Sequential associations of myelin dysfunction with Aβ and Tau pathology in Alzheimer's disease: a dynamic myelin perturbation

Abstract

Background

Myelin dysfunction has emerged as a potentially pivotal factor in Alzheimer’s disease (AD); however, its temporal dynamics and relationship with hallmark pathological processes remain incompletely understood. This study aimed to investigate dynamic demyelination and its association with the progression of Aβ and Tau pathology in AD.

Methods

A total of 1,237 participants from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) with 18F-florbetapir (FBP) PET imaging were included. White matter SUVR (WM SUVR), defined as the adjusted FBP signal in white matter, was used as a proxy measure of myelin integrity. Cross-sectional and longitudinal analyses were conducted to examine associations between WM SUVR and AD progression, as well as AD-related biomarkers. Using cortical Aβ accumulation as a temporal reference, we further evaluated two-year changes in cortical Aβ and cumulative Tau PET signals in individuals with and without reduced myelin integrity (WM − vs. WM + uptake).

Results

WM SUVR demonstrated a gradual decline with advancing AD severity. Lower WM SUVR was associated with changes in Aβ_42/40 and pTau₁₈₁ levels in both cerebrospinal fluid (CSF) and plasma. In early-stage AD, the WM − group exhibited a higher two-year rate of cortical Aβ accumulation compared with the WM + uptake group. At comparable levels of cortical amyloid, the WM − group showed significantly higher Tau PET signals in temporal meta-regions of interest and more extensive cortical spread in later stages.

Conclusions

Progressive myelin dysfunction is observed across the course of AD as a notable perturbation within neurodegenerative processes. Early alterations in myelin integrity are associated with subsequent Aβ accumulation, whereas later changes coincide with increased Tau aggregation. These findings support a conceptual framework in which myelin degeneration shows temporally structured associations with AD pathology and may serve as a sensitive marker of vulnerability, particularly in early disease stages.

Supplementary Information

The online version contains supplementary material available at https://doi.org/10.1186/s13195-026-02139-9.

Introduction

Alzheimer’s disease (AD), one of the leading causes of dementia in aging populations worldwide, is pathologically characterized by progressive accumulation of neuritic beta-amyloid (Aβ) plaques and hyperphosphorylated Tau neurofibrillary tangles (NFTs) in the brain [1]. Despite extensive efforts, most therapeutic interventions targeting Aβ plaques removal have yielded limited to disappointing clinical efficacy [2]. Recent neuroimaging studies have increasingly highlighted the frequent occurrence of macrostructural and microstructural white matter (WM) abnormalities in AD, with the loss of myelin integrity emerging as one of the most prominent pathological features [3, 4]. Multiple neuropathological studies have also shown reduced myelin integrity in the AD brain [5–7], suggesting that myelin degeneration may represent a critical component of AD pathophysiology.

Myelin, a multilayered insulating membrane, ensheathes axons to enable rapid and efficient neural conduction [8]. In healthy adults, myelin demonstrates complex plasticity, undergoing continuous renewal and remodeling through ongoing oligodendrocyte generation and turnover [9]. In contrast, the myelin system in AD is profoundly impaired, showing extensive pathological myelin loss [10]. Cerebral accumulations of Aβ and Tau have been found to correlate with decreased myelin levels, particularly in the tracts rich in late-developing lower myelinated fibers [11–13]. Previous studies have suggested that myelin defects might enhance Aβ production by neurons and oligodendrocytes [14–18], while neuroinflammation triggered by myelinated axonopathy may promote Tau hyperphosphorylation [19–21]. However, due to the limited sample size, the independent and joint contributions of demyelination to Aβ and Tau pathologies remain unclear, particularly across different stages of AD progression.

Recently, WM uptake of [18F] florbetapir (FBP), measured by florbetapir positron emission tomography (FBP-PET), has been proposed as a potential proxy for assessing brain myelin in AD [11, 12, 22]. This interpretation is based on the reported affinity of stilbene derivatives for β-sheet structures present not only in fibrillar Aβ but also in specific proteins such as myelin basic protein (MBP) [23–25], thereby enabling the estimation of myelin loss [26, 27]. However, in the context of AD, Aβ-related signal contributions must be appropriately accounted for [11, 12, 22].

Here, we leveraged a large cohort from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) with cross-sectional and longitudinal data to investigate dynamic alterations in myelin content across the AD spectrum, as assessed by FBP-PET. The present study provides novel perspectives on the dynamic relationships between myelin homeostasis and AD-related pathology, including Aβ deposition and NFTs, and offers integrated insights into myelin-associated processes underlying AD pathology.

Methods

Ethical approval

All research involving human participants adhered to the ethical principles of the Institutional and National Research Committees and complied with the Declaration of Helsinki (1964) and its subsequent revisions or equivalent ethical standards. For the ADNI cohort, written informed consent was obtained from all participants, their study partners, or legally authorized representatives prior to the conduct of any study-related procedures. This study involved only secondary analysis of ADNI data; no local patient data were included. Further details on ADNI’s ethical oversight and study protocol are available at http://www.adni-info.org.

ADNI participants

Data used in the preparation of this article were obtained from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database (adni.loni.usc.edu). The ADNI was launched in 2003 as a public-private partnership, led by Principal Investigator Michael W. Weiner, MD. The primary goal of ADNI has been to test whether serial magnetic resonance imaging (MRI), positron emission tomography (PET), other biological markers, and clinical and neuropsychological assessment can be combined to measure the progression of mild cognitive impairment (MCI) and early Alzheimer's disease (AD). In this study, participants who had FBP-PET, T1-weighted (T1w) MRI and clinical diagnosis were included. Some participants had data from concurrent CSF and plasma examinations (Aβ₄₂, Aβ₄₀, and pTau₁₈₁), [18F] flortaucipir (FTP) PET, Diffusion-weighted Imaging (DWI), and Fluid Attenuated Inversion Recovery (FLAIR) scans. Researchers at each ADNI study site involved in the ADNI project obtained ethical approval. All participants provided written informed consent.

Acquisition and processing of PET imaging data

Details on descriptions of PET imaging procedures were obtained from the processed results provided by UC Berkeley and the Lawrence Berkeley National Laboratory, which are publicly available in the ADNI database. Amyloid PET images using FBP were fully pre-processed by ADNI, including frame realignment and averaging, linear transformation to correct head positioning, standardization of voxel size, and smoothing to a uniform resolution of 6 mm. T1w MRI was first segmented and parcellated with Freesurfer (FS) using the Desikan-Killiany atlas, and then coregistered to the amyloid PET image with SPM. Positive amyloid was defined as a cortical-cerebellar average standardized uptake value ratio (SUVR) ≥ 1.11 [28]. Cortical FBP SUVR was defined as integrated values from meta-regions of the frontal, anterior and posterior cingulate, lateral parietal, and lateral temporal regions. The mean cortical SUVR values (mean ± standard error [SE]) were 1.11 ± 0.005 for CN, 1.20 ± 0.013 for MCI, and 1.42 ± 0.007 for dementia. Centiloid (CL) values were derived from cortical SUVR using the equation: CL = 188.22 × cortical SUVR − 189.16. Tau PET images using FTP were also smoothed to a uniform resolution of 6 mm, then coregistered in SPM to the bias-corrected T1w created by FS. The intensity was normalized by the inferior cerebellar gray matter (GM), and the mean Tau uptake (SUVR) within each Desikan-Killiany region and the temporal-meta regions of interest (MetaROI) (including the entorhinal cortex, amygdala, fusiform gyrus, inferior temporal gyrus, and middle temporal gyrus) was calculated. The specific processing methods are available at https://adni.loni.usc.edu/.

Adjustment of FBP-PET SUVR in WM

The primary variable analyzed in this study is the FBP-PET SUVR in WM, serving as a measure of myelination. To reduce the impact of FBP binding to Aβ plaques on FBP binding in WM, we adjusted the raw FBP SUVR in WM as previously described [11, 22]. In brief, in the CN group, which included both Aβ + and Aβ- participants, we used the integrated cortical FBP SUVR as a predictor and FBP SUVR in WM as the dependent variable to fit a linear regression model. We then adjusted the FBP SUVR in WM by subtracting the predicted SUVR (using the estimated linear model) to account for the cortical FBP-PET signal. To quantify the deviation of FBP SUVR in WM in Aβ + participants of the CI group compared to the CN group, we calculated the z-score of adjusted FBP SUVR in WM using the CN group as a reference, resulting in the final myelination measure WM SUVR, also described as Cerebral WM SUVR (Z-score). In this study, the median WM SUVR was used as the cutoff to define WM + and WM − uptake, representing higher and lower white matter uptake (Fig. S1), respectively, rather than a biologically defined positive or negative status.

In sensitivity analysis, the WM calibration procedure was repeated using only Aβ − CN participants as the reference group. The resulting z-score metric was termed Aβ − CN–recalibrated WM SUVR. A two-component Gaussian mixture model was fitted to the distribution of Aβ − CN–recalibrated WM SUVR, and the intersection point between the two fitted Gaussian components was used as a data-driven threshold for WM classification. In the present analysis, this threshold was estimated at − 1.987 (Fig. S2A).

Processing of diffusion tensor imaging (DTI) data

The raw DWI data were preprocessed using PreQual for denoising, artifact correction, and signal dropout imputation. T1w images were processed using SLANT-TICV for GM segmentation and MaCRUISE for cortical surface parcellation. Then these preprocessed DWI data were registered to the Johns Hopkins Eve Type 3 white matter atlas within the brain mask of the T1w image, through a combined use of FSL and ANTs, with diffusion scalar metrics calculated for each region of interest (ROI), including fractional anisotropy (FA), mean diffusivity (MD), radial diffusivity (RD), and axial diffusivity (AxD). The processed data were acquired from publicly available data in ADNI from ADSP (Alzheimer’s Disease Sequencing Project) Phenotype Harmonization Consortium. For each metric (RD, FA, MD, AxD), the median values from the left and right hemispheres were averaged to create four separate composite measures.

Total white matter hyperintensities (WMH) volume data acquisition

The WMH volume data were available in the ADSP shared dataset in ADNI, and were calculated from FLAIR data by FS processing. WMH volume was log-transformed for statistical model due to its skewed distribution.

GM volume data acquisition

The GM volume data were extracted from the ADSP-analyzed T1w MRI data using FS processing, which is publicly available in ADNI. We standardized the data employed in the comparisons by calculating the z-score of the ratio between the volumes of ROIs (grey matter regions defined by the FreeSurfer Desikan–Killiany cortical atlas and subcortical segmentation) and intracranial volume (ICV), a commonly used approach to compensate for head size variability [29, 30]. Neurodegeneration status (N + and N −) were defined according to residualized hippocampal volume (rHCV) adjusted for ICV (Fig. S3), as previously reported [31].

Acquisition of CSF data

CSF Aβ₄₂ and Aβ₄₀ were obtained from the ADNI database, provided by the University of Pennsylvania using 2D-UPLC-tandem mass spectrometry. CSF pTau₁₈₁ was achieved using the fully automated Roche Elecsys and Cobas e 601 immunoassay analyzer system, with detailed methods available at https://adni.loni.usc.edu/. In this study, CSF pTau₁₈₁ ≥ 23 pg/mL was used as the cutoff to define T + versus T–, as described previously [31].

Acquisition of plasma data

Plasma Aβ₄₂ and Aβ₄₀ were measured using a technique that combines immunoprecipitation and liquid chromatography-tandem mass spectrometry (LC–MS/MS) by Bateman Lab. Details of methods are available at https://adni.loni.usc.edu/.

Plasma pTau₁₈₁ was analyzed using the Single Molecule array (Simoa) technique, with an in-house assay developed by the Clinical Neurochemistry Laboratory at the University of Gothenburg, Sweden, as described in their previous study [32].

Cerebrovascular risk factors

Principal Component 1 of Cerebrovascular Risk Factors (CVR-PC1) and number of MRI infarcts were included as cerebrovascular risk factors in the analyses. CVR-PC1 was computed by the first principal component of heart disease, stroke history, hypertension, diabetes, and BMI in ADNI cohort and obtained from the ADSP dataset. Detailed methods have been described previously [33]. The number of MRI infarcts was derived from imaging data processed by the Charles DeCarli team at UC Davis, using standardized visual ratings of infarcts ≥ 3 mm based on T1- and T2-weighted MRI scans. This method has demonstrated excellent inter-rater reliability (κ = 0.73–0.90). The details are available at https://adni.loni.usc.edu/.

Data analysis

All data processing and analysis were performed using R version 4.4.1. Correlation analysis was generally conducted using linear correlation methods, with results reported as p-values and β-values. For Table 1 and Table S1−4, P-values were calculated using Chi-squared tests for categorical variables and t-tests for continuous variables. General group differences in medians were assessed using the Mann–Whitney U test. A/T and A/T/N group comparisons used median comparisons via the Kruskal–Wallis test. For multiple comparisons between groups, the Kruskal–Wallis test followed by Dunn's test was applied, with FDR correction. Multiple spatial comparisons of DTI metrics were performed using linear models with FDR correction. Mediation analysis was conducted using the mediation package to examine whether the association between the independent variable and the outcome variable is statistically mediated by the hypothesized mediator, adjusting for age, sex, APOE4 status, and cerebrovascular risk factors. All mediation results, including estimated direct and indirect effects, should be interpreted as statistical associations, particularly in analyses based on cross-sectional measures, where key assumptions required for causal mediation inference cannot be fully verified.

Table 1.

Characteristics of participants in ADNI cohort

Characteristic at baseline	WM+uptake	WM-uptake	p
n(%)	584 (47.2)	653 (52.8)
Diagnosis, n(%)			<0.001
CN	320 (54.8)	169 (25.9)
CI	264 (45.2)	484 (74.1)
Age, years	72.77 (7.49)	74.27 (7.83)	0.001
Sex, n(%)			0.315
Male	290 (49.7)	344 (52.7)
Female	294 (50.3)	309 (47.3)
APOE ε4 status, n(%)			<0.001
Non-carriers	395 (67.6)	320 (49.0)
Carriers	189 (32.4)	333 (51.0)
Education, years	16.40 (2.57)	16.11 (2.70)	0.056
CVR-PC1	0.02 (1.69)	-0.52 (1.31)	<0.001
Number of MRI infarcts	0.10 (0.52)	0.19 (1.09)	0.139
FBP-PET
Duration of follow up, year	3.42 (3.24)	2.46 (2.87)	<0.001
Baseline cortical SUVR	1.14 (0.23)	1.26 (0.26)	<0.001
741 participants with consurrent CSF Aβ42, Aβ40
n(%)	326 (44.0)	415 (56.0)
CSFAβ42/Aβ40	0.16 (0.05)	0.13 (0.05)	<0.001
947 participants with concurrent CSF pTau181
n(%)	444 (46.9)	503 (53.1)
CSF pTau181, pg/ml	22.55 (11.10)	30.07 (15.27)	<0.001
319 participants with concurrent plasma Aβ42, Aβ40
n(%)	157 (49.2)	162 (50.8)
Plasma42/Aβ40	0.13 (0.01)	0.12 (0.01)	<0.001
938 participants with concurrent plasma pTau181
n(%)	407 (43.4)	531 (56.6)
Plasma pTau181, pg/ml	17.22 (25.09)	19.98 (11.35)	0.025
503 participants with concurrent DWI
n(%)	292 (58.1)	211 (41.9)
Temporal-MetaROI SUVR	1.26 (0.23)	1.60 (0.75)	<0.001
450 participants with concurrent DWI
n(%)	239 (53.1)	211 (46.9)
WM FA Z-score	0.11 (1.03)	-0.24 (1.03)	<0.001
WM MD Z-score	-0.19 (0.96)	0.25 (1.06)	<0.001
WM RD Z-score	-0.18 (0.97)	0.25 (1.07)	<0.001
WM AxD Z-score	-0.20 (0.94)	0.19 (1.05)	<0.001
1114 participants with concurrent FLAIR
n(%)	557 (50.0)	557 (50.0)
WMH volume cm3	0.93 (1.30)	144 (1.25)	<0.001
1191 participants with concurrent TI w MRI (GM volume)
n(%)	581 (48.8)	610 (51.2)

*Continuous variables are presented as mean ± standard deviation (SD)

*Baseline refers to the time point of the first FBP-PET scan

Notably, a linear mixed-effects model was used to examine the relationship between WM SUVR and follow-up time, including group (CN/Aβ-, CI/Aβ-, Aβ +) as a fixed effect and individual variability as a random effect, adjusting for age, sex, APOE4 status. Cerebrovascular risk factors and WMH volume were not included due to insufficient data in the longitudinal cohort. The model was specified as follows:

• Model 1: WM SUVR ~ Time * Group + APOE4 + Age + Sex + (1 | Individual)

Generalized linear models (GLMs) were used to assess the relationships between CSF and plasma biomarkers and WM SUVR, adjusting for age, sex, APOE4 status, number of MRI infarcts, CVR-PC1, and WMH volume.

• Model 2: Biomarkers ~ WM SUVR + Age + Sex + APOE4 + Number of MRI Infarcts + CVR-PC1 + WMH Volume

GLMs were used to assess CSF and PET biomarkers across CN/CI and WM groups, adjusting for age, sex, APOE4, number of MRI infarcts, CVR, and WMH volume. Pairwise comparisons were based on estimated marginal means with FDR correction.

• Model 3: Biomarkers ~ Group + Age + Sex + APOE4 + Number of MRI Infarcts + CVR-PC1 + WMH Volume

Model 3 was also used for multiple spatial comparisons of Aβ PET SUVR slope, Tau PET SUVR, and GM volume between WM groups, adjusting for covariates with FDR-corrected. Additionally, within each WM group, LOESS (locally weighted scatterplot smoothing) modeling was used to estimate the relationship between the predictor and covariate-adjusted response variables. The Kolmogorov–Smirnov test (KS) was then used to assess whether the distributions of the fitted values differed significantly between groups. The linear relationships between the predictor and response variables were estimated using Model 2.

The interaction model was developed using a linear regression framework to investigate the association between Temporal MetaROI Tau SUVR and cortical FBP SUVR, WM group, and their interaction, with adjustment for multiple covariates.

• Model 4: Temporal MetaROI Tau SUVR ~ Cortical FBP SUVR × WM group + Age + Sex + APOE4 + MRI Infarct Number + CVR-PC1 + WMH Volume

Additionally, we conducted subsampling-based sensitivity analyses using 100 iterations of random subsampling. In each iteration, Aβ + and Aβ − groups, as well as WM + and WM − uptake groups, were randomly subsampled to achieve balanced group sizes, and Model 4 was refitted in each subsample.

In sensitivity analysis, Pearson correlation analysis was performed to assess the association between Aβ − CN-recalibrated WM SUVR and cortical FBP SUVR. The distributions of both measures were examined using density plots. Concordance between WM classification and amyloid status was assessed using cross-tabulation and Cohen’s kappa statistic.

Linear regression interaction models were fitted using both the original CN-calibrated WM classification and the Aβ − CN–recalibrated WM classification. Separate models were constructed for cortical SUVR 2-year slope and Temporal MetaROI Tau SUVR. Each model included clinical diagnosis, WM group, and their interaction term, with adjustment for APOE ε4 status, age, sex, WMH volume, MRI infarct number, and CVR-PC1.

• Model 5: Cortical SUVR 2-year slope ~ Clinical diagnosis × WM group + APOE ε4 status + Age + Sex + WMH volume + MRI Infarct Number + CVR-PC1

• Model 6: Temporal MetaROI Tau SUVR ~ Clinical diagnosis × WM group + APOE ε4 status + Age + Sex + WMH volume + MRI Infarct Number + CVR-PC1

The significance of model interaction terms was evaluated using analysis of variance (ANOVA). Estimated marginal means were derived from each fitted model. Pairwise comparisons between WM groups were performed within each diagnostic category. To further assess stage-related differences within each WM group, estimated marginal means across clinical diagnosis stages were evaluated using joint tests.

Results

Demographics

Demographical characteristics were summarized in Table 1. This study included 1,237 participants from the ADNI database with FBP PET scans. Based on a predefined WM SUVR threshold (Fig. S1), 47.2% were classified as WM + uptake, i.e. without myelin damage, and 52.8% as WM- uptake, i.e. with myelin damage, at baseline. WM + uptake individuals were more likely to be cognitively normal (CN: 54.8% vs. 25.9%) and less likely to be cognitively impaired (CI: 45.2% vs. 74.1%) compared to WM- uptake (p < 0.001). The WM + uptake group was slightly younger (mean age = 72.77 ± 7.49 vs. 74.27 ± 7.83 years, p = 0.001). APOE ε4 carriers were more frequent in the WM- uptake group (51.0% vs. 32.4%, p < 0.001). WM + uptake individuals exhibited significantly lower vascular burden, as reflected by a higher CVR-PC1 score (mean 0.02 vs. − 0.52, p < 0.001). No significant difference was found in sex distribution, education years, and number of MRI infarcts (all p > 0.05). Regarding FBP PET, WM + uptake individuals had a significantly longer duration of follow-up (mean 3.42 vs. 2.46 years, p < 0.001) and lower baseline cortical SUVR (mean 1.14 vs. 1.26, p < 0.001). During the FBP-PET follow-up, concurrent data for CSF Aβ₄₂/₄₀, CSF pTau₁₈₁, plasma Aβ₄₂/₄₀, plasma pTau₁₈₁, FTP-PET, DWI, FLAIR (WMH volume), and T1w MRI (GM volume) were also comprehensively summarized in Table 1.

WM SUVR as a candidate marker of myelin integrity

DTI is commonly used to assess the WM microstructural abnormalities and is sensitive to myelin-related changes [34]. Here, we found that lower WM SUVR was associated with reduced FA (estimate = 0.222 [95% confidence interval: 0.138 to 0.307], p < 0.001) and increased RD (estimate = −0.265 [95% confidence interval: −0.348 to −0.182], p < 0.001), MD (estimate = −0.262 [95% confidence interval: −0.344 to −0.180, p < 0.001), and AxD (estimate = −0.265 [95% confidence interval: −0.348 to −0.182], p < 0.001) (Fig. 1). While decreased FA and increased MD and RD are consistent with reduced myelin integrity, the observed increase in AxD may reflect axonal degeneration or neuroinflammation, rather than myelin loss per se [35]. Additionally, WM SUVR decreases were associated with increased WMH volume (estimate = −0.349 [95% confidence interval, −0.418 to −0.280], p < 0.001) (Fig. S4), consistent with the hypothesis that WMH indicates demyelinating lesions. These results suggest that WM SUVR may reflect aspects of myelin integrity, with lower values indicating myelin disruption.

Fig. 1.

Evaluation of myelin structural integrity using WM SUVR consistent with DTI Data. A-D Cerebral WM SUVR shows a significant correlation with the FA (A), RD (B), MD (C), and AxD (D) of cerebral WM. The legend displays β and p values, with red representing the Aβ + group and green representing the Aβ- group

Progressive disruption of myelin integrity in AD

A stepwise WM SUVR decline was seen across clinical groups: dementia < MCI < CN, with estimate = 0.776 [95% confidence interval, 0.619 to 0.931], p < 0.001, between dementia and MCI, and estimate = 0.495 [95% confidence interval, 0.380 to 0.613], p < 0.001, between MCI and CN, respectively (Fig. 2A). Furthermore, APOE4 carrier exhibited decreased WM SUVR compared to APOE4 non-carrier (estimate = 0.467 [95% confidence interval, 0.349 to 0.584], p < 0.001) (Fig. 2B). Across different A/T profile, WM SUVR was also reduced in A +/T- and A +/T + groups vs. A-/T-, with the lowest in A +/T + (Fig. 2C). Notably, the A–/T+ group showed WM SUVR levels higher than in the A +/T- group (p = 0.007) and comparable to that in the A-/T- group (p = 0.064) (Fig. 2C). Longitudinal mixed-effects models (Model 1, see Methods) monitoring myelin status over 5–10 years confirmed lower WM SUVR in Aβ + than in CN/Aβ- (estimate = −0.346 [95% confidence interval, −0.683 to −0.008], p = 0.046) with a significant group × time interaction indicating a faster decline in Aβ + over time (estimate = −0.063 [95% confidence interval, −0.093 to −0.032], p < 0.001) (Fig. 2D, left). Cross-sectional analysis showed a negative association (Aβ-:estimate = −0.023 [95% confidence interval, −0.036 to −0.009], p = 0.001; Aβ + :estimate = −0.035 [95% confidence interval, −0.060 to −0.010], p = 0.006) between decreased WM SUVR and older ages in CN, suggesting that age remains a significant factor in accelerated myelin degeneration or a loss of the ability of remyelination [36] (Fig. 2D, right). These findings collectively suggest a progressive disruption of myelin integrity during AD progression, which may be due to abnormal Aβ and Tau pathology, as well as APOE4 allele.

Fig. 2.

Progressive disruption of myelin integrity in AD. A A comparison of WM SUVR between the CN, MCI, and dementia groups across the AD spectrum using Mann–Whitney U test. B A comparison of WM SUVR between APOE4 carriers and non-carriers using Mann–Whitney U test. C A comparison of WM SUVR differences between groups based on different A/T pathological stages using the Kruskal–Wallis test, followed by post-hoc Dunn's test with FDR-adjusted p-values. D Left: Fitted lines were derived from a linear mixed-effects model comparing CN/Aβ −, CI/Aβ −, and Aβ + groups over time, adjusting for age, sex, and APOE4 status. Curvature reflects covariate-adjusted effects with 95% confidence intervals. The dashed lines indicate the mean WM SUVR for the CN, MCI, and dementia groups. Right: Linear regression lines depicting the association between age and WM SUVR in Aβ + and Aβ − CN participants. The legend displays β and p values, with red representing the Aβ + group and green representing the Aβ- group

Association of myelin dysfunction with CSF and plasma biomarkers

GLMs results (Model 2, see Methods) showed that lower WM SUVR was associated with lower Aβ_42/40 levels and higher pTau₁₈₁ levels in both CSF and plasma (Fig. 3A-B), the changes indicative of cognitive impairment. Indeed, in the cognitive impairment (CI, referring to MCI or dementia) population, the WM- uptake group exhibited higher cortical Aβ SUVR and decreased CSF Aβ_42/40 compared to the WM + uptake group (Fig. 3C), along with significantly elevated Tau PET and CSF pTau₁₈₁ levels (Fig. 3D). In contrast, among CN individuals, these group differences were attenuated substantially (Fig. 3C-D). These findings highlight a significant association between myelin dysfunction and Aβ and Tau pathology, supporting the notion that demyelination may constitute a critical perturbation within the AD pathological landscape.

Fig. 3.

Association between demyelination and Aβ/Tau pathology. A-B GLMs were used to assess the relationships between CSF Aβ_42/40, pTau₁₈₁, and WM SUVR (A), as well as plasma Aβ_42/40, pTau₁₈₁, and WM SUVR (B), adjusting for age, sex, APOE4 status, number of MRI infarcts, CVR-PC1, and WMH volume. C The differences in Aβ pathology biomarkers among the four groups (WM + uptake/CN, WM- uptake/CN, WM + uptake/CI, and WM- uptake/CI), including CSF Aβ_42/40 and Aβ PET cortical SUVR. D The differences in Aβ pathology biomarkers among the four groups include CSF pTau₁₈₁ and Tau PET temporal-metaROI SUVR. Both (C) and (D) were analyzed using GLMs, adjusting for age, sex, APOE4 status, number of MRI infarcts, CVR-PC1, and WMH volume. Pairwise group differences were assessed using estimated marginal means with FDR correction

Associations between demyelination and early cortical Aβ accumulation in AD

To evaluate the relationship of demyelination on Aβ deposition, a relatively early event in AD pathogenesis, from a temporal perspective, we analyzed all cases with two years of follow-up data from the ADNI database. Our results (Model 2, see Methods) revealed that lower baseline WM SUVR was associated with faster rates of cortical Aβ accumulation (estimate = −0.014[95% confidence interval, −0.021 to −0.007], p < 0.001) (Fig. 4A). The LOESS-fitted curve also indicated that, prior to the plateau of Aβ accumulation, the WM- uptake group exhibited a higher two-year cumulative slope of cortical PBP SUVR compared with the WM + uptake group (Fig. 4B). The KS test confirmed a significant difference in the distributions of the LOESS-fitted values between the WM- and WM + uptake groups (D = 0.658, p < 0.001) (Fig. 4B), suggesting that lower myelin integrity is associated with earlier cortical Aβ accumulation. Fitting the clinical progression based on the average cortical SUVR values from the CN, MCI, and dementia groups showed that the curves for the WM − and WM + uptake groups differed prior to the clinical diagnosis of dementia [37], particularly during the early stages of AD (Fig. 4B). The brain region differential comparison (Model 3, see Methods) also showed that during the early stages of AD, the WM- uptake group exhibited extensive cortical and subcortical regions with higher 2-year Aβ accumulation compared to the reference (Fig. 4C, Fig. S5A). Mediation analysis incorporating longitudinal outcomes suggested that baseline cortical SUVR did not statistically mediate the association between WM SUVR and subsequent Aβ accumulation, while a significant direct association was observed (Fig. S6A), consistent with a potential association between myelin status and Aβ deposition. Thus, these results are consistent with a potential relationship between demyelination and a leftward shift of the Aβ temporal accumulation curve.

Fig. 4.

Associations between demyelination and early cortical Aβ accumulation in AD. A The 2-year slope of Aβ PET cortical SUVR shows a linear correlation with baseline WM SUVR, adjusting for age, sex, APOE4 status, number of MRI infarcts, CVR-PC1, and WMH volume. Dashed lines indicate the mean WM SUVR values for the CN, MCI, and dementia groups (B) A fitting curve of the 2-year slope (residuals adjusted for age, sex, APOE4 status, number of MRI infarcts, CVR-PC1, and WMH volume) is plotted with baseline Aβ PET cortical SUVR on the x-axis, where the red and blue fitting lines represent the WM- uptake and WM + uptake groups, respectively, with the dashed line representing the mean Aβ PET cortical SUVR for CN, MCI, and dementia stages. C At different stages of Aβ accumulation (defined by cortical FBP SUVR), significant differences in the 2-year slope of FBP SUVR across cortical ROIs were shown between groups, adjusting for age, sex, APOE4 status, number of MRI infarcts, CVR-PC1, and WMH volume. D At different stages of Aβ accumulation (defined by cortical FBP SUVR), RD values in the WM + uptake and WM- uptake groups show various changes in WM ROI regions. In (C) and (D), the cortical FBP SUVR ≤ 1.10 stage in the WM + uptake group was used as the reference group for multiple spatial comparisons. After FDR correction, the p-value was < 0.05

Subsequently, we conducted a spatial comparison of WM regions based on DTI metrics, focusing on RD, which is sensitive to white matter microstructural changes and often reflects myelin-related properties [38], to identify the myelin areas most closely associated with cortical Aβ accumulation. As Aβ accumulates, the RD in both superficial and deep white matter shows an increasing trend. However, under the same Aβ deposition, the WM- uptake group exhibited more widespread and severe increases in RD (Fig. 4D). In the early stages, more sensitive regions included deep WM areas such as the superior corona radiata, posterior thalamic radiation, superior fronto-occipital fasciculus, anterior limb of internal capsule, and body of corpus callosum, as well as superficial white matter areas such as the precentral, superior/middle occipital, and superior parietal regions (Fig. 4D). These results suggest that lower myelin integrity in both long-range and local conduction pathways is associated with earlier and more extensive cortical Aβ accumulation in demyelinated brains.

Associations between demyelination and increased NFTs in mid-to-late Aβ phases

Mediation analysis showed that the association between WM SUVR and Temporal MetaROI Tau SUVR was partially statistically mediated by cortical FBP SUVR, while also showing a direct association with Temporal MetaROI Tau SUVR independent of cortical FBP SUVR (Fig. S6B). This observation indicates that the relationship between myelin integrity and Tau pathology is not solely explained by Aβ accumulation, as suggested by the traditional Aβ-Tau interaction hypothesis. Using Aβ accumulation as a timeline, we observed significant differences in Tau SUVR values in the temporal MetaROI regions of the WM- uptake group specifically during the progression from MCI to dementia (Fig. 5A). The significant interaction between cortical FBP SUVR and WM group (estimate = 0.870 [95% confidence interval, 0.465 to 1.276], p < 0.001) (Model 4, see Methods) indicated that the relationship between cortical FBP SUVR and Temporal MetaROI Tau SUVR differs significantly between the groups, with a stronger effect in the WM- uptake group (Fig. 5A). The LOESS fitted curve of the CSF biomarker data also highlighted that the pTau₁₈₁ in the WM- uptake group increased markedly once Aβ₄₂ levels fell below the normative threshold (Fig. 5B). Furthermore, the KS test results revealed a significant difference in the distributions of the LOESS fitted values between the WM- uptake and WM + uptake groups (D = 0.606, p < 0.001) (Fig. 5B). Those observations suggest that the association of demyelination with increased Tau pathology is most pronounced in the middle and late phases of Aβ pathology progression.

Fig. 5.

Associations between demyelination and increased NFTs in mid-to-late Aβ phases. A The differences in the changes of Tau PET temporal-metaROI SUVR relative to Aβ PET cortical SUVR were observed between the WM + uptake group and WM- uptake group, adjusted for age, sex, APOE4 status, number of MRI infarcts, CVR-PC1, and WMH volume. The dashed lines represent the mean Aβ PET cortical SUVR for CN, MCI, and dementia stages. B The differences in the changes of CSF pTau₁₈₁ (residuals adjusted for age, sex, APOE4 status, number of MRI infarcts, CVR-PC1, and WMH volume) relative to Aβ₄₂ were observed between the two groups through fitted curves. C At different stages of Aβ accumulation, Tau PET SUVR in cortical ROI regions exhibits differential changes between the WM + uptake group and WM- uptake group. Using the cortical SUVR ≤ 1.10 stage in the WM + uptake group as the reference, multiple spatial comparisons with other groups were conducted, adjusting for age, sex, APOE4 status, number of MRI infarcts, CVR-PC1, and WMH volume. After FDR correction, the p-value was < 0.05

The brain region differential maps (Model 3, see Methods) showed that, during the first stage of Aβ accumulation (cortical FBP SUVR ≤ 1.10, a sub-threshold range within the Aβ-negative group), there was no significant difference in Tau-PET brain regions between the WM- uptake and WM + uptake groups (Fig. 5C). Additionally, Braak I-II regions, such as the entorhinal cortex and hippocampus, did not exhibit noticeable differences (Fig. 5C, Fig. S5B). In the second stage of Aβ accumulation (1.10 < cortical FBP SUVR ≤ 1.42), the WM- uptake group exhibited relatively faster increases in Braak III regions, with even faster progression compared to the third stage WM + uptake group in certain brain areas, such as the temporal lobe and parts of the parietal lobe (Fig. 5C). In the third stage of Aβ accumulation (cortical FBP SUVR > 1.42), the brain region progression difference between the WM- uptake and WM + uptake groups was most pronounced, with the entorhinal cortex, temporal lobe, parietal lobe, and frontal lobe showing the most severe NFTs aggregation in the WM- uptake group (Fig. 5C). The WM- uptake group also showed a comprehensive RD increasing in WM during this period (Fig. 4D). In addition, the brain regional volume differences calculated from T1-weighted MRI (Model 3, see Methods) aligned with the Tau-PET results (Fig. 6), supporting that Tau pathology likely drives cortical atrophy. These results suggest that lower myelin integrity is associated with greater cortical NFTs aggregation and propagation during AD progression, which may reflect increased vulnerability to cognitive decline.

Fig. 6.

Associations between demyelination and GM atrophy in the progression of AD. During different stages of Aβ accumulation (defined by cortical FBP SUVR), GM volume in cortical and subcortical ROI regions shows distinct changes between the WM + uptake group and the WM- uptake group. Using the cortical SUVR ≤ 1.10 stage in the WM + uptake group as the reference, multiple spatial comparisons with other groups were conducted, adjusting for age, sex, APOE4 status, number of MRI infarcts, CVR-PC1, and WMH volume. After FDR correction, the p-value was < 0.05

Discussion

WM SUVR, which is influenced in part by MBP-related binding and myelin content, has been utilized previously in both multiple sclerosis and AD [22, 39], as a proxy measure of WM integrity for the present analyses. In this study, “myelin integrity” refers specifically to myelin content and sheath status [40], rather than broader microstructural properties inferred from DTI metrics such as FA [41]. Building on prior work [22, 42], we examined the relationship between WM SUVR and FLAIR-based WMH and DTI metrics, and observed consistent patterns across these modalities. We acknowledge that DTI metrics and WMH are influenced by heterogeneous microstructural and vascular processes; however, they may provide convergent, albeit non-specific support for myelin-related alterations within white matter. The association between lower WM SUVR and increased AxD may reflect the combined effects of reduced myelin content and disruption of sheath integrity. Overall, these findings support the use of WM SUVR as a proxy measure, enriched in myelin-related signal, for assessing myelin integrity in the human brain.

Through cross-sectional and longitudinal analyses across the AD spectrum, we observed a progressive decline in myelin integrity with AD progression, consistent with previous studies [12, 22]. While we observed an overall decline in myelin integrity across the AD spectrum, this pattern is not uniform across all subgroups. Within the A/T classification, the A − T + group—often considered representative of non-AD conditions such as suspected non-Alzheimer’s pathology (SNAP) or primary age-related tauopathy (PART)—does not show a clear reduction in WM SUVR. However, when neurodegeneration is taken into account (A/T/N framework), a different pattern emerges. The A − T + N + group exhibits significantly lower WM SUVR compared with the A − T − N − group, and even lower than the A + T − N − group (Fig. S7A). This suggests that in SNAP- or PART-like conditions, myelin alterations become evident primarily when neurodegeneration is present, indicating that myelin dysfunction may reflect a broader vulnerability associated with neurodegenerative processes rather than Tau pathology alone. At the same time, myelin integrity appears to be sensitive to Aβ burden. Across all T/N strata, A + groups consistently show lower WM SUVR than A − groups, with the greatest reduction observed in the A + T + N + group (Fig. S7A–C). Together, these findings suggest a dual pattern: myelin alterations may represent a general marker of neurodegenerative vulnerability, which is further exacerbated in the presence of Aβ-related pathology, particularly when multiple AD-related processes co-occur.

Within this vulnerability framework, we examined how myelin integrity interacts with the temporal dynamics of Aβ and Tau pathology across disease stages. Using a unified temporal framework, we identified vulnerable time windows during which Aβ and Tau are more susceptible to myelin dysfunction, addressing limitations of prior studies that focused on isolated pathological events [11, 12, 22, 43]. During the early Aβ accumulation phase, lower myelin integrity was associated with a higher rate of Aβ accumulation within two years. This pattern may reflect a leftward shift along the Aβ accumulation trajectory, indicating a shorter time interval to reach the plateau phase. In this context, reduced myelin integrity may be linked to accelerated progression of Aβ pathology and earlier attainment of the plateau stage, highlighting inter-individual variability in the timing of Aβ progression.

Importantly, in the context of reduced myelin integrity (WM- uptake), Tau aggregation and propagation were observed prior to the Aβ plateau, whereas under relatively preserved myelin conditions (WM + uptake), Tau spreading typically followed the plateau. These findings suggest that demyelination is associated with an earlier onset of Tau pathology, and may relate to inter-individual variability in the timing of transition from early Aβ positivity to more advanced pathological stages [44]. Notably, increased Tau signal elevations were not evident during the Aβ − stage, even in vulnerable regions. The association between myelin integrity and Tau pathology emerged only with increasing Aβ burden, indicating a stage-dependent relationship. Together, these results support a temporally structured pattern in which reduced myelin integrity is associated with faster Aβ accumulation during the early phase and earlier Tau emergence at higher levels of Aβ. Notably, this overall pattern remained robust across alternative WM calibration strategies. When calibration was restricted to biologically defined Aβ − CN participants, WM-related differences in both amyloid accumulation and Tau pathology became detectable at earlier disease stages while preserving the same sequential pattern. These findings indicate that the principal conclusions are not dependent on calibration strategy and suggest that an amyloid-negative reference may increase sensitivity to early pathological deviations (Fig. S8).

While several experimental studies have suggested that oligodendrocyte abnormalities and myelin dysfunction may contribute to Aβ accumulation [14–18], our data are observational in nature and do not allow inference of a causal relationship between these processes. Nonetheless, lower myelin integrity was associated with faster cortical Aβ accumulation (Fig. 4A), while in Aβ-positive individuals, higher cortical Aβ was in turn associated with a greater rate of WM SUVR decline (Fig. S9), suggesting a potential bidirectional relationship, possibly driven by shared underlying processes. Consistent with this, demyelination was associated with aging and APOE ε4 status (Fig. 2B, D), both of which are linked to oligodendrocyte dysfunction and reduced myelin maintenance [45, 46]. In contrast, the relationship between myelin integrity and Tau pathology was stage-dependent, emerging only in the presence of Aβ and becoming more apparent after the plateau phase. This is consistent with models in which Aβ deposition facilitates Tau propagation from the entorhinal cortex to the neocortex [47]. The Aβ plateau may therefore represent a transition to a later stage of disease, marked by the emergence of myelin–Tau associations and coordinated structural degeneration. Early myelin loss may shift this transition earlier, lowering the threshold for myelin–Tau coupling.

Rather than implying a direct mechanistic coupling, we conceptualize myelin dysfunction as a perturbation of the brain’s stability landscape. From this perspective, AD reflects a progressive loss of brain stability under accumulating Aβ and Tau pathology, rather than a strictly linear cascade driven by either process alone. Within this framework, myelin dysfunction may represent an amplification of underlying structural vulnerability. At the individual level, myelin integrity may serve as an index of resilience to pathological stress, capturing variability in the capacity to maintain structural stability despite comparable Aβ and Tau burden. In this sense, myelin integrity extends the A/T/N framework by introducing a complementary, non-specific dimension of system-level vulnerability, reflecting the brain’s structural robustness in the face of neurodegenerative processes. Consistent with this framework, the WM classification recalibrated using Aβ − CN identified a subgroup with preserved myelin integrity and slower pathological progression (Fig. S8). Importantly, although recalibrated WM classification showed substantial concordance with amyloid status (κ = 0.727), a considerable proportion of Aβ + individuals remained classified as WM + uptake (Fig. S2C). This observation suggests that WM status is not reducible to amyloid positivity alone and may capture inter-individual variability in structural vulnerability or resilience beyond that reflected by amyloid burden.

Given the potential confounding effects of cerebrovascular factors on the assessment of myelin integrity [48], we included cerebrovascular risk factors—CVR-PC1 and the number of MRI-detected infarcts—as covariates in our analyses [33, 49]. Additionally, despite its association with myelin-related change, WMH is also considered to reflect brain vascular injury [49]. Recognizing the reduced FBP uptake in WMH relative to normal-appearing white matter (NAWM) [22], we further adjusted for WMH volume in the models. Importantly, even after controlling for these variables, as well as age, sex, and APOE4 status, the model estimates of WM SUVR remained robust. However, in the longitudinal analyses, WMH and cerebrovascular risk variables were not included because of substantial missing data over the 5–10-year follow-up period, and their omission may result in residual confounding.

A major strength of this study lies in its temporal analysis of how myelin disruption is associated with Aβ and Tau dynamics during AD, supported by large-scale cohort data. Myelin loss has notable associations with both pathologies, with a sequential pattern observed. Nevertheless, important caveats should be noted when interpreting the current findings: (1) The potential impact of Aβ plaque deposition on FBP-PET binding in WM is inevitable. To mitigate this, we employed a previously established method in which WM FBP signals were adjusted based on cortical GM signals [11, 22]. This approach ensures that WM FBP signals are independent of cortical GM FBP signals, thereby minimizing the confounding influence of Aβ-related signals in WM. That said, a small degree of residual Aβ-related signal cannot be entirely excluded. However, as the association with cortical SUVR was markedly attenuated after adjustment, any residual effect is likely limited and unlikely to materially affect the interpretation that the adjusted WM metric primarily reflects white matter integrity. (2) We employed WM SUVR corrected for cortical GM signals, which may introduce some deviation in the linear relationship with the Aβ PET biomarkers. However, despite this correction, the association between myelin alterations and the rate of Aβ accumulation, as well as NFTs signals, remains robust under equivalent cortical signal conditions. Therefore, this does not compromise the validity of our primary conclusions. (3) The imaging data used in this study were obtained from processed data in the ADNI database. Although it cannot be confirmed whether images from different modalities were registered to the same original T1-w MRI, it is confirmed that all original data were collected within a comparable time frame. Due to the inability to perform large-scale spatial registration across modalities, myelin content was treated as a binary measure, mitigating small spatial biases in the large dataset. (4) Residual selection bias due to non-random missingness across imaging modalities cannot be fully excluded (Tables S1–S4). Such missingness may preferentially exclude more severely affected individuals, potentially attenuating the observed associations. Importantly, subsampling-based sensitivity analyses indicate that the key findings in Model 4 remain robust and are unlikely to be driven by sample imbalance (Fig. S10). (5) Future studies incorporating alternative myelin-sensitive imaging techniques (e.g., T1/T2 myelin mapping) will be important for validating and contextualizing WM SUVR. Finally (6), given the observed spatial patterns of demyelination, further work should investigate the network-level interactions among fiber tract demyelination, Aβ deposition, and NFT aggregation across connected brain regions to better characterize the progression of AD pathology.

Conclusions

In summary, despite caveats discussed above, our investigation shows that progressive myelin degeneration occurs throughout the course of AD. Notably, declines in myelin integrity are associated with earlier Aβ deposition in the early stages and greater Tau accumulation in later stages. This sequential, structured pattern suggests that myelin dysfunction is linked to broader system-level perturbations in neurodegenerative processes. Collectively, these results underscore the relevance of myelin in AD and support its potential role as an indicator of structural vulnerability across disease progression.

Abbreviations

AD

Alzheimer’s disease

ADNI

Alzheimer’s Disease Neuroimaging Initiative

Aβ

Beta-amyloid

ANOVA

Analysis of variance

AxD

Axial diffusivity

CN

Cognitively normal

CI

Cognitive impairment (referring to MCI or dementia)

CL

Centiloid

CSF

Cerebrospinal fluid

CVR-PC1

Principal component 1 of cerebrovascular risk factors

DTI

Diffusion tensor imaging

DWI

Diffusion-weighted imaging

FA

Fractional anisotropy

FBP

Florbetapir

FLAIR

Fluid-attenuated inversion recovery

FTP

Flortaucipir

GLMs

Generalized linear models

GM

Gray matter

LOESS

Locally estimated scatterplot smoothing

MCI

Mild cognitive impairment

MD

Mean diffusivity

MetaROI

Meta-region of interest

MRI

Magnetic resonance imaging

NFTs

Neurofibrillary tangles

NAWM

Normal-appearing white matter

PART

Primary age-related tauopathy

PET

Positron emission tomography

RD

Radial diffusivity

rHCV

Residualized hippocampal volume

ROI

Region of interest

SE

Standard error

SNAP

Suspected non-Alzheimer’s pathology

SUVR

Standardized uptake value ratio

WM

White matter

WMH

White matter hyperintensity

Authors’ contributions

JW, BX, and JZ contributed to the conception and design of the study. BX and JZ were responsible for project administration. JW was responsible for data analysis and contributed to drafting the manuscript. GL, YF, WJ, BX, and JZ contributed to the manuscript revision. JZ supervised the study. All authors read and approved the final manuscript.

Funding

This work was supported by the National Natural Science Foundation of China under Grant 32530027, Key R&D Program of Zhejiang under Grant 2024C03098, Natural Science Foundation of Zhejiang Province under Grant LY24H090006 and ZCLQ24H2801, and Innovative Institute of Basic Medical Science of Zhejiang University. The funding body didn’t play a role in the design of the study and collection, analysis, interpretation of data, or in writing the manuscript.

Data availability

The data used in the current study were obtained from the ADNI database (available at https://adni.loni.usc.edu). Derived data are available from the corresponding author on request by any qualified investigator subject to a data use agreement.

Declarations

Ethics approval and consent to participate

Data used in the preparation of this article were obtained from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database (https://adni.loni.usc.edu). The ADNI study was approved by the Institutional Review Boards of all participating institutions, and written informed consent was obtained from all participants or their authorized representatives. The present study complies with the ADNI Data Use Agreement and involved only secondary analysis of existing data, and thus did not require additional ethical approval.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.