Non-negative matrix factorization evaluation of patterns of brain Aβ deposition in Parkinson’s disease, Alzheimer’s disease and normal controls on [¹⁸F]florbetapir PET

Objectives: The prevalence of cerebral beta-amyloid (Aβ) in Parkinson’s disease (PD) dementia has been shown to be higher, but variable, than in age-matched controls but lower than in Alzheimer’s disease (AD) patients.^1-6 The distribution and role of Aβ in non-demented PD remains uncertain. In this study, differential patterns of [¹⁸F]florbetapir PET brain uptake, a measure of brain Aβ deposition, were examined in non-demented PD patients, AD patients, and age-matched normal controls (CON) using a data-driven algorithm based on non-negative matrix factorization (NNMF).⁷ NNMF partitions the brain into regions with correlated Aβ deposition and is thought to provide improved interpretability compared to Principal Component Analysis (PCA) and Independent Component Analysis (ICA).

Methods: This study included 120 subjects who underwent [¹⁸F]florbetapir PET brain imaging: 40 recruited from a local cohort of non-demented PD patients (67±8 years; 35% females), 40 age-matched AD patients (67±7.9 years; 43% females) obtained from Alzheimer's Disease Neuroimaging Initiative (ADNI), and 40 normal CON (69±6.7 years; 45% females) also from ADNI. The [¹⁸F]florbetapir PET studies of the PD cohort were smoothed using ADNI scanner-specific parameters based on Hoffman phantom to match the uniform effective isotropic resolution of 8 mm FWHM used in the ADNI data. [¹⁸F]Florbetapir PET data were then transformed to common MNI space using a T1-weighted MRI based method and normalized to whole cerebellum to calculate standardized uptake value ratio (SUVR) maps.^8,9 NNMF was applied to [¹⁸F]florbetapir SUVR maps over the entire population, resulting in components that represent brain regions with correlated Aβ deposition. The loading coefficients of each NNMF component for each subject is a surrogate measure of SUVR for the brain regions of that component. For each NNMF component, differences in loading coefficients of the subjects grouped by diagnosis were examined using ANOVA with Tukey post-hoc test.

Results: NNMF components showed multiple patterns of cortical brain regions with correlated [¹⁸F]florbetapir uptake, where loading coefficients as a measure of uptake significantly differed by diagnosis on ANOVA (p<0.00001; see Figure 1 for examples). One component consisted of brain regions typically positive for Aβ in AD, including bilateral anterolateral frontal lobes, superolateral temporal lobes, lateral parietal lobes, precuneus, and posterior cingulate gyrus. On post-hoc analysis, the loading coefficients of this component were significantly higher in AD than CON (p<0.001) and PD (p<0.001), but not different between PD and CON (p=0.22). Another component, consisting of bilateral medial temporal lobes, hippocampi, and insula, showed significantly higher loading coefficients both in AD than CON (p<0.0001) and in PD than CON (p=0.0024), but not different between PD and AD (p=0.53). There were also components where the loading coefficients were significantly higher in AD than PD and in PD than CON (i.e., AD > PD > CON in uptake). This included a component involving bilateral inferolateral temporal lobes and anteroinferior frontal lobes and a component involving bilateral posterior occipital lobes (p<0.005 for all three post-hoc pairwise comparisons for each NNMF component).

Conclusions: Through NNMF, this study demonstrates that cerebral Aβ deposition in PD diverges from that of normal aging and has a different pattern of distribution in PD compared to AD. This study lends further support for the need to investigate the potential role of Aβ in the pathophysiology of PD. RESEARCH SUPPORT: This study was funded or supported by the National Institutes of Health: P50-NS053488, P30-AG010124, R43-NS063607, R01-AG054409, and K08-NS093127, and by Avid Radiopharmaceuticals, Inc.