What Is Conserved Across a Hundred Vertebrate Brains

Cells Cluster by Cell Type, Not by Species

After batch-effect correction with Harmony and unsupervised clustering, the 492,121-cell analysis layer resolves into 43 atlas-level clusters. The most direct reading of integration quality is an inverse Local Inverse Simpson Index (iLISI): a high species iLISI means cells from many species mix together, while a low cell-type iLISI means distinct cell types remain separated. Here the species iLISI is 7.76 against a cell-type iLISI of 1.07, showing that the same cell type from different species shares a common molecular blueprint while remaining sharply distinct from other cell types. Consistent with this, large clusters each span a median of 99 of the 100 analysis-layer species.

Program Conservation Across About 450 Million Years

To ask how much of cell-type identity survives deep evolutionary time, we compared molecular profiles of the white-spotted bamboo shark (Chiloscyllium plagiosum), a cartilaginous fish, with humans. Despite an estimated divergence of about 450 million years, ultra-conserved molecular signatures are preserved at the program level across sharks and humans, from the oldest jawed vertebrates to ourselves.

The Ligand-Receptor Interactome Is More Conserved Than Cell-Internal Programs

Cell-cell communication is coordinated by ligand-receptor pairs, and we tested whether this communication layer is more evolutionarily conserved than the cell-type-internal programs that execute it. Across 64 ordered cell-type pairs spanning 8 broad cell classes in 99 species, ligand-receptor conservation was approximately twice that of the matched differentially expressed gene programs: a median pairwise Jaccard of 0.218 versus 0.108. The two quantities are not measured on the same scale, so the comparison reflects relative ordering rather than absolute magnitude, and the margin is threshold-dependent.

Of the 64 cell-type pairs, 55 showed significantly higher ligand-receptor conservation than DEG-program conservation (paired Wilcoxon, FDR < 0.05; global rank-biserial 0.849). The most conserved pairs were dominated by synaptic adhesion molecules such as NRXN1-NLGN1, CLSTN1-NRXN1, NCAM1-NCAM1 and CADM1-CADM1.

Identity Genes Are as Old as the Brain

Phylostratigraphic age analysis across 15 branches reveals that identity genes are drawn from the oldest stratum that already dominates the brain transcriptome. All 46 core cell-type markers trace to the Euteleostomi-age phylostratum (ps1), as old as the brain-expressed background itself. Identity genes are therefore as old as, not older than, the brain background, and diversity arises from regulatory recombination of this ancient toolkit rather than from new gene birth.

Where Conservation Is Assembled

The comparative picture is built from cell-type-specific stories. Each lineage reveals how an ancient program is conserved across species and where it has been remodeled, from the Shh-class lineage logic of neurons to the two-layer microglial program detected even in the lamprey.