, and site were included as covariates to control for potential confounding effects, including site-related variability introduced by direct multi-center data merging. The significance threshold was set at p < 0.05, and corrections for multiple comparisons were applied using the false discovery rate (FDR) method.

2.5. Dynamic functional network connectivity

The dFNC analysis was conducted using the dynamic FNC Toolbox incorporated within the GIFT package, following methodologies established in previous studies (Allen et al., 2014, Espinoza et al., 2019). Initially, to diminish the effects of various artifacts, such as those generated by heartbeat, respiration, and movement, additional post-processing techniques were applied to all time courses. These techniques included linear, quadratic, and cubic detrending. Multiple regression analysis was performed on the six realignment parameters—which represent head motion corrections—and their temporal derivatives. Despiking was carried out using the 3D-despike method, a procedure designed to reduce spike artifacts. Finally, low-pass filtering was applied with a high-frequency cutoff set at 0.15 Hz.

Subsequently, a sliding window approach was employed to compute the dFNC. Data from subjects in the multisite study were acquired with different TRs; therefore, prior to postprocessing, the component time courses were temporally resampled to the shortest TR (2 s) (Faghiri et al., 2018). For postprocessing, a window size of 30 TRs was selected based on prior findings suggesting that a window length in the range of 30–60 s is optimal for effectively capturing dynamic information (Hutchison et al., 2013, Faghiri et al., 2018). The window was advanced by a step size of 1 TR, meaning the window moved forward by one time repetition at each step. A tapered window was constructed by convolving a Gaussian function with a standard deviation (σ) of 3 TRs with a rectangular function. For each window, a comprehensive correlation matrix was generated across 16 ICs. The graphical least absolute shrinkage and selection operator (LASSO) method was subsequently utilized to regularize the covariance matrices, as this approach has demonstrated effectiveness in minimizing noise impact associated with a limited number of time points (Faghiri et al., 2018). Within each time window, the temporal correlation between the IC time courses was assessed using Pearson’s correlation coefficient for each IC pair. To evaluate the recurrence of FNC patterns, k-means clustering was conducted on the time-varying covariance matrices to categorize the observed patterns into four distinct FNC states. The optimal cluster count (k = 4) was determined through the elbow criterion, a method that identifies the point where adding more clusters does not significantly improve clustering performance, as outlined in prior studies (Allen et al., 2014, Espinoza et al., 2019). During post-processing, age, sex, and site were regressed out as nuisance covariates in the dFNC correlation analysis to account for potential confounding effects, including site-related variability arising from the merging of multi-center data.

For each subject, a representative dFNC matrix for each state was computed based on the average dFNC matrix corresponding to that state. These matrices were then compared across groups using two-sample t-tests to identify differences in dFNC at specific states. Given that not all subjects exhibited all states, only the available representative dFNC matrices were utilized for conducting t-tests at each state. P-value correction was executed employing the FDR method, with differences considered statistically significant at p < 0.05.

To further characterize these states, three temporal metrics were evaluated: the fraction of time (FT) spent in each state relative to the total scan duration, the mean dwell time (MDT) remained in a specific state, and the number of transitions (NT) between states, where a higher count suggests reduced stability of each state over time. The nonparametric Mann-Whitney U test was employed to compare these dFNC indices between the two groups, with a significance threshold set at p < 0.05.

2.6. Statistical analysis

The demographic attributes, specifically age and gender, were assessed using SPSS version 26.0 (IBM Corp., Armonk, NY, USA), with a statistical significance threshold set at p < 0.05. Comparative evaluations between individuals diagnosed with PD and the HCs were performed at each research location. Differences in sex distribution were analyzed using the χ² test, whereas differences in age were investigated using independent samples t-tests. Subsequently, network metrics between the PD group and HCs group were compared by two-sample t-test (p _FDR < 0.05). Temporal metrics were compared between the two groups using the nonparametric Mann-Whitney U test (p < 0.05).

3. Results

3.1. Demographic and clinical characteristics

In the PPMI study, a total of 162 individuals diagnosed with PD and 52 HCs were initially identified. However, 58 PD patients and 18 HCs were subsequently excluded from the analysis due to incomplete or suboptimal quality of rs-fMRI data. Additionally, participants with head motion exceeding 2.0 mm displacement during scanning were excluded, resulting in the removal of 6 PD patients and 10 HCs. After applying these exclusion criteria, the final dataset for analysis comprised 122 participants, including 98 PD patients and 24 HCs. In the OpenfMRI dataset, 30 participants were incorporated, consisting of 15 PD patients and 15 HCs. Regarding the FCP/INDI dataset, it included a total of 83 participants, comprising 47 PD patients (27 from NEUROCON and 20 from Tao Wu) and 36 HCs (16 from NEUROCON and 20 from Tao Wu). Notably, across all datasets utilized in this research, no statistically significant differences were observed in terms of age or sex between the PD and HC groups (p > 0.05). The demographic and clinical statistics are summarized in Table 1.

Table 1.

Demographic information of patients with PD and HCs.

	Participants		Sex (M/F)			Age (years, mean [SD])			Scanner	Voxel(mm)/ TR(s)	Scan time (min)/ (#vols)
Dataset	PD	HCs	PD	HCs	P-value	PD	HCs	P-value
PPMI	98	24	65/33	14/10	0.46	63.4 (9.7)	63.2 (12.9)	0.94	Siemens Tim Trio 3 T	3.3 × 3.3 × 3.3/ 2.4	8.4/ 210
OpenfMRI	15	15	6/9	7/8	0.71	64.4 (7.2)	63.3 (5.2)	0.65	Siemens Magnetom 3 T	3.0 × 3.0 × 3.5/ 2.5	8.25/ 198
NEUROCON	27	16	16/11	5/11	0.08	68.7 (10.6)	67.6 (11.9)	0.76	Siemens/ Avanto 1.5 T	3.8 × 3.8 × 5.0/ 3.48	8.05/ 137
Tao Wu	20	20	11/9	12/8	0.75	65.2 (4.4)	64.8 (5.6)	0.78	Siemens Magnetom 3 T	4.0 × 4.0 × 5.0/ 2.0	8.0/ 239

Abbreviations: F, female; M, male; PD, Parkinson’s disease; HCs, healthy controls; SD, standard deviation; TR, repetition time.

3.2. Networks of interests

From the 26 ICs, 16 were selected as regions of interest and categorized into seven RSNs: SMN (IC14), AN (IC2,15,16), VN (IC7,8,11), DMN (IC18,19,21,22), CCN (IC10,13), SC (IC6,26), and CB (IC9) (Table 2, Fig. 1).

Table 2.

Resting-state networks classified based on functional domains, independent component numbers, and Montreal Neurological Institute peak coordinates.

Functional domains and resting-state network		IC number	MNI peak coordinate (x, y, z)
SMN	L postcentral gyrus	14	−39, −40, 56
AN	R temporal pole: superior temporal gyrus	2	39, 14, −25
	L superior temporal gyrus	15	−63, −22, 8
	L Temporal pole: superior temporal gyrus	16	−36, 8, −28
VN	R lingual gyrus	7	27, −88, −19
	L lingual gyrus	8	0, −70, 5
	R middle occipital gyrus	11	30, −85, 20
CB	R cerebellum crus	9	6, −82, −25
DMN	R angular gyrus	18	42, −58, 47
	L precuneus	19	0, −64, 50
	L anterior cingulate and paracingulate gyri	21	−3, 53, 2
	L cuneus	22	0, −73, 35
CCN	L inferior parietal gyri	10	−33, −67, 47
	L inferior frontal gyrus	13	−48, 20, −10
SC	R thalamus	6	3, −16, 11
	L caudate nucleus	26	−3, 14, −1

IC, independent component; MNI, Montreal Neurological Institute; SMN, sensorimotor network; AN, auditory network; VN, visual network; DMN, default mode network; CCN, cognitive control network; SC, subcortical network; CB, cerebellar network; L, left; R, right.

Fig. 1.

Spatial maps of the 16 independent components that were observed in the resting-state network and grouped into seven functional domains according to their anatomical and functional properties: sensorimotor (SMN), auditory (AN), visual (VN), default mode (DMN), cognitive control (CCN), subcortical (SC), and cerebellar (CB) networks.

3.3. Differences between groups in sFNC

For sFNC analysis, the PD group exhibited significantly decreased interaction between the VN and CCN/AN/SMN, and between the AN and SMN; along with a specific increase connectivity between the AN and SC/CB, compared to the HCs group (p < 0.05, FDR correction; Fig. 2).

Fig. 2.

Significant differences in the static functional network connectivity between Parkinson’s Disease (PD) and healthy controls (HCs) groups. Statistical threshold was established at a false discovery rate (FDR)-corrected p < 0.05.

3.4. Differences between groups in dFNC

Four states of FNC that occurred repeatedly were derived. State 1 (pan-network hyperconnectivity state): Representing 17% of windows, this state exhibited high connectivity within and between most networks, with the SC showing mild negative couplings with other RSNs (Fig. 3A). State 2 (relatively modular network state): Occurring in 28% of windows, this state featured strong connectivity within RSNs, particularly within the DMN, with relatively weak inter-network connectivity (Fig. 3B). State 3 (sparse connectivity state): The most frequent state, occurring in 44% of windows, characterized by marked sparsity in FNC both within and between RSNs (Fig. 3C). State 4 (differentiated network state): Occurring in 11% of windows, this state exhibited strong positive FNC among sensory-related networks (VN, AN, SMN), together with selective positive couplings between these networks and the DMN (Fig. 3D).

Fig. 3.

Functional network connectivity matrix of the clustering analysis in each state. The total number of occurrences and percentage of total occurrences for each state are displayed.

Significant dFNC differences between groups were observed in states 1, 3, and 4 (Fig. 4). In state 1, connectivity between the DMN and SC, and between the VN and AN/SMN decreased; conversely, connectivity within the AN, and between the AN and SC increased (p < 0.05, FDR correction; Fig. 4A). In state 3, altered connectivity distinguished PD and HC groups across connections between the DMN and CCN/CB/VN, between the AN and CCN, between the VN and SMN, as well as between the SC and CCN/SMN/VN (Fig. 4B). In state 4, PD patients exhibited connectivity reductions, primarily involving connections between the VN and AN/SMN, as well as within the VN, and AN; notably, the sole significantly increased connection was observed within the CCN (Fig. 4C). Collectively, altered functional connections predominantly involved the DMN, AN, VN, CB, CCN, SC, and SMN.

Fig. 4.

Compared to the healthy controls (HCs) group, the Parkinson’s Disease (PD) group exhibited altered dynamic functional network connectivity in state 1, state 3 and state 4.

3.5. Temporal properties of FNC states

Mean dwell time in state 4 was significantly lower in the PD group relative to the HCs group (PD: 0.50 [0.00, 1.00]; HCs: 1.50 [0.00, 15.38]; p = 0.046; Fig. 5; Table S1), while exhibiting significantly increased mean dwell time in state 1 (p = 0.023; Fig. 5; Table S1). Additionally, the PD group showed significantly reduced fractional time in states 4 compared with the HC group, while exhibiting significantly increased fractional time in states 1 and 3 (all p < 0.05; Fig. S1A; Table S1), but the number of transitions was not significantly different between PD and HCs groups (Fig. S1B; Table S1).

Fig. 5.

Violin plots displaying mean dwell times in States 1–4 for healthy controls (HCs) and Parkinson’s Disease (PD) groups. Mann-Whitney U test results showed a significant difference in mean dwell time between the HCs and PD groups in state 1 (p value = 0.023) and state 4 (p value = 0.046).* p < 0.05; ns: Not Significant.

4. Discussion

This multisite rs-fMRI investigation provides a comprehensive examination of sFNC and dFNC alterations in PD using one of the larger aggregated neuroimaging cohorts. Through rigorous analysis of data from 160 PD patients and 75 HCs across multiple sites, we demonstrate that PD is characterized by: 1) predominantly decreased sFNC involving visual, auditory, and sensorimotor networks alongside specific auditory-subcortical and auditory-cerebellar hyperconnectivity; 2) state-specific dFNC alterations characterized by predominantly decreased connectivity in multiple states, particularly showing reduced connectivity between higher-order, sensory, and cerebellum networks; and 3) significantly reduced whole-brain network flexibility characterized by reduced mean dwell time and fractional time in the differentiated state, and increased mean dwell time and fractional time in the hyperconnected and sparse states. Consistent with our hypothesis, these findings indicate that PD is characterized by widespread network-level dysfunction, particularly marked by state-specific alterations in dynamic connectivity. These results substantially advance understanding of PD as a large-scale network disorder and provide novel mechanistic insights, with potential implications for biomarkers monitoring disease progression and treatment response.

Although these results provide only partial support for our hypothesis regarding widespread network-level dysfunction, the sFNC results revealed a more nuanced dysconnectivity pattern in PD. While a specific increase in AN-SC/CB connectivity was observed, the most prominent findings were significant decreases in connectivity between the VN and SMN/CCN, aligning with the hypothesis that PD pathology selectively disrupts specific sensory-cognitive integration pathways (Helmich et al., 2010, Jouve et al., 2010). In contrast to prior studies that focused on posterior cortical atrophy in various dementia subtypes, our data indicate alterations in the interaction between auditory processing networks and basal ganglia loops even in non-demented PD cohorts, suggesting these alterations may serve as an early biomarker preceding overt cognitive decline (Devignes et al., 2022). Furthermore, the heightened connectivity between the auditory network and subcortical structures may reflect a compensatory recruitment of alternative sensory channels aimed at bypassing impaired dopaminergic gating mechanisms. Notably, this phenomenon appears distinct from the specificity observed in chemosensory networks documented in olfactory dysfunction, where trigeminal sensitivity is often preserved despite profound olfactory loss (Tremblay et al., 2020). Moreover, this specific disconnection between visual and sensorimotor regions supports emerging evidence linking such dysconnectivity to impaired visual-motor integration, a core neural mechanism underpinning symptoms such as gait instability and freezing of gait (Kicik et al., 2025). Importantly, the diminished functional connectivity between the VN and the CCN identified in individuals with PD may significantly compromise the neural integration of visual inputs into advanced executive and attentional functions. This weakened interaction is posited to disrupt the essential top-down regulation required for sensorimotor transformation, consequently exacerbating fundamental motor impairments including bradykinesia and postural instability (Schettino et al., 2006, Si et al., 2023). These static network anomalies suggest a large-scale reorganization in which primary sensory inputs increasingly fail to modulate higher-order cognitive control systems, thereby establishing a neural substrate for the heterogeneous motor and non-motor symptoms observed in PD.

In dFNC analysis, four recurrent brain states were identified. Our results demonstrate that PD patients exhibit diminished flexibility in transitioning between brain states, characterized by prolonged dwell times in the hyper-connected state, and reduced occupancy of a differentiated state. This propensity to sustain a highly synchronized configuration is consistent with observations in levodopa-induced dyskinesia, where excessive coupling between visual and sensorimotor networks dominates the dynamic profile (Si et al., 2023). This similarity implies that the inflexible brain dynamics in PD may share pathophysiological mechanisms with hyperkinetic movement disorders despite differing clinical manifestations (Hirato et al., 1995, Dwi Wahyu et al., 2021). Moreover, the shorter dwell time in the differentiated state (state 4) align with findings in patients with freezing of gait, who similarly exhibit a tendency toward rigid connectivity patterns, suggesting a potential common pathophysiology involving visuomotor network dynamics (Su et al., 2025). While previous studies have reported that dopaminergic medication can partially normalize dwell times in states related to top-down control, our cohort, comprising untreated or mixed-medication patients, demonstrates that this intrinsic dynamic rigidity persists as a core pathological feature of the disease (Chen et al., 2021, Averna et al., 2026). In contrast to the sparse connectivity patterns frequently associated with excessive daytime sleepiness (Verma et al., 2026), the predominance of a hyper-connected state in our PD cohort points to a maladaptive over-synchronization that impairs the brain’s ability to flexibly reallocate resources for complex cognitive and motor tasks (Malagurski et al., 2017, Lench et al., 2021). Consequently, these dynamic metrics provide more sensitive indicators of neural inflexibility than static connectivity indices by capturing the failure to disengage from inefficient network configurations that may contribute to motor dysfunction.

The analysis of the specific neural network underlying these dynamic aberrations revealed the engagement of the CB, CCN and the DMN, highlighting the critical role of extra-nigral pathways in the broader pathophysiology of PD. This key finding directly challenges the long-standing conventional focus predominantly centered on the basal ganglia-thalamocortical circuit. In the identified aberrant dynamic states, the attenuated functional connectivity specifically associated with the CB contradicts the view of the cerebellum as a primarily compensatory hub that stabilizes information processing during cortical network impairment (Shen et al., 2024b, Antonucci and Jay, 2025). Subsequent graph theoretical analyses further support this interpretation, demonstrating that significant decreases in global network efficiency are accompanied by compromised local topological reorganization in both cerebellar and sensorimotor networks, ultimately leading to markedly reduced functional output and overall system flexibility (Zhang et al., 2023). Furthermore, the increased functional connectivity with the CCN alongside decreased connectivity with the DMN revealed a critical imbalance between externally directed cognitive control and internally directed mental operations in PD. The heightened CCN connectivity contradicts the typical notion of uniform network degeneration, instead reflecting a potentially maladaptive hyperactivation. This hyperconnectivity likely represents a failed compensatory mechanism, wherein the already strained executive system disproportionately recruits neural resources to counteract ongoing dopaminergic denervation and maintain cognitive-motor output, particularly under the demands of complex tasks (Amboni et al., 2015). Clinically, this aberrant CCN hyper-connectivity aligns with the pervasive executive dysfunction observed in PD patients—such as impaired set-shifting and deficits in working memory—where the brain's excessive effort to maintain goal-directed behavior paradoxically results in cognitive fatigue and inflexibility rather than enhanced performance (Ratajska et al., 2025, Zhang et al., 2024). Conversely, the concurrent attenuation of DMN connectivity aligns with the aforementioned network destabilization, underscoring a progressive loss of intrinsic functional integrity that mirrors the clinical emergence of apathy and impaired episodic memory retrieval (Lee et al., 2020). Crucially, the opposing trajectories of CCN and DMN connectivity strongly indicate a severe disruption of their physiological anti-correlation, a dynamic antagonism essential for efficient cognitive flexibility (Putcha et al., 2015). The inability to flexibly suppress DMN activity while abnormally amplifying CCN engagement locks the brain into a rigid, resource-depleted dynamic state. This loss of dynamic balance not only exacerbates the characteristic bradyphrenia—evidenced by the prolonged latency observed in patients in transitioning between spontaneous thought and focused problem-solving—but also hastens the progression from isolated motor deficits to widespread cognitive deterioration (Breukelaar et al., 2020). Consequently, this CCN-DMN dysregulation characterizes a maladaptive process where the diminished capacity to sustain functional homeostasis manifests clinically as a pathologically rigid and inflexible cognitive-behavioral repertoire.

Our results robustly support conceptualization of PD as a network disorder in which pathology disrupts both local and global brain communication. The observed sFNC and dFNC alterations likely result from specific neurodegenerative processes characteristic of PD. Histopathological evidence demonstrates that alpha-synuclein aggregate accumulation and consequent striatal dopaminergic depletion lead to long-range white matter tract degeneration, compromising the structural scaffolding necessary for efficient functional synchronization (Bellucci et al., 2016). For example, decreased VN-SMN connectivity aligns with post-mortem and diffusion tensor imaging studies showing microstructural damage to the superior longitudinal fasciculus and optic radiation in PD, white matter structures physically linking visual and motor cortices (Gu et al., 2014, Lee et al., 2014, Chondrogiorgi et al., 2016). Thus, observed functional dysconnectivity likely reflects downstream impacts of structural disconnection rather than representing a mere epiphenomenon. Future research should explore how these functional alterations relate to structural connectivity, as structural-functional network decoupling has been reported in PD (Zou et al., 2024).

Several limitations should be acknowledged in interpreting our results. First, although our sample was large and drawn from multiple sites, the cross-sectional design prevents causal inferences about disease progression. Second, PD heterogeneity presents a significant obstacle in neuroimaging research (Dorsey et al., 2018b), with subgroup-specific connectivity patterns being essential for tailored medical interventions. However, subgroup analyses—differentiating early-stage from late-stage PD or distinguishing between tremor-dominant and akinetic-rigid forms—proved impracticable due to the scarcity of phenotypic data in public datasets. Recent evidence indicates that different motor subtypes exhibit distinct dynamic connectivity characteristics (Shen et al., 2024a, Shen et al., 2024b, Pan et al., 2025). Future studies with even larger samples would enable more detailed subgroup analyses and potentially identify distinct network phenotypes corresponding to different clinical manifestations of PD. Third, our multisite data acquired at 1.5 T and 3 T may be subject to systematic differences in BOLD contrast and noise levels. Although site was included as a covariate in the regression, residual scanner effects cannot be fully eliminated, potentially confounding group comparisons. Therefore, these findings should be interpreted with caution. Future studies employing harmonized acquisition protocols was needed to validate these results across field strengths. Additionally, due to the absence of standardized clinical severity data (e.g., Unified Parkinson’s Disease Rating Scale scores) in the aggregated public datasets, we could not perform correlation analyses between sFNC/dFNC measures and clinical symptoms. Therefore, claims regarding the direct relationship between our findings and specific clinical symptoms remain speculative. Future work should integrate prospective clinical assessments, such as standardized motor and non-motor evaluations, to address this gap.

In conclusion, this large multisite study demonstrates that PD is associated with significant alterations in the sensorimotor, cognitive, auditory, visual, and default mode networks. Additionally, dFNC analysis revealed connectivity reductions that were not apparent in the static FNC analysis, along with altered temporal dynamics characterized by reduced dwell time in a differentiated sensory state. These findings provide new insights into the neural mechanisms of PD and highlight the potential of dFNC as a biomarker for the disease. Future research integrating multimodal neuroimaging and clinical data will be essential to translate these findings into clinical practice and develop network-based biomarkers for diagnosis and treatment monitoring in PD.

Funding

No funding was received for this manuscript.

Ethics approval and consent to participate

Participants' data were sourced from the PPMI, OpenfMRI, FCP/INDI database. Local research ethics committees approved all PPMI-, OpenfMRI-, FCP/INDI-affiliated studies.

Patient consent

Participants' data were sourced from the PPMI, OpenfMRI, FCP/INDI database. Prior to the study's commencement, every participant furnished written informed consent.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Appendix A. Supplementary material

Supplementary material

Data Availability

Data used in this Article are publicly available from the Parkinson’s Progression Markers Initiative (PPMI) database (http://www.ppmi-info.org/data), the OpenfMRI database (accession number: ds000245) (https://openfmri.org/dataset/ds000245), and FCP/INDI [NEUROCON dataset and Tao Wu dataset] (http://fcon_1000.projects.nitrc.org/indi/retro/parkinsons.html).