Regulation of glioblastoma stem cells by retinoic acid: role for Notch pathway inhibition

Abstract

It is necessary to understand mechanisms by which differentiating agents influence tumor-initiating cancer stem cells. Toward this end, we investigated the cellular and molecular responses of glioblastoma stem-like cells (GBM-SCs) to all-trans retinoic acid (RA). GBM-SCs were grown as non-adherent neurospheres in growth factor supplemented serum-free medium. RA treatment rapidly induced morphology changes, induced growth arrest at G1/G0 to S transition, decreased cyclin D1 expression and increased p27 expression. Immunofluorescence and western blot analysis indicated that RA induced the expression of lineage-specific differentiation markers Tuj1 and GFAP and reduced the expression of neural stem cell markers such as CD133, Msi-1, nestin and Sox-2. RA treatment dramatically decreased neurosphere-forming capacity, inhibited the ability of neurospheres to form colonies in soft agar and inhibited their capacity to propagate subcutaneous and intracranial xenografts. Expression microarray analysis identified ~350 genes that were altered within 48 h of RA treatment. Affected pathways included retinoid signaling and metabolism, cell-cycle regulation, lineage determination, cell adhesion, cell–matrix interaction and cytoskeleton remodeling. Notch signaling was the most prominent of these RA-responsive pathways. Notch pathway downregulation was confirmed based on the downregulation of HES and HEY family members. Constitutive activation of Notch signaling with the Notch intracellular domain rescued GBM neurospheres from the RA-induced differentiation and stem cell depletion. Our findings identify mechanisms by which RA targets GBM-derived stem-like tumor-initiating cells and novel targets applicable to differentiation therapies for glioblastoma.

Introduction

Glioblastoma multiforme (GBM) is the most aggressive primary brain tumor in adults with a 2-year survival rate of 28% following surgical resection, chemotherapy and radiotherapy (Stupp et al., 2009). Recurrence is nearly certain after initial treatment and there is currently no therapy proven to prolong survival after tumor recurrence. The dismal prognosis associated with GBM has fostered aggressive investigations into alternative therapeutic paradigms. Radical improvements in clinical outcomes will require a better understanding of the molecular and cell biological bases of glioblastoma propagation and therapeutic resistance (Kumar et al., 2008).

Small subpopulations of neoplastic cells with stem-like properties have been identified in leukemia and solid tumors including glioblastoma (Galli et al., 2004; Singh et al., 2004). These stem-like cells display the characteristic cardinal features of unlimited growth potential, self-renewal and multilineage differentiation. GBM stem-like cells (GBM-SCs) grow in vitro as non-adherent clonal multicellular spheroids (variably referred to as neurospheres or oncospheres) and efficiently initiate tumor xenografts that recapitulate the genetic and histopathological features of the original neoplasm from which they were derived (Lee et al., 2006). Thus, GBM-SCs form highly infiltrative orthotropic xenografts that are excellent models of the human disease. Conventional radiotherapy and chemotherapy appear to mainly target the most proliferative cancer cells and spare the less proliferative neoplastic stem-like cells that appear to be relatively resistant to current cytotoxic therapeutics due to upregulated anti-apoptotic proteins, multidrug transporters and DNA repair enzymes (Bao et al., 2006; Chalmers, 2007; Johannessen et al., 2008). It is currently hypothesized that targeting GBM-SCs or their tumor-initiating capacity will be more effective than current treatment regimens. Thus, understanding the mechanisms that regulate neoplastic stem-like cell growth and differentiation could lead to new and particularly effective anti-cancer strategies (Al-Hajj et al., 2004; Massard et al., 2006; Piccirillo and Vescovi, 2007).

Strategies to induce cancer cell differentiation have been applied to select malignancies (Sell, 2004, 2006). Retinoic acid (RA) is the most common differentiating agent in clinical practice and all-trans RA has been successfully used to treat acute promyelocytic leukemia, a stem cell malignancy (Jurcic et al., 2007). The anti-tumor activity of RA is due to the activation of an as yet partially defined genetic program that modulates cell proliferation, differentiation and death (Dragnev et al., 2003; Mongan and Gudas, 2007). RA activates nuclear retinoid receptors, and genes most proximately modulated by RA contain retinoid responsive elements. In embryonic carcinoma (EC) cells and embryonic stem (ES) cells, two peaks of gene expression change occur in response to RA treatment (Niles, 2004; Soprano et al., 2007). One set of affected genes codes for regulatory proteins that function at early stages of lineage commitment. These include Notch, Wnt, Hedgehog and TGF-β, most of which have prominent functions as development regulators. A second wave of RA-induced gene regulation occurs at 24–48 h. Many of the second wave gene products, such as matrix metalloproteinases and integrins, function in extracellular matrix remodeling, cell adhesion and cell–cell/cell–matrix interactions. Thus, RA influences multiple signaling pathways involved in stem cell maintenance, lineage-specific cell differentiation and organogenesis. There have been mixed results regarding RAs efficacy in most other solid malignancies including glioblastoma (Yung et al., 1996). Understanding the molecular RA response in these less responsive cancers could lead to more innovative and effective applications of RA.

We examined the molecular and cellular responses of GBM-SCs grown as neurospheres to RA. We show that RA potently inhibits GBM neurosphere proliferation and induces neurosphere cell differentiation. RA was found to reduce the pool of tumor-initiating cells as evidenced by the inhibition of clonogenic and tumorigenic potentials. The global gene expression profile of GBM neurosphere cells in response to RA was examined by microarray analysis and several RA-responsive molecular pathways are implicated to mediate the GBM stem cell responses.

Results

RA inhibits GBM neurosphere growth and clonogenicity

We studied the effect of all-trans RA on the growth of four neurosphere cell lines established from human glioblastomas. GBM-derived neurosphere cells became adherent to the tissue culture substratum and generated processes within 48 h of exposure to RA (10 µmol/l, Figure 1a). RA was also found to inhibit GBM neurosphere cell growth as determined by cell counting (Figure 1b) and to induce cell growth arrest at the G1-S phase transition as determined by flow cytometry cell-cycle analysis (Figure 1c). As one example, exposure of HSR-GBM1B neurosphere cells to RA for 48 h increased the G1/G0 fraction from 58 to 74% (P<0.05) and decreased the S and G2/M fractions from 21 to 14% and 20 to 11%, respectively (Figure 1c). Similar responses were seen in all of four additional neurosphere lines examined, GBM-KK190156 (KK), HSR-GBM1A and GBM-DM140207 (DM) neurospheres (Supplementary Figure 1). Thus, neurospheres that are enriched in GBM-SCs appear to be uniformly sensitive to RA.

Figure 1.

All-trans RA induces growth arrest of GBM neurospheres. (a) Phase contrast photomicrographs of neurospheres before and after RA treatment for 48 h. Cells change morphology and attach to culture surface and form processes in response to RA. Bar=100 µm. (b) Cell growth curve after RA treatment. Cells were treated with RA continuously for 96 h. Compared with control, RA inhibited neurosphere cell growth (n=6, ***P<0.001). (c) Cell-cycle analysis shows that RA (10 µmol/L, 48 h) induced G1-S arrest in GBM neurospheres (n=4, *P<0.05). (d, e) Immunoblot analysis shows that cyclin D1 was significantly downregulated (~50%, n=4, ***P<0.001), and p21 was significantly upregulated by RA (twofold, n=4, ***P<0.001). The expression of cyclin A, B1, E and p27 was not changed by RA. (f) RA reduces neurosphere number and size. Neurospheres were treated with RA for 5 days. RA dramatically inhibited the number of neurospheres between 50 and 100 µm by >90%; and RA completely inhibited the formation of neurospheres bigger than 100 µm (n=6, ***P<0.001). (g) RA inhibits GBM neurosphere cell colony formation in soft agar. RA (10 µmol/l) was added in both culture medium and soft agar and incubated for 7–10 days. Colonies >100 µm were counted (n=6, ***P<0.001).

We examined the effects of RA on principal cell-cycle regulatory proteins including cyclins, cyclin-dependent kinase and the cyclin-dependent kinase inhibitors p21 and p27 in GBM neurosphere cultures. Western blot analysis revealed a 50% decrease in cyclin D1 and twofold increase in p21 protein levels in response to RA (Figures 1d and e, P<0.001). RA had no effect on the expression of cyclin A, B1, E and p27 (Figures 1d and e). Thus, RA-induced G1/G0 arrest is likely secondary to cyclin D1 inhibition and p21 induction.

Neurosphere size and number were used to assess the effects of RA on cells expressing the stem-like phenotype, as most stem-like neoplastic cells generate large neurospheres (>100 µm, Galli et al., 2004; Bar et al., 2007; Sun et al., 2009) in contrast to small neurospheres generated by progenitor cells. RA treatment for 5 days dramatically reduced neurosphere formation. RA completely inhibited the formation of neurospheres >100 µm (70 neurospheres to none) (Figure 1f, P<0.001). RA was also found to significantly inhibit neurosphere clonogenicity in soft agar (Figure 1g, P<0.001).

We also examined the effects of RA on neurosphere cell viability and caspase activation. There was no significant increase in Annexin V-positive cells up to 48 h of RA exposure. There was a slight increase in caspase 3 cleavage after 4 days of RA treatment (Supplementary Figure 2).

RA induces GBM neurosphere cell differentiation and inhibits stem cell marker expression

RA induced a dramatic change in neurosphere cell morphology. In response to RA, >90% of the suspended neurosphere cells became adherent to the tissue culture substratum and generated processes within 48 h of exposure to RA (Figure 1a). The effects of RA on neurosphere cell expression of the progenitor cell marker nestin, the neuronal marker Tuj1 and the astrocytic marker GFAP were examined by using multiple methodologies. Immunofluorescence revealed that RA generated cells with prominent Tuj1 and GFAP expression and decreased nestin expression, consistent with differentiation along neuronal and astroglial lineages (Figure 2a). The oligodendroglial marker Gal-c was not detected in either control or RA-treated cells (data not shown). Established GBM neurosphere lines and low passage primary GBM-derived neurospheres (JHH551, passage <5) responded similarly (Figure 2b). Western blot analysis showed that RA increased the expression of GFAP and Tuj1 after normalized to total protein and actin expression (Figure 2c). Flow cytometry analysis demonstrated an increase in number of GFAP⁺ cells (from 31 to 44% in HSR-GBM1B cells and from 9 to 32% in HSR-GBM1A cells) and an increase in number of Tuj1⁺ cells (from 1 to 7% in HSR-GBM1B cells and from 9 to 25% in HSR-GBM1A cells) (Figures 2d and e, P<0.001). In contrast to these neurosphere cell responses, RA had no effect on differentiation marker expression in U87 glioblastoma cells (Figure 2c).

Figure 2.

RA induces GBM neurosphere differentiation. (a, b) GBM neurospheres were treated with RA for 96 h and stained with anti-nestin, anti-GFAP and anti-Tuj1. Nuclei were stained with DAPI. Nestin was downregulated while GFAP and Tuj1 were upregulated in response to RA. Bars=25 µm. (c) Immunoblot analysis showed increased GFAP and Tuj1 expression in neurosphere cultures after RA treatment (10 µmol/l, 96 h). In contrast, in U87 human glioblastoma cells, RA (10 µmol/l, 96 h) did not increase the expression of GFAP and Tuj1. (d, e) Flow cytometry analysis shows that RA increased the number of GFAP⁺ cells and Tuj1⁺ cells in neurosphere cultures (n=3, ***P<0.001). (f) RT–PCR reveals that RA significantly decreased the expression of stem cell markers including CD133, nestin and Msi-1 (vs 18S, n=4, ***P<0.001). (g) Flow cytometric analysis of CD133 expression shows a significant decrease in cells expressing CD133 after RA treatment (from 44 to 10%, n=4, ***P<0.001).

Quantitative RT–PCR was used to examine the effects of RA on neurosphere cell expression of msi-1, nestin and CD-133, markers associated with the stem and progenitor cell pools. RA treatment for 48 h dramatically reduced CD-133, msi-1 and nestin expression by 65, 50 and 75%, respectively (Figure 2f, P<0.001). Flow cytometry with propidium iodide-conjugated anti-CD133 antibody confirmed that the fraction of CD133+ cells decreased from ~43 to ~10% in response to RA (Figure 2g, P<0.001).

It has been reported that RA can induce a ‘differentiation-like’ state in tumor cells, but the effect is often reversible (Tang and Gudas, 2011). To determine if the effect of RA on GBM-SC growth and colony formation is readily reversible, we pretreated GBM neurospheres with RA for 96 h and then cultured the cells in normal medium without RA. RA pretreatment for 96 h dramatically inhibited cell growth (Supplementary Figure 3A). When RA-pretreated cells were plated in soft agar in the absence of RA, the number of large colonies (>100 µm diametre) decreased by nearly 50% (Supplementary Figure 3B, P<0.001). In GBM-DM and KK cultures, RA and RA pretreatment both dramatically decreased their colony formation in soft agar (Supplementary Figures 3C and D, P<0.001). Flow cytometry analysis also indicated that the number of CD133+ cells in RA-pretreated GBM-SCs were significantly decreased compared with control, indicating that RA depletes the stem cell pool in the neurosphere cultures (Supplementary Figure 3E).

RA inhibits GBM neurosphere tumor propagation

We examined the effect of RA on the capacity of GBM-derived neurospheres to support and initiate xenograft growth. Multiple complementary in vivo experiments were performed. First, viable GBM neurosphere cells were implanted subcutaneously to immunodeficiency (SCID) mice. After several days, when tumors reached a measurable size, RA administration was initiated at 1.5 µg/kg/day i.p. RA significantly inhibited the growth of these preestablished neurosphere-derived xenografts (Figure 3a). By day 12 of treatment, RA-treated tumors were ~75% smaller than the controls (n=5, P<0.01) (Figure 3b). Western blot analysis of tumor extracts revealed that in response to RA, Tuj1 and GFAP expression significantly increased. Although there is some degree of variability in GFAP/Tuj1 expression among different xenografts, which is not unusual for in vivo studies, we found that in response to RA, the expression of GFAP and Tuj1 increased ~5.4- and 3.2-fold, respectively (Figure 3c, P<0.05, n=4). The decrease in tumor growth and increase in differentiation marker expression are consistent with the depletion of xenografts of their tumor-propagating stem-like cells. To test this hypothesis, primary neurospheres were established from control and RA-treated xenografts. Tumors from RA-treated group generated 70% fewer neurospheres than controls (Figure 3d, n=4, P<0.001).

Figure 3.

RA inhibits subcutaneous tumor xenograft growth. Subcutaneous tumors derived from GBM-KK neurosphere cells were treated daily with RA (1.5 µg/kg, i.p.). Control animals received DMSO. (a) Tumors retrieved from animals after 12 days of RA treatment are shown. Bar=5 mm. (b) Tumor was measured daily and volumes calculated as described in Materials and methods. By treatment day 12, RA significantly inhibited the growth of tumors by 75% (n=5, **P<0.01). (c) Proteins from tumor samples were subjected to immunoblot analysis. There was a significant increase in GFAP and Tuj1 expression in RA-treated subcutaneous tumors. Each lane represents individual xenograft. (d) Cells were dissociated from subcutaneous tumors and grown in neurosphere culture medium. RA-treated tumors generated fewer neurospheres after reculturing (n=4, ***P<0.001).

GBM neurosphere cells were implanted intracranially and animals were treated with RA (1.5 µg/kg/day i.p.) beginning on postimplantation day 14. After 2 months, animals were sacrificed and tumor sizes quantified in brain sections. Tumors were significantly smaller in RA-treated animals compared with dimethyl sulfoxide (DMSO)-treated controls (5.7 vs 18.1mm³, n=6, P<0.001) (Figures 4a–c). Ki-67 staining showed that there are about 34.3% Ki-67+ cells in control xenografts vs 10.2% Ki-67+ cell in RA-treated tumors (Figure 4b, P<0.001, n=6), consistent with the decreased tumor size after RA treatment. In addition, RA treatment increased median survival by ~50%, from 33.1±3.5 to 47.5±3.3 days (Figure 4d, P<0.05). Immunofluorescence was performed to determine if RA induced differentiation of orthotropic xenografts. Figures 4e and f show increased staining for GFAP in RA-treated brain slices. Co-labeling with anti-human nuclear antigen localized GFAP expression to the human tumor cells.

Figure 4.

RA inhibits intracranial tumor xenograft growth. Intracranial tumors derived from HSR-GBM1B neurosphere cells (a, b, d–g) or GBM-KK (c) were treated daily with RA (1.5 µg/kg, i.p.). Control animals received DMSO. (a) H&E stained brain sections from animals after 30 days of treatment. Bar=1 mm. Insets in (a) show a higher magnification of tumor staining. Bar=20 µm. (b) Ki-67 staining of DMSO and RA-treated xenografts. There are about 34.3% Ki-67+ cells in control xenografts vs 10.2% Ki-67+ cell in RA-treated tumors (P<0.001, n=6). Bar=20 µm. (c). Tumor cross-sectional areas were measured and tumor volumes were calculated as described in Materials and methods. RA reduced tumor size by 75% (n=6, ***P<0.001). (d) RA treatment prolonged median survival (33.1 vs 47.5 days, P<0.05, n=6). Data obtained from GBM-KK neurosphere cells are shown. (e, f) Immunofluorescent staining shows that GFAP expression by human glioma cells in orthotropic xenografts was increased after RA treatment. Bar=20 µm (d) and 5 µm (e). (g, h). Neurosphere cells were pretreated with RA for 96 h before intracranial implantation. Equal numbers of viable control or RA-pretreated cells were implanted to brain. Nine of 10 animals (90%) injected with control cells developed tumors. RA-pretreated cells formed detectable tumors in 3 of 10 animals (30%, n=3, **P<0.01). Tumors from RA-pretreated cells were also significantly smaller than controls (*P<0.05).

Neurosphere cells were pretreated with RA for 96 h before intracranial implantation to determine if RA reduced the capacity of neurospheres to propagate xenografts. Equal numbers of viable control or RA-pretreated cells were implanted to brain. After 2 months, animals were sacrificed and brain sections were analyzed for tumor formation. Nine of 10 animals (90%) injected with control cells developed tumors. RA-pretreated cells formed detectable tumors in only 3 of 10 animals (30%) (Figure 4g, P<0.01). Tumors propagated from RA-pretreated cells were also significantly smaller than controls (Figure 4h, P<0.05).

Gene expression response to RA

The response of GBM-derived neurospheres to RA provides an excellent model for identifying molecular mechanisms that regulate the proliferation, differentiation and tumor-propagating capacity of GBM-SCs. The transcriptional responses to RA at 8 h (early response) and 48 h (late response) after RA treatment were examined using Affymetrix gene expression arrays. A comprehensive list of gene expression changes can be found in Supplementary Table 1 and a short gene list is in Table 1a. The expression of ~350 genes was significantly altered by RA after 48 h treatment, with ~20% downregulated and 80% upregulated. Prominent were changes in the expression of genes coding for proteins involved in RA signaling and metabolism, cell-cycle regulation, neuronal differentiation, cell–cell and cell–matrix interaction and cytoskeleton remodeling (Table 1a). RA-induced expression of both RA receptor α and β (RARα and RARβ) indicating positive feedback for RA signaling. RA also altered the expression of basic helix–loop–helix DNA binding proteins, including ATOH8 (atonal homolog 8, induced ~3-fold), oligodendrocyte lineage transcription factor 1 and 2 (OLIG1, OLIG2, inhibited ~50%), and basic helix–loop–helix family member e40 (BHLHE40, induced ~2-fold). This is consistent with the preferential differentiation response along neural and astroglial lineage (not oligodendrocytic lineage) observed by immunofluorescent analysis. RA was found to inhibit cyclin D1 expression by ~40%, independently validating the decrease in cyclin D1 protein levels (Figures 1d and e). We selected 10 genes for further validation using quantitative RT–PCR. These include those coding for regulators of RA-mediated gene transcription (RARs and co-activators: Sox6, RARα, RARβ and nuclear receptor co-activator 3 (NCOA3)), neuronal differentiation (basic helix–loop–helix member: ATOH8) and cell adhesion (tenascin, cadherin, integrin β3 (ITGB3), intercellular adhesion molecule 1 (ICAM1) and fibronectin leucine rich transmembrane protein 1 (FLRT1)). All showed the same pattern of upregulation or downregulation in HSR-GBM1A and 1B. We further validated the expression change of 10 genes in two other GBM neurosphere lines as well as in low passage GBM-derived neurospheres. The results are summarized in Table 1b.

Table 1.

a Gene expression change after retinoic acid treatment for 48 h
Gene symbol	Gene title	Fold change	P-value
Lineage determination-related genes
ASCL1	Achaete–scute complex homolog 1	0.26	4.57E-07
OLIG1	Oligodendrocyte transcription factor 1	0.45	4.89E-06
OLIG2	Oligodendrocyte transcription factor 2	0.49	1.57E-05
SOX6	SRY-box 6	0.78	5.48E-06
SOX9	SRY-box9	0.86	6.23E-06
DCX	Doublecortin	0.38	9.04E-06
ATOH8	Atonal homolog 8	2.79	9.59E-08
KCNE4	Potassium voltage-gated channel	2.26	1.3E-05
KCNJ5	Potassium inwardly-rectifying channel	2.55	7.82E-06
Retinoids metabolism-related genes
DHRS3	Dehydrogenase/reductase (SDR family) member 3	4.25	8.21E-10
RBP1	Retinol binding protein 1	2.32	1.8E-07
RLBP1	Retinaldehyde binding protein 1	0.44	1.01E-06
Retinoids signaling-related transcription factors/co-activators
ZNF436	Zinc finger protein 436	5.13	6.03E-09
NCOA3	Nuclear receptor co-activator 3	2.37	6.73E-06
NRIP1	Nuclear receptor interacting protein 1	2.56	1.13E-05
RARB	Retinoic acid receptor, β	2.11	1.13E-05
BHLHE40	Basic helix–loop–helix family member 40		4.19E-05
Cell adhesion-related genes
FLRT1	Fibronectin leucine rich transmembrane protein 1	3.18	6.73E-08
ICAM1	Intercellular adhesion molecule 1	6.92	6.44E-10
CDH6	Cadherin 6, type 2, K-cadherin	3.45	4.04E-08
CHL1	Cell adhesion molecular, homolog of L1	2.49	1.28E-05

1b Microarray analysis confirmed by RT–PCR
Gene symbol (NCBI ref sequence)	Fold change in microarray	Fold change in HSR-GBM1A	Fold change in HSR-GBM1B	Fold change in GBM-KK	Fold change in GBM-DM	Fold change in JHH-551
SOX6 (NM_19829)	0.58	0.2	ND	ND	ND	ND
ATOH8 (NM_1267389)	2.79	39.3	20.63	ND	ND	39.47
ICAM1 (NM_1267389)	6.92	71.4	23.08	3.44	1.46	15.9
FLRT1 (NM_1267389	3.18	4.26	ND	ND	ND	28.6
ITGB3 (NM_1267389)	4.89	13.4	6.09	0.79	6.55	27.26
TNC (NM_1267389)	0.47	0.18	0.63	ND	ND	0.45
RARA (NM_1267389)	NS	2.64	18.86	ND	ND	ND
RARB (NM_1267389)	2.11	ND	13.6	5.59	2.67	ND
NCOA3 (NM_1267389)	2.38	2.07	1.06	1.23	0.99	5.33
CDH6 (NM_1267389)	3.46	4.53	2.25	3.88	1.28	1.64

Abbreviations: ND, not determined; NS, not significant.

Fold change is shown here.

A web-based pathway analysis (http://www.ingenuity.com) identified several molecular pathways significantly affected during RA-induced neurosphere differentiation (Tables 2a and b). Among them, Notch signaling was one of the most prominently altered pathways (P-value 3.006e-5). Six of the 34 known members of this pathway were altered by RA. Of these, Delta-like 1 (DLL1), Notch 1, NCID1, NEXT and Notch 1 precursor were downregulated by RA after 48 h. Only Delta-like 4 (DLL4) was upregulated by RA.

Table 2.

a Pathways affected by RA in GBM neurospheres (48 h vs control), analyzed with the ingenuity systems

b Members in Notch pathway affected by retinoic acid (RA)
Pathway name	P-value	Network objects altered by RA	Genes	Expression change
				8 h vs control	48 h vs control
Development Notch signaling pathway	3.006e-5	Six out of 34 members in this pathway	DLL1	Down	Down
			DLL4	Up	Up
			Notch1	Down	Down
			NCID	Down	Down
			NEXT	Down	Down
			Notch1 precursor	Down	Down

Abbreviations: GBM, glioblastoma multiforme; RA, retinoic acid.

Role for Notch pathway inhibition in GBM neurosphere response to RA

Based on the gene expression results, we hypothesized that RA promotes GBM growth inhibition and neurosphere differentiation, at least in part, by inhibiting Notch signaling. RA was found to downregulate neurosphere cell expression of the Notch pathway targets Hes2, Hey1 and Hey2 (Figure 5a). As the Notch receptor intracellular domain (NICD1) is a proximal activator of notch signaling, constitutive expression of NICD1 would be expected to rescue GBM neurospheres from RA responses mediated by the downregulation of Notch receptor expression or activation by RA. The forced expression of NICD1 in GBM neurosphere cells partially abrogated the inhibition of Hes2, Hey1 and Hey2 expression by RA (Figure 5a, P<0.01). We observe no obvious morphology change in cells transfected with NICD1. However, forced NICD expression partially reversed the RA-induced decrease in the expression of several stem cell markers (Figure 5b, P<0.05). Flow cytometric analysis of CD133 expression revealed that RA in conjunction with NICD1 overexpression increased CD133+ cells to 28.5%, compared with 16% in the presence of RA (Figure 5c, P<0.01). In cells treated with RA, NICD1 over-expression increased cell growth as evidenced by a decrease in cells at G1/G0 phase (from 70.4 to 65.5%) and an increase at S phase (from 10.1 to 15%, Figure 5d). Compared with RA alone, NICD1 increased cell number by 40% (Figure 5e, P<0.05) and colonies >100 µm in soft agar by 1.5-fold (Figure 5f, P<0.01). As RA has a stronger influence on glial differentiation, we examined the effect of NICD1 on GFAP expression. NICD1 expression decreased upregulation of GFAP in response to RA (~40% less vs RA, Figure 5g, P<0.01). Similar results were found in other GBM neurosphere lines (Supplementary Figure 4). These data indicate that Notch pathway downregulation mediates RA effects on GBM-SCs including cell growth arrest, differentiation and stem cell pool loss.

Figure 5.

NICD rescues RA-induced neurosphere differentiation. (a) HSR-GBM1B neurosphere cells were treated with lentivirus containing empty control (con) or NICD1 for 48 h before adding RA. RT–PCR shows that RA downregulated the Notch pathway targets Hes2, Hey1 and Hey2, which were rescued by NICD1 over-expression. (b) NICD1 partially restored the expression of msi-1, nestin and Sox-2, which was downregulated by RA, as revealed by RT–PCR (n=9, *P<0.05; **P<0.01; ***P<0.001). (c) Flow cytometric analysis of CD133 expression indicated that NICD1 over-expression increased CD133+ cells to 28.5%, compared with 16% with RA alone (n=4, **P<0.01). (d) Cell-cycle analysis of GBM neurosphere cells treated with RA±NICD1 over-expression. Compared with RA alone, NICD1 over-expression decreased cells at G1/G0 phase from 70.4 to 65.5% and increased cells at S phase from 10.1 to 15% (n=4). (e, f) In the presence of RA, NICD1 increased cell number in medium by 40% (e, n=4, *P<0.05) and colony number in soft agar by 1.5-fold (f, n=6, **P<0.01). (g) Western blot analysis shows that RA induced differentiation marker GFAP expression, which was partially reversed by NICD1 over-expression (n=3, **P<0.01).

Discussion

In this study, we examined the biological effects of RA on glioblastoma-derived stem-like cells grown as neurospheres and identified transcriptional responses to RA that mediate its biological functions. We found that RA induces cell growth arrest and differentiation of GBM neurosphere cells in vitro and in vivo. More importantly, we found that RA changes the expression of several groups of genes in GBM-SCs. These RA-responsive genes and pathways can guide further studies of GBM-SC differentiation. Moreover, we also found that Notch pathway downregulation by RA mediates the GBM-SC response. Developing these results further could lead to novel drug combinations for targeting GBM-SCs through differentiating mechanisms.

Several aspects of the cellular response of GBM-SCs to RA need to be commented. We found that RA induced differentiation of GBM-SCs as evidenced by the expression of neuronal markers and decrease of stem cell markers. The marker expression change is consistent in various cell lines we have studied, including one low passage primary neurosphere isolates directly from patient tumor. It has been reported that RA mainly induces glial differentiation in mouse neural stem cells by upregulating GFAP expression (Ray and Gage, 2006). We also found that RA has a stronger influence on GFAP expression vs Tuj1 expression, the marker of neurons, suggesting that RA mainly induces glial differentiation in human GBM-SCs as well. As marker expression is only one indication of the differentiation effect of RA, we further performed functional analysis of neurosphere formation, clonogenic and tumor propagation potential. All assays showed that RA treatment depletes the stem cell pool in the neurosphere cultures. The neurosphere assay used to identify the stem cell pool is based on the fact that most stem-like neoplastic cells generate large neurospheres in contrast to small neurospheres generated by progenitor cells (Galli et al., 2004; Bar et al., 2007; Sun et al., 2009). As RA also significantly inhibits cell growth, we used a secondary neurosphere/clonogenic assay, in which we pretreated GBM-SCs with RA, and analyze their colony formation ability and marker expression. There was persistent inhibition of clonogenic ability after RA withdrawal compared with controls, indicating fewer stem-like cells in the cultures. This also suggests that the effect of RA on GBM-SCs is stable. The in vitro results are also confirmed in the in vivo setting in which pretreated cells were injected into mouse brain and both the tumor size and tumor propagation rate are significantly decrease. Therefore, the results of marker expression, neurosphere/clonogenic assay, tumor propagation assay agree in concert that RA depletes stem cell pool and induces differentiation in GBM neurosphere cultures.

RA induces differentiation of EC cells and ES cells in vitro (Soprano et al., 2007) and is used to treat acute promyelocytic leukemia in clinical practice (Mongan and Gudas, 2007; Nasr et al., 2009). Previous studies examined the gene expression changes concurrent with RA-induced differentiation of EC and ES cells. A variety of genes were found to respond either directly or indirectly to RA including transcription factors, RA metabolism and transporter proteins, extracellular matrix proteins, proto-oncogenes, growth factors and their receptors, cytoskeletal proteins, apoptosis-related proteins, cell-cycle control proteins and proteins that mediate intracellular and extracellular signaling (Bain et al., 2000; Harris and Childs, 2002; Wei et al., 2002; Sangster-Guity et al., 2004; An et al., 2005). Our gene expression data in GBM neurosphere cells partially overlaps with the findings in EC and ES cells. For example, we found that RA upregulates genes related to retinol signaling and metabolism such as RAR α/β, NCOA3, retinol binding protein 1, nuclear receptor interacting protein 1 (NRIP1), suggesting there is a positive feedback for RA signaling (Mark et al., 2006). We also found that RA downregulates some genes involved in lineage determination, including OLIG1/2, achaete–scute complex homolog 1 (ASCL1), SRY-box 6/9 (Sox6/9) and so on. In the meantime, RA reduced neural progenitor marker expression (CD133, nestin, Sox-2, Msi-1 and doublecortin) and upregulated genes expressed in differentiated cells, KCNE4 and KCNJ5, which encode potassium channels. We also confirmed that RA regulates cell-cycle control proteins (cyclin D and cyclin D1), and proteins that mediate intracellular and extracellular signaling, including ICAM1, cadherin 6 (CDH6), ITGB3, FLRT1 and so on.

In our gene expression array data, we found that several Notch pathway components were significantly downregulated by RA. Notch signaling was indeed found to be downregulated by determining the expression of its downstream target genes Hes2, Hey1, Hey2 and Hey5. The fact that expression of the active form of Notch 1 (NICD1) can partially reverse the GBM stem cell response to RA further confirmed that Notch pathway mediates RA-induced differentiation and stem cell depletion. The involvement of Notch is particularly interesting because of the high level of conservation of this signaling pathway in cell fate determination and pattern formation across metazoans (Wilson and Radtke, 2006). It is well known that Notch signaling has a role in cell fate determination during development. Notch signaling also has an important role in the maintenance of neural stem cells (Kageyama et al., 2009). Our findings also agree with the results published by others (Murata-Ohsawa et al., 2005; Fan et al., 2010; Wang et al., 2010). Murata-Ohsawa et al. (2005) showed that delta-1 ligand activated Notch signaling alters RA-induced differentiation responses and reduces RA-induced apoptosis in myeloid leukemia cells. Fan et al. (2010) proposed that Notch pathway blockade inhibits the tumor-propagating capacity of GBM-derived neurospheres and depletes CD133+ stem-like cells.

Our gene expression profile also indicated that downregulation of several genes by RA may be mediated by Notch pathway inhibition. For example, nestin is shown to be downregulated by RA and restored by NICD. Nestin is in fact a direct Notch/RBPJk signaling target (Shih and Holland, 2006). The extracellular matrix protein tenascin C is significantly downregulated by RA. Studies from Sivasankaran et al. (2009) have suggested a possible link between Notch signaling and tenascin C in glioblastoma.

Wnt and Notch signaling pathways have important roles in regulating stem cell proliferation and cell differentiation. In general, Wnt activation tends to promote while Notch tends to inhibit neuronal differentiation (Shi et al., 2008). In our GBM neurospheres, some Wnt pathway members were upregulated by RA. The relationship between RA and the Notch and Wnt pathways is variable, contextual and cell type specific. In pluripotent human EC cells (Walsh and Andrews, 2003), RA treatment alters expression of Wnt receptor frizzled family members, the Frizzled Related Protein family, and receptors of the Notch pathway. Ginestier et al. (2009) found a role for retinoid signaling in the regulation of breast cancer stem cell self-renewal and differentiation and analyzed gene sets and pathways associated with retinoid signaling. They found Wnt to be upregulated and the Notch pathway unaffected in breast cancer stem cells treated with RA. The interplay between the Wnt and Notch pathways and their role in cancer stem cell maintenance and differentiation requires more detailed work in the future.

RA has been used to target GBM in the preclinical and clinical settings, but responses have been modest at best. Ray and colleagues have shown that RA downregulates telomerase activity and survival pathways and enhances chemotherapy-induced apoptosis in the U87 GBM cells grown under traditional serum-containing conditions (Zhang et al., 2007; Das et al., 2008; Karmakar et al., 2008). Chearwae and Bright (2008) found that RA inhibits the growth of U87-derived CD133+ cells grown as neurospheres and concluded that this activity was related to the peroxisome proliferator-activated receptor-γ agonist activity of RA. See et al. (2004) performed a retrospective analysis of patients treated with RA for recurrent glioblastoma. Phase II studies combining temozolomide, radiotherapy and cis-RA (cRA) in patients with GBMs have demonstrated mixed results. For instance, Butowski et al. (2005) demonstrated that cRA in combination with chemotherapy and radiotherapy had no effect on newly diagnosed malignant gliomas, whereas both Yung et al. (1996) and Jaeckle et al. (2003) concluded that cRA in combination with current standard of care was beneficial in recurrent gliomas. Our findings regarding the biological and transcriptional responses to RA in GBM-SCs suggest new avenues for exploiting RA or its targets for brain cancer therapeutics. Further studies are needed to identify the mechanistic basis for these effects in hopes of elucidating the best therapeutic strategy for induced cell-differentiation therapy.

In conclusion, the GBM neurosphere response to RA provides a platform for understanding the pathways important for cancer stem cell maintenance, proliferation, differentiation and tumorigenicity. Figure 6 presents a summary of our findings. RA binds to its nuclear receptors RARα or RARβ. The subsequent upregulation of these receptors generates a positive feedback loop for RA signaling. RARs regulate gene expression including those related to lineage determination and cell differentiation. This results in loss of stem cell markers and differentiation. RA also regulates cellcycle control genes with cell-cycle arrest. In addition, RA alters genes coding for proteins that regulate cell– cell and cell–matrix interactions. These adhesion interactions have the capacity to further enhance cellular differentiation. All these responses to RA act together to reduce tumor propagation and inhibit tumor growth. More specifically, we identified the Notch pathway as a downstream target of retinoids and the inhibition of Notch signaling by RA mediates, in part, the biological effect of RA in GBM neurospheres. A more detailed understanding of the molecular responses of neoplastic stem-like cells to RA may ultimately contribute to novel therapeutic approaches to stem cell-based cancer therapy.

Figure 6.

Schematic diagram of the molecular and cellular response of GBM-SCs to all-trans RA. RA binds to its nuclear receptors to activate gene expression including those related to lineage determination and cell differentiation, cell cycle and cell–matrix interaction. These result in loss of stem cell markers, differentiation, cell-cycle arrest and morphology change. All these cellular responses to RA act together to reduce tumor propagation and inhibit tumor growth. In addition, Notch pathway is the downstream target of retinoids and the inhibition of Notch signaling partially mediates the biological effect of RA in GBM neurospheres. RARE, RA response element.

Materials and methods

Reagents

All reagents were purchased from Sigma Chemical Co. (St Louis, MO, USA) unless otherwise stated. All-trans RA was prepared as stock solutions in DMSO and diluted in cell culture medium. In all the experiments, the final DMSO concentration was ≤0.1% and DMSO had no demonstrable effect on neurosphere cultures.

Cell culture

The human glioblastoma neurosphere lines HSR-GBM1A (20913) and HSR-GBM 1B (10627) were originally derived by Vescovi and colleagues and maintained in serum-free medium supplemented with epidermal growth factor and fibroblast growth factor (Vescovi et al., 1999; Galli et al., 2004; Bar et al., 2007; Sun et al., 2009). Cells were incubated in a humidified incubator containing 5% CO₂/95% air at 37 °C, and passaged every 4–5 days. The GBM-DM140207 and GBM-KK190156 neurosphere lines were derived from glioblastomas at the University of Freiburg and kindly provided by Dr Jaroslaw Maciaczy. The primary neurospheres JHH551 was derived from a malignant glioma at Johns Hopkins using the same methods and culture conditions as described in Galli et al. (2004). JHH551 neurospheres were used at passage ≤10. All human materials were obtained and used in compliance with the Johns Hopkins IRB.

Clonogenic assays

For soft agar clonogenic assays, 1% agarose in Dulbecco’s Modified Eagle’s medium was cast on the bottom of plastic 6-well plates. Dissociated neurosphere cells (5×10³) were suspended in neurosphere culture medium containing 0.5% agarose and placed on top of the bottom layer. Cells were incubated in neurosphere culture medium±RA for 7–14 days and colonies were fixed and stained with Wright’s solution (1%). Colony formation was analyzed by computer-assisted morphometry (MCID) by measuring the number of neurospheres >100 µm in diameter in three random microscopic fields per well.

Flow cytometric analysis

Cell cycle was analyzed by flow cytometry on a FACSCalibur (Becton-Dickinson, Mountain View, CA, USA) (Reznik et al., 2008). Briefly, GBM neurosphere cells were collected and dissociated by vigorously pipetting. Cells were fixed with 75% ethanol at 4 °C for 30 min. Cells were incubated with DNase-free RNase at 37 °C for 30 min followed by propidium iodide (100 ng/ml) for 1 h at 37 °C. The percentage of cells at each cell-cycle phase (G1/G0, S and G2/M) was analyzed using CellQuest software (Becton-Dickinson).

The number expression of CD133, Tuj1 and GFAP were analyzed by flow cytometry. For the cell surface marker CD133, dissociated neurosphere cells (1×10⁶) were suspended in 100 µl assay buffer (phosphate buffered saline (PBS) pH 7.2, 0.5% bovine serum albumin and 2mM EDTA) and incubated with 10 µl of phycoerythrin-conjugated anti-CD133 antibody (Miltenyi Biotec, Auburn, CA, USA) for 10 min in the dark at 4 °C. For Tuj1 and GFAP expression, cells were first fixed with 4% paraformaldehyde for 30 min at 4 °C and permeabilized with PBS containing 0.5% Triton X-100 for 5 min. The cells were then incubated with primary antibodies (anti-Tuj1, 1:1000, Millipore, Billerica, MA, USA; anti-GFAP, 1:5000, DAKO, Carpinteria, CA, USA) for 2 h and then incubated with appropriate corresponding secondary antibodies conjugated with fluorescent dye (1:1000, The Jackson Laboratory, Bar Harbor, ME, USA) for 30 min. Cells were rinsed and pelleted by centrifugation. The cell pellet was resuspended in PBS and analyzed by FACSCalibur.

Western blot analysis

Cells were lysed with radioimmunoprecipitation assay buffer (50mM Tris–HCl, pH 7.4, 150mM NaCl, 1% NP-40 and 0.25% Na-deoxycholate) containing 1× protease and phosphatase inhibitor cocktail (Calbiochem, San Diego, CA, USA). After sonication for 15 s, the suspensions were centrifuged at 3000 g for 10 min. Protein concentrations were determined using the Coomassie Protein Assay Reagent (Pierce, Rockford, IL, USA). SDS-PAGE was performed on 30 µg of cellular protein per lane using 4–20% gradient Trisglycine gels according to the method of Towbin et al. (1979) with some modifications (Reznik et al., 2008). Proteins were electrophoretically transferred onto nitrocellulose membranes (GE Healthcare, San Francisco, CA, USA). Membranes were incubated for 1 h in Odyssey Licor Blocking Buffer (LI-COR Biosciences, Lincoln, NE, USA) at room temperature and then overnight with primary antibodies at 4 °C in Odyssey Blocking Buffer. After rinsing, membranes were incubated with IRDye secondary antibodies (1:15 000–1:20 000, LI-COR Biosciences) and protein expression changes were quantified by dual wavelength immunofluorescence imaging (Odyssey Infrared Imaging System, LI-COR Biosciences) scanning of the membranes as previously described (Goodwin et al., 2010). The primary antibodies were anti-GAPDH (1:7500, Santa Cruz Biotechnologies, Santa Cruz, CA, USA), anti-β-actin (1:6000), anti-GFAP (1:500) and anti-Tuj1 (1:1000). Antibodies against cyclin, cyclin A, cyclin E, p21 and p27 were purchased from Santa Cruz and were diluted according to the manufacturers’ recommendations.

Immunofluorescence and immunohistochemistry

Neurosphere cells were grown on coverslips or collected by cytospin onto glass slides. The cells were fixed with 4% paraformaldehyde for 30 min at 4 °C and permeabilized with PBS containing 0.5% Triton X-100 for 5 min. The cells were then incubated with primary antibodies for 2 h and then incubated with appropriate corresponding secondary antibodies conjugated with fluorescent dye (1:1000) for 30 min. The primary antibodies were anti-nestin (1:200, Sigma), GFAP (1:400) and Tuj1 (1:1000). Coverslips were mounted with Vectashield Antifade solution containing 4′-6-Diamidino-2-phenylindole (DAPI) (Vector Laboratories, Burlingame, CA, USA) and observed under fluorescent microscopy. Immunofluorescent images were taken and analyzed using Axiovision software (Zeiss, Oberkochen, Germany).

Tumor xenografts

Female 4- to 6-week-old Athymic Nude mice were injected s.c. in the flank with 5×10⁶ viable cells in 0.1 ml of PBS. When tumors reached ~50mm³, the mice were randomly divided into two groups and treated with RA 1.5 µg/kg, i.p. daily or with solvent DMSO as control. Animal were treated with RA during the entire period of experiments because of rapid metabolism of some retinoids and the duration of RA treatment is based on previous published RA work on glioblastoma xenografts (Karmakar et al., 2008). Tumor sizes were determined daily by measuring two dimensions (length (a) and width (b)) and volumes (V) were estimated using the formula V=ab²/2 (Lal et al., 2005). At the end of each experiment, tumors were minced and dissociated tumor cells were cultured in neurosphere culture medium.

For intracranial xenografts, SCID immunodeficient mice received 5000 (GBM-KK) or 10 000 (HSR-GBM1B) viable neurosphere cells in 5 µl of culture medium by stereotactic injection to the right caudate/putamen. Cell viability was determined by trypan blue dye exclusion. Groups of mice (n=6) were sacrificed at the indicated times and the brains were removed for histological studies. The tumor cross-sectional areas were measured based on H&E stained cryostat sections from perfusion-fixed brains using computer-assisted image analysis as previously described (Lal et al., 2005). Tumor sizes were determined according to the following formula: tumor volume=(square root of maximum cross-sectional area)³. All animal protocols used in this study were approved by the Johns Hopkins School of Medicine Animal Care and Use Committee.

Microarray analysis

Total cellular RNA was extracted using the RNeasy Mini kit (Qiagen, Inc., Chatsworth, CA, USA) and purified using RNeasy columns according to the manufacturer’s instructions. The integrity of rRNA was checked using agarose gel electrophoresis. Two GBM neurosphere lines, HSR-GBM1A and HSR-GBM1B were used. For each cell line, three time points (0, 8 and 48 h after RA treatment) were analyzed. The hybridizations were performed in the Johns Hopkins Microarray core facilities using ExonArray (Affymetrix, Santa Clara, CA, USA). We combined the data set for each condition from the two cell lines (n=4) and analyzed the data with Significant Analysis of Microarrays (http:// www-stat.stanford.edu/~tibs/SAM/) with the false discovery rate set at 20%. The raw data have been submitted to the NCBI website (http://www.ncbi.nlm.nih.gov/geo/, accession number GSE21336). Pathways affected by RA were analyzed using the Ingenuity Systems (http://www.ingenuity.com).

Quantitative real-time PCR

Quantitative real-time PCR (RT–PCR) was performed according to Pfaffl (2001). The primers used are listed in Supplementary Table 2. Total RNA (1 µg) was reverse transcribed using the oligo(dT)12–18 primer and Superscript II (Life Technologies, San Diego, CA, USA) according to the manufacturer’s instructions. RT–PCR was performed with an Applied Biosystems Prism 7900 HT Sequence Detection System using SYBR Green PCR Master Mix (Life Technologies). The thermal cycling conditions were as follows: 95 °C for 5 min, followed by 40 cycles of 95 °C for 10 s, 55 °C for 10 s and ended with 72 °C for 30 s. Samples were amplified in triplicate and data were analyzed using the Applied Biosystems Prism Sequencer Detection Software Version 2.3 (Life Technologies). Human 18S rRNA was amplified as endogenous control. Relative expression of each gene was normalized to the 18S rRNA control.

Viral expression

Viral expression vector FugW and FugW containing the Notch intracellular domain 1 were obtained from Dr LinZhao Chen (Johns Hopkins University) (Yu et al., 2008). Virus was packaged using the viral power package system (from Invitrogen) according to the manufacturer’s instructions. Virus was collected by centrifuging at 3000 r.p.m. for 10 min. Neurosphere cells were transfected with virus containing FugW or NICD 48 h before RA was added.

Statistical analysis

Data were analyzed using parametric statistics with one-way analysis of variance. Post hoc tests included the Student’s t-Test and the Tukey multiple comparison tests as appropriate using Prizm (GraphPad, San Diego, CA, USA). All experiments reported here represent at least three independent replications. All data are represented as mean value±s.e. significance was set at P<0.05.