The study, led by Jonathan Rittmo and Jacob W. Vogel of Lund University together with an international team, analyzed resting-state functional magnetic resonance imaging data from the BioFINDER-2 cohort in Sweden, comprising 973 participants with complete cerebrospinal fluid amyloid measurements, tau PET imaging and fMRI scans. To guard against the possibility that the results were an artifact of one dataset, the team replicated their core findings in 129 participants from the Alzheimer’s Disease Neuroimaging Initiative, a public–private partnership that has collected multimodal brain data since 2003. Rather than examining individual connections between predefined regions—a approach the authors argue has produced a fragmented and often inconsistent picture—they measured something called functional connectivity similarity, or nodal affinity, at each of 1,000 cortical parcels.

Nodal affinity is a technically elegant measure. After parcellating the cortex into 1,000 regions using the Schaefer atlas, the researchers computed pairwise Pearson correlations between the activity time series of every pair of regions, producing a 1,000-by-1,000 connectivity matrix for each person. For each parcel, they then calculated how similar its entire connectivity profile was to those of all other parcels, using cosine similarity after retaining the strongest 25 percent of connections. A high affinity value means a region’s pattern of communication is generic and shared with much of the cortex; a low value means the region is functionally distinctive. This single number per region captures both local granularity and whole-cortex context, bridging the two dominant traditions in connectivity research.

The crucial analytical move came next. Instead of treating functional gradients—the principal components of the healthy connectome—as outcomes to be measured, the team used them as a reference frame for interpreting where connectivity changes fall. The first and strongest gradient, known as the sensory–association axis, runs from unimodal sensory and motor cortices at one end to transmodal association areas supporting higher-order cognition, such as the medial prefrontal cortex, posterior cingulate and angular gyrus, at the other. The third gradient, which the authors call the representational–executive axis, separates the lateral frontoparietal executive system from representational regions including parts of the default mode network. The researchers correlated spatial maps of age- and pathology-related connectivity effects against these axes, assessing significance with spin tests that account for spatial autocorrelation.

The results revealed a double dissociation that is as clean as such findings ever get in human neuroscience. Connectivity changes tied to Alzheimer’s pathology—quantified as a continuous composite score derived from the CSF amyloid-beta 42/40 ratio and tau PET using the SCORPIUS trajectory-inference algorithm—aligned strongly with the sensory–association axis, with a correlation of 0.74 in BioFINDER-2 and 0.54 in ADNI. As pathology accumulated, connectivity similarity decreased in sensorimotor regions and increased in associative cortex. Age, by contrast, aligned with the representational–executive axis, with correlations of 0.75 and 0.68 in the two cohorts, reflecting rising connectivity similarity in executive areas and falling similarity in representational areas as people got older. Critically, age effects did not track the sensory–association axis and pathology effects did not track the executive axis, and the two predictors showed no problematic collinearity.

The team then asked when, across the course of disease and lifespan, these gradient-aligned patterns emerge. Fitting generalized additive models with nonlinear smooth terms for age and pathology, they found that the alignment between pathology effects and the sensory–association axis rose steeply during early pathology accumulation, remained high through intermediate stages, and then declined sharply as pathology burden grew severe. A parallel longitudinal analysis of 378 BioFINDER-2 participants with repeat imaging confirmed the story within individuals: within-participant increases in tau pathology were associated with concurrent gradient-aligned connectivity changes, and a sliding-window analysis showed that these within-participant effects actually peaked at lower pathology levels than the corresponding between-participant effects. In other words, the signature is detectable earliest precisely when intervention might matter most.

Perhaps the most clinically provocative results concern cognition. Among 310 cognitively healthy, amyloid-negative individuals who do not carry the APOE ε4 allele, age effects again aligned with the executive axis, and executive-axis-like connectivity patterns were associated with worse performance on a modified Preclinical Alzheimer’s Cognitive Composite. Meanwhile, even subtle tau pathology in these apparently healthy people produced sensory–association-aligned connectivity changes. But in the 258 participants with mild cognitive impairment or Alzheimer’s dementia, the pathology-to-gradient link vanished entirely; instead, worse cognition itself, regardless of pathology load, was associated with the sensory–association pattern. The authors suggest that gradient-aligned connectivity changes may signal cognitive strain or system-level stress—a general neural signature of demands exceeding available resources—rather than pathology alone.

Domain-specific analyses sharpened this interpretation. Using a bifactor model to decompose ten cognitive tests into general and domain-specific factors, the team confirmed that aging disproportionately erodes executive function while Alzheimer’s pathology preferentially damages memory. Connectivity changes aligned with the sensory–association axis were more strongly tied to memory, and executive-axis alignment to executive performance, mirroring the cognitive profiles of the two processes. Mediation analyses hinted that pathology-related connectivity changes along the sensory–association axis were not detrimental to memory—and may even have been beneficial—while effects that appeared helpful for one cognitive domain tended to harm the other. This trade-off is consistent with the idea that cognitive processes draw on shared, limited neural resources, and that the brain’s reorganization reflects a reallocation of those resources under pressure.

The study is not without limitations, which the authors acknowledge candidly. Resting-state connectivity is a weaker probe of cognition than task-based paradigms; the observational design precludes causal claims; and the cohorts, over 90 percent native Swedish speakers in BioFINDER-2 and predominantly white participants in ADNI, limit generalizability. Atypical Alzheimer’s phenotypes were too rare to test whether domain-predominant impairments produce distinct gradient signatures. Yet the robustness work is extensive: the findings survived replication across gradient-derivation methods, correlation thresholds, alternative connectivity measures, covariate models including cortical atrophy and vascular markers, and cross-cohort gradient validation. Whether the gradient-aligned patterns ultimately prove compensatory, harmful, or both, their emergence in healthy older adults, in pathology-positive but asymptomatic individuals, and in clinically impaired patients alike suggests a unifying framework—one that could allow clinicians to read the brain’s functional architecture not as a collection of isolated connections, but as a coherent map of where cognitive resources are being stretched thinnest.

Subject of Research: Functional connectivity gradients in aging and Alzheimer's disease

Article Title: Different functional connectivity gradients reflect aging and Alzheimer’s disease

Article References: Rittmo, J., Franzmeier, N., Strandberg, O., Chauveau, L., Satterthwaite, T. D., Wisse, L. E. M., Spotorno, N., Behjat, H. H., Dehsarvi, A., van Westen, D., Anijärv, T. E., The Alzheimer’s Disease Neuroimaging Initiative, Landau, S. M., Palmqvist, S., Janelidze, S., Stomrud, E., Ossenkoppele, R., Mattsson-Carlgren, N., Hansson, O., & Vogel, J. W. (2026). Different functional connectivity gradients reflect aging and Alzheimer’s disease. Nature Neuroscience. https://doi.org/10.1038/s41593-026-02402-0

Image Credits: AI Generated

DOI: 10.1038/s41593-026-02402-0

Keywords: Alzheimer's disease, functional connectivity, brain aging, cortical gradients, tau PET, amyloid-beta, resting-state fMRI, BioFINDER-2, ADNI, default mode network, cognitive decline, neurodegeneration