Connectivity as a universal predictor of tau progression in atypical Alzheimer's disease

Abstract

The link between regional tau load and clinical manifestation of Alzheimer's disease (AD) highlights the importance of characterizing spatial tau distribution across disease variants. In typical (memory-predominant) AD, the spatial progression of tau pathology mirrors the functional connections from temporal lobe epicentres. However, given the limited spatial heterogeneity of tau in typical AD, atypical (non-amnestic-predominant) AD variants with distinct tau patterns provide a key opportunity to investigate the universality of connectivity as a scaffold for tau progression. In this large-scale, multicentre study across 14 international sites, we included cross-sectional tau-PET data from 320 individuals with atypical AD (n = 139 posterior cortical atrophy/PCA-AD; n = 103 logopenic variant primary progressive aphasia/lvPPA-AD; n = 35 behavioural variant AD/bvAD; n = 43 corticobasal syndrome/CBS-AD), with a subset of individuals (n = 78) having longitudinal tau-PET data. Additionally, as an independent sample, we included regional post-mortem tau stainings from 93 atypical AD patients from two sites (n = 19 PCA-AD, n = 32 lvPPA-AD, n = 23 bvAD, n = 19 CBS-AD). Gaussian mixture modelling was used to harmonize different tau-PET tracers by transforming tau-PET standardized uptake value ratios to tau positivity probabilities (a uniform scale ranging from 0% to 100%). Using linear regression, we assessed whether brain regions with stronger resting-state functional MRI-based functional connectivity, derived from healthy elderly controls in the Alzheimer's Disease Neuroimaging Initiative (ADNI), showed greater covariance in cross-sectional and longitudinal tau-PET and post-mortem tau pathology. Furthermore, we examined whether functional connectivity of tau-PET epicentres (i.e. the top 5% of regions with the highest baseline tau load) and tau-PET accumulation epicentres (i.e. the top 5% of regions with the highest tau accumulation rates) was associated with cross-sectional and longitudinal tau patterns. Our findings show that tau-PET epicentres aligned with clinical variants, e.g. a visual network predominant pattern in PCA-AD ('visual AD') and left-hemispheric temporal predominance, particularly within the language network, in lvPPA-AD ('language AD'). Moreover, more strongly functionally connected regions showed correlated concurrent tau-PET levels (confirmed with post-mortem data) and tau-PET accumulation rates. The functional connectivity profile of tau-PET epicentres and accumulation epicentres corresponded to tau-PET progression patterns, with higher tau-PET levels and accumulation rates in functionally close regions, and lower tau-PET levels and accumulation rates in functionally distant regions. Our data are consistent with the hypothesis that tau propagation occurs along functional connections originating from local epicentres, across all AD clinical variants. Since tau proteinopathy is a major driver of neurodegeneration and cognitive decline, this finding may advance personalized medicine and participant-specific end points in clinical trials.

Keywords: PET; atypical Alzheimer's disease; connectivity; fMRI; heterogeneity; tau.

Figure 1

Tau-PET epicentres and positivity across AD variants. Tau epicentres (i.e. the regions with the assumed earliest and greatest tau burden) were defined at the subject level as the 5% regions with the highest tau-PET SUVRs at baseline. Group-average epicentre probabilities (A) indicate the likelihood of a region being part of the epicentre, with only epicentre probabilities ≥20% shown. Group-average baseline tau-PET positivity probabilities (a uniform tau-PET scale ranging from 0% to 100%) across AD variants are shown in (B). AD = Alzheimer’s disease; bvAD = behavioural variant Alzheimer’s disease; CBS = corticobasal syndrome; L = left; lvPPA = logopenic variant primary progressive aphasia; PCA = posterior cortical atrophy; R = right; SUVR = standardized uptake value ratio.

Figure 2

Association between functional connectivity and covariance in tau-PET across variants of AD. Surface rendering of the 200 ROI brain atlas used for tau-PET and resting-state functional MRI (fMRI) data in ROI-based analyses (A). Functional connectivity was defined as Fisher z-transformed Pearson correlations between fluctuations in the BOLD signal of all possible 200 Schaefer ROI pairs in 42 CN Aβ-negative individuals from ADNI. The 200 × 200 ROI functional connectivity matrix was density thresholded at 30% (i.e. 30% of the strongest positive connections were retained) and transformed to functional connectivity-based distance (strongly connected regions are ‘close’, while weakly or indirectly connected regions are ‘distant’). Tau-PET covariance was defined as AD variant-average Fisher z-transformed partial Pearson correlations between tau positivity probabilities of all possible ROI pairs, while adjusting for age, sex and site. The association between inter-regional functional connectivity-based distance and inter-regional tau-PET covariance was assessed using linear regression for all AD variants, both across the whole brain (C–H) and in seven individual resting-state fMRI networks separately (A and B). To test the robustness of these findings, we re-ran the whole-brain analysis 1000 times, each time using a different connectivity null model from the set of 1000 null models that were generated by shuffling the connectivity matrix while preserving the weight and degree distribution. This procedure resulted in a distribution of β-values based on the null models, as depicted in the beeswarm panels in C–H, where the actual β-value (furthest data-point) always exceeded the null model β-values. Aβ = amyloid-β; AD = Alzheimer’s disease; ADNI = Alzheimer’s disease neuroimaging initiative; BOLD = blood oxygen level-dependent; bvAD = behavioural variant Alzheimer’s disease; CBS = corticobasal syndrome; CN = cognitively normal; DAN = dorsal attention network; DMN = default mode network; FPCN = frontoparietal control network; lvPPA = logopenic variant primary progressive aphasia; PCA = posterior cortical atrophy; ROI = region of interest; VAN = ventral attention network.

Figure 3

Association between functional connectivity and covariance in post-mortem tau pathology in atypical AD. Using established cortical and subcortical brain atlases (i.e. AAL, CoBrA, Julich, Neuromorphometrics), we created a bilateral MRI brain atlas for the regions with post-mortem tau assessment (n = 9, see A). Functional connectivity was defined as Fisher z-transformed Pearson correlations between functional MRI (fMRI) time series (reflective of fluctuations in the BOLD signal) of all ROI pairs in 42 CN Aβ-negative individuals from ADNI. Tau covariance was defined as Fisher z-transformed partial Spearman correlations between semi-quantitative tau pathology ratings of all ROI pairs, while adjusting for age and sex. We pooled the data from all AD variants to increase statistical power. The association between inter-regional functional connectivity and inter-regional tau pathology covariance was assessed using linear regression (B). AAL = automated anatomical labelling; Aβ = amyloid-β; AD = Alzheimer’s disease; ADNI = Alzheimer’s disease neuroimaging initiative; BOLD = blood oxygen level-dependent; CN = cognitively normal; CoBrA = computational brain anatomy laboratory; ROI = region of interest.

Figure 4

Association between tau epicentre connectivity and tau-PET across AD variants. Tau epicentre connectivity was determined by taking the functional connectivity-based distance (see Fig. 2 for method specifications) of each non-epicentre ROI (n = 190) to the epicentre (n = 10). For each individual, linear regression was used to assess the association between functional connectivity-based distance to the tau epicentre and tau-PET SUVR. Subject-level β-values are visualized per AD variant in the notched boxplots in A–E. Additionally, all non-epicentre regions were grouped into quartiles based on their functional proximity to the epicentre (quartile 1 = shortest functional connectivity-based distance, quartile 4 = longest functional connectivity-based distance), and tau positivity probabilities across quartiles were compared using paired Wilcoxon signed-rank tests. AD = Alzheimer’s disease; bvAD = behavioural variant Alzheimer’s disease; CBS = corticobasal syndrome; lvPPA = logopenic variant primary progressive aphasia; PCA = posterior cortical atrophy; Q = quartile; ROI = region of interest; SUVR = standardized uptake value ratio.

Figure 5

Association between functional connectivity and covariance in tau-PET percentage change in PCA-AD and lvPPA-AD. Surface rendering of the 200 ROI brain atlas used for tau-PET and resting-state functional MRI (fMRI) data in ROI-based analyses (A). We computed annual tau-PET SUVR change for each individual by fitting 200 linear models (one for each ROI), using follow-up time as the independent variable and tau-PET SUVR as the dependent variable. We then normalized each ROI’s rate of change by the individual’s initial SUVR (at follow-up time = 0) to express it as a relative percentage change per year. Covariance in tau-PET percentage change was determined by calculating AD variant-average Fisher z-transformed partial Pearson correlations between the percentage change rates of all ROI pairs while adjusting for age, sex and site. Using the functional connectivity-based distance matrix described in Fig. 2, we assessed the association between inter-regional functional connectivity-based distance and inter-regional tau-PET percentage change covariance through linear regression, both across the whole brain (C and D) and in seven individual resting-state fMRI networks separately (A and B). We re-ran the analysis 1000 times (same procedure as described in Fig. 2) to test the robustness of our findings, as illustrated in the beeswarm panels in C and D, where the actual β-value (furthest data-point) always exceeded the null model β-values. AD = Alzheimer’s disease; DAN = dorsal attention network; DMN = default mode network; FPCN = frontoparietal control network; lvPPA = logopenic variant primary progressive aphasia; PCA = posterior cortical atrophy; ROI = region of interest; SUVR = standardized uptake value ratio; VAN = ventral attention network.

Figure 6

Association between tau accumulation epicentre connectivity and tau-PET change in PCA-AD and lvPPA-AD. Tau accumulation epicentres were defined as the top 5% of ROIs (i.e. 10 ROIs in total) with the highest annual percentage change in tau-PET SUVR. Group-average epicentre probabilities indicate the likelihood of a region being part of the epicentre, with only epicentre probabilities ≥10% shown. Tau accumulation epicentre connectivity was determined by taking the functional connectivity-based distance (see Fig. 2 for method specifications) of each non-accumulation-epicentre ROI (n = 190) to the accumulation epicentre (n = 10). For each individual, linear regression was used to assess the association between functional connectivity-based distance to the tau accumulation epicentre and tau-PET annual percentage change. Subject-level β-values are visualized per AD variant in the notched boxplots in A and B. Additionally, all non-accumulation-epicentre regions were grouped into quartiles based on their functional proximity to the accumulation epicentre (quartile 1 = shortest functional connectivity-based distance, quartile 4 = longest functional connectivity-based distance), and tau-PET percentage change rates across quartiles were compared using paired Wilcoxon signed-rank tests. AD = Alzheimer’s disease; lvPPA = logopenic variant primary progressive aphasia; PCA = posterior cortical atrophy; Q = quartile; ROI = region of interest; SUVR = standardized uptake value ratio.