Preservation of co-expression defines the primary tissue fidelity of human neural organoids

Abstract

Human neural organoid models offer an exciting opportunity for studying often inaccessible human-specific brain development; however, it remains unclear how precisely organoids recapitulate fetal/primary tissue biology. Here, we characterize field-wide replicability and biological fidelity through a meta-analysis of single-cell RNA-sequencing data for first and second trimester human primary brain (2.95 million cells, 51 datasets) and neural organoids (1.63 million cells, 130 datasets). We quantify the degree to which primary tissue cell-type marker expression and co-expression are recapitulated in organoids across 12 different protocol types. By quantifying gene-level preservation of primary tissue co-expression, we show neural organoids lie on a spectrum ranging from virtually no signal to co-expression near indistinguishable from primary tissue data, demonstrating high fidelity is within the scope of current methods. Additionally, we show neural organoids preserve the cell-type specific co-expression of developing rather than adult cells, confirming organoids are an appropriate model for primary tissue development. Overall, quantifying the preservation of primary tissue co-expression is a powerful tool for uncovering unifying axes of variation across heterogeneous neural organoid experiments.

Cross-ancestry atlas of gene, isoform, and splicing regulation in the developing human brain

“…Gene expression and splicing heritability were estimated using a restricted maximum likelihood algorithm implemented in Julia ( 85 ). Cell-type proportions were estimated for seven major cell classes with CIBERSORTx ( 53 ) using a single-cell reference panel of meta-analyzed cell-type markers (“metamarkers”) from 37 primary fetal brain scRNA-seq datasets comprising ~2.95 million individual nuclei and cells ( 54 ).…”

“…Thus, we hypothesized that the observed decrease in heritability could reflect genetic regulation in specific cell types as their proportions become more heterogeneous. To assess this, we deconvoluted bulk gene expression using a reference panel of cell type–specific marker genes derived from a meta-analysis of 2.95 million single cells and nuclei from the developing human brain ( 53 , 54 ). As expected, estimated cell-type proportions were highly dynamic during this window, with a notable increase in excitatory neurons and a concomitant decrease in progenitor populations (Fig.…”

Identifying Reproducible Transcription Regulator Coexpression Patterns with Single Cell Transcriptomics

“…Similarly, Suresh et al (2023) built aggregate networks using middle temporal gyrus scRNA-seq data from five primate species, demonstrating the conservation of coexpression signals across primates and highlighting human-novel patterns. Lastly, Werner and Gillis (2023) explored the commonalities and differences of single cell coexpression in neural primary versus organoid tissues. Importantly, these works typically focused on the preservation of the global coexpression network structure, rather than any specific gene profile.…”

Cross-site reproducibility of human cortical organoids reveals consistent cell type composition and architecture

“…Three dimensional hCO differentiations have been previously compared across labs using pre-existing single cell-RNA-seq (scRNA-seq) datasets. These studies have compared either consistency in developmental trajectories across hCO protocols (Tanaka et al 2020;He et al 2023) or fidelity of hCO cell types to the in vivo human brain (Bhaduri et al 2020;Werner and Gillis 2023), but could not disambiguate differences caused by cell lines, protocols, or technical artifacts from site-specific implementations of the protocol. Additionally, hCO phenotypes other than scRNAseq have not been compared across multiple labs.…”