Modeling brain‐predicted age from functional connectivity in early‐stage and preclinical Alzheimer disease

Abstract

AbstractBackgroundThe “brain‐predicted age” approach summarizes complex age relationships with neuroimaging features by quantifying apparent brain age compared to normative trajectories. Advanced brain‐predicted age has been observed in many studies of symptomatic Alzheimer disease (AD), but the effects of preclinical AD have not been investigated. Prior studies have typically modeled brain aging with structural MRI, but other modalities, including resting‐state functional connectivity (FC), are underexplored. Here we modeled FC‐predicted brain age and tested its relationships with preclinical and symptomatic AD.MethodWe calculated FC correlation matrices between 300 regions of interest in 493 cognitively normal (clinical dementia rating [CDR]=0), amyloid‐negative (CSF pTau/Aβ42 ratio and/or amyloid PET Centiloid) control participants (18‐91 years old). We trained a Gaussian process regression model to predict age from these FC features using 10‐fold cross‐validation. We applied the trained model to a separate set of 153 amyloid‐negative, 143 preclinical AD (CDR=0, amyloid‐positive), and 183 symptomatic AD participants (CDR>0). Finally, we tested group differences in FC‐predicted brain age between the AD samples and continuous relationships with biomarkers of amyloid (CSF Aβ42, amyloid PET Centiloid), tau (CSF pTau‐181, tau PET [AV‐1451]), and neurodegeneration (CSF NfL, hippocampal volume), controlling for true age as a covariate.ResultThe FC‐based model accurately predicted age in the training set (R2 =.72, mean absolute error=8.14, Figure 1). FC‐predicted brain age was 2.43 years older in symptomatic AD compared to cognitively normal controls (p<.001, Cohen’s d=0.38, Figure 2A), but was 1.90 years younger in preclinical AD compared to amyloid‐negative controls (p=.011, Cohen’s d=‐0.29, Figure 2B). Greater FC‐predicted brain age was associated with smaller hippocampal volume (p=.009, Figure 3F), but also with lower CSF pTau‐181 (p=.012, Figure 3C). Relationships between FC‐predicted age and other AD biomarkers were not significant (p’s>.20, Figure 3).ConclusionElevated FC‐predicted brain age may reflect advanced brain aging in FC networks during symptomatic AD. This finding is consistent with previous models based on structural MRI. Reduced FC‐predicted brain age in amyloid‐positive participants may reflect compensation during early preclinical stages, but this interpretation requires further investigation. FC‐predicted brain age may be a sensitive biomarker in future models of healthy aging and symptomatic AD.