Sequencing cortical microstructural changes along the Alzheimer’s disease continuum

Abstract

AbstractBackgroundMicrostructural abnormalities likely precede macrostructural changes in the Alzheimer’s disease (AD) cascade. Diffusion MRI (dMRI) is sensitive to microstructural properties of brain tissue but few studies have evaluated dMRI measures in cortical gray matter where many early AD histopathological changes occur. Event‐based modeling (EBM) is a data‐driven approach for probabilistically sequencing cross‐sectional biomarkers in the order that they likely become abnormal. Here, we used EBM to examine the sequence of changes in cortical dMRI measures relative to more widely used amyloid, tau, brain volume, and cognitive biomarkers in two independent AD studies.MethodT1w, dMRI, and amyloid‐PET (FBB/FBP) data were analyzed in 461 ADNI3 participants and 188 OASIS3 participants (Figure 1A). Some ADNI3 participants also had tau‐PET (AV‐1451) and CSF pTAU‐181 and Aβ1‐42 data. In addition to DTI, novel single‐shell adaptations of multi‐shell models, NODDI‐DTI and MAP‐AMURA, were fit to the dMRI data. 10 mean cortical dMRI measures were extracted using FreeSurfer parcellated T1w‐images (Figure 1B). T1w cortical and hippocampal volumes were also extracted, and all MRI measures harmonized using ComBat. Mean cortical amyloid‐PET SUVRs were converted to centiloids. Associations between each of 17 biomarkers (Figure 1C) and cognitive impairment (CI –MCI+Dementia– vs CN) were tested in ADNI3. Significant biomarkers were included in an ADNI3 EBM to determine event ordering.The EBM was validated in OASIS3 and used to classify clinical diagnosis.ResultAll biomarkers except DTI FA showed significant differences between CN and CI ADNI3 participants (Figure 2A). ADNI3 EBM confirmed existing AD models: CSF amyloid abnormalities preceded PET, followed by CSF pTau, while cognitive and volumetric abnormalities occurred at later stages. Changes in all dMRI biomarkers preceded CSF or PET Tau, cognitive, and volumetric abnormalities (Figure 2B). Estimated disease stages in ADNI3 distinguish dementia from CN with an AUC = 0.96 (10‐fold cross‐validation). OASIS3 disease stages, estimated using ADNI3 biomarker distributions excluding CSF measures and tau‐PET, classified diagnosis with an AUC = 0.85 (Figure 3).ConclusionCortical microstructural measure abnormalities, including greater diffusivity, lower restriction and lower dispersion potentially due to neurite loss and lower dendritic arborization complexity, may precede those detectable with conventional T1w biomarkers.