Targeting Cancer Stem Cells through L1CAM Suppresses Glioma Growth

The cancer stem cell hypothesis posits that brain tumors contain a subset of neoplastic cells that propagate and maintain tumors through sustained self-renewal and potent tumorigenicity (1–6). Cancer stem cells can be enriched from many human brain tumor biopsy specimens using the cell surface marker CD133 (prominin 1) although there are tumors that may contain CD133⁻ tumor-initiating cells (6, 7), albeit often with lower tumor potency. We previously showed that CD133⁺ glioma cells derived from human tumors are radioresistant (5), promote angiogenesis (6), and display greater tumorigenic potential in immunocompromised mice relative to CD133⁻ glioma cells (5). Thus, targeting CD133⁺ glioma cell survival and protumorigenic behaviors by identifying novel molecular targets specific in cancer stem cells may improve patient outcome.

Molecular targets that are secreted or located on the cell surface are particularly enticing as antibody and small-molecule inhibitor therapies that do not have to cross the cell membrane and can be developed for clinical use. To select a potential target, we consulted the literature to select a cell surface protein that has been reported to be expressed in malignant gliomas and contribute to tumor malignancy. Based on these criteria, the neural cell adhesion molecule, L1CAM (L1, CD171), has been identified as a potential therapeutic target in neuro-oncology. L1CAM regulates neural cell growth, survival, migration, and axonal outgrowth and neurite extension during central nervous system development (8). Although the role of L1CAM in the normal adult nervous system is not well defined, L1CAM is overexpressed in gliomas and other solid cancers (9–15), including colorectal cancer where L1CAM is a prognostic indicator (15). Although L1CAM has many potential biological effects that can contribute to tumor formation and maintenance, recent studies suggest that overexpression of L1CAM in cancer can protect cells from apoptosis induced by nutrient deprivation (16) or chemotherapeutics (17). As glioma stem cells display a chemoresistant phenotype, we sought to determine if L1CAM mediates survival and tumorigenic potential of CD133⁺ glioma cell subpopulation.

Cell isolation and culture. Matched glioma cell populations enriched or depleted in cancer stem cells were isolated and cultured as previously described (5, 6) from human glioma surgical specimens (designated Txxx, Supplementary Table S1) or from the D456MG pediatric glioblastoma xenograft. Briefly, tumors were immediately dissected with removal of gross necrosis; washed in Earle's balanced salt solution; subjected to a papain digestion followed by tituration, filtering, and lysis of RBC in PBS/water (1:3) solution; and then cultured overnight (∼12 h) in neurobasal medium (with B27 and epidermal growth factor and basic fibroblast growth factor at 20 ng/mL) for recovery of cellular surface antigens before cell sorting. Primary glioma tumor samples were obtained from patients undergoing resection in accordance with a protocol approved by the Duke University Medical Center Institutional Review Board. Normal human fetal neural progenitor cells were purchased from Lonza. To control for differences in medium conditions, all cells were cultured in stem cell medium for experiments.

Fluorescence-activated cell sorting analysis. Because papain digestion may cause loss of some cellular surface antigens such as CD133, it is critical to allow the isolated total tumor cells to reexpress their surface antigens in the neurobasal medium overnight (12 h) before sorting for CD133⁺ glioma cells. Tumor cell cultures were subjected to fluorescence-activated cell sorting (FACS) analysis and cell sorting after the antigen recovery period. Human-specific anti-CD133 (293C3) conjugated to allophycocyanin (APC) or anti-CD133 (293C3) conjugated to phycoerythrin (PE; Milyteni) was used for cell sorting and FACS analysis as previously described (5, 6). Anti–L1CAM-PE used for the FACS was generated using the Lightning-Link PE kit (Innova Biosciences) in combination with a GeneTex monoclonal antibody against human L1CAM (UJ127 clone).

Short hairpin RNA L1CAM targeting. Knockdown of L1CAM was achieved through the use of lentiviral vector–mediated short hairpin RNA (shRNA) interference using Mission RNAi system clones (Sigma-Aldrich). L1CAM-targeting lentivirus and the nontargeting control lentivirus were produced in HEK293FT cells with the ViraPower Lentiviral Expression System (Invitrogen). Five different shRNA clones were characterized in terms of knockdown efficiency using L1CAM immunoblotting of glioma stem cells infected with lentiviruses encoding L1CAM shRNA or a nontargeting control shRNA. The most efficient clone was used for all further experiments. Produced lentiviruses were concentrated by using Centricon Plus-20 centrifugal filter device (Millipore). To ensure the same number of L1CAM-targeting and the control lentiviruses were used in the same experiment, produced lentiviral stock was titered and stored according to the manufacturer's instruction (Invitrogen). For in vitro infection of glioma stem cells with the lentivirus, cultured neurospheres were disaggregated before infection to increase the infection efficiency and uniformity.

Antibodies and Western blotting. Specific monoclonal antibodies for human L1CAM (UJ127) were purchased from GeneTex and NeoMarker. L1CAM-PE used for FACS analysis was generated using the Lightning-Link PE kit (Innova Biosciences). Anti-Olig2 goat IgG was obtained from R&D Systems. Anti-OCT4 monoclonal antibody (MAB4401) was purchased from Millipore. Anti-p21 and anti-p27 antibodies were obtained from Cell Signaling. Immunoblotting was done as previously described (5).

In vivo tumor formation assays. Intracranial or s.c. transplantation of glioma stem cells into nude mice was done as described (5, 6). Glioma stem cells were infected with lentivirus expressing nontargeting shRNA or L1CAM-targeting shRNA for 24 h, and then 10⁵ cells per mouse of viable brain tumor stem cells transduced with control lentivirus or the L1CAM-targeting lentivirus were transplanted into athymic BalbC nu/nu mice through intracranial or s.c. injection. To establish glioma xenografts for lentiviral-mediated L1CAM shRNA treatment, glioma stem cells were injected into the athymic nude mice (10⁵ cells per mouse, 10 mice per group) first, and the same number of nontargeting lentivirus or L1CAM-targeting lentivirus were delivered to tumor sites through intratumoral injection once every 2 d.

Neurosphere formation efficiency. To determine the effect of knockdown of L1CAM on the ability of glioma stem cells to form neurospheres, disaggregated cells were infected with nontargeting or L1CAM-targeting lentivirus for 48 h and then plated into 96-well plates at a density of 1 cell per well via a flow cytometry cell sorter (FACSAria, BD Bioscience) or through serial dilution. The percentage of wells with neurospheres was determined on day 10 as shown.

Immunofluorescent staining. CD133⁺ and CD133⁻ glioma cells were fixed with 4% paraformaldehyde, washed with TBS, incubated with anti-L1CAM monoclonal antibody (UJ127, GeneTex), and FITC-conjugated donkey anti-mouse IgG secondary antibody. Cells were costained with 4′,6-diamidino-2-phenylindole (DAPI) and mounted with the anti-fade medium. Stained cells were examined under a fluorescent microscope (Zeiss Axiovert 200).

Annexin V–FITC staining. Annexin V–FITC staining for detecting apoptosis in CD133⁺ and CD133⁻ glioma cells and normal neural progenitor cells before and after L1CAM targeting for 48 h was done with the Annexin V–FITC Apoptosis Detection Kit (BD PharMingen) according to the manufacturer's instructions. The cells stained with Annexin V–FITC were analyzed by FACS, or fixed and costained with DAPI, and then examined under a fluorescent microscope (Zeiss Axiovert 200).

Real-time PCR. To examine the transcriptional regulation of L1CAM on Olig2 and p21^WAF1/CIP1 in CD133⁺ cells, CD133⁺ glioma cells were treated with nontargeting lentivirus or L1CAM-targeting lentivirus for 48 h, and RNA samples were isolated from these cells using RNeasy Kit (Qiagen), and then used for real-time PCR analysis with pairs of specific primers for Olig2, p21^WAF1/CIP1, L1CAM, and glyceraldehyde-3-phosphate dehydrogenase (GAPDH; internal PCR control).

Statistical analyses. Descriptive statistics and significance were determined as described (5, 6).

L1CAM is differentially overexpressed in CD133⁺ glioma cells. We examined L1CAM expression in CD133⁺ glioma cells derived from 12 human primary tumors (Supplementary Table S1) and one xenograft through complementary techniques. Using flow cytometry, we determined that the majority of CD133⁺ cells were doubly positive for L1CAM⁺ when isolated directly from human patient specimens or passaged in short-term cell culture (Fig. 1A; Supplementary Fig. S1A). In striking contrast, the vast majority of CD133⁻ cells were L1CAM negative (>99% in T3565, T4121, and T4577; Fig. 1A; Supplementary Fig. S1B). In addition, the total percentage of CD133⁺ and L1CAM⁺ cells in patient glioma specimens were very similar (Fig. 1A; Supplementary Fig. S1B). L1CAM expression on the surface of CD133⁺ cells, but not CD133⁻ cells, was further verified using immunofluorescence (Fig. 1B), and Western analysis confirmed that L1CAM protein was expressed at levels 18- to 42-fold higher in CD133⁺ cells from glioma patient specimens relative to matched CD133⁻ tumor cells (Fig. 1C). In addition, CD133⁺ glioblastoma cells expressed L1CAM at significantly higher levels than normal neural progenitor cells derived from human fetal tissue that were enriched for CD133⁺ cells (Fig. 1D). Together, these data suggest that L1CAM is a surface glycoprotein specifically expressed by glioma stem cells.

Targeting L1CAM decreases growth and survival of CD133⁺ glioma cells. To assess the functional significance of the relative overexpression of L1CAM in CD133⁺ glioma cells compared with CD133⁻ glioma cells and normal neural progenitors, we targeted L1CAM expression using lentivirus expressing shRNA directed against L1CAM, causing a 90% reduction of L1CAM expression relative to nontargeting control shRNA (Fig. 2A). As neurosphere formation is a key behavior of neural stem cells and brain tumor stem cells (1–6, 18) and is used as a measure of stem cell capacity for self-renewal, we determined neurosphere formation competence in cells with knockdown of L1CAM. Targeting L1CAM expression markedly decreased the ability of CD133⁺ glioblastoma cells to form neurospheres as indicated by the reduction in neurosphere formation efficiency (Fig. 2B; Supplementary Fig. S2A and B) and the size of the neurospheres formed (Fig. 2B). In addition, L1CAM knockdown resulted in significant growth inhibition in CD133⁺ glioblastoma cells, only modest attenuation in normal neural progenitors, and no effect on CD133⁻ glioma cells (Supplementary Fig. S3). To determine if the reduction in neurosphere formation and growth of CD133⁺ glioma cells with L1CAM knockdown was due to decreased cell survival, we determined the percentage of apoptotic cells using Annexin V staining. Similar to the effects of L1CAM-directed shRNA on cell growth, L1CAM shRNA significantly increased the percentage of Annexin V–positive cells in CD133⁺ glioblastoma cells when compared with the effects of a nontargeting shRNA control (Fig. 2C; Supplementary Fig. S2C). In contrast, L1CAM knockdown had essentially no significant effect on apoptosis in CD133⁻ glioblastoma cells (Fig. 2C; Supplementary Fig. S2C) and a moderate induction of apoptosis in normal neural progenitor cells (Fig. 2C and D), which was significantly less than that induced in CD133⁺ glioma cells. These data support a significant role for L1CAM in maintaining the growth of CD133⁺ glioblastoma cells and suggest that targeting L1CAM decreases CD133⁺ glioma self-renewal due to decreased survival.

Reduction of L1CAM protein induces down-regulation of Olig2 and up-regulation of p21^WAF1/CIP1. To delineate the molecular mechanisms through which L1CAM regulates CD133⁺ glioma cell survival, we interrogated intracellular signaling pathways in L1CAM knockdown cells with a focus on regulators of neural stem cell maintenance. As the neural stem cell transcription factor Olig2 was recently found to regulate neurosphere formation in vitro and glioma formation in vivo (19), we determined Olig2 expression in CD133⁺ glioma cells infected with lentivirus expressing no shRNA as a vector control, nontargeting control shRNA, or L1CAM shRNA. L1CAM knockdown reduced Olig2 protein expression in CD133⁺ glioma cells derived from primary glioma surgical specimens and a glioblastoma xenograft (Fig. 3A,-i, B). The suppression of Olig2 induced by L1CAM knockdown is expected to have specific effects on CD133⁺ glioblastoma cells, as minimal Olig2 was detected in matched CD133⁻ glioma cells (Fig. 3A,-ii). In contrast, L1CAM reduction did not alter the protein levels of other stem cell transcription factors such as Oct4 in CD133⁺ glioma cells (data not shown). As Olig2 may mediate its growth effects by suppressing expression of the cyclin-dependent kinase inhibitor p21^WAF1/CIP1 (19), we further determined the effect of L1CAM shRNA on p21^WAF1/CIP1 expression. Knockdown of L1CAM in CD133⁺ glioblastoma cells led to a specific up-regulation of p21^WAF1/CIP1, but not p27^KIP1, protein (Fig. 3B). Real-time PCR further confirmed that L1CAM knockdown was associated with Olig2 down-regulation and p21^WAF1/CIP1 up-regulation at the RNA level (Fig. 3C). These data suggest that the poor growth and survival of CD133⁺ glioma cells with L1CAM knockdown results, at least in part, from the loss of Olig2 expression and resulting increase in p21^WAF1/CIP1. This model was further supported by data demonstrating that Olig2 overexpression rescued CD133⁺ glioblastoma cells from L1CAM shRNA–induced reductions in cell growth (Fig. 3D). Thus, our data provide the first evidence that L1CAM can mediate cancer stem cell self-renewal and survival by regulating Olig2 expression with associated changes in the downstream effector, p21^WAF1/CIP1.

Targeting L1CAM in CD133⁺ glioma cells suppresses glioma growth in vivo and increases survival of mice bearing glioma xenografts. As L1CAM shRNA-mediated decreases in CD133⁺ glioma cell growth were associated with reduced Olig2 expression and Olig2 is critical for the growth of gliomas in vivo (19), we next sought to determine if targeting L1CAM expression could attenuate the tumorigenic potential of CD133⁺ glioma cells. To accomplish this goal, L1CAM expression was modulated in CD133⁺ glioma cells before implantation into immunocompromised mice (Fig. 4A and B; Supplementary Fig. S4). Knockdown of L1CAM expression in CD133⁺ cells derived from glioma surgical biopsy specimens before intracranial injection reduced tumor volumes (Fig. 4A; Supplementary Fig. S4A) and significantly increased survival until the development of neurologic signs (Fig. 4B; Supplementary Fig. S4B) relative to CD133⁺ glioma cells expressing nontargeting control shRNA. Furthermore, the tumorigenic capacity of CD133⁺ glioma cells was decreased by targeting L1CAM expression before intracranial injection as shown in the limiting dilution assay (Supplementary Figs. S4C, D). Although these in vivo data show a requirement for L1CAM in CD133⁺ glioma cell–mediated tumorigenesis, they do not address the effects of targeting L1CAM in established tumors. To determine if targeting L1CAM could represent a potential therapeutic paradigm targeting CD133⁺ glioma cells, we intracranially implanted CD133⁺ glioma cells into immunocompromised mice and allowed tumors to establish for 5 days. Tumor-bearing mice were then intracranially injected with lentivirus expressing nontargeting shRNA as a control or L1CAM shRNA as a novel therapy. Brains of mice treated with nontargeting shRNA displayed aggressive high-grade gliomas (Fig. 4C), whereas the brains of mice receiving intracranial injections of lentivirus expressing L1CAM shRNA showed significantly reduced tumors with small extraparenchymal tumor in most cases (Fig. 4C). Most importantly, receiving lentivirus expressing L1CAM shRNA nearly doubled the lifespan of tumor-bearing mice relative to mice receiving the nontargeting shRNA-expressing lentivirus (Fig. 4D). A similar decrease in tumor growth was observed in s.c. tumors injected with lentivirus expressing L1CAM shRNA (Supplementary Fig. S5A, B) in which L1CAM expression was decreased compared with mice receiving nontargeting shRNA-expressing lentivirus (Supplementary Fig. S5C). FACS analysis further confirmed that the CD133⁺ glioblastoma cell population was reduced in tumors treated with lentivirus expressing L1CAM shRNA (Supplementary Fig. S5D). Together, these data show that molecular targeting of L1CAM in CD133⁺ glioma cells reduces tumor growth and increases survival in immunocompromised mice in vivo and may be a novel therapeutic paradigm with important clinical implications.

The recent ability to prospectively identify cell subpopulations that are highly resistant to cancer therapies, drive tumor angiogenesis, and promote tumor spread (1–6, 20) makes it possible to identify molecular targets specific to these highly tumorigenic subpopulations for novel therapeutic targets. We have discovered a novel molecular target that had not been linked to CD133⁺ glioma cells, the cell adhesion molecule L1CAM. The dramatic overexpression of L1CAM in CD133⁺ brain tumor cells, the increased survival of mice bearing gliomas in which L1CAM has been targeted, and the decreased percentage of CD133⁺ cells in tumors treated with lentivirus expressing L1CAM shRNA all suggest that targeting L1CAM may be useful as a cancer stem cell–directed therapy. As there is a very small percentage of (<1%) L1CAM⁺ cells that are CD133⁻, it remains possible that the CD133⁻ cells reported in the literature to be tumorigenic (7) may be L1CAM⁺. Future studies will be needed to determine if L1CAM itself may be useful as a cancer stem cell marker in ex vivo and immunohistochemical studies. Regardless of whether L1CAM can be used to prospectively identify cancer stem cells, our studies provide evidence that L1CAM should be considered for further exploitation in therapeutic development and biological investigation.

No potential conflicts of interest were disclosed.

Grant support: Goldhirsh Foundation; Childhood Brain Tumor Foundation; the Pediatric Brain Tumor Foundation of the United States; Accelerate Brain Cancer Cure; Alexander and Margaret Stewart Trust; Brain Tumor Society; Duke Comprehensive Cancer Center Stem Cell Initiative Grant (J.N. Rich); NIH grants NS047409, NS054276, CA116659 and CA129958 (J.N. Rich). J.N. Rich is a Damon Runyon-Lilly Clinical Investigator supported by the Damon Runyon Cancer Research Foundation and a former Sidney Kimmel Foundation for Cancer Research Scholar.

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.

We thank Y.H. Sun, S. Keir, D. Satterfield, L. Ehinger, and J. Faison for technical assistance; M. Cook, B. Harvat, and T.R. Dissanayake for assistance with flow cytometry; Z. Lu for assistance with fluorescent microscopy; and R. Wechsler-Reya and D. Bigner for helpful discussions.