Spatial gradients of gene expression, amyloid PET and functional connectome for predicting regional rates of tau‐PET increase: towards tau‐PET outcome ROIs in clinical trials.

Abstract

AbstractBackgroundIn clinical trials, an important question is which brain regions show high tau accumulation within 1‐2 years. Building upon our previous epicenter‐connectivity‐based distance model for predicting the progression of tau‐PET accumulation from initial seed regions of high tau (epicenters) to other regions (Franzmeier et al. Sci Adv. 2020), we incorporate here epicenter‐based distance within spatial gradients of gene expression and amyloid‐PET to enhance the predictive power.MethodWe included 351 participants encompassing Ab(+) non‐demented subjects (controls & MCI), characterized by cross‐sectional 11‐F florbetaben amyloid‐PET and longitudinal 11‐F flortaucipir tau‐PET (longitudinal, n = 172, mean/range = 1,7 yrs (0.7 – 5.2 yrs)) from ADNI. For transcriptomic gene expression maps obtained from the Allen Human Brain Atlas, we computed for each of the 18,686 genes maps spatial correlations with the average tau‐PET map in a randomly selected training sample (n = 106), and retained the top 5% of positively/inversely tau‐correlating gene maps. For each imaging modality, spatial gradients were subsequently computed across amyloid‐PET scans in the training sample, the subset of gene expression maps, and functional connectivity (FC) matrices from Human Connectome project. In the test sample (n = 245), tau‐PET epicenters at baseline (Franzmeier et al. Sci Adv 2020) were projected onto the gradient maps and epicenter‐based gradient distance was computed. In step‐wise spatial regression analyses, we tested which gradient distances predicted the spatial patterns of cross‐sectional and longitudinal tau‐PET accumulation, accounting for spatial autocorrelations.ResultFor cross‐sectional tau‐PET (Fig. 1A), a total of 54.6% (p < 0.0001) of the spatial tau pattern could be explained, where an amyloid‐PET gradients and a FC gradient (spanning the transmodal axis from primary sensory to DMN regions) showed the strongest effects (Fig. 2, upper row & Fig. 3A). For longitudinal tau‐PET (Fig. 1B), a total of 47.8% (p < 0.0001) of tau increases could be explained, where gene‐expression gradients showed the strongest effects (Fig. 2 lower row & Fig. 3B).ConclusionEpicenter‐based gradient distances of connectivity and molecular brain properties show additive, high predictive accuracy for predicting regions of high future tau‐accumulation, providing an attractive approach to define tau‐PET outcome ROIs in clinical trials.