Dynamic functional networks in idiopathic normal pressure hydrocephalus: Alterations and reversibility by CSF tap test

Abstract Idiopathic Normal Pressure Hydrocephalus (iNPH)—the leading cause of reversible dementia in aging—is characterized by ventriculomegaly and gait, cognitive and urinary impairments. Despite its high prevalence estimated at 6% among the elderlies, iNPH remains underdiagnosed and undertreated due to the lack of iNPH‐specific diagnostic markers and limited understanding of pathophysiological mechanisms. INPH diagnosis is also complicated by the frequent occurrence of comorbidities, the most common one being Alzheimer's disease (AD). Here we investigate the resting‐state functional magnetic resonance imaging dynamics of 26 iNPH patients before and after a CSF tap test, and of 48 normal older adults. Alzheimer's pathology was evaluated by CSF biomarkers. We show that the interactions between the default mode, and the executive‐control, salience and attention networks are impaired in iNPH, explain gait and executive disturbances in patients, and are not driven by AD‐pathology. In particular, AD molecular biomarkers are associated with functional changes distinct from iNPH functional alterations. Finally, we demonstrate a partial normalization of brain dynamics 24 hr after a CSF tap test, indicating functional plasticity mechanisms. We conclude that functional changes involving the default mode cross‐network interactions reflect iNPH pathophysiological mechanisms and track treatment response, possibly contributing to iNPH differential diagnosis and better clinical management.

The diagnosis of iNPH is challenging because symptoms and radiological features can be confused with alternate neurological conditions; for example, Parkinson's disease or Alzheimer's disease (AD).
Diffusion-weighted magnetic resonance imaging studies highlighted periventricular, frontoparietal, and cortico-subcortical white matter (WM) microstructural alterations compatible with the mechanical and ischemic damage associated with iNPH (Kamiya et al., 2017;Siasios et al., 2016) and able to discriminate iNPH from other neurodegenerative disorders (Hattori et al., 2011;Hattori, Sato, Aoki, Yuasa, & Mizusawa, 2012;Kim et al., 2011). A subset of studies also demonstrated WM changes after shunt surgery or CSF tap test, indicating a possible short-term structural response to CSF drainage (Jurcoane et al., 2014;Kamiya et al., 2017;Kanno et al., 2017;Saito et al., 2020). However, the associations of WM features with iNPH symptoms, or with clinical improvement after treatment, are not reproducible across studies. Little is known about brain functional alterations in iNPH. As we have recently reviewed, few electroencephalography and functional magnetic resonance imaging (fMRI) studies report alterations of brain dynamics across multiple brain regions including frontal, motor, occipital, temporal, and cingular areas, but findings remain poorly consistent (Griffa et al., 2020).
A better characterization of brain functional alterations in iNPH is desirable for several reasons. First, the comprehension of the functional dynamics associated with iNPH symptoms and response to treatment, could shed new light on the pathophysiological mechanisms, and contribute to improve diagnostic guidelines and clinical management. Functional connectivity features have higher behavior predictive capacity than brain structural features (Amico & Goñi, 2018;Finn et al., 2015;Lin, Baete, Wang, & Boada, 2020) and are, therefore, likely to have better correlation with symptoms. In particular, including both spatial and temporal (besides static connectivity) features of the occurring functional reorganizations allows to achieve a better classification of neurodegenerative disorders (de Vos et al., 2018;Liégeois et al., 2019;Preti, Bolton, & Van De Ville, 2017).
Moreover, functional features may be more sensitive to short-term plasticity mechanisms occurring after CSF drainage. Lastly, the lack of specificity of symptoms and the high prevalence of comorbidities observed in iNPH suggests, on one side, the presence of dimensional symptom-level functional abnormalities (possibly shared across disorders), and, on the other side, the co-existence of multiple pathological pathways leading to possibly distinct functional alterations. The analysis of CSF molecular markers, including the core AD biomarkers amyloid-β 42, phosphorylated tau and total tau, has proved effective for the differential diagnosis of iNPH from cognitive, movement and cerebrovascular mimic disorders (Jeppsson et al., 2019;Manniche, Hejl, Hasselbalch, & Simonsen, 2019). However, it is currently unknown how comorbid pathologies, and, in particular, amyloid-and tau-pathways, contribute to the brain functional alterations observed in iNPH. In sum, functional networks in iNPH might not only contribute to a better understanding of iNPH mechanisms but also more importantly help the clinicians to predict the best responders to shunt surgery.
Here we tackle the open questions of iNPH disorder-the need for quantitative neuroimaging markers, the comprehension of brain changes following treatment, and the influence of AD pathology on its pathophysiological mechanisms-from the perspective of brain functional circuits probed with resting-state fMRI. To this end, iNPH patients are enrolled into a well-established 2-day protocol with gait, neuropsychological and MRI assessments before and after a CSF tap test (Allali et al., 2017).

| Assessment of gait and cognition
Quantitative spatiotemporal gait assessment was performed in a kinesiology laboratory. Subjects were equipped with reflective markers placed on the heels and asked to walk at their self-selected speed on a 10-m walkway. Reflective marker trajectories were recorded with a 12-camera optoelectronic system to compute average gait parameters including walking speed, stride time, stride length, and step width (Allali et al., 2013;Armand et al., 2011 (Table S2). Moreover, a double-echo gradient echo field map was acquired to estimate field inhomogeneities inside the scanner (Hutton et al., 2002) (Table S3).
During the fMRI acquisition, subjects were instructed to keep their eyes closed without focusing on any specific task and without falling asleep.
Spatial normalization and brain segmentation outcomes were visually inspected: one iNPH patient was excluded because of segmentation failure.
Note: Group differences between iNPH and HC groups were assessed with Student's t test for continuous variables and chi-square test for categorical variables. Group differences for gait and cognitive scores were assessed with ANCOVA, including age, gender, and education level as covariates. Differences between pre-and post-CSF tap test data in iNPH patients were assessed with paired Student's t test. Gait and cognitive data were missing for some subjects, who were discarded from related group-comparisons: gait assessment (n = 3 pre-/n = 2 post-CSF tap test); MMSE (n = 1); FCSRT, all scores (n = 2); Color Trail test, Part B and Index (n = 4 pre-/n = 3 post-CSF tap test); WAIS-III all scores (n = 1 pre-/n = 1 post-CSF tap test  (Esteban et al., 2019), temporally realigned to correct for head motion (Jenkinson et al., 2002), and slice-timing corrected (Cox & Hyde, 1997) using a unique composite geometric transformation. The first six volumes were discarded to allow signal stabilization and the remaining volumes were warped to the MNI space (Greve & Fischl, 2009;Jenkinson & Smith, 2001).
Single-voxel time series were corrected for nuisance contributions by regressing out the average WM and CSF signals, the six motion signals (three translations and three rotations) and low-frequency components estimated with the discrete cosine transform (Lund, Madsen, Sidaros, Luo, & Nichols, 2006;Power et al., 2014). In particular, the three nonzero lowest frequency basis vectors of the discrete cosine transform were removed, corresponding to oscillations at 0.0008, 0.0017, and 0.0025 Hz. Finally, a spatial Gaussian smoothing of 6 mm full-width-at-half-maximum and a temporal band-pass filter with 0.01-0.15 Hz band limits were applied. Analyses were repeated while including the global signal (average of fMRI time series over brain voxels) as additional regressor (Murphy & Fox, 2017). Registrations and susceptibility distortion corrections were visually inspected.

| Global signal
The fMRI global signal has been associated with both artifacts (respiratory and heartbeat oscillations, hardware fluctuations) (Chang & Glover, 2009;Fox, Zhang, Snyder, & Raichle, 2009;Murphy, Birn, Handwerker, Jones, & Bandettini, 2009) and neural activity (Schölvinck, Maier, Ye, Duyn, & Leopold, 2010). Moreover, recent studies associate the global signal with behavioral traits (Li et al., 2019b), pathological conditions (Scalabrini et al., 2020;Wang et al., 2019;Yang et al., 2014), and more generally to the anticorrelation patterns observed between the default mode and taskpositive network. The interpretation of the global signal is therefore debated (Murphy & Fox, 2017). For this reason, we adopted two strategies to assess the influence of the global signal on our data. First, we repeated all the analyses while including the global signals as an additional regressor (besides the WM, CSF, motion, and low-frequency components) in the preprocessing pipeline. Second, we investigated whether the representation of the global signal in individual subjects is modulated by iNPH pathology (the computation of the global signals representation maps is described below).

| Head motion
Head motion is a severe confounding factor of functional connectivity analyses and needs to be carefully considered, particularly when investigating elderly populations prone to high levels of motion (Gratton et al., 2020). The framewise displacement (FD) was used as motion indicator (Power, Barnes, Snyder, Schlaggar, & Petersen, 2012) and fMRI volumes with FD > 0.7 mm, together with their precedent and two following volumes, were discarded from the computation of functional measures. Moreover, 4 iNPH patients with less than 4 min of uncorrupted fMRI recoding after FD censoring were excluded. The final iNPH (N = 26) and HC (N = 48) groups, and the iNPH pre-and post-CSF tap test (N = 21) groups did not differ in terms of average FD, although a larger proportion of fMRI volumes was discarded in iNPH patients compared with controls (Table 1). All statistical analyses were replicated including the number of excluded fMRI volumes as covariate.

| Analysis of resting-state functional dynamics
To identify the brain circuits affected in iNPH and characterize their spatial and temporal characteristics during resting-state, we developed a three-step methodological approach. First, we performed a whole-brain analysis of within-and between-RSNs functional connectivity. Results identified the default mode (DMN) as the central network vulnerable to iNPH. Second, we investigated the spatial characteristics of the average co-activation map of the precuneus, the main DMN hub (Raichle, 2015), and the spatial representation of the global signal (Murphy & Fox, 2017). Third, we clustered in time the precuneus co-activations to unravel the dynamic interactions between this hub and other DMN regions, as well as other functional circuits.
The latter analysis delivers interpretable statistics that can be compared before and after the CSF tap test and related to AD pathways and clinical dimensions.

| Step 3: Temporal characterization of DMN dynamics
The fMRI volumes selected by precuneus activations were clustered into three states using the k-means algorithm. The z-scored centroids of the clusters represent distinct precuneus co-activation patterns

| Statistical analyses
Statistical differences between iNPH and control groups were assessed with analysis of covariance (ANCOVA) including age, gender, and education level as covariates. Statistical differences between preand post-CSF tap test assessments were evaluated with paired t test.
Effect sizes were quantified with the Cohen's d. Pair-wise relationships were assessed with the Pearson's correlation coefficient (r). Multiple comparison correction was applied when necessary using the Benjamini-Hochberg procedure to control the false discovery rate (FDR) at q < 0.05 (Benjamini & Hochberg, 1995;Benjamini & Yekutieli, 2001).

| Voxel-wise precuneus co-activation and global signal representation maps comparison
Voxel-wise group differences were assessed with permutation testing and threshold-free cluster enhancement using FSL randomize (Smith & Nichols, 2009;Winkler, Ridgway, Webster, Smith, & Nichols, 2014). Age, gender, and education level were included as covariates and the analyses were limited to gray matter voxels (n = 60 0 013) ( Figure S3). Multiple comparison-corrected statistical maps were thresholded at p < .05 to identify significant clusters (Winkler et al., 2014). Only clusters larger than 20 voxels were retained.  and limbic-somatomotor connections) tended to be decreased in patients, but these comparisons did not survive FDR-correction ( Figure 1). Results were replicated using an alternative cortical parcellation ( Figure S6).  Table S4.

| The precuneus has increased co-activation with dorsolateral and medial frontal cortices
3.2.3 | iNPH is characterized by abnormal dynamics measured by CAPs between the default mode, executive-control, and salience-attention networks The fMRI volumes corresponding to precuneus activations were clustered into three CAPs encoding distinct functional configurations ( Figure S7). The first CAP ("CAP1 DMN ") exhibits a DMN activation pattern with co-deactivation of the dorsal attention, ventral attention and salience networks (Figures 3a,b and S8). The second CAP ("CAP2- indicating a within-subject balancing effect of CAP1 DMN -CAP3 ECN dynamics ( Figure S12).

| CAPs' occurrences reflect gait and executive dimensions of iNPH
Four distinct PLSC analyses were run to assess the relationships between the three CAPs' occurrences and the gait, executive, memory, and attention domains in iNPH patients. These analyses revealed two significant correlations between the CAPs' occurrences, and the gait and executive domains (p = .004, p = .008, respectively; surviving FDR-correction). Figure 4 illustrates the PLSC weights for the signifi-

| CAPs' occurrences normalize after CSF tap test
FMRI volumes corresponding to precuneus activations were extracted from post-CSF tap test data and classified in the three CAP-clusters, established from the pre-CSF tap test data as outlined above, based on Pearson's correlation values ( Figure S13).

| Global signal
The group-average global signal representation map is shown in Figure 2b: regions with highest global signal representation encompass occipital and posterior-medial regions, in agreement with previous reports (Li, Bolt, et al., 2019b;Power et al., 2017). Individual global signal representation maps of patients and controls were com-

| DISCUSSION
INPH is a prevalent but poorly understood neurological disorder. In this cohort of iNPH patients with mostly poor gait and cognitive functions, which are improved after CSF tap test, we demonstrate an alteration of higher order resting-state network dynamics compared with normal aging, involving the default mode, executive-control, and salience/attention systems. We show that these functional alterations underlie gait and executive disturbances in patients and are not driven by AD pathology or white matter changes. On the contrary, CSF tau levels relate to functional alterations opposite to iNPH. Finally, we demonstrate for the first time the presence of resting-state functional MRI plasticity mechanisms 24 hr after a CSF tap test, with partial normalization of functional dynamics.
We followed a hierarchical analytical approach from whole-brain connectivity to specific network-dynamics investigation to characterize the functional circuits implicated in iNPH pathophysiology. The whole-brain analysis pointed out a major implication of the DMNconfirming previous data (Khoo et al., 2016)-, with iNPH-related hyper-connectivity between the DMN, and the ECN and SAL ( Figure 1). We then investigated the spatial and temporal connectivity features of the main DMN hub, located in the precuneus/posterior cingulate cortex (Greicius, Krasnow, Reiss, & Menon, 2003;Raichle, 2015), by assessing the precuneus average co-activation maps, and co-activation patterns (CAPs) dynamics (Liu et al., 2018). In line with the whole-brain analysis, the precuneus demonstrated  Spreng, 2014;Raichle, 2015). We also note that the effect size of the iNPH-HC CAPs' differences is larger than the effect size of the static functional connectivity differences, with 94% of patients having CAP3 ECN occurrence above the mean of the control group (d = 1.56).
These considerations highlight the relevance of considering temporal features (besides static connectivity values) of resting-state functional patterns to disentangle the specific consequences of iNPH on brain activity in the absence of task.
The DMN has previously been related to iNPH (Khoo et al., 2016;Ogata et al., 2017) as well as to AD (Greicius, Srivastava, Reiss, & Menon, 2004;Jones et al., 2016;Pievani, Filippini, van den Heuvel, Cappa, & Frisoni, 2014). A study found decreased functional connectivity within the DMN in 17 iNPH patients compared with 15 healthy controls, but other resting-state networks were not taken into account (Khoo et al., 2016). Counterintuitively, within-DMN connectivity increased with worsening of clinical symptoms in the same patients (Khoo et al., 2016). Our analyses show abnormal interactions between the precuneus and higher order cortical regions, but not between the precuneus and other regions of the DMN (Figure 2), suggesting that iNPH is mainly characterized by abnormal crossnetwork dynamics involving the DMN, rather than by an intrinsic impairment of the DMN itself. This becomes particularly relevant when contrasting the DMN, CAP1 DMN and CAP3 ECN findings in iNPH with AD pathology. While AD is a frequent comorbid disease in iNPH and share similar cognitive and behavioral changes (Malm et al., 2013), AD has been consistently identified as a "DMN disorder." First, the functional connectivity within the DMN is impaired in AD and preclinical AD (Badhwar et al., 2017;Jones et al., 2016;Pievani et al., 2014;Sheline & Raichle, 2013), and the precuneus shows decreased functional MRI activity matching PET hypometabolism (Greicius et al., 2004), consistent with the accumulation of amyloid in the DMN regions (Buckner et al., 2005). In contrast, within-DMN connectivity and precuneus activity are not reduced in our iNPH cohort (Figures 2and 3c). Second, according to the cascading network failure and the prion-like propagation models of misfolded beta amyloid and tau proteins (Jones et al., 2016;Ossenkoppele et al., 2019;Raj, Kuceyeski, & Weiner, 2012;Vogel et al., 2020;Zhou, Gennatas, Kramer, Miller, & Seeley, 2012), AD brain alterations are theorized to spread from the DMN to multimodal functional hubs of the brain network, including salience and executive regions (Crossley et al., 2014;Miši c et al., 2015). These mechanisms can result in impairments of the SAL and ECN and decreased DMN inter-network connectivity in AD (Agosta et al., 2012;Brier et al., 2012;Li et al., 2019a). On the contrary, iNPH patients demonstrate increased DMN-ECN coupling (increased occurrence of CAP3 ECN ). Taken together, these considerations suggest that AD and iNPH pathologies may relate to distinct functional connectivity alterations, with possible implications for differential diagnosis (Andersson, Rosell, Kockum, Söderström, & Laurell, 2017;Lu et al., 2020). This is supported by the fact that the observed CAPs alterations in iNPH are not referable to Aβ-42 and pTau levels in the CSF (Table 2) Figure 2d). The global signal has been related to ongoing neural activity (Schölvinck et al., 2010;Turchi et al., 2018) and to changes in baseline glucose metabolism in the brain (Thompson et al., 2016). An increased global signal representation may therefore be interpreted as an increased baseline activity in the frontal lobe of iNPH patients, compatible with (ineffective) compensatory mechanisms in response to motor and executive deficits. However, this conclusion should be taken with caution. While recent studies have associated individual global signal topographies to behavioral traits (Li, Bolt, et al., 2019b) and neuropsychiatric pathologies (Scalabrini et al., 2020;Wang et al., 2019;Yang et al., 2014), it is well understood that the global signal contains artefactual contributions (Power et al., 2017) and research is needed to elucidate its interpretation.
The resting-state fMRI alterations observed in this study are coherent with our knowledge on iNPH pathophysiological mechanisms and brain changes (Griffa et al., 2020;Keong et al., 2016;Siasios et al., 2016;Tarnaris et al., 2009 (Griffa et al., 2020), with substantial overlap with our findings ( Figure 2). It is also well understood that the interaction between the SAL, DMN, and ECN is central to the accomplishment of higher order functions and integration of internal and external stimulusdriven mental processes (Goulden et al., 2014;Jilka et al., 2014;Menon, 2011;Menon & Uddin, 2010 CSF withdrawal induces functional plasticity with a partial normalization of resting-state brain dynamics (Figure 3e). Eighty-one percentage of patients experienced a decreased occurrence of CAP3 ECN after the tap test indicating a normalization of the DMN-ECN interaction. Moreover, CAP2 VSM occurrence was increased in a similar proportion of patients, despite the absence of significant differences at baseline. A previous study showed restored somatomotor activity during a finger-tapping test in iNPH responders to CSF drainage, conceptually in line with our findings (Lenfeldt et al., 2011). Few studies investigated changes of EEG activity before and after CSF tapping or shunting, but their results are inconsistent (Aoki et al., 2013;Aoki et al., 2019;Sand, Bovim, & Gimse, 1994). In our cohort, patients improved in gait, but not in cognitive performances after CSF tap test, suggesting that resting-state functional dynamics may recover quicker than cognitive functions. The validation of this hypothesis would require a longitudinal follow-up of patients. Finally, it is interesting to note that amyloid-and tau-positivity assessed in the CSF do not seem to prevent brain functional and gait changes in response to CSF tap test ( Figure S14). This is in line with recent studies showing no differences of Aβ 1-42 CSF levels between iNPH responders and nonresponders to CSF tap test or shunting (Hamdeh et al., 2018). However, findings on AD biomarkers assessed with PET imaging are more controversial (Hiraoka et al., 2015;Jang et al., 2018).
There are limitations to the current study. First, our analyses focused on cortical circuits only, although iNPH pathophysiology is likely to also involve subcortical and cerebellar circuits. However, our choice was dictated by the intention of maximizing the robustness of results, by minimizing spatial normalization biases that can be severe in periventricular/subcortical regions. In the same line, we paid particular attention to the correction of head motion artifacts and excluded patients with mediocre-quality MRI data. Second, even though we adopted a hierarchical analytical approach, the dynamic functional connectivity analysis was limited to the DMN interactions; future work should tackle whole-brain network dynamics. Third, molecular markers were assessed in the CSF rather than in brain tissues. Although this is the first study showing a relationship between molecular markers and functional dynamics in iNPH patients, availability of PET data would allow a better understanding of the relationship between iNPH and amyloid/tau deposition topographies. The available sample size was too small to reliably assess the effect of amyloid and tau pathways on CSF tap test response. However, it should be noted that this was a representative clinical sample and not a dataset drawn from less specific databases comparing large versus small ventricles. Fourth, changes of cognitive performances in iNPH patients were assessed through replication of neuropsychological tests before and after CSF tapping and may therefore be affected by learning biases (Benedict et al., 2017;Calamia, Markon, & Tranel, 2013). The inclusion of a test-retest reliability assessment of neuropsychological tests in an iNPH sample could deliver a better picture of possible cognitive changes in response to treatment.

| CONCLUSION
This study offers an in-depth characterization of resting-state dynamics in iNPH. Alterations of cross-network interactions between the default mode, and the executive control and salience networks partially explain iNPH symptoms, tend to normalize after CSF tap test, and are distinct from AD features. These results may contribute to the development of iNPH biomarkers for differential diagnosis and to the improvement of iNPH clinical management.

ACKNOWLEDGMENTS
We would like to thank all the patients and healthy volunteers for their participation in this study. We thank Dr. Stéphane Armand for his assistance and expertise on the gait analysis, Rachel Goldstein for her help with the clinical database, Dr. Daniela Zöller and Dr. Lorenzo Pini for their critical suggestions and helpful discussions. We thank the Swiss National Science Foundation for funding this work (SNSF grant #320030_173153).

CONFLICT OF INTEREST DISCLOSURE
All the authors report no conflict of interest to disclose.

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request and in accordance