Prediction of Regional Tau Spread Using Individualized Tau Epicenters and Structural Connectomes
Christopher Brown1, Sandhitsu Das2, Ilya Nasrallah2, John Detre3, Paul Yushkevich2, Corey McMillan2, David Wolk2
1Hospital of the University of Pennsylvania, 2University of Pennsylvania, 3Hosp of the Univ of Penn
Objective:
To predict the spread of tau in vivo based on structural connectivity with individualized epicenters of tau pathology.
Background:
Despite the fundamental association of tau pathology with functional impairment in Alzheimer’s disease, the factors that drive regional spread and the considerable individual differences in patterns of tau pathology remain poorly understood. Prior studies support spread of neurodegenerative pathology within brain networks and provide evidence of cell-to-cell transmission of pathologic proteins, which requires the transport of proteins along axons. Therefore, we hypothesized that spread of neuropathology requires a scaffolding of structural connectivity to propagate within the brain.
Design/Methods:
Tau PET (AV-1451) and DTI data from 29 participants across the clinical spectrum were included in the present study. T1-weighted and FLAIR images were used for FreeSurfer cortical segmentation. DTI and PET data were registered to T1-space using ANTs. Standardized uptake value ratios (SUVR) were calculated for each region using a cerebellar gray matter reference region. Multi-shell DTI data were collected in 113 non-collinear directions and 1.5mm isotropic voxels prior to pre-processing with a custom pipeline of tools to perform denoising, bias-correction, and eddy-current/motion correction with outlier replacement. BEDPOSTX and PROBTRACKX2 were used to perform probabilistic tractography between each FreeSurfer region. Network analysis was used to calculate the structural connectivity-based distance between all regions. The average connectivity-based distance from tau epicenters (regions with top 10% of tau SUVR in each hemisphere) was used in a linear-mixed model to predict regional SUVR.
Results:
A linear mixed-model demonstrated that regions with shorter connectivity-based distances from tau epicenters have higher tau burdens (F1,2362.5 = 298.39, p < 0.001, β = -0.022 [ -0.0246, -0.0197]). This model accounted for 40% of the total variance in tau SUVR.
Conclusions:
Individualized structural connectomes and epicenters of pathology predict regional tau burden and provide a tool for assessing expected propagation of pathology in vivo.