scPCIP: cross-species annotation of plant cell types with conserved gene programs

Single-cell atlases have been developed for various plant species and organs, yet cell types are still annotated using marker genes that do not always transfer between species. This study demonstrates that plant cell identities are characterized by gene programs that are conserved across different species. The researchers identified programs for root, leaf, flower, and seed cell types using ortholog groups extracted from 112,905 author-labeled cells.

They then created scPCIP, a tool that can annotate cells from species not included in the training data without the need for a reference. In comparison to 26 held-out datasets, scPCIP outperformed marker-based methods and plant deep-learning approaches, and it also matched supervised classifiers trained on the same atlas. When applied to an independent study, scPCIP transferred its cell type annotations, albeit with slightly lower accuracy.

The conserved core of each program was more effective at separating cell types than the peripheral regions. Surprisingly, most species-specific markers were found to be unstable and turned over across species. Notably, across different organs, leaf bundle-sheath cells exhibited a root endodermis program in 8 out of 12 studies.