Gene expression profiling enables refined parcellation of cortical layers in the heterogeneous human cerebral cortex
Precise delineation of cortical layers is fundamental for understanding human brain organization, cell-type architecture, and disease-related tissue alterations. However, traditional anatomy-based methods often lack molecular resolution and suffer from inter-observer subjectivity. Here, we present gene expression-defined cortical layers (GD-Ls) using the BayesSpace algorithm, a high-resolution framework for cortical parcellation based on spatial transcriptomics. Compared with traditional anatomy-based approaches, GD-Ls more accurately resolve laminar boundaries and capture fine-scale laminar heterogeneity, including sublayer-like domains within L1, L3, and L6, as well as a molecularly distinct transition zone at the gray-white matter interface. Validation across diverse cortical lobes, multiple spatial platforms, and independent healthy postmortem datasets demonstrates that GD-Ls capture the intrinsic molecular architecture of the cortex irrespective of tissue source. Furthermore, cross-species analyses show that this framework is extensible to macaque and mouse cortices. Crucially, GD-Ls successfully identify subtle laminar disorganization and aberrant cellular and molecular signatures in pathologically altered tissues, which are often missed by conventional histology. Together, GD-Ls provide an objective and reproducible tool for standardized cortical mapping and for identifying early pathological signatures in the human brain. The source code is available on GitHub (https://github.com/YanrongWei/GD-Ls).