Premium
Stages of tau aggregation associated with amyloidosis reflected in non‐negative matrix factorization components
Author(s) -
Srinivasan Dhivya,
TrueloveHill Monica,
Erus Guray,
Sotiras Aristeidis,
Doshi Jimit,
Wolk David A.,
Davatzikos Christos,
Nasrallah Ilya M.
Publication year - 2020
Publication title -
alzheimer's and dementia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 6.713
H-Index - 118
eISSN - 1552-5279
pISSN - 1552-5260
DOI - 10.1002/alz.046728
Subject(s) - non negative matrix factorization , neuroimaging , psychology , alzheimer's disease , pathology , disease , medicine , neuroscience , physics , matrix decomposition , eigenvalues and eigenvectors , quantum mechanics
Background Tau aggregation, when accompanied by amyloid deposition, is a key neuropathological signature of Alzheimer’s disease (AD). The regional distribution of tau may be a more sensitive marker of AD pathology than total tau level, however the spread of disease is unlikely to conform to common atlas‐based regions of interest. As such, methods for evaluating tau’s regional distribution could be useful for predicting AD staging or disease progression. In this analysis, we derive regional patterns of tau aggregation in AD using nonnegative matrix factorization (NNMF), a data‐driven method of identifying areas of the brain that consistently covary across individuals. Method We used data from 269 ‐amyloid (A) positive subjects with 18 F‐flortaucipir‐PET data from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) database and the AVID study (138 females, mean age = 73.09; 110 healthy controls, 118 MCI, and 41 AD). PET data were registered to Standard Montreal Neurological Institute (MNI152) 182x218x182 grid with 1mm resolution. SUVR maps were computed with a cerebellar gray matter reference and NNMF was used to estimate common patterns of tau aggregation. Once these patterns were identified, mean SUVR for each component computed for each participant. We performed Gaussian mixture modeling for each component to determine optimal tau positivity cutpoint, and subjects were classified as positive or negative for each component. Components were ordered by frequency of appearance in subjects in order to determine staging. Result 12 regional patterns of tau binding were identified (Figure 1). The patterns most commonly expressed among subjects reflected regions of the temporal lobe (Figure 2a). This was particularly noted in those with AD (Figure 2b). These components were followed by a component including the lingual gyrus, and then by components involving frontal lobes, precuneus, and parietal lobes. Conclusion NNMF‐derived tau PET components provide a data‐driven method for identifying common patterns of tau aggregation, producing stages that are consistent with previous research on tau pathology in AD. These components may provide a more sensitive method for identifying disease progression in AD.