Testing bidirectional relationships between functional brain networks and preclinical Alzheimer disease biomarkers

Abstract

AbstractBackgroundResting‐state functional connectivity (RSFC) provides estimates of brain networks using fMRI. There are conflicting reports regarding the effects of amyloid‐β (Aβ) and tau on RSFC networks in cognitively normal participants. These inconsistencies might be explained by a recently‐proposed theoretical model that Aβ and tau have an interactive effect on RSFC, marked first by hyper‐connectivity in association with Aβ, then by hypo‐connectivity in association with tau. Although there is preliminary support for this model, independent validations and further characterizations of the effects of Aβ and tau on RSFC are needed.MethodWe analyzed RSFC, Aβ PET, and tau PET data from the Knight Alzheimer Disease Research Center (ADRC). All participants were cognitively normal (Clinical Dementia Rating®: 0). RSFC was collected on two scanners in partly‐overlapping samples (PETMR N: 190, TRIO N: 149). RSFC data were processed with framewise censoring and global signal regression. RSFC networks were defined using 298 regions of interest. Aβ was measured with PET PiB and AV‐45 tracers, which were harmonized using a Centiloid conversion. Tau was measured with a PET AV‐1451 tracer and was summarized in pre‐defined regions. Main effect and interaction relationships between Aβ, tau PET, and RSFC were tested with linear regression models, controlling for age, sex, education, and head motion as covariates, with a false discovery rate correction.ResultWe did not observe a bidirectional interaction effect in any of the hypothesized networks (default mode [DMN], salience [SAL], frontoparietal [FPN]). Aβ was positively associated with RSFC, independent of tau, in the DMN and SAL, while tau was negatively associated with RSFC, independent of Aβ, in the SAL only. Exploratory analyses of FC within and between all networks did not identify clear patterns of interactive biomarker effects, nor consistent patterns of main effect relationships within any of the other networks.ConclusionThese results do not support an interactive model of RSFC in preclinical AD. Inconsistencies with previous interactions may be driven by sample‐specific or methodological differences, or the low reliability of FC and brain‐behavior relationships. Prominent biomarker relationships in the DMN and SAL are consistent with regional overlap with Aβ deposition in early AD.