|Home | About | Journals | Submit | Contact Us | Français|
Origins of the brain tumor, medulloblastoma, from stem cells or restricted progenitor cells are unclear. To investigate this, we activated oncogenic Hedgehog (Hh) signaling in multipotent and lineage-restricted CNS progenitors. We observed that normal unipotent cerebellar granule neuron precursors (CGNP) derive from hGFAP+ and Olig2+ RL progenitors. Hh activation in a spectrum of early and late stage CNS progenitors generated similar medulloblastomas, but not other brain cancers, indicating that acquisition of CGNP identity is essential for tumorigenesis. We show in human and mouse medulloblastoma that cells expressing the glia-associated markers Gfap and Olig2 are neoplastic and that they retain features of embryonic-type granule lineage progenitors. Thus, oncogenic Hh signaling promotes medulloblastoma from lineage-restricted granule cell progenitors.
Cancer stem cells have been proposed as therapeutic targets in human brain cancer. Although such populations have been reported in medulloblastoma, tumor progenitor origins remain unclear. Here we show that medulloblastoma (MB) is generated from cells related to several stages of cerebellar granule neuron precursor (CGNP) development, and that some tumor cells, demonstrable in murine models and human MB, retain primitive features analogous to precursors of the embryonic brain. These data highlight the importance of a developmental perspective in understanding cells involved in CNS tumor progression and the histopathological composition of brain cancers. They suggest that a potential therapeutic target is the process regulating acquisition of CGNP phenotype.
It has been proposed that cancer can be viewed as aberrant organogenesis initiated by a multipotent or restricted progenitor cell that acquires infinite capacity for self-renewal through accumulated mutations (Reya et al., 2001). Principles of normal hematopoietic stem cell development (Barabe et al., 2007; Cobaleda et al., 2007) have provided insight into the mechanisms of self-renewal and multi-lineage differentiation in hematopoietic malignancies (Faber and Armstrong, 2007; Warner et al., 2004). However, for most organ systems parallels between normal stem cell development and solid tumor formation remain poorly understood.
The central nervous system (CNS), with its highly diversified classes of neuronal and glial progenitor cells, provides a suitable model to study such mechanisms. Several studies have documented the existence of cancer stem cells for glioma and the cerebellar tumor, medulloblastoma (MB) with properties reminiscent of normal neural precursors (Singh et al., 2004; Galli et al., 2004; Hemmati et al., 2003). Expression profiling has indicated correlations between specific stages of cerebellar development and MB (Lee et al., 2003; Kho et al., 2004). The Sonic hedgehog (Shh) pathway has critical functions in cerebellar development and mutations of pathway are etiologic in MB (Pietsch et al., 1997).
The cerebellum (CB) develops from several progenitor regions: the rostral rhombic lip (rRL) and cells of the ventricular zone (VZ) surrounding the IVth ventricle (Altmann and Bayer, 1997; Morales and Hatten, 2006; Kawauchi et al., 2006; Casper and McCarthy, 2006; Malatesta et al., 2003) (Figure 1A). The VZ of the cerebellar anlage generates Purkinje neurons (PN), interneurons (IN), and glia cells. Progenitors within the rRL express the radial glial markers, RC2 and Nestin, and give rise to all granule lineage cells (GC) of the EGL and IGL. We recently reported that cells which express cre recombinase under control of human regulatory sequences for Glial Fibrillary Acidic Protein (hGFAP), are most probably radial glia and generate most cerebellar cell types including CGNP (Spassky et al., 2008). Subsequently, CGNP migrate rostrally along the surface of the cerebellar anlagen to form the external granule layer (EGL) (Lee et al., 2005; Machold and Fishell, 2005; Wang et al., 2005) (Figure 1A).
PN (Figure 1B) produce Shh, which binds the transmembrane protein repressor Patched (Ptc), relieving inhibition of Smoothened (Smo) activity, which is essential for CGNP proliferation in the EGL (Wechsler-Reya and Scott, 1999; Dahmane and Altaba, 1999; Lewis et al., 2004). CGNP of the EGL gradually show reduced proliferation levels, migrate internally and differentiate into glutamatergic neurons in the internal granule layer (IGL) (Figure 1B), a process that is largely completed by postnatal day 15 (P15) in mice (Borghesani et al., 2002), and by about a year of age in the human CB (Abraham et al., 2001). Transcriptional control of CGNP development involves functions of bHLH protein Math1 (Ben Arie et al., 1997; Machold and Fishell, 2005; Wang et al., 2005), and homeodomain protein Pax6 (Engelkamp et al., 1999; Yamasaki et al., 2001).
Inherited activating mutations of the Shh-Smo pathway in humans are etiologic in MB but not glioma (Hahn et al., 1999; Johnson et al., 1996). This cancer spectrum is faithfully modeled in the CNS of mice heterozygous for a mutation of the Hedgehog (Hh) pathway suppressor Ptc1 (Johnson et al., 1996) or expressing an activated allele of Smoothened (SmoM2) (Mao et al., 2006), supporting the proposal that CGNP are particularly susceptible to transformation by oncogenic Hh signaling (Shih and Holland, 2004). Nevertheless, the cellular origins of MB from unipotent precursors in the EGL remain controversial. The identification of multipotent MB stem cells (Hemmati et al., 2003; Singh et al., 2004) capable of multi-lineage differentiation, in particular, has challenged the hypothesis of a unipotent cell-of-origin. To investigate these issues, we defined distinct progenitor populations for CGNP within the rRL as well as in the EGL and systematically assessed their tumorigenic potential with respect to Hh signaling.
In order to determine possible parallels between progenitors during cerebellar development and tumorigenesis, we first characterized relevant embryonic hindbrain VZ and RL populations. Nuclei of cells that express cre proteins in hGFAP-cre mice are observed along the VZ and the ventral aspect of the rRL of the midline cerebellar anlage at E14.5 (Figure 1C-E). Cre proteins in these mice mark nestin+ cells that have been proposed to be radial glia of the rRL that give rise to CGNP (Spassky et al., 2008), which show robust Pax6 expression. We next investigated expression of the bHLH protein, Olig2, which marks multilineage progenitor populations of the forebrain and the spinal cord (Petryniak et al., 2007; Lu et al., 2002; Zhou and Anderson, 2002). Olig2 expression identified a distinct cell population predominantly restricted to the rRL and extending to the boundary of the EGL (Figure 1F). The cerebellar VZ contained rare Olig2+ cells and expression of Olig2 was rapidly down regulated within the EGL.
To establish a possible lineage relationship between hGFAP-cre and Olig2+ cells, we fate mapped hGFAP-cre-expressing precursors with a conditional ROSA26-eYPFFl/Fl reporter (Srinivas et al., 2001). As shown (Figure 1G, white line), hGFAP-cre-expressing precursors give rise to the majority of CGNP of the primitive EGL, and some Olig2+ cells of the rRL (Figure 1H). Most Olig2+ cells in the rRL co-expressed the CGNP marker, Pax6 (Figure 1I) while Olig2 expression was gradually lost and Pax6 retained as CGNP migrated into the EGL. Olig2+Pax6+ cells could be identified within the rRL and EGL until E18.5, but were never observed at P7 or later stages (data not shown; Figure 6E,F).
To establish Olig2+ progenitor contributions during CGNP development, we constructed a multi-functional mouse line by inserting tva, an avian-specific retroviral receptor, and an IRES-cre recombinase cassette into the endogenous Olig2 locus by homologous recombination (Figure 2A). When Olig2-tva-cre+/- mice were crossed to ROSA26-eYFPFl/Fl conditional reporter mice, we found that 98% of YFP+ cells in the P7 cerebellar white matter (WM) expressed Olig2, consistent with oligodendroglial identity (Figure 2B, C, yellow arrows). At P21 most of these cells co-expressed myelin basic protein, and we observed small numbers of YFP+ PN and IN (Supp. Figure 1E, F). At P7, we detected a small subpopulation of YFP+ CGNP within the EGL, confined mainly to posterior cerebellar Lobes IX/X (Figure 2C-E, white arrows). At P21, we observed fate mapping to GC within the IGL (Figure 2F-I). Interestingly, only 3% of YFP+ cells expressed the GC marker Zic in the Lobe II IGL, but 58% of YFP+ cells were double positive in Lobe X (Figure 2G and I), indicating that Olig2 contributions to GC of the IGL are restricted mainly to the posterior lobes.
We found Olig2+ cells in the rRL from E14.5-18.5 (Figure 1 and data not shown). To test whether these cells in particular were a source of lobe IX-X CGNP, we injected the rRL of E15.5 Olig2-tva-cre+/- animals with an avian retrovirus (RCAS-GFP), which can only transduce tva-expressing cells (Figure 2J). Fate mapping results analyzed at P7 indicated that infected rRL Olig2-tva+ cells generate Pax6+ CGNP of the EGL and Zic+ GC within the IGL (Figure 2K-M, Suppl. Figure 1) of lobes IX and X. Taken together, these data indicate that Olig2+ progenitors of the rRL make restricted contributions to the GC lineage, particularly in posterior lobes of the CB.
To determine the significance of hGFAP- and Olig2-expressing progenitor populations for CNS tumorigenesis, we used hGFAP-cre and Olig2-tva-cre drivers, respectively, to conditionally express an activated Smo (SmoM2) allele fused with yellow fluorescent protein (YFP) sequences (Mao et al., 2006). All hGFAP-cre:SmoM2 mice developed diffuse tumors with the characteristic “small round blue cell” histology of MB (Figure 3A-D, Table 1). Olig2-tva-cre:SmoM2 mice developed tumors with similar histology, except that they were focal, localized to the posterior-lateral lobes and had significantly (p <0.01) later mortality (Figure 3E-H, Table 1), which was likely a result of their focal nature and the small numbers of CGNP targeted by Olig2-tva-cre.
Although hGFAP+ and Olig2+ progenitor populations give rise, collectively, to CGNP as well as IN, astrocytes (AS) and oligodendrocytes (OL), we only observed cerebellar tumors with histological features of MB. Additionally, activated Hh signaling in cerebellar PN driven by Shh-creERT2 never resulted in MB (Table 1). These data demonstrate that hGFAP+ and Olig2+ multipotent progenitor populations can produce MB.
Hh signaling can have mitogenic effects in spinal cord and forebrain (Mao et al., 2006; Rowitch et al., 1999; Fuccillo et al., 2006), and roles in forebrain glioma have been reported (Clement et al., 2007; Ehtesham et al., 2007). However, no evidence of gliomas or other primitive neuroectodermal tumors (PNET) in the forebrain was observed despite forced expression of SmoM2 in the subventricular zone, in NG2+ progenitors, and in oligodendrocyte precursor cells (OPC) of the forebrain (Table 1, Figure 3I-L). Together, these data indicate a hindbrain- as well as a CGNP/lineage-restricted oncogenic response to Hh signaling, and raised the question of whether this was determined by the particular environmental niche of the EGL. To assess this, we transplanted freshly dissociated cells from Olig2-tva-cre:SmoM2 tumors into the forebrains and cerebella of SCID mice. Because tumors formed equally well in both locations (Figure 3M-P, Table 1), suggesting tumor propagation is governed primarily by cell-intrinsic factors rather than environment.
In myeloid oncogenesis, multipotent cells rather than more restricted progenitor cells are thought to be the origin of leukemia stem cells (Faber and Armstrong, 2007). To test whether a restricted neuronal precursor can also give rise to MB, we activated SmoM2 in committed CGNP. Math1-cre drives expression within CGNP of the rRL and the entire EGL until P7-10, but not in PN, AS or IN (Machold and Fishell, 2005; Wang et al., 2005). In contrast, expression of the transcription factor Tlx3 (Hox11L2) is only seen in the EGL of Lobes VI-IX, but never in other lobes or the rRL (Suppl. Figure 2). Transgenic Math1-cre mice were generated using regulatory sequences 4 kb upstream of Math1 (Matei et al., 2005; Schüller et al., 2007) and Tlx3-cre mice were generated by gene targeting of Tlx3 (Xu et al., 2008). Fate mapping crosses involving a conditional ROSA26-eYFPFl/Fl reporter confirmed that Math1+ and Tlx3+ precursors contributed solely to the GC lineage (Figure 4A-H and data not shown).
When crossed with SmoM2Fl/Fl mice, all Math1-cre animals developed MBs that could be transplanted into the cerebella of SCID mice (Figure 4I, J, Table 1, also see (Yang et al., 2008)). Tlx3-cre:SmoM2 animals developed tumors in a posterior distribution in accord with the Tlx3 expression domain (Figure 4K, L). These data show that unipotent CGNP are capable of tumor formation when targeted with oncogenic SmoM2. The Shh transcriptional target, Gli1, is expressed in proliferating CGNP (Wechsler-Reya and Scott, 1999) and we could induce MB formation in Gli1-creERT2:SmoM2 mice until P14, but not at later postnatal stages.
To determine whether significant cellular (i.e. glial and neuronal diversity) and gene expression differences might exist in tumors derived from multipotent (e.g. hGFAP-cre) versus unipotent (e.g. Math1-cre) progenitor populations we performed Affymetrix gene expression profiling analyses, previously demonstrated to aid in distinguishing MB from other histologically similar primitive neuroectodermal tumors (Pomeroy et al., 2002). Comparison of expression profiles of the different murine tumors and Ptc+/- medulloblastoma using principal component (PC) analysis confirmed that all tumors had a gene expression signature of medulloblastomas and were highly related to each other (Figure 5A). Along the PC1 coordinate — the direction of greatest transcriptomic sample variance in cerebellar development —centroids of the MB tumor profiles were significantly different from matched normal samples (p < 10-6). However, PC1 coordinate centroids were not significantly different for tumors derived from early (i.e., hGFAP- or Olig2-cre) versus late/unipotent (i.e., Math1- or Tlx3-cre) GC lineage progenitors (Suppl. Figure 3). In fact, only a very small number of genes (~150 of 54,675 probes tested) were found to be uniquely and differentially expressed in each tumor sub-group. In line with this, extensive histological (Figures 3C, 3G, 4J, 4L) and immunohistochemical analysis (Figure 5B) failed to reveal any clear differences in the cellular composition of the tumors. Furthermore, hierarchical cluster analysis based on the 150 genes differentially expressed amongst the tumors did not show strict clustering of tumors based on cell of origin (Suppl. Figure 4). Together, these findings indicate that tumors derived from early (i.e., hGFAP-cre or Olig2-cre) or late/unipotent (i.e., Math1-cre or Tlx3-cre) GC lineage progenitors arrive at a common phenotypic endpoint.
All mouse tumors analyzed showed histological features in common with some human MBs including expression of CGNP markers (Math1, Zic, Pax6), Gfap, a marker usually found to be expressed in astrocytes, and Olig2, which is required for OL development (Lu et al., 2002) (Figure 5C). Clinical symptoms of tumor burden in Math1-cre:SmoM2 tumors developed at an average age of 41 days (Table 1), however we found evidence of hyperplasia in the EGL at P0 and more clearly at P7 (Figure 6A, C). Strikingly, high expression of Olig2 was also found in these regions (Fig 6B, D). At P7, when the EGL is clearly discerned, excessive numbers of Olig2+ cells were observed in the EGL but not RL of Math1-cre:SmoM2 mice (Figure 6D). Indeed, although Olig2+ cells that co-expressed Pax6 (Figure 1I) or a Math1 proxy marker (Figure 6E) could be observed at antenatal stages until E18, such co-labeled cells were not detectable at P7 (Figure 6F) or later stages. Thus, persistent Olig2 expression in the EGL of Math1-cre:SmoM2 mice represents an early molecular marker of a pre-neoplastic state.
We found that only about 2% of mitotically active phosphoH3+ (pH3) cells were Olig2+ in fully developed Math1-cre:SmoM2 tumors (Figure 6G). Indeed, the vast majority of proliferating cells were Cre+ (Figure 6H), a proxy for Math1 expression, indicating that most cells contributing to MB growth are CGNP-like. In human tumors it is controversial whether cells expressing GFAP (or OLIG2) comprise tumor, or alternatively, represent intermingled cells from the normal brain. This is a critical point given that glial differentiation within MB would provide further evidence for a multipotent stem cell-like progenitor. To establish the nature of cells comprising MB, we dissociated Math1-cre:SmoM2 tumors and found that the SmoM2-YFP fusion protein, a marker of cells bearing the tumorigenic mutation, co-localized in cells expressing either Pax6, NeuN, Gfap or Olig2. Quantification revealed that YFP proteins co-localized with most if not all Pax6+, Olig2+ and Gfap+ (100%, 90%, 96%, respectively) cells in acutely dissociated tumors (Figure 6I-L). In addition, we identified Olig2 and Gfap+ cells that co-expressed the YFP reporter in vivo (Figure 6M, N). Thus, Olig2- and Gfap+ cells are tumor cells rather than infiltrated/normal glial cells.
The Gfap+ and Olig2+ cells might represent simple astroglial and oligodendroglial tumor components. However, examination of Gfap+ and Olig2+ cells of the cerebellar anlagen (Figure 1) suggested that such tumor cells could, alternatively, retain features of primitive CGNP. In support of the latter possibility, we found that 81% of Olig2 cells expressed Pax6 and that 100% of Gfap+ cells expressed Pax6 in acutely disassociated Math1-cre:SmoM2 tumors in vitro (Figure 6O, P). These findings indicate that cells with an immature/progenitor phenotype are present in Hh-activated MB.
In order to determine the significance our findings might have for human MB, we examined OLIG2+ cell populations within a cohort of human MBs. Although our original findings in mice were all in the setting of Shh/Smo pathway activation, we found that OLIG2+ cells were present in both classic (40%) as well as desmoplastic (Shh-related) (77%) subtypes of human MBs (Fig 7A-D; Suppl. Table 1), and comprised less than 5% of cells in all tumors. Fluorescent in situ hybridization (FISH) analysis showed that OLIG2+ cells in human MB contained copy number gain at the C-MYC locus, confirming them as tumor cells and arguing against the possibility that they might represent intermingled normal cells (Figure 7E, F). As further support for OLIG2+ cells being tumor cells we identified co-expression of OLIG2 and the terminal neuronal differentiation marker, NEUN, in rare cells within human tumors (Figure 7G), and we found rare OLIG2+ cells that expressed the proliferation marker, Ki67 (Figure 7H). Together, these data suggest the possibility that some OLIG2+ cells are tumor cells with immature features and may represent a type of MB progenitor cell.
To mechanistically assess the lineage differentiation potential of Olig2+ cells in the context of MB, we acutely disassociated tumors from Olig2-tva-cre:SmoM2 mice and infected these cultures with RCAS-GFP viruses, to perform fate mapping of tumor stem/progenitor cells (Figure 7I). In our experimental paradigm, cells must be Olig2-tva+ for infection with RCAS-GFP to occur. After six days in culture, we found that cells derived from Olig2+ progenitors expressed Pax6, Zic or NeuN (Figure 7K-N). Significantly, of the 561 GFP+ cells counted, only 68 (12%) expressed Olig2 on Day 6 (Figure 7J, N), indicating differentiation along GC lineage axis with down-regulation of Olig2. Upon acute harvest and disassociation of tumors (Day 0), we found that only 8% expressed Olig2. Of these, 66% co-expressed Pax6 and 12% expressed NeuN (Figure 7O-Q). Contrasting this, on Day 6 in the tva fate map experiment, 52% of Olig2-derived progeny expressed NeuN. These data are consistent with the proposal that Olig2+Pax6+ cells lose Olig2 expression and progress to express the more mature markers such as Zic and NeuN during the 6 days in culture. We conclude that Olig2+ cells in Olig2-tva-cre:SmoM2 tumors have the potential to produce GC-like progeny therefore supporting the possibility that they represent a subset of MB progenitor cells.
Here we show that unipotent CGNP derive from multilineage embryonic CNS progenitors and that both populations are capable of generating MB when targeted with oncogenic SmoM2. Despite diverse origins, resulting tumors were similar leading to the conclusion that acquisition of CGNP identity is a critical determinant of progenitor cell competence to form Hedgehog-induced MB. These findings provide insight into MB origins with implications for development of targeted cellular therapies.
By creating a knock-in transgenic mouse line, Olig2-tva-cre, we targeted Olig2+ cells using two fate map approaches. Olig2-tva-cre intercrossed with conditional reporters (e.g., ROSA26-YFP) mapped a subset of GC in the EGL and IGL of posterior CB. We then confirmed that these cells are derived from Olig2+ progenitors in the rRL by in utero infection of Olig2-tva-cre mice with an avian RCAS retrovirus reporter. Although we cannot exclude that Olig2+ cells outside of rRL could contribute to CGNP, this possibility seems unlikely given previous work that has established the rRL as the source of the GC lineage. These data indicate that rRL Olig2+ progenitors give rise to a subset of GC (Figure 8A), which might contribute to functional heterogeneity (Zong et al., 2005).
Human inherited cancer syndromes show tumor formation in only a subset of cells within specific lineages despite germline transmission of tumor promoting mutations (e.g., p53, pRB-loss-of-function). Although several reports indicate that the Hh pathway is active in low- and high-grade gliomas (Ehtesham et al., 2007; Clement et al., 2007), in human Goltz-Gorlin syndrome, inherited heterozygous mutation of PTCH or SUFU lead to MB, but not gliomas. We failed to detect gliomas despite robust SmoM2 activation in forebrain progenitors, including developing neural stem cells and persistent progenitors of the SVZ and cortex expressing GFAP (Doetsch et al., 1999), NG2/Olig2 (Ligon et al., 2006) and Gli1 (Machold et al., 2003; Ahn and Joyner, 2005). Further work is needed to establish whether additional mutations (e.g., loss of p53) in concert with Hh activation might be necessary to enable glioma formation. Another possibility is that Shh activation biases towards neuronal lineage versus glial differentiation and therefore a MB phenotype and not glial lineages or glial tumor phenotype. Although we cannot rule this out, we think this explanation is unlikely for the following reasons. First, assuming that Shh promotes neuronal tumorigenesis, then one might expect to see development of such tumors at other CNS sites (e.g. supratentorial PNETs); this is not observed in human Goltz-Gorlin syndrome patients or in the mouse models we have studied. Second, we did not see other neuronal tumor types in the CB despite activation of Shh in PN and IN (Table 1). Third, we have recently shown that SmoM2 activation in the dentate gyrus causes hypertrophy but not cancer (Han et al., 2008).
To identify progenitors competent for MB formation, we activated SmoM2 in a spectrum of stage- and position-specific progenitors during CGNP development. Both multipotent (e.g., hGFAP+ and Olig2+) progenitor populations within the cerebellar anlage as well as unipotent/restricted CGNP readily gave rise to MB with 100% penetrance. Despite these diverse origins, all SmoM2-activated MBs we analyzed exhibited a strikingly similar tumor phenotype at histological, immunohistochemical and expression profiling levels. The common final tumor phenotype is consistent with (1) a common cell or origin for all tumors, or (2) initial origins from any stem/progenitor which later converge on an amplifying CGNP-like tumor cell that dominates the ultimate phenotype (see Figure 8C). In either case, these findings support the conclusion that acquisition of CGNP identity is important for Hh-induced MB formation. Because orthotopically transplanted Olig2-tva-cre:SmoM2 tumors were able to grow within the forebrain and CB, restriction of tumor susceptibility is evidently governed by cell-intrinsic factors. Further work is needed to identify the underlying molecular determinants of CGNP “tumor competence” that lead to MB formation.
Human desmoplastic and classic MB have been suggested to arise from CGNP EGL (Schüller et al., 2005; Bühren et al., 2000) or progenitors of the VZ (Salsano et al., 2007; Katsetos et al., 1995), respectively. Our studies demonstrate that Hh-induced MB can be initiated by precise restriction of activating mutations to the EGL using Tlx3-cre:SmoM2 and postnatally-induced Gli1-creERT2:SmoM2 without activation of SmoM2 in the less differentiated cell populations of the rRL or VZ. Furthermore, while the time of tumor onset and the degree of tumor distribution varied depending on the cre drivers used (Table 1, Figure 8B), these variations appear to be based on the number/extent/ location of the total CGNP population targeted (i.e., broad diffuse versus small lateral populations). In addition, tumor masses never developed in the VZ or rRL proper, suggesting that further work will be needed to determine whether additional oncogenic pathways (e.g. MYC, Wnt) might lead to MB formation in these regions.
Our data highlight that both mouse and human MBs exhibit extensive cellular heterogeneity. GFAP+ cells are commonly identified in human MB (Schüller et al., 2004) and the morphology of those cells ranges from small tumor cells with a relatively large nucleus and a sparse, ring-like cytoplasm to cells with a rather differentiated phenotype and branched processes. We observed that OLIG2+ cells, present in human MB, also varied in size and morphology. Using MYC FISH in human MB and YFP tracer expression in murine MB we demonstrate that OLIG2+ and GFAP+ cells are bona fide tumor cells and not just infiltrating cells arising from the normal surrounding brain. In addition we found small numbers of proliferating Olig2+ and Gfap+ cells within mouse and to a lesser degree human MB as further evidence that these heterogeneous elements represent tumor cells.
In human gliomas OLIG2 is expressed in tumor stem/progenitor cells and maintains the proliferation competence of both normal and tumorigenic progenitors of the murine forebrain (Ligon et al., 2007). However, given our data that OLIG2 is only rarely expressed in proliferating cells and is only present in 40% of classic and 77% of desmoplastic human MB cases, it seems clear that MB tumor growth is largely driven by rapidly dividing progenitor cells with CGNP characteristics, rather than OLIG2+ cells. Might GFAP+ and OLIG2+ cells in MBs have other roles as relatively quiescent tumor progenitor cells? Our in vitro tumor fate mapping demonstrates that Hh-transformed Olig2+ cells from murine MB can produce GC lineage-like progeny with advanced differentiation measured by NeuN labeling. Whether Olig2+ tumor cells from other mouse models (i.e., non-Olig2-cre derived) have this capability remains to be determined.
Our data suggest that GFAP and OLIG2 expression in MB is not associated with “differentiated” astrocytes and oligodendrocytes. While almost all Gfap+ and Olig2+ cells in Hh-activated murine MB co-expressed Pax6 and other CGNP markers, the “biphenotypic” (e.g., OLIG2-NEUN) staining pattern is unlikely to be a specific consequence of Hh signaling, because we also observed this phenomenon in classic MBs, which are not associated with Hh mutations (Pietsch et al., 1997). We suggest that GFAP and OLIG2 label cells with a phenotype reminiscent of radial glia and less differentiated CGNP progenitors of the rRL (Figure 8A).
MB was one of the first solid tumors for which “cancer stem cells” were reported (Hemmati et al., 2003). Cancer stem cells are classically described to be a tumor subpopulation with (1) unlimited capacity for self-renewal and the ability to regenerate tumor upon transplantation, and (2) the potential to produce multilineage progeny similar to distinct cell types of a cognate organ (Reya et al., 2001). The question of authentic glial cell production has critical implications for understanding MB progenitors and their characterization as multipotent cancer stem cells. However, evidence for multilineage differentiation of MB stem cells has been based on the expression of relatively few neuronal and glial cell markers with uncertain relation to either differentiated cell types of the CB or CNS multipotent stem cells. For example, the commonly used marker, GFAP, marks Type B stem cells of the subventricular zone (Doetsch et al., 1999). We show that Gfap and Olig2 mark cells in MBs reminiscent of those found during early CGNP development, but not differentiated glia. These findings argue against existence of MB progenitors with multilineage (neuronal and glial) contributions to the tumor mass. However, they are consistent with the possibility that MB could originate from either a multipotent or unipotent progenitor (Figure 8C).
We used lineage tracing to clarify the possible relationship between cells found during GC lineage development and those that are tumor-competent to produce MB. Our findings highlight distinctions between “cell-of-origin” (that acquires the oncogenic mutation) and “tumor-propagating” cell, responsible for major growth and phenotypic characteristics of the ultimate tumor. For instance, if unipotent cell identity (“Type C2”) is necessary for tumor growth (Figure 8C), then a tumor arising from mutation of a Type A or Type C2 cell would appear identical notwithstanding their different cells-of-origin. This also raises the possibility of a “de-differentiation” pathway to produce tumors with mixed multipotent and unipotent features, as has been suggested for certain leukemia stem cells (Krivtsov et al., 2006). Although our data do not provide direct evidence for de-differentiation, tumors from postnatally-induced Gli1-creERT2:SmoM2 (J. Mao, A. McMahon and K. Ligon, unpublished observations) and Math1-creERT2:PtcFl/Fl animals (Z. J. Yang and R. J. Wechsler-Reya, unpublished observations) contain many Olig2+ cells, consistent with the possibility that late-stage CGNP can adopt less-differentiated characteristics (Figure 8). Further investigation is needed to determine whether dedifferentiation of mature granule cells (D2 >C2) constitutes a biologically feasible or clinically relevant pathway of tumor formation.
For fate mapping, various cre-strains (Table 1) were intercrossed with ROSA26-eYFPFl/Fl or CAG-CAT-eGFPFl/Fl conditional reporter mice. At least three cerebella were harvested for each different time point fixed overnight in 4% paraformaldehyde/PBS. For tumor generation, cre-strains were intercrossed with SmoM2Fl/Fl mice and followed for survival and subsequent Kaplan-Meier analysis. For injections into forebrains or cerebella of SCID mice, tumor cells were freshly dissociated and resuspended in 1 μl PBS at a concentration of 105/μl (see Suppl. Methods). See Supplemental Methods for detailed information on generation, acquisition and genotyping of transgenic mouse strains and procedures including RCAS infection of Olig2-tva-cre embryonic hindbrain. All animal procedures were approved by the Institutional Animal Care and Use Committees at DFCI and UCSF.
Human tissue samples were obtained in accordance with the rules and regulations for human tissue collection; use in this study was approved by the Institutional Review Boards of the Brigham and Women’s Hospital/Partners Healthcare and Children’s Hospital Boston. Tissue samples were processed using standard techniques. Detailed protocols for in situ hybridization and immunohistochemistry, immuno-FISH as well as all antibodies used in this study are listed in Supplemental Methods and are available on request. Photomicrographs were taken with Zeiss Axioskop/Axiocam imaging or Nikon E600/SPOT imaging systems. For whole mount pictures of mouse MBs, fluorescent and bright-field images were overlaid using Photoshop software. A Leica TCS SL system was used for confocal microscopy.
Cultures from Math1-cre:SmoM2 tumors were established by triturating tumor tissue in DMEM medium and plating the cells on poly-D-Lysine coated dishes. Acutely dissociated Math1-cre, SmoM2 tumor cells were fixed 12 hours after plating. Long-term cultures were analyzed after 4 weeks in culture and 3 passages. Immunostaining was performed as described above. Cells from Olig2-tva-cre:SmoM2 tumors were infected with virus carrying RCAS(A)-GFP (2×107 cfu/ml; gift of C. Cepko, Harvard Medical School) after 24 hr culture in serum-free medium. Serum-containing medium was then added; after a total culture time of 6 days, cells were fixed and stained.
For in vivo fate mapping in Olig2-tva-cre mice, counts were derived from analysis of 3 independent animals, and at least 10 20X fields of labeled sections of CB focused around Lobe X. For cell culture counts, cells were derived from pups of at least four separate litters. Each litter yielded material for two six-well plates of cerebellar and one six-well plate of ventricular cultures. Three wells of a six-well plate were used for each treatment (EGF, Shh, vehicle). At least 100 cells were assessed in each sample to determine the fraction of Olig2+, Zic+ or Gfap+ cells. Statistical analysis was performed using the unpaired Student’s t-test with a two-tailed p-value.
The transcriptome of 21 cre-activated tumors and control cerebellar tissues were profiled using the Affymetrix GeneChip Mouse Genome 430 2.0 Plus [http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE11859]. Transcriptomes of the developing mouse CB P1-60 were profiled using the Affymetrix Murine Mu11K set as described previously (Kho et al., 2004) and principal component analysis (PCA) was used to determine the global similarity and variation between the transcriptome profiles of the different tissues (Johnson and Wichern, 2002). For detailed description, see supplementary methods.
The authors are grateful to Rob Wechsler-Reya for sharing unpublished data, to Charles D. Stiles for comments and to Jason Ling, Sovann Kaing, Dong-in Yuk, Natalie Vena and Emily Learner for expert technical assistance. U.S. was supported by a fellowship from the Dr. Mildred-Scheel-Stiftung für Krebsforschung and by the Max-Eder-Nachwuchsgruppenprogramm of the Deutsche Krebshilfe. V.M.H. thanks the Netherlands Organization for Scientific Research (NWO) for a TALENT-stipend. J.M. was supported by a grant from the Charles King Trust Medical Foundation. Y-G. H., A.K.D. and E.H. thank the American Brain Tumor Foundation for support. This work was supported by grants from the NIH to K.L.L. (NS047213), A.K. (NS040828). A.P.M. (NS033642) and D.H.R. (NS047527), the James S. McDonnell Research Foundation (to D.H.R.), the March of Dimes Foundation (to D.H.R.), and a grant from the Pediatric Brain Tumor Foundation of the United States (to A.A-B. and D.H.R.). D.H.R is a HHMI Investigator. All authors declare that they have no competing financial interests.
Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final citable form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.