Abnormal characteristic static and dynamic functional network connectivity in idiopathic normal pressure hydrocephalus

Abstract Aims Idiopathic Normal pressure hydrocephalus (iNPH) is a neurodegenerative disease characterized by gait disturbance, dementia, and urinary dysfunction. The neural network mechanisms underlying this phenomenon is currently unknown. Methods To investigate the resting‐state functional connectivity (rs‐FC) abnormalities of iNPH‐related brain connectivity from static and dynamic perspectives and the correlation of these abnormalities with clinical symptoms before and 3‐month after shunt. We investigated both static and dynamic functional network connectivity (sFNC and dFNC, respectively) in 33 iNPH patients and 23 healthy controls (HCs). Results The sFNC and dFNC of networks were generally decreased in iNPH patients. The reduction in sFNC within the default mode network (DMN) and between the somatomotor network (SMN) and visual network (VN) were related to symptoms. The temporal properties of dFNC and its temporal variability in state‐4 were sensitive to the identification of iNPH and were correlated with symptoms. The temporal variability in the dorsal attention network (DAN) increased, and the average instantaneous FC was altered among networks in iNPH. These features were partially associated with clinical symptoms. Conclusion The dFNC may be a more sensitive biomarker for altered network function in iNPH, providing us with extra information on the mechanisms of iNPH.

7][8] Previous largescale network analysis studies using independent component analysis (ICA) and region of interest (ROI)-based approaches in iNPH revealed the alterations in intra-and internetwork connectivity, which might be related to the severity of symptoms and the efficacy of shunt surgery. 6,7,9However, the temporal resolution of rs-fMRI is longer than necessary to study the dynamics of brain function, and most previous rs-fMRI studies were based on the assumption of spatial and temporal stationarity in the resting-state, which may limit their results. 10Bommarito G et al. provided what is, to our knowledge, the first in-depth description of iNPH through resting-state dynamics.
Alterations in interactions between large-scale networks explained iNPH symptoms, and the normalization of these alterations after CSF-TT may contribute to prognosis prediction and differential diagnosis. 9framework for dFNC and static FNC (sFNC) based on group ICA has been proposed to compare the whole-brain FNC features between Alzheimer's disease (AD) and subcortical ischemic vascular disease (SIVD).11 DFNC focuses on connectivity between networks 13,14 and has been widely employed in identifying dFNC biomarkers for different psychiatric and neurodegenerative diseases.13,15,16 The primary aim of the current work was to investigate the resting-state functional connectivity (rs-FC) abnormalities of iNPH brain connectivity and the correlations of these abnormalities with clinical symptoms from static and dynamic perspectives.We hypothesized the presence of changes in sFNC and dFNC between iNPH patients and healthy controls (HCs).We also assumed that the altered FNC features would be associated with the clinical symptoms of iNPH and that dFNC could provide additional biomarkers for supplying sFNC.

| Participants
This study was performed between February 2021 and March  2 See Appendix S1 for details.Finally, 33 patients with diagnosed iNPH (26 male, 7 female; 74.03 ± 8.94) were included in this study.Twenty-three elderly healthy control subjects (14 male, 9 female; 75.57± 6.66) were included for comparison.The two groups had no significant demographic differences in age (p > 0.05).
Clinical symptoms assessments can be found in Appendix S1.

| MR imaging acquisition and functional data preprocess
MR imaging data were obtained with a 3-T whole-body MRI scanner (MAGNETOM Prisma, Siemens Healthcare, Erlangen, Germany).The scanning parameters of functional images and structure images can be found in Appendix S1.Imaging data were preprocessed using a standard pipeline (Appendix S1).

| Group independent component analysis
After data preprocessing, resting-state data of all participants were analyzed using spatial ICA as implemented in the Group ICA Of fMRI Toolbox (GIFT v4.0b, http://mialab.mrn.org/software/gift) to create the intrinsic connectivity networks used in our analysis.Detailed process displayed in Appendix S1.After that, we sorted 10 meaningful ICs into seven functional networks, based on the spatial correlation values between ICs and the template. 17

| Statistic functional network connectivity (sFNC) analysis
We calculated sFNC using the Pearson correlation coefficients between the time courses of ICs, then obtained an sFNC matrix with the dimension of 10 × 10 (10 is the number of ICs) and an sFNCdomain matrix with the dimension of 5 × 5 (5 is the number of largescale networks which contain all the ICs) for each participant.The coefficients were extracted when statistically significant group differences were observed in sFNC.After that, correlation analysis was used to compute the correlation between coefficients and clinical scores.

| Dynamic functional network connectivity (dFNC) analysis
The common two ways to investigate dFNC are the sliding window approach and k-means clustering.The sliding-window approach was used to explore time-varying changes of FC within 10 ICs during functional MRI scans.Resting-state time series data were segmented into a 30TR window with a Gaussian of σ = 3 TRs, 18 which resulting in 170 consecutive windows across the entire scan.As covariance estimation using shorter time series can be noisy, the regularized inverse covariance matrix was used. 19To promote sparsity in estimation, a penalty on the L1-norm (Manhattan distance) was imposed in the graphic LASSO framework with 100 repetitions. 20After computing dFNC, all the FC matrices were transformed to z-scores using Fisher's z-transformation to stabilize variance and were residualized with nuisance variables (such as age and gender) prior to further analysis.

| Functional connectivity state analysis
We applied k-means clustering methods on 170 window FC matrices for all subjects to estimate reoccurring FC patterns (states), which could be characterized by both frequency and structure based on previous methods. 10K = 4 was determined using the elbow criterion. 10We repeated the clustering algorithm 500 times to increase the chance of escaping the local minima. 10All-time windows of all subjects were clustered into four states based on the similarity with the cluster centroid, then we investigated the average instantaneous functional connectivity and temporal properties of dFNC states by computing the fractional time (i.e., the number of total windows belonging to one state) and mean dwell time (i.e., the number of consecutive windows belonging to one state before changing to the other state) in each state as well as the transition number (i.e., the number of transitioning between states and represents the reliability of each state) between different states for each subject.Group differences in mean dwell time, fractional windows, and transition number, between age-matched iNPH patients and HCs were examined using two-sample t-test [p < 0.05, false discovery rate (FDR) correction].

| Temporal variability of independent components
We investigated the between-group differences in the temporal variability of each IC.Specifically, we first measured the temporal stability of FNC in IC, that is, k, by the average Pearson correlation coefficient between the k-th row of every two correlation matrices of each participant, and then, the temporal variability V k of IC k can be described by one minus the temporal stability. 21,22 which n = 170 is the total number of windows and F j,k ,F j,k is the Pearson correlation coefficient between the FNC of IC k in the correlation matrices derived from the i-th and the j-th windows (i, j = 1, 2, …, n; i ≠ j; k = 10).

| Demographic and clinical characteristics differences between groups
The demographic and clinical characteristics of the diagnosed iNPH and HC groups are summarized in Table 1.There was no significant intergroup difference in age or frame-wise displacement (FD) distribution (p1 = 0.612; p2 = 0.439).As there was a significant difference (p = 0.003) in gender distribution, we included gender as a covariable in all statistical analyses.Poor performance was found in patients with iNPH in the Mini-Mental State Examination (MMSE), Time Up and Go Test (TUG) and 3 subitems of iNPH grading scale (iNPHGS), that is, cognition (iNPHGS_c), motion (iNPHGS_m), and urination (iNPHGS_u) (p < 0.001).

| Intrinsic connectivity networks
Spatial maps of all 10 ICs defined using the group ICA are shown in The detailed information of independent components was listed in Table S1 in Supplementary Materials.

| Group-discriminating sFNC
The sFNC between all ICs and large-scale functional networks is presented as matrices in Figures 1B and 3A

| State analysis and functional connectivity strength in the dFNC
The centroids and the count of dFNC states are displayed in Figure 4A.The most frequent connected state-2 (31%), and a least frequent interconnected state-1 (17%).The percentage of total occurrences of another two states was similar with state-2, with state-3 (24%) and state-4 (28%).Figure 4B showed the corresponding visualized connectivity patterns of these four FC states.
In state-3, connections between ICs were located only within networks with positive correlations, especially between and within DAN and VN.Similarly, the connections between ICs in state-2 and state-4 were mainly located within networks with positive couplings, especially connections between IC13 in DAN and IC14 in DMN (state-4) and within DMN (state-2).In contrast, the least and sparsely connected state (state-1) was characterized by connections between functional connectivity networks with positive and negative correlations.

| Average instantaneous functional connectivity of ICs in the iNPH and HC groups
Figure 5A showed state-and group-specific cluster centroids.
Average instantaneous FCs of ICs were calculated and compared between the iNPH and HC groups, and we observed that compared with HCs, the iNPH group had a significantly decreased average instantaneous FC in most ICs for both positive and negative correlation networks, except for IC18 vs.IC11 (iNPH: 0.518, HC: −0.237, p = 0.002) and IC5 vs.IC14 (iNPH: 0.337, HC: 0.119, p = 0.009) in state-3.Detailed information was displayed in Table S2 in Supplementary Materials.

| Temporal properties of states in the iNPH and HC groups
We observed that in the iNPH and HC participants, state  Between-state comparisons in the iNPH group revealed that iNPH participants spent significantly most time in state-4 and least time in state-1 (fraction time: 0, 0-12.06;mean dwell time: 0, 0-11.67).
The distribution of state transition times displayed no difference between iNPH and HC group (p = 0.406, Figure 5B).In a further analysis of correlations between dFNC properties, that is, fraction time, mean dwell time and transition number, and clinical characteristics in the iNPH group, we found no significant results after multiple comparison correction expect for the mean dwell time in state-4 and the TUG score (r = 0.368, p = −0.035).
The urination performance, as measured by the iNPHGS_u, also showed a significant negative correlation with regional temporal variability in IC8 (r = −0.539,p = 0.001).The abovementioned results may indicate an association between the worsening of clinical function and group-discriminating temporal variability.

| DISCUSS ION
The present study employed both sFNC and dFNC analysis to clarify rs-FC changes in iNPH; another purpose was to determine the association between these changes and clinical presentation.
A dynamic analysis framework, based on ICA, Pearson correlation coefficient, sliding window approach and k-means clustering, is a method to analyze the dynamic changes in relative FNC among networks. 13,14This is the first study in iNPH cohort to combine a dynamic analysis framework with classical sFNC to investigate changes in brain connectivity from both static and dynamic per-  Dynamic FC may reflect aspects of the capacity of neural system functional 27 and serve as another physiological biomarker of neurodegenerative disease. 28,29The highly variable dFNC depart significantly from those of sFNC, which imply a flexible diversity of functional coordination between subsystems. 30The dFNC analyses of the present study suggested four discrete connectivity configurations.The temporal properties of the iNPH and HC groups differed across dFNC states.The results showed that both iNPH and HC group had the shortest dwell times in state-1 with sparse connections.It highly indicated that FC and network integrity changes in iNPH overlap with normal aging, but the decline was more rapid in iNPH patients.Only high-strength positive connections were present in state-3, or the integration state.The reduced number of transitions to the integration state in iNPH is consistent with results obtained in other neurological disorders. 29,31This may suggest that the segregation of neural network and functional connection may be closely related to the pathogenesis of iNPH.
Importantly, iNPH patients had the longest dwell time at state-4, named the iNPH-abnormality state.The longer dwell time in the iNPH-abnormality state in iNPH suggested that it may cost more energy to process the externally derived input information and execute the output.Taking the above evidence together, we suggest that the iNPH-abnormality state may be a core working state specific to iNPH and that an increased frequency of this state would facilitate functional classification.In addition, we observed that iNPH patients transitioned more frequently than HCs did, and the distribution of transition frequencies indicated that iNPH patients were more likely to transition to the iNPH-abnormality state than HCs, reflecting a propensity for this state.The iNPH patients with longer mean dwell time in the iNPH-abnormality state presented poorer gait performance, the cluster result showed the decrease in FC between the DMN and DAN was more significant in the iNPH-abnormality state than in other states.We speculated that the increased time spent in the iNPH-abnormality state would affect the exchange of information between the DMN and DAN, and that the decline of cognition (associated with the DMN) and attention (associated with the DAN) may be the cause of motor disturbance in iNPH patients.In addition, the DAN is involved in the formation of the somatormotor system, consistent with the local processing of motor output, 32 which supports the former conjecture.
INPH patients experienced more state transitions and had higher regional temporal variability than HCs, probably because the FC matrices derived from the iNPH group tended to have higher overall diversity.The increased temporal variabilities in the SMN, DMN and DAN were related to compensation for declines in sensory, motor, and cognitive function, consistent with a previous report that the brain may compensate for cognitive decline by enhancing the variability of corresponding networks. 33The temporal variability in IC8 of the DAN was significantly correlated with clinical triad symptoms in iNPH patients, implying that the vulnerability of the DAN may be closely related to disease expression.The DAN is sensitive to the effects of aging and is considered to underlie the degradation of attention that occurs with aging. 34e accelerated degeneration of iNPH and the clinical correlation of increased temporal variability of the DAN suggested that impaired attention in iNPH may be the core cause of symptom emergence.The instability of interactions between the DAN and other networks may hinder the expression of corresponding cognitive, motor, and urinary functions.
The decrease in instantaneous FC in dFNC was consistent with the intrinsic FC in sFNC.FC reduction was demonstrated from both static and dynamic perspectives, not only among networks but also within the DMN.This may be one of the characteristic changes in iNPH, which is related to the damage to the brain networks.
Decreased FC of the SMN, DMN, and DAN was found in all four FC states, suggesting that these three networks may be the key to the whole brain affects of iNPH.Decreases in FC in the DAN could underlie the impairments in sustained attention, which is one of the most affected cognitive functions that occurs with aging. 34The regulation of activity in the DMN is also critical and closely related to the key interactive functional network that coordinates motor and cognitive functions. 35The interaction between the DMN and SMN is impaired in elderly, suggesting that DMN-SMN co-deactivation may relate to gait impairments and fall risk in aging. 36Attention, sensory and cognition dysfunction interact with each other, further aggravating the clinical symptoms of iNPH, which provides additional support for our previous view.
The increased connectivity between the VAN and VN in iNPH 2022, which approved by the Ethics Committee of HuadongHospital affiliated with Fudan University (approval number: 2017K027).A total of 80 participants, diagnosed iNPH patients (n = 47) and healthy persons of a similar age (n = 23) were included in this retrospective cross-sectional study.INPH patients were recruited at the Department of Neurosurgery of Huadong Hospital affiliated to Fudan University to undergo shunt surgery.All iNPH patients fulfilled the criteria for possible, probable, and definite iNPH based on the third edition of the guidelines for management of iNPH.
Demographic and clinical data, evaluated with the Kolmogorov-Smirnov test for normality, were reported as means and standard deviations or medians and interquartile ranges for continuous variables, or as percentages for categorical variables.All functional data (including sFNC, dFNC, mean dwell time, fractional windows, transition number and average instantaneous functional connectivity) followed normal distribution because Fisher's z-transformation had been performed on all functional data.The two-sample t-test was used to analyze the normalized data, otherwise the Mann-Whitney non-parametric tests for nonnormalized data.The paired-sample t-test was used to analyze the clinical symptoms change in the iNPH patients before and 3 months after shunting, and the change was presented as percentage (ratio of difference to preoperative score).The correlation analysis p < 0.05 was considered statistically significant.IBM SPSS software (v26.0) was used for statistical analyses.

TA B L E 1
-3 was characterized by positive internetwork connections, involving the DAN, VN and DMN.In the iNPH group, state-4 was most frequently observed (n = 25); whereas in HCs, state-4 occurred less frequently (n = 12).State-1 was the least frequently observed state in both the iNPH (n = 11) and HC groups (n = 11) (Figure 5A).Pairwise group comparisons in the fraction time of occurrences and the mean dwell time are shown in Figure 5B.The results indicated a significant group difference in fractional time and mean dwell time of state-4 (p < 0.05, Mann-Whitney U-test; median, 25%-75%), which suggests that iNPH exhibited an increased proportion of time spent only in state-4 compared to HC.Specifically, among the four states, iNPH had a significantly longer fraction time (iNPH: 21.18, 5.295-59.71;HC: 10.59, 0-30.59;p = 0.023) and mean dwell time (iNPH: 22.5, 7-42; HC: 9, 0-26; p = 0.029) in state-4 than in HC.Demographic and clinical data of iNPH patients and participants.
Correlation analyses were carried out to test whether sFNC and dFNC properties were associated with clinical characteristics.A marginally significant positive correlation was observed between the preoperative MMSE scores of iNPH patients and the median F I G U R E 1 (A) Independent components (n = 10) identified by group independent component analysis.Independent component special maps divided on seven functional networks [somatormotor network (SMN), dorsal attention network (DAN), visual network (VN), default mode network (DMN), ventral attention network (VAN)] based on their anatomical and functional properties.(B) Group averaged static FC between IC pairs was computed using the entire resting state data.The value in the correlation matrix represents the Fisher's z-transformed Pearson correlation coefficient.Each of the 10 ICs was rearranged by network group based on the seven functional networks.Color-coded legends of ICs were pasted on the left side of the matrix, which matches to the overlaid colors of the spatial maps in figure A.strength of the significantly decreased sFNC between IC12 vs. IC14 (r = 0.349, p = 0.046).The preoperative iNPHGS_c, iNPHGS_u, and TUG scores showed a significant negative correlation with sFNC between IC15 vs. IC16, respectively (r1 = −0.373,p1 = 0.032; r2 = −0.397,p2 = 0.022; r3 = −0.409,p3 = 0.018, Figure2B).After correction for multiple comparisons, there was a no significant association between group-discriminating sFNC in the large-scale network and clinical symptoms.
spectives.Our results highlight that (1) the sFNC of brain networks is generally decreased in iNPH patients, and the reduction within the DMN and between the SMN and VN is related to clinical symptoms; (2) state-4 is the iNPH-abnormality state, and the temporal properties and temporal variability in state-4 are sensitive to identify iNPH; (3) the dFNC analysis can supplement the sFNC with additional information, and investigate rs-FC from both static and dynamic perspectives can help to depict a full picture of rs-FC abnormalities in iNPH.The sFNC of brain networks in iNPH patients is generally decreased.The dysfunction that occurs in iNPH could stem from distinct changes in the information processing in the brain underlying disparate network damage.Decreased FC within the DMN significantly correlated with cognitive decline, which is consistent with the results of previous studies.6,[23][24][25]Interestingly, we found that the F I G U R E 2 Static functional network connectivity (sFNC) analysis in independent components (ICs).(A) SFNC significantly decreased in iNPH group than HC group among partially ICs after false discovery rate (FDR) correction.(B) Correlation analysis between sFNC among ICs and pre-and post-operational clinical symptoms in iNPH group.F I G U R E 3 (A) Static functional network connectivity (sFNC) analysis in large-scale networks.(B) SFNC significantly decreased in iNPH group than HC group among partially large-scale functional networks, between DAN and VN (five-pointed star), SMN and DAN (fourpointed star), and within DMN (seven-pointed star).F I G U R E 4 (A) Results of the clustering analysis per state.The total number of occurrences and percentage of total occurrences are listed above each cluster median [percentage of total occurrences for state-1, state-2, state-3, state-4: 17%, 31%, 24%, and 28%].(B) The corresponding visualized connectivity patterns of these four FC states.decrease in internal connectivity of the DMN was more significant than that in FC among other networks.The reduction in FC within the DMN showed weaker significance than it in other networks.A study of iNPH in 2020 found that changes in FC in the DMN detected by PET were associated with the course of iNPH, and the decreased FC occurred after the onset of clinical symptoms, which indicated that the disease had entered a decompensated stage.26The abnormality of connectivity intensity within the DMN can help to capture reliable biomarkers of the severity of clinical symptoms and advance the current understanding of the pathological mechanism underlying iNPH.On the other hand, a reduction in connectivity between the SMN and VN was associated with the poor cognitive and motor performance.The disconnection between SMN and VN may be the key to the appearance of symptoms in iNPH.

F I G U R E 5
Comparison of the temporal properties of the dynamic FC states between the iNPH and HC groups.(A) The cluster centroid matrices of four dFNC states, along with the count of participants that have at least one window clustered into each state [total occurrences for state-1, state-2, state-3, state-4: 11, 23, 15, and 25 in the iNPH group and 11, 16, 13, and 12 in the HC group, respectively]; group comparison in average instantaneous FC of ICs between iNPH and HC group, along with the average instantaneous FC values with significant between-group differences are annotated in the matrixes (p < 0.05).(B) Comparison of group level and state level in fraction time, mean dwell time and transition number of four dynamic functional network connectivity (dFNC) states.Asterisks indicating significant group difference.Only the mean dwell time in state-4 had significant correlation with the TUG score.

F I G U R E 6
Results of the temporal variability analysis.(A) Group comparison in temporal variability of all ICs.Asterisks indicating significant group difference.(B) Correlation analysis between all group discriminating temporal variability of ICs and preoperational clinical symptoms score of iNPH.
group may be related to compensatory mechanisms.Combined with the results of sFNC, we speculate that the VN is another major sensory domain affected by iNPH.It has been proven that visual-sensory-motor interaction is essential for motion control37 ; thus, visual dysfunction may be one of the causes of gait disorder in iNPH.This result partially supports our previous findings regarding sFNC, that is, decreased FC between the SMN and VN is correlated with poor motion performance.Dynamic FNC may be more sensitive than sFNC to changes in overall FNC, providing us with extra information on the mechanisms underlying the effects of iNPH on networks.There are a few limitations that should be considered in the interpretation of our results.First, due to the limited sample size, we did not have sufficient statistical power to investigate the association between dFNC characteristics and clinical scores separately.As more scan data are collected in the future, the exact relationship between dFNC features and symptoms can be more accurately demonstrated.Another limitation is the difference in gender distribution between patients and HCs.Although we corrected for the effect of gender by including it as a covariate of no interest in the regression model, residual effects of gender may still exist or have nonlinear effects on functional networks.Finally, future studies could use rs-fMRI data in different stages of diagnosed iNPH to examine the relationship of functional integration and separation with symptoms, which could contribute to the understanding of the mechanisms of FNC change in iNPH.