Aberrant dynamic functional network connectivity in progressive supranuclear palsy

Background: The clinical symptoms of progressive supranuclear palsy (PSP) may be mediated by aberrant dynamic functional network connectivity (dFNC). While earlier research has found altered functional network connections in PSP patients, the majority of those studies have concentrated on static functional connectivity. Nevertheless, in this study, we sought to evaluate the modifications in dynamic characteristics and establish the correlation between these disease-related changes and clinical variables. Methods: In our study, we conducted a study on 53 PSP patients and 65 normal controls. Initially, we employed a group independent component analysis (ICA) to derive resting-state networks (RSNs), while employing a sliding window correlation approach to produce dFNC matrices. The K-means algorithm was used to cluster these matrices into distinct dynamic states, and then state analysis was subsequently employed to analyze the dFNC and temporal metrics between the two groups. Finally, we made a correlation analysis. Results: PSP patients showed increased connectivity strength between medulla oblongata (MO) and visual network (VN) /cerebellum network (CBN) and decreased connections were found between default mode network (DMN) and VN/CBN, subcortical cortex network (SCN) and CBN. In addition, PSP patients spend less fraction time and shorter dwell time in a diffused state, especially the MO and SCN. Finally, the fraction time and mean dwell time in the distributed connectivity state (state 2) is negatively correlated with duration, bulbar and oculomotor symptoms. Discussion: Our findings were that the altered connectivity was mostly concentrated in the CBN and MO. In addition, PSP patients had different temporal dynamics, which were associated with bulbar and oculomotor symptoms in PSPRS. It suggest that variations in dynamic functional network connectivity properties may represent an essential neurological mechanism in PSP.


Introduction
Progressive supranuclear palsy (PSP) is a rare neurodegenerative disease characterized by tau inclusion bodies in neurons and glial cells, which is the most common primary four repeats (4R) tauopathy, accompanied by neurofibrillary tangles (NFTs) or neuropil threads (tau protein) in the basal ganglia and brainstem (Armstrong, 2018;Tinaz, 2021).It consist of a huge spectrum of motor, cognitive, and behavioral impairments.These encompass akinesia, early falls due to postural instability, oculomotor deficits, frontal-executive dysfunction, and neuropsychiatric facets such as apathy and impulsivity (Boxer et al., 2017;Höglinger et al., 2017).PSP Richardson's syndrome (PSP-RS) is the most frequent form of medical phenotype in the disease (Williams et al., 2005).The PSP study confirmed by autopsy pathology displayed that about 40% of PSP patients showed atypical clinical variants (vPSP), including PSP with predominant parkinsonism (PSP-P), PSP with progressive gait freezing (PSP-PGF), and PSP with corticobasal syndrome (PSP-CBS), as well as primary lateral sclerosis (PSP-PLS) (Höglinger et al., 2017).Despite significant attempts over the previous decade, the pathophysiological mechanism of PSP is still largely unknown.
It is now possible to effectively and non-invasively examine the human brain due to recent developments in neuroimaging technology.Resting-state fMRI is an method that shows promise for exploring brain function (Barkhof et al., 2014;Guo et al., 2012;Liu et al., 2017).Instead of isolated brain regions, abnormal multiple interconnected brain systems are believed to be associated with functional impairments in PSP.Several studies have discovered aberrant functional connectivity (FC) measurements in patients with PSP. Brown et al. (Brown et al., 2017) and Gardner et al. (Gardner et al., 2013) mapped intrinsic connectivity to the midbrain tegmentum (MT) network including rostral midbrain tegmentum (rMT) and dorsal midbrain tegmentum (dMT).Most of the FC alterations have been localized to the corticosubcortical and corticobrainstem interactions.Furthermore, Whitwill et al. (Whitwell et al., 2011) found significantly reduced FC between the thalamus and premotor cortex including supplemental motor area (SMA), striatum, thalamus and cerebellum in PSP.However, recent research indicate that abnormal temporal dynamics of brain networks may act as a mediator for PSP, that the impact, increased the proportion of time in networks associated with higher cognitive abilities, was correlated with decreased neural signal complexity and clinical severity as measured by the PSPrating-scale (Whiteside et al., 2021).
Resting-state functional connectivity (rs-FC), which represents spontaneous brain activity, is created by observing interactions between signals that are reliant on blood oxygen levels in various brain regions when a person is resting (Barkhof et al., 2014;Biswal et al., 1995;Shen, 2015).Besides, the time scale variations are eliminated by this technique because it assumes that the connectivity strength between various areas remains largely constant throughout the entire scan.However, some researchers claim that dynamic brain activity occurs during resting conditions (Hutchison et al., 2013).Dynamic FC has gained interest as a result of developments in fMRI study (Allen et al., 2014;Calhoun et al., 2014;Damaraju et al., 2014).Dynamic functional network connectivity (dFNC) studies have gained popularity recently, and the framework has been used extensively to study neuropsychiatric disorders like schizophrenia (Damaraju et al., 2014;Yang et al., 2022), Alzheimer's disease (Fu et al., 2019), Parkinson's disease (Fiorenzato et al., 2019), and major depression disorder (Wu et al., 2019).For instance, Whiteside et al. (Whiteside et al., 2021) found that participants with PSP spent less time in states with subcortical and posterior activation, and more time in frontoparietal states.
We compared the dFNC in a relatively large group of 53 PSP patients with that in 65 age and handedness-matched normal controls, drawing inspiration from earlier researches.Concisely, we firstly utilized a group independent component analysis (ICA) to extract resting-state networks (RSNs), and sliding window correlation approach was used to generate dFNC matrices.These matrices were then clustered using the K-means algorithm into various dynamic states, and state analysis was subsequently employed to analyze the dFNC and temporal metrics between the two groups.Based on earlier research (Whiteside et al., 2021), they discovered that PSP had been in an inefficient brain state for a long time.However, our primary objectives in this research were divided into two aspects.One is to determine whether PSP had altered dynamic properties and, if so, whether those disease-related changes were related to clinical variables.Therefore, we hypothesize that PSP varies in dFNC, that PSP performs differently in various functional states, and that the time index of dFNC is related to the severity of the disease and other symptoms, such as bulbar and oculomotor symptoms.In contrast to earlier PSP studies, which frequently focused on networks of interest, in this work, we focused on the whole-brain FC at the network level.This investigation was useful in identifying possibly altered network interactions in PSP and may offer fresh perspectives for extending our understanding of the neuropathological causes of this illness.

Participants
The current study was authorized by the Ethics Committee of Qilu Hospital, Shandong University.Before beginning any study procedures, each individual provided written informed consent.Based on the new MDS-PSP clinical requirements (Höglinger et al., 2017), 53 PSP patients were enrolled, and all of them were diagnosed as probable PSP or possible PSP: (1) presence of typical clinical symptoms of PSP, including vertical supranuclear gaze palsy, repeated unprovoked falls within years, progressive gait freezing within 3 years and frontal cognitive/ behavioral presentation; (2) the patients are older than 45 and <80 years old; (3) the dose of Parkinson's treatment is stable one month before joining the group.In addition, to ensure the accuracy of the diagnosis, we had continuous clinical follow-up of all included patients.In the course of the follow-up, all five patients who were initially diagnosed with possible PSP have progressed to probable PSP.These were the exclusion criteria: (1) younger than 18 years or older than years; (2) having magnetic resonance contraindications; (3) diagnosis of other neurological disorders; (4) MRI quality insufficient for analysis.Furthermore, sixty-five age-matched normal controls (NC) were enlisted.They were questioned to make sure there were no obvious anomalies and that none of them had a history of neurological or mental disorders.Based on the Chinese updated version of the Edinburgh Handedness Inventory, all of the patients and controls were righthanded (Oldfield, 1971).
We evaluated daily activities and behavioral symptoms, ocular motor deficits, motor impairment causing bulbar symptoms, limb motor and gait deficits using the Progressive Supranuclear Palsy Rating Scale (PSPRS), which was developed as a quantitative measure of disease severity and disability (Golbe and Ohman-Strickland, 2007).Moreover, the Unified Parkinson's Disease Rating Scale-I (UPDRS-I) scores and Unified Parkinson's Disease Rating Scale-II (UPDRS-II) scores were applied to asses mental behavior, emotions and activities of daily life.The Unified Parkinson's Disease Rating Scale-III (UPDRS-III) scores and the Hoehn and Yahr (H -Y) stage were used to evaluate the intensity of motor symptoms.The Freezing of Gait Questionnaire (FOGQ) scale was developed to measure the severity of freezing of gait.The Montreal Cognitive Assessment Scale (MoCA) and Mini Mental State Examination (MMSE) were applied to assess cognitive function.The Hamilton Anxiety Rating Scale (HAM-A) and Hamilton Depression Rating Scale (HAM -D) were administered to estimate the presence and severity of anxiety and depression, respectively.The health related quality of life was assessed with the Spanish version of the 39-item Parkinson's Disease Questionnaire (PDQ-39).The overview of methodology is shown in Fig. 1.

Data acquisition
A Siemens verio 3.0 T MR scanner (Erlangen, Germany) was used to capture the MR pictures with a eight-channel head coil at Qilu Hospital, Jinan, China.Tight and snug foam padding was accustomed to minimize head movement.To reduce scanner noise, we provided earplugs to each individual.Gradient-Echo Single-Shot Echo-Planar Imaging sequence (GRE-SS-EPI) with the following imaging parameters was used to acquire resting-state functional MRI (fMRI) data: repetition time/echo time = 2000/30 ms; field of view = 220 mm × 220 mm; matrix = 64 × 64; flip angle = 90 • ; slice thickness = 3 mm; slice gap = 1 mm; transversal slices; 180 volumes.All participants were told to close their eyes, to be as still as possible, to think of nothing specific, and to not fall asleep during the fMRI scans.A magnetization-prepared rapid acquisition gradient echo (MPRAGE) sequence was employed to obtain sagittal 3D T1-weighted images (repetition time/echo time = 2000/2.3ms; inversion time = 900 ms; flip angle = 9 • ; matrix = 256 × 256; slice thickness = 1 mm, no gap; 192 slices).

Data preprocessing
The fMRI data were preprocessed using the SPM12 (http://www.fil.ion.ucl.ac.uk/spm) and DPABI software (Yan et al., 2016).The first 10 volumes of each functional time series were discarded in order to achieve the signal equilibrium.After being realigned to the last volume, the remaining 170 volumes had their acquisition time delay between various segments adjusted.Through estimating the translation in each direction and the angular revolution on each axis for each volume, head movement parameters were obtained.Ultimately, Eight people were left out of the study because of maximum displacement in any orthogonal directions exceeding 3 mm, maximum head rotation exceeding 3 • , or average frame displacement exceeding 0.5 mm.small amounts of head motion, as mentioned in a prior research, may have an effect on the functional connectivity findings (Power et al., 2012).After motion was corrected with a linear transformation, the individual high-resolution T1 images were coregistered to the mean resting-state fMRI image.The gray matter, white matter, and cerebrospinal fluid were then separated from the transformed structural images using a new segmentation algorithm and the Diffeomorphic Anatomical Registration Through Exponentiated Lie Algebra (DARTEL) method.Additionally, utilizing exponential Lie algebra and diffeomorphic anatomical registration, a brain template was generated.Through the normalization parameters calculated by DAR-TEL, the movement-corrected functional volumes were spatially normalized to the Montreal Neurological Institute (MNI) space and resampled to 3mm 3 isotropic voxels.4 mm full width at half maximum was used by DARTEL to smooth the normalized fMRI data.

Group independent component analysis
The Group independent component analysis (ICA) of the fMRI Toolbox (Version4.0b,http://mialab.mrn.org/software/) was adopted to conduct the independent component analysis (ICA) for each individual participant in three steps: data dimension reduction, independent components estimation and back-reconstruction.The data was reduced into 16 components using a two-step principal component analysis, with the component number chosen by the minimum description length standard (Li et al., 2007).Then, to determine the most steady and dependable components, the Infomax algorithm (Bell and Sejnowski, 1995) was applied in independent components (ICs) estimation, which was executed 100 times using the ICASSO method (Himberg et al., 2004).The components of each unique topic were then backreconstructed using a dual-regression technique.Subsequently, All individuals' time courses and spatial maps from the ICs were acquired after back-reconstruction, and the subject-specific maps were then transformed to Z-scores.According to the following criteria (Beckmann et al., 2005;Cordes et al., 2000;Damoiseaux et al., 2006;Zuo et al., 2010), all ICs were assessed here based on the group IC maps: RSNs displayed peak activations in gray matter, had time courses dominated by lowfrequency variations and demonstrated low spatial overlap with known white matter structures, vascular, encephalocoeles, motion, and susceptibility artifacts.Ultimately, seven functionally significant RSNs were found for the studies that followed.

dFNC calculation and construction
The sliding window technique is the most popular method for studying dFNC; we performed this analysis using GIFT's dFNC toolbox.As in earlier research (Allen et al., 2014;Damaraju et al., 2014; The analysis includes the following steps: (1) fMRI data were preprocessed; (2) ICA was conducted and 7 RSNs were identified; (3) dFNC matrices were calculated and constructed in all different sliding windows for each subject; (4) clustering and state analyses were conducted to investigate the dFNC changes.Abbreviations: dFNC, dynamic functional network connectivity; ICA, independent component analysis; RSN, resting-state network Fiorenzato et al., 2019), resting state data were split into windows of 22 repetition times (44 s) in steps of one repetition time because it has been shown that this section length offers a reasonable balance between the accuracy of correlation matrix estimation and the capacity to resolve dynamics (Allen et al., 2014).In addition, in order to increase sparsity in the graphic LASSO structure with 100 repeats, we added a second L1 norm of the accuracy matrix (Friedman et al., 2007).Fisher's z-transformation was used to convert all functional connectivity matrices to zscores after calculating dFNC in order to fix variance before further analysis.Following Fisher z-transformation, matrices were regressed out with nuisance factors like age, gender and head motions.

Clustering analysis
All dFNC matrices were clustered applying the k-means method, and the frequency and structure of recurrent functional network connectivity patterns were evaluated.Since the L1 (Manhattan) distance has been shown to be a effective measure for high-dimensional data, we used it to determine how comparable the dFNC matrices are to one another (Aggarwal et al., 2001).Further, a cluster analysis was performed on the examples of all the participants to estimate the ideal number of clusters.Specifically, a subsampling study along the temporal dimension was performed within each subject to lessen computational requirements and window duplication.In brief, we chose subject exemplars as those windows with local peaks in FC variation in the first step, led by earlier research (Allen et al., 2014;Shen et al., 2016).Second, in order to avoid local minima, k-means clustering was applied to all exemplars and repeated 500 times with random starting cluster centroid locations.The elbow criterion, which calculates as the ratio of within-cluster distance to between-cluster distance, was used to identify the optimum number of clusters.

State analysis
We were able to better understand the temporal characteristics of dFNC states by three measures, such as the mean dwell time, fractional of time, and number of transitions between each state.The "mean dwell time" is the total number of windows that belong to one state over time, the "fraction time" is the total number of windows that belong to one state, and the "number of transitions" is the number of windows that switch between states.These definitions indicate the reliability of each state.A two-sample t-test [False discovery rate (FDR) correction, P < 0.05] was used to investigate group differences in dwell time, fraction time, and the number of transitions between PSP patients and normal subjects.

Correlation analysis
The associations between dynamic measures and clinical factors such as disease duration, start age, disease severity, cognitive status and each item with PSPRS were evaluated in the patient group using Spearman's rank correlation coefficient once substantial between-group differences were discovered in any dynamic measures.The FDR adjustment was applied to all findings with a p-value <0.05 to account for multiple comparisons.

Demographics and clinical characteristics of the participants
This study had 118 participants in total, including 53 patients with progressive supranuclear palsy and 65 normal controls.Due to head motion exceeding a translation of 3 mm or an angular rotation of 3 • , 6 PSP and 2 NC were removed.The remaining individuals consisted of 47 PSP patients and 63 normal controls.The gender difference (χ 2 = 6.15, p = 0.013) between the PSP patients and normal controls was statistically significant.The findings showed that the two groups were comparable in terms of age (64.13 ± 5.60 years for the PSP and 64.66 ± 6.54 years for the NC; p = 0.647) and handedness (all participants in both groups are right-handed).The detailed demographics and clinical data are displayed in Table 1.

Resting-state networks
Figure 2 displays spatial maps of each of the 11 independent components discovered via group independent component analysis.Based on previous investigations of resting-state brain networks (Allen et al., 2014;Fiorenzato et al., 2019;Liu et al., 2017;Yang et al., 2022), independent components were organized into the following six networks: subcortical cortex network (SCN, IC 1 and 5), sensorimotor network (SMN, IC 2), visual network (VN, IC 6), cerebellum network (CBN, IC 10), attentional network (AN, IC 8 and 12), and default mode network (DMN, IC 7, 9, and 13).In addition, based on their anatomical and assumed functional characteristics, we identified a novel network in the medulla oblongata (MO, IC 3) into different groups.

dFNC state analysis
We used the k-means algorithm to cluster the dFNC matrices from all patients.The cluster centroid is represented by each matrix in Fig. 3, which is sorted in the order of emergence, and it indicates the FC state of the data.The sign and the amount of connectivity between RSNs are distinct between different matrices.Although there were other disparities as well, we were able to clearly define the differences between the various states here in terms of their strong linkages.To more clearly illustrate the varied pattern among FC states, we maintained the strongest 5% connections from each state for better presentation (Fig. 3).State 1 revealed positive connectivity among the VN, SMN, and CBN but widespread negative connectivity of the DMN and the VN, SMN, AN, and CBN.State 2 displayed distributed strong connections between RSNs, especially the MO and SCN.State 3 shared the strong correlations mainly concentrated among the SMN, VN, CBN and AN.In state 4, the whole network showed strong connectivity, as characterized by positive correlations between the AN and DMN.However, the whole network showed sparse and weak connectivity in state 5, with the DMN being negatively correlated with the SMN.

Differences between groups in dFNC
After FDR adjustment for multiple comparison, the group differences in state 5 (FDR p < 0.05) were discovered using the two sample t-test approach in Fig. 4. Compared with normal controls, We found the patients with PSP showed significantly increased connections between MO and VN/CBN.However, only significantly decreased connections were found between DMN and VN/CBN, SCN and CBN in state 5.There was no significant alteration of dFNC in other states.

Temporal properties
As shown in Fig. 5, there are differences of dFNC features within each state between both groups.
The fraction time of PSP patients in state 2 was significantly lower than that of the normal controls.The results of mean dwell time indicated that patients with PSP spend less time in the state of distributed connectivity.However, we found no significant group differences in the number of transitions in any state (all p > 0.05).

Correlation analysis
To determine if dFNC features were correlated with clinical characteristics, correlation analyses were performed.We discovered that in state 5, the connection between IC 2 and IC 10 shows a positive correlation with freezing gait (r = 0.5920, p = 0.004), whereas the connection between IC 3 and IC 10 is negatively correlated with motor impairment (r = − 0.6597, p = 0.001).The connections between IC 4 and IC 9, 10 are positively correlated with freezing gait and ocular movement respectively (r = 0.5431, p = 0.009; r = 0.6305, p = 0.002), and the connection between IC 7 and IC 10 is positively correlated with motor impairment (r = 0.5763, p = 0.005).In addition, the connectivity between IC 7 and IC 4,5 is positively correlated with freezing gait and eye movement respectively (r = 0.5812, p = 0.005; r = 0.4425, p = 0.039).Furthermore, we found that mean dwell time in state 2 was negatively correlated with duration, bulbar and oculomotor symptoms (r = − 0.332, p = 0.039; r = − 0.3921, p = 0.014; r = − 0.322, p = 0.046).Additionally, it is worth noting that the temporal properties of dFNC did not correlate with other clinical traits like H -Y, MMSE, FOGQ and so on.

Discussion
This study investigated the alterations of dFNC and the relationship between the time index of dynamic network in patients with PSP and the bulbar and oculomotor dysfunction.Our research produced the following conclusions: (1) patients with PSP showed increased connectivity strength between MO and VN/CBN and decreased connections were found between DMN and VN/CBN, SCN and CBN.(2) PSP patients spend less fraction time and shorter dwell time in a diffused state, especially the MO and SCN.(3) significant correlations were observed among the changes in dFNC, dynamic indicators, and clinical characteristics in patients.
As noted above, we discovered significant between-group functional  network connectivity (FNC) differences in a number of RSNs, including the DMN, SCN, CBN, VN, and MO, when comparing dFNC differences within RSNs across groups.It should be noted that the disrupted FNC in state 5 was predominantly related to the CBN, and recent studies have demonstrated that the cerebellum plays a role in a number of activities, including processing cognition and emotion, pain, movement, and thirst (Diedrichsen et al., 2019;Li et al., 2020).Some studies on Parkinson's disease (PD) have shown a decrease in the functional connectivity between the basal ganglia and the cerebellum, mainly in certain subregions of the cerebellum or in the advanced stage of Parkinson's disease (Hacker et al., 2012;Kawabata et al., 2020;Sako et al., 2019).It is still unclear what these relationships mean functionally because certain alterations may be pathogenic while others may be compensatory mechanisms (Pelzer et al., 2019;Sako et al., 2019).But when the illness worsens, the pathological alterations in the cerebellar circuitry may eventually result in a reduction in functional connectivity (Hacker et al., 2012;Simioni et al., 2016).The finding that functional connectivity in the cerebellar circuitry is inversely connected with the intensity of motor symptoms in PD patients lends credence to this compensatory theory (Kawabata et al., 2020;Simioni et al., 2016).Our findings indicated that the dFNC between CBN and DMN as well as SCN was disrupted, suggesting that the PSP may be related with the disruption of CBN.At the same time, the destruction of dFNC between CBN and other networks can provide a new understanding of the pathophysiological mechanism of PSP.Additionally, the separation of DMN and SCN from other networks is consistent with other studies (Brown et al., 2017;Whiteside et al., 2021;Whitwell et al., 2011).Likewise, we observed the interference of DMN with VN, whereas medulla oblongata exhibited enhanced connectivity with CBN and VN.According to our search of the known articles, as an important part of brainstem, medulla oblongata controls a variety of autonomic nervous functions, including the regulation of respiration, swallowing, digestion and blood pressure.We found that in a PET study of PSP, the mean [ 18 F]-THK5351PET signal increased significantly in the midbrain, bilateral globus pallidus, bilateral frontal cortex and medulla oblongata (Brendel et al., 2018).This distribution pattern is very consistent with the known tau deposition topology in postmortem histological examination (Williams et al., 2007).We speculate that there may be an abnormal activation mechanism of medulla oblongata in the process of PSP disease, which may lead to an abnormal increase in connection between medulla oblongata and other networks in dFNC, but it is not clear and we need to further explore in the future.
One possible reason for alterations in dFNC could be disruptions in the communication between different brain regions.This could be caused by changes in the strength or timing of connections between regions, leading to abnormal patterns of neural activity and information processing.These disruptions could be the result of various factors, such as changes in neurotransmitter levels, structural abnormalities in the brain, or dysfunction in specific brain networks (Hutchison et al., 2013).
Ultimately, these alterations in dFNC could contribute to cognitive and behavioral symptoms seen in neurological and psychiatric disorders.Furthermore, due to the division of IC 9,10 into DMN, IC 2,3 into SCN, IC 7 into AN, IC 4 into SMN, IC 5 into VN, the main relationship between changes in dFNC and specific clinical symptoms manifests as eye movements, motor disorders, and freezing gait primarily associated with the connectivity of DMN and SCN as well as AN, while freezing gait and eye movements are also associated with the connections of AN and SMN as well as VN.It is noteworthy that these results are primarily associated with DMN and AN.Convergent evidence from functional brain imaging indicates a high spatial overlap between functional hubs and DMN regions, suggesting a crucial role of DMN in the entire network structure (Liu et al., 2015;Tomasi and Volkow, 2011;van den Heuvel and Sporns, 2013).Additionally, DMN is one of the most important RSN as it integrates information from major functional and cognitive networks (Liao et al., 2009).Functional abnormalities in DMN may affect its communication with other networks, leading to impaired functional integration between DMN and other RSNs, resulting in abnormal clinical symptoms.AN primarily involves frontal and parietal regions and participates in top-down modulation of attention and working memory tasks (Cole and Schneider, 2007); VN and SMN are involved in sensory perception and motor processes, responsible for exchanging information with the external environment.Decision-making guided by objectives affects our perception and leads to corresponding adjustments in the sensory cortex activity, which is actually top-down control enabling us to flexibly navigate multiple streams of sensory information (Gazzaley et al., 2005a;Gazzaley et al., 2005b).Thus, the lack of functional connectivity between the AN and perceptual networks presents deficits in higher-order control over sensory processes in PSP, potentially resulting in ocular movements, motor impairments, and freezing gait.These findings, along with the altered dFNC and its relationship with clinical variables, provide crucial insights for a better understanding of PSP.In addition, the temporal features of dFNC are notably different in PSP patients.The PSP patients had lower fraction time and mean dwell time in state 2 than normal controls.Our findings showed that state 2 is characterized by a distributed strong connections across RSNs, particularly the MO and SCN.These findings consistent with those from other Fig. 5. Temporal metrics produced from state transition vector are compared between groups.Asterisk represent statistically significant differences between the two groups in bar plots.The standard error is shown by error bars.neurological (Parkinson's and Alzheimer's disease) and mental (schizophrenia) diseases that all exhibit altered mean dwell time (Damaraju et al., 2014;Fiorenzato et al., 2019;Fu et al., 2019;Kim et al., 2017;Yang et al., 2022).In another study on PSP (Whiteside et al., 2021), they used the HMM method to explore the stability of the network, found that patients with PSP spent longer in inefficient brain states, and found that PSP increased the proportion of time in the frontoparietal network, while our research methods were different.At the same time, we found that in patients with PSP, the dFNC between MO and VN/CBN was increased and PSP patients spend less fraction time and shorter dwell time in the MO and SCN.Furthermore, in our research, the fraction time and mean dwell time of state 2 in PSP patients were negatively connected with the duration of PSP, bulbar and oculomotor symptoms in PSPRS.According to earlier research (Golbe and Ohman-Strickland, 2007), bulbar symptoms are associated with dysarthria and dysphagia.One of the potential pathoanatomical causes of bulbar dysfunction may be periventricular white matter (PVWM) damage (Chankaew et al., 2016).Levine et al. (Levine et al., 1992) have proved that the increase of PVWM lesions can affect the prolongation of oropharyngeal swallowing time.Other studies have shown that impaired PVWM can lead to tongue dyskinesia, delayed oral food mass transit, and dysphagia (Cola et al., 2010;Daniels et al., 1999).The subcortical cortex is the transmission pathway of sensory and motor information related to swallowing.When the subcortical cortex is damaged, it destroys the swallowing information pathway between the cortex and the brainstem, which will affect the swallowing function of pharynx and oral phase (Toogood et al., 2017).These changes may explain what we found that the dysconnectivity of the MO and SCN may be related to the bulbar and oculomotor dysfunction.
There are some limitations to this research.First, most PSP patients were taking dopaminergic medication.Thus, the use of dopaminergic medications might have affected the functional connectivity (Berman et al., 2016;Fiorenzato et al., 2019) and potentially enhanced connection (Esposito et al., 2013;Prodoehl et al., 2014), confounding our findings and lowering the strength of the observed impact.Second, there were notable differences in gender between the two groups.Although we removed the influence of gender by regression, gender may still have an impact on our results.Third, due to the differences in potential pathogenic mechanisms and the propagation of neurodegenerative diseases, different subtypes of PSP may have varying effects on dFNC.Therefore, our study is not applicable to all types of PSP.Fourth, Neuropsychological data investigate a limited amount of domains, possibly missing more cognitive dysfunction (dysexecutive, visuospatial) to correlate to disrupted functional connectivity.Fifth, because this is a preliminary cross-sectional investigation of dFNC alterations in PSP, it is challenging to draw conclusions about a causal link between clinical characteristics in PSP and the brain's functional network of PSP status.Furthermore, because there are no longitudinal data in our investigation, it is not able to evaluate whether bulbar impairment is a significant predictor of PSP survival.In the future, longitudinal design will be used to investigate this issue more thoroughly.However, our research findings should still be interpreted cautiously due to the cross-sectional method's inherent limitations.In addition, previous studies (Brown et al., 2017;Gardner et al., 2013;Whiteside et al., 2021;Whitwell et al., 2011) have all demonstrated that PSP severely disrupts the connectivity between cortical regions and the subcortex/brainstem, but they have not assessed whether this is stronger than associations found in structural imaging.In the future, we can investigate this issue.Finally, it has been recommended that dFNC analysis should be applied to resting state that has been acquired for at least 10 min (Hindriks et al., 2016).Our resting state acquisition period was about 8 min, allowing us to get stable resting-state fMRI data (Tomasi et al., 2016).

Conclusion
Our main findings were that the altered connectivity was mostly concentrated in the CBN and MO.In addition, as compared to normal controls, PSP patients had different temporal dynamics (fraction time and mean dwell time), which were associated with bulbar and oculomotor symptoms in PSPRS.Overall, these results offer a novel perspective on the pathophysiology of PSP and suggest new directions for future studies using dFNC to investigate other types of Parkinsonism.

Fig. 1 .
Fig. 1.An overview of analysis steps of dynamic functional network connectivity.The analysis includes the following steps: (1) fMRI data were preprocessed; (2) ICA was conducted and 7 RSNs were identified; (3) dFNC matrices were calculated and constructed in all different sliding windows for each subject; (4) clustering and state analyses were conducted to investigate the dFNC changes.Abbreviations: dFNC, dynamic functional network connectivity; ICA, independent component analysis; RSN, resting-state network

Fig. 3 .
Fig. 3.The total number and the percentage of occurrences are displayed in (A), along with the six cluster medians for all participants.The group-specific centroids of the states for PSP patients and normal controls are presented in (B) and (C), respectively, along with the number of participants who had at least one window in each state.The color bar depicts the dFNC's z value.PSP = progressive supranuclear palsy; NC = normal controls.

Table 1
Demographics and clinical characteristics of the PSP patients and normal controls.
A two-sample t-test for age and a chi-square test for gender were used to get the P values.PSP = progressive supranuclear palsy; NC = normal controls; PSP-RS = PSP Richardson's syndrome; PSP-P = PSP with predominant parkinsonism; PSP-PGF = PSP with progressive gait freezing; PSP-SL = PSP with predominant speech/language disorder; MDS-UPDRS = Movement Disorder Society Unified Parkinson's Disease Rating Scale; MoCA = Montreal Cognitive Assessment; MMSE = Mini-Mental State Examination; HAMD = Hamilton Depression Scale; HAMA = Hamilton Anxiety Scale; FOGQ = Freezing of Gait Questionnaire; PDQ-39 = Parkinson's Disease Questionnaire; H-Y = Hoehn-Yahr; PSPRS = Progressive Supranuclear Palsy Rating Scale.