In this study, an in vitro culture system was developed to differentiate diploid and trisomic hESCs into astrocytic progenitor cells (APCs), which were used to determine if gene expression profiles of trisomic APCs remain similar to, or deviate from, diploid APCs following astrocytic differentiation. The data indicate that expression profiles of trisomic BG01V APCs diverge considerably from diploid H9 APCs. Analysis of high-density microarray data revealed numerous, highly significant differences in transcript expression levels in trisomic BG01V APCs relative to diploid H9 APCs. Similar differences were observed when the human astrocytoma cell line, CCF-STTG1, was compared to diploid H9 APCs. Several expression level changes, initially detected by microarray analysis, were subsequently confirmed by qRT-PCR validation. A remarkably similar trend was observed when trisomic BG01V APCs were compared to human glioblastoma patient samples. Taken together, the data suggest that following differentiation along an astrocytic pathway trisomic BG01V APCs exhibit a global gene expression profile that is more similar to astrocytic cancer cells that to normal diploid hESC-derived APCs.
Although trisomic BG01V APCs continue to express markers of differentiated astrocytes, they are clearly distinct from diploid H9 APCs. Despite the high PROM1 expression in BG01V APCs when cultured under adherent conditions, there is insufficient evidence to classify trisomic BG01V APCs as brain tumor-initiating cells [25, 26]. Although trisomic BG01V APCs were derived from multipotent neurospheres that give rise to oligodendrocytes or neurons under different culture conditions (Gopalakrishna-Pillai and Iverson, unpublished), it is also incorrect to consider BG01V APCs equivalent to transformed neural stem cells. Unlike the CCF-STTG1 astrocytoma cells, BG01V APCs do not readily dedifferentiate into non-adherent neurospheres (not shown). Thus, it may be more appropriate to consider trisomic BG01V APCs as one type of premalignant astrocytic stem/progenitor cell. BG01V APCs exhibit an increased rate of proliferation relative to diploid H9 APCs and display gene expression patterns more similar to the astrocytoma cell line and glioblastoma patient samples. Trisomic BG01V APCs may be predisposed to becoming astrocytic cancer cells, but have not yet acquired the full spectrum of mutations required for malignant transformation.
Because the BG01V trisomic hESC line studied here was not derived from the diploid H9 hESC line used for comparison, the results of this analysis cannot be used to identify those changes in gene expression that are predicted to be initiators of premalignant transformation or tumorigenesis (i.e. changes in gene expression that cause transformation). However, group analyses of the data sets can be used to identify differentially expressed transcripts in normal differentiated astrocytes/astrocytic progenitors relative to astrocytic cancer cells. For example, highly significant decreases in expression levels of several transcripts encoding known markers of normal astrocytes including TRPA1, BDNF and MGMT were detected by the analyses. Down regulation of MGMT transcript levels, via hypermethylation of the MGMT promoter, is associated with malignant progression of astrocytomas and is being used in the clinic to identify patients that may benefit the most from treatment with temozolomide [29, 30]. Thus, transcripts exhibiting highly significant over expression in all classes of abnormal astrocytes, including trisomic BG01V APCs, CCF-STTG1 astrocytoma cells and glioblastoma patient samples with respect to diploid H9 APCs (Table 1, GND > C) may be diagnostic markers of transformation and/or potential therapeutic targets (i.e. changes in gene expression that are consequences of transformation). This list includes cell surface expressed markers such as GPRC5, signaling molecules such PIK3R1 and the histone deacetylase, HDAC9, for which small molecule inhibitors are currently under investigation in glioblastoma clinical trials [13, 31].
Neural stem cells with astrocytic character arising in the sub ventricular zone are thought to be one source of origin of gliomas [32, 33]. Donor-derived neural stem cells were recently demonstrated to give rise to tumors of glioneuronal origin in an Ataxia Telangiectasia patient . Slowly cycling cancer stem cells, which are refractory to conventional radiation and chemotherapy, are thought to be a source of both the original tumor as well as recurrent tumors. These astrocytic cancer stem cells create an obstacle for effective treatment of malignant gliomas, yet identification of molecular markers characteristic of brain tumor stem cells is a major challenge complicated by their low numbers, elusive nature, heterogeneity of brain tumors and the difficulty of obtaining abundant quantities of normal human astrocytes that can serve as controls for global expression analyses . As a result, gene expression profiles of rare subpopulations of brain tumor stem cells cannot be readily identified by microarray analyses of brain tumor samples. The system used in this study describes methods for both obtaining sufficient quantities of suitable control cells via directed differentiation of hESCs and using gene expression profiles of these cells to refine the list of putative biomarkers of the astrocytic cancer stem cells through rigorous bioinformatic analyses of the microarray data.
Given their astrocytic, premalignant and stem-like properties, the class of transcripts predicted to be biomarkers of premalignant astrocytic stem/progenitor cells are the GN > CD transcripts (Table 2, Additional file 6, Table S5, Figure 8), and include transcripts encoding numerous potential therapeutic targets and/or cell-surface expressed proteins. In addition to PROM1, CHI3L1 was also identified by the analysis. CHI13L1 (YKL-40) is expressed in a small percentage of glioblastoma cells upon initial diagnosis, but exhibits profound up regulation upon tumor recurrence . The regulator of G protein signaling 5, RGS5, is over expressed in highly angiogenic astrocytomas and RGS5 expression is specifically up regulated in the vasculature of premalignant lesions . FKBP5 is over expressed in gliomas and down regulation of FKBP5 expression using siRNAs suppresses glioma cell growth . IGFBP2 over expression has been demonstrated to promote glioma growth as well as progression from low to high grade in mouse models . Over expression of transcripts encoding transmembrane proteins KCNMB4, a neural-specific β subunit of a large-conductance, calcium-sensitive potassium channel associated with glioma cell growth , and LPHN2, a putative G-protein coupled receptor, were both identified by the comparative microarray analysis. As were ASTN1, an adhesion molecule associated with neuronal migration along astroglial fibers , and GPC4, a cell-surface expressed proteoglycan that may play a role in controlling cell division . Additional over expressed transcripts identified by comparative global microarray analysis using in vitro differentiated trisomic BG01V APCs include those encoding signaling molecules such as the protein phosphatase, PPP2R2B, the purinergic receptor, P2RY5, the ras homolog, RHOU, and others not previously associated with astrocytomas or astrocytic cancer stem cells.
Inherent genetic instability of cultured hESCs renders them susceptible to gain or loss of entire chromosomes and/or discrete chromosomal regions. Gain of chromosomes 12 or 17 has been reported in several other hESC lines [2, 3, 7]. Amplification or deletion of discrete regions of chromosome 20 and 5 as well as mosaic gain of chromosome 12 has also been reported to arise spontaneously in cultured hESC lines . Significantly, neural stem cells derived from some of these hESC variants have been clearly demonstrated to exhibit several features of neoplastic transformation in vivo. Thus, it is highly unlikely that trisomic BG01V hESCs line are unique in their ability to differentiate into premalignant astrocytic stem/progenitor cells upon in vitro directed differentiation. The in vivo evidence of neoplastic transformation of differentiated hESC variants suggests that the propensity toward transformation may be a somewhat common occurrence and underscores the absolute necessity of subjecting all hESC-derived cells to functional characterization prior to their use in therapeutic regimens . Although BG01V hESCs may not be unique in exhibiting features of premalignant transformation following differentiation, the conspicuous differences in expression profiles of BG01V APCs and H9 APCs, combined with the striking similarities in expression profiles of BG01V APCs and glioblastoma samples suggest that gain of chromosomes X, 12 and/or 17 may be one of several routes by which transformation can be initiated in astrocytic progenitor cells. However, we cannot rule out the possibility that genetic events other than trisomy played a role in the initiation of premalignant transformation observed here since the trisomic hESC line, BG01V, is not a derivative of the diploid hESC line, H9.
That a constellation of gain-of-function and/or loss-of-function mutations in multiple genes is required for malignant transformation has been known for decades [41, 42]. Aneuploidy, however, has been associated with cancer for more than a century . Given the high-degree of aneuploidy observed in glioblastoma patient samples, it is difficult to distinguish those genes or chromosomal regions associated with tumor initiation or propagation from those representing random events arising from the inevitable genetic instability common to these high-grade tumors. Comprehensive analysis of chromosomal aberrations in 141 glioma samples identified approximately 35 broad and focal regions of gene amplifications and deletions demonstrating statistically significant associations in human gliomas , and revealed that amplification of a number of chromosomal regions, including chromosomes 12 and 17, met the threshold for significance in these glioma samples, including secondary glioblastomas arising from low-grade gliomas. High-resolution copy number analysis of glioma samples also revealed recurrent gain of multiple sub-regions of chromosome 12 in secondary glioblastomas arising from lower grade astrocytomas . Since recurrent gain of chromosomes 12 or 17 has been observed in a number of karyotypically abnormal hESC lines [2–4, 7], this also suggests that other aneuploid hESC variants might exhibit properties similar to trisomic BG01V cells upon differentiation into astrocytes.
Dot plots for several additional markers associated with gliomas are shown in Additional file 7, Figure S2. Amplification of EGFR (or gain of chromosome 7) is often seen in high-grade gliomas . Not surprisingly, over expression of EGFR in some glioblastoma samples is also observed here (Additional file 7, Figure S2, panel A). Loss of chromosome 10 (or regions of chromosome 10 encoding the PTEN locus) is associated with some de novo glioblastomas, and reduced PTEN expression in some glioblastoma samples is seen here (panel B). IL13RA2 is reportedly over expressed in ~90% of glioblastoma samples. Exon microarray analysis indicates bifurcation of patient samples into two distinct groups; one group exhibits significant over expression of IL13RA2 transcripts, whereas IL13RA2 is expressed at levels similar to controls in the other group of patient samples (D). Relative expression levels of PDGFB (F), PDGFRA (G) PDGFRB (H) are also shown in Additional file 7, Figure S2. Although median expression level of PDGFRB is decreased relative to controls, there is a wide range of expression levels, which may also reflect distinct glioma subtypes. Mutations in a number of genes have also been associated with gliomagenesis, although many are somatic mutations that do not necessarily result in expression level changes . Several are shown in Additional file 7, Figure S2, including AKT1 (C), NRG1 (E), TP53 (I), MDM2 (J), NF1 (K) and RB1 (L). There is no obvious correlation in relative expression levels of these markers in glioblastoma samples and/or premalignant astrocytic progenitors, with perhaps the exception of NRG1 (HER2 ligand) where expression levels are lower in all samples relative to the diploid H9 APCs. Although trisomy for chromosomes 12 and 17 may be one potential precipitating event resulting in transformation of BG01V hESCs into premalignant APCs during differentiation, not all differences in relative gene expression levels can be explained by the trisomy. Dot plots for a number of genes that are known to be expressed in astrocytes, known to be associated with cancer and known to map to chromosomes 12 or 17 are also shown in Additional file 7, Figure S2, including STAT3 (panel M, chromosome 17) where expression levels in glioblastoma, BG01V APCs and CCF-STTG1 samples are higher than H9 APCs, ERBB2 (HER2 - panel N - chromosome 17) where expression levels are higher in BG01V APCs and glioblastoma samples but also high in diploid H9 APCs, and both RARA (panel O, chromosome 17) and RARG (panel P, chromosome 12), where expression levels are higher in BG01V APCs relative to H9 APCs but also lower in glioblastoma samples. Thus, not all transcripts encoded by genes that map to chromosomes X, 12 and/or 17 are over expressed in BG01V APCs and not all over expressed gene transcripts in BG01V APCs map to chromosomes X, 12 and/or 17. We think that most gene transcripts that are over expressed in BG01V APCs relative to H9 APCs are consequences of premalignant transformation rather than causes of premalignant transformation.
Nonetheless, gain of chromosomes X, 12 and/or 17 has also been observed in carcinoma in situ of the testis, which is thought to be the common pre-invasive progenitor cell (or precancerous stem cell) that gives rise to both seminoma and non-seminomatous testicular germ cell tumors . Recurrent amplification of 17q23.2 in some breast tumors has recently been reduced to a 249 kb minimal region including potential tumor driver genes, RPS6KB1 and mir-21 . Finding gains of similar chromosomal regions in tumors whose origins are unrelated to the nervous system suggests that trisomy for these chromosomal regions is a type of moderate aneuploidy common to other precancerous progenitor or stem cells and hypothesized to be an initiator of tumorigenesis [48–50]. This suggests that directed differentiation of trisomic hESC variants along other lineages might be used to simulate early molecular events occurring enroute to malignant transformation of other cancers and to identify diagnostic markers and/or molecular targets amenable to therapeutic intervention.