Hemispheric asymmetry of tau pathology is related to asymmetric amyloid deposition in Alzheimer's Disease

Abstract

The distribution of tau pathology in Alzheimer's disease (AD) shows remarkable inter-individual heterogeneity, including hemispheric asymmetry. However, the factors driving this asymmetry remain poorly understood. Here we explore whether tau asymmetry is linked to i) reduced inter-hemispheric brain connectivity (potentially restricting tau spread), or ii) asymmetry in amyloid-beta (Aβ) distribution (indicating greater hemisphere-specific vulnerability to AD pathology). We include 452 participants from the Swedish BioFINDER-2 cohort with evidence of both Aβ pathology (CSF Aβ42/40 or neocortical Aβ-PET) and tau pathology (temporal tau-PET), categorising them as left asymmetric (n = 102), symmetric (n = 306), or right asymmetric (n = 44) based on temporal lobe tau-PET uptake distribution. We assess edge-wise inter-hemispheric functional (RSfMRI; n = 318) and structural connectivity (dMRI; n = 352) but find no association between tau asymmetry and connectivity. In contrast, we observe a strong association between tau and Aβ laterality patterns based on PET uptake (n = 233; β = 0.632, p < 0.001), which we replicate in three independent cohorts (n = 234; β = 0.535, p < 0.001). In a longitudinal Aβ-positive sample, we show that baseline Aβ asymmetry predicts progression of tau laterality over time (n = 289; β = 0.025, p = 0.028). These findings suggest that tau asymmetry is not associated with a weaker inter-hemispheric connectivity but might reflect hemispheric differences in vulnerability to Aβ pathology, underscoring the role of regional vulnerability in determining the distribution of AD pathology.

Fig. 1. Grouping of the subjects.

a Participants were divided into three groups based on the distribution of temporal tau-PET uptake; b Average tau-PET SUVRs for each group; c Comparison of tau load between the groups. Boxplots in panel c represent tau uptake across the three tau asymmetry groups. The horizontal line within each box indicates the median, while the lower and upper box edges denote the first and third quartiles, respectively. Whiskers extend to 1.5 times the interquartile range, and dots represent individual data points. Statistical comparisons between the groups in panel c were performed using ordinary least squares multiple linear regression (tau load ~ age + sex + group). Visualised statistical annotations indicate significance levels Bonferroni-corrected for the number of group comparisons. The sample sizes for each group compared in panel c are following: n = 102 for LA, n = 306 for S, and n = 44 for RA. Laterality index (%) = 100 × (right tau – left tau) / (right tau + left tau). LA left tau asymmetric, S tau symmetric, RA right tau asymmetric, SUVR standardized uptake value ratio; *, p_Bonf < 0.05; ***, p_Bonf < 0.001.

Fig. 2. Association between average connectivity between hemispheres and asymmetry in tau distribution.

a Inter-hemispheric functional/structural connectivity vs absolute global tau laterality; b Fractional anisotropy in main white matter tracts vs absolute global tau laterality; c Inter-hemispheric functional/structural connectivity between tau asymmetry groups; d Fractional anisotropy in main white matter tracts between tau asymmetry groups. Panels a and b show regression lines with 95% confidence intervals, with statistical annotations indicating the standardized effect size and significance level of tau laterality as a predictor of connectivity in ordinary least squares multiple linear regression models (inter-hemispheric connectivity ~ age + sex + global tau load + global tau laterality). Boxplots in panels c and d represent inter-hemispheric connectivity across the three tau asymmetry groups, where the groups were statistically compared using ordinary least squares multiple linear regression models (inter-hemispheric connectivity ~ age + sex + global tau load + group), with the significance levels Bonferroni-corrected for the number of group comparisons. The horizontal line within each box indicates the median, while the lower and upper box edges denote the first and third quartiles, respectively. Whiskers extend to 1.5 times the interquartile range, and dots represent individual data points. Note: 16 subjects were dropped from the analyses within panels b and d after visual quality control of the tract segmentation, resulting in n = 336. LA left tau asymmetric, S tau symmetric, RA right tau asymmetric.

Fig. 3. Association between asymmetry in Aβ and tau distribution.

a Aβ and tau laterality averaging the PET uptake over the whole hemisphere; b Regional-specific asymmetries in Aβ and tau. Panel a shows a regression line with 95% confidence interval, with statistical annotation indicating the standardized effect size and significance of Aβ laterality as a predictor of tau laterality in an ordinary least squares multiple linear regression model (global tau laterality ~ age + sex + global Aβ laterality). Panel b depicts similar models for all homotopic regions, with significance levels FDR-corrected and visualized on brain surface and bar plots. Aβ amyloid-beta; *, p_FDR < 0.05; **, p_FDR < 0.01; ***, p_FDR < 0.001.

Fig. 4. Association between global Aβ laterality and tau laterality in external cohorts.

a All cohorts combined; b Cohorts separately. Both panels show regression lines with 95% confidence intervals, with statistical annotations indicating the standardized effect size and significance of Aβ laterality as a predictor of tau laterality in ordinary least squares multiple linear regression models (global tau laterality ~ age + sex + cognitive impairment status + global Aβ laterality). Aβ amyloid-beta, OASIS-3 Open Access Series of Imaging Studies, A4 Anti-Amyloid Treatment in Asymptomatic Alzheimer’s Disease, ADNI Alzheimer’s Disease Neuroimaging Initiative.

Fig. 5. Longitudinal analysis of the association between baseline Aβ laterality and changes over time in tau laterality.

a Whole A+ sample at global meta-ROI (i.e., whole-brain for each hemisphere); b A+T- subsample at Braak meta-ROIs; c A+T+ subsample at Braak meta-ROIs. The statistical analyses were performed using linear mixed effects models with random intercepts and slopes for time and participants (tau LI ~ time * (age_baseline + sex + Aβ LI_baseline) + [1 + time | participant]), with p-values Bonferroni-corrected for the number of meta-ROIs tested in each subsample. The statistical annotations indicate the standardized effect size and significance level of the interaction between time and baseline Aβ laterality on tau laterality. For visualisation, regression lines represent the modelled mean tau laterality with 95% confidence intervals, plotted for LI_ref ± 2 SD of baseline Aβ laterality, where LI_ref = 0 (i.e., perfect Aβ symmetry). Colorbar indicates baseline Aβ laterality index. Aβ amyloid-beta, LI laterality index.

Fig. 6. Longitudinal analysis of the association between baseline Aβ laterality and changes over time in tau laterality at Braak meta-ROIs, stratified by conversion to T+ status.

a A+T- subsample who stay A+T- throughout their follow-up; b A+T- subsample who progress to A+T+ during their follow-up. The statistical analyses were performed using linear mixed effects models with random intercepts and slopes for time and participants (tau LI ~ time * (age_baseline + sex + Aβ LI_baseline) + [1 + time | participant]), with p-values Bonferroni-corrected for the number of meta-ROIs tested in each subsample. The statistical annotations indicate the standardized effect size and significance level of the interaction between time and baseline Aβ laterality on tau laterality. For visualisation, regression lines represent the modelled mean tau laterality with 95% confidence intervals, plotted for LI_ref ± 2 SD of baseline Aβ laterality, where LI_ref = 0 (i.e., perfect Aβ symmetry). Colorbar indicates baseline Aβ laterality index. Aβ amyloid-beta, LI laterality index.

Fig. 7. Longitudinal analysis within A+ sample over Braak meta-ROIs predicting mPACC score over time with.

a Absolute baseline tau laterality; b Absolute baseline tau laterality after adjusting for tau load; c Absolute baseline Aβ laterality after adjusting for tau laterality, tau load, and Aβ load. The statistical analyses were performed using linear mixed effects models with random intercepts and slopes for time and participants (e.g., model depicted in panel a: mPACC ~ time * (age_baseline + sex + tau LI_baseline) + [1 + time | participant]). The statistical annotations indicate the standardized effect size and significance level of the interaction between time and baseline Aβ/tau laterality on cognitive test score, with p-values Bonferroni-corrected for the number of meta-ROIs tested in each model. For visualisation, regression lines represent the modelled mean cognitive test score with 95% confidence intervals, plotted for LI_ref and LI_ref + 2 SD of baseline Aβ/tau laterality, where LI_ref = 0 (i.e., perfect Aβ/tau symmetry). Aβ amyloid-beta, mPACC modified Preclinical Alzheimer Cognitive Composite, LI laterality index.

Fig. 8. An overview of the data processing steps and analyses.

PET positron emission tomography, dMRI diffusion magnetic resonance imaging, RSfMRI resting-state functional magnetic resonance imaging, FMM flutemetamol, SUVR standardized uptake value ratio, Meta-ROI meta region of interest, EC entorhinal cortex, LA left tau asymmetric, S tau symmetric, RA right tau asymmetric, DTI diffusion tensor imaging, MD mean diffusivity, FA fractional anisotropy, FMin Forceps Minor, CC Corpus Callosum, FMaj Forceps Major, ACT anatomically constrained tractography, SIFT2 spherical-deconvolution informed filtering of tractograms, LH/RH left/right hemisphere, SC structural connectivity, BOLD blood-oxygen-level-dependent imaging, FC functional connectivity, Aβ amyloid-beta.