Understanding Mechanistic Role of Structural and Functional Connectivity in Tau Propagation Through Multi-Layer Modeling

TL;DR

Beyond showing that connectome architecture constrains tau spread, this model reveals a regionally asymmetric contribution of SC and FC, with FC predominantly drives tau spread in subcortical areas, the insula, frontal and temporal cortices, whereas SC plays a larger role in occipital, parietal, and limbic regions.

Abstract

Emerging neuroimaging evidence shows that pathological tau proteins build up along specific brain networks, suggesting that large-scale network architecture plays a key role in the progression of Alzheimer's disease (AD). However, how structural connectivity (SC) and functional connectivity (FC) interact to influence tau propagation remains unclear. Leveraging an unprecedented volume of longitudinal neuroimaging data, we examine SC-FC interactions through a multi-layer graph diffusion model. Beyond showing that connectome architecture constrains tau spread, our model reveals a regionally asymmetric contribution of SC and FC. Specifically, FC predominantly drives tau spread in subcortical areas, the insula, frontal and temporal cortices, whereas SC plays a larger role in occipital, parietal, and limbic regions. The relative dominance of SC versus FC shifts over the course of disease, with FC generally prevailing in early AD and SC becoming primary in later stages. Spatial patterns of SC- and FC-dominant regions strongly align with the regional expression of AD-associated genes involved in inflammation, apoptosis, and lysosomal function, including CHUK (IKK-alpha), TMEM106B, MCL1, NOTCH1, and TH. In parallel, other non-modifiable risk factors (e.g., APOE genotype, sex) and biological mechanisms (e.g., amyloid deposition) selectively reshape tau propagation by shifting dominant routes between anatomical and functional pathways in a region-specific manner. Findings are validated in an independent AD cohort.

Figures (14)

• Figure 1: Typical trajectory of tau evolution in aging brains.(a) Left: Scatter of participant ages at longitudinal tau PET scan, with lines linking multiple scans per individual. The cortical mappings represent the average tau SUVR for each clinical cohort. Right: Regional tau SUVR values grouped by clinical diagnosis (CN, SMC, EMCI, LMCI, AD), with black dots marking group means, illustrating the progressive increase in tau burden across disease stages. (b) Surface map of regional SUVR differences (CN vs. AD) by a linear mixed‐effects model (random intercept per subject; sex covariate; Bonferroni‐corrected $p<5\times10^{-5}$, two‐sided). Higher positive $t$‐values (darker blue) indicate stronger tau propagation. (c,d) Community‐level aggregation of tau progression. Regions were assigned to anatomical lobes based on the Destrieux atlas destrieux2010automatic (c) and to functional modules by Power et al.power2011functional (d). For each community, mean $t$‐values were converted to $z$‐scores via 1,000 spin‐permutation tests to correct for spatial autocorrelation alexander2018testingliang2024structural, generating a null‐model of regional $t$‐values. Positive $z$‐scores denote accelerated tau deposition relative to the null model. Asterisks indicate $p_{\mathrm{spin}}<0.05$ (e.g. Temporal lobe: $p=0.024$, Subcortical: $p<10^{-5}$). SUVRs mapped in ParaView (v5.10.1) .
• Figure 2: Tau propagation is anchored to structural and functional network backbones.(a). Group‐average structural (left) and functional (right) connectivity matrices ($160\times 160$) thresholded to their backbone edges. (b). Schematic illustrating how we relate regional tau propagation to network structure: (1) for each node $i$ (red), we compute the tau extent $T_i$ ($t$-value derived from mix-effect model of predicting diagnostic label using longitudinal regional tau SUVR); (2) compute averaged tau extent of its directly connected neighbors $\hat{T}_i^S$ based on SC (green edges) and $\hat{T}_i^F$ based on FC (gray edges), then correlate $\hat{T}_i^S\sim T_i$ and $\hat{T}_i^F\sim T_i$, respectively. (c). $T_i\sim \hat{T}_i^S$ (left) and $T_i\sim \hat{T}_i^F$ (right) across all nodes reveals strong positive associations (Pearson correlation, SC: $r{_{\text{adj}}=0.783, p<0.01}$, FC: $r{_{\text{adj}}=0.736, p<0.01}$, two sided). Lines show least‐squares fits; shaded areas denote 95% confidence intervals. (d). Lobe‐level spin‐permutation testing (FDR corrected at $p{_{\text{spin}}=0.01},$ one‐tailed) confirms that only multi-sensory area (frontal, temporal, and parietal) and subcortical systems exhibit connectivity–tau correlations exceeding spatial‐null expectations (colored boxes). Multiple linear regression analyses of FC and SC associations with (e) tau accumulation rate ($\Delta$Tau) and (f) MMSE, respectively. Both models include age, sex, and APOE genotype as covariates to control for potential confounding effects. Our analyses revealed that FC primarily drives tau propagation, while SC is more closely linked to cognitive performance, indicating distinct roles in disease spread and resilience.
• Figure 3: Age‐related and individual‐level analyses of tau propagation along SC and FC.Panels a–c: GAM with age as a continuous predictor.(a) Cortical surface map of the effect size of age on tau accumulation, Bonferroni‐corrected at $p<5\times10^{-5}$, two‐sided. (b) Example GAM fits for entorhinal cortex, middle temporal gyrus, amygdala, and hippocampus: green lines denote the estimated age‐trajectory of tau SUVR; shaded gray bands show 95% confidence intervals. (c) Spatial maps of the first derivative of the GAM fit (tau propagation rate) plotted for four age windows: $<$60, 61–75, 76–85, and $>$85 years. Panels d–e: individual‐level longitudinal analysis.(d) Illustration of how nodal tau propagation rates are computed from each subject’s serial scans. ($\Delta$Tau/$\Delta$Age), and the resulting group‐mean surface map (averaged over all 210 participants) showing the spatial distribution of propagation velocities—highest in medial temporal and subcortical regions. (e) Scatter plot of nodal rate versus neighbor‐mean rate in SC (gray) and FC (green), with linear fits and 95% confidence intervals (red shading). Reported correlations: SC $r_{\rm adj}=0.855$, FC $r_{\rm adj}=0.835$, both $p<0.001$, two‐sided.
• Figure 4: Mechanistic role of SC and FC on tau propagation.(a) Main findings.Left: TProportions of tau propagation attributed to structural connectivity ($u_s$, blue) and functional connectivity ($u_f$, red) across different brain lobes. Right: Comparative mapping of $u_s$ versus $u_f$, where blue marks regions with stronger SC‐driven propagation and red highlights those with stronger FC‐driven propagation. Node sizes scale with the absolute difference in magnitude, $|u_s - u_f|$. (b) Dynamic contribution of SC and FC to tau propagation. SC and FC contributions to tau propagation across four age stages: $<$60, 61–75, 76–85, and $>$85 years. Age-dependent shift in tau propagation, with younger individuals showing FC-dominant spread and older individuals showing increasing reliance on SC pathways. (c) Stratification analysis by biological sex. Sex-dependent effect only manifests a minor modulatory effect on tau propagation, with a modest occipital difference that does not alter the overall SC-to-FC propagation architecture. (d) Impact of amyloid-$\beta$ deposition. SC vs. FC comparisons between $A\beta+$ and $A\beta-$ individuals suggest that A$\beta$ burden “boosts” tau spread along functional circuits (especially in frontal cortex). (e) The effect of APOE4 status on region-specific network conduit in tau propagation.APOE4 carriers show a transition from FC- to SC-dominant tau spread in the frontal, occipital, and insula cortex. (f) Stratified results on disease phrase. Group-wise propagation patterns across clinical diagnoses (e.g., CN, AD). Except for the temporal and subcortical regions, all other lobes exhibited a reversal in SC–FC dominance between AD and CN groups. .
• Figure 5: Gene correlates of SC- and FC-dominant tau propagation on ADNI and OASIS datasets.Top: LASSO selection results using tau propagation associated with FC ($u_f$) (left), associated with SC ($u_s$) (middle), and the selection pattern of SC-FC dominance that both $u_s$ and $u_f$ exhibit significant difference between CN and AD (right). Pie chart--selection paths across regularization ($\lambda$) for the top 10 genes (first six genes are more stable). Violin diagram: Gene selection frequency over 100 non-negative LASSO bootstrap resamples. Brain mapping under the violin diagram: $t$-value in each CN vs AD group comparison ($p<0.05$). Shaded genes are the common genes on both ADNI and OASIS datasets. Star denotes the common genes across $u_s$ and $u_f$. Bottom: Abagen‐derived expression maps of nine consensus genes plotted on Destrieux cortical (top rows) and subcortical (bottom rows) surfaces. Color scale denotes normalized expression (0–0.9). CHUK: Conserved Helix-Loop-Helix Ubiquitous Kinase, TH: Tyrosine Hydroxylase, TMEM106B: Transmembrane Protein 106B, MCL1: Myeloid Cell Leukemia 1, NOTCH1: Notch receptor 1