Table 1.
Published transdifferentiation protocols used to evaluate our genomic model of cell identity.
Figure 1.
Transdifferentiation factors are more highly expressed in target cell types and more PcG repressed in source cell types.
(A) Gene expression and H3K27me3 histone modification levels as measured by RNA-seq and ChIP-seq, respectively, are shown for MyoD1, a factor that converts fibroblasts to myoblasts [12]. Reads are displayed (units of reads per ten million mapped reads) across the MyoD1 locus and 1 kb regions flanking the gene. The arrow above the gene structure denotes direction of transcription. Data from [55], [56]. (B) Differential expression and modification levels are shown for all transcription factors [15] (n = 1,356) annotated in the mouse genome (grey points), including MyoD1 (blue point). (C,D) Similar plots to (A,B) are shown for factors that convert fibroblast to neural stem cells (SOX2, FOXG1, POU3F2 [14]) and a TF with an opposite genomic pattern (TWIST1). The box in the lower right-hand quadrant highlights nine other TFs (black points) (of 1,447 total annotated human TFs [15]) with differential expression and modification levels similar to the transdifferentiation factors (blue points). Data from ([57], [58]) and the Roadmap Epigenomics Project (http://roadmapepigenomics.org). (E,F) Similar plots to (A, B) are shown for factors that convert liver to pancreas (Pdx1 [21]; blue point), pancreas to liver (Cebpa and Cebpb [22]; orange points), and three other TFs with similar genomic patterns (Id1, Pax6, Nkx6-1; black points). Data from [59]–[61] and two other public datasets (Table S1).
Figure 2.
Genome-wide screening with differential expression and H3K27me3 levels significantly enriches for transdifferentiation factors.
(A) We compared six human and eight mouse pairs of cell types with known transdifferentiation factors (Table 1) and plotted the differential expression (x-axis) and differential H3K27me3 modification (y-axis) for these factors (black points) as well as all other TFs in the genome (grey points). We also indicated (blue points) the subset of these other TFs that were experimentally tested for their ability to convert cell types, but not included in the conversion protocols (Table S2). Each point represents 1 gene in 1 pair of cell types. Most transdifferentiation factors are more highly expressed in the target cell (right side of the y-axis) and more highly H3K27me3 modified in the source cell (below the x-axis). The marginal distributions depict the differential expression (top) and modification (right) of transdifferentiation factors (black curve) compared to all other transcription factors (grey curve), as well as the subset that were experimentally tested (blue curve). (B) We ranked all transcription factors by differential expression (orange), differential H3K27me3 modification (blue), or a combination of both (black) and then calculated the percentage of tested transdifferentiation factors (y-axis) as a function of the position in the ranked list. (C) Plot similar to (A) showing differential H3K9me3 levels in place of H3K27me3. In contrast to H3K27me3 levels, transdifferentiation factors exhibit on average only a slight differential H3K9me3 modification. Six human and four mouse pairs of cell types were used in this analysis, as H3K9me3 modification profiles were not available for three mouse tissues (myoblast, heart, neuron; Table 1). (D) H3K9me3 modification provides minimal information beyond expression for identifying transdifferentiation factors. Data sources are listed in Table S1.
Figure 3.
Similar genes are differentially expressed and PcG modified when comparing fibroblasts and pancreatic islets in both human and mouse.
(A) Genome browser view of RNA-seq and ChIP-seq data over several genes that are more highly expressed in human pancreatic islet cells and more strongly H3K27me3 modified in fibroblasts. Axes similar to Figure 1A. (B,C) Similar sets of genes are differentially expressed and modified when comparing fibroblasts and pancreatic islets in both human and mouse. Plots similar to Figure 1B. Bold labels denote genes shown on both human and mouse plots. Data from [57]–[64], the Illumina Human Bodymap (http://www.illumina.com), the Roadmap Epigenomics Project (http://roadmapepigenomics.org), and two other public datasets (Table S1).
Figure 4.
Reprogramming factors SOX2 and POU5F1 are PcG repressed and DNA methylated, respectively, in fibroblasts.
We compared PcG repression and DNA methylation in fibroblasts and an embryonic stem cell line (H1 ESC). (A) Of the six original reprogramming factors [31], [33] (labeled points), SOX2 is the most significantly PcG repressed in fibroblasts. Plot layout similar to Figure 1B. (B) In contrast, POU5F1 is the most differentially methylated reprogramming factor. Differential methylation is shown in units of absolute methylation score (ams). Data from [57], [58] and the Roadmap Epigenomics Project (Table S1).
Figure 5.
PcG repression of key transcription factors form epigenetic barriers between adult cell types.
(A) Our results suggest that PcG repression of key transcription factors help form the barriers between adult cell types, as depicted by Waddington's epigenetic landscape [41]. Conceptual model based on Fig. 1F. (B) Ectopic expression of endogenously repressed transcription factors overcomes these barriers to convert one cell type to another. For example, expressing Pdx1 in liver cells, where it is PcG repressed, converts them to pancreatic islet cells [21], where it is expressed. Expressing Cebpa in pancreatic islet cells, where it is PcG repressed, converts them to liver cells [22], where it is expressed. Positive autoregulation of transdifferentiation factors would stabilize the newly converted cell identity [7].