Periventricular Diffusivity Reflects APOE4-modulated Amyloid Accumulation and Cognitive Impairment in Alzheimer’s Continuum

32

Altered glymphatic-related fluid dynamics are increasingly recognized as a key 33 feature of Alzheimer’s disease (AD). We generalized an established diffusion imaging technique to 34 estimate periventricular diffusivity (PVeD) , hypothesizing that fast diffusion signal s in the 35 periventricular region can reflect amyloid-beta (Aβ) deposition across the Alzheimer’s continuum. 36

Participants from two multi -site cohorts (n = 440 and 414), comprising cognitively 37 unimpaired individuals, those with mild cognitive impairment, and patients with AD, were included. 38 We tested and validated the association of PVeD with Aβ burden and core AD characteristics. 39

Lower PVeD was extensively associated with greater A β burden, neurodegeneration, 40 cognitive impairment, and clinical severity. Importantly, the relationship between PVeD and Aβ burden 41 was significantly modulated by APOE4 status, with APOE4 carriers showing a stronger negative 42 association. Baseline PVeD also predicted longitudinal cognitive decline. 43

These findings suggest that periventricular fast diffusion signals can reflect APOE4-44 modulated Aβ burden and cognitive decline in AD. 45 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 3 Research-in-Context 46 Systematic review: A comprehensive PubMed literature search indicates that fluid movement 47 related to glymphatic activity assessed by diffusion tensor image analysis along the perivascular 48 space (DTI-ALPS) is associated with amyloid-beta deposition in Alzheimer’s disease (AD). 49 However, recent evidence underscores certain limitations of DTI-ALPS, suggesting that it may not 50 fully capture the diffusion processes involved in amyloid clearance. Moreover, no previous studies 51 have investigated the role of APOE4 in modulating the relationship between glymphatic-related fast 52 diffusion signals and amyloid-beta deposition. 53 Interpretation: The transverse diffusion process along the perivenous space in the periventricular 54 region appears to reflect glymphatic-related dysfunction manifested by amyloid-beta deposition. 55 Reduced periventricular diffusivity is associated with greater amyloid burden across the AD 56 continuum. This association is notably enhanced in APOE4 carriers, who exhibit higher amyloid 57 accumulation for a given reduction in the periventricular diffusivity. Besides, periventricular 58 diffusivity is related to other pathological markers of AD, including clinical symptom severity and 59 neurodegeneration, and may also predict subsequent cognitive decline. 60 Future directions: Although diffusion-based neuroimaging metrics hold promise as surrogate 61 imaging biomarkers for glymphatic-related activity, they do not comprehensively capture the 62 complex fluid dynamics such as convective bulk flow within the glymphatic system. By leveraging 63 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 4 multimodal neuroimaging techniques and advanced analytic approaches, future research can refine 64 these metrics into more sensitive, non-invasive tools capable of evaluating fluid dynamics related to 65 glymphatic dysfunction in AD. 66 67 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 5 1 Background 68 Dementia is characterized by progressive impairments in multiple cognitive domains driven by 69 neurodegenerative processes.1, 2 A critical focus of investigation is the glymphatic system, a brain-wide 70 network of perivascular pathways facilitating the exchange of cerebrospinal fluid (CSF) and interstitial 71 fluid (ISF) within the brain.3, 4 Glymphatic dysfunction can lead to the accumulation of toxic proteins 72 such as amyloid-beta (A β) and tau proteins, which contributes to the pathogenesis of 73 neurodegenerative diseases such as Alzheimer's disease (AD) and other dementias .2, 3, 5 The 74 apolipoprotein E ε4 (APOE4) allele, a known major genetic risk factor for AD, has been linked to 75 glymphatic dysfunction, impairing perivascular clearance and promoting Aβ deposition.6 The interplay 76 between impaired glymphatic function and Aβ accumulation may create a feedback loop where waste 77 build-up further compromises glymphatic function, leading to greater neuronal damage and cognitive 78 decline.5 Thus, elucidating the mechanisms underlying glymphatic dysfunction , particularly 79 glymphatic fluid dynamics (e.g., convection and diffusion), is critical for identifying therapeutic targets 80 that address the causes related to AD.2, 7 81 The original model of the glymphatic system proposed a convective bulk flow of CSF through 82 perivascular spaces (PVS) driven by arterial pulsations, facilitating the exchange with ISF and the 83 clearance of solutes including Aβ.2, 3, 8 -11 However, accumulating evidence suggests that fluid 84 movement in the brain parenchyma (i.e. interstitial space , ISS ) occurs via the combination of 85 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 6 convection and diffusion since bulk flow is potentially limited to the ISS.2, 4, 8, 12-14 Also, the impairment 86 of CSF-ISF exchange has been linked to the accumulation of Aβ.6, 14-17 Current imaging techniques 87 including contrast-enhanced magnetic resonance imaging ( MRI) and diffusion tensor imaging (DTI) 88 have been employed to investigate fluid dynamics relevant to CSF-ISF exchange in the brain .4, 18-22 89 While contrast-enhanced MRI visualizes fluid movement using tracers, DTI assesses the diffusion of 90 water molecules associated with ISF dynamics in perivascular spaces.18, 23, 24 91 The diffusion tensor image analysis along t he perivascular space (DTI-ALPS) index, calculated 92 as the ratio of diffusi vity along PVS to that perpendicular to main directions of white matter tracts 93 within a certain region of interest (ROI) in the periventricular area , has been hypothesized and 94 investigated as a surrogate marker of glymphatic -related integrity.25-27 However, DTI -ALPS has 95 received criticism as a glymphatic-specific marker as it does not distinguish Brownian motion of water 96 along perivascular spaces from other sources of water motion such as neighboring white matter fiber 97 tracts and their structure.28, 29 Even so, the DTI -ALPS index has shown significant correlations with 98 cognitive decline, Aβ deposition, and neurodegeneration in AD, suggesting DTI-ALPS as a potential 99 marker for studying the disease.26, 30, 31 Nevertheless, the accuracy of DTI-ALPS is highly dependent 100 on predefined small ROIs, making it susceptible to inter-rater variation and image registration errors 101 in automatic analysis, especially with severe brain atrophy, thereby insufficiently capturing effective 102 perivascular diffusion signals. Moreover, while a significant negative association between DTI-ALPS 103 and Aβ burden has been reported, the potential influence of the genetic risk factor, APOE4 allele, on 104 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 7 this relationship remains unclear. 105 Building upon the DTI -ALPS framework, to better capture diffusion signals in periventricular 106 regions relevant to ISF dynamics, we hypothesize that the transverse water diffusion signal at the scale 107 of fast diffusion process within periventricular regions may serve as an imaging correlate of the ISF 108 efflux along the perivenous conduits 2, which is associated with Aβ clearance and potentially links to 109 pathophysiological characteristics of AD (Figure 1). We proposed a method to estimate an apparent 110 transverse portion of periventricular diffusivity (PVeD) as a diffusion proxy. Our method involves 111 quantifying the voxel-wise transverse diffusivity components in regions adjacent to the lateral 112 ventricles covering main deep medullary veins. Furthermore, we hypothesize the APOE4 allele could 113 modulate the association between ISF and Aβ deposition. The objectives of this stud y are threefold: 114 (1) to develop and validate the effectiveness of PVeD metrics using clinical DTI data; (2) to associate 115 the PVeD metrics with A β burden and other clinical characteristics reflecting neuropathological 116 changes in the AD continuum; and (3) to examine the genetic modulation of APOE4 on the relationship 117 between PVeD and Aβ deposition.118 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 8 119 Figure 1. Overview of the study. A conceptual framework of fluid dynamics describes how, within the 120 glymphatic system, solutes such as metabolic waste are propelled toward the perivenous conduit, accompanied 121 by cerebrospinal fluid (CSF) and interstitial fluid (ISF) exchange from the brain parenchyma. This study builds 122 on the method proposed by Taoka et al., who introduced diffusion tensor image (DTI) analysis alon g the 123 perivascular space (DTI-ALPS) to estimate diffusion-based proxies related to glymphatic integrity given this 124 framework. Rather than targeting predefined regions of interest, we developed an automated approach to 125 delineate the periventricular region using DTI data. We then calculated the transverse tensor ratio (TTR) on a 126 voxel basis to capture the transverse component of fast diffusion signals within the periventricular region, 127 which primarily contains deep medullary veins oriented along the left -right axis. We hypothesize that such 128 periventricular diffusivity (PVeD) can reflect glymphatic -related activities such as amyloid-beta (A β) 129 deposition. To test this, we analyzed two multi-site, multi-modal datasets that included cognitively unimpaired 130 (CU) individuals, individuals with mild cognitive impairment (MCI), and patients with AD and other types of 131 dementia for exploration and replication. Our investigation focused on (1) t he association of PVeD with Aβ 132 burden, symptom severity, neurodegen eration, and cognitive function, and (2) t he modulatory effect of 133 apolipoprotein E ε4 (APOE4) on the association between PVeD and A β burden. Additionally, we tested 134 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 9 whether the baseline PVeD metrics can predict longitudinal cognitive change. Abbreviations: CDR-SB, 135 Clinical Dementia Rating Sum of Boxes ; CL, Centiloid; MD, mean diffusivity; MMSE, Mini -Mental State 136 Examination; PET, positron emission tomography; SWI, susceptibility-weighted imaging; T1w, T1-weighted 137 imaging. 138 139 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 10 2 Methods 140 2.1 Study Cohorts 141 This study utilized two large-scale, multimodal neuroimaging datasets to test our hypotheses and 142 independently replicate the findings. Specifically, we analyzed data from the BICWALZS and MCSA 143 cohorts, which collectively cover the spectrum of dementia ranging from cognitively unimpaired (CU) 144 individuals to those with clinically diagnosed dem entia (Figure 1). These datasets, derived from two 145 distinct national populations , could provide a robust framework for evaluating the link between 146 periventricular diffusion signals and Aβ deposition reflecting the glymphatic -related integrity and 147 clinical characteristics of dementia. 148 2.1.1 BICWALZS dataset for main analyses 149 This study leveraged clinical data from the Biobank Innovations for Chronic Cerebrovascular 150 Disease With Alzheimer’s Disease Study (BICWALZS) and the Centre for Convergence Research of 151 Neurological Disorders.32, 33 Initiated in 2016 by the Korea Disease Control and Prevention Agency as 152 part of the Korea Biobank Project, BICWALZS supports systematic collection and utilization of human 153 biological specimens and real-world clinical data.32, 33 Participants were recruited from memory clinics 154 and a geriatric me ntal health center with centralized protocols ensuring standardized data collection 155 and harmonization. Clinical diagnoses were established using internation ally recognized criteria to 156 identify subjective cognitive decline (SCD), mild cognitive impairment (MCI), Alzheimer ’s disease 157 (AD), subcortical vascular dementia (SVaD) , etc. Exclusion criteria included major neurological or 158 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 11 systemic conditions that could confound findings. The detailed recruitment criteria can be found in 159 Supplementary Material S1. Comprehensive a ssessments involved symptom severity, 160 neuropsychological testing, amyloid positron emission tomography (PET), APOE genotyping, and 161 brain MRI exam ination including T1-weighted imaging and diffusion MRI . A longitudinal protocol 162 included annual brief assessments such as Mini -Mental State E xamination (MMSE) for the 163 participants and biannual comprehensive evaluations for those with Aβ burden, vascular pathology, 164 APOE4 positivity, or cognitive decline. The BICWALZS study was approved by the Institutional 165 Review Board of Ajou University Hospital (AJOUIRB-SUR-2021-038) and registered in the Korean 166 National Clinical Trial Registry (KCT00 03391). Written informed consent was obtained from all 167 participants and caregivers. 168 2.1.2 MCSA dataset for replication analyses 169 To validate the findings found in the BICWALZS dataset, we also collected the data from t he 170 Mayo Clinic Study of Aging (MCSA), which is a longitudinal population-based cohort study designed 171 to investigate the prevalence, incidence, and risk factors of MCI and dementia among residents of 172 Olmsted County, Minnesota.34, 35 The MCSA aims to assess the conversion rates from MCI to dementia 173 or AD, identify risk factors for cognitive decline, and develop predictive models to aid in early 174 detection and prevention strategies. The study cohort consists of individuals mainly aged 50 and older, 175 identified through the Rochester Epidemiology Project. Participants were randomly selected using an 176 age- and sex-stratified sampling frame and underwent a structured baseline evaluation betwee n 2004 177 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 12 and 2020. The assessment included a physician -conducted neurological examination, administration 178 of the Short Test of Mental Status, neuropsychological testing, and structured interviews conducted by 179 researchers to collect demographic data, medical history, and clinical dementia ratings. Standardized 180 diagnostic criteria were applied to determine cognitive status, and vascular risk factors were 181 systematically extracted from electronic health records. In addition to clinical evaluations, a subset of 182 participants underwent multimodal neuroimaging including brain MRI examination and amyloid PET 183 of which imaging data were processed and de-identified for research use.36 The neuroimaging data is 184 publicly accessible upon request through the Image & Data Archive run by the Laboratory of Neuro 185 Imaging (https://ida.loni.usc.edu/). All study protocols were approved by the Mayo Clinic Institutional 186 Review Board, and informed consent was obtained from participants or their surrogates.34, 35 187 2.2 Neuroimaging Acquisition and Processing 188 We used DTI data from the BICWALZS and MCSA projects to quantify diffusion signals related 189 to ISF. Other neuroimaging modalities were used to evaluate both neurodegeneration via hippocampal 190 volume (structural MRI) and Aβ deposition (amyloid PET). 191 2.2.1 MRI imaging 192 Participants enrolled in the BICWALZS project completed the baseline brain MRI scans on 3T 193 MRI scanners including T1-weighted imaging (in-plane voxel size 0.5-1.0 mm2 with slice thickness 194 1.0-1.3 mm) and single-shell DTI (in-plane voxel size 0.88-1.8 mm2 with slice thickness 2.0-3.0 mm, 195 b-value 600-1,000 s/mm2). The details of imaging parameters are provided in Supplementary Table 1. 196 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 13 T1-weighted images went through image quality assurance procedure, image preprocessing, 197 segmentation, and regional sampling to estimate bilateral hippocampal volume by using CAT12 198 toolbox (https://neuro-jena.github.io/cat/).37 We used multiple anatomical atlases including the AAL3, 199 IBSR, LPBA40, MORI, and Neuromorphometrics built in the CAT12 to obtain unbiased average 200 estimates of hippocampal volume to represent a quantitative measure of neurodegeneration. The 201 hippocampal volume was further adjusted by the total intracranial volume. 202 Similarly, participants from the MCSA dataset completed the baseline brain MRI scans on 3T 203 MRI scanners including structural MRI (in-plane voxel size 1.0 mm2 with slice thickness 1.2 mm) and 204 DTI (in-plane voxel size 1.37 mm2 with slice thickness 2 .7 mm, b-value 1,000 s/mm 2). Instead of 205 estimating hippocampal volume using our analytical pipeline , we directly used the estimated 206 hippocampal volume (variable name SPM12_HPVOL) and total intracranial volume (variable name 207 SPM12_TIV) provided by the MCSA online data repository for analyses. 208 2.2.2 PET imaging 209 For the BICWALZS cohort, all participants in this study underwent 18F-flutemetamol PET 210 imaging using PET/computed tomography scanners for amyloid imaging . A standardized imaging 211 protocol was implemented across all participants (deta iled imaging parameters are provided in 212 Supplementary Table 1). 18F-flutemetamol was administered as a single bolus injection (mean dose: 213 185 MBq) into the antecubital vein. PET acquisition was initiated 90 minutes’ post-injection and 214 continued for 20 minutes, comprising four 5-minute frames. Individual PET scans were co-registered 215 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 14 to corresponding MRI scans and subsequently normalized to a standardized T1 -weighted MRI 216 template using spatial transformation parameters. 18F-flutemetamol retention was quantified using the 217 standard uptake value ratio (SUVR) by which the pons served as the reference region. Global cortical 218 18F-flutemetamol retention was calculated using the automated anatomical labeling (AAL) atlas ,38 219 deriving volume-weighted average SUVRs from multiple bilateral cortical ROIs encompassing the 220 frontal, cingulate, temporal, parietal, and occipital lobes. The Centiloid (CL) method was used to 221 convert SUVR values into a standardized, comparable scale for subsequent analyses.39, 40 222 For the MCSA cohort, amyloid PET imaging was performed using 11C-Pittsburgh compound 223 B (PiB) according to previously published protocols.41 PET images were processed by the MCSA, and 224 the quantification of Aβ burden was performed by calculating the SUVR for each participant using 225 11C-PiB retention. Similarly, w e used the global CL SUVR values (variable name 226 SPM12_PIB_CENTILOID) provided by the MCSA online data repository for analyses. 227 2.3 DTI Processing 228 The diffusion weighted images (DWIs) used in this study (both BICWALZS and MCSA datasets) 229 were processed using the DSI studio (version: Chen, March 18 2023, https://dsi-studio.labsolver.org/). 230 The DWIs were first corrected for phase distortion through FSL’s TOPUP if the images acquired with 231 the reverse phase encoding direction were available. This correction was followed by eddy current and 232 head movement corrections using FSL ’s EDDY . These FSL correction functions 233 (https://fsl.fmrib.ox.ac.uk/fsl/) were impleme nted in the integrated preprocessing in the DSI studio. 234 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 15 The corrected DWIs were further reconstructed by using q-space diffeomorphic reconstruction (QSDR) 235 method,42 which is an extension of generalized q-sampling imaging approach43 that can non-linearly 236 transform the images to the ICBM152 adult template and reconstruct the DWIs in the MNI space. To 237 calculate the diffusivity of interstitial fluid, the DWIs with a b -value of no more than 1,10 0 s/mm 2 238 were used to form the diffusion tensor model. The accuracy of b -table orientation was verified by 239 comparing the fiber orienta tions with those in a population -averaged brain template.44 The diffusion 240 sampling with a length ratio of 1.25 was used. The voxel size of the diffeomorphic reconstruction 241 output was 2 mm isotropic. The color -coded fractional anisotropy (FA), mean diffusivity (MD), and 242 axis-specific diffusivity (Dxx, Dyy, and Dzz) maps were estimated based on the tensor model. The 243 DWIs exhibiting a low signal-to-noise ratio, poor spatial correlation with neighboring samples, failure 244 in image correction, or visible artifacts were excluded, and the reconstructed images with low spatial 245 consistency relative to the template were also removed. 246 2.4 Quantification of Periventricular Diffusivity 247 2.4.1 DTI-ALPS calculation 248 To define the ROI for DTI-ALPS calculation,25, 27 we first created a group-averaged template 249 in the MNI space by using the reconstructed DWIs from the HCP -Aging project45 accessed through 250 the DSI studio data sharing portal ( https://brain.labsolver.org/hcp_a.html). We identified projection 251 and association fibers on the color-coded FA maps of the template and placed spherical ROIs with a 3 252 mm radius in these fiber areas, specifically at the level of the lateral ventricle body on both sides of 253 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 16 the brain (Supplementary Material S2). A total 4 ROIs were applied to the axis-specific diffusivity 254 maps of each subject, and the diffusivity values of Dxx, Dyy and Dzz within those ROIs were sampled 255 for the DTI-ALPS calculation. The DTI-ALPS is a ratio of the average axis-specific diffusivities in the 256 area of projection fibers (Dxx −proj) and association fibers (Dxx −assoc) on the x -axis to that of the 257 projection fibers (Dyy −proj) on the y -axis and the association fibers (Dzz −assoc) on the z -axis 258 (Supplementary Material S2). The DTI-ALPS indices (left-side, right-side, and mean values) were 259 calculated for each subject to reflect the diffusion signals related to the glymphatic system.25 260 2.4.2 Estimation of PVeD 261 Inspired by the DTI-ALPS calculation,25 we devised an automatic method to approximate 262 the transverse portion of fluid diffusivity in the periventricular area (i.e. PVeD) that is hypothesized to 263 reflect fast diffusion process in the perivenous space (Figure 2), assessing whether the derived measure 264 could reflect biosignals related to the clearance integrity of the glymphatic system .46 To capture the 265 fast diffusion signals related to the ISF dynamics along with the perivenous space around deep 266 medullary veins that are predominantly oriented in the left -right direction, we defined an image 267 contrast called transverse tensor ratio (TTR), which is a voxel -wise ratio calculated by divid ing the 268 tensor element on the x-axis (Dxx) by the Euclidean norm of tensor elements on three axes (Dxx, Dyy, 269 and Dzz), capturing the transverse portion of water diffusivity on a voxel basis compared to other 270 orthogonal directio ns. We also developed an automatic region growing algorithm to label 271 periventricular regions based on the MD maps to target the periventricular areas that contain deep 272 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 17 medullary veins. In practice, the MD maps were segmented to generate white matter (WM) m asks, 273 and a predefined lateral ventricle (LV) prior mask would be registered onto the MD maps to contour 274 individual LV morphology, and the generated LV posterior mask would be used to create initial 275 periventricular areas (PVeAs) as the sampling ROI through an optimized region growing algorithm. 276 The algorithm implemented an anatomically -guided dilation of the lateral ventricle (LV) posterior 277 mask along its transverse axis, with differential expansion coefficients that correspond to the known 278 lateral ventricular morphology. Maximum dilation occurs at the ventricular body, while minimal 279 expansion is applied at the peripheral ventricular margins. The hyperparameters of the region growing 280 algorithm were empirically optimized through a multimodal image analysis, which incorporated 281 susceptibility-weighted images, T1-weighted images, and DWIs to achieve optimal delineation of the 282 periventricular region with specific emphasis on deep medullary vein localization ( Supplementary 283

S3). The initial PVeA mask was further overlapped with the WM mask to ensure that the ROI 284 only contains the WM parenchyma, and their intersect ion was the final PVeA mask, which was used 285 to sample the TTR values. Here, we calculated the median value of the TTR within the final PVeA 286 mask to represent the average transverse portion of interstitial fluid diffusivity in the PVeA, namely 287 periventricular diffusivity (PVeD). Processing was performed using the SPM12 and CAT12 packages 288 37, 47 in MATLAB (R2023a) (MathWorks, Natick, MA, USA). The s cripts and codes of PVeD 289 estimation are online available through our GitHub repository 290 (https://github.com/ChangleChen/EstPVeD). 291 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 18 292 Figure 2. Flowchart illustrating the proposed method for estimating the transverse portion of periventricular diffusivity (PVeD). Diffusion-weighted 293 images (DWIs) were preprocessed including phase distortion, eddy current, and motion corrections (a). The c orrected DWIs were reconstructed using q -space 294 diffeomorphic reconstruction (QSDR) to map images into the MNI space and estimate diffusion tensors (b). Meanwhile, mean diffusivity (MD) and axis-specific 295 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 19 diffusivity maps (Dxx, Dyy, and Dzz) were calculated (c). By using axis-specific diffusivity maps, a novel transverse tensor ratio (TTR) contrast was defined to 296 reflect transverse water diffusivity potentially along the perivascular space in deep medullary veins (c). Periventricular area (PVeA) was identifie d using an 297 automated region growing algorithm applied to MD maps, incorporating lateral ventricle priors and anatomically guided dilation (d). The final PVeA mas k was 298 intersected with white matter masks, and average TTR values within the PVeA mask were calculated to approximate the apparent PVeD (e), which is used as an 299 overall metric to represent interstitial fluid properties that could be related to glymphatic function. 300 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 20 2.5 Neuropsychological, Laboratory, and Clinical Assessments 301 2.5.1 Neuropsychological measurement 302 For the BICWALZS cohort, c ognitive assessment was conducted using a comprehensive 303 battery of standardized instruments. General cognitive status was evaluated using the MMSE .48 304 Dementia severity was assessed using the Clinical Dementia Rating Sum of Boxes (CDR-SB) and the 305 Global CDR score .49 Domain-specific cognitive functions were evaluated using the Seoul 306 Neuropsychological Screening Battery ,50 which comprises validated neuropsychological measures 307 across multiple cognitive domains. Language function was examined using the Korean version of the 308 Boston Naming Test (K -BNT). Visuospatial abilities and memory were assessed using the Rey 309 Complex Figure Test and Recognition Trial (RCFT). Verbal memory was measured using the Seoul 310 Verbal Learning Test Elderly's version (SVLT-E). Executive function, specifically inhibitory control, 311 was evaluated using the Korean version of the Color Word Stroop Test Color Reading (K-CWST CR). 312 All cognitive test scores were standardized to z -scores using demographic-matched healthy controls 313 as the reference. A subset of the participants had the secon d time-point measures of MMSE, which 314 would be used to correlate with the imaging-derived measures. 315 Participants in the MCSA were evaluated by a comprehensive neuropsychological battery 316 comprising 9 standardized tests to assess 4 cognitive domains 34: (1) memory, evaluated using the 317 Auditory Verbal Learning Test Delayed Recall Trial as well as the Logical Memory II and Visual 318 Reproduction II subtests from the Wechsler Memory Scale-Revised (WMS-R); (2) language, assessed 319 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 21 with the Boston Naming Test and the Category Fluency Test; (3) attention/executive function, 320 measured using the Trail Making Test Part B and the Digit Symbol subtest of the Wec hsler Adult 321 Intelligence Scale -Revised (WAIS-R); and (4) visuospatial abilities, assessed through the Picture 322 Completion and Block Design subtests of the WAIS -R. Raw scores for each test were standardized 323 into z-scores, which were then averaged within each domain to compute domain-specific cognitive z-324 scores (i.e. memory, language, attention/executive function, and visuospatial abilities). A global 325 cognitive z-score, reflecting overall cognitive performance, was derived by averaging the 4 domain-326 specific z-scores. Similarly, general cognitive status was evaluated using the MMSE , and dementia 327 severity was assessed using the CDR -SB and Global CDR score s. We used the domain-specific 328 cognitive scores, MMSE, Global CDR, and CDR -SB (variable name pzmemory, pzlangu age, 329 pzattention, pzvisualsp, MMSEcalc, CDRGLOB, and CDRSUM ) provided by the MCSA online data 330 repository for analyses. The participants also underwent longitudinal assessment of MMSE, which was 331 utilized to examine correlations with imaging-derived measures. 332 2.5.2 APOE genotyping 333 All study participants in the BICWALZS provided written informed consent for blood 334 collection and genomic DNA analysis. Genomic DNA was extracted from peripheral blood samples , 335 and single-nucleotide polymorphism (SNP) genotyping was conducted using the Affymetrix Axiom 336 KORV1.0-96 Array (Thermo Fisher Scientific, Waltham, MA, USA). All genotyping procedures were 337 performed by DNA Link, Inc. (Seoul, Korea) in accordance with the manufacturer's standardized 338 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 22 protocols. APOE genotypes were determined through the analysis of two SNPs: rs429358 and rs7412, 339 both of which were incorporated in the genotyping array. The participants with APOE4 allele were 340 labeled as APOE4 carriers. 341 In the MCSA dataset, blood samples from the participants collected in clinic were used to 342 determine APOE genotype. We used the binary status of APOE4 allele positivity (variable name 343 Any_E4) provided by the MCSA online data repository for analyses. 344 2.5.3 Clinical assessment 345 T1-weighted images acquired in the BIC WALZS were evaluated by trained clinicians to 346 assess medial temporal lobe atrophy (MTA) scores. To minimize intra- and inter-rater variability, the 347 assessments were conducted by a team of two psychiatrists and six neurologists who underwent 348 standardized training in the visual rating procedure .51 Coronal T1-weighted images were utilized for 349 the evaluation with left and right MTA scored independently. The severity of MTA was graded on a 350 standardized scale ranging from 0 (no atrophy) to 4 (severe atrophy) ,52 which was used to represent 351 medial temporal lobe degeneration. 352 2.6 Statistical Analysis 353 Since the neuroimaging data were acquired a t multiple sites, to mitigate the scanner -related 354 variability,53 the ComBat harmonization procedure was performed to harmonize the imaging-derived 355 measures (e.g. DTI-ALPS, PVeD, PET CL SUVR, and hippocampal volume) across scanners, while 356 preserving the information of biological covariates used in the statistical analyses for each dataset.54, 357 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 23 55 358 All statistical analyses were performed by using MATLAB R2023a (The MathWorks Inc., Natick, 359 MA, USA) and R software (version 4.4.0). Images that passed quality assurance procedures were used 360 to generate imaging-derived data, and tabular data were screened for outliers; values exceeding three 361 standard deviations (SDs) from the mean were excluded from analysis. Data analyzed in this study are 362 reported in tables as mean (±SD) or number (%) unless otherwise specified. The chi-square test (for 363 categorical variables) , two -sample t test (for continuous variables with two groups), and one -way 364 analysis of variance ( ANOV A, for continuous variables with three group) were used to test group 365 difference for demographic characteristics. Group comparisons of neuropsychological and imaging 366 characteristics were exa mined by using analysis of covariance (ANCOV A) controlling for age, sex, 367 and education. 368 To explore the association of PVeD with amyloid -beta deposition and other core AD 369 characteristics, we conducted an exploratory partial Spearman correlation analysis in the BICWALZS 370 dataset while adjusting age, sex, and education. The core AD characteristics covered three domains 371 including symptom severity (CDR-SB and Global CDR), cognitive outcomes (MMSE, K-BNT, RCFT, 372 SVLT-E, and K-CWST CR), and neurodegeneration (average hippocampal volume and MTA scores). 373 In addition, we also performed the same analysis on DTI-ALPS metrics and conventional DTI metrics 374 including FA and MD for comparison. Given that previous studies have suggested that MD may act as 375 a potential confoundi ng factor in the interpretation of DTI -ALPS,29 we conducted an additional 376 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 24 correlation analysis while adjusting for MD values extracted from the ROIs used in DTI-ALPS and 377 PVeD calculations. The significance threshol d was adjusted for multiple comparisons using the 378 Bonferroni method, accounting for the combination of image feature types (i.e. DTI-ALPS, PVeD, 379 MD, and FA ) and the number of domains (i.e. symptom severity, cognitive outcomes, 380 neurodegeneration, and Aβ burden) to mitigate Type I error inflation in the exploratory analysis. A post 381 hoc analysis was conducted to examine bilateral alterations in DTI -ALPS and PVeD measures. The 382 partial correlation analysis was also applied to the MCSA replication cohort. 383 If a significant association between PVeD and Aβ burden was observed, a mediation analysis was 384 conducted to assess whether Aβ deposition mediated the relationship between PVeD and cognitive 385 change. The mediation model was structured as a three-variable regression framework, with PVeD as 386 the independen t variable, cognitive performance as the dependent variable, and amyloid -beta 387 deposition as the mediator. Covariates including age, sex, and education were controlled in the analysis. 388 Unstandardized regression coefficients were reported for each path, and the proportion of mediation 389 (PM) was calculated to quantify the relative contribution of the indirect effect to the total effect. 390 Multiple comparison correction was applied using the Benjamini-Hochberg method to account for the 391 number of image measures. All analyses were performed using the lavaan package (version 0.6.18) in 392 R. 393 To examine the effect of APOE4 status on the association between PVeD and Aβ deposition, we 394 employed robust local weighted regression modeling in which amyloid CL SUVR values served as the 395 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 25 dependent variable while PVeD, APOE4 positivity, and their interaction were included as independent 396 variables. Additional covariates including age, sex, and education were also included. The Benjamini-397 Hochberg method was applied for multiple comparison correction given the number of image measures. 398 Furthermore, the interaction analysis was validated in the MCSA cohort to assess the genetic 399 modulation of APOE4 on the PVeD-Aβ relationship across the Alzheimer’s disease continuum. 400 To assess whether PVeD metrics at baseline could predict longitudinal cognitive change, we 401 collected a subset of participants who had the second time -point measurement of MMSE from the 402 BICWALZS study and calculated the annual change rate of MMSE by subtracting the baseline MMSE 403 score from the second time-point measurement and then dividing by the time interval in years. Linear 404 regression analysis was conducted in which the MMSE annual change rate was the dependent variable 405 while the independent variables included baseline PVeD and other covariates involving baseline age, 406 sex, and education. The Benjamini-Hochberg method was applied to address multiple comparisons 407 given the number of image measures . The analysis was also performed for DTI -ALPS metrics. To 408 validate the findings, we selected participants from the MCSA who had longitudinal MMSE 409 assessments. Since the MCSA primarily includes cognitively unimpaired individuals, we identified a 410 subset who exhibited cognitive decline at the second time point to replicate the experiment. 411 412 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 26 3 Results 413 3.1 Clinical Characteristics of Study Cohorts 414 Table 1 summarizes the clinical characteristics of the study cohort from the BICWALZS project. 415 Among 440 participants, 302 (68.6%) were female, and 124 (28.2%) were APOE4 carriers. The mean 416 (±SD) of age and education were 73.1 (±6.47) and 7.78 (±4.79) years, respectively. The majority of 417 participants were diagnosed with mild cognitive impairment (n = 275, 62.5%). The dementia group 418 contains the participants diagnosed with Alzheimer’s disease (n = 81, 18.4%), vascular dementia (n = 419 28, 6.4%), frontotemporal lobe dementia (n = 4, 0.9%), and other degenerative conditions (n = 7, 1.6%). 420 The symptom severity, neuropsychologica l measures, and imaging -derived phenotypes are also 421 displayed in Table 1. Within this cohort, there were 77 participants who had longitudinal measurement 422 of MMSE and their demographic information is shown as follows; age at baseline: 72.2 (±5.8) years, 423 56 females (72.7%), and education: 7.6 (±5.1) years. The MMSE scores at baseline and at the second 424 time point were 24.1 (±4.4) and 23.8 (±5.7), respectively. The time interval was 2.0 (±0.4) years, and 425 the average annual change rate was -0.064 (±2.15). 426 For replication, following the data quality a ssurance process, a total of 414 participants were 427 selected from the MCSA study in which 351 cognitively unimpaired individuals, 57 individuals with 428 MCI, and 6 patients diagnosed with dementia were included. Due to the limited number of participants 429 in the dementia group, individuals with MCI or dementia were combined into one group for analysis. 430 Among these participants, 189 (45.7%) were female, and 106 (25.6%) were APOE4 carriers. The mean 431 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 27 (±SD) of age and education were 75.3 (±8.50) and 14.4 (±2.77) years, respectively. The details of 432 demographics, symptom severity, neuropsychological measures, and imaging-derived phenotypes are 433 provided in Supplementary Table 2. 434 435 436 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 28 Table 1 Characteristics of the study cohort from BICWALZS 437 Demographic characteristics CU MCI Dementia Comparis ons post hoc: CU vs. MCI post hoc: CU vs. Dementia post hoc: MCI vs. Dementia N 45 275 120 - - - - Age, year 70.8 (±6.2) 73.0 (±6.2) 74.3 (±7.0) p = 0.008 p = 0.091 p = 0.006 p = 0.243 Sex, female, N (%) 33 (73.3%) 192 (69.8%) 77 (64.2%) p = 0.416 - - - Education, year 7.6 (±4.4) 7.7 (±4.8) 8.1 (±4.8) p = 0.728 - - - MMSE 27.0 (±2.5) 24.3 (±3.9) 19.5 (±5.0) p < 0.001 p < 0.001 p < 0.001 p < 0.001 Global CDR 0.42 (±0.18) 0.52 (±0.14) 0.88 (±0.44) p < 0.001 p = 0.071 p < 0.001 p < 0.001 CDR-SB 0.88 (±0.65) 2.00 (±1.36) 5.15 (±2.59) p < 0.001 p < 0.001 p < 0.001 p < 0.001 APOE4 +, N (%) 3 (6.7%) 61 (22.2%) 60 (50.0%) p < 0.001 p = 0.016 p < 0.001 p < 0.001 Neuropsychological characteristics K BNT z-score 0.470 (±0.64) -0.487 (±1.34) -1.577 (±1.77) p < 0.001 p < 0.001 p < 0.001 p < 0.001 RCFT delayed recall z-score 0.404 (±1.00) -0.556 (±0.98) -1.505 (±0.88) p < 0.001 p < 0.001 p < 0.001 p < 0.001 RCFT recognition z- score -0.062 (±1.29) -0.511 (±1.02) -1.790 (±1.42) p < 0.001 p = 0.006 p < 0.001 p < 0.001 SVLT-E delayed recall z-score 0.215 (±0.95) -0.882 (±1.14) -1.893 (±0.82) p < 0.001 p < 0.001 p < 0.001 p < 0.001 SVLT-E recognition z- score 0.341 (±0.95) -0.764 (±1.31) -1.957 (±1.38) p < 0.001 p < 0.001 p < 0.001 p < 0.001 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 29 K CWST color reading z-score 0.194 (±0.91) -0.689 (±1.31) -1.909 (±1.49) p < 0.001 p < 0.001 p < 0.001 p < 0.001 Imaging characteristics MTA scale L (Scheltens’s scale) 1.200 (±0.79) 1.691 (±0.75) 2.367 (±0.82) p < 0.001 p < 0.001 p < 0.001 p < 0.001 MTA scale R (Scheltens’s scale) 1.044 (±0.77) 1.564 (±0.74) 2.283 (±0.87) p < 0.001 p < 0.001 p < 0.001 p < 0.001 Adjusted average left hippocampal volume 3.073 (±0.33) 2.815 (±0.39) 2.484 (±0.41) p < 0.001 p < 0.001 p < 0.001 p < 0.001 Adjusted average right hippocampal volume 3.183 (±0.33) 2.951 (±0.38) 2.553 (±0.43) p < 0.001 p = 0.002 p < 0.001 p < 0.001 Amyloid PET CL-SUVR- Cingulum 17.486 (± 29.33) 46.587 (± 57.71) 88.473 (± 73.39) p < 0.001 p = 0.005 p < 0.001 p < 0.001 Amyloid PET CL-SUVR- Frontal -1.466 (± 22.83) 20.540 (± 43.09) 57.710 (± 61.29) p < 0.001 p = 0.004 p < 0.001 p < 0.001 Amyloid PET CL-SUVR- Occipital 6.398 (± 16.35) 22.777 (± 33.50) 51.772 (± 47.03) p < 0.001 p = 0.006 p < 0.001 p < 0.001 Amyloid PET CL-SUVR- Parietal 7.628 (± 21.18) 28.087 (± 39.25) 65.485 (± 55.62) p < 0.001 p = 0.003 p < 0.001 p < 0.001 Amyloid PET CL-SUVR- Temporal 7.889 (± 16.56) 22.519 (± 30.43) 50.939 (± 44.02) p < 0.001 p = 0.008 p < 0.001 p < 0.001 Amyloid PET CL-SUVR- 18.655 (± 18.79) 36.919 (± 35.51) 69.912 (± 50.62) p < 0.001 p = 0.004 p < 0.001 p < 0.001 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 30 Global DTI-ALPS 1.598 (± 0.16) 1.523 (± 0.15) 1.452 (± 0.16) p < 0.001 p = 0.003 p < 0.001 p < 0.001 PVeD 0.459 (± 0.04) 0.454 (± 0.04) 0.438 (± 0.04) p = 0.003 p = 0.480 p = 0.002 p = 0.002 1. Comparisons were adjusted for age, sex, and education in the neuropsychological and imaging measures 438 2. Hippocampal volume was adjusted by total intracranial volume 439 Abbreviations: APOE, apolipoprotein E; CDR, Clinical Dementia Rating; CDR-SB, Clinical Dementia 440 Rating Sum of Boxes; CL, Centiloid; CU, cognitively unimpaired; DTI-ALPS, diffusion tensor image 441 analysis along the perivascular space; K-BNT, Korean version of the Boston Naming Test; K-CWST CR, 442 Korean version of the Color Word Stroop Test Color Reading; L, left; MCI, mild cognitive impairment; 443 MMSE, Mini-Mental State Examination; MTA, medial temporal lobe atrophy; PET, positron emission 444 tomography; PVeD, periventricular diffusivity; R, right; RCFT, Rey Complex Figure Test and Recognition 445 Trial; SUVR, standard uptake value ratio; SVLT-E, Seoul Verbal Learning Test Elderly's version. 446 447 448 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 31 3.2 Association of PVeD with Aβ Deposition, Neurodegeneration, Cognitive Outcomes , and 449 Symptom Severity in the BICWALZS Cohort 450 The partial correlation analysis revealed that PVeD metrics were primarily correlated with 451 symptom severity, cognitive performance, neurodegeneration, and Aβ burden after multiple 452 comparison correction (Figure 3a & Supplementary Table 3). Lower PVeD values were associated with 453 higher Global CDR (P = 2.9×10-4), CDR-SB (P = 3.8×10-5), MTA scores (P = 1.7×10-30 for the left 454 side; P = 3.4×10-23 for the right side ), and amyloid CL SUVRs (P = 8.1×10-4 for the global region) 455 whereas higher PVeD values were associated with better cognitive performance (P = 3.5 ×10-4 for 456 MMSE) across multiple domains and reduced hippocampal volume (P = 3.3×10-17 for the left side; P 457 = 3.7×10-18 for the right side). These findings suggest that the interpretation of PVeD aligns with that 458 of DTI-ALPS where higher values may indicate more efficient glymphatic-related clearance. However, 459 PVeD demonstrated overall stronger correlations with cog nitive decline, neurodegeneration, and 460 particularly Aβ deposition (Figure 3a). Notably, the PVeD metric in the left hemisphere exhibited a 461 preferential association with Aβ deposition compared to the right hemisphere (Figure 3b). On the other 462 hand, the DTI-ALPS metrics demonstrated a stronger association with symptom severity and cognitive 463 outcomes in the left -sided region compared to the right (Figure 3b) . Additionally, we examined 464 conventional DTI markers including FA and MD to show the association of white matter integrity with 465 clinical factors (Figure 3a). Given prior reports that DTI -ALPS may be influenced by white matter 466 diffusivity, we repeated the analysis with MD included as an additional covariate. The results indicated 467 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 32 that the correlation patterns between PVeD and c linical factors remained robust; however, the 468 associations between DTI -ALPS and cl inical measures were diminished particularly for cognitive 469 outcomes (Supplementary Material S4). 470 471 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 33 472 Figure 3. Partial correlation matrix of diffusion imaging-derived metrics with multifaceted clinical characteristics. Diffusion imaging-derived metrics refer 473 to the diffusion tensor imaging (DTI)-derived measures including DTI-ALPS, periventricular diffusivity (PVeD), fractional anisotropy (FA), and mean diffusivity 474 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 34 (MD). The multifaceted clinical characteristics include four domains; symptom severity, cognitive outcomes, neurodegeneration, and Aβ burden. The matrix was 475 generated using the BICWALZS dataset (n = 440). Panels (a) and (b) present the correlation matrix of overall diffusion imaging-derived metrics and that of 476 bilateral metrics for DTI-ALPS and PVeD, respectively. Values are Spearman’s correlation coefficients for each pairwise correlation. Covariates including age, 477 sex, and education were adjusted . Hippocampal volume was additionally adjusted by the total intracranial volume. Darker blue and red colors indicate greater 478 negative and positive correlations, respectively. Gray dots indicate those did not pass the significance threshold adjusted by multiple comparisons (adjusted alpha 479 threshold = 0.0031 ). Abbreviations: adj, adjusted; avg, average; CDR, Clinical Dementia Rating ; CDR-SB, Clinical Dementia Rating Sum of Boxes ; Cing, 480 cingulum; CL, Centiloid; CU, cognitively unimpaired; DTI, diffusion tensor image; DTI-ALPS, diffusion tensor image analysis along the perivascular space ; 481 Front, frontal; Hippo, hippocampal; K-BNT, Korean version of the Boston Naming Test; K-CWST, Korean version of the Color Word Stroop Test; L, left; MCI, 482 mild cognitive impairment; MMSE, Mini-Mental State Examination; MTA, medial temporal lobe atrophy; Occi, occipital; Parie, parietal; PET, positron emission 483 tomography; PVeD, periventricular diffusivity; R, right; RCFT, Rey Complex Figure Test and Recognition Trial; SUVR, standard uptake value ratio; SVLT-E, 484 Seoul Verbal Learning Test Elderly's version; Temp, temporal; V ol, volume; WM, white matter; z, z-score. 485 486 487 488 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 35 3.3 Mediation of Amyloid Accumulation between PVeD and Cognition in the BICWALZS Cohort 489 Given the observed significant associations of PVeD with cognitive outcomes and Aβ deposition, 490 we further investigated whether amyloid accumulation could potentially serve as a mediator in the 491 relationship between PVeD and cognitive decline. The mediation analysis in the BICWALZS cohort 492 demonstrated that both PVeD and DTI -ALPS significantly and directly explained cognitive changes 493 represented by MMSE and CDR -SB (Figure 4). However, global Aβ burden represented by global 494 PET CL SUVR partially mediated the relationship only between PVeD and cognitive decline (PFDR = 495 0.028 for MMSE, Figure 4a; PFDR = 0.028 for CDR-SB, Figure 4b) whereas no significant mediation 496 effect was observed in the relationship between DTI -ALPS and cognitive decline ( PFDR = 0.196 for 497 MMSE, Figure 4c; PFDR = 0.196 for CDR-SB, Figure 4d). These findings indicate that lower PVeD is 498 associated with lower MMSE scores and higher CDR -SB scores where this relationship is partially 499 mediated by greater global Aβ deposition. While DTI -ALPS exhibited a similar pattern of cascade 500 correlations, only the direct effect remained significant, which was likely due to the lack of a significant 501 association between DTI-ALPS and global PET CL SUVR (PFDR = 0.182). 502 503 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 36 504 Figure 4. Aβ burden mediates the relationship between periventricular diffusivity (PVeD) and cognitive decline in the BICWALZS cohort. Aβ burden is 505 quantified using the global amyloid PET standard uptake value ratio (SUVR), and cognitive decline is represented by Mini-Mental State Examination (MMSE) 506 scores and Clinical Dementia Rating Sum of Boxes (CDR -SB) scores. Panels (a), (b), (c), and (d) illustrate the conditions of the mediation analysis for the 507 following variable pairs: PVeD vs. MMSE, PVeD vs. CDR-SB, DTI-ALPS vs. MMSE, and DTI -ALPS vs. CDR-SB, respectively. Unstandardized coefficients 508 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 37 are reported, and the proportion of mediation (PM) represents the indirect effect relative to the total effect. Multiple comp arisons are controlled using the 509 Benjamini-Hochberg correction accounting for the number of imaging features tested. Abbreviations: CDR-SB, Clinical Dementia Rating Sum of Boxes ; CL, 510 Centiloid; DTI-ALPS, diffusion tensor image analysis along the perivascular space ; FDR, false discovery rate; MMSE, Mini-Mental State Examination; PET, 511 positron emission tomography; PM, proportion of mediation; PVeD, periventricular diffusivity; SUVR, standard uptake value ratio. 512 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 38 3.4 APOE4 Allele Moderates the A ssociation between PVeD and Aβ Deposition in the 513 BICWALZS Cohort 514 We further examined whether APOE4 status moderates the relationship between PVeD and Aβ 515 deposition. We observed significant interaction effects between PVeD and APOE4 status on amyloid 516 PET CL SUVRs across all ROIs including the cingulum ( PFDR = 5.7×10-4, Figure 5a ), frontal lobe 517 (PFDR = 2.6×10-4, Figure 5b), parietal lobe (PFDR = 1.3×10-4, Figure 5c), occipital lobe (PFDR = 1.3×10-518 4, Figure 5d), temporal lobe (PFDR = 1.3×10-4, Figure 5e), and global region ( PFDR = 1.3×10-4, Figure 519 5f). This result indicates that individuals carrying the APOE4 allele exhibit a stronger negative 520 correlation between PVeD and Aβ burden compared to non-carriers (Figure 5 & Supplementary Table 521 4). In other words, for each unit decrease in PVeD, individuals carrying the APOE4 allele exhibit an 522 additional 365.5-unit increase in global PET CL SUVR , denoting a stronger association be tween 523 reduced diffusivity and higher Aβ burden in APOE4 carriers. 524 Given that both PVeD and DTI -ALPS reflect diffusion signals associated with glymphatic 525 clearance efficiency, we repeated the analysis using DTI -ALPS. Although no significant correlation 526 was observed between DTI-ALPS and Aβ deposition in the BICWALZS cohort, significant interaction 527 effects between DTI-ALPS and APOE4 status on amyloid PET CL SUVR s were still present 528 (Supplementary Material S 5 & Supplementary Table 5). Notably, the interaction e ffects were more 529 pronounced for PVeD compared to DTI -ALPS, suggesting that PVeD may be more sensitive in 530 capturing the APOE4 modulation on the relationship between diffusion process and Aβ burden. 531 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 39 532 Figure 5. The presence of APOE4 allele enhances the association between periventricular diffusivity (PVeD) and Aβ deposition in the BICWALZS cohort. 533 The amyloid PET CL SUVRs were sampled from five anatomical regions including the cingulum (a), frontal lobe (b), parietal lobe (c), occipital lobe (d), and 534 temporal lobe (e), with the inclusion of the global measure (f). The models included age, sex, and education as covariates. Multiple comparisons were controlled 535 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 40 using the Benjamini-Hochberg correction accounting for the number of image measures analyzed. Green and red colors represent APOE4 non-carriers and APOE4 536 carriers, respectively. The interaction terms between PVeD and APOE4 are reported. Abbreviations: APOE, apolipoprotein E; CL, Centiloid; FDR, false discovery 537 rate; PET, positron emission tomography; PVeD, periventricular diffusivity; SUVR, standard uptake value ratio. 538 539 540 541 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 41 3.5 PVeD Predicts MMSE Change in the BICWALZS Cohort 542 We analyzed a subset of the BICWALZS cohort which had a second time -point MMSE 543 measurement to assess whether baseline PVeD could pr edict the MMSE annual change rate. We 544 observed that baseline PVeD metrics significantly predicted the MMSE annual change rate after 545 adjusting for covariates ( PFDR = 7.9×10-3 for left PVeD, Figure 6 a; PFDR = 7.9×10-3 for right PVeD, 546 Figure 6b; PFDR = 7.9×10-3 for mean PVeD, Figure 6c) (Supplementary Table 6). Lower baseline PVeD 547 values were associated with greater declines in MMSE, suggesting that PVeD may serve as a predictive 548 biomarker for cognitive decline. We also repeated the analysis using DTI-ALPS metrics. Unlike PVeD, 549 baseline DTI-ALPS metrics were not significantly associated with the MMSE annual change rate in 550 this cohort (PFDR = 5.9×10-1 for left DTI-ALPS, Figure 6d; PFDR = 2.1×10-1 for right DTI-ALPS, Figure 551 6e; PFDR = 2.6×10-1 for mean DTI-ALPS, Figure 6f). However, the trend in regression slopes suggested 552 a similar interpretation where lower DTI-ALPS values appeared to correspond with greater cognitive 553 decline, albeit without statistical significance (Figure 6). 554 555 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 42 556 Figure 6. Higher baseline periventricular diffusivity (PVeD) is associated with better longitudinal cognitive outcomes in the BICWALZS cohort. The 557 annual change rate of MMSE is used as the dependent variable with the baseline PVeDs including left PVeD (a), right PVeD (b), and mean PVeD (c) as the 558 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 43 primary independent variable. Covariates include baseline age, sex, and education. The Benjamini-Hochberg correction was applied to account for multiple 559 comparisons given the number of imaging measures analyzed. For reference, the analysis was also performed using DTI-ALPS metrics including left DTI-ALPS 560 (d), right DTI-ALPS (e), and mean DTI-ALPS (f). The coefficients of slope are reported. Abbreviations: DTI-ALPS, diffusion tensor image analysis along the 561 perivascular space; FDR, false discovery rate; MMSE, Mini-Mental State Examination. 562 563 564 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 44 3.6 Group Comparisons of PVeD and DTI-ALPS in the BICWALZS Cohort 565 We also conducted group comparisons of PVeD using ANCOV A while adjusting for age, sex, and 566 education in the BICWALZS cohort ( Supplementary Material S 6 & Supplementary Table 7 ). The 567

demonstrated that all PVeD metrics including mean, left, and right PVeD exhibited significant 568 differences among the groups; mean PVeD: F(2,434) = 8.14, PFDR = 3.8×10-4, left PVeD: F(2,434) = 8.10, 569 PFDR = 3.8×10-4, and right PVeD: F(2,434) = 8.03, PFDR = 3.8×10-4. The post hoc analysis further showed 570 that the significant differences existed between CU and dementia (mean PVeD: CU = 0.459 (±0.037), 571 dementia = 0.438 (±0.039), PBonferroni = 5.4×10-3; left PVeD: CU = 0.453 (±0.041), dementia = 0.430 572 (±0.042), PBonferroni = 5.6×10-3; right PVeD: CU = 0.465 (±0.035), dementia = 0.445 (±0.040), PBonferroni 573 = 5.9×10-3) and between MCI and dementia (mean PVeD: MCI = 0.454 (±0.037), PBonferroni = 7.0×10-574 4; left PVeD: MCI = 0.446 (±0.040), PBonferroni = 8.0×10-4; right PVeD: MCI = 0.461 (±0.038), PBonferroni 575 = 7.0×10-4). There was no significant difference between CU and MCI (mean PVeD: PBonferroni = 1.00; 576 left PVeD: PBonferroni = 1.00; right PVeD: PBonferroni = 1.00). Similarly, all DTI-ALPS metrics exhibited 577 significant differences among groups; mean PVeD: F(2,434) = 16.12, PFDR < 1.0×10-6, left PVeD: F(2,434) 578 = 15.62, PFDR < 1.0×10-6, and right PVeD: F(2,434) = 8.65, PFDR = 3.8×10-4. The post hoc analysis showed 579 that the significant differences existed between each pair of groups ; (1) between CU and dementia: 580 mean DTI-ALPS: CU = 1.598 (±0.162), dementia = 1.452 (±0.163), PBonferroni < 1.0×10-4; left DTI -581 ALPS: CU = 1.695 (±0.197), dementia = 1.519 (±0.198), PBonferroni = 1.0×10-4; right DTI-ALPS: CU = 582 1.520 (±0.189), dementia = 1.400 (±0.171), PBonferroni = 1.1×10-3, (2) between MCI and dementia: mean 583 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 45 DTI-ALPS: MCI = 1.523 (±0.149), PBonferroni < 1.0×10-4; left DTI -ALPS: MCI = 1.606 (±0.186), 584 PBonferroni = 1.0×10-4; right DTI-ALPS: MCI = 1.457 (±0.168), PBonferroni = 6.4×10-3, and (3) between 585 CU and MCI: mean PVeD: PBonferroni = 8.4×10-3; left PVeD: P = 1.1×10-2; right PVeD: PBonferroni = 8.7×586 10-2. 587 588 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 46 3.7 Replication in the MCSA Preclinical Cohort 589 To validate the findings id entified in the BICWALZS cohort, we utilized data from the MCSA 590 cohort to replicate the analyses. Partial correlation analysis successfully replicated the association 591 between PVeD metrics and cognitive outcomes, confirming that lower PVeD corresponds to po orer 592 cognitive performance (Figure 7a & Supplementary Table 8). However, the analysis did not identify 593 significant correlations between PVeD and symptom severity, neurodegeneration, or Aβ deposition 594 (Figure 7a). This discrepancy may be attributed to the co mposition of the MCSA cohort, which 595 primarily consists of cognitively unimpaired individuals, potentially limiting the ability of PVeD to 596 capture associations with AD signatures. A similar pattern was o bserved in the DTI -ALPS metrics 597 whereas DTI-ALPS demon strated a significant correlation with Global CDR scores (Figure 7a & 598 Supplementary Table 8). When controlling for MD values, PVeD metrics retained their association 599 with cognitive outcomes ; however, the associations of DTI -ALPS metrics with both cognitive 600 outcomes and symptom severity were compromised (Supplementary Figure 7a in Supplementary 601

S7). Due to the absence of a significant correlation between PVeD and Aβ deposition in the 602 MCSA cohort, mediation analysis was not conducted. 603 Additionally, we validated whether APOE4 status moderates the relationship between PVeD and 604 Aβ deposition in the MCSA cohort. We observed a significant interaction effect between PVeD and 605 APOE4 status on amyloid PET CL SUVR (P = 6.5×10-4), replicating a genetic influence effect of 606 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 47 APOE4 on the relationship between PVeD and Aβ deposition (Figure 7b & Supplementary Table 8). 607 The interaction term between DTI-ALPS and APOE4 status also significantly explained amyloid PET 608 CL SUVR (P = 1.9×10-3, Supplementary Figure 7b in Supplementary Material S7), further supporting 609 the APOE4 status ’ influence on the association between Aβ deposition and diffusion-based 610 glymphatic-related metrics. 611 To validate whether baseline PVeD could predict longitudinal cognitive decline, we selected 612 individuals from the MCSA cohort who exhibited a decline in MMSE over time (n = 152, Figure 7c). 613 Regression analysis confirmed that the baseline mean PVeD metric significantly predicted the MMSE 614 annual change rate after adjusting for covariates (PFDR = 7.0×10-3) (Figure 7d & Supplementary Table 615 8), reinforcing the potential utility of PVeD as a predictive biomarker for cognitive deterioration. In 616 contrast, the baseline mean DTI-ALPS metric was not significantly associated with the MMSE annual 617 change rate in th is cohort ( PFDR = 3.8 × 10-1) (Figure 7 e). Cross-sectional g roup comparisons 618 demonstrated that both PVeD and DTI -ALPS significantly differentiated CU individuals from those 619 with cognitive impairment (including MCI and dementia) while adjusting for age, sex, and education 620 (Supplementary Figure 8 in Supplementary Material S7); PVeD: CU = 0.474 (±0.035), MCI/dementia 621 = 0.460 (±0.039), F(1,408) = 7.38, PFDR = 6.9×10-3; DTI-ALPS: CU = 1.474 (±0.176), MCI/dementia = 622 1.392 (±0.152), F(1,408) = 11.38, PFDR = 1.6×10-3. 623 624 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 48 625 Figure 7. Replication of findings identified in the BICWALZS cohort using the MCSA cohort. The results including partial correlation analysis (a), interaction 626 analysis (b), and regression analysis for predicting longitudinal cognitive decline were d emonstrated (c-e). In the correlation matrix (a), values are Spearman’s 627 correlation coefficients for each pairwise correlation. Covariates including age, sex, and education were adjusted. Hippocampal volume was additionally adjusted 628 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 49 by the total intracranial volume. Darker blue and red colors indicate greater negative and positive correlations, respectively. Gray dots indicate those did not pass 629 the significance threshold adjusted by multiple comparisons. For the interaction analysis (b), the amyloid PET global CL SUVR was dependent variable, with the 630 primary independent variables including PVeD, APOE4 status, and their interaction. The model also included age, sex, and education as covariates. The 631 demographic information is displayed for those with longitudinal cognitive decline in the MCSA cohort (c). In the regression analysis for this sample, the annual 632 change rate of MMSE s erves as the dependent variable while baseline PVeD and DTI-ALPS are included as the primary independent variables with covariates 633 including baseline age, sex, and education (d & e). The coefficients of slope are reported. Abbreviations: adj, adjusted; APOE, apolipoprotein E; CDR, Clinical 634 Dementia Rating; CDR-SB, Clinical Dementia Rating Sum of Boxes ; CL, Centiloid; DTI-ALPS, diffusion tensor image analysis along the perivascular space ; 635 Hippo, hippocampal; L, left; MMSE, Mini-Mental State Examination; PiB, Pittsburgh Compound B; PVeD, periventricular diffusivity; R, right; SUVR, standard 636 uptake value ratio; TP, time-point; V ol, volume; z, z-score. 637 638 639 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 50 4 Discussion 640 In this study, we investigated the utility of periventricular diffusivity (PVeD) as an imaging proxy 641 to assess the relationship between diffusion signals that reflect fast diffusion process es within 642 perivenous spaces in the periventricular region and Aβ deposition in the brain parenchyma. Leveraging 643 data from two large -scale, multi-site cohorts, we demonstrated that PVeD is significantly associated 644 with core AD characteristics in which lower PVeD values are correlated with higher Aβ deposition, 645 reduced cognitive performance, and greater neurodegeneration across the AD continuum, suggesting 646 that this imaging-derived phenotype could potentially serve as an imaging marker for AD. Furthermore, 647 our study provides the first evidence of a replicable genetic modulation effect of the APOE4 allele on 648 the relationship between PVeD and Aβ deposition. We observed that APOE4 carriers showed a stronger 649 negative correlation between PVeD and Aβ burden, suggesting an interaction between genetic risk and 650 glymphatic-related dysfunction in AD. We also found that baseline PVeD is a replicable predictor of 651 longitudinal cognitive decline. Additionally, we developed an automated approach for measuring 652 PVeD and have made it openly accessible to the scientif ic community, aiming to facilitate broader 653 adoption and reproducibility in future research. Together, our findings advance the idea that 654 periventricular ISF diffusion impairment may be associated with the progression of AD pathology and 655 disease. 656 The observed stronger negative correlation between Aβ deposition and PVeD in APOE4 carriers 657 compared to the weak or absent correlation in non -carriers suggests a critical role of APOE4 in 658 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 51 modulating Aβ clearance dysfunction. Although several studies have demon strated a correlation 659 between Aβ deposition and the glymphatic -related marker, DTI-ALPS,30, 31, 56 they have primarily 660 treated APOE4 as a covariate rather than investigating its potential modulatory role. As a result, none 661 have explicitly explored how genetic variation may influence the relationship between glymphatic -662 related fluid movement and amyloid accumulation in humans . Our findings address this gap by 663 providing novel evidence of modulation effect of APOE4, suggestin g that its presence alters the 664 association between PVeD and Aβ burden. That this finding is consistent across cohorts with MCI and 665 cognitively unimpaired older adults underscores the genetic influence on glymphatic-related diffusion 666 processes and amyloid clearance. The APOE4 allele is the strongest known genetic risk factor for AD, 667 promoting Aβ accumulation and impeding its clearance through multiple mechanisms .6, 57 APOE4 668 accelerates early seeding of amyloid pathology by enhancing Aβ deposition and neuritic dystrophy.58 669 It also impairs Aβ clearance by reducing the size and function of meningeal lymphatic vessels .6, 59 670 Moreover, r ecent literature emphasizes the role of aquaporin -4 (AQP4) water channels in the 671 glymphatic system, which faci litates the exchange of CSF and ISF in the brain .60 APOE4 has been 672 linked to disrupted AQP4 polarization, potentially impairing perivascular clearance pathways and 673 promoting amyloid retention .61, 62 Finally, APOE4 impairs Aβ clearance more than other isoforms, 674 resulting in higher soluble A β concentrations in the brain ISF.63 Thus, the accumulated evidence 675 suggests that the APOE4 allele has strong influence on glymphatic dysfunction in AD, which c ould 676 exacerbate Aβ accumulation in the interstitial space. 677 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 52 As already discussed, the glymphatic system clears Aβ deposition through convective bulk flow 678 and diffusion.7, 15, 23 The CSF enters periarterial spaces, driven by arteria l pulsations, and mixes with 679 ISF via AQP4 channels, which further facilitate the waste transportation including A β toward 680 perivenous spaces.2, 15, 64 While bulk flow dominates perivascular spaces, diffusion also contributes to 681 solute movement from high - to low -concentration areas .2, 7, 20, 23 This combination enhances A β 682 removal, with fluid and solutes draining into meningeal lymphatics and other outflow routes. The 683 relative contributions of convection and diffusion vary based on brain region, solute size, and other 684 factors.2, 7, 15 Our results concur with prior studies30, 31 that the fast diffusion process observed through 685 PVeD is associated with the Aβ deposition, and we further identified that this association could be 686 moderated by the APOE4 allele status. Although the actual mechanism between impaired diffusion 687 and Aβ deposition remains uncertain, our results support that altered diffusion may contribute to Aβ 688 accumulation. This could potentially be due to structural changes in the vasculature, reduced 689 permeability, or an increased concentration of metabolites within the perivenous space,65-67 leading to 690 slowed or obstructed Aβ transport. Nevertheless, Aβ deposition itself may also obstruct pathways to 691 the perivenous space, further diminishing diffusion efficiency .7, 65 However, we did not observe a 692 significant association between PVeD and amyloid burden in the MCSA cohort, which mainly 693 comprised cognitively normal preclinical participants, compared to the BICWALZS clinical cohort 694 largely including MCI patients. This discrepancy may reflect that diffusion-related clearance becomes 695 more pronounced with accumulating amyloid pathology and neurodegeneration.30 In preclinical stages, 696 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 53 compensatory clearance mechanisms may still preserve so that PVeD alterations may emerge only with 697 advancing disease burden and clinical symptoms .68 Further investigations are necessary to elucidate 698 the fluid dynamics underlying these processes that govern Aβ clearance. 699 Recent studies have explored the potential of DTI-ALPS as a non-invasive imaging surrogate for 700 assessing glymphatic-related dysfunction.4, 18, 25 Lower DTI-ALPS indices have been associated with 701 increased Aβ deposition.30, 31, 56 However, the ROI seeding and calculation of DTI -ALPS are highly 702 susceptible to various factors such as the confounding effect of deep white matter structure in the ROI, 703 inter-rater variability, and image registration performance when automated methods are employed. 704 These factors can introduce inconsistencies and increase sensitivity to potential confounders, 705 potentially impacting the interpretation of the results.20, 24, 29 This is possibly reflected in our findings 706 where the DTI-ALPS metrics demonstrated greater efficacy in distinguishing group differences , but 707 were less likely to exhibit a specific relationship with Aβ deposition. Hence, w e proposed a new 708

to extend the DTI -ALPS framework by automatically localizing the periventricular region 709 encompassing the transverse perivascular space surrounding deep medullary veins. We also 710 generalized the index calculation (i.e. transverse tensor ratio, TTR) to approximate the transverse 711 component of fluid movement primarily along the perivenous space. Our results demonstrate that the 712 proposed metrics effectively capture core AD signatures involving Aβ burden, neurodegeneration, and 713 cognitive decline. Our findings in the mediation analysis also support this inference. We observed that 714 the association between PVeD and amyloid burden appears to be primarily driven by the left 715 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 54 hemisphere in the clin ical cohort. Although glymphatic transport is generally presumed to be 716 bilaterally symmetrical, subtle hemispheric variations in vascular and white matter geometry may 717

in asymmetric perivascular fluid dynamics. Further research is necessary to eluci date the 718 mechanisms underlying the hemispheric asymmetry. 719 Cross-sectional cognitive performance and subsequent cognitive decline were significantly 720 associated with PVeD metrics in two independent cohorts, which underlines its potential as an imaging 721 marker of cognitive trajectory . In line with previous literature of which DTI -ALPS can predict 722 cognitive changes,30, 31 PVeD values not only relate to current cognitive status but also predict future 723 changes from baseline, signifying the sensitivity of fast diffusion signals in the periventricular region 724 to shifts in cognitive function. This suggests that the investigation the biological underpinnings of 725 altered periventricular diffusion processes could facilitate targeted interventions aimed at preserv ing 726 cognitive health, particularly in vulnerable or at-risk populations.1, 69, 70 727 This study, while introducing a novel approach to measuring diffusion signals via the PVeD metric, 728 has several limitations. F irst, although the PVeD metric holds promise as a non -invasive proxy for 729 glymphatic-related integrity, it remains an indirect index, offering only an overall view of Brownian 730 motion of water within white matter in the periventricular area . Furthermore, the spatial resolution 731 constraints of clinical DTI data pose challenges for capturing the nuanced fluid flow in perivascular 732 spaces. Because the PVeD metric relies on quantifying transverse diffusivity in the periventricular 733 region, it may not sufficiently represent the complex, anisotropic nature of fluid transport within the 734 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 55 perivascular space. Partial volume effects stemming from low spatial resolution can also introduce 735 inaccuracies in PVeD estimates. Additionally, this method is limited to regions adjacent to the lateral 736 ventricles and thus cannot capture the full spectrum of glymphatic functionality across the entire brain. 737 Considering the regional heterogeneity of AD neuropathology, future studies should extend 738 investigations to other AD-relevant regions and integrate additional disease -specific biomarkers to 739 elucidate regional glymphatic integrity. Second, the PVeD metric rests on the assumption that water 740 diffusion within the perivascular space is Gaussian, which may not accurately characterize true 741 diffusion in this compartment. Similarly, the assumption of isotropic, fixed diffusivity for the free -742 water component may overlook important features of fluid dynamics in the narrow, tubular 743 perivascular spaces. While we have illustrated how fast diffusion signals might reflect amyloid 744 accumulation, the roles of othe r fluid mechanisms such as bulk flow in amyloid clearance warrant 745 further exploration. Third, our longitudinal assessment was restricted to MMSE measurements from a 746 subset of participants, limiting our ability to infer changes in amyloid clearance over time. Future 747 research should adopt more extensive longitudinal protocols, incorporating other modes of 748 neuroimaging, blood -based assays, and neuropsychological a ssessments to capture the evolving 749 clinical progression and neuropathological processes. Fourth, although this study examined the 750 influence of the APOE4 allele on the interaction between the diffusivity and Aβ deposition, it did not 751 investigate other APOE variants, notably the potentially protective APOE2 allele. Such investigations 752 may offer a more comprehensive understanding of the genetic factors influencing glymphatic function 753 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 56 and amyloid pathology. 754 Altogether, by leveraging multi -national, multi -site, an d multi -modal neuroimaging data 755 spanning the Alzheimer’s spectrum, t his study presents a novel ap proach to capture fast diffusion 756 signal along the perivenous space in the periventricular region. We demonstrate critical evidence of its 757 association with Aβ burden, neurodegeneration, and cognitive decline in the Alzheimer’s continuum. 758 Our findings suggest that lower periventricular diffusivity is linked to increased A β deposition 759 particularly in APOE4 carriers and could serve as a n imaging predictor of longitudinal cognitive 760 decline. These results indicate that PVeD may offer a non-invasive and clinically compatible tool for 761 tracking Aβ clearance dysfunction in AD and evaluating therapeutic interventions. Moreover, the 762 observed modulation by APOE4 status underscores an important connection between glymphatic -763 related diffusion processes and Aβ accumulation, warranting further investigation into individualized 764 risk stratification. Future research should explore the mechanistic underpinnings of periventricular 765 fluid transport and validate PVeD across diverse po pulations and imaging platforms. Additionally, 766 expanding the application of PVeD to other neurodegenerative conditions may further elucidate its role 767 in broader pathophysiological contexts. Integrating such diffusion imaging-derived metrics with 768 multimodal neuroimagi ng and fluid biomarkers could enhance early detection and intervention 769 strategies, ultimately contributing to a more comprehensive understanding of neurodegenerative 770 disease mechanisms. 771 772 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 57 5 Acknowledgements 773 Funding: This work was supported by the Nation al Institutes of Health/Na tional Institute on 774 Aging grant; the Normal Aging study (R01 2RF1AG025516). the SDAR study (R01AG067018), the 775 PPG4 study (P01 AG025204), the SERA study (R01 AG085566), and the Harmonization study (R01 776 AG063752). 777 Resources: This s tudy was conducted with biospecimens and data from the consortium of the 778 Biobank Innovations for Chronic cerebrovascular disease With ALZheimer's disease Study 779 (BICWALZS), which was supported by the National Institute of Health research project, Republic o f 780 Korea (Project No. 2024 -ER0505-00). The biospecimens and data used for this study were provided 781 by the Biobank of Ajou University Hospital, a member of Korea Biobank Network. This work was 782 supported by the National Research Foundation of Korea (NRF), fun ded by the Ministry of Science 783 and ICT (RS -2019-NR040055). This research was supported by a grant of the Korea Dementia 784 Research Project through the Korea Dementia Research Center (KDRC), funded by the Ministry of 785 Health & Welfare and Ministry of Science a nd ICT, Republic of Korea (RS -2024-00339665). This 786 research was supported by a grant of the Korea Health Technology R&D Project through the Korea 787 Health Industry Development Institute (KHIDI), funded by the Ministry of Health & Welfare, Republic 788 of Korea (HR21C1003, HR22C1734 and RS-2024-00406876). 789 The Mayo Clinic Study of Aging was supported by the NIH (U01 AG006786, P30 AG062677, 790 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 58 R37 AG011378, R01 AG041851, R01 NS097495), the Alexander Family Alzheimer's Disease 791 Research Professorship of the Mayo Clinic, the Mayo Foundation for Medical Education and Research, 792 the Liston Award, the GHR Foundation, the Schuler Foundation, and used the resources of the REP 793 medical records linkage system, which is supported by the National Institute on Aging (NIA: AG 794 058738), by the Mayo Clinic Research Committee, and by fees paid annually by REP users. Dr. 795 Devanand receives funding support from the National Institute on Aging (NIA) and the Alzheimer's 796 Association; serves as a scientific adviser for Acadia, GSK, Corium, Eisai; and serves on the data 797 safety and monitoring board for BioXcel. Dr. Motter reports funding from NIA. Dr. Lee reports funding 798 from NIA. Dr. Vassilaki has served as a consultant for F. Hoffmann -La Roche Ltd; she currently 799 receives research funding from NIH a nd has equity ownership in Amgen, Johnson and Johnson, 800 Medtronic, and Merck. Dr. Luchsinger receives research funding from NIH and receives a stipend 801 from Wolters Kluwer as editor in chief of a journal. Dr. Knopman has received research support from 802 the NIH and the Robert H. and Clarice Smith and Abigail Van Buren Alzheimer's Disease Research 803 Program of the Mayo Foundation; has served on a data safety and monitoring board for Lundbeck 804 Pharmaceuticals and for the Dominantly Inherited Alzheimer Network (DIAN) study; and has been an 805 investigator for clinical trials sponsored by Biogen, TauRx Pharmaceuticals, Lilly pharmaceuticals, 806 and the Alzheimer's Disease Treatment and Research Institute, University of Southern California. 807 808 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 59 6 Conflicts of Interest 809 The autho rs declare that they have no financial/non -financial and direct/potential conflict of 810 interest. 811 7 Author’s Contribution 812 (1) Research Project: A. Conception, B. Organization, C. Execution; (2) Statistical Analysis: A. 813 Design, B. Execution, C. Review and Cri tique; (3) Manuscript Preparation: A. Writing of the First 814 Draft, B. Review and Critique. 815 C.L. Chen: 1A, 1B, 1C, 2A, 2B, 2C, 3A; S.J. Son: 1B, 1C, 2A, 2C, 3B; N. Schweitzer: 2C, 3B; H. Jin: 816 2C, 3B; J Li: 2C, 3B; L. Wang: 2C, 3B; S. Yang: 2C, 3B; C.H. Hong: 1C, 3B; H.W. Roh: 1C, 3B; B. 817 Park: 1C, 3B; J.W. Choi: 1C, 3B; Y .S. An: 1C, 3B; S.W. Seo: 1C, 3B; Y .H. Cho: 1C, 3B; S. Hong: 1C, 818 3B; Y .J. Nam: 1C, 3B; D.S. Minhas: 2C, 3B; C.M. Laymon: 2C, 3B; G.D. Stetten: 2C, 3B; D.L. 819 Tudorascu: 2C, 3B; MCSA: 1C; H.J. Aizenstein: 1A, 1B, 1C, 2C, 3B; M. Wu: 1A, 1B, 1C, 2C, 3B. 820 8 Research data for this article 821 The BICWALZS datasets generated during and/or analyzed in the current study are available from 822 the corresponding author with the approval of the BICWALZS team upon re asonable request. The 823 MCSA dataset is open access and can be applied through their website 824 (https://www.mayo.edu/research/centers-programs/alzheimers-disease-research-center/research-825 activities/mayo-clinic-study-aging/overview). Scripts of analytical methods and c odes of imaging 826 process are online available through our GitHub repository 827 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 60 (https://github.com/ChangleChen/EstPVeD). 828 9 Ethical Approval 829 All procedures performed in this study involving human participants from the BICWALZS 830 project were in accordance with the ethical standards of the Institutional Review Boards of Ajou 831 University Hospital (AJOUIRB-SUR-2021-038) and with the 1964 Helsinki declaration and its later 832 amendments or comparable ethical standards. Informed consent in the study was obtained fr om all 833 individual participants who were recruited in the BICWALZS project. BICWALZS was registered in 834 the Korean National Clinical Trial Registry (KCT0003391||Registration Date: 2018/07/04|| 835 http://cris.nih.go.kr/cris/en/use_guide/cris_introduce.jsp). Similarly, all human participants provided 836 informed consent for the MCSA study. 837 838 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 61 10 References 839 1. Livingston G, Huntley J, Liu KY , et al. Dementia prevention, intervention, and care: 2024 report 840 of the Lancet standing Commission. The Lancet. 2024;404(10452):572-628. 841 2. Bah TM, Siler DA, Ibrahim AH, Cetas JS, Alkayed NJ. Fluid dynamics in aging-related dementias. 842 Neurobiol Dis. Feb 2023;177:105986. doi:10.1016/j.nbd.2022.105986 843 3. Rasmussen MK, Mestre H, Nedergaard M. The glymphatic pathway in neurological disorders. 844 Lancet Neurol. Nov 2018;17(11):1016-1024. doi:10.1016/S1474-4422(18)30318-1 845 4. Taoka T, Naganawa S. Glymphatic imaging using MRI. J Magn Reson Imaging . Jan 846 2020;51(1):11-24. doi:10.1002/jmri.26892 847 5. Carlstrom LP, Eltanahy A, Perry A, et al. A clinical primer for the glymphatic system. Brain. Apr 848 29 2022;145(3):843-857. doi:10.1093/brain/awab428 849 6. Mentis AA, Dardiotis E, Chrousos GP. Apolipoprotein E4 and meningeal lymphatics in Alzheimer 850 disease: a conceptual framework. Mol Psychiatry . Apr 2021;26(4):1075 -1097. 851 doi:10.1038/s41380-020-0731-7 852 7. Thomas JH. Fluid dynamics of cerebrospinal fluid flow in perivascular spaces. J R Soc Interface. 853 Oct 31 2019;16(159):20190572. doi:10.1098/rsif.2019.0572 854 8. Barisano G, Lynch KM, Sibilia F, et al. Imaging perivascular spac e structure and function using 855 brain MRI. Neuroimage. Aug 15 2022;257:119329. doi:10.1016/j.neuroimage.2022.119329 856 9. Iliff JJ, Wang M, Liao Y , et al. A paravascular pathway facilitates CSF flow through the brain 857 parenchyma and the clearance of interstitia l solutes, including amyloid β. Science translational 858 medicine. 2012;4(147):147ra111-147ra111. 859 10. Iliff J, Simon M. CrossTalk proposal: The glymphatic system supports convective exchange of 860 cerebrospinal fluid and brain interstitial fluid that is mediate d by perivascular aquaporin -4. J 861 Physiol. Sep 2019;597(17):4417-4419. doi:10.1113/JP277635 862 11. Daversin-Catty C, Vinje V , Mardal K-A, Rognes ME. The mechanisms behind perivascular fluid 863 flow. Plos one. 2020;15(12):e0244442. 864 12. Hladky SB, Barrand MA. The glymphatic hypothesis: the theory and the evidence. Fluids Barriers 865 CNS. Feb 3 2022;19(1):9. doi:10.1186/s12987-021-00282-z 866 13. Bohr T, Hjorth PG, Holst SC, et al. The glymphatic system: Current understanding and modeling. 867 iScience. Sep 16 2022;25(9):104987. doi:10.1016/j.isci.2022.104987 868 14. Mestre H, Mori Y , Nedergaard M. The Brain's Glymphatic System: Current Controversies. Trends 869 Neurosci. Jul 2020;43(7):458-466. doi:10.1016/j.tins.2020.04.003 870 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 62 15. Rasmussen MK, Mestre H, Nedergaard M. Fluid transport in the brain. Physiol Rev . Apr 1 871 2022;102(2):1025-1151. doi:10.1152/physrev.00031.2020 872 16. Lee MK, Cho SJ, Bae YJ, Kim JM. MRI-Based Demonstration of the Normal Glymphatic System 873 in a Human Population: A Systematic Review. Front Neurol . 2022;13:827398. 874 doi:10.3389/fneur.2022.827398 875 17. Yu L, Hu X, Li H, Zhao Y . Perivascular Spaces, Glymphatic System and MR. Front Neurol. 876 2022;13:844938. doi:10.3389/fneur.2022.844938 877 18. Lee DS, Suh M, Sarker A, Choi Y . Brain Glymphatic/Lymphatic Imaging by MRI and PET. Nucl 878 Med Mol Imaging. Oct 2020;54(5):207-223. doi:10.1007/s13139-020-00665-4 879 19. Klostranec JM, Vucevic D, Bhatia KD, et al. Current Concepts in Intracranial Interstitial Fluid 880 Transport and the Glymphatic System: Part II -Imaging Techniques and Clinical Applications. 881 Radiology. Dec 2021;301(3):516-532. doi:10.1148/radiol.2021204088 882 20. Wright AM, Wu YC, Feng L, Wen Q. Diffusion magnetic resonance imaging of cerebrospinal fluid 883 dynamics: Current techniques and future advancements. NMR Biomed. Sep 2024;37(9):e5162. 884 doi:10.1002/nbm.5162 885 21. Naganawa S, Taoka T. The Glymphatic System: A Review of the Challenges in Visualizing its 886 Structure and Function with MR Imaging. Magn Reson Med Sci . Mar 1 2022;21(1):182 -194. 887 doi:10.2463/mrms.rev.2020-0122 888 22. Kamagata K, Saito Y , Andica C, et al. Noninvasive Magnetic Resonance Imaging Measures of 889 Glymphatic System Activity. J Magn Reson Imaging . May 2024;59(5):1476 -1493. 890 doi:10.1002/jmri.28977 891 23. Wang DJ, Hua J, Cao D, Ho ML. Neurofluids and the glymphatic system: anatomy, physiology, 892 and imaging. Br J Radiol. Nov 2023;96(1151):20230016. doi:10.1259/bjr.20230016 893 24. Sepehrband F, Cabeen RP, Choupan J, et al. Perivascular space fluid contributes to diffusion tensor 894 imaging changes in white matter. Neuroimage. Aug 15 2019;197:243 -254. 895 doi:10.1016/j.neuroimage.2019.04.070 896 25. Taoka T, Masutani Y , Kawai H, et al. Evaluation of glymphatic system activity with the diffusion 897 MR technique: diffusion tensor image analysis along the perivascular space (DTI -ALPS) in 898 Alzheimer's disease cases. Jpn J Radiol. Apr 2017;35(4):172-178. doi:10.1007/s11604-017-0617-899 z 900 26. Steward CE, Venkatraman VK, Lui E, et al. Assessment of the DTI -ALPS Parameter Along the 901 Perivascular Space in Older Adults at Risk of Dementia. J Neuroimaging. May 2021;31(3):569-902 578. doi:10.1111/jon.12837 903 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 63 27. Liu X, Barisano G, Shao X, et al. Cross-Vendor Test-Retest Validation of Diffusion Tensor Image 904 Analysis along the Perivascular Space (DTI-ALPS) for Evaluating Glymphatic System Function. 905 Aging Dis. Aug 1 2024;15(4):1885-1898. doi:10.14336/AD.2023.0321-2 906 28. Ringstad G. Glymphatic imaging: a criti cal look at the DTI -ALPS index. Neuroradiology. 907 2024;66(2):157-160. 908 29. Taoka T, Ito R, Nakamichi R, Nakane T, Kawai H, Naganawa S. Diffusion Tensor Image Analysis 909 ALong the Perivascular Space (DTI -ALPS): Revisiting the Meaning and Significance of the 910 Method. Magn Reson Med Sci. Jul 1 2024;23(3):268-290. doi:10.2463/mrms.rev.2023-0175 911 30. Huang SY , Zhang YR, Guo Y , et al. Glymphatic system dysfunction predicts amyloid deposition, 912 neurodegeneration, and clinical progression in Alzheimer's disease. Alzheimers Dement . May 913 2024;20(5):3251-3269. doi:10.1002/alz.13789 914 31. Li Y , Wang L, Zhong J, et al. Impaired glymphatic function as a biomarker for subjective cognitive 915 decline: An exploratory dual cohort study. Alzheimers Dement . Sep 2024;20(9):6542 -6555. 916 doi:10.1002/alz.14149 917 32. Son SJ, Park B, Roh HW, et al. Biobank innovations for chronic cerebrovascular disease with 918 Alzheimer's disease study (BICWALZS) Development of new models and analysis 919 methods/screening studies/platforms. Alzheimer's & Dementia. 2020;16:e042259. 920 33. Schweitzer N, Son SJ, Thurston RC, et al. Sex-specific risk factors and clinical dementia outcomes 921 for white matter hyperintensities in a large South Korean cohort. Alzheimer's Research & Therapy. 922 2024;16(1):243. 923 34. Roberts RO, Geda YE, Knopman DS, et al. The Mayo Clinic Study of Aging: design and sampling, 924 participation, baseline measures and sample characteristics. Neuroepidemiology. 2008;30(1):58-925 69. 926 35. Petersen RC, Roberts RO, Knopman DS, et al. Prevalence of mild cognitive impairment is higher 927 in men: The Mayo Clinic Study of Aging. Neurology. 2010;75(10):889-897. 928 36. Schwarz CG, Kremers WK, Prakaashana CM, et al. A Large Public Release of Clinical and 929 Imaging Data from the Mayo Clinic Study of Aging. Alzheimer's & Dementia. 2024;20:e091346. 930 37. Gaser C, Dahnke R, Thompson PM, Kurth F, Luders E, Initiative AsDN. CAT: a computational 931 anatomy toolbox for the analysis of structural MRI data. Gigascience. 2024;13:giae049. 932 38. Tzourio-Mazoyer N, Landeau B, Papathanassiou D, et al. Automat ed anatomical labeling of 933 activations in SPM using a macroscopic anatomical parcellation of the MNI MRI single -subject 934 brain. Neuroimage. 2002;15(1):273-289. 935 39. Roh HW, Chauhan N, Seo SW, et al. Assessing cognitive impairment and disability in older adults 936 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 64 through the lens of whole brain white matter patterns. Alzheimer's & Dementia. 2024;20(9):6032-937 6044. 938 40. Tudorascu DL, Minhas DS, Lao PJ, et al. The use of Centiloids for applying [11C] PiB 939 classification cutoffs across region -of-interest delineation m ethods. Alzheimer's & Dementia: 940 Diagnosis, Assessment & Disease Monitoring. 2018;10:332-339. 941 41. Jack Jr CR, Wiste HJ, Weigand SD, et al. Defining imaging biomarker cut points for brain aging 942 and Alzheimer's disease. Alzheimer's & Dementia. 2017;13(3):205-216. 943 42. Yeh F-C, Tseng W -YI. NTU -90: a high angular resolution brain atlas constructed by q -space 944 diffeomorphic reconstruction. Neuroimage. 2011;58(1):91-99. 945 43. Yeh F-C, Wedeen VJ, Tseng W -YI. Generalized q -sampling imaging. IEEE transactions on 946 medical imaging. 2010;29(9):1626-1635. 947 44. Yeh F-C, Panesar S, Fernandes D, et al. Population -averaged atlas of the macroscale human 948 structural connectome and its network topology. Neuroimage. 2018;178:57-68. 949 45. Bookheimer SY , Salat DH, Terpstra M, et al. The lifespan human connectome project in aging: an 950 overview. Neuroimage. 2019;185:335-348. 951 46. Chen CL, Jin H, Schweitzer N, et al. Markers of Interstitial Fluid Dynamics from Diffusion MRI 952 Reveal Association with Amyloid Deposition in Cognitively Normal Older Adults. Alzheimer's & 953 Dementia. 2024;20:e093872. 954 47. Penny WD, Friston KJ, Ashburner JT, Kiebel SJ, Nichols TE. Statistical parametric mapping: the 955 analysis of functional brain images. Elsevier; 2011. 956 48. Oh E-A, Kang Y-W, Shin J-H, Yeon B-K. A validity study of K -MMSE as a screening test for 957 dementia: comparison against a comprehensive neuropsychological evaluation. Dementia and 958 Neurocognitive Disorders. 2010;9(1):8-12. 959 49. O’Bryant SE, Waring SC, Cullum CM, et al. Staging dementia using Clinical Dementia Rating 960 Scale Sum of Boxes scores: a Texas Alzheimer's research consortium study. Archives of neurology. 961 2008;65(8):1091-1095. 962 50. Ahn H-J, Chin J, Park A, et al. Seoul Neuropsychological Screening Battery -dementia version 963 (SNSB-D): a useful tool for assessing and monitoring cognitive impairments in dementia patients. 964 Journal of Korean medical science. 2010;25(7):1071-1076. 965 51. Yoon B, Shim YS, Cheong H-K, et al. White matter hyperintensities in mild cognitive impairment: 966 clinical impact of location and interaction with lacunes and medial temporal atrophy. Journal of 967 Stroke and Cerebrovascular Diseases. 2014;23(5):e365-e372. 968 52. Scheltens P, Leys D, Barkhof F, et al. Atrophy of medial temporal lobes on MRI in" probable" 969 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 65 Alzheimer's disease and normal ageing: diagnostic value and neuropsychological correlates. 970 Journal of Neurology, Neurosurgery & Psychiatry. 1992;55(10):967-972. 971 53. Chen C-L, Torbati ME, Minhas DS, et al. Superpixel -ComBat modeling: A joint approach for 972 harmonization and characterization of inter -scanner variability in T1 -weighted images. Imaging 973 Neuroscience. 2024;2:1-21. 974 54. Fortin J-P, Parker D, Tunç B, et al. Harmonization of multi -site diffusion tensor imaging data. 975 Neuroimage. 2017;161:149-170. 976 55. Fortin J-P, Cullen N, Sheline YI, et al. Harmonization of cortical thickness measurements across 977 scanners and sites. Neuroimage. 2018;167:104-120. 978 56. Kamagata K, Andica C, Takabayashi K, et al. Association of MRI Indices of Glymphatic System 979 With Amyloid Deposition and Cognition in Mild Cognitive Impairment and Alzheimer Disease. 980 Neurology. Dec 12 2022;99(24):e2648-e2660. doi:10.1212/WNL.0000000000201300 981 57. Liu C-C, Kanekiyo T, Xu H, Bu G. Apolipoprotein E and Alzheimer disease: risk, mechanisms 982 and therapy. Nature Reviews Neurology. 2013;9(2):106-118. 983 58. Liu C-C, Zhao N, Fu Y , et al. ApoE4 accelerates early seeding of amyloid pathology. Neuron. 984 2017;96(5):1024-1032. e3. 985 59. Guo X, Zhang G, Peng Q, Huang L, Zhang Z, Zhang Z. Emerging roles of meningeal l ymphatic 986 vessels in Alzheimer’s disease. Journal of Alzheimer's Disease. 2023;94(s1):S355-S366. 987 60. Silva I, Silva J, Ferreira R, Trigo D. Glymphatic system, AQP4, and their implications in 988 Alzheimer’s disease. Neurological research and practice. 2021;3:1-9. 989 61. Wang C, Nambiar A, Strickland MR, et al. APOE-ε4 synergizes with sleep disruption to accelerate 990 Aβ deposition and Aβ-associated tau seeding and spreading. The Journal of Clinical Investigation. 991 2023;133(14) 992 62. Sadleir KR, Vassar R. Connections be tween ApoE, sleep, and A β and tau pathologies in 993 Alzheimer’s disease. Journal of Clinical Investigation. 2023;133(14):e171838. 994 63. Castellano JM, Kim J, Stewart FR, et al. Human apoE isoforms differentially regulate brain 995 amyloid-β peptide clearance. Science translational medicine. 2011;3(89):89ra57-89ra57. 996 64. Klostranec JM, Vucevic D, Bhatia KD, et al. Current Concepts in Intracranial Interstitial Fluid 997 Transport and the Glymphatic System: Part I -Anatomy and Physiology. Radiology. Dec 998 2021;301(3):502-514. doi:10.1148/radiol.2021202043 999 65. Wardlaw JM, Benveniste H, Nedergaard M, et al. Perivascular spaces in the brain: anatomy, 1000 physiology and pathology. Nat Rev Neurol . Mar 2020;16(3):137 -153. doi:10.1038/s41582-020-1001 0312-z 1002 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 66 66. Mestre H, Kostrikov S, Mehta RI, Nedergaard M. Perivascular spaces, glymphatic dysfunction, 1003 and small vessel disease. Clinical science. 2017;131(17):2257-2274. 1004 67. Pinheiro A, Demissie S, Scruton A, et al. Association of Apolipoprotein E varepsilon4 Allele with 1005 Enlarged Perivascular Spaces. Ann Neurol. Jul 2022;92(1):23-31. doi:10.1002/ana.26364 1006 68. Hong H, Hong L, Luo X, et al. The relationship between amyloid pathology, cerebral small vessel 1007 disease, glymphatic dysfunction, and cognition: a study based on Alzheimer's disease contin uum 1008 participants. Alzheimers Res Ther. Feb 20 2024;16(1):43. doi:10.1186/s13195-024-01407-w 1009 69. Chen C-L, Cheng S-Y , Montaser-Kouhsari L, et al. Advanced brain aging in Parkinson ’s disease 1010 with cognitive impairment. npj Parkinson's Disease. 2024;10(1):62. 1011 70. Chen C -L, Kuo M -C, Chen P -Y , et al. Validation of neuroimaging -based brain age gap as a 1012 mediator between modifiable risk factors and cognition. Neurobiology of Aging. 2022;114:61-72. 1013 1014 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 1 Supplementary Materials 1 S1 Recruitment criteria of the BICW ALZS dataset 2 S2 Region sampling for DTI-ALPS calculation 3 S3 Hyperparameter optimization for targeting periventricular area 4 S4 Partial correlation analysis of DTI-ALPS and PVeD adjusted for mean diffusivity 5 S5 Interaction of DTI-ALPS with APOE4 on amyloid deposition 6 S6 Group comparisons of DTI-ALPS and PVeD between CU, MCI, and dementia 7 S7 Additional results in the replication cohort: the MCSA dataset 8 Supplementary Tables 9 S. Table 1 Neuroimaging data acquisition parameters 10 S. Table 2 Demographic information of subjects in the MCSA replication cohort 11 S. Table 3 Partial correlation analysis of diffusion imaging-derived measures 12 S. Table 4 Interaction effects between PVeD and APOE4 on amyloid deposition 13 S. Table 5 Interaction effects between DTI-ALPS and APOE4 on amyloid deposition 14 S. Table 6 Association between baseline PVeD and longitudinal cognitive changes 15 S. Table 7 Group comparisons of DTI-ALPS and PVeD 16 S. Table 8 Additional results in the replication cohort: the MCSA dataset 17 18 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 2 S1 Recruitment criteria of the BICWALZS dataset 19 The BICWALZS study’s primary objective is to facilitate the systematic collection and utilization 20 of human biological specimens and real-world clinical data for research on subjective cognitive decline 21 (SCD), mild cognitive impairment (MCI), Alzheimer ’s disease (AD), and subcortical vascula r 22 dementia (SVaD) (Son et al., 2020) . Clinical diagnoses were established using internationally 23 recognized criteria: SCD was defined as self - or informant -reported cognitive decline without 24

impairment in neurocognitive tasks (no less than −1.5 standard deviations in each domain 25 and Clinical Dementia Rating (CDR) = 0) (CHOI et al., 2001); MCI was diagnosed based on a CDR 26 score of 0.5 and the expanded Mayo Clinic criteria (Winblad et al., 2004); AD dementia was classified 27 using the National Institute on Aging-Alzheimer’s Association (NIA-AA) Core Clinical Probable AD 28 Dementia Criteria (McKhann et al., 2011) ; and SVaD was identified based on moderate -to-severe 29 white matter hyperintensity (WMH) burden and diagnostic c riteria outlined in the Diagnostic and 30 Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) (American Psychiatric Association and 31 American Psychiatric Association, 2013) . Patients with a history of neurological or systemic 32 conditions such as territorial cerebral infarction, intracranial hemorrhage, Parkinson ’s disease, heart 33 failure, renal failure, or any condition that could confound study findings were excluded. Clinical 34 assessments included CDR global scores and sum of boxes (CDR -SB), a validated measure of 35 cognitive and functional impairment used in dementia clinical trials (Morris, 1993) . In addition, 36 comprehensive evaluations were conducted, including blood pressure, pulse pressure, body mass index, 37 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 3 smoking status, amyloid positron emission tomography (PET), apolipoprotein E (APOE) genotyping, 38 and standardized neuropsychological assessments. A subset of participants was further classified based 39 on neuroimaging biomarkers including cortical amyloid burden and subcortical vascular burden, into 40 cognitively unimpaired (CU), Alzheimer’s disease-related cognitive impairment (ADCI), or vascular 41 cognitive impairment (VCI) groups. Participants with cortical amyloid pathology or infar ctions 42 unrelated to vascular dementia, such as those resulting from radiation injury, multiple sclerosis, 43 vasculitis, or leukodystrophy, were excluded from C U and VCI classifications. The study included 44 1,013 participants, with subsets undergoing additional evaluations such as brain MRI (n = 817), brain 45 amyloid PET (n = 713), single nucleotide polymorphism microarray analysis (n = 949), actigraphy 46 measurement (n = 200), and patient -derived dermal fibroblast sampling (n = 175). A structured 47 longitudinal follow-up protocol was established, with annual brief assessments for all participants and 48 biannual comprehensive evaluations —including neuropsychological testing, brain MRI, and 49 actigraphy measurements —for those exhibiting cortical amyloid burden, subcortical vascular 50 pathology, APOE e4 positivity, or significant cognitive decline. Among the participants, 336 were 51 followed longitudinally for cognitive decline using CDR -SB and clinical diagnoses. BICWALZS is 52 registered in the Korean National Clinical Trial Regi stry (Clinical Research Information Service, 53 identifier: KCT0003391) and was approved by the Institutional Review Board (IRB) of Ajou 54 University Hospital (AJOUIRB -SUR-2021-038). Written informed consent was obtained from all 55 participants and their caregivers.56 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 4 S2 Region sampling for DTI-ALPS calculation 57 To define the regions of interest (ROIs) for the diffusion tensor imaging –analysis along 58 the perivascular space (DTI -ALPS) calculation (Liu et al., 2024; Taoka et al., 2017) , we first 59 generated a group -averaged template in the Montreal Neurological Institute (MNI) space 60 (Supplementary Figure 1) . This was achieved using the reconstructed diffusion -weighted 61 images (DWIs) from the Human Connectome Proje ct–Aging (HCP -Aging) dataset 62 (Bookheimer et al., 2019) , accessed via the DSI Studio data -sharing portal 63 (https://brain.labsolver.org/hcp_a.html). Projection and association fibers were identified on 64 the color-coded fractional anisotropy (FA) maps of the template, and spherical ROIs with a 3 65 mm radius were manually placed within these fiber regions, specifically at the level of the 66 lateral ventricle body bilaterally. A total of four ROIs were applied to the axis -specific 67 diffusivity maps of each subject, and the diffusivity values along the principal axes (Dxx, Dyy, 68 and Dzz) were extracted for DTI-ALPS calculation. The DTI-ALPS index was derived as the 69 ratio of axis -specific diffusivities in (1) projection (Dxx_proj) and association fibers 70 (Dxx_assoc) along the x -axis to (2) projection fibers (Dyy_proj) along the y -axis and 71 association fibers (Dzz_assoc) along the z -axis (Supplementary Figure 2) . For each subject, 72 DTI-ALPS indices were computed for the left and right hemispheres, as well as a mean value, 73 to quantify diffusion signals associated with glymphatic function (Taoka et al., 2017) . To 74 validate the effectiveness of DTI-ALPS measures as described in previous literature (Huang et 75 al., 2024; Li et al., 2024), we performed a regression analysis by using the BICWALZS cohort. 76 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 5 The mean DTI -ALPS was an i ndependent variable, and the Mini -Mental State Examination 77 (MMSE) score was the dependent variable while controlling for age, sex, and education. The 78

showed a significant positive correlation between them (Supplementary Figure 1), which 79 is in line with the previous findings. 80 81 Supplementary Figure 1. The group-average template based on the diffusion -weighted 82 images acquired from the HCP-Aging (HCP-A) dataset for defining sampling regions of 83 interest in the DTI-ALPS calculation. 84 85 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 6 86 Supplementary Figure 2. Illustration of the Diffusion Tensor Imaging Analysis along the 87 Perivascular Space (DTI-ALPS) method designed to evaluate glymphatic function using 88 diffusion MRI. The method quantifies diffusivity in periventricular white matter regions 89 closely associated with the perivascular space (PVS). Spherical regions of interest are manually 90 placed in deep white matter adjacent to the lateral ventricles (a), covering both projection and 91 association fibers. The DTI-ALPS index is derived by comparing diffusiv ity along directions 92 parallel and perpendicular to the PVS (b), providing a metric for assessing glymphatic-related 93 function in the brain. 94 95 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 7 S3 Hyperparameter optimization for targeting periventricular area 96 We developed an automated region growth algorithm to delineate periventricular regions 97 from mean diffusivity (MD) maps, specifically targeting areas containing deep medullary veins. 98 The algorithm first segments MD maps to generate white matter (WM) masks, followed by the 99 registration of a predef ined lateral ventricle (LV) prior mask onto the MD maps to 100 accommodate individual LV morphology. A posterior LV mask is then derived and serves as 101 the basis for initializing periventricular areas (PVeAs), which function as the sampling regions 102 of interest (ROIs). To refine these ROIs, the algorithm applies an anatomically guided dilation 103 of the LV posterior mask along the transverse axis, incorporating differential expansion 104 coefficients that account for the known morphological variations of the lateral ven tricles. 105 Maximum dilation is applied at the ventricular body, whereas minimal expansion is enforced 106 at the peripheral ventricular margins to preserve anatomical fidelity. The hyperparameters of 107 the region growth algorithm were empirically optimized through multimodal image analysis, 108 integrating susceptibility -weighted imaging, T1 -weighted imaging, and diffusion -weighted 109 imaging (DWI) to enhance the accuracy of periventricular region delineation, with particular 110 emphasis on deep medullary vein localization (Supplementary Figure 3) . This analysis was 111 performed using a proprietary multimodal dataset, and the performance of the ROI generation 112 process was rigorously assessed by experienced neuroimaging experts. 113 114 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 8 115 Supplementary Figure 3. Automatically generated r egions of interest (ROIs) for 116 delineating periventricular regions based on mean diffusivity (MD) maps. The proposed 117 region growth algorithm was optimized using multimodal imaging data, including 118 susceptibility-weighted imaging (SWI), T1 -weighted imaging (T 1w), and MD maps. The 119 resulting ROIs serve as sampling regions for transverse tensor ratio (TTR) maps, which is used 120 to represent periventricular diffusivity. 121 122 123 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 9 S4 Partial correlation analysis of DTI-ALPS and PVeD additionally adjusted for mean diffusivity 124 125 Supplementary Figure 4. Partial correlation matrix of imaging-derived metrics with multifaceted clinical characteristics, with the additional statistical 126 adjustment for mean diffusivity in the BICWALZS cohort. Imaging-derived metrics refer to the diffusion tensor imaging (DTI)-derived measures including 127 DTI-ALPS and periventricular diffusivity (PVeD). The multifaceted clinical characteristics include four domains; symptom severity, cognitive outcomes, 128 neurodegeneration, and Aβ burden. The matrix was generated using the BICWALZS dataset (n = 440). Values are Spearman’s correlation coefficients for each 129 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 10 pairwise correlation. Covariates including age, sex, education , and mean diffusivity were adjusted. Hippocampal volume was additionally adjusted by the total 130 intracranial volume. Darker blue and red colors indicate greater negative and positive correlations, respectively. Gray dots indicate those did not pass the 131 significance threshold adjusted by multiple comparisons. Abbreviations: adj, adjusted; avg, average; CDR, Clinical Dementia Rating; CDR-SB, Clinical Dementia 132 Rating Sum of Boxes; Cing, cingulum; CL, centiloid; CU, cognitively unimpaired; DTI, diffusion tensor image; DTI-ALPS, diffusion tensor image analysis along 133 the perivascular space; Front, frontal; Hippo, hippocampal; K-BNT, Korean version of the Boston Naming Test ; K-CWST, Korean version of the Color Word 134 Stroop Test; L, left; MCI, mild cognitive impairment; MMSE, Mini-Mental State Examination; MTA, medial temporal lobe atrophy; Occi, occipital; Parie, parietal; 135 PET, positron emission tomography; PVeD, periventricular diffusivity; R, right; RCFT, Rey Complex Figure Test and Recognition Trial; SUVR, standard uptake 136 value ratio; SVLT-E, Seoul Verbal Learning Test Elderly's version; Temp, temporal; V ol, volume; WM, white matter; z, z-score. 137 138 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 11 S5 Interaction of DTI-ALPS with APOE4 on amyloid deposition 139 140 Supplementary Figure 5. The presence of APOE4 allele enhances the association between DTI-ALPS and Aβ deposition in the BICWALZS cohort. 141 Details of the statistical tests used are p rovided in Supplementary Table 5. The amyloid PET SUVRs were sampled from five anatomical regions including the 142 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 12 cingulum (a), frontal lobe (b), parietal lobe (c), occipital lobe (d), and temporal lobe (e), with the inclusion of the global measure (f). The models included age, 143 sex, and education as covariates. Multiple comparisons were controlled using the Benjamini-Hochberg correction accounting for the number of image measures 144 analyzed. Green and red colors represent APOE4 non-carriers and APOE4 carriers, respectively. The interaction terms between DTI-ALPS and APOE4 are 145 reported. Abbreviations: APOE, apolipoprotein E; CL, centiloid; DTI-ALPS, diffusion tensor image analysis along the perivascular space; FDR, false discovery 146 rate; PET, positron emission tomography; SUVR, standard uptake value ratio. 147 148 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 13 S6 Group comparisons of DTI-ALPS and PVeD between CU, MCI, and dementia 149 150 Supplementary Figure 6. Beeswarm plots of PVeD and DTI-ALPS measures for group comparison in the BICWALZS cohort. Details of the statistical 151 tests used are provided in Supplementary Table 7. Diffusion imaging-derived metrics for group comparisons include mean PVeD (a), left PVeD (b), right PVeD 152 (c), mean DTI-ALPS index (d), left DTI-ALPS index (e), and right DTI-ALPS index (f). 153 154 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 14 S7 Additional results in the replication cohort: the MCSA dataset 155 156 Supplementary Figure 7. Partial correlation matrix of imaging-derived metrics with multifaceted clinical characteristics, with the additional statistical 157 adjustment for mean diffusivity in the MCSA cohort (a), and the presence of APOE4 allele enhances the association between DTI-ALPS and Aβ 158 deposition in the MCSA cohort (b). 159 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 15 160 Supplementary Figure 8. Beeswarm plots of PVeD (a) and DTI-ALPS (b) measures for group comparison in the MCSA cohort. 161 162 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 16 Supplementary Table 1. MRI, DTI, and PET acquisition parameters for the BICWALZS and DTI acquisition parameters for the 163 MCSA 164  Diffusion weighted imaging 165 166 167  T1-weighted imaging 168 169 Site MRI manufacturer Magnetic field Acquisition scheme b-value (s/mm^2) Number of diffusion directions Number of b0 image TR (msec) TE (msec) Matrix size In-plane resolution (mm) Slice thickness (mm) Sequences A GE DISCOVERY MR750w 3T single shell DTI 1000 24 1 10,800 16 256 x 256 0.875 2 SE EPI B GE DISCOVERY MR750w 3T single shell DTI 800 24 1 10,800 16 256 x 256 0.875 2 SE EPI C GE DISCOVERY MR750w 3T single shell DTI 600 60 1 10,000-15,000 75-89 128 x 128 1.797 3 SE EPI D Philips Achieva 3T single shell DTI 600 45 1 8,700-11,000 60-90 128 x 128 1.719 2 SE EPI E Siemens TrioTim 3T single shell DTI 1000 30 1 9,000-10,000 60-130 256 x 256 0.938 2 SE EPI F Philips Achieva 3T single shell DTI 600 45 1 8,700-11,000 60-90 128 x 128 1.719 2 SE EPI G GE DISCOVERY MR750w 3T single shell DTI 1000 24 1 13,700-15,600 75-89 256 x 256 0.875 2 SE EPI Site MRI manufacturer Magnetic field Acquisition scheme b-value (s/mm^2) Number of diffusion directions Number of b0 image TR (msec) TE (msec) Matrix size In-plane resolution (mm) Slice thickness (mm) Sequences A GE MEDICAL SYSTEMS 3T single shell DTI 1000 41 5 10,000 63 256 x 256 1.3672 2.7 SE EPI B GE MEDICAL SYSTEMS 3T single shell DTI 1000 41 4 10,000 69 256 x 256 1.3672 2.7 SE EPI C GE MEDICAL SYSTEMS 3T single shell DTI 1000 41 4 10,000 69 256 x 256 1.3672 2.7 SE EPI D GE MEDICAL SYSTEMS 3T single shell DTI 1000 41 4 10,200 69 256 x 256 1.3672 2.7 SE EPI E GE MEDICAL SYSTEMS 3T single shell DTI 1000 41 5 11,300 69 256 x 256 1.3672 2.7 SE EPI F GE MEDICAL SYSTEMS 3T single shell DTI 1000 41 5 10,000 63 256 x 256 1.3672 2.7 SE EPI BICWALZS MCSA Site MRI manufacturer Magnetic field TR (msec) TE (msec) Matrix size In-plane resolution (mm) Slice thickness (mm) Flip angle (degree) A GE DISCOVERY MR750w 3T 7,100-9,900 2.8-4.8 256 x 256 0.78 x 0.78 1 12 B GE DISCOVERY MR750w 3T 7,400-7,600 2.7-2.8 256 x 256 1.0 x 1.0 1.2 11 C GE DISCOVERY MR750w 3T 7,500-8,600 2.8-3.3 256 x 256 0.88 x 0.88 1 12 D Philips Achieva 3T 9,900 4.6 480 x 480 0.5 x 0.5 1 8 E Siemens TrioTim 3T 1,800 1.8 256 x 256 0.98 x 0.98 1 9 F Philips Achieva 3T 9,900 4.6 480 x 480 0.5 x 0.5 1 8 G GE DISCOVERY MR750w 3T 8,200-14,000 3.2-6.0 512 x 512 0.5 x 0.5 1.3 9 BICWALZS .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 17 170  Amyloid PET imaging 171 172 Site Manufacturer Acquisition matrix Voxel size (mm) Tracer Machine model A GE 128 x 128 x 47 1.95 x 1.95 Flutmetamol GE, Discovery STE B GE 128 x 128 x 47 2.34 x 2.34 Flutmetamol GE, Discovery 690 C GE 128 x 128 x 47 1.95 x 1.95 Flutmetamol GE, Discovery STE/GE, Discovery ST D GE 128 x 128 x 188 2 x 2 Flutmetamol GE, Discovery STE E Siemens 128 x 128 x 148 1.95 x 1.95 Flutmetamol Siemense, Biograph40 TruePoint F GE 128 x 128 x 188 2 x 2 Flutmetamol GE, Discovery STE G GE 256 x 256 x 47 0.98 x 0.98 Flutmetamol GE, Discovery STE BICWALZS .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 18 Supplementary Table 2. Characteristics of the replication cohort from MCSA. 173 Demographic characteristics CU MCI & Dementia Comparisons N 351 63 - Age, year 74.7 (±8.5) 79.1 (±7.0) p < 0.001 Sex, female, N (%) 168 (48.0%) 21 (33.3%) p = 0.031 Education, year 14.5 (±2.7) 13.9 (±3.0) p = 0.125 MMSE 28.3 (±1.2) 25.5 (±2.4) p < 0.001 Global CDR 0.0 (±0.1) 0.4 (±0.3) p < 0.001 CDR SB 0.0 (±0.2) 1.3 (±1.5) p < 0.001 APOE4 +, N (%) 83 (23.7%) 23 (36.5%) p = 0.036 Neuropsychological characteristics Population z-score memory -0.028 (±1.000) -1.869 (±0.994) p < 0.001 Population z-score language -0.096 (±0.966) -1.399 (±1.265) p < 0.001 Population z-score attention -0.113 (±0.957) -1.368 (±1.263) p < 0.001 Population z-score visuospatial 0.077 (±0.992) -0.754 (±0.964) p < 0.001 Population z-score global -0.037 (±0.935) -1.646 (±0.873) p < 0.001 Imaging characteristics Adjusted bilateral hippocampal volume 7.296 (±0.745) 6.772 (±1.005) p = 0.001 Amyloid PET PiB global CL SUVR 28.573 (±32.343) 60.220 (±47.745) p < 0.001 DTI-ALPS 1.474 (±0.176) 1.392 (±0.152) p < 0.001 PVeD 0.474 (±0.035) 0.460 (±0.039) p = 0.007 1. Comparisons were adjusted for age, sex, and education in the neuropsychological and imaging measures 174 2. Hippocampal volume was adjusted by total intracranial volume 175 Abbreviations: APOE, apolipoprotein E; CDR, Clinical Dementia Rating; CDR-SB, Clinical Dementia Rating 176 Sum of Boxes; CL, centiloid; CU, cognitively unimpaired; DTI-ALPS, diffusion tensor image analysis along 177 the perivascular space ; MCI, mild cognitive impairment ; MMSE, Mini -Mental State Examination; PET, 178 positron emission tomography ; PiB, Pittsburgh Compound B; PVeD, periventricular diffusivity ; SUVR, 179 standard uptake value ratio. 180 181 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 19 Supplementary Table 3. Partial correlation analysis of diffusion imaging-derived measures 182  Partial correlation coefficients in Figure 3a. 183 184 185 * Values in cells: Correlation (raw p-values) 186 ** Corrected Significance Threshold = 0.0031 187 188 189 190 Global CDR CDR SB MMSE K BNT z RCFT Delayed Recall z RCFT Recognition z SVLT E Delayed Recall z SVLT E Recognition z K CWST Color Reading z socre DTI-ALPS -0.21067 (8.94E-06) -0.19071 (6.02E-05) 0.15671 (1.01E-03) 0.15645 (2.80E-03) 0.15349 (1.37E-03) 0.16117 (8.18E-04) 0.16481 (5.42E-04) 0.12786 (7.45E-03) 0.22191 (4.88E-06) PVeD -0.17251 (2.91E-04) -0.19571 (3.80E-05) 0.17032 (3.48E-04) 0.20663 (7.30E-05) 0.19603 (4.09E-05) 0.11531 (1.70E-02) 0.18250 (1.25E-04) 0.17476 (2.41E-04) 0.16337 (8.24E-04) WM FA -0.21184 (7.95E-06) -0.16898 (3.88E-04) 0.19223 (5.24E-05) 0.15223 (3.64E-03) 0.15883 (9.24E-04) 0.12969 (7.22E-03) 0.12990 (6.54E-03) 0.09327 (5.14E-02) 0.24649 (3.56E-07) WM MD 0.27423 (5.60E-09) 0.27165 (7.87E-09) -0.26566 (1.71E-08) -0.22717 (1.24E-05) -0.24123 (3.89E-07) -0.19666 (4.19E-05) -0.21435 (6.16E-06) -0.17454 (2.46E-04) -0.29799 (5.61E-10) MTA scale L MTA scale R adj avg Hippo Vol L adj avg Hippo Vol R CL SUVR Cing CL SUVR Front CL SUVR Parie CL SUVR Temp CL SUVR Occi CL SUVR Global DTI-ALPS -0.36897 (1.53E-15) -0.33793 (3.91E-13) 0.23112 (1.17E-06) 0.18070 (1.57E-04) -0.02783 (5.62E-01) -0.05675 (2.36E-01) -0.07589 (1.13E-01) -0.05333 (2.66E-01) -0.10901 (2.27E-02) -0.07477 (1.19E-01) PVeD -0.51149 (1.67E-30) -0.45025 (3.35E-23) 0.39028 (3.33E-17) 0.40096 (3.74E-18) -0.13306 (5.34E-03) -0.15557 (1.10E-03) -0.15539 (1.12E-03) -0.12641 (8.16E-03) -0.18715 (8.29E-05) -0.15970 (8.07E-04) WM FA -0.33181 (1.09E-12) -0.34649 (9.01E-14) 0.25865 (4.77E-08) 0.21065 (9.85E-06) 0.00843 (8.61E-01) 0.00078 (9.87E-01) -0.03579 (4.55E-01) -0.02899 (5.46E-01) -0.08487 (7.64E-02) -0.03322 (4.89E-01) WM MD 0.52663 (1.50E-32) 0.53456 (1.16E-33) -0.46818 (5.67E-25) -0.43703 (1.27E-21) 0.04226 (3.78E-01) 0.04962 (3.01E-01) 0.06885 (1.51E-01) 0.07665 (1.10E-01) 0.13837 (3.75E-03) 0.08132 (8.95E-02) .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 20  Partial correlation coefficients in Figure 3b. 191 192 193 * Values in cells: Correlation (raw p-values) 194 ** Corrected Significance Threshold = 0.0031 195 196 Global CDR CDR SB MMSE K BNT z RCFT Delayed Recall z RCFT Recognition z SVLT E Delayed Recall z SVLT E Recognition z K CWST Color Reading z socre DTI-ALPS L -0.20967 (9.88E-06) -0.19646 (3.54E-05) 0.17247 (2.92E-04) 0.15180 (3.74E-03) 0.17919 (1.81E-04) 0.16532 (5.95E-04) 0.14562 (2.28E-03) 0.14612 (2.20E-03) 0.20433 (2.68E-05) DTI-ALPS R -0.15907 (8.47E-04) -0.12775 (7.50E-03) 0.10293 (3.14E-02) 0.11378 (3.02E-02) 0.07916 (1.00E-01) 0.11410 (1.82E-02) 0.14087 (3.17E-03) 0.07575 (1.14E-01) 0.17130 (4.49E-04) PVeD L -0.18195 (1.31E-04) -0.20977 (9.78E-06) 0.16901 (3.87E-04) 0.20270 (1.01E-04) 0.19496 (4.51E-05) 0.11389 (1.84E-02) 0.16990 (3.61E-04) 0.16864 (3.99E-04) 0.15653 (1.36E-03) PVeD R -0.17277 (2.85E-04) -0.18384 (1.11E-04) 0.16996 (3.59E-04) 0.21461 (3.74E-05) 0.19567 (4.22E-05) 0.12324 (1.07E-02) 0.18647 (8.81E-05) 0.17850 (1.76E-04) 0.16822 (5.70E-04) MTA scale L MTA scale R adj avg Hippo Vol L adj avg Hippo Vol R CL SUVR Cing CL SUVR Front CL SUVR Parie CL SUVR Temp CL SUVR Occi CL SUVR Global DTI-ALPS L -0.33435 (7.13E-13) -0.30187 (1.17E-10) 0.21664 (5.39E-06) 0.14988 (1.76E-03) -0.03671 (4.44E-01) -0.05887 (2.19E-01) -0.08824 (6.53E-02) -0.07282 (1.29E-01) -0.11322 (1.79E-02) -0.08323 (8.22E-02) DTI-ALPS R -0.32089 (6.36E-12) -0.29853 (1.91E-10) 0.20355 (1.97E-05) 0.17460 (2.61E-04) -0.01002 (8.35E-01) -0.04093 (3.93E-01) -0.04757 (3.21E-01) -0.02126 (6.58E-01) -0.08017 (9.42E-02) -0.04884 (3.08E-01) PVeD L -0.49154 (5.80E-28) -0.41837 (6.05E-20) 0.37936 (2.87E-16) 0.36343 (5.75E-15) -0.14992 (1.67E-03) -0.17673 (2.05E-04) -0.17286 (2.83E-04) -0.15228 (1.41E-03) -0.20003 (2.53E-05) -0.17946 (1.62E-04) PVeD R -0.51027 (2.41E-30) -0.46665 (5.14E-25) 0.38586 (8.04E-17) 0.41645 (1.36E-19) -0.10816 (2.37E-02) -0.12709 (7.82E-03) -0.13584 (4.44E-03) -0.09853 (3.95E-02) -0.16771 (4.30E-04) -0.13457 (4.83E-03) .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 21 Supplementary Table 4. Interaction effects between PVeD and APOE4 status on amyloid deposition 197  Interaction effects in Figure 5. 198 199 * Other covariates included age, sex, and education 200 PVeD APOE4 PVeD*APOE4 PVeD APOE4 PVeD*APOE4 Coefficients -46.16 281.34 -518.73 Coefficients -52.12 172.34 -336.66 SE 82.05 66.23 149.36 SE 45.42 36.67 82.69 p FDR 0.6889 < 0.0001 0.0006 p FDR 0.6889 < 0.0001 0.0001 PVeD APOE4 PVeD*APOE4 PVeD APOE4 PVeD*APOE4 Coefficients -40.60 226.06 -415.26 Coefficients -8.07 167.41 -313.70 SE 61.17 49.38 111.35 SE 42.71 34.47 77.74 p FDR 0.6889 < 0.0001 0.0003 p FDR 0.8502 < 0.0001 0.0001 PVeD APOE4 PVeD*APOE4 PVeD APOE4 PVeD*APOE4 Coefficients -48.50 231.94 -437.88 Coefficients -37.24 195.30 -365.45 SE 56.47 45.59 102.80 SE 50.70 40.92 92.28 p FDR 0.6889 < 0.0001 0.0001 p FDR 0.6889 < 0.0001 0.0001 Response: CL SUVR global Response: CL SUVR cingulum Response: CL SUVR frontal lobe Response: CL SUVR parietal lobe Response: CL SUVR occipital lobe Response: CL SUVR temporal lobe .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 22 Supplementary Table 5. Interaction effects between DTI-ALPS and APOE4 status on amyloid deposition 201 202 * Other covariates included age, sex, and education 203 204 DTI-ALPS APOE4 DTI-ALPS*APOE4 DTI-ALPS APOE4 DTI-ALPS*APOE4 Coefficients 9.36 191.80 -93.80 Coefficients -7.32 103.64 -54.11 SE 20.02 54.31 36.21 SE 11.13 30.19 20.13 p FDR 0.8286 0.0005 0.0099 p FDR 0.8286 0.0007 0.009 DTI-ALPS APOE4 DTI-ALPS*APOE4 DTI-ALPS APOE4 DTI-ALPS*APOE4 Coefficients 8.92 161.79 -80.76 Coefficients 2.91 108.12 -53.45 SE 15.20 41.25 27.50 SE 10.51 28.52 19.02 p FDR 0.8286 0.0003 0.0077 p FDR 0.8286 0.0003 0.0077 DTI-ALPS APOE4 DTI-ALPS*APOE4 DTI-ALPS APOE4 DTI-ALPS*APOE4 Coefficients 3.76 152.94 -77.38 Coefficients 2.76 131.23 -65.61 SE 14.13 38.34 25.56 SE 12.74 34.57 23.05 p FDR 0.8286 0.0003 0.0077 p FDR 0.8286 0.0003 0.0077 Response: CL SUVR cingulum Response: CL SUVR occipital lobe Response: CL SUVR frontal lobe Response: CL SUVR temporal lobe Response: CL SUVR parietal lobe Response: CL SUVR global .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 23 Supplementary Table 6. Association between baseline PVeD and longitudinal cognitive changes 205  Regression analysis in Figure 6. 206 207 *Response: MMSE annual change rate 208 ** Other covariates included age, sex, and education 209 210 PVeD L PVeD R PVeD DTI-ALPS L DTI-ALPS R DTI-ALPS Coefficients 14.734 16.425 16.698 0.716 1.756 1.784 SE 4.781 5.512 5.515 1.327 1.170 1.436 p FDR 0.00785 0.00785 0.00785 0.59123 0.20669 0.26180 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 24 Supplementary Table 7. Group comparisons of DTI-ALPS and PVeD between CU, MCI, and dementia 211 212 213 * Other covariates included age, sex, and education 214 ** Following a significant ANCOV A, pairwise comparisons were conducted using the Bonferroni method to examine differences between adjusted group 215 means while controlling for multiple comparisons. 216 PVeD PVeD L PVeD R CU 0.459 (0.037) 0.453 (0.041) 0.465 (0.035) MCI 0.454 (0.037) 0.446 (0.040) 0.461 (0.038) Dementia 0.438 (0.039) 0.430 (0.042) 0.445 (0.040) Main effect F (2,434) = 8.14, P FDR = 0.000376 F (2,434) = 8.10, P FDR = 0.000376 F (2,434) = 8.03, P FDR = 0.000376 CU vs. Dementia P Bonferroni = 0.0054 P Bonferroni = 0.0056 P Bonferroni = 0.0059 CU vs. MCI P Bonferroni = 1.0000 P Bonferroni = 1.0000 P Bonferroni = 1.0000 MCI vs. Dementia P Bonferroni = 0.0007 P Bonferroni = 0.0008 P Bonferroni = 0.0007 DTI-ALPS DTI-ALPS L DTI-ALPS R CU 1.598 (0.162) 1.695 (0.197) 1.520 (0.189) MCI 1.523 (0.149) 1.606 (0.186) 1.457 (0.168) Dementia 1.452 (0.163) 1.519 (0.198) 1.400 (0.171) Main effect F (2,434) = 16.12, P FDR < 0.000001 F (2,434) = 15.62, P FDR < 0.000001 F (2,434) = 8.65, P FDR = 0.000376 CU vs. Dementia P Bonferroni < 0.0001 P Bonferroni < 0.0001 P Bonferroni = 0.0011 CU vs. MCI P Bonferroni = 0.0084 P Bonferroni = 0.0105 P Bonferroni = 0.0869 MCI vs. Dementia P Bonferroni < 0.0001 P Bonferroni = 0.0001 P Bonferroni = 0.0064 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 25 Supplementary Table 8. Additional results in the replication cohort: the MCSA dataset 217  Partial correlation coefficients in Figure 7a. 218 219 220 * Values in cells: Correlation (raw p-values) 221 ** Corrected Significance Threshold = 0.0063 222 223 224 225 Global CDR CDR SB MMSE PZ memory PZ language PZ attention PZ visuospatial PZ global adj Hippo Vol PIB CL SUVR DTI-ALPS -0.1423 (3.93E-03) -0.1183 (1.67E-02) 0.1327 (7.35E-03) 0.1058 (3.43E-02) 0.2001 (6.08E-05) 0.1419 (4.98E-03) 0.1500 (2.94E-03) 0.1831 (3.20E-04) 0.0053 (9.14E-01) -0.1094 (2.66E-02) PVeD -0.0897 (6.99E-02) -0.0758 (1.26E-01) 0.1645 (8.66E-04) 0.1388 (5.36E-03) 0.1552 (1.95E-03) 0.1929 (1.26E-04) 0.1326 (8.67E-03) 0.1916 (1.65E-04) -0.0425 (3.90E-01) -0.0463 (3.49E-01) Global CDR CDR SB MMSE PZ memory PZ language PZ attention PZ visuospatial PZ global adj Hippo Vol PIB CL SUVR DTI-ALPS L -0.1411 (4.26E-03) -0.1123 (2.32E-02) 0.1303 (8.49E-03) 0.1163 (1.98E-02) 0.2108 (2.34E-05) 0.1531 (2.43E-03) 0.1081 (3.26E-02) 0.1928 (1.50E-04) 0.0176 (7.22E-01) -0.0866 (7.97E-02) DTI-ALPS R -0.1076 (2.96E-02) -0.0943 (5.68E-02) 0.1007 (4.24E-02) 0.0611 (2.22E-01) 0.1334 (7.87E-03) 0.1033 (4.15E-02) 0.1495 (3.05E-03) 0.1265 (1.33E-02) 0.0064 (8.98E-01) -0.1009 (4.10E-02) PVeD L -0.0952 (5.45E-02) -0.0831 (9.33E-02) 0.1778 (3.14E-04) 0.1906 (1.23E-04) 0.1913 (1.28E-04) 0.1740 (5.55E-04) 0.1445 (4.19E-03) 0.2206 (1.35E-05) -0.0265 (5.92E-01) -0.0423 (3.93E-01) PVeD R -0.0652 (1.88E-01) -0.0498 (3.15E-01) 0.1271 (1.03E-02) 0.0816 (1.03E-01) 0.1056 (3.58E-02) 0.1953 (1.04E-04) 0.1114 (2.77E-02) 0.1497 (3.37E-03) -0.0490 (3.22E-01) -0.0579 (2.42E-01) .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 26  Interaction effects in Figure 7b. 226 227 * Other covariates included age, sex, and education 228 229  Regression analysis in Figure 7d & 7e. 230 231 *Response: MMSE annual change rate 232 ** Other covariates included age, sex, and education 233 PVeD APOE4 PVeD*APOE4 Coefficients 19.64 185.69 -326.88 SE 38.92 44.29 95.09 p- value 0.6141 < 0.0001 0.0006 Response: PiB CL SUVR global PVeD DTI-ALPS Coefficients 4.832 0.294 SE 1.618 0.334 p FDR 0.0066 0.3800 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 27 Supplementary References 234 American Psychiatric Association, D., American Psychiatric Association, D., 2013. 235 Diagnostic and statistical manual of mental disorders: DSM-5. American psychiatric 236 association Washington, DC. 237 Bookheimer, S.Y ., Salat, D.H., Terpstra, M., Ances, B.M., Barch, D.M., Buckner, R.L., 238 Burgess, G.C., Curtiss, S.W., Diaz-Santos, M., Elam, J.S., 2019. The lifespan human 239 connectome project in aging: an overview. Neuroimage 185, 335-348. 240 CHOI, S.-H., NA, D.-L., LEE, B.-H., HAHM, D.-S., JEONG, J.-H., YOON, S.-J., YOO, K.-241 H., HA, C.-K., HAN, I.-W., 2001. Estimating the validity of the Korean version of 242 expanded clinical dementia rating (CDR) scale. Journal of the Korean Neurological 243 Association, 585-591. 244 Huang, S.Y ., Zhang, Y .R., Guo, Y ., Du, J., Ren, P., Wu, B.S., Feng, J.F., Alzheimer's Disease 245 Neuroimaging, I., Cheng, W., Yu, J.T., 2024. Glymphatic system dysfunction predicts 246 amyloid deposition, neurodegeneration, and clinical progression in Alzheimer's disease. 247 Alzheimers Dement 20, 3251-3269. 248 Li, Y ., Wang, L., Zhong, J., Xu, H., Han, Y ., Alzheimer's Disease Neuroimaging, I., Zuo, C., 249 Jiang, J., 2024. Impaired glymphatic function as a biomarker for subjective cognitive 250 decline: An exploratory dual cohort study. Alzheimers Dement 20, 6542-6555. 251 Liu, X., Barisano, G., Shao, X., Jann, K., Ringman, J.M., Lu, H., Arfanakis, K., Caprihan, A., 252 DeCarli, C., Gold, B.T., Maillard, P., Satizabal, C.L., Fadaee, E., Habes, M., Stables, L., 253 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 28 Singh, H., Fischl, B., van der Kouwe, A., Schwab, K., Helmer, K.G., Greenberg, S.M., 254 Wang, D.J.J., Mark, V .C., 2024. Cross-Vendor Test-Retest Validation of Diffusion 255 Tensor Image Analysis along the Perivascular Space (DTI-ALPS) for Evaluating 256 Glymphatic System Function. Aging Dis 15, 1885-1898. 257 McKhann, G.M., Knopman, D.S., Chertkow, H., Hyman, B.T., Jack Jr, C.R., Kawas, C.H., 258 Klunk, W.E., Koroshetz, W.J., Manly, J.J., Mayeux, R., 2011. The diagnosis of 259 dementia due to Alzheimer’s disease: Recommendations from the National Institute on 260 Aging-Alzheimer’s Association workgroups on diagnostic guidelines for Alzheimer's 261 disease. Alzheimer's & Dementia 7, 263-269. 262 Morris, J.C., 1993. The Clinical Dementia Rating (CDR) current version and scoring rules. 263 Neurology 43, 2412-2412-a. 264 Son, S.J., Park, B., Roh, H.W., Kim, N., Jo, Y .H., Lee, E., Lee, K., Lee, H., Ha, J.H., Lee, 265 K.S., 2020. Biobank innovations for chronic cerebrovascular disease with Alzheimer's 266 disease study (BICWALZS) Development of new models and analysis 267 methods/screening studies/platforms. Alzheimer's & Dementia 16, e042259. 268 Taoka, T., Masutani, Y ., Kawai, H., Nakane, T., Matsuoka, K., Yasuno, F., Kishimoto, T., 269 Naganawa, S., 2017. Evaluation of glymphatic system activity with the diffusion MR 270 technique: diffusion tensor image analysis along the perivascular space (DTI-ALPS) in 271 Alzheimer's disease cases. Jpn J Radiol 35, 172-178. 272 Winblad, B., Palmer, K., Kivipelto, M., Jelic, V ., Fratiglioni, L., Wahlund, L.O., Nordberg, 273 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint 29 A., Bäckman, L., Albert, M., Almkvist, O., 2004. Mild cognitive impairment–beyond 274 controversies, towards a consensus: report of the International Working Group on Mild 275 Cognitive Impairment. Journal of internal medicine 256, 240-246. 276 277 .CC-BY-NC-ND 4.0 International licensemade available under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is The copyright holder for this preprintthis version posted April 30, 2025. ; https://doi.org/10.1101/2025.04.28.651021doi: bioRxiv preprint