Defining developmental potency and cell lineage trajectories by expression profiling of differentiating mouse embryonic stem cells

Abstract

Biologists rely on morphology, function and specific markers to define the differentiation status of cells. Transcript profiling has expanded the repertoire of these markers by providing the snapshot of cellular status that reflects the activity of all genes. However, such data have been used only to assess relative similarities and differences of these cells. Here we show that principal component analysis of global gene expression profiles map cells in multidimensional transcript profile space and the positions of differentiating cells progress in a stepwise manner along trajectories starting from undifferentiated embryonic stem (ES) cells located in the apex. We present three 'cell lineage trajectories', which represent the differentiation of ES cells into the first three lineages in mammalian development: primitive endoderm, trophoblast and primitive ectoderm/neural ectoderm. The positions of the cells along these trajectories seem to reflect the developmental potency of cells and can be used as a scale for the potential of cells. Indeed, we show that embryonic germ cells and induced pluripotent cells are mapped near the origin of the trajectories, whereas mouse embryo fibroblast and fibroblast cell lines are mapped near the far end of the trajectories. We suggest that this method can be used as the non-operational semi-quantitative definition of cell differentiation status and developmental potency. Furthermore, the global expression profiles of cell lineages provide a framework for the future study of in vitro and in vivo cell differentiation.

Figure 1

Global gene expression profiles of 32 different cell types. Names of the cells and their abbreviated forms are shown. Color coding corresponds with the colors in all other figures. DNA microarray analysis was carried out in duplicate: for each cell type, two independent cell cultures were prepared and used for separate DNA microarray analysis. Reproducibility between two microarray data for each cell type was very high as shown by the Pearson correlation coefficient (mean 0.9927; SD 0.0043; range 0.9788–0.9966). The average of two samples is shown here. (A) Hierarchical clustering analysis. (B) Principal component analysis (PCA). Individual cell types are mapped in the 3D space according to the first three principal components (PC1, PC2 and PC3). A movie file is available as Supplementary Video S1.

Figure 2

Average expression levels of selected genes correlate well with the locations of each cell type in PCA coordinates. See Fig. 1A for the abbreviated names of the cells. Each point shows DNA microarray data for each sample; therefore, each cell type is represented by two points. (A) Primitive endoderm lineage, represented by the list of 1738 genes (Supplementary Table S1). (B) Trophoblast lineage, represented by the list of 2311 genes (Supplementary Table S2). (C) Primitive ectoderm/neural lineage, represented by the list of 2463 genes (Supplementary Table S3).

Figure 3

Global gene expression profiles of 44 different cell types. In addition to 32 cell types in the Fig. 1, 12 cell types are added for the analysis. Names of the cells and their abbreviated forms are shown. Color coding of cells are consistent throughout the paper. Average of two samples is shown. (A) Hierarchical clustering analysis. (B) Principal component analysis (PCA). Individual cell types are mapped in the 3D space according to the first three principal components (PC1, PC2 and PC3). A movie file is available Supplementary Video S2.