Parkinson’s disease (PD) is a progressive neurological disorder characterized by motor symptoms, such as bradykinesia, rigidity, tremor, and postural instability, as well as non-motor features, including hyposmia, sleep disorders, depression, and constipation (Jankovic, 2008; Schapira et al., 2017). Its pathological process is mainly attributed to disruptions in the nigrostriatal dopamine system (Hayes, 2019). Currently, diagnosis and severity assessment of PD are still predominantly based on clinical symptoms (Gelb et al., 1999; Postuma et al., 2015). Developing reliable non-invasive biomarkers to monitor disease severity is important for future diagnosis and disease-modifying therapies in PD.

Since the 1990s, advancements in the physics of complex systems (Strogatz, 2001; Albert and Barabási, 2002; Boccaletti et al., 2006) and the rise of network science (Börner et al., 2008) have allowed us to study the structure and function of brain networks in terms of small-world topology, highly connected hubs, and modularity. In recent years, graph theory analysis has been widely employed in the study of complex brain networks (Bullmore and Sporns, 2009). By using graph theory analysis to study the interactions between neurons and the structure and function of neural networks, it is possible to better understand the information transmission between neurons and how neural networks operate, thereby making deeper contributions to the function of the nervous system and the mechanism of disease (Craddock et al., 2013). Neuroimaging has shown that PD is a neurodegenerative disorder involving many neurotransmitters, brain regions, structural and functional connections, and neurocognitive systems (Weingarten et al., 2015). Therefore, the integrated analysis of the whole brain networks may provide a more comprehensive understanding of brain abnormalities in PD.

Altered topological properties in functional networks have been reported in PD by several studies (Sang et al., 2015; Luo et al., 2015; Fang et al., 2017; Göttlich et al., 2013; Suo et al., 2017). The majority of these studies have focused on the alteration of brain functional networks in early-stage PD. For example, the study examining the topological configuration of brain networks in early-stage PD patients who received antiparkinson treatment found that global properties, module structure, and hub distribution were markedly altered in these patients (Sang et al., 2015). The results of these studies were inconsistent due to clinical heterogeneity (dopaminergic medication) among participants. Furthermore, these studies mainly focused on early-stage patients and were therefore unable to identify progressive brain changes across different stages. The configurations of the brain functional connectome in patients with PD were perturbed and correlated with disease severity (Suo et al., 2017). However, the topological properties of large-scale brain functional networks in patients with different stages of PD are unknown.

Our study aims to use resting-state functional magnetic resonance imaging (rs-fMRI) and graph analysis approaches to compare changes in brain functional network topology in PD at different disease stages.

This study was conducted in accordance with the principles of the Declaration of Helsinki and was approved by the local human research ethics committee. Written informed consent was obtained from all participants (or their legal guardians) before enrollment.

In this study, we investigated the topological properties of brain functional networks in PD patients at different disease stages and HC. Patients with PD showed abnormalities at both the global level (decreases in the Cp, Elocal, and Lambda) and the nodal level (decreased nodal centrality metrics in the temporal-occipital regions, but increased in brain regions related to the default mode network and the frontoparietal control network). It is worth noting that patients with PD-M exhibit more extensive changes in topological attributes compared to those with PD-E. Furthermore, nodal centrality metrics of the MFG.L and ORBmid.R were positively correlated with the H&Y stage and UPDRS-III score. These findings may enhance our understanding of the mechanisms underlying the progression of PD and contribute to the development of non-invasive neuroimaging biomarkers for monitoring disease progression.

Networks with low construction costs but high efficiency in disseminating information are called economic small-world networks. Since the combination of high local clustering and short path length supports the two fundamental organization principles of functional segregation and functional integration in the brain, the small-world is an attractive model to describe brain networks (Zhang et al., 2011). In our study, the networks in PD patients were perturbed in a way that may reflect the underlying pathophysiologic abnormalities and disease progression. Regarding global topologic properties, Cp and Elocal were significantly decreased, whereas Lp and Eglobal were not significantly different in PD-M patients compared to HC. Since both Cp and Elocal measure local cliquishness of the network, and Lp and Eglobal are the most commonly used measures of functional integration (Latora and Marchiori, 2001), our results indicate a disturbance in the normal balance in functional brain networks of PD-M patients. These results are consistent with one functional connectome study, which reported decreased network segregation in drug-naive PD with no significant change in network integration (Luo et al., 2015). However, other studies (Suo et al., 2017; Ma et al., 2017) reported decreased local specialization and global integration in early-to-mid-stage PD and a progressive impairment in local specialization with an additional loss of global integration in PD. In accordance with reports, dopaminergic antagonists can reduce both global and local efficiency in healthy subjects (Achard and Bullmore, 2007), while dopamine-based medications can increase the functional connectivity between dopamine-related cognitive and motor pathways in healthy individuals (Kelly et al., 2007). Subsequent studies in PD patients have found that dopaminergic medication is also considered to partially restore the deficits in brain functional networks in PD patients (Delaveau et al., 2010; Palmer et al., 2010). In the above study, medication was withdrawn at least 12 h (off-state) in PD patients before resting-state functional MR imaging, and we were in the on-state, so we hypothesized that medication might contribute to the difference. In addition, PD-E patients only had lower Cp than HC participants in our study, which might be attributed to the effective symptom management of PD-E patients following medication. Future research could conduct medication-specific subgroup analyses, such as comparing medicated and unmedicated patients, to further elucidate the impact of pharmacological interventions on these findings.

In addition to these altered global topologic properties, PD patients at different disease stages have selectively and significantly impaired nodal centrality metrics in several regions of the brain’s functional networks, mainly including the decreased temporal-occipital regions and the increased brain regions related to the default mode network and the frontoparietal control network. Abnormalities of gray matter (Goldman et al., 2014), neuronal activity (Meppelink et al., 2009), and nodal centrality metrics (Luo et al., 2015) in temporal-occipital regions have been reported in PD patients compared to HC. The temporal-occipital regions are important for visual object recognition, so these changes are thought to be related to abnormal bottom-up visual processing of visual information in these regions and visual-cognitive deficits in PD patients (Meppelink et al., 2009). We found that PD-E patients had decreased nodal centrality metrics in the temporo-occipital brain regions, and the abnormalities were more obvious in PD-M patients. This finding is consistent with previous findings that have shown PD patients exhibit functional changes in the cortical visual system before visual symptoms are clinically evident (Cardoso et al., 2010; Uc et al., 2005). The default mode network and frontoparietal control network serve as core neural networks for cognitive control (Menon and D'Esposito, 2021). Previous research has indicated that even patients with PD-E, who have not yet been formally diagnosed with mild cognitive impairment, may experience subtle cognitive declines (Kandiah et al., 2009; Aarsland et al., 2017). Interestingly, although our study’s PD patients did not exhibit overt cognitive deficits, we observed significantly increased nodal degree and nodal efficiency within brain regions associated with the default mode network and the frontoparietal control network at different disease stages. This finding may reflect a potential adaptive response, characterized by increased connectivity and efficiency, which could support cognitive stability. However, further validation through targeted neuropsychological assessments is required to confirm compensatory processes in future studies.

The middle frontal gyrus is a key part of the prefrontal network that responds to inhibit responses to inappropriate stimuli by overriding the motor system’s automatic response tendency, and one study has shown that with age and the development of behavioral control, the specialization and organization of the response inhibition network improve. In contrast, the activation of the right middle frontal gyrus decreases significantly (Hardee et al., 2014). Inhibition control is an important executive function that involves inhibiting dominant responses or allowing appropriate actions to meet complex task requirements and adapt to different environments (Li and Sinha, 2007). A recent study on impulse control disorders in PD has shown that PD patients with impulse control disorders show reduced voxel-mirrored homotopic connectivity (VMHC) values in the MFG, middle and superior orbital frontal gyrus, inferior frontal gyrus, and caudate, which can detect altered interhemispheric connectivity by quantifying functional connections between the corresponding voxels in the two hemispheres of the brain (Gan et al., 2021). Therefore, this result suggested that the bilateral connections between the cerebral hemispheres of PD patients with impulse control disorders are altered, and the severity of impulse control disorders is correlated with the mean VMHC values of MFG. Our research findings indicate that as the H&Y stage and UPDRS-III scores advance in patients with PD, there is a corresponding increase in the nodal degree and nodal efficiency of the MFG.L and ORBmid.R. This result may complement previous studies suggesting that PD patients may compensate for the weak anatomical connections between the inhibited areas by increasing executive control.

Additionally, in module analysis, reduced functional connectivity was observed between the sensorimotor and visual modules in PD-M patients. Sensorimotor network plays a crucial role in integrating sensory inputs and facilitating motor execution, whereas the visual network contributes to spatial orientation and the processing of visual feedback (Bertoni et al., 2025; Kim et al., 2024). Previous research has shown that the sensorimotor network in PD patients is unable to effectively integrate feedback from the visual network, leading to motor dysfunction (Caspers et al., 2021). However, the results of our study did not survive multiple comparison correction and warrant further investigation to validate its clinical relevance.

There are several limitations to our study. First, the sample size was relatively small, which may have constrained our ability to detect the expected sensorimotor network impairments consistent with typical motor manifestations of PD. In the future, the sample size will be increased to continue the research. Second, the observation of reduced functional connectivity between sensorimotor and visual modules in PD-M patients did not survive multiple comparison correction and should be considered exploratory. Larger and independent datasets are needed to validate these findings, alongside studies using advanced neuroimaging approaches (e.g., task-based fMRI) to uncover their functional roles and biological foundations. Finally, all rs-fMRI data were acquired during the dopaminergic “ON” state to optimize patient comfort and scan quality. While clinically practical, this approach precludes the dissociation between medication effects and intrinsic PD-related network pathology. Previous studies demonstrate that dopaminergic agents can partially normalize aberrant functional connectivity patterns in PD patients (Delaveau et al., 2010; Palmer et al., 2010). Consequently, future studies should implement counterbalanced ON/OFF state designs with standardized medication withdrawal protocols. Additionally, the potential dose-dependent effects of levodopa equivalent daily dose (LEDD) on functional network topology were not examined due to incomplete pharmacological records. Future studies should incorporate stratified LEDD cohorts to resolve this critical confounding.

In conclusion, we used rs-fMRI and graph analysis approaches to compare changes in brain functional network topology in PD at different disease stages. We observed that patients with PD-M exhibited a reduction in Cp, Elocal, and Lambda. Additionally, both PD-E and PD-M patients demonstrated decreased nodal centrality metrics within the temporal-occipital regions, alongside an enhanced brain region related to the default mode network and frontoparietal control network. Notably, patients with PD-M exhibit more extensive changes in topological attributes compared to those with PD-E. Furthermore, nodal centrality metrics of the MFG.L and ORBmid.R were positively correlated with the H&Y stage and UPDRS-III score. These findings contribute to our understanding of the topological changes in the neural networks associated with the severity of PD.