Module map of stem cell genes guides creation of epithelial cancer stem cells

Abstract

Self-renewal is a hallmark of stem cells and cancer, but existence of a shared stemness program remains controversial. Here, we construct a gene module map to systematically relate transcriptional programs in embryonic stem cells (ESCs), adult tissue stem cells, and human cancers. This map reveals two predominant gene modules that distinguish ESCs and adult tissue stem cells. The ESC-like transcriptional program is activated in diverse human epithelial cancers and strongly predicts metastasis and death. c-Myc, but not other oncogenes, is sufficient to reactivate the ESC-like program in normal and cancer cells. In primary human keratinocytes transformed by Ras and I kappa B alpha, c-Myc increases the fraction of tumor-initiating cells by 150-fold, enabling tumor formation and serial propagation with as few as 500 cells. c-Myc-enhanced tumor initiation is cell-autonomous and independent of genomic instability. Thus, activation of an ESC-like transcriptional program in differentiated adult cells may induce pathologic self-renewal characteristic of cancer stem cells.

Figure 1. Gene Module Analysis of Stem Cells

(A) Flow chart of the steps in our gene module analysis of stem cells. (B) Mouse stem cell gene module map: a matrix of gene sets (rows) versus expression arrays (columns), where a red (or green) entry indicates that the individual genes within the gene set are significantly induced (or repressed) in the expression array more than expected by chance (FDR<0.05, p<0.01). The intensity of each entry corresponds to the average fold change of all of the genes within the gene set in the array. The arrays are labeled as differentiated cells or the type of stem cell. A subset of significant gene sets is shown; redundant gene sets were removed for clarity. Gene sets defined by gene ontology terms are listed in black; published gene sets defined by expression arrays of stem cells and differentiated cells are listed in purple; gene set defined by RNA interference experiments is listed in orange; gene set defined by genome-wide chromatin immunoprecipitation experiments is listed in blue. Note the clustering of two groups of stem cells, labeled “ESC-like” and “adult tissue stem”.

Figure 2. Activation of ESC Module in Human Cancers

(A) Cancer gene module map: a matrix of stem cell gene modules (columns) and array clinical annotations (rows), where a red (or green) entry indicates that the arrays in which the corresponding module was significantly induced (or repressed) contained more arrays with a given annotation than would be expected by chance (FDR<0.05, p<0.01). The intensity of the entries corresponds to the significance, i.e. −log₁₀ (p-value). A subset of significant annotations is shown; redundant annotations were removed for clarity. (B-D) Expression patterns of ESC-like genes in primary human cancers and corresponding normal tissues are organized by hierarchical clustering. Note the coordinate regulation of the genes in the normal and cancer tissues. The dendrogram at the top of each data represent the similarities among the samples in their expression of ESC-like genes. Tumors are indicated by black branches, and normal tissues by green branches. Kaplan-Meier survival curves of tumors stratified into two classes based on expression of the ESC-like module are shown for data in (C) and (D). (E) Average log₂ expression value of all the genes in the ESC-like module in previously published expression arrays of normal breast and FACS-sorted CD44+CD24-/low tumorigenic breast cancer cells (Liu et al., 2007). A line indicating mean expression value among all samples is shown for each. (F) Correlation of Pearson value to CD44+CD24-/low tumorigenic breast cancer cell signature with average log₂ expression value of ESC-like module in the primary human breast cancers shown in (C).

Figure 3. c-Myc Activates ESC Module in Epithelial Cells

(A) Expression of ESC-like module genes (excluding the proliferation genes) in previously published expression arrays of primary human mammary epithelial cells transduced with indicated oncogenes relative to those cells transduced with GFP (Bild et al., 2006). Each row represents an individual gene within the ESC-like module and each column is an expression array. The level of expression of each gene in each sample relative to the mean level of expression of that gene in the cells transduced with GFP is represented using a red-green color scale. (B) Average log₂ expression values of all of the genes in the ESC-like gene module are expressed as mean ± SEM. (C) Average log₂ expression values of all of the genes in the adult tissue stem gene module are expressed as mean ± SEM. (D) Expression of ESC-like module genes (excluding the proliferation genes) in previous published expression arrays of c-Myc induction in the epidermis of K14-Myc-ER transgenic mice, one and four days after topical application of tamoxifen (Frye et al., 2003). (E) Average log₂ expression values of all of the genes in the ESC-like gene module are expressed as mean ± SEM. (F) Average log₂ expression values of all of the genes in the adult tissue stem gene module are expressed as mean ± SEM.

Figure 4. c-Myc Induces ESC-like State in Cancer

Subcutaneous scid mouse tumors derived from transduced primary human keratinocytes expressing the indicated proteins: (A-C) Each row represents an individual gene and each column is an expression array of a subcutaneous tumor. The level of expression of each gene in each sample was normalized relative to its average expression in all the samples and is represented using a red-green color scale. (A) RNA expression of genes in the ESC-like module. (B) RNA expression of genes with bivalent chromatin domains that are repressed in ESCs (Bernstein et al., 2006). (C) RNA expression of genes in a signature of CD44+CD24-/low cancer cell subpopulation enriched for cancer stem cells (Liu et al., 2007). The CD44+CD24-/low signature is composed of genes that are induced and repressed which is shown on the bar to the right (red is induced; green is repressed). (D) Histology of subcutaneous tumors. (E) Grades of subcutaneous tumors. (F) Subcutaneous tumor tissue: protein expression of keratinocyte differentiation markers keratin 5 (K5), keratin 10 (K10), transglutaminase 1 (Tgase), and keratin 8 (K8), all in orange. Hoechst 33342 nuclear staining in blue.

Figure 5. c-Myc Increases Fraction of Tumor-Initiating Cells

Left: Growth kinetics of tumors formed after subcutaneous injection of transduced primary human keratinocytes expressing the indicated proteins (mean ± SEM). Right: Representative pictures of subcutaneous tumors in scid mice: c-Myc+Ras+IκB tumors on left flank and GFP+Ras+IκB tumors on right flank. (A) 5×10⁵ cells per subcutaneous injection. Picture is at day 18 post-injection. (B) 5×10⁴ cells. The growth kinetics shown for GFP+Ras+IκB is of the one tumor that formed. The growth kinetics shown for E2F3+Ras+IκB is the mean ± SEM of the two tumors that formed. Picture is at day 20. (C) 5×10³ cells. Picture is at day 28. (D) 5×10² cells. The growth kinetics shown for Myc+Ras+IκB is the mean ± SEM of the three tumors that formed. Picture is at day 45.

Figure 6. c-Myc-mediated Tumor Initation Is Cell-Autonomous and Not Linked with Genomic Instability

(A) Subcutaneous scid mouse tumors derived from injection of 5×10⁴ transduced primary human keratinocytes expressing GFP, Ras, and IκB (100% GFP+Ras+IκB), 5×10⁴ cells expressing c-Myc, Ras, and IκB (100% Myc+Ras+IκB), or 2.5×10⁴ cells expressing GFP, Ras, and IκB plus 2.5×10⁴ cells expressing c-Myc, Ras, and IκB (50% GFP+Ras+IκB, 50% Myc+Ras+IκB). Histology and immunohistochemistry with antibody against GFP. (B) Graphical displays of genome-wide DNA copy number alteration of subcutaneous tumors using array-based comparative genomic hybridization. Ratios are plotted on a log₂ scale according to chromosome position.