Normative Modeling of Molecular-Enriched Functional Connectivity for Detecting Deviations from Healthy Brain Aging

Inter-individual variability in adult brain aging can obscure early pathological alterations and is only partly represented by population-average or scalar brain-aging measures. We combined Receptor-Enriched Analysis of functional Connectivity by Targets (REACT) with normative modeling (NM) to derive spatially resolved, molecularly informed deviation scores for dopamine transporter (DAT)-, norepinephrine transporter (NET)-, and serotonin transporter (SERT)-enriched resting-state functional connectivity (FC). We first evaluated whether these normative models could be transferred to independently processed external data through local calibration and then explored whether the resulting deviation profiles differed with cerebral amyloid burden in cognitively unimpaired older adults.

Hierarchical Bayesian regression (HBR) models with a sinh-arcsinh likelihood (SHASHb) were estimated separately for 204 cortical molecular-enriched FC features in 4,152 healthy adults (18.0-89.8 years) from seven publicly available neuroimaging datasets and evaluated in a held-out healthy test set. Pretrained models were locally adapted to three Amyloid Imaging to Prevent Alzheimer's Disease Prognostic and Natural History Study (AMYPAD-PNHS) acquisition batches using cognitively unimpaired, amyloid-negative participants (global Clinical Dementia Rating (CDR) = 0; Centiloid (CL) < 10). The independent primary comparison contrasted participants with intermediate amyloid burden (10 ≤ CL < 30; n = 111) and amyloid-positive participants (CL ≥ 30; n = 66); secondary analyses tested linear associations with continuous CL within participants with CL ≥ 10.

Of 204 normative reference models, 199 (97.5%) met the predefined diagnostic criteria; median held-out explained variance (EXPV) was 0.174. Of 612 feature-by-batch transfers, 541 (88.4%) met the strict transfer-diagnostic criteria. In the primary false discovery rate (FDR)-controlled regional analysis, DAT-enriched right caudal middle frontal cortex showed lower locally standardized deviation scores in the intermediate-amyloid-burden group than in the amyloid-positive group (adjusted difference = -0.582, q = 0.016). DAT extreme-deviation burden was also greater in the intermediate-amyloid-burden group (difference = 0.032, q = 0.044). The right frontal effect was reproduced with model-native scores across the three acquisition batches (pooled standardized effect = -0.698, q = 0.010). No NET- or SERT-enriched regional difference between these two groups survived FDR-correction, and no regional or subject-level linear association with continuous CL values survived FDR-correction in the CL ≥ 10 group.

NM can provide spatially resolved reference distributions of molecular-enriched FC that are deployable in independently processed external data when local adaptation, calibration, and feature-level transfer diagnostics are incorporated. The AMYPAD-PNHS application identified modest, spatially selective, and molecular-system-specific categorical differences during a cognitively unimpaired stage of amyloid accumulation, while continuous analyses within CL ≥ 10 did not support linear association. These findings support a methodological basis for distributed application of molecular-enriched normative models and motivate independent and longitudinal evaluation of their biological relevance.

Hierarchical normative models characterized 204 regional DAT-, NET-, and SERT-enriched FC features across 4,152 healthy adults; 199/204 models met the predefined normative-reference-model diagnostic criteria.Local hierarchical adaptation enabled the pretrained normative models to be applied to independently processed AMYPAD-PNHS data without centralized reprocessing of the target raw images or explicit feature-level harmonization; 541/612 feature-batch transfers met the strict transfer criteria.In cognitively unimpaired AMYPAD-PNHS participants, the exploratory amyloid application identified a spatially selective DAT-enriched right frontal difference and greater DAT extreme-deviation burden in the intermediate-amyloid-burden group than in the amyloid-positive group. No FDR-corrected regional NET or SERT differences between these groups and no FDR-corrected linear associations with continuous CL were identified.