Disrupted rich club organization of white matter networks in de novo Parkinson's disease with REM sleep behavior disorder

Introduction: Rapid eye movement (REM) sleep behavior disorder (RBD) is common in Parkinson's disease (PD) but the pathophysiological mechanism is still unclear in PD with RBD. The purpose was to explore the alterations of white matter network topology in de novo PD, with or without RBD. Methods: In total, this study contained 153 de novo individuals with PD and 59 healthy controls recruited from the Parkinson’s Progression Markers Initiative (PPMI) database. In such patients, 56 were probable RBD (PD-pRBD) and 97 were non-probable RBD (PD-npRBD) according to the RBD screening questionnaire (RBDSQ). All the recruited members had diffusion tensor imaging and complete neuropsychological assessments. The structural network of brain and organization of rich club were constructed based on deterministic ber tracking at the individual level. Results: Compared with PD-npRBD patients, the PD-pRBD patients showed diminished global eciency and elevated shortest path length. It was found that both the PD-pRBD and PD-npRBD had common hubs including putamen, middle occipital gyrus, precuneus, besides, PD-pRBD mainly recruited additional hubs in the frontotemporal paralimbic regions. Furthermore, the PD-pRBD showed decreased rich-club connections in comparison to PD-npRBD. Additionally, the nodal shortest path length in the right superior frontal gyrus, orbital part (ORBsup.R) was positively correlated to RBDSQ scores in PD-pRBD patients(r=0.25, p=0.01). Conclusions: The results indicated the PD-pRBD patients showed disrupted topological organization of white matter in the whole-brain and rich-club connections which mainly located in frontotemporal paralimbic regions, thus partially helps to illustrate the pathological processes of RBD in PD patients.


Introduction
Patients with Parkinson's disease (PD) have many non-motor symptoms, and the rapid eye movement (REM) sleep behavior disorder (RBD) is one of the most common ones, which is suggested by dream-related motor behavior and accompanied by phasic tension electromyography during REM sleep (Gagnon et al. 2002;Sudarsky and Friedman 2006). Recently, a meta-analysis showed that RBD occurs in as much as 25% of new PD patients and the percentage elevates as the symptom progresses, with an overall prevalence of 42.3% in the PD population (Destrieux et al. 2010; Kalia and Lang 2015;Marion et al. 2008). In addition, there is growing evidence indicating the exhibition of RBD in PD patients is associated with poorer motor and non-motor symptoms, namely, smell disorders, constipation, visual hallucinations, cognitive impairment, depression, impulse control disorders (Agosta et al. 2015; Arnaldi et al. 2016). However, the potential neurologic mechanisms of RBD in such patients remain to be elucidated.
Dopaminergic denervation of the striatum is a common feature in PD (Kalia and Lang 2015). Idiopathic RBD (iRBD) may be induced by neuronal degeneration in brain regions regulating the tone of skeletal muscle during REM sleep, including the pontine nucleus, lateral dorsal tegmental nucleus and locus coeruleus (Sudarsky and Friedman 2006). However, only a subset of PD patients develops RBD and half of iRBD patients develop PD (Marion et al. 2008). Therefore, other potential factors are likely to play key roles in RBD pathogenesis in PD.
A handful of magnetic resonance researches have been conducted to explore the probable pathogenesis of PD with RBD. A longitudinal study compared the changes in the thickness of cortex and the volume of subcortex in PD cases with or without RBD. In comparison to the non-probable RBD (PD-npRBD) subset, the thinning rate of the left insula in the probable RBD (PD-pRBD) subset was signi cantly increased, and the volume of the left caudate nucleus, globus pallidus and amygdala decreased over time(Yoon and Monchi 2021). Some studies have suggested that the neuropathological mechanism of PD with RBD may be related to abnormal spontaneous neuron activity patterns in cerebellum and visuomotor cortex, as well as enhanced connections between cerebellum and occipital and motor areas (Liu et al. 2021a). In addition, the differences in white matter microstructure in PD with RBD were explored by using diffusion tensor imaging(DTI), and it was shown that compared with PD-npRBD the fractional anisotropy(FA) of bilateral cingulate and left suboccipital tract was signi cantly reduced (Rahmani et al. 2016).
In contrast to approaches analyzing localized brain changes, lately, some new studies describe the human brain as a network model, and the structural and functional systems of the brain have the characteristics of complex networks. Graph theory is considered a good way to analyze such network. In graph theory, a network stands for a group of nodes or vertices and the edges or lines between them, and brain networks could be described quantitatively through multiple measurements (Bullmore and Sporns 2009;Sporns et al. 2005). Van and Power JD et al. (Power et al. 2013; van den Heuvel and Sporns 2011) demonstrated the existence of "rich club" formed by brain hubs of high-degree nodes. These nodes show a tendency to be more closely connected than those with lower degree values. Rich club provides signi cant information about the higher-level topology of brain networks.
In recent years, the white matter structural network connectome constructed by DTI has been increasingly applied in graph theory to reveal the alteratioons in the topological properties of brain structures, especially in PD patients (Li et al. 2017;Pereira et al. 2015). Several neuroimaging studies using DTI have shown that compared with health contrast (HC), PD patients had lower whole-brain clustering coe cient and lower global e ciency (Abbasi et al. 2018;Nigro et al. 2016). Liu et al (Liu et al. 2021b) revealed that in early PD patients, the rich club organization was disorganized and the connection strength between the peripheral regions was abnormal. However, it is still not clear whether PD-pRBD and PD-npRBD differ in brain white matter connectivity network.
To address this issue, this study focused on the alterations of structural network in the white matter of PD sufferers with or without RBD by using DTI. Based on the aforementioned clinical heterogeneity differences between PD-pRBD and PD-npRBD, it was hypothesized that the PD-pRBD could present convergent and divergent neural and anatomical abnormalities in the regions, which were correlated to the performance. These comparisons can probably contribute to clarify the common points and differences in neurobiological characteristics between PD-pRBD and PD-npRBD cases.

Participants and Clinical Evaluation
Marker Initiative 2011). Each PPMl site was approved by its own ethical committee. All PPMI subjects provided written informed consents prior to participating in the program. The analysis procedure in this study was carried out basing on the approved PPMI guidelines.
PD cases included in the database were newly diagnosed and treatment-naïve and we only selected data from the baseline visits. Given that polysomnography was not currently available in the database, the presence of RBD in PD individuals was decided by RBD screening questionnaire (RBDSQ). The RBDSQ is highly sensitive and reasonably speci c for RBD (Stiasny-Kolster et al. 2007). In the present study, PD patients with RBDSQ score ≥ 5 was de ned as PD-pRBD, < 5 was de ned as PD-npRBD(Kamps et al. 2019). HC with RBDSQ score ≥ 5 were excluded. Patients with cognitive impairment (Montreal Cognitive Assessment (MoCA) score < 23) were also excluded (Nasreddine et al. 2005). Finally, 56 PD-pRBD patients,97 PD-npRBD patients, and 59 HC were enrolled.

DTI Preprocessing and Tractography
PANDA software (Cui et al. 2013) was used to preprocess DTI images, mainly including: First, the non-brain tissue was stripped off. Secondly, the eddy-current effect and mild head motion were corrected. Then the diffusion tensor (DT) matrix was calculated based on the voxel. Finally, the Fiber Assignment by Continuous Tracking (FACT) algorithm was applied to conduct deterministic tractography. If the curvature angle was greater than 45• or the voxel fractional anisotropy (FA) < 0.2, the trace was terminated.

Network Construction
The T1WI images of all individual subjects were rst mapped to the corresponding B = 0 s/mm2 images to obtain jointly calibrated T1 images in the DTI space. Then the transformed TI images were nonlinearly transformed into ICBM152 TI template in Montreal Neuroscience Institute (MNI) space. According to automated anatomical labeling (AAL) atlas of MNI, the brain cold be divided into ninety regions, which were nodes of the brain structural network(supplementary table 1). A 90×90 weighted matrix was obtained by taking the number of bers (FN) in each pair of brain regions as the threshold value T. We set T ≥3 to indicate the existence of ber connections in brain regions, which was de ned as 1; otherwise, it was de ned as 0 (Shu et al. 2009) , and then PANDA was used to establish the binary matrix of the whole brain network.

Graph analysis of network topology
Global and nodal properties All network analyses were performed using the graph theory network analysis toolbox GRETNA (https://www.nitrc.org/projects/gretna)and the results were visualized by using the BrainNet Viewer toolbox (http://www.nitrc.org/projects/bnv).
The global properties of network include global e ciency (Eg), local e ciency (Eloc), cluster coe cient (Cp), shortest path length (Lp), and "small-world attribute" (σ)(Wang et al. 2020). Eg is considered as a signi cant parameter to estimate the transmission e ciency of global network. Eloc describes the network interconnection between nodes. Cp represents the average clustering coe cient of all nodes, mainly to measure the level of network interconnection. Lp refers to the average length of the shortest path between any two nodes in the network. A smaller Lp means a higher e ciency of network information transmission. The "small worldness" has a higher clustering coe cient and a smaller shortest path length, which ensures the transmission e ciency globally and locally. The study generated 1,000 random networks compared them with real ones to explore the subjects "small worldness". γ and λ respectively represent the ratios of Cp and Lp of real networks to that of random networks. σ is de ned as the ratio of γ to λ. When γ>>1 and λ≈1, or σ >1, the network was regarded as "small worldness" (Li et al. 2017).
The nodal properties of network include nodal e ciency (Ne), nodal shortest path length (Nlp), nodal cluster coe cient (Ncp), nodal degree centrality (Dc) and node betweenness (Bc). Ne refers to the e ciency of information transfer between node (i) and all other nodes in the network. Nlp refers to the shortest path from one node to all other nodes. Ncp refers to the density of connections between the neighbors of a node. Dc is de ned as the number of edges shared by node (i) and other nodes in the network. Bc de nes the fraction of the shortest path through node (i)(Wang et al. 2020).

Rich-club analysis
The central hub of global communication in the brain network described by rich club (van den Heuvel et al. 2012; van den Heuvel and Sporns 2011), stands for nodes with higher degrees within brain networks and nodes with a higher connection intensity(van den Heuvel and Sporns 2011). In this study, the group-averaged FN structural networks were constructed for the three groups separately. The weighted rich club coe cient φw(k), randomly rich club coe cient φ rand(k), and normalized rich club coe cient φ norm (k) for each group were calculated to describe rich club organization(van den Heuvel et al. 2012). The rich club regions refer to the node with the most consistent top 10% of the three groups(van den Heuvel et al. 2013). Three groups of connections were used to characterize the edge architecture of brain networks, which are: rich club connections (between rich club nodes), feeder connections (between rich and non-rich club nodes) and local connections (between non-rich club nodes) (van den Heuvel et al.

Statistical analysis
The demographic characteristics and clinical data were compared by using IBM SPSS 22 software among the three groups. Distribution-based Chi-square test was applied to compare qualitative data, and one-way ANOVA or Kruskal-Wallis H test was used for quantitative data. The signi cance level was p <0.05. Signi cance values were adjusted by the Bonferroni correction for multiple tests. After controlling age, sex, and education years, one-way ANOVA was applied to compare differences in global and local attributes among the groups. For global parameters, p < 0.05 was considered signi cant; for nodal parameters, signi cance values were adjusted by the Bonferroni correction for multiple tests. Post hoc analysis was performed to further explore differences among the groups. At the same time, Bonferroni correction was also carried out. Comparison of rich club, feeder and local connection among groups was analyzed by one-way ANOVA and post hoc two-sample T test. All graph theory parameters were analyzed using GRETNA statistical model. At last, multiple linear regression analysis was applied to test the relationship between structural network parameters and RBDSQ scores in the PD-pRBD group.

Demographic and clinical variables
No signi cant differences were found in age, gender and education level among the three subsets (p>0.05). There were no signi cant differences in HVLT Immediate/Total Recall, HVLT Discrimination Recognition, HVLT Retention, MoCA, SCOPA-AUT, UPSIT and UPDRS-III between PD-pRBD and PD-npRBD patients (p>0.05). (Table 1). Table 1 Characteristics of HC, PD-npRBD patients, and PD-pRBD patients Signi cant values were adjusted by the Bonferroni correction for multiple tests. *The signi cance threshold was set at p<0.05.

Global properties of network
No signi cant differences were found in the global properties of Cp and Eloc among HC, PD-npRBD and PD-pRBD groups, but they were found in Eg and LP. Compared with the Hc group, PD-npRBD patients exhibited signi cantly increased Eg and decreased Lp in post-hoc analysis. However, in comparison to PD-npRBD, PD-pRBD patients exhibited signi cantly lower Eg but higher Lp (p< 0.05). All groups showed that σ > 1 represented the small worldness property of WM connection network, and there is no statistical difference among the three groups (Fig. 1).

Nodal properties of network
Signi cant differences were con rmed in Ncp among the three groups. Compared with HC, the Ncp of PD-npRBD patients was all increased in the right superior occipital gyrus, left middle occipital gyrus, bilateral precuneus, bilateral putamen, bilateral superior temporal gyrus, and left middle temporal gyrus (p<0.05, Bonferroni correction).
However, there were no signi cant differences between PD-pRBD and PD-npRBD group, HC and PD-pRBD group (Table 2). Other nodal properties, such as Ne, Np, Dc and Bc, showed no signi cant difference within such groups.  The signi cance threshold was set at p<0.05. **post hoc Bonferroni test, * Bonferroni uncorrected.

Rich-club analysis
It was illustrated the rich club organization found in the structural networks of HC, PD-npRBD and PD-pRBD groups.
For the three groups, the normalized rich club coe cients φ w norm (k) >1 over the range of degrees from k = 1-15 were found (Fig. 2). On contrast, there were no signi cant differences in φ w norm (k) among the three groups(p>0.05, Bonferroni-corrected). The brain network hubs of the three groups were determined as the top 10% of the most connected region and the brain center region according to node degree. The following hubs were discovered in HC: left superior frontal gyrus, of PD-RBD do not exist in PD-npRBD, whereas the temporal hubs (bilateral temporal pole: superior temporal gyrus (TPOsup), left middle temporal gyrus (MTG.L)) of PD-nRBD do not exist in PD-pRBD. Noticeably, these different hub regions were mainly located in the fronto-temporal paralimbic system. (Fig. 3d-f). PUT.R Subcortical 12 The regions in bold represent different hub regions between the PD-npRBD and PD-pRBD groups. The abbreviations of the 90 brain regions are given in supplementary materials (Online Resource).
When describing the edge structure in the context of rich club organization, it was found that rich club connections (p = 0.021) increased in the PD-PRBD group compared with HC, while rich club connections (p = 0.018) decreased signi cantly in the PD-pRBD group compared with PD-npRBD group (permutation test, 10 000 times, Fig. 3C). There was no signi cant group effect being found in the feeder connection between rich and non-rich club regions (p = 0.093) and the local connection between non rich club regions (p = 0.075) (Fig. 4).

Correlation between graph parameters and clinical variables
Multiple linear regression analysis showed that no signi cant correlation was found between RBDSQ scores and global parameters. Nlp of the right superior frontal gyrus, orbital part (ORBsup.R)was positively associated with RBDSQ scores (r=0.25, p=0.01). p < 0.05 was considered as signi cant (Fig. 5).

Discussion
In this study, DTI and graph theory methods were applied to evaluate the topological properties of WM networks in The ndings of decreased Eg and increased Lp may indicate that the inter-nodal organization of the whole brain was excessively destructed in PD-pRBD cases. Even through, the altered structural topological properties in PD with RBD were barely explored, former researches (Arnaldi et al. 2016) have indicated that PD with pRBD presenting more severe motor and non-motor symptoms than PD without pRBD, suggesting more severe and widespread neurodegeneration. It might indicate that normal topological network organization was impacted, namely, the separation and integration of brain networks was disrupted.
Our results con rmed that all three groups of brain networks demonstrate the presence of network hubs, comprising RBD symptoms exhibited decreased cholinergic innervation in the neocortical, limbic and thalamic cortex. In addition to molecular level, there are also some radiographic evidences that the paralimbic cortex may play a vital role in PD-pRBD. An MRI structure study of PD with RBD showed that, in comparison to the control group, inferior temporal cortex, supra-frontal bilaterally, and left rostral middle frontal cortices were extensively atrophied in PD patients, which suggested that RBD symptoms of PD patients appeared with signi cant paralimbic/limbic cortical changes (Pereira et al. 2019). In accordance with the previous studies, our results also revealed that the disruption of WT in the frontotemporal paralimbic regions may be associated with the neuropathological process of PD with RBD.
In this study, we discovered a decrease of rich-club connections in the PD-pRBD compared with PD-npRBD, while no signi cant differences of feeder connections and local connections were found between the two groups. It probably revealed that the rich-club connections among central hubs, which worked as high-cost and high-capacity backbone of global brain communication (Power et  The study also had several limitations. First, we grouped PD using RBDSQ scores rather than polysomnography, which is the standard for diagnosis and quantitative description of RBD. However, RBDSQ score has high sensitivity and speci city for identifying RBD(Stiasny-Kolster et al. 2007). Second, a cross-sectional study with patients in the early stage of disease was conducted, which should be included in a longitudinal follow-up study thereafter to further explain the pathogenesis of PD-RBD. Third, we adopted the deterministic ber tracking method to estimate the structural connectivity between regions, and the intersecting directions of multiple optical bers cannot be determined. Therefore, in order to accurately measure the ber trajectory, probabilistic tractography algorithm and diffusion spectral imaging technology will be used to solve these problems in the future. Fourth, other non-motor features may affect brain network structure were not considered, which should be taken into consideration as control variables in the future.

Conclusion
Page 12/16 Taken as a whole, our study indicated the impaired global properties of network and diminished rich-club connections in PD-pRBD patients.  Figure 1 One-way ANCOVA of global parameters among the three groups. Signi cant differences of global parameters of the structural network were detected among the three groups, performed by one-way ANCOVA, adjusting for age, gender and education years. p < 0.05 was set signi cant. Afterwards, two-sample t-test was employed to explore further differences. **Post hoc tests were corrected by Bonferroni correction with a signi cant different p < 0.017 (0.05/3).

Figure 2
The normalized rich club coe cient φ w norm (k) curves of the three groups were > 1 at the level of k=1-15 (p< 0.05, Bonferroni corrected). Different colored curves represent different groups of φ w norm (k). Black: HC; blue: PD-npRBD; red: D-pRBD. No signi cant difference in rich club organization was found among the three groups.   Table 3.L left hemisphere, R right hemisphere.

Figure 4
Comparison of group-averaged rich-club organization between HC, PD-npRBD and PD-pRBD groups. Edges across individual brain networks were divided into three distinct classes: rich club connections, feeder connections and local connections. Signi cant difference was shown in rich-club connections between HC, PD-npRBD and PD-pRBD groups, while no signi cant group effect was found in feeder connections and the local connections. Correlation between signi cant topology properties and severity of RBD in PD-PRBD group. A positive correlation between Nlp of ORBsup.R and RBDSQ scores was found in PD-PRBD patients. p < 0.05 was set signi cant. The association was explored by multiple linear regression. p < 0.05 was set signi cant.

Supplementary Files
This is a list of supplementary les associated with this preprint. Click to download. supplementarytable1.docx