Extended Data Fig. 8: Molecular axis predicting cortical mean SA gradient and its biological interpretation across age.

a, Parcel-wise score map at 25 years based on the first component of a partial least squares (PLS) model, relating adult human cortical gene expression (AHBA, parcellated) to the mean SA axis. The component is sign-oriented so that corr(PLS1 score, mean gradient) ≥ 0: warm colours mark parcels whose multigene expression profile predicts higher mean gradient; cool colours mark parcels aligned with lower mean gradient. b, Spatial projection of the top enriched GO term, trans-synaptic signalling. A parcel-wise gene-set score was computed (z-scored expression, coefficient-weighted by PLS1) and mapped to cortex. Co-localization of this map with the PLS1 map (and with the phenotype) links the molecular axis in a to synaptic biology. c, GO themes across age. For each age, genes were ranked by signed PLS1 coefficients and analysed with GOrilla. Cells show the median −log₁₀(FDR q) of terms within each theme, summarizing which processes are most consistently associated with higher mean gradient at different developmental stages (e.g., strong synaptic/vesicle signatures with age-specific modulation of ion transport and occasional extracellular matrix/immune/cell-cycle signals). d, Gene–gradient coupling as a function of age: the Pearson correlation between the PLS1 parcel scores and the mean gradient, with points marked when permutation q < 0.05. This curve indicates when adult-like molecular patterns best predict the spatial distribution of the gradient (stronger earlier, declining into later adulthood in this dataset).