• We are sorry, but NCBI web applications do not support your browser and may not function properly. More information
Logo of pnasPNASInfo for AuthorsSubscriptionsAboutThis Article
Proc Natl Acad Sci U S A. Jan 25, 2011; 108(4): 1591–1596.
Published online Jan 10, 2011. doi:  10.1073/pnas.1018696108
PMCID: PMC3029739
Medical Sciences

A CD133-related gene expression signature identifies an aggressive glioblastoma subtype with excessive mutations

Abstract

Cancer cells are heterogeneous and, it has been proposed, fall into at least two classes: the tumor-initiating cancer stem cells (CSC) and the more differentiated tumor cells. The transmembrane protein CD133 has been widely used to isolate putative CSC populations in several cancer types, but its validity as a CSC marker and hence its clinical ramifications remain controversial. Here, we conducted transcriptomic profiling of sorted CD133+ and CD133 cells from human glioblastoma multiforme (GBM) and, by subtractive analysis, established a CD133 gene expression signature composed of 214 differentially expressed genes. Extensive computational comparisons with a compendium of published gene expression profiles reveal that the CD133 gene signature transcriptionally resembles human ES cells and in vitro cultured GBM stem cells, and this signature successfully distinguishes GBM from lower-grade gliomas. More importantly, the CD133 gene signature identifies an aggressive subtype of GBM seen in younger patients with shorter survival who bear excessive genomic mutations as surveyed through the Cancer Genome Atlas Network GBM mutation spectrum. Furthermore, the CD133 gene signature distinguishes higher-grade breast and bladder cancers from their lower-grade counterparts. Our systematic analysis provides molecular and genetic support for the stem cell-like nature of CD133+ cells and an objective means for evaluating cancer aggressiveness.

Keywords: molecular profiling, systems biology

The identification of tumor-initiating cancer stem cell (CSC) subpopulations in leukemia (1) and subsequently in breast cancer (2) has led to the hypothesis that tumor proliferation arises largely from these stem/progenitor-like cells and raises the possibility that a similar hypothesis can be applied to a number of solid tumor types (3) including glioblastoma multiforme (GBM) (4), the most deadly form of brain cancer with a diverse cellular phenotype and genetic heterogeneity (57). A CD133+ cell population has been isolated from brain as well as other cancers and has been shown to possess stem-cell properties (4, 8), to be more tumorigenic than the CD133 cells in xenografted animal models (4, 8, 9), and to confer radiation resistance (10). Gene-expression signatures derived from both tumorigenic and normal stem cells have been correlated to expression profiles of patient samples in an attempt to stratify these patients according to their disease aggressiveness (presumably more CSCs indicate greater aggressiveness). For instance, an invasiveness gene signature derived from tumorigenic breast CSCs has been related to poor survival in breast cancer patients, suggesting that CSCs may contribute to tumor invasion and metastasis (11). A recent computational analysis linking an ES cell-like gene expression signature to the poorly differentiated states of aggressive human tumors, including breast cancer and GBM, further strengthens the putative stem-cell origins of cancer cell proliferation (12). However, this ES cell signature was unable to stratify the tumor-initiating CSCs from the nontumorigenic cells in breast cancer (11, 12), implying that tissue type-specific information embedded in the tumor of origin may be lacking in the ES cell signature. Furthermore, CD133 tumor cells also reportedly are capable of initiating tumors in xenografted animal models (1315). Thus, the stem-cell origin of cancer, as viewed through the window of CD133+ tumor cell isolation, and the clinical implications remain to be fully elucidated.

The Cancer Genome Atlas (TCGA) Network recently characterized genomic abnormalities in more than 200 GBM patients (6, 16), laying the foundation for systematic integration of genetic variations, molecular profiles, and clinical phenotypes. We here established a GBM CD133 gene expression signature and conducted extensive computational analyses to relate it to a wide collection of stem-cell and cancer-cell profiles including the TCGA dataset, providing key initial insights to an understudied subpopulation of cells with critical medical importance. We were able to identify a molecular subtype of GBM reflecting aggressive clinical behavior and hypermutated genetic background, both characteristics of CSCs. Because of the difficulty in obtaining pure CD133+ cells in sufficient volume from patients to perform transcriptomic profiling, no such data linking this key cell population to genomic mutations exist to date.

Results

Establishment of a GBM CD133+/− Cell Gene Expression Signature.

To explore the underlying molecular difference between phenotypically distinct CD133+ and CD133 subpopulations in GBM and the stem-cell relatedness of each population, we conducted gene expression profiling experiments comparing the two cell populations by using DNA microarrays. Freshly excised GBM tumor samples were enzymatically dispersed into single-cell suspensions and subjected to immunostaining with an anti-CD133 antibody (Miltenyi Biotec Inc.) and FACS analysis (Fig. 1A). Both CD133+ and CD133 cell populations were isolated from each tumor tissue. We processed 21 fresh GBM tumor samples over the course of 18 mo and observed CD133+ populations (>0.05%) in 15 samples. Sufficient total RNA was obtained from five patient samples for DNA microarray analysis. To derive the GBM CD133 gene signature, we first applied a Wilcoxon rank-sum test to the microarray data with a cutoff P value of 0.05; we then retained genes exhibiting at least twofold difference between the CD133+ and the CD133 cells. After we removed lower-abundance genes (those with the sum of all expression values below 10), we obtained a final list consisting of the 214 most differentially expressed genes, which we designated as the CD133 gene expression signature. The CD133 signature contains two subsets of transcripts: the “CD133-up” subset of 89 transcripts elevated in the CD133+ population (Table S1) and the “CD133-down” subset of 125 transcripts decreased in the CD133+ population (Table S2). (Given that elevated and decreased transcription levels may arise from different mechanisms, we use the terms “up” and “down” transcript levels merely for the convenience of narration without implying mechanism.)

Fig. 1.
Isolation of CD133+/− GBM cells and establishment of a CD133 gene-expression signature. (A) Gating used to isolate CD133+/− cells. Because the samples exhibited significant autofluorescence, gating was performed in the 530/30 (FL1) vs. ...

We hypothesize that correlating the enrichment patterns of the CD133 up and/or down signatures with the compendium of existing expression profiles in a variety of cancers and stem cells will help resolve whether the signature has stem cell-like features and whether it is useful in stratifying tumors as more and less aggressive. To that end, we used and improved an unbiased algorithm developed by Setlur et al. (17) to assess whether a particular gene set (e.g., a gene ontology category or a predefined gene list) was statistically over- or underexpressed in a given mRNA transcript-expression profile. In brief, we calculated Z scores for each gene in a given expression profile—assuming that these expressions have a normal distribution—to minimize the noise arising from different expression profiles obtained across diverse platforms. These Z scores then were converted into corresponding P values to which a logarithmic transformation was applied and were designated as individual gene scores. For a given subset of genes (e.g., the CD133-up set with 89 genes), the gene scores were summed to compute a score for the gene set. The significance of the gene set scores then was determined by running 106 iterations on randomly selected gene sets of the same size to simulate probability. These probabilistic values were used to generate heat maps throughout this study (Figs. 1B, ,2A,2A, ,3,3, and and4)4) with a probabilistic value of 0 representing enrichment of overexpression (red), a probabilistic value of 1 representing enrichment of underexpression (green), and a probabilistic value 0.5 representing nonsignificant changes (black). In essence, in the expression profile for a given patient a red line represents overexpression of either the CD133-up or -down gene set; a green line represents underexpression of the corresponding gene set. This algorithm allows us to investigate the relatedness of our CD133 transcript signature to those from other cancer cell- and stem cell-expression studies.

Fig. 2.
Stem cell relatedness of the GBM CD133 signature. (A) Expression patterns of CD133 signatures in seven replicate hESC samples. (B) Hierarchical clustering of gene-expression profiles of primary GBM samples cultured in serum-containing medium (Serum Group) ...
Fig. 3.
Using the CD133 gene signature for molecular stratification of gliomas. (A) Stratification of different grades of glioma samples using the CD133-up and CD133-down signatures. AC2, grade 2 astrocytoma; GBM, grade 4 astrocytoma; ODG2, grade 2 oligodendroglioma. ...
Fig. 4.
The GBM CD133 gene signature identifies an aggressive subgroup of GBM with hypermutations. (A) Survival curve of CD133-active vs. CD133-inactive/others in three independent GBM datasets. (B) Distribution of genomic mutations among three CD133 classes ...

We first applied this strategy, as proof of principle, to determine how closely the expression patterns of the CD133-up and -down signatures are related to our original microarray dataset comparing the five pairs of CD133+ and CD133 GBM cell populations. An overall overexpression pattern of the CD133-up gene set in the CD133+ populations and of the CD133-down gene set in the CD133 populations was observed in all five paired samples (Fig. 1B). One validation we performed was to compare two signatures obtained using the same algorithm before and after the quantile normalization. These two signatures were almost identical in both size (214 vs. 212) and component (95% overlapped), suggesting that our signature selection algorithm is tolerant of slight variations in sample normalization. The signature also was found to have a higher enrichment score for the CD133-up set than for the CD133-down set for CD133+ cell populations, and vice versa for all CD133 cell populations in each patient. In cross validation, using only four paired samples in each training set, this simple classifier approach achieved 80% performance in classifying the CD133+ and CD133 cell populations (and 100% when a floating P value cutoff was used to maintain the same size of signature in each cross-validation loop). Estimates of cross-validation error on sample sizes this small have high variance and bias, so we focused primarily on leveraging the signature learned from the five paired samples on much larger sets of different but related phenotypes. The small sample size necessitated in the current study was offset by (i) the high quality and novelty of the data, (ii) the choice of a simple signature approach to mitigate overfitting, and (iii) the integration with numerous orthogonal large datasets (see below) to augment our findings and provide strong confirmation of the predictive power of this signature. Although the ordered list of differentially expressed genes certainly would be expected to change significantly with increasing numbers of samples, the differences observed, combined with orthogonal data sets, already provide important insights into the molecular differences between the CSCs and the other cells in the bulk tumor.

Resemblance of the GBM CD133 Signature to Signatures of Human ES Cells and Neural Stem Cells.

To assess the potential stem-cell origins of the CD133-derived cell populations, we evaluated whether the CD133-up and -down signatures display expression patterns that are congruent with the transcription patterns seen in human ES cells (hESCs) (18). A direct comparison of the CD133 signatures with published hESC expression profiles showed that the CD133-up gene set also was overexpressed in all seven hESC cell lines examined (including four FES lines from the University of Helsinki, Helsinki, Finland, and three HS lines from Karolinska University Hospital, Stockholm, Sweden), whereas the CD133-down gene set was underexpressed in all seven hESC cell lines (Fig. 2A) (18). These data support the existence of a shared transcription pattern in the CD133+ GBM cells and hESCs.

To explore further the stem-cell origin of the CD133+ GBM population in the context of brain tissues, we measured the relative transcriptional relationships among the expression profiles of various brain cell samples, including neural stem cells (NSC) and primary GBM tumor cells cultured in NSC-enriching medium or in regular serum medium, by means of our CD133 signatures. Primary GBM cells cultured in neurobasal medium supplemented with basic FGF and EGF (NBE medium) retain an undifferentiated status—a key CSC feature—and recapitulate the GBM phenotype more faithfully than cells cultured in regular serum medium, which promotes cellular differentiation (19). We set forth to study whether our ES cell-resembling CD133 signature is capable of distinguishing GBM cells cultured in NBE or regular serum medium. We used hierarchical clustering to analyze expression profiles of 22 serum-cultured GBM samples, 28 NBE-cultured GBM samples, and three NSC samples as described by J. Lee et al. (19). By using Ward's minimum variance method as clustering algorithm, Pearson correlation as distance function, and our CD133 gene sets as clustering features, we were able to accomplish striking separation of the NBE group from the serum group (Fig. 2B). Of particular interest are the three NSC cultures, which were placed decisively in the middle of the NBE group (red), as they should be if our hypothesis is correct. Another intriguing finding is that four cases in the serum group (Fig. 2B, far left, purple) were separated from both the serum group and the NBE group. These four cases are the lowest-passage GBM cells cultured in serum and have expression patterns most similar to the parent GBM samples because they have not yet diverged away from primary GBM samples (19). Taken together, these data strongly support the idea that the NBE medium does expand a NSC-like population out of primary GBM tumor cells, and these in vitro-enriched CSCs bear an intrinsic connection with freshly sorted GBM CD133 populations from which the CD133 gene signatures were generated.

Stratification of Glioma Patient Populations via the GBM CD133 Gene Signature.

Two major ramifications of the CSC hypothesis can be exploited for clinical applications. (i) CSCs are poorly differentiated and phenotypically emulate more aggressive stage and higher-grade tumors. (ii) The imperative function of CSCs to sustain the cancer and the experimental requirement for CSCs to cause new tumors in transplanted foci suggest a potential role in tumor metastasis. Thus, mapping the stem cell-like signature with gene expression profiles in various cancers may afford objective molecular criteria for tumor grading and for predicting clinical outcomes. The recently established ES cell-like gene signature from 13 ES cell-defining gene sets (each containing dozens to hundreds of genes) has been shown to stratify different grades of gliomas to a certain extent, with the strongest presence in GBM, the highest grade (most malignant) astrocytoma (12). Our CD133+ and CD133 cell populations exhibited similar enrichment patterns of the ES cell signature (Fig. S1), suggesting that GBM per se already is poorly differentiated and that a CSC signature specific for GBM may provide a better approach to the stratification of gliomas. We thus evaluated whether our GBM CD133 signature can serve as a metric to stratify gliomas of distinct clinical grades effectively by examining its enrichment pattern in 181 brain samples encompassing nontumor brain tissues and glioma of varying WHO grades (20). As shown in Fig. 3A, none of the nontumor control samples and lower-grade gliomas (e.g., grade 2 astrocytoma and grade 2 oligodendroglioma) overexpressed the CD133-up signature. As tumors progress toward higher grades (e.g., grade 3 astrocytoma and grade 3 oligodendroglioma), enrichment of the CD133-up signature starts to accrue, with a predominant presence occurring in the GBM (grade 4 astrocytoma) samples. Concordantly, the CD133-down signature was enriched predominantly in the control and lower-grade gliomas as compared with GBMs (Fig. 3A, Lower). These data demonstrate that the presence of the CD133-up gene set is well correlated with the higher-grade aggressive astrocytomas, suggesting that the CD133+ GBM cells closely resemble poorly differentiated GBM cells, positing better representation of the stem-cell origin of GBM than the CD133 cells, which mimic normal brain tissue and lower-grade tumors.

Identification of GBM Subpopulations via the GBM CD133 Gene Signature.

Although the CD133 signatures effectively separate GBM from lower-grade gliomas (Fig. 3A), the variation of their enrichment patterns within GBM patients remains conspicuous (Fig. 3A, Right). Recently, the heterogeneous GBM populations have been clustered into four molecular subtypes, namely Proneural, Neural, Classical, and Mesenchymal, based on gene expression profiles (16, 21). We mapped our CD133 gene signatures onto the four molecular subgroups defined by the TCGA network with a total of 173 patients (16). The most prominent enrichment of the CD133 signature occurs in the Proneural cluster, with diminishing appearance in other subtypes (Fig. 3B). Interestingly, it has been reported that the Proneural cluster was more less responsive than the other clusters to a more intensive treatment regime and demonstrated a paradoxical (but statistically nonsignificant) trend toward longer survival (16). We address this paradox by reclassifying the GBM patients into the molecularly based subclasses described below.

The overexpression of the CD133-up signature and the underexpression of the CD133-down signature in stem cells and more aggressive gliomas prompted us to search for GBM molecular subtypes based on the enrichment pattern of our CD133-up and -down gene sets. We devised the following criteria to regroup the 173 TCGA GBM samples into three classes.

  • i)The CD133-active class (43 patients): Either of the two signatures supports the activation of CD133, and the nonsupporting signature does not oppose it.
  • ii)The CD133-inactive class (16 patients): Either of the two signatures supports the inactivation of CD133, and the nonsupporting signature does not oppose it.
  • iii)The CD133-semiactive class (114 patients): All remaining patients that fall outside the active and inactive classes.

We then explored the clinical relevance of these GBM classes by correlating them with reported patient outcomes from the TCGA data. We observed that the CD133-active class contains a higher number of younger patients. However, in contrast to patients classified in the Proneural subtype, who survive longer (16), the patients in the CD133-active class exhibited shorter survival than patients in the CD133-inactive class (Fig. 4A, Left). We validated this finding in two additional datasets using survival curves (21, 22); in both datasets, the CD133-active class showed much shorter survival than the rest of patients (inactive and semi-active classes combined) (Fig. 4A, Center and Right). Thus, the CD133 signature identifies a set of younger patients with a more aggressive subtype within GBM.

Two clinical benefits from the identification of the CD133-active class suggest themselves. First, one can use this classification to design trials more effectively for both old and new drugs. Second, although currently there is no effective treatment for the CD133-active class, this information can be useful in providing patients with a realistic assessment so they can decide how to move forward with their disease.

To seek the genetic root supporting the aggressive phenotype of the CD133-active GBM subtype, we examined genomic abnormalities underlying the three CD133 GBM subclasses. TCGA efforts unveiled a total of 747 mutations of 414 genes in 114 patient samples. Strikingly, the CD133-active class, with only 28 patients (25% of the 114 patients for whom mutation data were available), accounts for more than half (399/747) of all the mutations identified. The average mutation rate per patient in the CD133-active class is four times higher than in the CD133- inactive class and three times higher than in the CD133-semiactive class. Fig. 4B illustrates the distribution of all gene mutations among the three CD133 GBM classes; frequently mutated genes [e.g., epidermal growth factor receptor (EGFR), isocitrate dehydrogenase 1 (IDH1), Neurofibromin 1 (NF1), platelet-derived growth factor receptor (PDGFR), phosphatase and tensin homolog deleted on chromosome 10 (PTEN), and tumor protein p53 (TP53)] are highlighted. Although the majority of mutations occur in the phenotypically aggressive CD133-active subtype, no particular mutation pattern of any specific genes across the three subtypes is observed, suggesting that combinatorial stochastic (as opposed to particular) genetic aberrations contribute, in a quantitative manner, to the tumorigenic properties of CSCs. The bottom line is that these tumors appear to have an increased general mutation rate.

Staging Other Cancer Types by Using the GBM CD133 Signature.

To explore whether the GBM CD133 signature has a far-reaching discriminating power in other tumor types, we performed the same enrichment analyses using expression profiles obtained from 189 breast cancer (23) and 157 bladder cancer samples (24). A compelling enrichment pattern of the CD133-up signature was observed in higher-grade breast cancers (grade 3) compared with lower grades (Fig. S2), whereas the CD133-down signature showed an opposite but less prominent expression-pattern association. A similar observation was made in bladder cancers. These results support the notion that cancers originating from distinct tissue types may share some fundamental properties, purportedly at the stem-cell level; our CD133 gene signature represents at least part of these fundamental stem-cell features. Nonetheless, tissue-specific factors derived from corresponding tumor tissues will lead to more precise classification of the respective tumors.

Discussion

Genetic alterations and cytopathological heterogeneity are hallmarks of GBM and contribute to the intractability of the disease. The hypothesis that cancer arises from a tumorigenic subpopulation denoted as “cancer stem cells” affords a unique developmental perspective to interrogate cancer cell heterogeneity for better understanding disease pathogenesis and for developing targeted therapy (25). Full realization of CSC clinical potential is dependent on the rigorous delineation of CSC identity in terms of cell-surface markers and the underlying molecular compositions. Of particular interest and the subject of much debate is the transmembrane protein CD133, the most commonly adopted CSC marker for several cancer types, including GBM. To test its validity as a GBM stem-cell marker and its clinical utility, we conducted transcriptomic profiling of highly purified CD133+/− GBM cells to establish a CD133 transcript signature. We then computationally mapped this signature to a score of published expression profiles of stem cells and cancers of various differentiation stages. We observed significant enrichment of the GBM CD133 signature in stem cells and higher-grade human cancers. Furthermore, the CD133 signature allows us to identify a clinically aggressive molecular subtype of GBM characterized by excessive genomic mutations. This integrated systems approach using CSC sorting, expression profiling, and computation comparison with public domain transcriptome and genome data enables an initial comprehensive view of the CD133+/− cell populations, lending support to the stem-cell origin of CD133+ cells. The CSC gene panel established in this work also lays the groundwork for developing further clinical tests for multiple cancers.

Gene-expression profiling is powerful for classifying diseased populations, and a wealth of transcriptomic data from cancer patients has been produced and deposited into the public domain. Deriving stem-cell gene signatures for cancer classification is relatively new, with varying demonstrable clinical utility (26). Among the pioneering works are an invasiveness gene signature consisting of 186 genes derived from breast cancer-initiating cells (11), an hESC-like signature derived from 13 ES cell-related subsets ranging in size from hundreds to 1,000 genes (12), and a list of 117 genes differentially expressed in CD133+ and CD133 GBM cells (13). We evaluated the effectiveness of these stem-cell signatures in classifying the same cohort of glioma and astrocytoma patient samples by using the same enrichment algorithm employed in this study; none achieved the same level of separation as our GBM CD133 signature (Figs. S3 and S4), suggesting that unique features embodied in our gene signature contribute to its robust performance. A gene ontology analysis revealed that more than one third of the CD133-up genes are involved in cell cycle processes, in particular mitosis, whereas a significant number of CD133-down genes are involved in immune response. Cross comparison of our GBM CD133 signature genes with the other gene lists identified only one gene overlapping with the invasiveness breast cancer signature (11) and three overlapping with the GBM CD133 panel reported earlier (13). Although the former result is not too surprising, because different cell-surface marker panels (i.e., CD44 and CD24) were used for isolating the breast CSCs, the mere three-gene overlap (including CD133 itself) between the two GBM CD133 panels is indeed perplexing, because both datasets were generated from GBM tumors. One possibility is that we are sampling very different cellular populations. Our strategy of using freshly removed tumor samples rather than cultured primary cell lines suggests that our signature is more relevant to in vivo disease. In addition, the different cell-sorting schema (e.g., flow cytometer versus magnetic beads) used by different research groups may lead to varying purity of cellular components with corresponding phenotypes. The fact that we observed a CD133+ cell population in fewer GBM patients and at a smaller fraction among total tumor cells than observed in other reports suggests a more stringent cell-sorting scheme and possibly reflects a higher cell purity in our study. More recently, additional gene expression profiling data comparing CD133+ vs. CD133 GBM cells have become available, including molecular signatures containing 41 genes and 24 genes, respectively (27, 28). A direct comparison with our GBM CD133 gene signature revealed no overlapping genes at all, further demonstrating the heterogeneous nature of the disease and the need for continued intensive study of increasing numbers of samples. Moreover, none of the published GBM CD133 signatures to date leveraged the vast amount of TCGA GBM genomic data illustrated in this study.

The true origin of CSCs remains elusive: Both normal stem cells and dedifferentiated tumor cells have been proposed as possible sources. In any case, it is believed that during the course of tumorigenesis CSCs imperatively accumulate multiple genetic alterations which would lead to an altered gene expression spectrum in CSCs. No direct link between gene expression profiles of CSCs and the genomic mutation spectrum has been established. We took advantage of the comprehensive efforts by the TCGA network, which built a valuable framework for integrating genomic and transcriptomic datasets, and discovered that the CD133-active GBM subtype harbors dramatically more genetic mutations than the CD133-inactive and CD133-semiactive subtypes, lending strong genetic support for the stem-cell origin of CD133+ cells. The increased mutation spectra raise the fascinating possibility that the loads of somatic mutation vary enormously in different glioblastoma types, and patients who have higher mutation loads may have shorter survival than patients with lower mutation load.

Although CD133 is not a universal marker for brain CSCs, transcriptomic profiling of cell populations based on the presence or absence of this single protein gave rise to a 214-transcript signature related to cancer patient stratification that provided fresh insight into tumorigenesis. Although establishing the CD133 signatures entailed the use of highly purified GBM CD133+/− cells, all subsequent cross-comparison analyses were performed with legacy expression profiles generated from bulk tumor samples. In this manner, we leveraged the necessarily small sample size of our highly purified CD133+/− cells with validation in much larger sample sets. Thus, clinically amenable molecular tests may be developed by profiling unsorted tumor cells. By virtue of its similarity with stem cells and aggressive human cancers, the CD133 gene signature supports the stem-cell origin of CD133+ GBM cells. The aggressive molecular subtype of GBM identified in this study with the hypermutated genotype will be subjected to future clinical studies for developing targeted therapy.

Materials and Methods

All tumor specimens were collected from the Swedish Neuroscience Institute (Seattle, WA). Fresh GBM single cell suspension were stained with a phycoerythrin (PE)-conjugated CD133 antibody and sorted with BD Influx cell sorter. DNA microarray analysis was performed as described previously (29). Details are provided in SI Materials and Methods.

Supplementary Material

Supporting Information:

Acknowledgments

We thank Dr. David Galas, Dr. Gilbert S. Omenn, Dr. Leslie Chen, and Christopher Lausted for critical reading of the manuscript. This work was supported by grants from the National Institutes of Health (NanoSystems Biology Cancer Center/U54 CA119347; NIGMS Center for Systems Biology/P50 GM076547; National Center for Integrative Biomedical Informatics/U54 DA021519; P01 DK053074; R21 CA135339-01A2, and a Howard Temin Pathway to Independence Award in Cancer Research/R00 CA126184-03), the James S. McDonnell Foundation, the Roy J. Carver Charitable Trust, the University of Luxembourg, and the Swedish Neuroscience Institute Foundation.

Footnotes

The authors declare no conflict of interest.

This article contains supporting information online at www.pnas.org/lookup/suppl/doi:10.1073/pnas.1018696108/-/DCSupplemental.

References

1. Lapidot T, et al. A cell initiating human acute myeloid leukaemia after transplantation into SCID mice. Nature. 1994;367:645–648. [PubMed]
2. Al-Hajj M, Wicha MS, Benito-Hernandez A, Morrison SJ, Clarke MF. Prospective identification of tumorigenic breast cancer cells. Proc Natl Acad Sci USA. 2003;100:3983–3988. [PMC free article] [PubMed]
3. Reya T, Morrison SJ, Clarke MF, Weissman IL. Stem cells, cancer, and cancer stem cells. Nature. 2001;414:105–111. [PubMed]
4. Singh SK, et al. Identification of human brain tumour initiating cells. Nature. 2004;432:396–401. [PubMed]
5. Furnari FB, et al. Malignant astrocytic glioma: Genetics, biology, and paths to treatment. Genes Dev. 2007;21:2683–2710. [PubMed]
6. Anonymous; Cancer Genome Atlas Research Network Comprehensive genomic characterization defines human glioblastoma genes and core pathways. Nature. 2008;455:1061–1068. [PMC free article] [PubMed]
7. Parsons DW, et al. An integrated genomic analysis of human glioblastoma multiforme. Science. 2008;321:1807–1812. [PMC free article] [PubMed]
8. O'Brien CA, Pollett A, Gallinger S, Dick JE. A human colon cancer cell capable of initiating tumour growth in immunodeficient mice. Nature. 2007;445:106–110. [PubMed]
9. Singh SK, et al. Identification of a cancer stem cell in human brain tumors. Cancer Res. 2003;63:5821–5828. [PubMed]
10. Bao S, et al. Glioma stem cells promote radioresistance by preferential activation of the DNA damage response. Nature. 2006;444:756–760. [PubMed]
11. Liu R, et al. The prognostic role of a gene signature from tumorigenic breast-cancer cells. N Engl J Med. 2007;356:217–226. [PubMed]
12. Ben-Porath I, et al. An embryonic stem cell-like gene expression signature in poorly differentiated aggressive human tumors. Nat Genet. 2008;40:499–507. [PMC free article] [PubMed]
13. Beier D, et al. CD133(+) and CD133(-) glioblastoma-derived cancer stem cells show differential growth characteristics and molecular profiles. Cancer Res. 2007;67:4010–4015. [PubMed]
14. Shmelkov SV, et al. CD133 expression is not restricted to stem cells, and both CD133+ and CD133- metastatic colon cancer cells initiate tumors. J Clin Invest. 2008;118:2111–2120. [PMC free article] [PubMed]
15. Chen R, et al. A hierarchy of self-renewing tumor-initiating cell types in glioblastoma. Cancer Cell. 2010;17:362–375. [PubMed]
16. Verhaak RG, et al. Cancer Genome Atlas Research Network Integrated genomic analysis identifies clinically relevant subtypes of glioblastoma characterized by abnormalities in PDGFRA, IDH1, EGFR, and NF1. Cancer Cell. 2010;17:98–110. [PMC free article] [PubMed]
17. Setlur SR, et al. Integrative microarray analysis of pathways dysregulated in metastatic prostate cancer. Cancer Res. 2007;67:10296–10303. [PubMed]
18. Skottman H, et al. Gene expression signatures of seven individual human embryonic stem cell lines. Stem Cells. 2005;23:1343–1356. [PubMed]
19. Lee J, et al. Tumor stem cells derived from glioblastomas cultured in bFGF and EGF more closely mirror the phenotype and genotype of primary tumors than do serum-cultured cell lines. Cancer Cell. 2006;9:391–403. [PubMed]
20. Sun L, et al. Neuronal and glioma-derived stem cell factor induces angiogenesis within the brain. Cancer Cell. 2006;9:287–300. [PubMed]
21. Phillips HS, et al. Molecular subclasses of high-grade glioma predict prognosis, delineate a pattern of disease progression, and resemble stages in neurogenesis. Cancer Cell. 2006;9:157–173. [PubMed]
22. Murat A, et al. Stem cell-related “self-renewal” signature and high epidermal growth factor receptor expression associated with resistance to concomitant chemoradiotherapy in glioblastoma. J Clin Oncol. 2008;26:3015–3024. [PubMed]
23. Sotiriou C, et al. Gene expression profiling in breast cancer: Understanding the molecular basis of histologic grade to improve prognosis. J Natl Cancer Inst. 2006;98:262–272. [PubMed]
24. Sanchez-Carbayo M, Socci ND, Lozano J, Saint F, Cordon-Cardo C. Defining molecular profiles of poor outcome in patients with invasive bladder cancer using oligonucleotide microarrays. J Clin Oncol. 2006;24:778–789. [PubMed]
25. Tu LC, Foltz G, Lin E, Hood L, Tian Q. Targeting stem cells-clinical implications for cancer therapy. Curr Stem Cell Res Ther. 2009;4:147–153. [PMC free article] [PubMed]
26. Price ND, Foltz G, Madan A, Hood L, Tian Q. Systems biology and cancer stem cells. J Cell Mol Med. 2008;12:97–110. [PMC free article] [PubMed]
27. Lottaz C, et al. Transcriptional profiles of CD133+ and CD133- glioblastoma-derived cancer stem cell lines suggest different cells of origin. Cancer Res. 2010;70:2030–2040. [PubMed]
28. Garcia JL, et al. Molecular analysis of ex-vivo CD133+ GBM cells revealed a common invasive and angiogenic profile but different proliferative signatures among high grade gliomas. BMC Cancer. 2010;10:454. [PMC free article] [PubMed]
29. Foltz G, et al. Genome-wide analysis of epigenetic silencing identifies BEX1 and BEX2 as candidate tumor suppressor genes in malignant glioma. Cancer Res. 2006;66:6665–6674. [PubMed]

Articles from Proceedings of the National Academy of Sciences of the United States of America are provided here courtesy of National Academy of Sciences

Formats:

Related citations in PubMed

See reviews...See all...

Cited by other articles in PMC

See all...

Links

Recent Activity

Your browsing activity is empty.

Activity recording is turned off.

Turn recording back on

See more...