Migratory jackpot individuals fuel rapid ecotype shifts in Galaxias fishes.

The advent of single-cell RNA sequencing has significantly revolutionized our understanding of cellular diversity in the brain. Large-scale transcriptomic atlases provide invaluable resources for dissecting cellular heterogeneity. Validating atlas consistency is vital for their practical application in neuroscience research. Two recent mouse brain atlases, each independently identifying over 5000 cell clusters using single-cell and nucleus sequencing, underscore the magnitude of this challenge. Assessing cluster replicability, or the degree to which cell clusters identified in one study are consistent with those found in another, is critical to any claim that they represent reproducible biology, such as cell types. Prior work by the BRAIN Initiative Cell Census Network (BICCN) has identified 70 highly replicable cell types in the mouse primary motor cortex. This analysis leveraged seven transcriptomic datasets that differed in preparation methods (single nuclei vs. whole cells) and sequencing platforms (10x Genomics v2, v3, and SMART-Seq). Notably, the authors observed that replicability decreased as cells were subdivided into finer partitions. While single-nucleus sequencing detects fewer transcripts per cell, previous investigations have found it to possess comparable sensitivity to single-cell sequencing for identifying cell types in brain tissue. However, these comparisons were limited to fewer than 12 coarse cell types, and single-nucleus sequencing may not be sufficient to detect disease-associated cell states. From an evolutionary perspective, six retinal cell classes were shown to be highly conserved across 17 species, with transcriptomic similarity decreasing as evolutionary distance increased. These observations raise the following questions: Do these findings extend to the thousands of cell clusters identified in the two recently published whole-brain studies? If so, what are the properties of these replicable clusters? This study deeply characterizes the replicability of cell clusters identified in the two central BICAN reference large-scale mouse brain atlases. We systematically evaluate cluster similarity through marker gene analysis and MetaNeighbor, an expression learning formalism using transcriptome-wide neighbor voting between all cells. We further characterize the replicable clusters, focusing on their regional enrichment, spatial alignment, coordinated gene expression, and representation across diverse mouse and cross-species brain datasets. By rigorously assessing cluster agreement, we aim to highlight both the limitations and strengths of large-scale transcriptomic atlases in capturing cellular heterogeneity as well as providing a firm foundation for subsequent work.