Aging and Alzheimer's Disrupt Completely Different Brain Communication Pathways

Understanding why brain connectivity studies in aging and Alzheimer's disease (AD) so often produce conflicting results — some showing hyperconnectivity, others hypoconnectivity, in overlapping regions — has been a persistent challenge. This study, published in Nature Neuroscience, offers a conceptual resolution by showing that age- and AD-related changes in functional connectivity (FC) are not random or contradictory but are instead organized along two distinct, orthogonal axes of brain architecture. Using data from the BioFINDER-2 cohort (N=973) and a validation cohort from the Alzheimer's Disease Neuroimaging Initiative (ADNI, N=129), investigators quantified nodal affinity — a parcel-wise measure of how similar each brain region's connectivity profile is to all others — across 1,000 cortical parcels. Rather than asking how gradient expression changes, they used canonical functional gradients as a spatial reference framework to interpret where connectivity alterations land on the cortical map.

The primary finding was a striking double dissociation. Increasing AD pathology (quantified via a continuous SCORPIUS-derived score combining CSF Aβ42/40 and tau PET) was associated with FC similarity (FCS) changes strongly aligned with the sensory–association (SA) axis (BioFINDER: r=0.74, P_spin<0.001; ADNI: r=0.54, P_spin<0.001): sensorimotor regions lost FCS while associative and transmodal regions gained it. By contrast, increasing age was associated with changes aligned with the representational–executive (RE) axis (BioFINDER: r=0.75, P_spin<0.001; ADNI: r=0.68, P_spin<0.001): executive regions gained FCS while representational regions lost it. Crucially, age effects showed no strong alignment with the SA axis, and AD pathology showed no strong alignment with the RE axis, confirming orthogonality. All variance inflation factors were below 1.2, ruling out multicollinearity.

Nonlinear generalized additive models revealed that these gradient alignments are not static. SA axis alignment for AD pathology peaked at moderate pathology levels (~score 0.25–0.50) and declined sharply thereafter, mirroring the often-debated 'phasic' hyperconnectivity-to-hypoconnectivity trajectory. RE axis alignment for age was strongest between ages 55–70 and essentially vanished after 80. Longitudinal analyses in 378 BioFINDER participants confirmed that within-participant increases in tau pathology (Δpathology) produced SA-aligned FCS changes (r=0.59, P_spin<0.001), not merely cross-sectional confounds. A sliding-window approach further showed that Δpathology effects peaked and declined earlier along the pathology spectrum than baseline pathology effects, suggesting gradient-aligned FC changes are detectable at very early, preclinical stages.

Cognitive analyses added further clinical weight. In cognitively healthy, amyloid-negative, APOE ε4 noncarriers (N=310), poorer cognition was associated with RE-like FCS changes (r=0.47, P_spin<0.001), and early tau pathology — even in this 'healthy' group — showed SA alignment (r=0.78, P_spin<0.001). In patients with MCI or AD dementia (N=258), the SA-like pattern shifted from tracking pathology to tracking cognitive performance directly, suggesting that once clinical impairment emerges, connectivity reorganization reflects cognitive burden more than underlying pathology load. Domain-specific bifactor analyses showed executive performance aligning more strongly with the RE axis and memory performance aligning more with the SA axis, consistent with the known differential cognitive vulnerabilities of aging versus AD. Mediation analyses suggested that pathway-aligned FCS changes may be simultaneously beneficial for one cognitive domain and detrimental for another, consistent with limited shared neural resources.

This study's strengths include its large single-site discovery cohort, external replication, longitudinal within-participant design, continuous biomarker-based pathology staging, and extensive sensitivity analyses across gradient derivation methods and FC measures. Limitations include the cross-sectional nature of the ADNI replication, limited cohort diversity (>90% of BioFINDER participants were native Swedish speakers; >90% of ADNI participants self-identified as white), inability to pinpoint specific connections driving gradient-aligned patterns, and limited representation of atypical AD phenotypes. Causal inferences about whether connectivity changes are compensatory or pathological remain open. Overall, this work positions whole-brain functional gradient alignment as a powerful interpretive and potentially diagnostic framework for distinguishing aging from AD in living humans.