Spatially Distributed Amyloid-β Reduces Glucose Metabolism in Mild Cognitive Impairment

Abstract

Background: Several positron emission tomography (PET) studies have explored the relationship between amyloid-β (Aβ), glucose metabolism, and the APOEɛ4 genotype. It has been reported that APOEɛ4, and not aggregated Aβ, contributes to glucose hypometabolism in pre-clinical stages of Alzheimer's disease (AD) pathology.

Objective: We hypothesize that typical measurements of Aβ taken either from composite regions-of-interest with relatively high burden actually cover significant patterns of the relationship with glucose metabolism. In contrast, spatially weighted measures of Aβ are more related to glucose metabolism in cognitively normal (CN) aging and mild cognitive impairment (MCI).

Methods: We have generated a score of amyloid burden based on a joint singular value decomposition (SVD) of the cross-correlation structure between glucose metabolism, as measured by [18F]2-fluoro-2-deoxyglucose (FDG) PET, and Aβ, as measured by [18F]florbetapir PET, from the Alzheimer's Disease Neuroimaging Initiative study. This SVD-based score reveals cortical regions where a reduced glucose metabolism is maximally correlated with distributed patterns of Aβ.

Results: From an older population of CN and MCI subjects, we found that the SVD-based Aβ score was significantly correlated with glucose metabolism in several cortical regions. Additionally, the corresponding Aβ network has hubs that contribute to distributed glucose hypometabolism, which, in turn, are not necessarily foci of Aβ deposition.

Conclusions: Our approach uncovered hidden patterns of the glucose metabolism-Aβ relationship. We showed that the SVD-based Aβ score produces a stronger relationship with decreasing glucose metabolism than either APOEɛ4 genotype or global measures of Aβ burden.

Keywords: Alzheimer’s disease; amyloid-β; cross-correlation; glucose metabolism; positron emission tomography; singular value decomposition.

Fig.1

SVD analysis for the cross-correlation between Aβ and metabolism across the whole sample of florbetapir and FDG PET images. Cortical surface projections for the pair of spatial loadings of (A) Aβ and (B) glucose metabolism corresponding to the first principal component, which accounted for 27.84% of the total co-variability. The strongest positive weights in (A) are regions maximally related to the reduction of glucose metabolism observed in (B), namely, in the angular and inferior temporal gyri. C) Histograms and empirical density distribution of the SVD-based Aβ scores for each clinical sub-population. The CN and EMCI scores show unimodal distributions with maximum peaks around SUVR = 1.2. The LCMI scores show a bi-modal distribution with two peaks around SUVR = 1.2 and SUVR = 1.6. D) Box-plots of the Aβ subject loadings for each cognitive group. The mean SUVR_SVD amyloid increases with the cognitive decline (F = 30.84, p < 0.001).

Fig.2

Statistical assessment of APOE ɛ4 and two different Aβ predictors on glucose metabolism. A) Cortical surface projections of t-statistic maps for the main effect of APOE ɛ4. FDR-based thresholding showed no significant regions. FDR-thresholded statistically significant regions for the main effect of SVD-based Aβ scores (B) and composite ROI amyloid burden (C) as predictors of cortical metabolism. The effect of composite ROI Aβ burden on metabolism is less spatially extended compared to the SVD-based Aβ score, particularly in the bilateral right angular gyrus, inferior temporal gyrus, and precuneus. D) Regions surviving FDR thresholding for the main effect of SVD-based Aβ scores corresponding to the LMCI cohort. Significant regions appear to be weaker and less spatially extended as compared to the whole sample.

Fig.3

Statistical assessment of different Aβ predictors on glucose metabolism for the cohort of Aβ- and Aβ+ subjects. A) Strong areas of relationship between SVD-based Aβ scores and glucose metabolism are observed in the bilateral angular gyri and the posterior cingulate cortex for the Aβ- subjects. B) A much weaker association is observed for the composite ROI Aβ score. C) A weak association between SVD-based scores and metabolism is also observed for Aβ+ subjects. D) Boxplots for the SVD-based Aβ scores and composite ROI amyloid show similar mean values in both cohorts of subjects.

Fig.4

The Aβ seed-based network corresponding to (A) left fusiform gyrus (LFUSI) seed, (B) right angular gyrus (RANG) seed, and (C) composite ROI Aβ burden. High correlation values are observed around the seed locations represented by black arrows. The Aβ network relative to the LFUSI seed is weaker and less spatially extended than that of the RANG seed. D) Box-plots of the seed-based Aβ measurements for each cognitive group. For each of the three clinical cohorts, the LFUSI seed region shows lower amyloid values than the RANG seed region.

Fig.5

Statistical assessment of seed-based Aβ predictors on glucose metabolism. Cortical surface projections of FDR-thresholded statistically significant regions for the main effect of (A) LFUSI amyloid seed and (B) RANG amyloid seed on metabolism. The Aβ LFUSI seed predicts a significant reduction of glucose metabolism mainly in the inferior temporo-parietal cortex, posterior cingulate cortex, and precuneus. The Aβ RANG seed only predicts a spatially concurrent significant reduction of metabolism in the right angular gyrus itself. C) Effect of LFUSI amyloid seed in metabolism corresponding to the LMCI cohort. Significant regions essentially match those for the case of the whole sample.