Evaluating brain cell marker genes based on differential gene expression and co-expression

Abstract

Reliable identification of brain cell types is necessary for studying brain cell biology. Many brain cell marker genes have been proposed, but their reliability has not been fully validated. We evaluated 540 commonly-used marker genes of astrocyte, microglia, neuron, and oligodendrocyte with six transcriptome and proteome datasets from purified human and mouse brain cells (n=125). By setting new criteria of cell-specific fold change, we identified 22 gold standard marker genes (GSM) with stable cell-specific expression. Our results call into question the specificity of many proposed marker genes. We used two single-cell transcriptome datasets from human and mouse brains to explore the co-expression of marker genes (n=3337). The mouse co-expression modules were perfectly preserved in human transcriptome, but the reverse was not. Also, we proposed new criteria for identifying marker genes based on both differential expression and co-expression data. We identified 16 novel candidate marker genes (NCM) for mouse and 18 for human independently, which have the potential for use in cell sorting or other tagging techniques. We validated the specificity of GSM and NCM by in-silico deconvolution analysis. Our systematic evaluation provides a list of credible marker genes to facilitate correct cell identification, cell labeling, and cell function studies.