Fig. 3: Inter-regional biological similarity and atrophy.
From: Network spreading and local biological vulnerability in amyotrophic lateral sclerosis
Inter-regional similarity networks reflect the similarity of brain regions according to multiple biological features. We analyze transcriptomic (gene expression) similarity, receptor similarity, laminar similarity, metabolic similarity, and hemodynamic similarity51. The heatmaps visualize these networks. Negative-valued elements are excluded from all analyses. The node-neighbour atrophy correlations are estimated for each inter-regional similarity network. The group-averaged ALS atrophy pattern in Fig. 1c, d is used to derive the correlations. The significance of node-neighbour correlations is assessed with respect to spatial autocorrelation preserving spin tests (nspin = 10,000; FDR corrected; transcriptomic similarity: pspin = 1.55 × 10−2, receptor similarity: pspin = 1.55 × 10−2, laminar similarity: pspin = 0.14, metabolic similarity: pspin = 9.99 × 10−4, hemodynamic similarity: pspin = 0.03). Red dots depict node-neighbour correlations, and boxplots depict the corresponding spin test-estimated null distributions. Additionally, we construct linear regression models to predict ALS cortical atrophy using two variables: (1) weighted mean neighbour atrophy, where the weights are derived from the structural connectome, and (2) weighted mean neighbour atrophy, where the weights come from the interregional similarity matrices. We assess whether adding the second regressor (weighted mean neighbour atrophy based on a specific interregional similarity matrix) improves model fit (adjusted R2), compared to adding a regressor with the same spatial organization. Adjusted R2 is improved only when incorporating data from metabolic similarity matrix. See Fig. S5 for further details.
