Abstract
Deregulation of the polycomb group gene BMI-1 is implicated in the pathogenesis of many human cancers. In this study, we have investigated if the Ewing sarcoma family of tumors (ESFT) expresses BMI-1 and whether it functions as an oncogene in this highly aggressive group of bone and soft tissue tumors. Our data show that BMI-1 is highly expressed by ESFT cells and that, although it does not significantly affect proliferation or survival, BMI-1 actively promotes anchorage-independent growth in vitro and tumorigenicity in vivo. Moreover, we find that BMI-1 promotes the tumorigenicity of both p16 wild-type and p16-null cell lines, demonstrating that the mechanism of BMI-1 oncogenic function in ESFT is, at least in part, independent of CDKN2A repression. Expression profiling studies of ESFT cells following BMI-1 knockdown reveal that BMI-1 regulates the expression of hundreds of downstream target genes including, in particular, genes involved in both differentiation and development as well as cell-cell and cell-matrix adhesion. Gain and loss of function assays confirm that BMI-1 represses the expression of the adhesion-associated basement membrane protein nidogen 1. In addition, although BMI-1 promotes ESFT adhesion, nidogen 1 inhibits cellular adhesion in vitro. Together, these data support a pivotal role for BMI-1 ESFT pathogenesis and suggest that its oncogenic function in these tumors is in part mediated through modulation of adhesion pathways. [Cancer Res 2008;68(16):6507–15]
Introduction
Members of the Ewing sarcoma family of tumors (ESFT) are characterized by the expression of chimeric fusion oncogenes, most commonly EWS-FLI1 (reviewed in ref. 1). EWS-FLI1 transforms NIH-3T3 fibroblasts and its knockdown in ESFT cells dramatically inhibits tumorigenicity (2). In contrast, EWS-FLI1 induces a p53-dependent cell cycle arrest in primary human fibroblasts and loss of p16 is required for the transformation of primary murine fibroblasts (reviewed in ref. 1). Appropriate suppression of innate tumor suppressor pathways is, therefore, necessary for EWS-FLI1–mediated malignant transformation. Unfortunately, evaluation of primary ESFT samples has, thus far, yielded little insight into the mechanism of this inactivation in vivo. Only 25% of ESFT cases exhibit either mutation of p53 or deletion of the p16INK4a/ARF locus, and although these patients exhibit a worse prognosis, they clearly represent a minority of cases (3).
Polycomb group family proteins are central players in the maintenance of stem cell self-renewal and pluripotency as well as in the control of cellular differentiation and development (4–6). Polycomb proteins assemble into large complexes, termed PRC1 and PRC2, to repress transcription through modulation of the chromatin structure and their expression is frequently deregulated in cancer (7–9). The PRC1 gene bmi-1 was first identified as an oncogene that collaborates with c-myc in a murine model of lymphomagenesis (10). Studies of bmi-1 knockout mice later showed that BMI-1 regulates the self-renewal of hematopoietic, neural, and neural crest stem cells (reviewed in ref. 9). Mechanistically, BMI-1 maintains stemness and prevents premature cellular senescence in large part through transcriptional repression of Cdkn2a (11). The CDKN2A locus encodes p16INK4a and p14ARF, genes that contribute to cell cycle regulation and apoptosis through modulation of the retinoblastoma and p53 pathways. In somatic stem cells, BMI-1 functionally inhibits both pathways thereby supporting self-renewal and immortality. Down-regulation of BMI-1 expression during cellular differentiation is associated with the release of this inhibition.
In addition to c-myc (10, 12), bmi-1 cooperates with the leukemia-associated translocations E2a-Pbx1 (13) and Hoxa9-Meis1 (14) in murine leukemogenesis, and it has also been implicated in the origin of nasopharyngeal carcinoma (15), neuroblastoma (16, 17), and medulloblastoma (18, 19). Moreover, the EWS-FLI1–related protein TLS-ERG can immortalize primary human hematopoietic stem cells that have up-regulated endogenous BMI-1 expression (20). Finally, the self-renewal and tumorigenicity of leukemia, neuroblastoma, and breast cancer stem cells has been linked to BMI-1 function (14, 17, 21). Thus, BMI-1 is expressed by and functions as an oncogene in many types of human cancer. Importantly, although direct transcriptional repression of the CDKN2A locus contributes to the oncogenic activity of BMI-1 in cellular models (12), recent studies have suggested that other mechanisms of BMI-1–mediated tumor promotion exist and that these are functionally independent of CDKN2A repression (19, 22–24).
In this study, we have investigated whether BMI-1 functions as an oncogene in ESFT. Our findings confirm that BMI-1 is highly expressed by ESFT cells and that it promotes anchorage-independent growth in vitro and tumor formation in vivo. Furthermore, our data also show that the mechanism of BMI-1–mediated tumorigenicity in ESFT is, at least in part, independent of CDKN2A repression and that BMI-1 regulates pathways involved in cell differentiation and development as well as cell adhesion. In particular, these data support the hypothesis that BMI-1 expression is central to the pathogenesis of ESFT and that modulation of cell adhesion pathways contributes to BMI-1–mediated tumorigenicity in this tumor family.
Materials and Methods
Cell lines and tumor specimens. ESFT cell lines obtained from Dr. Timothy Triche (Los Angeles, CA) and Dr. Heinrich Kovar (CCRI, St. Anna Kinderkrebsforschung, Vienna, Austria) were grown as described (25, 26). MRC5, Tera-2, and MCF-7 cells were obtained from American Type Culture Collection, and H9 human embryonic stem cells were from Wicell. Primary tumor RNA and tissue slides were obtained from Childrens Hospital Los Angeles (CHLA) and additional RNA samples and ESFT tissue microarrays were obtained from the Cooperative Human Tissue Network in Columbus, OH. All primary tissue samples were anonymized and acquired in accordance with approval from the CHLA Committee for Clinical Investigation.
Quantitative real-time reverse transcriptase PCR. cDNA was generated from DNase I–treated RNA (iScript; Bio-Rad) and quantitative real-time reverse transcriptase PCR (QRT-PCR) was performed using validated TaqMan Gene Expression Assays (Applied Biosystems). Assays were performed in triplicate on an Applied Biosystems 7900HT Fast real-time PCR system and average Ct values normalized relative to expression of ACTIN and/or GAPDH in the same sample. Expression data was collected from a minimum of three replicate experiments.
Western blot and immunohistochemistry. Western blots of whole cell lysates were performed using the following primary antibodies at 1:1,000 dilutions: BMI-1 (Millipore); PARP, phosphoretinoblastoma, total retinoblastoma, and p21 (Cell Signaling); p16, p14ARF, and ACTIN (Santa Cruz Biotechnology). Histologic sections and tissue microarrays were stained with anti–BMI-1 antibody (1:5,000; Millipore). Tissue microarrays were digitally scanned and images acquired by the Children's Oncology Group Biopathology Center (Columbus, OH) using Aperio ImageScope Scanner and software (Aperio).
Gene knockdown and overexpression. To knockdown BMI-1, cell lines were transfected with 50 nmol/L of BMI-1–targeted (siBMI1-A, 5′-CCGUCUUAAUUUUCCAUUG-3′; or siBMI1-B, 5′-GCGGUAACCACCAAUCUUC-3′), or negative control (siNS) Small interfering RNA (siRNA) oligonucleotides (Ambion). Knockdown of NID1 was achieved by transfection of 100 nmol/L of prevalidated siRNA oligonucleotides (Ambion). For stable BMI-1 knockdown, siBMI1-A and siBMI1-B DNA oligonucleotide sequences were cloned into the pSuper-retro-puro short-hairpin vector backbone (shBMI1-A and shBMI1-B; Oligoengine). A nonsilencing sequence (5′-ACGCATGCATGCTTGCTTT-3′) was similarly cloned to act as a negative control (27). For gain-of-function studies, full-length human BMI-1 cDNA was PCR-amplified from TC-71 cells, sequence-verified, and cloned into pBabe-Puro. Retroviral supernatants were produced through tritransfection of 293FT packaging cells (Invitrogen) with the appropriate retroviral construct along with the packaging plasmids pHit 60 (MLV gag-pol) and pHit 456 (VSV envelope; provided by Dr. C. Lutzko, CHLA). Viral supernatants were collected after 48 h for transduction of ESFT cells followed by selection in 2 μg/mL of puromycin.
Assessment of cell growth, adhesion, and tumorigenicity. Transfected/transduced cells were plated in triplicate wells and total cell counts and cell viability determined using a ViCell XR cell counter (Beckman Coulter). Flow cytometry studies of cell cycle and apoptosis were completed following standard protocols for Annexin V detection and DNA content (28) using a FACScan instrument (Becton Dickinson). For the study of anchorage-independent growth, transduced cells were plated as single cell suspensions in 0.35% noble agar (Difco) in Iscove's medium supplemented with 20% fetal bovine serum, 2 μg/mL of puromycin, 2 mmol/L of l-glutamine, 100 IU of penicillin, and 100 mg/mL of streptomycin. Cellular layers were sandwiched between feeder layers consisting of 0.7% noble agar in Iscove's medium supplemented with 10% fetal bovine serum, 2 μg/mL of puromycin, 2 mmol/L of l-glutamine, 100 IU of penicillin, and 100 mg/mL of streptomycin. In vitro cell adhesion assays were performed as described (29). Briefly, 5 × 104 cells/well were plated in 96-well plates and then washed with PBS and fixed at 30-min intervals. Adherent cells were stained with 0.2% crystal violet followed by several washes with water. Crystal violet was extracted with 2% SDS and the absorbance (570 nm) of each well was measured using a GeniosPro plate reader (Tecan). In vivo studies were performed with the assurance of the Institutional Animal Care and Usage Committee. Nonobese diabetic (NOD)-severe combined immunodeficiency (SCID) mice (Charles River Laboratories) were injected s.c. with 5 × 106 tumor cells and tumor volume was measured using calipers. Mice were sacrificed when tumors reached a size of ≥1.5 cm in any dimension.
Genomic real-time PCR determination of p16 status. Genomic real-time PCR was performed as previously described (30) and calculation of CDKN2A exon 1a gene copy number determined relative to GAPDH copy number in each sample (31).
Microarray analysis. RNA was extracted, DNase I–treated, and purified (RNeasy Minikit; Qiagen) from A4573 cells 48 h post-siRNA transfection. Samples were prepared in the CHLA genome core using the GeneChip whole transcript sense target labeling assay manual (Affymetrix) and hybridization to HuEx1.0 microarrays was carried out following the manufacturer's instructions. Cell intensity file data were quantile-normalized and summarized using the iterative probe logarithmic intensity error estimation (iterPLIER) within the Affymetrix expression console software. A linear model fit was determined for each transcript using the LIMMA package (linear models for microarray data; ref. 32) and differentially expressed genes between control and BMI-1 knockdown cells identified on the basis of statistically significant differences in transcript level signal intensity (P < 0.01) and a fold change of ≥1.5. Metacore4
(GeneGo, Inc.) analytical tools were used to identify functional gene ontology categories.Results
BMI-1 is expressed by Ewing tumors. Expression microarray data generated from primary tumors and ESFT cell lines were analyzed for BMI-1 expression. As shown in Fig. 1A, BMI-1 was variable but detectable in all samples. These data were corroborated by QRT-PCR analysis of an independent cohort of primary tumors and cell lines (Fig. 1B). Nontransformed control cells, consisting of primary human fibroblasts (MRC5), bone marrow–derived mesenchymal stromal cells (MSC), and human embryonic stem cells (hESC), were similarly evaluated. To ensure equivalent extracellular environmental stimuli, all cells were collected during logarithmic growth phase and, with the exception of the hESC, were grown in the same medium (RPMI with 10% fetal bovine serum) for 24 hours prior to harvesting. Western blot (Fig. 1C) and immunohistochemical analysis (Fig. 1D) confirmed BMI-1 protein expression in ESFT cell lines and primary tumors, respectively. Histologic evaluation of 67 ESFT tumor biopsies revealed BMI-1 to be diffusely and robustly expressed by tumor cells in 49 cases (73%), whereas endothelial cells and infiltrating lymphocytes were negative for the protein (Fig. 1D,, case I and II). In 18 cases (27%), only rare and/or very weakly stained BMI-1–positive tumor cells were detected (Fig. 1D , case III).
Ewing tumors (ESFT) express BMI-1. A, gene expression profiling using Affymetrix U133A GeneChips detects expression of BMI-1 (Pcgf-4; Probeset ID 202265_at) in 94 primary ESFT and 10 ESFT cell lines. Dashed lines, median expression levels; bars, interquantile ranges. B, QRT-PCR of 20 primary ESFT and 6 ESFT cell lines confirms detectable but variable expression of BMI-1 in ESFT. Expression in nonmalignant H9 human embryonic stem cells (hESC), four primary human bone marrow–derived mesenchymal stromal cell cultures (MSC1–MSC4), and MRC5 lung embryo fibroblasts is uniformly low. Expression levels were repeated in triplicate and normalized relative to the median expression of GAPDH and ACTIN in the same sample. C, Western blot analysis detects BMI-1 protein in ESFT cell lines whereas expression in normal fibroblasts (MRC5) and MSC is very low to undetectable. The variability of protein expression in both ESFT and nonmalignant cell lines correlates with transcript expression in B. D, immunohistochemical analysis of 67 ESFT biopsy samples reveals that in 49 cases, tumor cells are highly positive for BMI-1 (e.g., case I and case II). Unlike tumor cells, endothelial cells in case I (arrow) and infiltrating lymphocytes in case II (circled) are negative. In 18 cases (e.g., case III), BMI-1 is weakly expressed or is only expressed by rare cells.
Ewing tumors (ESFT) express BMI-1. A, gene expression profiling using Affymetrix U133A GeneChips detects expression of BMI-1 (Pcgf-4; Probeset ID 202265_at) in 94 primary ESFT and 10 ESFT cell lines. Dashed lines, median expression levels; bars, interquantile ranges. B, QRT-PCR of 20 primary ESFT and 6 ESFT cell lines confirms detectable but variable expression of BMI-1 in ESFT. Expression in nonmalignant H9 human embryonic stem cells (hESC), four primary human bone marrow–derived mesenchymal stromal cell cultures (MSC1–MSC4), and MRC5 lung embryo fibroblasts is uniformly low. Expression levels were repeated in triplicate and normalized relative to the median expression of GAPDH and ACTIN in the same sample. C, Western blot analysis detects BMI-1 protein in ESFT cell lines whereas expression in normal fibroblasts (MRC5) and MSC is very low to undetectable. The variability of protein expression in both ESFT and nonmalignant cell lines correlates with transcript expression in B. D, immunohistochemical analysis of 67 ESFT biopsy samples reveals that in 49 cases, tumor cells are highly positive for BMI-1 (e.g., case I and case II). Unlike tumor cells, endothelial cells in case I (arrow) and infiltrating lymphocytes in case II (circled) are negative. In 18 cases (e.g., case III), BMI-1 is weakly expressed or is only expressed by rare cells.
In summary, using both RNA and protein studies, we have found that BMI-1 is highly expressed by the vast majority of ESFT cells. We are currently investigating the potential clinical and/or biological significance of high versus low-level BMI-1 expression in primary ESFT.
BMI-1 knockdown does not induce ESFT cell death or CDKN2A expression. It has been previously shown that BMI-1 promotes the proliferation of normal human fibroblasts (33), and that loss of BMI-1 in multiple human cancer cell lines induces cell death (34). To determine if BMI-1 promotes the survival and/or proliferation of ESFT cells, we assessed the effects of altered BMI-1 expression on ESFT cells grown in standard culture conditions. Two different siRNA sequences were used and both sequences effectively repressed BMI-1 in ESFT and in MCF-7 breast cancer cells (Fig. 2A). In contrast to MCF-7 cells, in which BMI-1 knockdown was found to significantly inhibit both proliferation and survival, BMI-1 knockdown had no effect on the growth of ESFT cells (Fig. 2B; Supplementary Data Fig. S1). In corroboration with these findings, we also found that whereas overexpression of BMI-1 promoted the proliferation of MRC5 fibroblasts, it had no effect on ESFT cell proliferation (data not shown).
BMI-1 knockdown does not affect ESFT cell proliferation or death. A, QRT-PCR confirms the knockdown of BMI-1 following transfection of siRNA oligonucleotides (siBMI1A and siBMI1B) compared with mock or control (siNS) transfected cells. ESFT cell lines—A4573, StaET8.2, TC71, and MCF7: positive controls, breast cancer cell line. RNA was harvested from cells 48 h posttransfection and gene expression normalized to GAPDH expression levels in the same sample. Columns, averages of three replicate experiments; bars, SD. B, siRNA-mediated knockdown of BMI-1 significantly inhibits growth of MCF7 but not ESFT cells. Cells transfected with siRNA oligonucleotides were counted daily for 4 days and the number of viable cells plotted. BMI-1 knockdown (siBMI1) counts were compared at each time point to control (siNS) transfected cells. Points, average counts of three replicate experiments; bars, SD. Data is shown for siBMI1B. Equivalent results were obtained for the siBMI1A sequence (data not shown). C, QRT-PCR analysis shows no effect of BMI-1 knockdown on CDKN2A expression in serum-free culture. Stably transduced A4573 cells (shBMI1 and shNS control) were grown in serum-free conditions for 48 h prior to RNA isolation. Columns, average of replicate experiments with gene expression expressed relative to GAPDH in the same samples; bars, SD. D, knockdown of BMI-1 has no effect on ESFT cell proliferation or death in cells grown in serum-free conditions. A4573 cells stably transduced with a BMI-1 hairpin construct (shBMI1) and grown in serum-free conditions for 48 h show no change in cell death (left; fluorescence-activated cell sorting analysis of Annexin-V/propidium iodide–stained cells) or cell proliferation (right; fluorescence-activated cell sorting analysis of fixed, propidium iodide–stained cells).
BMI-1 knockdown does not affect ESFT cell proliferation or death. A, QRT-PCR confirms the knockdown of BMI-1 following transfection of siRNA oligonucleotides (siBMI1A and siBMI1B) compared with mock or control (siNS) transfected cells. ESFT cell lines—A4573, StaET8.2, TC71, and MCF7: positive controls, breast cancer cell line. RNA was harvested from cells 48 h posttransfection and gene expression normalized to GAPDH expression levels in the same sample. Columns, averages of three replicate experiments; bars, SD. B, siRNA-mediated knockdown of BMI-1 significantly inhibits growth of MCF7 but not ESFT cells. Cells transfected with siRNA oligonucleotides were counted daily for 4 days and the number of viable cells plotted. BMI-1 knockdown (siBMI1) counts were compared at each time point to control (siNS) transfected cells. Points, average counts of three replicate experiments; bars, SD. Data is shown for siBMI1B. Equivalent results were obtained for the siBMI1A sequence (data not shown). C, QRT-PCR analysis shows no effect of BMI-1 knockdown on CDKN2A expression in serum-free culture. Stably transduced A4573 cells (shBMI1 and shNS control) were grown in serum-free conditions for 48 h prior to RNA isolation. Columns, average of replicate experiments with gene expression expressed relative to GAPDH in the same samples; bars, SD. D, knockdown of BMI-1 has no effect on ESFT cell proliferation or death in cells grown in serum-free conditions. A4573 cells stably transduced with a BMI-1 hairpin construct (shBMI1) and grown in serum-free conditions for 48 h show no change in cell death (left; fluorescence-activated cell sorting analysis of Annexin-V/propidium iodide–stained cells) or cell proliferation (right; fluorescence-activated cell sorting analysis of fixed, propidium iodide–stained cells).
Because BMI-1 has been reported to exert its effects through the repression of CDKN2A and its protein products p16 and p14ARF, we reasoned that the lack of effect of BMI-1 modulation on ESFT cell proliferation and death may be a consequence of CDKN2A deletion in the cell lines. Using a combination of QRT-PCR and Western blot, we found that A4573, StaET8.2, and StaET7.2 express the CDKN2A transcript and both A4573 and StaET8.2 express the p16 protein (Supplementary Data Fig. S2A and B). Genomic real-time PCR (30) revealed that TC71, TC32, and TC252 cells all have homozygous loss of CDKN2A, StatET8.2, and StatET7.2 are wild-type for the locus, and A4573 is hemizygously deleted (Supplementary Data Fig. S2C and D). Thus, BMI-1 knockdown does not inhibit ESFT cell growth irrespective of their p16 status. Next, we set out to determine whether BMI-1 represses CDKN2A transcript expression in p16+ ESFT cells. Unexpectedly, we found that although BMI-1 repressed the expression of CDKN2A/p16 in human MSC and Tera2 embryonal carcinoma cells, no significant or consistent effect was observed in ESFT cell lines (Supplementary Data Fig. S3A, B, and C). The alternate protein product encoded by the CDKN2A locus, p14ARF, was not expressed to detectable levels by ESFT (data not shown). BMI-1–dependent repression of a p21-retinoblastoma pathway has recently been implicated in the control of embryonic neural stem cell self-renewal (35). To determine if BMI-1 may be targeting p21 rather than p16 in ESFT cells, we also evaluated CDKN1A expression and p21 protein levels following BMI-1 gain and loss of function. p21 protein levels were unaffected by BMI-1 knockdown in all six ESFT cell lines tested whereas CDKN1A was up-regulated in A4573 cells and down-regulated in StaET8.2 cells (data not shown). No change in pRB-phosphorylation was observed in any ESFT cell line following BMI-1 knockdown (Supplementary Data Fig. S3D). Together, these data indicate that BMI-1 does not significantly modulate either CDKN2A or CDKN1A expression in ESFT and the failure of BMI-1 to promote ESFT proliferation and/or survival is not a consequence of CDKN2A/p16 loss.
The mechanism of BMI-1–mediated repression of p16 has been shown to involve direct binding of the CDKN2A promoter and to depend on the presence of a functional retinoblastoma protein (pRB; ref. 36). ESFT cells display hyperphosphorylated (inactive) pRB when grown in standard culture conditions and although serum withdrawal leads to a reduction in pRB phosphorylation, this is accompanied by a reduction in total pRB expression (Supplementary Data Fig. S3E). To determine if BMI-1 knockdown is able to effect changes in ESFT growth in conditions in which pRB may be relatively more active, we repeated cell growth assays in serum-free conditions. Consistent with our earlier findings, we observed no significant effects of BMI-1 knockdown on CDKN2A expression level, cell proliferation, or cell death (Fig. 2C and D; data not shown).
BMI-1 promotes the anchorage-independent growth and tumorigenicity of Ewing tumor cells. Although BMI-1 loss did not alter ESFT proliferation or death, we observed that knockdown dramatically altered the morphologic characteristics of A4573 cells in vitro. Whereas these cells normally grow as three-dimensional clusters, knockdown of BMI-1 reproducibly resulted in conversion to growth as adherent cellular monolayers (Fig. 3A). We therefore reasoned that BMI-1 might contribute to anchorage-independent colony formation. To assess whether the anchorage-independent growth of ESFT cells in vitro is affected by altered BMI-1 expression levels, we performed soft agar assays of cells that were genetically modified to express altered levels of BMI-1. As shown, knockdown of BMI-1 inhibited colony formation (Fig. 3B), whereas overexpression of BMI-1 led to a more rapid formation of macroscopic colonies and an increase in macroscopic colony number (Fig. 3C). Confirmation of the specificity of the effects of BMI-1 knockdown was achieved by demonstrating equivalent effects using a second shRNA sequence (Supplementary Data Fig. S4).
BMI-1 promotes the anchorage-independent growth of ESFT cells. A, phase contrast images of A4573 cells growing in standard culture conditions. siRNA-mediated knockdown of BMI-1 (siBMI1) causes A4573 cells to flatten whereas Lipofectamine-treated and control transfected (siNS) cells continue to grow as three-dimensional clusters. B, stable knockdown of BMI-1 (shBMI1B) impairs colony formation of ESFT cells in soft agar. Images are of macroscopic colonies on the day at which control cell (shNS) media was depleted for each of the three respective cell lines. Each well is representative of six replicate wells for each condition. Similar results were obtained with a second BMI-1–targeted shRNA construct (see Supplementary Data Fig. S4). C, ESFT cells transduced to overexpress BMI-1 (pBp-BMI1) have a reduced time to macroscopic colony formation and increased colony number in soft agar when compared with control, empty vector–transduced cells (pBp-V). Images are of macroscopic colonies on the day at which pBp-BMI1 media was depleted for each of the three respective cell lines. Each well is representative of six replicate wells for each condition.
BMI-1 promotes the anchorage-independent growth of ESFT cells. A, phase contrast images of A4573 cells growing in standard culture conditions. siRNA-mediated knockdown of BMI-1 (siBMI1) causes A4573 cells to flatten whereas Lipofectamine-treated and control transfected (siNS) cells continue to grow as three-dimensional clusters. B, stable knockdown of BMI-1 (shBMI1B) impairs colony formation of ESFT cells in soft agar. Images are of macroscopic colonies on the day at which control cell (shNS) media was depleted for each of the three respective cell lines. Each well is representative of six replicate wells for each condition. Similar results were obtained with a second BMI-1–targeted shRNA construct (see Supplementary Data Fig. S4). C, ESFT cells transduced to overexpress BMI-1 (pBp-BMI1) have a reduced time to macroscopic colony formation and increased colony number in soft agar when compared with control, empty vector–transduced cells (pBp-V). Images are of macroscopic colonies on the day at which pBp-BMI1 media was depleted for each of the three respective cell lines. Each well is representative of six replicate wells for each condition.
To determine if BMI-1 promotion of ESFT colony formation in vitro correlates with in vivo tumorigenicity, we evaluated xenograft tumor formation in NOD-SCID mice. As shown in Fig. 4A and B, the rate of engraftment of TC71 tumor cells directly correlated with BMI-1 expression levels. For shBMI1 cells, the median time to measurable tumor was 16 days compared with only 13 days for shNS cells (P < 0.05). In contrast, BMI-1–overexpressing cells formed tumors within 10.5 days compared with 15 days for empty vector–transduced cells (P = 0.005). Similarly, whereas A4573 shNS control cells formed tumors with a median time of 15 days, only one of five shBMI1-recipient mice developed a tumor and this did not appear until 37 days postimplantation (data not shown). To confirm that the xenografts did not form from cells that had escaped genetic modification, variable expression of BMI-1 was confirmed in the excised tumors (Fig. 4C).
BMI-1 promotes ESFT tumorigenicity in vivo. A, NOD-SCID mice injected with TC71 cells that express reduced levels of BMI-1 (shBMI1) show delayed engraftment and slowed in vivo growth compared with mice injected with control (shNS) transduced cells (two-tailed paired t test, P < 0.001). B, time to tumor cell engraftment is decreased and tumor growth enhanced in NOD-SCID mice injected with TC71 cells that overexpress BMI-1 (pBp-Bmi1) compared with empty vector (pBp-V) transduced cells (two-tailed paired t test, P = 0.01). C, QRT-PCR analysis of RNA harvested from tumor xenografts confirms differential expression of BMI-1.
BMI-1 promotes ESFT tumorigenicity in vivo. A, NOD-SCID mice injected with TC71 cells that express reduced levels of BMI-1 (shBMI1) show delayed engraftment and slowed in vivo growth compared with mice injected with control (shNS) transduced cells (two-tailed paired t test, P < 0.001). B, time to tumor cell engraftment is decreased and tumor growth enhanced in NOD-SCID mice injected with TC71 cells that overexpress BMI-1 (pBp-Bmi1) compared with empty vector (pBp-V) transduced cells (two-tailed paired t test, P = 0.01). C, QRT-PCR analysis of RNA harvested from tumor xenografts confirms differential expression of BMI-1.
BMI-1 knockdown effects significant changes in developmental and cell adhesion pathways. Having established that BMI-1 promotes ESFT tumorigenicity in the absence of CDKN2A modulation, we initiated studies to identify novel effectors of BMI-1 action. To achieve this, we performed expression profiling of A4573 cells following acute knockdown of BMI-1 and found that nearly 900 known genes were significantly affected (P < 0.01), and 245 of these were altered at least 2-fold (Supplementary Table S1A; GEO accession series no. GSE12064). Assessment of the gene ontology designations of these BMI-1–responsive genes found several pathways to be significantly overrepresented, in particular, pathways involved in cellular development as well as adhesion and invasion (Table 1). In addition, whereas many genes in these overrepresented pathways were induced by BMI-1 knockdown (e.g., NOTCH1, WNT5A, TIMP2, and TIMP3), others were down-regulated (e.g., COL5A2, COL1A2, and ICAM2) demonstrating that although BMI-1 is known to function as a transcriptional repressor, inhibition of its expression does not exclusively lead to gene induction and genes that are indirectly regulated by BMI-1, either up-regulated or down-regulated, may contribute to its function as an oncogene.
Gene ontology biological processes significantly affected by BMI-1 knockdown in A4573 ESFT cells
No. . | Gene ontology biological process . | Douglas et al. (P) . | Overlap with Wiederschain et al. . |
---|---|---|---|
1 | Development: neurogenesis in general | 3.2E−07 | * |
2 | Proteolysis: connective tissue degradation | 9.9E−05 | * |
3 | Cell adhesion: cell-matrix interactions | 1.0E−04 | * |
4 | Reproduction: progesterone signaling | 2.5E−04 | NS |
5 | Cell adhesion: amyloid proteins | 2.6E−04 | * |
6 | Signal transduction: WNT signaling | 4.2E−04 | NS |
7 | Cytoskeleton: regulation of cytoskeleton rearrangement | 5.5E−04 | NS |
8 | Proteolysis: ECM remodeling | 2.3E−03 | * |
9 | Development: skeletal muscle development | 2.3E−03 | * |
10 | Cell adhesion: integrin-mediated cell-matrix adhesion | 2.6E−03 | NS |
11 | Signal transduction: neuropeptides signaling pathways | 3.1E−03 | NS |
12 | Signal transduction: NOTCH signaling | 4.5E−03 | * |
No. . | Gene ontology biological process . | Douglas et al. (P) . | Overlap with Wiederschain et al. . |
---|---|---|---|
1 | Development: neurogenesis in general | 3.2E−07 | * |
2 | Proteolysis: connective tissue degradation | 9.9E−05 | * |
3 | Cell adhesion: cell-matrix interactions | 1.0E−04 | * |
4 | Reproduction: progesterone signaling | 2.5E−04 | NS |
5 | Cell adhesion: amyloid proteins | 2.6E−04 | * |
6 | Signal transduction: WNT signaling | 4.2E−04 | NS |
7 | Cytoskeleton: regulation of cytoskeleton rearrangement | 5.5E−04 | NS |
8 | Proteolysis: ECM remodeling | 2.3E−03 | * |
9 | Development: skeletal muscle development | 2.3E−03 | * |
10 | Cell adhesion: integrin-mediated cell-matrix adhesion | 2.6E−03 | NS |
11 | Signal transduction: neuropeptides signaling pathways | 3.1E−03 | NS |
12 | Signal transduction: NOTCH signaling | 4.5E−03 | * |
Abbreviation: NS, nonsignificant overlap.
Processes in which statistically significant overlap (P < 0.01) exists in BMI-1–modulated genes between ESFT and medulloblastoma cells (19).
In an attempt to identify the downstream genes that are most likely to mediate tumorigenicity, we compared our data to a previously published analysis of BMI-1 knockdown in DAOY medulloblastoma cells (19). Raw data from this study was extracted from the National Center for Biotechnology Information GEO database (series no. GSE7578) and processed using the same methodology used for analysis of our ESFT data (described in Materials and Methods). Comparison of the two independent gene lists reveals that expression of 101 genes was significantly and commonly altered by BMI-1 knockdown in both A4573 and DAOY cells (Supplementary Table S1B). Importantly, whereas A4573 cells express p16, DAOY are p16-null, further supporting the designation of these BMI-1 targets as p16-independent (37). Moreover, DAOY cells were studied following shRNA transduction and antibiotic selection, whereas A4573 cells were analyzed following acute siRNA-mediated knockdown. Thus, the common effects on gene expression cannot be attributed to the consequences of experimental design. Rather, these 101 genes are likely to represent true BMI-1 targets. A high degree of overlap in overrepresented biological processes between the two model systems is also observed, with the same developmental and cell adhesion genes featuring prominently (Table 1). To ensure that the overlap between the gene sets was not the result of chance, we computed the probability of obtaining the observed number of overlaps under a hypergeometric distribution and confirmed the degree of overlap to be highly statistically significant (P < 0.001).
In view of the striking overrepresentation of genes involved in cellular adhesion, we evaluated the consequences of BMI-1 modulation on the adhesion of A4573 cells in vitro. As shown in Fig. 5A, adhesion is promoted by BMI-1 overexpression and inhibited by BMI-1 knockdown in these cells. To validate that adhesion genes identified by microarray analysis are bona fide targets of BMI-1, we performed QRT-PCR and Western blot to assess the expression levels of NID1 and VEZT, and their respective protein products, nidogen 1 and vezatin. Although expression of both genes was consistently and reproducibly altered by BMI-1 knockdown in ESFT cell lines (Fig. 5B), only nidogen 1 expression was altered at the level of protein expression (Fig. 5C). Nidogen 1 is a cell adhesion protein and integral component of basement membranes (38). To determine if nidogen 1 levels affect ESFT cell adhesion, we examined the consequences of NID1 knockdown. As shown in Fig. 5D, NID1 knockdown accelerated ESFT cell adhesion, suggesting that repression of nidogen 1 may be functionally important in BMI-1–mediated adhesion and tumorigenicity. More extensive studies are now required to test this possibility.
BMI-1 accelerates ESFT cell adhesion in vitro. A, A4573 cells were plated in 96-well plates and monitored for adhesion. Cells with stable knockdown of BMI-1 (shBMI1) show reduced adhesion relative to control cells (shNS; top), whereas adhesion is accelerated in BMI-1–overexpressing cells (pBp-BMI1) in comparison with empty vector cells (pBp-V; bottom). Points, mean absorbance of crystal violet taken up by adherent cells in at least seven replicate wells from duplicate experiments at 30-min intervals. B, QRT-PCR validates that two cell adhesion genes identified by microarray analysis are truly BMI-1–responsive. Note that although basal levels of the two genes differ, knockdown of BMI-1 leads to significant up-regulation of NID1 and down-regulation of VEZT in all three ESFT cell lines. C, Western blot confirms the up-regulation of nidogen 1 protein in A4573 cells following BMI-1 knockdown (siBMI1A and siBMI1B) and down-regulation in the presence of BMI-1 overexpression (pBp-BMI1). D, A4573 cells were transfected with siNID1 or siNS control oligonucleotides and knockdown confirmed after 48 h by QRT-PCR (left). Transfected cells were plated in 96-well plates and adhesion measured as described in A. Reduced expression of NID1 accelerates adhesion.
BMI-1 accelerates ESFT cell adhesion in vitro. A, A4573 cells were plated in 96-well plates and monitored for adhesion. Cells with stable knockdown of BMI-1 (shBMI1) show reduced adhesion relative to control cells (shNS; top), whereas adhesion is accelerated in BMI-1–overexpressing cells (pBp-BMI1) in comparison with empty vector cells (pBp-V; bottom). Points, mean absorbance of crystal violet taken up by adherent cells in at least seven replicate wells from duplicate experiments at 30-min intervals. B, QRT-PCR validates that two cell adhesion genes identified by microarray analysis are truly BMI-1–responsive. Note that although basal levels of the two genes differ, knockdown of BMI-1 leads to significant up-regulation of NID1 and down-regulation of VEZT in all three ESFT cell lines. C, Western blot confirms the up-regulation of nidogen 1 protein in A4573 cells following BMI-1 knockdown (siBMI1A and siBMI1B) and down-regulation in the presence of BMI-1 overexpression (pBp-BMI1). D, A4573 cells were transfected with siNID1 or siNS control oligonucleotides and knockdown confirmed after 48 h by QRT-PCR (left). Transfected cells were plated in 96-well plates and adhesion measured as described in A. Reduced expression of NID1 accelerates adhesion.
Discussion
We have found that BMI-1 is highly expressed by ESFT cells and that it promotes anchorage-independent growth and in vivo tumorigenicity. Importantly, our studies reveal that these tumorigenic properties of BMI-1 are modulated independent of CDKN2A repression, indicating that novel mechanisms of BMI-1 oncogenic activity exist. In fact, in contrast to normal mesenchymal stem cells, altering expression levels of BMI-1 has no significant or consistent effect on CDKN2A or p16 expression in ESFT. Although this may be an artifact of in vitro culture, it is also possible that nonfunctional retinoblastoma family proteins prevent BMI-1–mediated repression of CDKN2A in these cells (36). Although previous reports have shown that pRB is only rarely mutated in ESFT (25, 39), recent work suggests that pRB function may be functionally inactivated by EWS-FLI1 itself (40). Further studies are now required to determine if pRB inactivation contributes to the dissociation of BMI-1 from p16 regulation in ESFT.
Although the histogenesis of ESFT remains a mystery, recent studies implicate somatic stem cells as cells of origin (reviewed in ref. 41). Given that most somatic stem cells express high-levels of BMI-1 and that expression diminishes during differentiation (42), it is possible that the high level of BMI-1 expression we observe in ESFT cells is an inherent feature of their cellular origin. Alternatively, expression of the EWS-FLI1 fusion oncogene may be able, in some cell types, to induce BMI-1 as was recently shown in NIH-3T3 cells (43). Cell type– and differentiation state–appropriate experimental models are now required to test which of these situations exists in the initiation of ESFT. EWS-FLI1–mediated transformation of primary fibroblasts requires the inactivation of p16-retinoblastoma and/or p53 pathways (44, 45). We speculate that BMI-1–mediated repression of CDKN2A may confer cellular tolerance to EWS-FLI1 in the ESFT cell of origin and that this epigenetic inactivation of tumor suppressor pathways could explain the relatively low incidence of secondary genetic mutations in primary ESFT (3). In support of this, we find that BMI-1 levels are, in general, higher among primary tumors compared with ESFT cell lines (Fig. 1A), suggesting that mutations in p16 and/or p53 that are more commonly present in cell lines (25, 46) may at least partially compensate for BMI-1 expression. It has been previously documented that p16 loss in lung tumors correlates with low BMI-1 expression (47). We are now testing whether there is a relationship between BMI-1 expression and p16 and/or p53 status in primary ESFT and whether differences exist in clinical presentation or outcome between tumors that express high versus low levels of BMI-1.
We have shown that altering BMI-1 expression affects the ability of both p16-null and p16-positive cells to form anchorage-independent colonies in vitro and tumors in vivo. In corroboration with our findings, several recently published reports have revealed that, in cooperation with other oncogenic lesions such as mutated epidermal growth factor receptor or H-RAS, BMI-1 can transform both CDKN2A wild-type and CDKN2A null cells (22, 23). In addition, BMI-1 knockdown in p16-null DAOY medulloblastoma cells significantly impedes tumor formation in vivo (19, 37). Thus, although initial studies of BMI-1 implicated the repression of the p16Ink4a/p14ARF-encoding CDKN2A locus as the primary mechanism of oncogenic action (11, 12), more recent data from our lab and others show a pivotal role for p16-independent mechanisms.
In order to identify potentially novel downstream targets of BMI-1, we performed gene expression profiling of ESFT cells following BMI-1 knockdown and compared BMI-1–responsive genes to those genes similarly affected by BMI-1 knockdown in human medulloblastoma cells (19). Although a significant subset of genes was commonly regulated by BMI-1 in both tumor types, others were uniquely altered in only one of the model systems. This implies that although some biological pathways are shared among different tumor types, it is likely that at least some downstream effectors of BMI-1 differ among tumors of different cellular origins. Nevertheless, our findings show that significant commonalities exist. In particular, direct comparison between ESFT and medulloblastoma cells (19) reveals that alterations in cell adhesion and extracellular remodeling processes are highly overrepresented and common to both tumor types. For the current study, we have validated that expression of the basement membrane protein nidogen 1 is repressed by BMI-1 in ESFT cells. Nidogen 1 acts as a linker between laminins, collagens, and proteoglycans in the extracellular matrix and binds to cell surface integrins (38). Interestingly, it has recently been reported that NID1 is frequently silenced in colon cancer, suggesting that nidogen 1 may have a role as a tumor suppressor gene, preventing invasion and metastasis (48). In support of this possibility, we have found that down-regulation of NID1 promotes adhesion of ESFT cells in vitro, recapitulating the effects of BMI-1 overexpression. Therefore, we hypothesize that the effect of BMI-1 knockdown on cell adhesion, through modulation of nidogen 1 and/or other adhesion-related proteins, is likely to underlie the delay to in vivo tumor engraftment that we observe in ESFT cells with reduced levels of BMI-1. Consistent with this hypothesis, delayed engraftment and altered adhesion pathways have also been shown to be a feature of bmi-1–deficient murine glioma cells (22). Thus, the cumulative evidence suggests that modulation of adhesion molecules such as nidogen 1 is likely to contribute to the oncogenic function of BMI-1 in ESFT as well as other tumor types.
Finally, polycomb genes, including BMI-1, play a central role in the repression of differentiation and in the controlled orchestration of normal development (5). Our finding that developmental pathways are significantly affected by BMI-1 knockdown in ESFT cells suggests that the embryonic function of BMI-1 is being recapitulated in these undifferentiated tumor cells. It is particularly noteworthy that both WNT and NOTCH pathway genes are highly affected by BMI-1 knockdown as both of these developmental pathways have been previously implicated in ESFT growth and tumorigenicity (49, 50). Intriguingly, our data also show that among the affected developmental processes, BMI-1 loss has its most profound effect on genes that are involved in neural development with BMI-1 knockdown leading to increased expression of neural markers (Table 1; Supplementary Table S1). These findings corroborate recent documentation of the effects of bmi-1 loss on the phenotype and neural differentiation capacity of murine gliomas (22). One of the many clinical mysteries surrounding ESFT is the observation that they vary from highly undifferentiated tumors to tumors with obvious neural features. It is tempting to speculate that the phenotype of a particular tumor is either (a) predetermined by the BMI-1 expression level in the parent cell that originally acquires the EWS-ETS translocation or (b) a consequence of the tumor microenvironment and its downstream effects on BMI-1 expression. More extensive studies are required to evaluate these hypotheses.
In summary, we have shown that BMI-1 functions as an oncogene in ESFT and that it promotes tumorigenicity in a CDKN2A-independent manner, influencing pathways involved in cell adhesion, differentiation, and development. Given its central role in the regulation of multiple developmental processes in normal stem cells, we expect that no single gene will be uniquely responsible for the oncogenic effects of BMI-1. Nevertheless, our findings support the hypothesis that regulation of adhesion pathways is central to BMI-1–mediated tumor promotion. Future studies directed at understanding this relationship are likely to proffer attractive and novel targets for therapeutic intervention that may be common to the multiple human cancers that deregulate and overexpress BMI-1.
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).
D. Douglas and J.H-R. Hsu contributed equally to this work.
Acknowledgments
Grant support: V Foundation and the Margaret E. Early Medical Research Trust (E.R. Lawlor), NIH grants U01 CA88199 and U01 CA115757 (T.J. Triche), California Institute for Regenerative Medicine training grants CHLA T2-00005, USC T1-00004 (D. Douglas and J.H-R. Hsu), and NIH training grant T32 CA 09659 (J. van Doorninck). Support from the My Brother Joey, Stop Cancer, and T.J. Martell Foundations is also gratefully acknowledged.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
The authors thank Heinrich Kovar, Darwin Prockop, Carolyn Lutzko, Gerard Evan, and Robert Ilaria for cells and reagents. We also thank members of the Triche & Lawlor labs for helpful discussion, and Betty Schaub and members of the genomics, animal, FACS, and vector cores for technical assistance. Tumor samples were provided by the CHLA Department of Pathology and by the Children's Oncology Group Biorepository in Columbus, OH. We gratefully acknowledge the staff of these facilities for their efforts.