Sonic hedgehog (SHH) medulloblastoma originates from the cerebellar granule neuron progenitor (CGNP) lineage, which depends on Hedgehog signaling for its perinatal expansion. Whereas SHH tumors exhibit overall deregulation of this pathway, they also show patient age-specific aberrations. To investigate whether the developmental stage of the CGNP can account for these age-specific lesions, we analyzed developing murine CGNP transcriptomes and observed highly dynamic gene expression as a function of age. Cross-species comparison with human SHH medulloblastoma showed partial maintenance of these expression patterns, and highlighted low primary cilium expression as hallmark of infant medulloblastoma and early embryonic CGNPs. This coincided with reduced responsiveness to upstream SHH pathway component Smoothened, whereas sensitivity to downstream components SUFU and GLI family proteins was retained. Together, these findings can explain the preference for SUFU mutations in infant medulloblastoma and suggest that drugs targeting the downstream SHH pathway will be most appropriate for infant patients.
Medulloblastoma is a malignant tumor of the cerebellum that frequently affects children. It consists of four main transcriptional subgroups that can be further subdivided upon additional molecular profiling (Cavalli et al., 2017; Hovestadt et al., 2019; Northcott et al., 2012a, 2017; Vladoiu et al., 2019). The Sonic hedgehog (SHH) subgroup of medulloblastoma, which accounts for 30% of all medulloblastoma cases, is believed to originate from the cerebellar granule neuron progenitor (CGNP) population that critically depends on SHH pathway signaling for its perinatal expansion (Dahmane and Ruiz, 1999; Hovestadt et al., 2019; Salsano et al., 2004; Vladoiu et al., 2019; Wallace, 1999; Wechsler-Reya and Scott, 1999; Yang et al., 2008; Yokota et al., 1996). SHH is a secreted morphogen that controls development and patterning in many organs, including the central nervous system (Fuccillo et al., 2006; Jiang and Hui, 2008). The pathway becomes active when SHH binds to its receptor Patched 1 (PTCH1), which relieves inhibition of Smoothened (SMO) and induces MYCN (Fuccillo et al., 2006; Jiang and Hui, 2008; Kenney, 2003; Knoepfler et al., 2002). Activated SMO inhibits the tumor suppressor protein SUFU and stimulates processing of the GLI factors into transcriptional activators. In the absence of the SHH morphogen, transcription of SHH target genes is actively repressed by SUFU-mediated formation of GLI repressors or direct transcriptional repression (Cheng and Bishop, 2002; Monnier et al., 1998; Pearse et al., 1999).
In line with the above, SHH medulloblastoma is characterized by an overall deregulation of SHH signaling that is often accompanied by mutually exclusive mutations in SHH pathway components, underlining the importance of this pathway in driving tumorigenesis (Kool et al., 2014; Northcott et al., 2012a). Interestingly, there is also patient age-related heterogeneity within this subgroup, suggesting that developmental factors affect tumor biology (Cavalli et al., 2017; Kool et al., 2014; Northcott et al., 2011, 2012b, 2017; Rausch et al., 2012; Wefers et al., 2014). For instance, infant and adult medulloblastoma display distinct gene expression patterns, as well as differences in copy number alterations and tumor localization. Strikingly, mutations in SHH pathway genes are also correlated with patient age (Kool et al., 2014). Whereas PTCH1 mutations occur across all age groups, SUFU mutations are almost exclusively found in infant patients, GLI2, NMYC and TP53 mutations in older children, and SMO mutations in adults.
These latter findings raise the question of why the pathway is differentially perturbed depending on patient age, as these mutations would presumably have identical outcomes (i.e. enhanced activation of SHH target genes). One explanation is that the tumor cells-of-origin undergo changes in sensitivity to, or usage of, the SHH signaling pathway during development, which would then provoke age-specific oncogenic lesions in the pathway (Jessa et al., 2019). The presumed cell-of-origin for SHH medulloblastoma is part of the CGNP lineage, a highly dynamic cell population that is present over 3 weeks of mouse and 2 years of human development (Carter et al., 2018), with the precise moment of transformation remaining under debate (Gibson et al., 2010; Haldipur et al., 2012; Leto et al., 2016; Lewis et al., 2004; MacHold and Fishell, 2005; Miale and Sidman, 1961; Schüller et al., 2008; Selvadurai et al., 2020; Wang et al., 2005; Wechsler-Reya and Scott, 1999; Yang et al., 2008). In mice, future CGNPs become specified in the upper rhombic lip (uRL) of the hindbrain around embryonic day (E)13.5, from where they migrate across the surface of the cerebellar primordium (Leto et al., 2016; MacHold and Fishell, 2005; Miale and Sidman, 1961; Wang et al., 2005). Here, they form a secondary germinal zone termed external granular layer (EGL), with a peak in proliferation occurring around birth, which is driven by Purkinje neuron-secreted SHH. Once they reach maturity, terminally differentiating granule neurons cease proliferation and migrate inwards until they reach their final destination in the internal granular layer (IGL) of the cerebellum (Dahmane and Ruiz, 1999; Leto et al., 2016; Lewis et al., 2004; Wallace, 1999; Wechsler-Reya and Scott, 1999). Besides the longer gestational period, the development of human CGNPs generally resembles that of the mouse (Haldipur et al., 2012).
To study the potential impact of the CGNP developmental age on medulloblastoma outcome, we have taken advantage of a transgenic mouse model that allows the prospective isolation of the entire CGNP cell lineage during embryonic and postnatal development (Hatten and Roussel, 2011; Machold et al., 2011). As reported before, we confirmed that CGNPs exhibit dynamic changes in gene expression in time, highlighting the identity changes of the CGNP population as neural development progresses (Hovestadt et al., 2019; Machold et al., 2011; Vladoiu et al., 2019). We also confirmed that CGNPs resemble human SHH medulloblastoma samples and, importantly, that early embryonic CGNPs co-segregate with the youngest patients, corroborating a linear relationship between cell-of-origin and patient age. In particular, we found that primary cilium expression was low across young CGNPs and young medulloblastoma patients, which was somewhat unexpected given the importance of primary cilia for SHH pathway activity (Bangs et al., 2015; Chang et al., 2019; Corbit et al., 2005; Haycraft et al., 2005; Huangfu and Anderson, 2005; Huangfu et al., 2003; May et al., 2005; Ong et al., 2020; Rohatgi et al., 2007; Sasai and Briscoe, 2012). In line, early embryonic CGNPs displayed a partial unresponsiveness to SMO-mediated (i.e. upstream) pathway stimulation, preventing increased proliferation. They were, however, sensitive to inhibition of Sufu or GLI proteins, both downstream SHH components, which in contrast to SMO, are known to have cilium-independent functions (Chen et al., 2009; Jia et al., 2009). These observations might have clinical implications, as patients with early developmental medulloblastoma are likely to benefit from drugs targeting the downstream primary cilium-independent part of the SHH pathway (Di Magno et al., 2015).
Developing CGNPs undergo dynamic changes in the expression of genes implicated in medulloblastoma
To address whether intrinsic changes in the developing medulloblastoma cell-of-origin could contribute to the age-dependent mutations in SHH pathway components found in SHH medulloblastoma, a transgenic mouse model was employed that allows lineage tracing and prospective isolation of the developing murine CGNP cell lineage from their specification in the cerebellar primordium onwards (Fig. 1A) (Feil et al., 1997; Machold et al., 2011; Machold and Fishell, 2005). This transgenic model consisted of the previously published Math1CreERT2 transgenic strain (Machold and Fishell, 2005) and a tdTomato reporter transgenic strain Ai14 (Madisen et al., 2010). Accordingly, we observed that treatment of pregnant transgenic mice with a single dose of tamoxifen at E13.5 induced acute and stable labeling of CGNPs with tdTomato in the offspring (Fig. 1B). We dissected fluorescent cerebella at E15.5, E17.5, P0 (day of birth), postnatal day (P)7, P14 and P30, purified tdTomato-expressing CGNPs and generated transcriptomes (Fig. 1A,C–E; Table S1). Subsequent principal component analysis revealed that more than 70% of all variation in gene expression between samples could be explained by developmental age, demonstrating that CGNPs indeed undergo significant changes in gene expression as a function of time (Fig. 1C).
To further explore these dynamic changes, we performed unsupervised hierarchical clustering analysis to identify genes that were differentially expressed (DE) between at least two time points (Fig. 1D; Table S2). We next divided the DE genes into five major gene clusters according to the clustering tree. The yellow cluster contained genes that were highest expressed at embryonic time points (E15.5 and E17.5, n=471 genes), the orange cluster genes that were high during embryonic and early postnatal time points (E15.5–P7, n=1541 genes), genes in the red cluster were highest expressed at early postnatal time points (P0 and P7, n=619 genes), and light and dark blue clusters contained the genes that were induced upon granule neuron maturation (P14 and P30, n=763 and 943 genes, respectively) (Fig. 1D,E). To identify enriched biological processes (Gene Ontology) within these gene clusters, we used the Database for Annotation, Visualization and Integrated Discovery (DAVID), and visualized the data using Cytoscape (Fig. 2; Fig. S1, Table S3) (Zomerman et al., 2018). At the early stages of CGNP development (yellow cluster), there is a strong enrichment of biological processes associated with early neural development such as axon and dendrite formation, which consist of genes involved in the attraction and repulsion of migrating progenitor cells, as well as developmental transcription factor-driven gene expression and glycolysis (Fig. 2; Fig. S1). In comparison, processes enriched throughout embryonic and neonatal stages (orange cluster) are biased towards cell proliferation and mitosis. This surge in proliferation might be accompanied by increased DNA damage, as also processes related to the DNA damage response (DDR) are frequent in this cluster. The end of cerebellar proliferation is heralded by the appearance of cell cycle arrest processes, which are specific to early neonatal CGNPs (red cluster). Of note, the vast majority of orange cluster genes show the highest relative gene expression levels at P0, suggesting that proliferation of CGNPs born around E13.5 peaks around birth and declines at P7 (Fig. 1D). Two orange cluster processes (GO terms covalent chromatin modification and ATP-dependent chromatin remodeling) stand out, as genes involved in chromatin regulation like Arid2, are known to be mutated in medulloblastoma (Table S3) (Roussel and Stripay, 2018; Zhu et al., 2016). During the final stages of CGNP differentiation (light and dark blue clusters), processes related to neuronal connectivity, differentiation and regeneration are enriched, as well as major signal transduction routes like Bmp, TGF-β and phosphoinositide 3-kinase (PI3K) signaling, which are also known to play a role in medulloblastoma (Angley et al., 2003; Grimmer and Weiss, 2008; Northcott et al., 2012b).
To further explore a potential link between temporal CGNP gene expression and medulloblastoma, we extracted the individual gene expression profiles of n=40 genes commonly mutated in medulloblastoma and grouped them as suggested by Northcott et al. (Fig. 3A; Fig. S2; Northcott et al., 2017). We found striking peaks in the expression of the genes involved in cell cycle and genome maintenance at P0, and genes involved in chromatin and transcription regulation at P7, indicating that the age of the cell-of-origin could be related to mutations found in medulloblastoma. Interestingly, we observed the highest induction of TP53 expression at birth, which is most frequently mutated in children over the age of three (Kool et al., 2014).
Age-specific CGNP gene expression is preserved in human medulloblastoma
To investigate whether overall CGNP gene expression is reflected in human SHH medulloblastoma, we performed an independent cross-species comparison between an existing cohort of human SHH medulloblastoma and our murine CGNP samples (Fig. 3B) (Kool et al., 2014). For this, we selected the human genes that were differentially expressed between SHH medulloblastoma patients of three age groups, i.e. 0–3 years of age (infants/toddlers), 4–11 years of age (children), and 12 years and older (older children/adults). In agreement with earlier publications, we confirmed that infant and adult medulloblastomas have distinct gene expression patterns, with childhood medulloblastoma forming an intermediate group (Kool et al., 2014; Northcott et al., 2011). Interestingly, unsupervised hierarchical clustering analysis of these human genes together with their murine counterparts showed that embryonic mouse CGNPs co-cluster with the infant medulloblastomas, and older CGNPs with tumors from older patients (Fig. 3B). This demonstrates that CGNPs resemble human medulloblastoma, and that the age of the patient is partially reflected by the age of the medulloblastoma cell-of-origin.
Cell cycle regulation and primary cilium biogenesis are age-related processes in CGNPs and medulloblastoma
We next re-examined the CGNP transcriptome data with the purpose of finding an explanation for the existence of age-specific SHH pathway mutations. We hereby hypothesized that these genes are differentially susceptible to oncogenic mutation as a function of time because they are not equally important at all stages of development. If true, there could be differences in peak gene expression levels for the mutated SHH pathway genes as we had observed for Tp53 (Fig. 3A). To test this, we extracted the individual gene expression profiles from genes associated with the SHH signaling pathway from our CGNP transcriptome data. However, although we observed that SHH target genes like Gli1, Gli2, Ccnd1, Ccnd2, Ptch1 and Ptch2 (Ccnd1 and Ccnd2 encode cyclin D1 and D2, respectively), and to a lesser extent Mycn, had clear expression peaks at P0, this was not evident for either the Sufu or Smo genes, which exhibited striking age-specific mutation patterns (Fig. 3C; Fig. S2). Thus, only Gli2 and Mycn, the homologs of which are predominantly mutated in children over the age of three in conjunction with TP53, show age-specific gene expression peaks (Kool et al., 2014).
We subsequently searched for alternative age-related processes that could impose differential SHH pathway usage on the CGNPs. For this, we subjected the gene clusters from the cross-species comparison to gene ontological analysis (Fig. 3B,D; Table S4). We identified only five enriched biological processes, which were all enriched in the older CGNP or patient group (Fig. 3B, lower gene cluster indicated by red bar). These processes were either involved in cell proliferation or in primary cilium formation (Fig. 3D; Table S5). Given that it is known that primary cilia are required for SHH signaling, it seemed paradoxical that younger patients have relatively low primary cilia gene expression. We stained a small panel of SHH medulloblastoma samples derived from young patients for ARL13B, a marker for primary cilia (Caspary et al., 2007), and confirmed that there is large variation in the number of ciliated cells between different patients, with some tumors hardly expressing any primary cilia (Fig. 3E) (Gate et al., 2015; Han et al., 2009). Thus, low primary cilium expression can occur in very young SHH medulloblastoma patients.
We next investigated primary cilium expression in normal developing CGNPs. In human second trimester fetal cerebellum, we found ciliated cells in both the EGL and IGL, the latter being most prominent (Fig. 3E, upper left panel). As we had no access to first trimester embryonic human cerebellum that harbors the early specified uRL and presumptive EGL cells, we subsequently turned to the developing murine cerebellum to establish the dynamics of primary cilium expression from the uRL stage onwards. In contrast to the general view that most cells have a primary cilium, we found that in the early E12.5–E13.5 mouse cerebellum, cells in the uRL and especially in the future EGL appear to have fewer primary cilia compared to surrounding brain structures (Fig. 4A) (Bangs et al., 2015; Michaud and Yoder, 2006). However, we cannot exclude that some of the apically localized primary cilia contacting the ventricle are expressed by developing CGNPs. Hence, to assess primary cilia expression specifically in developing CGNPs, we reverted to the transgenic mouse model to label Math1-expressing cells and their descendants with tdTomato at E13.5. This permitted quantifying CGNPs carrying a primary cilium from E15.5 onwards. We found that the number of ciliated cells in the EGL and later also the IGL, increases over time (Fig. 4B,C). At neonatal stages, ciliated CGNPs were more abundant towards the outer EGL in agreement with earlier studies (Chang et al., 2019; Ong et al., 2020). However, ciliated CGNPs were most frequent in the IGL, and these cilia were also significantly longer (Fig. 4B–D). Altogether, this demonstrates that both primary cilium expression and length are dynamic during CGNP development, and that at early phases of medulloblastoma cell-of-origin development, primary cilium expression is less pronounced.
SUFU and SMO expression is dynamic during cerebellar development
Both Smo and Sufu function is essential for SHH signaling in the cerebellum, with SMO acting upstream in relaying the signals from SHH morphogen-bound PTCH proteins, and SUFU acting downstream in receiving signals from activated SMO to cease the inhibition of GLI factors (Blaess, 2006; Kim et al., 2011). A major difference though is their dependence on the primary cilium for pathway activation. SMO requires the primary cilium for its function, whereas SUFU is also known to have cilium-independent activity (Chen et al., 2009; Haycraft et al., 2005; Huangfu and Anderson, 2005; Huangfu et al., 2003; Jia et al., 2009; Rohatgi et al., 2007; Spassky et al., 2008). Hence, if a subset of infant medulloblastoma is derived from the earliest specified CGNPs, this could explain the increased frequency of SUFU mutations, as these cells are less frequently ciliated. We therefore checked whether during cerebellar development, SUFU and SMO proteins were expressed in patterns consistent with this hypothesis (Fig. 5A,B). In line with cilium-dependent functions, we occasionally detected SUFU and SMO expression in primary cilia (Fig. S3A). Interestingly, SUFU was most prominently expressed in the presumptive EGL at embryonic day E15.5 and was also found in the uRL (Fig. 5A). At postnatal stages, SUFU expression became more restricted towards the outer EGL and upper layers of the IGL. In contrast, SMO expression appeared to be most highly expressed in the area where the IGL forms (Fig. 5B). These differences in expression pattern could indicate differences in activity during cerebellar development.
Embryonic CGNPs have reduced sensitivity to SMO manipulation but can be activated by downstream SHH signaling
We next set out to functionally test whether embryonic CGNPs are differentially sensitive to SHH pathway component manipulation compared to early postnatal CGNPs, as suggested by the differential expression of SUFU, SMO and primary cilia. For this, we isolated Math1-CreERT2; tdTomato CGNPs from either E15.5 or P7 cerebellum and cultured them for brief periods of time to preserve their primary status (48–72 h) (Anne et al., 2013; Hatten, 1985; Lee et al., 2009; Salero and Hatten, 2007). We observed ciliated CGNPs in both E15 and P7 cultures (Fig. 5C), suggesting that cilia are not completely absent from E15.5 CGNPs even though they are expressed less frequently (Fig. 5D,E). We then explored SHH pathway activation in these cells. In agreement with earlier publications, we found a significant increase in P7 cell proliferation upon overexpression of oncogenic Smo (SmoM2) and a trend towards increased proliferation upon knockdown of Sufu, indicating engagement of upstream and downstream SHH pathway segments at this time (Fig. 5F; Fig. S3B,C) (Dahmane and Ruiz, 1999; Kim et al., 2011; Wechsler-Reya and Scott, 1999). However, in E15.5 cells, only Sufu knockdown cells showed increased proliferation, whereas SmoM2 overexpression had no effect (Fig. 5F; Fig. S3B,C).
Prompted by these results, we wanted to address whether E15.5 CGNPs are sensitive to treatment with the SMO inhibitor cyclopamine, derivatives of which are being used in the treatment of relapsed SHH medulloblastoma (Taipale et al., 2000). Whereas total cell counts of P7 CGNPs were reduced by cyclopamine treatment as was shown previously by others, E15.5 CGNPs appeared less sensitive (Fig. 6A) (Kenney, 2003; Lafranchi et al., 2014). Likewise, treatment with purmorphamine, a SMO agonist, had no effect on E15.5 CGNP proliferation whereas P7 CGNPs were highly sensitive (Fig. 6B; Fig. S3D). In contrast, when we treated E15.5 CGNPs with HPI1, a compound that inhibits the downstream SHH target GLI1, we did see a significant reduction in proliferation, implying that the downstream SHH pathway is functional in early embryonic CGNPs (Fig. 6C). To further address this at the molecular level, we analyzed the expression of key SHH target genes (e.g. Gli1, Ptch1, Nmyc, Ccnd1 and Ccnd2) in purmorphamine- or HPI1-treated E15.5 and P7 CGNPs (Fig. 6D–D″; Fig. S3E,F). Surprisingly, whereas E15.5 CGNPs did not increase proliferation upon purmorphamine treatment, they did show significant induction of Gli1 gene expression, suggesting that these early cells are not completely unresponsive to upstream SHH signaling (Fig. 6D). However, in line with the absence of a proliferative response, they did not increase expression of Ccnd2 as did P7 CGNPs (Fig. 6D″). Together, these data show that the age of the medulloblastoma cell-of-origin determines the extent of the response to SHH pathway activation, and as a consequence has an impact on SHH drug sensitivity (Fig. 6E).
Whereas SHH medulloblastoma is characterized by an overall deregulation of the SHH signaling pathway, there are also abnormalities specific to patient age (Gibson et al., 2010; Kho et al., 2004; Kool et al., 2014; Northcott et al., 2011; Northcott et al., 2017). In this study, we set out to address whether intrinsic changes in developmental processes between young and older CGNPs, a cell lineage known to be highly susceptible to SHH medulloblastoma formation, can account for the age-specific characteristics of this tumor type (Hatten and Roussel, 2011; Yang et al., 2008).
The developmental age of the medulloblastoma cell-of-origin is partially reflected in tumors
The idea that naturally occurring changes in identity of the tumor cell lineage-of-origin impact on tumor outcome, is supported by studies showing differences in tumor onset and phenotype in relation to timing of tumor induction, although the type of genetic lesion and intra-tumoral heterogeneity are also expected play a role (Dey et al., 2012; Gibson et al., 2010; Hovestadt et al., 2019; Ohli et al., 2015; Swartling et al., 2012; Tan et al., 2018; Vladoiu et al., 2019). Indeed, when we compared gene expression patterns between developing CGNPs, we found unique enriched biological processes at each developmental stage, and several of those have been linked to SHH medulloblastoma. A linear chronological relationship between CGNP and patient age was evident in some but not all cases (Cavalli et al., 2017; Kho et al., 2004; Kool et al., 2014; Northcott et al., 2011; Northcott et al., 2017). For instance, neurotransmission activity and neural development are enriched in both early embryonic CGNPs and infant medulloblastoma (Cavalli et al., 2017). PI3K signaling, on the other hand, which is also associated with infant medulloblastoma, is mostly enriched in the late postnatal CGNP samples representing maturing granule neurons (Kool et al., 2014). Intriguingly, there was a strong enrichment for processes related to cell cycle control and the DNA damage response throughout early CGNP development with a noticeable peak around birth. As these processes are overrepresented within the older SHH medulloblastoma subtypes, this suggests that the SHH-induced perinatal surge in CGNP proliferation, possibly coinciding with increased DNA damage due to replication stress, is a critical event during cerebellar development that brings along the risk of developing child and adulthood medulloblastoma (Cavalli et al., 2017; Hovestadt et al., 2019; Northcott et al., 2017; Vladoiu et al., 2019). The strong increase in TP53 expression at birth is particularly interesting, as it might be related to the high incidence of TP53 mutations and occurrence of chromothripsis in children (Rausch et al., 2012).
Differential SHH pathway regulation during cerebellar development might impact on infant SHH medulloblastoma outcome
An important remaining question is the precise nature of the cell-of-origin for infant SHH medulloblastoma. In our cross-species comparison, we found that the infant SHH medulloblastomas clustered towards the earliest embryonic CGNPs, suggesting an early embryonic granule neuron progenitor (up to E15.5) as the cell-of-origin (Gibson et al., 2010). This is in line with elegant single-cell sequencing studies, which showed that a subset of SHH medulloblastoma clustered with early embryonic cerebellar cells that were identified by the authors as CGNPs and unipolar brush cell precursors of comparable age (Hovestadt et al., 2019; Vladoiu et al., 2019). However, initially we were puzzled by the fact that both infant SHH tumors and early embryonic CGNPs were characterized by low primary cilium expression, as the latter is believed to be indispensable for SHH signaling (Huangfu et al., 2003; Sasai and Briscoe, 2012; Wheway et al., 2018); how can we reconcile early embryonic CGNPs being the cell-of-origin for infant medulloblastoma exhibiting deregulated SHH signaling, if these cells are not capable of SHH signaling? And moreover, how can we then explain that there is a higher incidence of SUFU mutations, and complete lack of SMO mutations, in these tumors (Kool et al., 2014)?
We now think that the unique role of SUFU in the SHH pathway might be the answer to these questions. It had already been shown that in contrast to Smo, Sufu does not absolutely depend on the primary cilium to exert its control over SHH target gene expression (Chen et al., 2009; Corbit et al., 2005; Jia et al., 2009; Rohatgi et al., 2007). In addition, another vital difference between Smo and Sufu is their effect on SHH signaling, as SMO activates and SUFU represses target gene expression (Cheng and Bishop, 2002; Fuccillo et al., 2006; Jiang and Hui, 2008; Kenney, 2003; Knoepfler et al., 2002). We therefore favor a model in which transformation of early stage CGNPs into infant medulloblastoma requires SUFU deletion, as this would induce precocious pathway activation independently of a SHH signal, but not SMO activation, which would not have a proliferative effect, possibly as the result of less frequent primary cilium expression (Han et al., 2009). This idea is supported by studies from others showing a significantly earlier role in cerebellar development for Sufu compared to Smo and Shh (Blaess, 2006; Corrales, 2004; Corrales, 2006; Kim et al., 2011; Lewis et al., 2004; Svärd et al., 2006; Varjosalo et al., 2006). Furthermore, mutant cerebella lacking primary cilia also show relatively late developmental phenotypes, in line with primary cilia being most important during perinatal SHH-induced CGNP proliferation (Chizhikov et al., 2007; Spassky et al., 2008). Whether the frequency or maturity the primary cilium itself plays a functional role in this, awaits further investigation.
Implications for future medulloblastoma research
Only recently, it was demonstrated that the four consensus subgroups of medulloblastoma can be further subclassified using a combination of molecular profiling techniques (Cavalli et al., 2017; Kool et al., 2014; Lin et al., 2016; Northcott et al., 2011; Northcott et al., 2017). These studies have provided a wealth of information on the molecular genetics, as well as putative regions of origin for these tumors, which are both essential pieces of information for developing accurate preclinical models and subsequential drug testing. For instance, although several mouse models for medulloblastoma have been generated, none of them has faithfully recapitulated infant SHH medulloblastoma, suggesting that the correct cell-of-origin has not been properly targeted (Pöschl et al., 2014). The importance of bona fide preclinical modeling is underscored by our finding that early embryonic CGNPs, the putative cells-of-origin for infant SHH medulloblastoma, are less sensitive to SMO inhibition. Thus, in addition to taking into account the level at which the SHH pathway is compromised when designing targeted therapy, intrinsic characteristics of the cell-of-origin preserved in the tumor should also be taken into consideration (Hassounah et al., 2012; Zhao et al., 2017). Especially in infants, who typically do not receive radiotherapy, it is crucial to use drugs precisely tailored to their specific medulloblastoma subtype (Sharpe et al., 2015).
MATERIALS AND METHODS
The Math1-CreERT2; tdTomato compound transgenic mouse strain was derived from the Math1CreERT2 (Machold and Fishell, 2005) and Ai14 mouse strains (Madisen et al., 2010) (The Jackson Laboratory, strains #007684 and #007914), and was in a C57BL6/mixed background. Mice were conventionally housed, fed ad libitum and routinely genotyped by PCR. Timed matings were performed overnight, with the following morning considered E0.5.
Pregnancies were detected by measuring female weight gain at E13.5. A subset of pregnant females received a single dose of tamoxifen (2 mg/100 μl peanut oil, Sigma) by oral gavaging at E13.5. Pregnant females were killed by asphyxiation (CO2), neonatal mice until the age of P7 were killed by decapitation. Offspring from different gender were randomly assigned. All animal experiments were approved by the Institutional Animal Care and Use Committee of the University Medical Center Groningen, The Netherlands.
CGNP single-cell suspension preparation
For transcriptional analyses, CGNPs were harvested from E15.5, E17.5, P0, P7, P14 and P30 tdTomato+ dissected cerebella from offspring from tamoxifen-treated females. For primary cell cultures, CGNPs were harvested from E15.5 and P7 days old cerebella from offspring of non-treated females. Cerebellar dissection was performed using a stereomicroscope. E15–P30 cerebella were dissociated with a papain dissociation kit following the manufacturer's instructions (Worthington). Following papain treatment, ovomucoid was added to stop the reaction. For P14 and P30 cerebella, an additional Percoll gradient step was performed to remove myelin and debris. Cell suspensions were filtered through a 40 μm cell strainer prior to further processing.
Primary CGNP cell cultures
For embryonic CGNP cultures (E15.5), n=8–20 dissected cerebella were pooled. For P7 CGNP cultures, n=2–4 dissected cerebella were pooled. Single-cell suspensions were prepared as described above. Cells were pelleted and resuspended in the appropriate culture media. For E15.5 CGNPs, this was Basal medium Eagle (BME) supplemented with 1% N2, 2% B27 (Invitrogen) and 0.25 µg ml−1 SHH (R&D systems). For P7 CGNPs, Dulbecco's modified Eagle's medium and Ham's F-12 (DMEM-F12) supplemented with 1% N2, 2% B27, 1.5% sucrose, 5 μm HEPES (Invitrogen) and 0.25 µg ml−1 SHH (R&D systems) was used. For proliferation assays, 1 µM 4-hydroxytamoxifen (Sigma) was added to the culture media. For proliferation assays and RNA isolation, 250,000 E15.5 CGNPs or 500,000 P7 CGNPs cells were seeded into poly-D-lysine (100 µg ml−1, Sigma)-coated 24-well plates (E15.5) or 12-well plates (P7). For confocal imaging, 100,000 E15.5 CGNPs or 60,000 P7 CGNPs were seeded onto Ibidi 8-chamber glass slides coated with poly-D-lysine.
For drug treatments, E15.5 or P7 CGNPs were plated as described above in the presence 5 µM cyclopamine (Merck), 250 nM purmorphamine (Stem Cell Technologies), 2.5 µM HPI1 (Tocris) or DMSO, and incubated for 48 h. Cells were seeded as one or more (if sufficient cells were available) replicates per experiment, and at least three biological replicates (e.g. samples derived from different mice) were assayed as indicated in the figure legends.
The Sufu shRNA construct was generated by cloning a Sufu-targeting 22-mer oligonucleotide into a modified pRRL-SFFV-IRES-GFP plasmid (restriction sites XhoI and EcoRI), with tNGFR replaced by GFP (kind gift from Dr Hein Schepers, Dept of Cell Biology, University Medical Center Groningen, The Netherlands; see also Table S6) (Korthuis et al., 2015).
293T producer cells (ATCC, routinely tested free from mycoplasm) were transfected with prrl-SFFV-IRES-GFP, prrl-SFFV-IRES-GFP-Sufu or SmoM2 (W535L)-pcw107-V5 (Addgene #64628) using Fugene HD transfection reagent (Promega). Appropriate CGNP culture media supplemented with B27 was added to the producer cells 16 h prior to virus harvest. CGNPs were subsequently incubated with lentiviruses for 2 h, after which virus was gradually replaced with culture medium to prevent cell death. Cells were fixed at 48 h following transduction.
Quantification of cells and primary cilia
For quantification of PCNA-positive CGNPs in primary cultures, random fluorescence microscopic images (n=10 per sample) were taken and processed using CellProfiler 3.1.9 software (Broad Institute, cellprofiler.org) using an adapted Counting and Scoring pipeline. Average ratios of double positive cells (PCNA+/tdTomato+) per field of view were calculated and tested for significance (P<0.05), as stated in the figure legend.
For quantification of total CGNP numbers after cyclopamine treatment, the average number of tdTomato-positive cells per field of view was determined using Cell Profiler. Cell counts were subsequently normalized to EV or DMSO control.
For cells lentivirally transduced with SmoM2 or Sufu shRNA (prrl-SFFV-IRES-GFP-Sufu), the PCNA+/DAPI+ fraction was counted manually using Fiji software.
For primary cilium quantification or length measurement, ARL13B and tdTomato (double) positive cells were counted manually per developmental time point in cultured CGNPs, or in RL, EGL and IGL, using confocal imaging and Fiji software. Average ratios of double-positive cells per field of view were determined and tested for significance as described in the figure legend.
Tissue treatments and immunofluorescence
CGNPs were fixed with 100% methanol (for PCNA staining) or 4% formaldehyde (for ARL13B), and blocked with 0.1% Triton X-100, 1% bovine serum albumin (BSA) and 0.05% Tween 20 in PBS. Primary antibodies used were against: ARL13B (Proteintech, 17711-1-AP, lot #00049885, 1:400), PCNA (Abcam, ab29, lot #GR3195972-1, 1:1000) and RFP (Rockland, 600-401-379, lot #36233, 1:500). Secondary antibodies were Alexa Fluor 488 (1:500) and Alexa Fluor 568 (1:500) conjugated Invitrogen). Cells were counterstained with DAPI (Sigma).
Brains from tamoxifen-treated Math1-CreERT2; tdTomato mice were dissected from the skull, fixed with 4% formaldehyde, and cryoprotected with a sucrose gradient (10%, 20% and 30% sucrose in PBS). They were embedded into Tissue-Tek O.C.T. compound (Sakura-FineTek) and snap frozen in liquid nitrogen. Human fetal brain and medulloblastoma tissue was obtained through the University Medical Center Groningen, The Netherlands and the Princess Máxima Center for Pediatric Oncology, Utrecht, The Netherlands. Tumor samples and clinical data were provided by collaborating centers with approval from the respective institutional review boards/ethics committees [Medical Ethics Review Board (METc) UMCG Groningen, and the Medical Ethics Review Board (METc) Utrecht] in agreement with the Declaration of Helsinki; and informed consent from all patients or their guardians.
Cryosections were generated on a Leica cryostat. Paraffin sections were deparaffinized and rehydrated prior to further processing. Antigen retrieval was performed by boiling tissue sections in citrate buffer (100 mM, pH 6.0), except for ARL13B and mCherry. Sections were blocked with 5% normal goat serum and 0.1% Triton X-100 (Cell Signaling) in PBS. Primary antibodies used were against: ARL13B (Proteintech, 17711-1-AP, lot #00049885, 1:100), ARL13B (for co-staining with SUFU or SMO; Abcam, ab136648, lot #GR3394308-1, 1:100), mCherry (SICGEN, AB0040-200, lot #0081030119, 1:200), PCNA (Abcam, ab29, lot #GR3195972-1, 1:1000), RFP (Rockland, 600-401-379, lot# 36233, 1:500), SUFU (Abcam, ab28083, lot #GR3263748-3, 1:50), and SMO (Abcam, ab38686, lot #GR198520-1, 1:500). Secondary antibodies were Alexa Fluor 488 (1:500) and Alexa Fluor 568 (1:500) conjugated (Invitrogen). Slides were counterstained with DAPI (Sigma), and mounted with Vectashield (Vector Laboratories) or Prolong Diamond Antifade Mountant (Thermo Scientific).
Whole-mount images of fluorescent embryonic and postnatal mouse brain were made on an Olympus stereozoom SZX-16 microscope. Primary cells were imaged on an EVOS FL inverted fluorescence microscope (Life Technologies). Fluorescent tissue sections were imaged on Leica TCS SP8 or SP8X DLS confocal microscopes, and analyzed using Fiji (Schindelin et al., 2012).
RNA from E15.5 and P7 CGNPs was isolated using the RNeasy Mini Kit (Qiagen). cDNA was synthesized using random hexamer primers, dNTPs and Ribolock (Thermo Fisher Scientific). Reverse transcription was performed using reverse transcriptase (Thermo Fisher Scientific). Quantitative RT-PCR was performed on a LightCycler 480 System (Roche) system using universal SYBR® Green supermix (Bio-Rad). Relative gene expression was calculated using the 2−ΔΔCT method. Expression levels were normalized to housekeeping gene Gapdh. For primers, see Table S6.
CGNP FACS and RNA isolation for RNA-Seq
Cerebellar single-cell suspensions (E15.5–P30) were subjected to fluorescence-activated cell sorting (FACS) for tdTomato+ cells, using a MoFlo Astrios cell sorter (Beckman Coulter). tdTomato+ cells (CGNPs) were collected in 1.5 ml tubes, centrifuged (500 g, 10 min at 4°C) and snap frozen. For the E15.5 and E17.5 time points, tdTomato-positive cerebella from n=4–8 embryos from the same litter were pooled prior to FACS to constitute one biological replicate; for P0–P30 time points, individual mice were sorted to constitute a biological replicate.
RNA was extracted using the NucleoSpin RNA XS kit (Macherey-Nagel), following the manufacturer's instructions. RNA concentration was measured with a QuBit RNA HS assay kit (Thermo Fisher Scientific) and analyzed on a bioanalyzer using the Agilent RNA 6000 Pico Kit (Agilent). 2.5–10 ng of RNA per sample was used as input for library preparation.
Library preparation and RNA-seq
For library prepping, the Clontech SMARTer stranded total RNA-seq kit-pico was used according to the manufacturer's instructions for mammalian sample input. Three biological replicates were sequenced for each time point (Illumina HiSeq 2500). On average, 16.7 million reads (63 bp single-end) were generated for each replicate. Reads were aligned to the mouse reference genome (GRCm38 assembly, gene annotation from Ensembl release 84, http://www.ensembl.org) and quantified using STAR 2.5.3a (Dobin et al., 2013).
Hierarchical clustering analysis and gene expression analysis
For hierarchical clustering analysis, differentially expressed genes were called for all possible pairwise comparisons of developmental stages (three replicates each) using the edgeR package (Robinson et al., 2010). Genes were selected that significantly changed their expression [false discovery rate (FDR)<10−5] in one or more comparisons. Log2 ratios were calculated relative to the average FPM expression of a gene. Unsupervised hierarchical clustering and heatmap plotting was performed using the gplots library (https://github.com/talgalili/gplots). The Canberra distance and Ward clustering method was used.
Cross species comparison
For comparison of gene expression profiles between mouse CGNP and human medulloblastoma patients, a published human medulloblastoma data set was downloaded from GEO (accession number GSE49243). Expression values were transformed into log2 fold change (compared to average expression of a gene across all patients). Unambiguous orthologs (one to one orthology) were determined using the Ensembl Biomart tool (http://www.ensembl.org/biomart). Unsupervised hierarchical clustering and heatmap plotting was performed using the gplots library. The Euclidean distance and average clustering method was used.
Pathway enrichment analysis was performed by uploading lists of differentially expressed genes (gene clusters), to the Database for Annotation, Visualization and Integrated Discovery v6.8 (DAVID), and subsequent analysis for gene ontology of biological processes was performed (Huang et al., 2009). Lists of biological processes were imported into the Enrichmentmap app in Cytoscape v3.2.1 (Merico et al., 2010; Shannon et al., 2003). Biological processes were Benjamini-corrected with a moderately permissive q value of <0.1 and a P-value of <0.01. Enrichment maps represent biological processes enriched in the gene clusters. Each node represents a biological process grouped and labeled by biological theme. Biological processes connected by edges have genes in common using a Jaccard and Overlap coefficient combined with a similarity cutoff value of 0.375.
We are grateful to Dr Ilia Vainchtein and the UMCG Flow cytometry center for assistance on cerebellar granule neuron progenitor sorting; to Petra Bakker BSc, Dr Seka Lazare, Dr Marianna Bevova, Dr Diana Spierings and the ERIBA sequencing facility for assistance with RNA-sequencing; and to the UMCG Central Animal Facility (CDP) for assistance with animal experiments. We also thank the UMCG Microscopy and Imaging Center (UMIC), at which part of the work presented in this study has been performed. We are further very grateful to Dr Mariëtte Kranendonk and Dr Pieter Wesseling from the Princess Máxima Center for Pediatric Oncology for providing tissue samples.
Conceptualization: M.J.S., S.W.M.B.; Methodology: M.J.S., T.E.I.M., J.T.M.L.P., V.G., S.W.M.B.; Software: W.W.Z., V.G.; Validation: M.J.S., S.W.M.B.; Formal analysis: M.J.S., W.W.Z., V.G., S.W.M.B.; Investigation: M.J.S., T.E.I.M., I.A., I.B., W.W.Z., E.S.d.C.M., Z.S., T.G.J.M., F.J.G.S., M.R., S.W.M.B.; Resources: M.H.S., W.F.A.d.D., E.W.H., J.T.M.L.P., G.d.H., S.W.M.B.; Data curation: V.G.; Writing - original draft: M.J.S., S.W.M.B.; Writing - review & editing: M.J.S., T.E.I.M., I.B., W.W.Z., J.T.M.L.P., S.W.M.B.; Visualization: M.J.S., W.W.Z., V.G., S.W.M.B.; Supervision: S.W.M.B.; Project administration: M.J.S., S.W.M.B.; Funding acquisition: M.J.S., I.A., E.S.d.C.M., J.T.M.L.P., G.d.H., S.W.M.B.
This work was funded by the UMCG Kanker Researchfonds (KRF), a Stichting De Cock-Hadders grant to M.S.; an Indonesian Endowment Fund for Education (LPDP) grant (20151022024475) to I.A.; a Coordenação de Aperfeiçoamento de Pessoal de Nível Superior, Brasil, CAPES, grant, Finance Code 001 to E.C.M.; a Rosalind Franklin fellowship from the University of Groningen and European Union to J.P.; a Mouse Clinic for Cancer and Ageing/Large Infrastructure grant from the Netherlands Organisation for Scientific Research (NWO) to G.H., a Stichting Vrienden Beatrix Kinderziekenhuis grant, a Rosalind Franklin fellowship from the University of Groningen, and a KWF Kankerbestrijding Cancer Research career award (RUG 2014-6903) to S.B. Open Access funding provided by University Medical Center Groningen (Universitair Medisch Centrum Groningen). Deposited in PMC for immediate release.
The authors declare no competing or financial interests.