A central confounding factor in the development of targeted therapies is tumor cell heterogeneity, particularly in tumor-initiating cells (TIC), within clinically identical tumors. Here, we show how activation of the Sonic Hedgehog (SHH) pathway in neural stem and progenitor cells creates a foundation for tumor cell evolution to heterogeneous states that are histologically indistinguishable but molecularly distinct. In spontaneous medulloblastomas that arise in Patched (Ptch)+/− mice, we identified three distinct tumor subtypes. Through cell type–specific activation of the SHH pathway in vivo, we determined that different cells of origin evolved in unique ways to generate these subtypes. Moreover, TICs in each subtype had distinct molecular and cellular phenotypes. At the bulk tumor level, the three tumor subtypes could be distinguished by a 465-gene signature and by differential activation levels of the ERK and AKT pathways. Notably, TICs from different subtypes were differentially sensitive to SHH or AKT pathway inhibitors, highlighting new mechanisms of resistance to targeted therapies. In summary, our results show how evolutionary processes act on distinct cells of origin to contribute to tumoral heterogeneity, at both bulk tumor and TIC levels. Cancer Res; 74(17); 4864–74. ©2014 AACR.

While the existence of tumor-initiating cells (TIC) in some tumor types remains controversial, their existence has been well established for brain tumors, both in human and mouse models (1–4). Because TICs have been shown to be more resistant to chemotherapy and radiotherapy than bulk tumor cells and are thought to be responsible for tumor recurrence, there is great interest in developing novel therapies to target these cells (5). However, progress is impeded by both intra- and intertumoral heterogeneity in TIC phenotypes even within the same clinical tumor type (3). Hence, elucidating the source and functional consequences of TIC heterogeneity is a critical step in advancing this field.

Medulloblastomas account for approximately 20% of all brain tumors in children (6), and they form in the cerebellum, a brain region that develops postnatally through massive expansion and differentiation of neural progenitors in the external granule layer (EGL; ref. 7). Molecularly, human medulloblastomas can be divided into four subgroups: SHH, WNT, Group 3, and Group 4 (8). The “SHH (sonic hedgehog)” subgroup of tumors constitute approximately 25% of human medulloblastomas, and are recognized by high levels of the SHH pathway activity. During normal cerebellar development, the SHH pathway drives rapid proliferation of EGL progenitor cells. Hence, it is not surprising that mutations that activate SHH pathway genes, such as PTCH, GLI1/2, SUFU, or SMO, are common in the SHH subgroup of medulloblastomas (9, 10). Interestingly, even among the SHH subgroup of patients with medulloblastoma, there is heterogeneity in terms of prognosis, age of onset, and clinical pathology (11), suggesting that further analyses of tumor heterogeneity within this subgroup is necessary to better understand and treat these tumors.

Patched (Ptch) is a negative regulator of the SHH pathway, and the Ptch+/− mice develop spontaneous medulloblastomas, similar to patients with Gorlin syndrome (12, 13). Consequently, Ptch+/− mice have been extensively used to understand the etiology of medulloblastomas (14) and to test the efficacy of SHH pathway inhibitors for cancer treatment (15, 16). Previous studies reported that TICs in this model are CD15+ cells (4, 17). However, the reported ex vivo cellular phenotype of TICs in Ptch tumors was inconsistent: Read and colleagues indicated that TICs from Ptch medulloblastomas do not form tumorspheres in culture (4), whereas Ward and colleagues reported that they formed tumorspheres in serum-free neural stem cell (NSC) medium (17). While tumorsphere formation is not strictly correlated with tumor initiation, tumorsphere cultures are commonly used to propagate TICs ex vivo, based on the observations that NSCs can be propagated and maintained as spheres in serum-free culture medium (18). Variable sphere formation abilities from individual tumors of the same clinical subtype have been reported by several laboratories, indicating that it is not due to lab-to-lab variations in techniques or reagents but may represent inherent differences among tumors.

We hypothesized that the differential ability of TICs to survive and proliferate in culture may reflect intrinsic biologic differences among different tumors. Specifically, the fact that cellular transformation can occur in either NSCs or neural progenitor cells (NPC) in the Ptch+/− mouse suggests that TICs in these medulloblastomas may have different cellular behaviors depending on the cell of origin. We tested this hypothesis through systematic analyses of a large number of spontaneous Ptch tumors and by cell type–specific activation of the SHH pathway in vivo.

We now report that medulloblastomas in Ptch+/− mice are composed of three subtypes, distinguishable at both TIC and bulk tumor levels. This study directly demonstrates the differential outcomes of transforming stem versus progenitor cells by the same oncogenic event in vivo, and shows that the cell of origin is a major contributor of tumor heterogeneity. Furthermore, it shows that the pathways driving TICs and bulk tumor cell proliferation and survival can be distinct in the same tumor, cautioning against selection of targeted therapies solely based on the dominant molecular signature of bulk tumor cells.

Intracranial injections and limiting dilution assays

Freshly dissociated tumor cells were injected into the cerebella of NSG mice using a stereotaxic device (Bregma: +1/−6.5/−3.0). TIC frequency was calculated using the ELDA software (19).

Microarray

Microarray analysis was performed using Affymetrix St1.0 chips using independent no growth (NG; n = 7), growth factor–dependent (GFD; n = 6), and growth factor–independent (GFI; n = 4) samples. One-way ANOVA was used to identify genes that distinguish the three tumor subtypes. A list of 465 genes identified in this analysis was used for principal component analysis and to identify potentially enriched biologic processes using the GSEA software (20). Ingenuity Pathway Analysis of the 465 genes was performed on the Spring 2012 release by pairwise comparisons of GFD-GFI, GFD-NG, and NG-GFI tumors. The Affymetrix gene expression dataset has been deposited in GEO.

Real-time Reverse Transcription-PCR

Real-time RT-PCR analyses were performed using the iQ5 Optical System from Bio-Rad with SYBR Green mix and primers indicated in the Supplementary Materials and Methods.

Spectral karyotyping

Metaphase spreads were prepared from p1 to p3 tumorsphere cells from different culture subtypes and prepared using SkyPaint following the manufacturer's instructions (Applied Spectral Imaging). Captured images were analyzed using Applied Spectral Imaging HiSKY5.0 software.

Immunohistochemistry

Antibodies listed in Supplementary Materials and Methods were used on frozen and paraffin sections following standard protocols.

Immunoblotting

Thirty to 50 μg of total lysates from tumor tissues were separated on 10% SDS-PAGE gel. Standard immunoblotting technique was used.

Statistical analysis

GraphPad Prism software was used to generate survival curves and calculate statistical significance. For additional information, please see Supplementary Materials and Methods.

Spontaneous medulloblastomas in Ptch+/− mice are composed of three distinct subtypes

To systematically analyze cellular and molecular phenotypes of tumor cells in each spontaneous Ptch+/− medulloblastoma, we performed multiple parallel analyses in vivo and ex vivo (Fig. 1A). First, from each spontaneous (Generation1: G1) tumor, we propagated the tumor by directly injecting freshly dissociated tumor cells, without culture, orthotopically into cerebella of NOD-SCID;Il2gr−/− (NSG) mice. In parallel, single tumor cells were plated at a clonal density (1 cell/μL) in serum-free NSC media containing basic fibroblast growth factor (bFGF) and EGF or in serum-free medium containing no exogenous growth factors (NSC-GF medium) and tumorsphere formation was measured 7 days later. If no spheres were present in either media, the tumor was designated as the NG subtype; if spheres were present only in NSC medium, the tumor was designated as the GFD subtype; and if spheres were present in NSC-GF media, the tumor was designated as the GFI subtype for its potential to grow in the absence of exogenous growth factors (Table 1). As shown in Table 1, 38% of all tumors analyzed were classified as NG and 38% and 24% were classified as GFD and GFI, respectively. Interestingly, the distribution of the subtypes did not vary significantly from Ptch;p53+/+, Ptch;p53+/−, and Ptch;p53−/− mice (Table 1). However, the loss of p53 accelerated tumor latency in Ptch+/− mice (Supplementary Fig. S1A), consistent with a previous report (21).

Figure 1.

Systematic analysis of TICs in Ptch+/− medulloblastomas reveals three distinct subtypes. A, a schematic of in vivo and ex vivo analyses performed with spontaneous medulloblastomas that formed in Ptch+/−, Ptch;p53+/−, and Ptch;p53−/− mice. Spontaneous tumors (G1) were serially passaged orthotopically without culture in vivo. In parallel, freshly isolated tumor cells were tested for sphere formation at a clonal density. Similar analyses were performed at subsequent generations (G2 to G5). B, hematoxylin and eosin staining of the three subtypes of G1 and G3 tumors. Scale bar, 100 μm. C, survival analysis of mice injected with 1,000 G1 tumor cells from different subtypes of Ptch+/−;p53+/+ medulloblastomas. G2 mice injected with GFI cells die significantly earlier (P = 0.0554). D, G2 mice injected with 100,000 G1 cells show significantly longer survival for mice bearing NG tumors (P < 0.0105). See also Supplementary Fig S2. E, GFD and GFI tumorsphere cells express markers for neurons (MAP2), astrocytes (GFAP), and oligodendrocytes (NG2) when induced to differentiate as adherent cultures. Nuclei were stained with 4′,6-diamidino-2-phenylindole (DAPI; in blue).

Figure 1.

Systematic analysis of TICs in Ptch+/− medulloblastomas reveals three distinct subtypes. A, a schematic of in vivo and ex vivo analyses performed with spontaneous medulloblastomas that formed in Ptch+/−, Ptch;p53+/−, and Ptch;p53−/− mice. Spontaneous tumors (G1) were serially passaged orthotopically without culture in vivo. In parallel, freshly isolated tumor cells were tested for sphere formation at a clonal density. Similar analyses were performed at subsequent generations (G2 to G5). B, hematoxylin and eosin staining of the three subtypes of G1 and G3 tumors. Scale bar, 100 μm. C, survival analysis of mice injected with 1,000 G1 tumor cells from different subtypes of Ptch+/−;p53+/+ medulloblastomas. G2 mice injected with GFI cells die significantly earlier (P = 0.0554). D, G2 mice injected with 100,000 G1 cells show significantly longer survival for mice bearing NG tumors (P < 0.0105). See also Supplementary Fig S2. E, GFD and GFI tumorsphere cells express markers for neurons (MAP2), astrocytes (GFAP), and oligodendrocytes (NG2) when induced to differentiate as adherent cultures. Nuclei were stained with 4′,6-diamidino-2-phenylindole (DAPI; in blue).

Close modal
Table 1.

Cellular characteristics of TICs in Ptch medulloblastoma subtypes

Orthotopic tumor formationp53 statusTIC frequency
Subtypes (n = 68)Tumorsphere cultureGrowth factor dependenceDirect transferCultured cellsWt (n = 23)Het (n = 35)Homo (n = 10)95% confidence interval
NG (38%) No n.a. Yes (n = 18) n.a. 34% 43% 30% 1/18,395 
GFD (38%) Yes Yes Yes (n = 19) Yes (n = 5) 39% 31% 60% 1/918 
GFI (24%) Yes No Yes (n = 11) Yes (n = 6) 26% 26% 10% 1/217 
Orthotopic tumor formationp53 statusTIC frequency
Subtypes (n = 68)Tumorsphere cultureGrowth factor dependenceDirect transferCultured cellsWt (n = 23)Het (n = 35)Homo (n = 10)95% confidence interval
NG (38%) No n.a. Yes (n = 18) n.a. 34% 43% 30% 1/18,395 
GFD (38%) Yes Yes Yes (n = 19) Yes (n = 5) 39% 31% 60% 1/918 
GFI (24%) Yes No Yes (n = 11) Yes (n = 6) 26% 26% 10% 1/217 

To test whether this cellular phenotype (sphere formation in different culture conditions) is a cell-intrinsic characteristic of tumor-propagating cells in each tumor, we isolated and tested cellular phenotypes of Generation 2 (G2) tumors that formed from direct transfer of G1 tumors (Fig. 1A). Again, acutely dissociated cells from G2 tumors were orthotopically transplanted into NSG cerebella to generate G3 tumors, and, in parallel, were plated at a clonal density in NSC and NSC-GF media to test their growth properties ex vivo (Fig. 1A). We consistently observed that G2 tumor culture phenotypes remained identical to those of the G1 tumors. Specifically, the NG subtype of G1 tumors generated NG-subtype G2 tumors, GFD-subtype G1 tumors generated GFD G2 tumors, and GFI-subtype G1 tumors generated GFI G2 subtypes. The ex vivo growth phenotype of tumors tested in G3, G4, and G5 tumors remained identical to those of G1 tumors in >83% (39/47) of tumors tested, indicating that ex vivo sphere formation and growth factor dependence are variable and cell-intrinsic phenotypes of tumor-propagating cells in the original tumor.

More than 93% (44/47) of spontaneous tumors in Ptch mice formed G2 tumors upon orthotopic transplantation without culture. Hundred percent (35/35) of G2 tumors tested formed G3, G4, and G5 tumors via serial transplantation. At the histologic level, these transplanted tumors were identical to the G1 tumors and indistinguishable from each other (Fig. 1B). Together, these observations indicate that regardless of tumor subtype, Ptch medulloblastomas contain long-term, self-renewing stem cell–like cells that can initiate tumors upon serial transplantation for at least five generations in vivo.

TIC frequencies in vivo and TIC phenotypes ex vivo

To further analyze the three tumor subtypes, we measured the TIC frequency of G1 tumors by limiting dilution analysis in vivo. To do this, we injected 105, 104, 103, and 102 G1 tumor cells orthotopically into NSG mice cerebella (unlike cortical injections previously reported; ref. 22) and measured the incidence of tumor formation in recipient mice at each cell dose. The TIC frequency was 1 in 18,395 cells in the NG subtype, in which TICs require the in vivo microenvironment to proliferate and/or survive; 1 in 918 in the GFD subtype, in which the TICs require exogenous growth factors for proliferation and survival ex vivo; and 1 in 217 in the GFI subtype, in which forms TICs proliferate and self-renew in the absence of exogenous growth factors ex vivo (95% confidence interval; Table 1). Consistently, Kaplan–Meier survival analyses of G2 mice injected with 100 or 1,000 uncultured G1 tumor cells show that mice bearing the GFI subtype showed shorter survival time than the other two subtypes (P = 0.0554; Fig. 1C and Supplementary Fig. S2A). When survival times of G2 mice injected with 100,000 G1 tumor cells were compared, mice bearing NG tumor cells survived significantly longer than the other two subtypes (P = 0.0105; Fig. 1D). Analyses of cell proliferation and apoptosis in G1 and G2 tumors of each subtype showed no significant differences (Supplementary Fig. S2B), suggesting that the major driver for differential G2 tumor latency is the frequency of TICs in different G1 tumor subtypes.

We next analyzed self-renewal and multipotentiality of TICs ex vivo. As the NG subtype of tumorspheres cannot be cultured, we limited our ex vivo analyses to GFD and GFI tumorspheres. Acutely dissociated tumor cells were plated at a clonal density (1 cell/μL) and the resulting clonally derived tumorspheres were analyzed. Morphologically, GFD tumorspheres resembled normal neurospheres (Fig. 1E). In contrast, many GFI tumorsphere lines formed loosely adherent, amorphic spheres composed of smaller, EGL progenitor-like cells (Fig. 1E). Both GFD and GFI tumorspheres could be propagated for >6 passages at a clonal density (Supplementary Fig. S2C), suggesting that they are long-term self-renewing cells. The frequency of self-renewing cells in each tumorsphere line ranged from 0.5%–6%, depending on the individual tumor. We also tested whether tumorsphere cells were multipotential, that is, able to give rise to neurons, astrocytes, and oligodendrocytes, by plating dissociated GFD and GFI tumorsphere cells on poly-d-lysine–coated coverslips and culturing them in differentiation medium. GFD and GFI culture cells expressed markers for neurons (MAP2, ß-III-tubulin), astrocytes (GFAP), and oligodendrocytes (NG2, O4, CNPase), suggesting that they can differentiate along both neuronal and glial lineages (Fig. 1E and Supplementary Fig. S1B).

Finally, we tested the tumorigenic potential of tumorspheres cells by injecting low passage (<passage 5) cells into NSG mouse cerebella. Both GFD and GFI tumorsphere cells produced tumors that phenocopied the original tumor (Table 1; and Supplementary Fig. S1C and S1D). These observations indicate that tumorsphere cultures from both GFD and GFI subtypes contain long-term self-renewing, multipotential, and tumorigenic cells (TICs) that can be propagated ex vivo.

GFD tumorspheres do not depend on the SHH pathway

Because Ptch-induced medulloblastomas are thought to represent the SHH subgroup of human medulloblastoma (9) and bulk tumor cells in all tumors express high levels of the SHH pathway genes (Fig. 2A and Supplementary Fig. S3), we tested whether TICs in all subtypes also depend on the SHH pathway for self-renewal, proliferation, or survival. We plated GFD and GFI tumorsphere cells at a clonal density and treated them with a SHH pathway inhibitor, cyclopamine. All GFI cultures tested were sensitive to SHH pathway inhibition as anticipated (Fig. 2B); however, despite high levels of SHH pathway activation in GFD tumor tissues (Fig. 2A), all GFD TIC cultures tested were resistant to cyclopamine treatment (Fig. 2B). To test whether the dependence of GFI cultures on the SHH pathway is due to lack of exogenous growth factors (bFGF and EGF) in the medium, we cultured GFI TICs in the NSC medium and then tested their dependence on the SHH pathway. GFI TICs cultured in NSC medium remained exquisitely sensitive to cyclopamine treatment (Fig. 2C). These results showed that extrinsic growth factors, bFGF and EGF, do not affect TIC phenotypes but rather GFD TICs are resistant to SHH inhibitor through a cell-intrinsic mechanism. This surprising result suggested that while bulk tumor cells in GFD tumors express high levels of SHH pathway genes and Atoh/Math1, a marker for EGL progenitors (Fig. 2A and Supplementary Fig. S3), TICs in GFD tumors do not depend on the SHH pathway for proliferation, self-renewal, or survival.

Figure 2.

All subtypes express high levels of SHH pathway genes, but GFD TICs are resistant to Smo inhibition. A, expression levels of Gli1, Gli2, and Atoh1 in G1 tumors, measured by real-time RT-PCR and normalized to 18S levels. Values represent relative fold change compared with wild-type postnatal day 8 cerebellum. Each colored dot represents an independent tumor. Also see Supplementary Fig S3. B, self-renewal assay of low passage (<p5) GFD and GFI tumorsphere cells in the presence or absence of 10 μmol/L cyclopamine for 7 days in vitro. C, self-renewal of GFI cells is blocked by cyclopamine treatment even when cultured in NSC medium (n = 4). D, RT-PCR analysis of low passage (p<5) GFD and GFI tumorsphere cells in culture. Relative fold changes shown in a log scale, compared with wild-type cerebellar NSCs isolated from wild-type p8 cerebellum. Each bar represents tumorspheres from independent tumors.

Figure 2.

All subtypes express high levels of SHH pathway genes, but GFD TICs are resistant to Smo inhibition. A, expression levels of Gli1, Gli2, and Atoh1 in G1 tumors, measured by real-time RT-PCR and normalized to 18S levels. Values represent relative fold change compared with wild-type postnatal day 8 cerebellum. Each colored dot represents an independent tumor. Also see Supplementary Fig S3. B, self-renewal assay of low passage (<p5) GFD and GFI tumorsphere cells in the presence or absence of 10 μmol/L cyclopamine for 7 days in vitro. C, self-renewal of GFI cells is blocked by cyclopamine treatment even when cultured in NSC medium (n = 4). D, RT-PCR analysis of low passage (p<5) GFD and GFI tumorsphere cells in culture. Relative fold changes shown in a log scale, compared with wild-type cerebellar NSCs isolated from wild-type p8 cerebellum. Each bar represents tumorspheres from independent tumors.

Close modal

To test whether cultured tumorspheres, particularly GFD TICs, lose expression of the SHH pathway genes, we measured SHH pathway gene expression levels in GFD and GFI tumorsphere cells at low passage (<passage 5). Real-time RT-PCR analysis of multiple independent tumorsphere lines showed that compared with normal cerebellar NSCs, GFD tumorsphere cells express higher levels of Gli1 and SHH (Fig. 2D). Interestingly high-level Atoh1 expression was unique to GFI tumorspheres, suggesting that GFD tumorspheres may be blocked in their differentiation to EGL progenitor-like cells ex vivo.

NSCs and NPCs generate TICs with distinct cellular phenotypes

On the basis of sphere morphology, gene expression patterns, and growth factor requirements, we hypothesized that the GFD subtype originates from transformed NSCs. To directly test this hypothesis, we activated the SHH pathway in a cell type–specific manner in vivo. In SmoM2 mice, a constitutively active form of Smo is expressed upon Cre-mediated recombination (23). To express SmoM2 in early NSCs, we crossed SmoM2 mice with hGFAP-cre mice. The hGFAP-cre driver activates the transgene expression in cerebellar NSCs as early as E14.5 (24). Each SmoM2;GFAP-cre mice developed medulloblastomas by 3 to 4 weeks of age, as previously reported (23, 24). To analyze the cellular phenotypes of these tumor cells ex vivo, we dissociated SmoM2;hGFAP-cre tumor cells and plated them at a clonal density in NSC and NSC-GF media. Strikingly, 100% of the tumors tested formed tumorspheres only in NSC medium but not in NSC-GF medium, indicating that they are GFD tumors (Fig. 3A, n = 8). In contrast, when SmoM2 was activated only in committed EGL progenitors, by administering tamoxifen to SmoM2;Atoh1-CreER mice at postnatal days 7 to 10, none of the tumors generated tumorspheres in either NSC or NSC-GF medium, indicating they are NG tumors (Fig. 3B, n = 6). This result indicates that NG tumors arise from transformed EGL progenitor cells. Molecularly and histologically, SmoM2;GFAP-cre and SmoM2;Atoh1-CreER tumors could not be distinguished from each other or from Ptch tumors (Figs. 1B, 2A, and 3C–E and Supplementary Fig. S3). Together, these observations indicate that both NSCs and NPCs can be transformed by the SHH pathway activation in vivo but transformation of NSCs and NPCs result in TICs with drastically different cellular phenotypes.

Figure 3.

Cell type–specific activation indicates that GFD tumors arise from transformed NSCs while NG tumors arise from NPCs. A, activation of the SHH pathway by expressing SmoM2 (activated Smoothen) in NSCs in vivo (SmoM2;hGFAP-cre) generated only GFD tumors (8/8). Also see Supplementary Fig. S6 for histology of tumor derived from injecting SmoM2;hGFAP-cre tumorsphere cells. B, expression of SmoM2 in committed EGL progenitors (SmoM2;Atoh-creER) by tamoxifen treatment from p7–p10 generated only NG tumors (6/6). C and D, hematoxylin and eosin staining of SmoM2;hGFAP-cre and SmoM2;Atoh-creER tumors. E, SHH pathway gene expression in independent SmoM2;hGFAP-cre and SmoM2;AtohCreER tumor tissues. Relative fold change compared with wild-type p8 cerebellum.

Figure 3.

Cell type–specific activation indicates that GFD tumors arise from transformed NSCs while NG tumors arise from NPCs. A, activation of the SHH pathway by expressing SmoM2 (activated Smoothen) in NSCs in vivo (SmoM2;hGFAP-cre) generated only GFD tumors (8/8). Also see Supplementary Fig. S6 for histology of tumor derived from injecting SmoM2;hGFAP-cre tumorsphere cells. B, expression of SmoM2 in committed EGL progenitors (SmoM2;Atoh-creER) by tamoxifen treatment from p7–p10 generated only NG tumors (6/6). C and D, hematoxylin and eosin staining of SmoM2;hGFAP-cre and SmoM2;Atoh-creER tumors. E, SHH pathway gene expression in independent SmoM2;hGFAP-cre and SmoM2;AtohCreER tumor tissues. Relative fold change compared with wild-type p8 cerebellum.

Close modal

Gain of chromosome 6 is associated with the most aggressive subtype, GFI

Interestingly, neither the SmoM2;hGFAP-cre nor SmoM2;Atoh-CreER mice produced GFI tumors. As mentioned above, approximately 83% of Ptch tumors retained their phenotype through several generations of in vivo passaging; however, approximately 17% (8/47 tested) switched cellular phenotypes around G4 in vivo, always to the GFI subtype. We postulated that this represents tumor evolution in vivo.

To investigate whether a recurring genetic alteration is associated with GFI tumors, we performed high-resolution spectral karyotyping (SKY) analysis on low-passage tumorsphere cells from GFD, GFI, and GFD-to-GFI transformed tumors (Fig. 4A). Multiple genetic alterations were present in each sample, but the only recurring alteration associated with the GFI subtype was a gain of chromosome 6 (Fig. 4A and B). All GFI samples tested (n = 4) were composed of >80% trisomy 6 cells. In contrast, a GFD sample that remained GFD through five generations of in vivo transfer contained no trisomy 6 cells at G1. Interestingly, GFD tumors that progressed to GFI at G4 during in vivo passage (tumor numbers 4951 and 5333) contained a minor fraction (<30%) of trisomy 6 cells at G1. When these tumors became GFI at G4 in vivo (ICF918 and ICF937, respectively), trisomy 6 cells dominated the tumorsphere cultures (>80%; Fig. 4B). Together, these observations suggest that acquisition of the GFI phenotype is associated with the gain of chromosome 6.

Figure 4.

Trisomy 6 is associated with the GFI subtype. A, SKY analysis of GFD and GFI tumorsphere cells at low passage (<passage 3) shows gain of chromosome 6 in all GFI tumorspheres (n = 4). B, quantitation of trisomy 6 cells in GFD and GFI tumorspheres by SKY. When G1 GFD tumors 4951 and 5333 progressed to the GFI subtype at G4 in vivo (ICF918 and ICF937, respectively), trisomy 6 cells dominated the tumorsphere cultures. C, a representative image of FISH analyses on G1 tumor tissues with Cy-3 labeled chromosome 6 and Cy5-labeled chromosome 13 probes. D, quantitation of chromosome 6 counts in multiple independent NG, GFD, and GFI tumor tissues at G1.

Figure 4.

Trisomy 6 is associated with the GFI subtype. A, SKY analysis of GFD and GFI tumorsphere cells at low passage (<passage 3) shows gain of chromosome 6 in all GFI tumorspheres (n = 4). B, quantitation of trisomy 6 cells in GFD and GFI tumorspheres by SKY. When G1 GFD tumors 4951 and 5333 progressed to the GFI subtype at G4 in vivo (ICF918 and ICF937, respectively), trisomy 6 cells dominated the tumorsphere cultures. C, a representative image of FISH analyses on G1 tumor tissues with Cy-3 labeled chromosome 6 and Cy5-labeled chromosome 13 probes. D, quantitation of chromosome 6 counts in multiple independent NG, GFD, and GFI tumor tissues at G1.

Close modal

Interestingly, gain of chromosome 6 has been observed in a subset of tumors in other medulloblastoma mouse models (25–27). To further test whether gain of chromosome 6 is associated with the GFI subtype, we performed FISH on G1 tumor tissues (Fig. 4C and D). Consistent with the SKY results, NG and GFD tumors that retained their original phenotype throughout in vivo passaging contained <10% trisomy 6 nuclei (Fig. 4D), whereas all GFI tumors contained >30% trisomy nuclei at G1. Interestingly, NG and GFD tumors that progressed to GFI phenotype during in vivo passaging already contained approximately 30% trisomy 6 cells at G1 (Supplementary Fig. S4). For example, 5333 GFD tumor, which progressed to GFI at G4 and showed approximately 30% trisomy 6 cells in low-passage tumorsphere culture by SKY analysis (Fig. 4B), also showed approximately 30% trisomy 6 cells in the G1 tumor tissue (Supplementary Fig. S4). These observations suggest that gain of chromosome 6 precedes acquisition of the GFI phenotype and that trisomy 6 is indeed closely correlated with the GFI subtype. Together, our analyses of tumorsphere cultures, genomic alterations detected by SKY and FISH, and results from serial in vivo transfer all indicate that the GFI subtype arises from either NSC or NPCs as tumors evolve to become more aggressive tumors.

AKT and ERK pathways are differentially activated in the three tumor subtypes

We reasoned that if TICs that propagate these tumors are molecularly and genetically distinguishable, the bulk tumor cells might also be molecularly distinct. To test this idea, we performed an unbiased transcriptome analysis of multiple independent Ptch+/−;p53+/+ G1 tumors. Using one-way ANOVA analysis, we identified 465 genes that can distinguish the three tumor subtypes (Supplementary Table S1). A principal component analysis using these genes showed a clear segregation of the three tumor subtypes (Fig. 5A). Gene-set enrichment analysis (GSEA) of the 465 genes showed that genes on chromosome 6 were overrepresented at the top of a ranked list of differentially expressed genes between GFD and GFI tumors (Fig. 5B, P<0.001). To test whether these subtypes model different human medulloblastoma subgroups, we compared gene expression profiles of Ptch tumors to those of human medulloblastomas. All three subtypes best matched the SHH subgroup of human medulloblastoma (Fig. 5C and D), consistent with high levels of the SHH pathway gene expression at the bulk tumor level in all three subtypes.

Figure 5.

The three Ptch medulloblastoma subtypes are molecularly distinguishable at the bulk tumor level and model the human SHH subgroup of medulloblastomas. A, principal component analysis with 465 genes identified in the ANOVA analysis of G1 tumor tissues. B, a GSEA shows that genes on chromosome 6 are highly ranked among differentially expressed genes between GFI and GFD tumors (P < 0.001). C, AGDEX comparison of mouse NG, GFD, and GFI tumors to human medulloblastoma subgroups. Bars represent the cosine similarity measure and reflect the similarity of global expression profile between each mouse tumor subtype and each human medulloblastoma subgroup. D, medulloblastoma subgroup assignment by differential marker expression and PAM classification. Heatmaps depict the differential expression of subgroup-specific markers for the human training cohort (left) and the mouse tumors (right). Subgroup assignments are indicated above the heatmap. All mouse tumors are classified with ≥99% confidence probability.

Figure 5.

The three Ptch medulloblastoma subtypes are molecularly distinguishable at the bulk tumor level and model the human SHH subgroup of medulloblastomas. A, principal component analysis with 465 genes identified in the ANOVA analysis of G1 tumor tissues. B, a GSEA shows that genes on chromosome 6 are highly ranked among differentially expressed genes between GFI and GFD tumors (P < 0.001). C, AGDEX comparison of mouse NG, GFD, and GFI tumors to human medulloblastoma subgroups. Bars represent the cosine similarity measure and reflect the similarity of global expression profile between each mouse tumor subtype and each human medulloblastoma subgroup. D, medulloblastoma subgroup assignment by differential marker expression and PAM classification. Heatmaps depict the differential expression of subgroup-specific markers for the human training cohort (left) and the mouse tumors (right). Subgroup assignments are indicated above the heatmap. All mouse tumors are classified with ≥99% confidence probability.

Close modal

To determine whether the three tumor subtypes can be distinguished by different signaling pathway activities, we performed Ingenuity Pathway Analysis with the 465-gene list. The top three molecular networks included AKT and ERK1/2 as major nodes (Fig. 6A and Supplementary Fig. S5). Immunoblotting analyses with antibodies that recognize activated AKT (phospho-AKT Thr308) and activated ERK (phospho-ERK Thr202/204) showed significantly lower levels of AKT activation in GFD tumors and significantly higher levels of pERK activation in GFI tumors (Fig. 6B).

Figure 6.

The three medulloblastoma subtypes are distinguishable by different levels of ERK and AKT activation. A, Ingenuity Pathway Analysis showing the top three networks across each of the three pairwise comparisons (GFD vs. GFI, GFD vs. NG, and NG vs. GFI) of bulk tumor tissue gene expression using the 465-gene list. Shown here is GFD versus GFI comparison, where red colored nodes indicate relatively higher expression and green nodes indicate relatively lower expression in GFI compared with GFD. Also see Supplementary Fig. S5 for GFD-NG and NG-GFI comparisons. B, immunoblot analysis of AKT and ERK pathway activation from G1 tumor tissues. GFD>GFI, a transplanted tumor sample that evolved from the GFD subtype to GFI subtype in vivo. C, in vitro secondary sphere formation assay on low passage (<p5) GFD and GFI tumorsphere cells in the presence or absence of triciribine (100 nmol/L). Four independent tumorspheres lines of each subtype were tested.

Figure 6.

The three medulloblastoma subtypes are distinguishable by different levels of ERK and AKT activation. A, Ingenuity Pathway Analysis showing the top three networks across each of the three pairwise comparisons (GFD vs. GFI, GFD vs. NG, and NG vs. GFI) of bulk tumor tissue gene expression using the 465-gene list. Shown here is GFD versus GFI comparison, where red colored nodes indicate relatively higher expression and green nodes indicate relatively lower expression in GFI compared with GFD. Also see Supplementary Fig. S5 for GFD-NG and NG-GFI comparisons. B, immunoblot analysis of AKT and ERK pathway activation from G1 tumor tissues. GFD>GFI, a transplanted tumor sample that evolved from the GFD subtype to GFI subtype in vivo. C, in vitro secondary sphere formation assay on low passage (<p5) GFD and GFI tumorsphere cells in the presence or absence of triciribine (100 nmol/L). Four independent tumorspheres lines of each subtype were tested.

Close modal

To test whether activated AKT plays a critical role in promoting survival and proliferation of Ptch medulloblastoma TICs, we treated GFD and GFI tumorsphere cells with an AKT inhibitor, triciribine. As shown in Fig. 6C, GFI tumorsphere self-renewal was completely blocked by 100nmol/L triciribine treatment. In contrast, GFD tumorsphere cells were relatively insensitive to AKT inhibition and formed >50% secondary spheres compared with DMSO-treated control cultures, even at 10-fold higher concentration of the drug (Fig. 6C and not shown). Together, these data indicate that both tumorspheres and bulk tumor cells of the GFI and GFD subtypes show differential activation level and dependence on the AKT pathway.

The Ptch+/− mouse model of medulloblastoma has been used as a single tumor model to elucidate the molecular and cellular mechanisms involved in medulloblastoma formation and to test SHH pathway inhibitors. Our finding that there are three distinct tumor subtypes in this model system has broad implications. First, it may explain the heterogeneity within the SHH subgroup: human SHH tumors are found predominantly in either infants (age <3 years) or adults (age >18 years), and these subgroups show different molecular and clinical phenotypes (11). Activation of the SHH pathway resulted in tumors in either weaning age mice (Smo-M2;hGFAP-cre) or in adult mice (Smo-M2;Atoh-CreER) depending on the timing of SHH pathway activation and target cell type (Fig. 3). Hence, one could speculate that infantile and adult human SHH medulloblastomas arise from different cells of origin. Second, this study may provide explanations for conflicting observations in the field, such as tumorsphere formation ability of tumorspheres in Ptch tumors (4, 17). Third, it provides a potentially novel mechanism of therapy resistance to SHH inhibitors. Previously, Buonamici and colleagues analyzed LDE225-resistant Ptch tumors to elucidate the molecular mechanism underlying the emergence of therapy-resistant tumors to SHH inhibitor treatment (16). They observed that while many resistant tumors developed new mutations in the SHH pathway genes as expected, approximately 50% of the tumors did not contain any new mutations in the SHH pathway genes. On the basis of our findings, a possible explanation for this observation is that these tumors correspond to the GFD subtype. TICs in GFD subtype tumors do not depend on the SHH pathway and therefore, there would be no selective pressure for cells with mutations within the SHH pathway for resistant tumors to emerge.

Our study provides evidence that activation of the AKT pathway may be a significant cooperating oncogenic event in progenitor-derived tumors, but not stem cell–derived tumors. AKT activation is commonly observed in many human cancers, and others have shown that it plays an important role in conferring radiotherapy resistance to perivascular cancer stem cells in another mouse model of medulloblastoma (28). While the effect of AKT inhibition on different subtypes of Ptch medulloblastoma needs to be tested in vivo, exquisite sensitivity of GFI tumorspheres to the AKT inhibitor Triciribine in vitro suggests that a combination therapy targeting the SHH and AKT pathways may be effective in treating the GFI subtype.

GFI tumors are associated with gain of chromosome 6, although the presence of trisomy 6 cells before acquisition of the GFI phenotype suggests that gain of chromosome 6 is not sufficient to confer the GFI phenotype on its own. However, it seems to be an early step in progression to the GFI tumor subtype. Notably, gain of chromosome 6 is a common genetic alteration observed in other mouse models of medulloblastoma (25–27). It will be informative to test if tumors with trisomy 6 cells are more aggressive in these models and to determine whether the heterogeneity of TICs we identified is recapitulated in these tumor models. Mouse chromosome 6 contains many oncogenes, including many that are upstream of ERK activation, such as Kras, BRAF, and MET, and epigenetic regulators such as EZH2. In other cancer types, ERK activation has been shown to be associated with tumor progression, metastasis, and invasion (29). Elevated EZH2 expression is correlated with human medulloblastomas with chromosome 7 gain (subgroup 3 and 4), and it has been reported that inhibition of EZH2 suppresses tumorsphere formation (30, 31).

Elucidating the etiology and functional consequences of TIC heterogeneity in a given clinical tumor type is critical for developing therapies that efficiently target TICs. This study demonstrates that the cell-of-origin has a significant determining role in molecular and cellular phenotypes of TICs and that transformation of NSCs or NPCs by the same oncogenic hit can produce distinct TICs and tumor subtypes. Importantly, our data show that pathways that drive TIC proliferation in some tumors are distinct from those that drive the bulk tumor cells, providing direct proof-of-principle evidence that eradicating TICs and bulk tumors may require different targeted therapies. This study has wide implications that range from highlighting the power and inherent limitations of genetically engineered cancer models to elucidating a source of TIC heterogeneity that may invoke new strategies for identifying pathway-specific inhibitors as anticancer agents.

No potential conflicts of interest were disclosed.

Conception and design: M.H. Jenkins, K. Yun

Development of methodology: K.-H. Chow, M.H. Jenkins, K. Yun

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): K.-H. Chow, M.H. Jenkins, E.E. Miller, S. Choi, B.E. Low, B. Rybinski, R.T. Bronson, M.D. Taylor, K. Yun

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): K.-H. Chow, D.-M. Shin, M.H. Jenkins, E.E. Miller, D.J. Shih, S. Choi, B.E. Low, V. Philip, R.T. Bronson, K. Yun

Writing, review, and/or revision of the manuscript: K.-H. Chow, D.-M. Shin, E.E. Miller, S. Choi, B.E. Low, M.D. Taylor, K. Yun

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): K.-H. Chow, B.E. Low, K. Yun

Study supervision: K. Yun

Other (conducted pathologic analyses): R.T. Bronson

The authors thank Ellen Akeson for SKY analysis, Jesse Hammer for assistance with figures, Pat Cherry for administrative assistance, Carol Bult for guidance on the use of MGI tools, Jane Johnson for the MATH1 antibody, Lindsay Shopland for guidance on FISH analysis, and Susan Ackerman, Patsy Nishina, Barbara Knowles, and Rob Burgess for critical reading of the article.

This work was supported in part by the National Cancer Institute Center Support Grant P30CA034196 and a gift from Walter and Dorsey Cabot (to K. Yun).

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.

1.
Singh
SK
,
Hawkins
C
,
Clarke
ID
,
Squire
JA
,
Bayani
J
,
Hide
T
, et al
Identification of human brain tumour initiating cells
.
Nature
2004
;
432
:
396
401
.
2.
Harris
MA
,
Yang
H
,
Low
BE
,
Mukherje
J
,
Guha
A
,
Bronson
RT
, et al
Cancer stem cells are enriched in the side population cells in a mouse model of glioma
.
Cancer Res
2008
;
68
:
10051
9
.
3.
Visvader
JE
,
Lindeman
GJ
. 
Cancer stem cells: current status and evolving complexities
.
Cell Stem Cell
2012
;
10
:
717
28
.
4.
Read
TA
,
Fogarty
MP
,
Markant
SL
,
McLendon
RE
,
Wei
Z
,
Ellison
DW
, et al
Identification of CD15 as a marker for tumor-propagating cells in a mouse model of medulloblastoma
.
Cancer Cell
2009
;
15
:
135
47
.
5.
Frank
NY
,
Schatton
T
,
Frank
MH
. 
The therapeutic promise of the cancer stem cell concept
.
J Clin Invest
2010
;
120
:
41
50
.
6.
Saran
A
. 
Medulloblastoma: role of developmental pathways, DNA repair signaling, and other players
.
Curr Mol Med
2009
;
9
:
1046
57
.
7.
Roussel
MF
,
Hatten
ME
. 
Cerebellum development and medulloblastoma
.
Curr Top Dev Biol
2011
;
94
:
235
82
.
8.
Taylor
MD
,
Northcott
PA
,
Korshunov
A
,
Remke
M
,
Cho
YJ
,
Clifford
SC
, et al
Molecular subgroups of medulloblastoma: the current consensus
.
Acta Neuropathol
2012
;
123
:
465
72
.
9.
Lau
J
,
Schmidt
C
,
Markant
SL
,
Taylor
MD
,
Wechsler-Reya
RJ
,
Weiss
WA
. 
Matching mice to malignancy: molecular subgroups and models of medulloblastoma
.
Childs Nerv Syst
2012
;
28
:
521
32
.
10.
Northcott
PA
,
Shih
DJ
,
Peacock
J
,
Garzia
L
,
Morrissy
AS
,
Zichner
T
, et al
Subgroup-specific structural variation across 1,000 medulloblastoma genomes
.
Nature
2012
;
488
:
49
56
.
11.
Northcott
PA
,
Hielscher
T
,
Dubuc
A
,
Mack
S
,
Shih
D
,
Remke
M
, et al
Pediatric and adult sonic hedgehog medulloblastomas are clinically and molecularly distinct
.
Acta Neuropathol
2011
;
122
:
231
40
.
12.
Hahn
H
,
Wicking
C
,
Zaphiropoulous
PG
,
Gailani
MR
,
Shanley
S
,
Chidambaram
A
, et al
Mutations of the human homolog of Drosophila patched in the nevoid basal cell carcinoma syndrome
.
Cell
1996
;
85
:
841
51
.
13.
Goodrich
LV
,
Milenkovic
L
,
Higgins
KM
,
Scott
MP
. 
Altered neural cell fates and medulloblastoma in mouse patched mutants
.
Science
1997
;
277
:
1109
13
.
14.
Corcoran
RB
,
Scott
MP
. 
A mouse model for medulloblastoma and basal cell nevus syndrome
.
J Neuro Oncol
2001
;
53
:
307
18
.
15.
Romer
JT
,
Kimura
H
,
Magdaleno
S
,
Sasai
K
,
Fuller
C
,
Baines
H
, et al
Suppression of the Shh pathway using a small molecule inhibitor eliminates medulloblastoma in Ptc1(+/−)p53(−/−) mice
.
Cancer Cell
2004
;
6
:
229
40
.
16.
Buonamici
S
,
Williams
J
,
Morrissey
M
,
Wang
A
,
Guo
R
,
Vattay
A
, et al
Interfering with resistance to smoothened antagonists by inhibition of the PI3K pathway in medulloblastoma
.
Sci Transl Med
2010
;
2
:
51ra70
.
17.
Ward
RJ
,
Lee
L
,
Graham
K
,
Satkunendran
T
,
Yoshikawa
K
,
Ling
E
, et al
Multipotent CD15+ cancer stem cells in patched-1-deficient mouse medulloblastoma
.
Cancer Res
2009
;
69
:
4682
90
.
18.
Reynolds
BA
,
Weiss
S
. 
Generation of neurons and astrocytes from isolated cells of the adult mammalian central nervous system
.
Science
1992
;
255
:
1707
10
.
19.
Hu
Y
,
Smyth
GK
. 
ELDA: extreme limiting dilution analysis for comparing depleted and enriched populations in stem cell and other assays
.
J Immunol Methods
2009
;
347
:
70
8
.
20.
Subramanian
A
,
Tamayo
P
,
Mootha
VK
,
Mukherjee
S
,
Ebert
BL
,
Gillette
MA
, et al
Gene set enrichment analysis: a knowledge-based approach for interpreting genome-wide expression profiles
.
Proc Natl Acad Sci U S A
2005
;
102
:
15545
50
.
21.
Wetmore
C
,
Eberhart
DE
,
Curran
T
. 
Loss of p53 but not ARF accelerates medulloblastoma in mice heterozygous for patched
.
Cancer Res
2001
;
61
:
513
6
.
22.
Kawauchi
D
,
Robinson
G
,
Uziel
T
,
Gibson
P
,
Rehg
J
,
Gao
C
, et al
A mouse model of the most aggressive subgroup of human medulloblastoma
.
Cancer Cell
2012
;
21
:
168
80
.
23.
Mao
J
,
Ligon
KL
,
Rakhlin
EY
,
Thayer
SP
,
Bronson
RT
,
Rowitch
D
, et al
A novel somatic mouse model to survey tumorigenic potential applied to the Hedgehog pathway
.
Cancer Res
2006
;
66
:
10171
8
.
24.
Schuller
U
,
Heine
VM
,
Mao
J
,
Kho
AT
,
Dillon
AK
,
Han
YG
, et al
Acquisition of granule neuron precursor identity is a critical determinant of progenitor cell competence to form Shh-induced medulloblastoma
.
Cancer Cell
2008
;
14
:
123
34
.
25.
Yan
CT
,
Kaushal
D
,
Murphy
M
,
Zhang
Y
,
Datta
A
,
Chen
C
, et al
XRCC4 suppresses medulloblastomas with recurrent translocations in p53-deficient mice
.
Proc Natl Acad Sci U S A
2006
;
103
:
7378
83
.
26.
Ayrault
O
,
Zindy
F
,
Rehg
J
,
Sherr
CJ
,
Roussel
MF
. 
Two tumor suppressors, p27Kip1 and patched-1, collaborate to prevent medulloblastoma
.
Mol Cancer Res
2009
;
7
:
33
40
.
27.
Zindy
F
,
Uziel
T
,
Ayrault
O
,
Calabrese
C
,
Valentine
M
,
Rehg
JE
, et al
Genetic alterations in mouse medulloblastomas and generation of tumors de novo from primary cerebellar granule neuron precursors
.
Cancer Res
2007
;
67
:
2676
84
.
28.
Hambardzumyan
D
,
Becher
OJ
,
Rosenblum
MK
,
Pandolfi
PP
,
Manova-Todorova
K
,
Holland
EC
. 
PI3K pathway regulates survival of cancer stem cells residing in the perivascular niche following radiation in medulloblastoma in vivo
.
Genes Dev
2008
;
22
:
436
48
.
29.
Wagner
EF
,
Nebreda
AR
. 
Signal integration by JNK and p38 MAPK pathways in cancer development
.
Nat Rev Cancer
2009
;
9
:
537
49
.
30.
Robinson
G
,
Parker
M
,
Kranenburg
TA
,
Lu
C
,
Chen
X
,
Ding
L
, et al
Novel mutations target distinct subgroups of medulloblastoma
.
Nature
2012
;
488
:
43
8
.
31.
Alimova
I
,
Venkataraman
S
,
Harris
P
,
Marquez
VE
,
Northcott
PA
,
Dubuc
A
, et al
Targeting the enhancer of zeste homologue 2 in medulloblastoma
.
Int J Cancer
2012
;
131
:
1800
9
.