Directed Functional Brain Connectivity is Altered in Sub-threshold Amyloid-β Accumulation in Cognitively Normal Individuals

Abstract

Several studies have shown that amyloid-β (Aβ) deposition below the clinically relevant cut-off levels is associated with subtle changes in cognitive function and increases the risk of developing future Alzheimer's disease (AD). Although functional MRI is sensitive to early alterations occurring during AD, sub-threshold changes in Aβ levels have not been linked to functional connectivity measures. This study aimed to apply directed functional connectivity to identify early changes in network function in cognitively unimpaired participants who, at baseline, exhibit Aβ accumulation below the clinically relevant threshold. To this end, we analyzed baseline functional MRI data from 113 cognitively unimpaired participants of the Alzheimer's Disease Neuroimaging Initiative cohort who underwent at least one ¹⁸F-florbetapir-PET after the baseline scan. Using the longitudinal PET data, we classified these participants as Aβ negative (Aβ-) non-accumulators (n = 46) and Aβ- accumulators (n = 31). We also included 36 individuals who were amyloid-positive (Aβ+) at baseline and continued to accumulate Aβ (Aβ+ accumulators). For each participant, we calculated whole-brain directed functional connectivity networks using our own anti-symmetric correlation method and evaluated their global and nodal properties using measures of network segregation (clustering coefficient) and integration (global efficiency). When compared to Aβ- non-accumulators, the Aβ- accumulators showed lower global clustering coefficient. Moreover, the Aβ+ accumulator group exhibited reduced global efficiency and clustering coefficient, which at the nodal level mainly affected the superior frontal gyrus, anterior cingulate cortex, and caudate nucleus. In Aβ- accumulators, global measures were associated with lower baseline regional PET uptake values, as well as higher scores on the Modified Preclinical Alzheimer Cognitive Composite. Our findings indicate that directed connectivity network properties are sensitive to subtle changes occurring in individuals who have not yet reached the threshold for Aβ positivity, which makes them a potentially viable marker to detect negative downstream effects of very early Aβ pathology.

Keywords: Functional MRI; amyloid-β positron emission tomography; anti-symmetric correlations; directed functional connectivity; longitudinal amyloid-β accumulation; preclinical AD.

Figure 1.

Sample differences between different groups in global and nodal directed functional network topology. (A) Overview of the calculation procedure for the anti-symmetric correlation. For simplicity, we illustrate the procedure using the time activation series for 5 nodes as an example. The lagged Pearson’s correlation coefficient between these time series at varying temporal lags allows the estimation of lagged correlation functional networks (in this example, at lag of 1). The lagged adjacency matrix can be written as a sum of symmetric and anti-symmetric and matrices. Here, we use the anti-symmetric matrices to estimate the functional connectivity of each individual. Differences between the different groups at lag 3 (calculated as group 2-group 1 in the figure) in the (B) clustering coefficient and (C) global efficiency. The differences in the corresponding network measures between groups are plotted in orange circles as a function of network density and the upper and lower bounds of the 95% confidence interval are plotted in blue. The differences are considered statistically significant if they fall outside the CIs. (D) Differences between Aβ−/NAc and Aβ+ groups in nodal network measures. The plotted regions remained significant after correcting for multiple comparisons using FDR at q < 0.05.

Figure 2.

Correlation between Aβ deposition levels and cognitive scores and global measures for Aβ−/Ac group. The plots show the best line fit for Aβ−/Ac group between baseline Aβ SUVR deposition levels and (A) clustering coefficient at lag 4, and global efficiency at lags (B) 2, (C) 3, and (D) 4. The correlation between PACCtrailsB test scores and global efficiency at lag 2 is shown in (F). The global measure values were regressed with age, sex, and education as covariates; the residuals of the regression were used to calculate the best line fit. A representative fit at density (A) 39% and (B-F) 25% is shown.