Norrin/Frizzled4 signalling in the preneoplastic niche blocks medulloblastoma initiation

The tumor microenvironment is a critical modulator of carcinogenesis; however, in many tumor types, the influence of the stroma during preneoplastic stages is unknown. Here we explored the relationship between pre-tumor cells and their surrounding stroma in malignant progression of the cerebellar tumor medulloblastoma (MB). We show that activation of the vascular regulatory signalling axis mediated by Norrin (an atypical Wnt)/Frizzled4 (Fzd4) inhibits MB initiation in the Ptch+/− mouse model. Loss of Norrin/Fzd4-mediated signalling in endothelial cells, either genetically or by short-term blockade, increases the frequency of pre-tumor lesions and creates a tumor-permissive microenvironment at the earliest, preneoplastic stages of MB. This pro-tumor stroma, characterized by angiogenic remodelling, is associated with an accelerated transition from preneoplasia to malignancy. These data expose a stromal component that regulates the earliest stages of tumorigenesis in the cerebellum, and a novel role for the Norrin/Fzd4 axis as an endogenous anti-tumor signal in the preneoplastic niche. DOI: http://dx.doi.org/10.7554/eLife.16764.001


Introduction
The tumor microenvironment is comprised of many stromal cell types, including the vasculature, which is well known to promote the growth and propagation of established tumors

Results
Ndp is expressed in GNPs and mouse and human Shh-MB To assess the stromal compartment at early stages of tumorigenesis in the Ptch +/À cerebellum, we sampled entire lesions at postnatal day (P) 14 by Collagen IV+ immunostaining, and observed an invasion of vasculature in a minority (24%; Figure 1A). Furthermore, lesion volume, which is a measure of neoplastic progression in the Ptch +/À model, was statistically larger in vascularized lesions (mean 0.18 mm 3 ) compared to non-vascularized ones (mean 0.029 mm 3 , Figure 1B). These observations are notable, considering that only a minority of lesions undergo malignant transformation and continue to grow into tumors (Kessler et al., 2009;Oliver et al., 2005). To explore this pre-tumor/ blood vessel interaction, we targeted Norrin signalling, a well-characterized regulator of neuralendothelial cell communication in the cerebellum (Wang et al., 2012;Zhou et al., 2014). Norrin (encoded by the X-linked gene Ndp) is a secreted atypical Wnt ligand that signals specifically through the Fzd4 receptor and the Lrp5/6 and Tspan12 co-receptors to activate the canonical Wnt pathway in endothelial cells ( Figure 1C), where it is required for blood brain barrier (BBB) integrity in the cerebellum (Junge et al., 2009;Wang et al., 2012;Xu et al., 2004). Using male Ndp -/y mice carrying an Ndp-lacZ knockout (KO) allele (Junge et al., 2009), we examined the cerebellar expression profile of Ndp. Consistent with alkaline phosphatase-based reporter data (Ye et al., 2010a), we detected Ndp expression in the Purkinje cell layer, presumably in Bergmann glia ( Figure 1D). We also detected Ndp expression in the cerebellar external granule layer (EGL), where GNPs, the Shh-MB cell of origin, reside throughout postnatal development ( Figure 1D). Combined X-gal staining and immunohistochemistry (IHC) during the peak period of GNP proliferation at P7 revealed that Ndp expression is concentrated in the outer region of the Pax6 + EGL, in the proliferative phosphohistone H3 (PH3) + compartment ( Figure 1D). b-gal + cells also overlapped with myelin basic protein counterstained with hematoxylin, to depict non-vascularized (red outline) and vascularized (green outline) lesions from 37 Ptch +/À mouse cerebellar lesions sampled by serial sections at P14. (B) Boxplot showing statistically larger volume in P14 Ptch +/À lesions assigned as vascularized versus those assigned as non-vascularized, based on immunostaining serial sections of each lesion for ColIV. (C) Schematic illustrating the machinery required for Norrin activation of b-catenin (b-cat)/T-cell factor (TCF)-dependent transcription via the Fzd4 receptor and Lrp5 and Tspan12 co-receptors. (D) X-gal staining (blue) of sagittally sectioned cerebella from male Ndp -/y mice carrying an Ndp-lacZ KO allele, at the ages indicated. Bottom row, combined X-gal staining and immunohistochemistry (IHC, brown) for phospho-histone H3 (PH3), Pax6, and myelin basic protein (MBP). (E) Box plot of qRT-PCR analysis of Ndp expression in mouse Ptch +/À purified GNPs and MB lysate from symptomatic animals ranging in age from 3 to 10 months. (F) Box plot of NDP expression obtained by array profiling of human cerebella and two different cohorts (Boston, left; Toronto, right) of human MB samples, categorized by molecular subgroup. (G) Kaplan-Meier survival curve illustrating overall survival of Shh-MB patients with high versus low NDP expression. EGL, external granule layer; GNP, granule neuron progenitor; PCL, Purkinje cell layer; ML, molecular layer; WM, white matter; IGL, internal granule layer; CB, cerebellum; MB, medulloblastoma. Scale bars, 100 mm. DOI: 10.7554/eLife.16764.002 (MBP) + white matter ( Figure 1D) implicating oligodendrocytes as another source of Norrin. To focus our expression analysis on the Ptch +/À tumor-relevant cell type, we examined Ndp expression in GNPs isolated from the cerebellar surface at various stages of tumorigenesis. While Ndp levels in Ptch +/À GNPs from pre-lesion (P7) and early lesion (P14) stages were comparable, Ndp expression exhibited a downward trend in GNPs at a lesion progression stage (P30) that reached significance by the tumor stage ( Figure 1E). In human MB, NDP expression is enriched specifically in the Shh subgroup compared to the other three molecularly distinct subgroups: Wnt, driven by aberrant Wnt pathway activation, and Group 3 and 4, which are less clearly defined biologically (Taylor et al., 2012) ( Figure 1F). Based on a limited sample size, we observed a trend towards reduced survival in Shh-MB patients with tumors exhibiting the lowest levels of NDP expression compared to those with the highest ( Figure 1G). Thus, given its known function in signalling to endothelial cells and its expression pattern in GNPs and MB, the Norrin/Fzd4 axis is well-positioned to mediate neural-endothelial cell crosstalk within the Shh-MB lesion and tumor microenvironment.

Norrin/Fzd4 loss of function dramatically enhances Ptch +/À MB formation
To explore the relevance of the stromal compartment during Shh-MB evolution, we disrupted the Norrin/Fzd4 signalling axis at the level of the ligand or receptor in Ptch +/À mice. Given that MB incidence in Ptch +/À mice is dependent on the genetic background strain (Mille et al., 2014), we performed all long and short-term studies of tumorigenesis by comparing animals within the same breeding cohort. While germ-line deletion of the ligand Ndp (Ndp KO ) or conditional deletion of Fzd4 in endothelial cells via the Tie2Cre driver (Tie2Cre+;Fzd4 fl/fl ) alone did not result in tumors, both mutations dramatically accelerated MB formation and significantly reduced survival in Ptch +/À mice ( Figure 2A-C), to an extent that has only previously been observed in this model upon DNA damage (Pazzaglia et al., 2006a;2006b;Wetmore et al., 2001). In both Ndp KO ;Ptch +/À and Tie2-Cre+;Fzd4 fl/fl ;Ptch +/À compound mutants, MB incidence increased by approximately two-fold and mean latency was reduced more than two-fold compared to Ptch +/À littermates ( Figure 2C). In isolated GNPs, we detected the expression of endogenous FZD4 protein ( Figure 2D), and transcripts of Fzd4, Lrp5 and Tspan12 ( Figure 2E); however, conditional deletion of Fzd4 from GNPs in the Ptch +/À model had no impact on survival and tumorigenesis ( Figure 2F). These results reveal a novel tumor inhibitory role for Norrin/Fzd4 signalling in Ptch +/À MB that, strikingly, is mediated by the endothelial cell component of the tumor stroma.
Loss of Ndp alters the stromal gene expression signature in Ptch +/À MB To gain insight into Norrin-mediated effects on Ptch +/À tumorigenesis, we examined established Ndp KO ;Ptch +/À and Ptch +/À MBs by histology and expression profiling. Both tumor types exhibited classic histology, and similar expression patterns for expected Shh-MB neural markers (Atoh1, TUBB3 and GFAP) and Hh target genes (Gli1, Mycn and Ccnd1) ( Figure 3A). Co-immunostaining for the ECM protein laminin and the vascular endothelial cell marker CD31 (PECAM1; platelet/endothelial cell adhesion molecule 1) revealed a stromal component in both tumors ( Figure 3B). However, tumor heterogeneity poses challenges to histological-based classification, therefore we turned to whole genome expression profiling, where hierarchical clustering and principal component analysis revealed that Ndp KO ;Ptch +/À and Ptch +/À tumors have clearly separable gene expression signatures ( Figure 3C and Figure 3-figure supplement 1A). Consistent with a stromal requirement for Norrin/Fzd4 signalling, tumors in Ndp KO ;Ptch +/À mutants were enriched for changes in stromal genes compared to their Ptch +/À counterparts, particularly ECM components ( Figure 3D,E). Up-regulated genes in Ndp KO ;Ptch +/À tumors included components of endothelial cells, including Esm1 (Endothelial cell-specific molecule 1), Plvap (Plasmalemmal vesicle associated protein) and Emcn (Endomucin) (Figure 3-figure supplement 1C,D), which we validated by qRT-PCR ( Figure 3F). The up-regulation of Plvap, a marker of fenestrated endothelial cells, was associated with increased vascular permeability in Ndp KO ;Ptch +/À versus Ptch +/À tumors, demonstrated by enhanced leakage of the serum protein binding dye Evans blue ( Figure 3G). Additional genes upregulated in Ndp KO ;Ptch +/À versus Ptch +/À tumors by qRT-PCR were Pecam1, suggesting increased vascularity in Ndp KO ;Ptch +/À tumors, and the angiogenic regulator Angpt2 (Angiopoietin-2) ( Figure 3F). These results suggest that rather than impacting Shh signalling, loss of Norrin modulates features of the tumor stroma. Kaplan-Meier survival curve to assess the impact of Ndp deletion in Ptch +/À mice. Ndp KO animals do not develop tumors but five of 44 animals were euthanized prematurely due to a skin condition. (B) Kaplan-Meier survival curve to assess the impact of endothelial cell-targeted (Tie2Cre-driven) Fzd4 deletion in Ptch +/À mice. Tie2Cre+; Fzd4 fl/fl animals do not develop tumors but exhibit reduced survival as reported in Fzd4 KO mice, which die with esophageal-related feeding defects and progressive auditory and cerebellar degeneration (Wang et al., 2001). (C) Summary of sample sizes, MB incidence and average latency in all animals from Kaplan-Meier survival curves in A and B. (D) Purified P10 GNPs immunostained for anti-FZD4 and anti-keyhole limpet hemocyanin (KLH) isotypematched control (green), counterstained with Hoescht (blue). White boxes are magnified at right. (E) Box plots of qRT-PCR analysis of Norrin receptor components in isolated mouse GNPs and Ptch +/À MB. (F) Kaplan-Meier survival curve to assess the impact of GNP-targeted (Atoh1Cre-driven) Fzd4 deletion in Ptch +/À mice. (G) Summary of sample sizes, MB incidence and average latency in all animals from Kaplan-Meier survival curve in F *Died with confirmed MB; †Do not develop MB. WT, wild-type; n.s., not significant. DOI: 10.7554/eLife.16764.003 Figure 3. Ptch +/À and Ndp KO ;Ptch +/À MBs have separable gene expression signatures enriched for stromal gene changes. (A) Sections of Ptch +/À and Ndp KO ;Ptch +/À established MBs stained by hematoxylin and eosin (H and E), in situ hybridization for Atoh1, Gli1, Mycn and Ccnd1 (purple), or immunostaining for class III b-tubulin (TUBB3) and glial fibrillary acidic protein (GFAP; green) counterstained with Hoescht (blue). n = 3 tumors of each genotype examined. (B) Sections of Ptch +/À and Ndp KO ;Ptch +/À established MBs co-immunostained for CD31 (green) and pan-laminin (red), counterstained with Hoescht (blue). n = 3 tumors of each genotype examined. (C) Whole genome expression profiles of Ptch +/À and Ndp KO ;Ptch +/À MBs and P6 WT (wild-type) GNPs were used for principal component analysis performed with the 1500 most variable probes across all samples. (D,E) Cellular component gene ontology (GO) analysis of differentially expressed genes between Ptch +/À and Ndp KO ;Ptch +/À MBs. (F) qRT-PCR analysis of Figure 3 continued on next page Stromal Norrin/Fzd4 signalling regulates lesion induction in the Ptch +/À cerebellum In the Ptch +/À model, MB progression is impacted by both the number of preneoplastic lesions, and the rate at which they transform to established tumors Pazzaglia et al., 2006a;2006b;Tanori et al., 2010). To determine the stage of tumorigenesis where stromal Norrin/ Fzd4 signalling is critical, we examined lesion formation in cerebella of Ndp KO ;Ptch +/À and Tie2Cre +;Fzd4 fl/fl ;Ptch +/À compound mutants. At P14, the earliest stage when lesions are consistently detected in Ptch +/À mice, the number of cerebellar lesions in Ndp KO ;Ptch +/À and Tie2Cre+;Fzd4 fl/fl ; Ptch +/À mutants was increased 3.9-fold and 2.4-fold, respectively, compared their Ptch +/À littermates ( Figure 4A,C). This effect on lesion number was not associated with changes to the Ptch +/À cerebellum at prior ages, as we failed to detect separable GNP expression profiles or general overgrowth of the EGL in Ndp KO ;Ptch +/À versus Ptch +/À cerebella at P6 (Figure 4-figure supplement 1). Thus, we have revealed an endogenous signalling axis in the preneoplastic microenvironment that plays a protective role during the early stages of Shh-induced tumorigenesis.
Loss of Norrin/Fzd4 signalling promotes stromal remodelling and angiogenesis in Ptch +/À lesions Together with the altered stromal expression signature in Ndp KO ;Ptch +/À tumors, our finding of enhanced lesion formation suggests that loss of Norrin/Fzd4 signalling may induce stromal perturbations that impact the earliest stages of MB progression. We therefore compared stromal characteristics in lesions from Ndp KO ;Ptch +/À and Tie2Cre+;Fzd4 fl/fl ;Ptch +/À compound mutants and their Ptch +/À littermates. Consistent with our previous observations ( Figure 1A), immunostaining for CD31 and laminin showed that the majority of Ptch +/À lesions were poorly vascularized, aside from existing blood vessels associated with the meninges at the cerebellar surface ( Figure 5A). In contrast, compound mutant lesions exhibited several hallmarks of angiogenic remodelling, including irregular deposition of ECM ( Figure 5A), an increased frequency of mitotic endothelial cells per vessel area ( Figure 5B), and an increase in vascular endothelial cell and laminin density ( Figure 5C). This vascular remodelling was restricted to lesions, with only rare examples of disrupted morphology associated with the EGL in single Tie2Cre+;Fzd4 fl/fl cerebella (Figure 5-figure supplement 1). Interestingly, while the vast majority of compound mutant lesions were vascularized at P14 (78% Ndp KO ;Ptch +/À and 95% Tie2Cre+;Fzd4 fl/fl ;Ptch +/À versus 24% Ptch +/À ; Figure 5D), compound mutant lesion volume was not increased at this early stage of tumor evolution ( Figure 4B,D). Accordingly, we observed vascular remodeling in very small (<0.02 mm 3 ) compound mutant lesions (Figure 5-figure supplement 2). Thus, vascular remodelling induced by loss of Norrin/Fzd4 is independent of the increase in lesion volume.
Consistent with the known role for Norrin/Fzd4 in normal BBB development and function [ (Wang et al., 2012) and Figure 6-figure supplement 1A,B], lesions in compound mutants showed a marked loss of vascular integrity, characterized by induction of PLVAP ( Figure 6A,B), variable loss of the tight junction protein Claudin-5 ( Figure 6A), and leakage of Evans blue ( Figure 6C). In contrast, single Ptch +/À mice never exhibited leaky vessels ( Figure 6A-C and Figure 6-figure supplement 1A,B), which is consistent with the positive regulation of BBB maintenance by active Hh signalling (Alvarez et al., 2011). Accordingly, the proportion of leaky lesions was significantly increased in compound mutants compared to Ptch +/À ( Figure 6D). Lesion-associated vessels in compound mutants did not show overt differences in pericyte coverage ( Figure 6-figure supplement 1D,E), however, they did exhibit a high frequency of perivascular accumulation of CD45+ leukocytes compared with Ptch +/À lesions ( Figure 6D,E). Together, these results show that loss of Norrin/Fzd4 signalling to endothelial cells accelerates the transition to a tumor-permissive stroma characterized by vascular permeability, inflammation and angiogenic remodelling. These changes are reminiscent of the switch to angiogenesis that promotes the preneoplastic progression of non-central nervous system (CNS) tumors (Baeriswyl and Christofori, 2009), where the range of cellular and molecular players involved include Angpt2 (Mazzieri et al., 2011;Rigamonti et al., 2014), consistent with our observation of increased Angpt2 expression in Ndp KO Ptch +/À tumors ( Figure 3F).
Our findings suggest that Wnt signaling in the endothelium is normally anti-angiogenic, particularly in the context of lesion formation. Therefore, we asked whether vascularization of Ptch +/À lesions was associated with altered Wnt signalling by examining expression of Lef1, a canonical Wnt target gene (Logan and Nusse, 2004). Endothelial Lef1 expression in P14 Ndp KO ;Ptch +/À lesions was significantly reduced compared to P14 Ptch +/À lesions and EGL ( Figure 6-figure supplement 2A,B), confirming the Ndp dependence of this marker in the endothelium. While Lef1 expression in the endothelium of non-vascularized Ptch +/À lesions was similar to that in adjacent EGL, we observed a downregulation of endothelial Lef1 expression in vascularized Ptch +/À lesions compared to EGL ( Figure 6-figure supplement 2B). These data suggest that lesion progression in Ptch +/À mice may be associated with a reduction in endogenous canonical Wnt activity in the endothelium, which is also consistent with the downward trend in Ndp expression during Ptch +/À lesion progression.
The tumor-protective microenvironment has an acute requirement for Fzd4 activity Genetic deletion of Ndp or Fzd4 impacts cerebellar vasculature throughout development and adulthood (Wang et al., 2012;Zhou et al., 2014). Thus, we determined whether we could recapitulate the effects of genetic inactivation on the development of a tumor-permissive stroma by short-term treatment of Ptch +/À mice with a function blocking Fzd4 antibody (aFzd4), which has been previously shown to effectively cross the BBB and impact retinal vasculature (Paes et al., 2011) (Figure 7-figure supplement 1). aFzd4 treatment of Ptch +/À mice at P7 and P16 was sufficient to phenocopy the enhanced tumor initiation phenotype of our compound mutant models, increasing MB incidence from 69% in animals injected with anti-keyhole limpet hemocyanin (aKLH) control antibody to 93% in those injected with aFzd4, and significantly reducing mean tumor latency from 167 to 104 days ( Figure 7A). Furthermore, a single dose of aFzd4 administered as late as P7 was sufficient to increase number (but not volume) of lesions at P14 ( Figure 7B), convert lesions to a leaky vasculature phenotype ( Figure 7C,I), and significantly increase the proportion of vascularized lesions compared to aKLH controls ( Figure 7D,I). Thus, acute disruption of Fzd4 promotes the conversion to an angiogenic, pro-tumor stroma within seven days, suggesting that continued maintenance of Norrin/Fzd4 signalling creates a tumor-protective microenvironment.

Ptx-Induced BBB disruption does not affect Ptch +/À tumorigenesis
To further examine what feature of the Norrin/Fzd4-deficient vasculature is pro-tumorigenic, we teased apart the impact of vessel permeability alone using pertussis toxin (Ptx), a stimulator of mouse BBB disruption (Clifford et al., 2007). Ptx treatment compromised paracellular permeability in the P14 cerebellum by disrupting endothelial cell tight junctions via reduced ZO-1 and Claudin-5 expression, accompanied by Evans Blue leakage (  (Stan et al., 2004). Ptx treatment of Ptch +/À mice did not affect MB incidence or latency compared to Ptch +/À PBS-injected controls ( Figure 7E). Furthermore, despite vascular permeability as seen by Evans Blue leakage in lesions of Ptx-treated Ptch +/À mice, the number, volume and proportion of vascularized lesions were not changed ( Figure 7F-I). Although Ptx has been shown in certain contexts to inhibit Hh signalling by disrupting Gai protein coupling to Smoothened (Riobo et al., 2006), our failure to observe changes in lesion number or MB latency suggests that GNP proliferation was not suppressed in this system. These data show that Ptx-induced BBB disruption alone does not drive lesion formation, lesion angiogenesis, or tumor initiation.
Loss of Norrin signalling accelerates Ptch LOH in Ptch +/À MB The striking acceleration of Ptch +/À MB initiation observed upon loss of Norrin/Fzd4 activity (Figure 2A-C) prompted us to assess indicators of GNP growth and malignant progression. We first observed a modest, but significant, increase in the mitotic index of P14 lesions of Ndp KO ;Ptch +/À mutants compared to their Ptch +/À counterparts ( Figure 8A), which was not counterbalanced by enhanced cell death or differentiation, as Ndp KO ;Ptch +/À lesions exhibited less Cleaved Caspase 3 immunostaining and comparable expression levels of the differentiation marker NeuN compared to their Ptch +/À littermates ( Figure 8A). Interestingly, this increased mitotic index did not translate into increased lesion volume at this stage ( Figure 4B,D). We examined the relationship between proliferating GNPs and lesion vasculature by Pax6/CD31/EdU triple staining, and found that Pax6 +EdU+ S-phase GNPs were present in the vicinity of blood vessels in Ptch +/À or Ndp KO ;Ptch +/À lesions (Figure 8-figure supplement 1). In the Ptch +/À MB context, enhanced proliferation and DNA damage in pre-malignant GNPs leads to increased incidence and decreased latency of MB (Ayrault et al., 2009;Leonard et al., 2008;Mille et al., 2014;Uziel et al., 2005). Here, Ptch loss of heterozygosity (LOH; where the remaining wild-type Ptch allele is inactivated) is a well-established indicator of malignant progression in this model (Mille et al., 2014;Pazzaglia et al., 2006a;2006b). To determine if Norrin/Fzd4 signalling affects the rate of Ptch LOH, we microdissected GNPs from lesions in Ptch +/À and Ndp KO ;Ptch +/À cerebella at P14, and determined Ptch LOH status by detection of wild-type Ptch transcript ( Figure 8B). At this stage, the frequency of Ptch LOH was increased in Ndp KO ;Ptch +/À lesions (9/17) compared to Ptch +/À lesions (3/14). Thus, the shorter MB latency observed upon Ndp loss-of-function is characterized by an accelerated rate of Ptch LOH.
To address whether the tumor inhibitory effect of Norrin/Fzd4 signaling extends to models of Shh-MB that do not require Ptch LOH for progression, we used the Neurod2-Smo A1 transgenic mouse, a Shh-MB model where an oncogenic form of Smoothened with an activating point mutation SmoA1 is expressed in GNPs (Hallahan et al., 2004). Deletion of Ndp dramatically accelerated cerebellar tumorigenesis in Neurod2-Smo A1+/À mice, reducing mean latency almost 5-fold, from 167 to 34 days ( Figure 8C). Thus, the tumor inhibitory effects of Norrin signaling extend to oncogenedriven Shh MB.

A novel link between stromal signalling and MB initiation
Stromal elements are increasingly recognized as critical to the initiation and progression of many tumor types, however little is known about the contribution of the microenvironment to MB evolution. By exploiting the multi-step nature of tumorigenesis in the Ptch +/À mouse model, we have described a novel stromal interaction regulating Ptch +/À MB initiation and identified a previously unrecognized angiogenic component to the earliest tumorigenic events in the MB microenvironment ( Figure 8D). We uncovered a cellular mechanism behind this effect, identifying endothelial cells as the mediators of a potent tumor inhibitory signal that acts during preneoplasia and oncogenic transformation. Furthermore, we identified the Norrin/Fzd4 pathway as an endogenous signalling axis that can be manipulated to target this pre-tumor/stromal relationship. The importance of Wnt signalling in brain tumor vasculature has been demonstrated in established human glioma, where tumor cell-derived Wnt ligand normalizes vasculature and inhibits angiogenesis (Reis et al., 2012), and in Figure 7 continued treated at P7 with aKLH (n = 5 lesions) or aFzd4 (n = 6 lesions). Mice were injected with Evans Blue dye prior to sacrifice and sampled by H and E staining followed by Evans Blue visualization as red fluorescence on adjacent sections. Lesions outlined in blue. (D) Lesion images from Ptch +/À mice treated at P7 with aKLH (n = 29 lesions) or aFzd4 (n = 52 lesions), immunostained for ColIV and counterstained with hematoxylin, to quantify the proportions of non-vascularized (outlined in red) and vascularized (outlined in green) lesions in each group. (E) Kaplan-Meier survival curve comparing Ptch +/À mice treated with Ptx or PBS vehicle control at P7, P9, P11 and P13. The sudden drop in Ptx-treated survival is a result of four animals dying from brain hemorrhages or seizures. Three other Ptx-treated animals were euthanized due to malocclusion or unknown causes, while all other mice in both groups died with confirmed MB. Sample sizes, MB incidence and average latency are indicated at the right. (F) Quantification of lesion number and volume from P14 cerebella of Ptch +/À mice treated as above with Ptx (n = 8 mice, 43 lesions total) or PBS (n = 7 mice, 26 lesions total). (G) Lesion images from Ptch +/À mice treated with Ptx (n = 13 lesions) or PBS (n = 13 lesions) as above. Mice were injected with Evans Blue dye prior to sacrifice and sampled by ColIV immunostaining followed by Evans Blue visualization as red fluorescence on adjacent sections. Lesions outlined in white. (H) Lesion images from Ptch +/À mice treated with Ptx (n = 13 lesions) or PBS (n = 13 lesions) as above, immunostained for ColIV to quantify the proportions of non-vascularized (outlined in brown) and vascularized (outlined in green) lesions in each group. (I) Summary of Ptch +/À lesion number and vessel parameters upon treatment with aFzd4 or Ptx. Scale bars, 100 mm. See also Wnt-MB, where tumor cells secrete Wnt inhibitors to disrupt BBB stability (Phoenix et al., 2016).
Here we have extended the concept of tumor-stroma crosstalk via Wnt signalling to the earliest stages of tumorigenesis in the brain.  Norrin/Fzd4 signalling and human tumor progression Mutations in Norrin/Fzd4 pathway components cause a spectrum of human pathologies, including vascular disorders with retinal and cerebellar phenotypes (Gilmour, 2015;Liu et al., 2010;Romaniello et al., 2013). However, the role of this signalling axis in the vasculature has not been explored in human tumorigenesis. Based on our data, one might predict an inverse relationship between NDP levels and outcome in human Shh-MB (such as earlier age of onset or reduced survival). While we did observe a trend towards reduced survival in Shh-MB patients with the lowest levels of NDP transcript, NDP mRNA levels are generally high in human Shh-MB, suggesting a potential protective role for this pathway. However, transcript abundance may not be a universal indicator of Norrin pathway activation. Notably, the activities of several angiogenic factors, including VEGF and FGF, are regulated at the level of protein bioavailability (Hynes, 2009), requiring release from the ECM to mediate angiogenesis in pre-tumor lesions (Bergers et al., 2000;Nozawa et al., 2006). Interestingly, Norrin also associates with the ECM (Perez-Vilar and Hill, 1997). We also note that, given the clonal evolution patterns resulting in tumor heterogeneity within established mouse and human Shh-MB (Wu et al., 2012), early initiation events will not necessarily be reflected in expression profiles of the evolved tumors in patient tumor banks. Nonetheless, given the known role of Wnt-mediated neural/endothelial crosstalk in CNS vascular development and pathology (Engelhardt and Liebner, 2014;Phoenix et al., 2016;Ye et al., 2010b), our work highlights the importance of further examining the Norrin/Fzd4 pathway in the human MB progression, including a potential prophylactic role in susceptible individuals.

The preneoplastic niche and Ptch +/À MB initiation
Ptch +/À MB incidence and latency can be altered by modulating several pathways involved in proliferation and DNA damage (Ayrault et al., 2009;Briggs et al., 2008;Uziel et al., 2005;Wetmore et al., 2001); however there are few factors known to conclusively impact the number of preneoplastic lesions or their transition to malignancy. Ptch +/À lesion induction is known to be mediated by growth or differentiation regulators [Igf1, Ccnd1, Tis21 (Farioli-Vecchioli et al., 2012;Pogoriler et al., 2006;Tanori et al., 2010)] and by radiation or interference with DNA repair (Malek et al., 2011;Pazzaglia et al., 2006a;2006b), whereas progression of lesions to malignancy is enhanced by the growth promoting factors Igf2, Mycn, Shh co-receptor Boc and by the evasion of p53-induced senescence Kessler et al., 2009;Mille et al., 2014;Tamayo-Orrego et al., 2016). Here, we introduce the concept that a permissive preneoplastic microenvironment will significantly enhance Ptch +/À lesion induction and tumor initiation. We have linked the protumor stroma induced by loss of Norrin/Fzd4 signalling to increased GNP proliferation and accelerated Ptch LOH, both events that are known to increase the incidence and decrease latency of Ptch +/À MB (Ayrault et al., 2009;Mille et al., 2014;Pazzaglia et al., 2006a;2006b;Uziel et al., 2005). These data add to accumulating evidence that development of an oncogenic niche supports the growth and transformation of preneoplastic cells (Barcellos-Hoff et al., 2013). Although beyond the scope of our study to investigate in detail, the persistence of stromal changes in Ndp KO ;Ptch +/À established tumors highlights the relevance of examining the ongoing role of tumor-stroma crosstalk in the growth of MB.

Blood vessels as mediators of tumorigenesis in the Ptch +/À cerebellum
We have identified several Norrin/Fzd4-dependent stromal abnormalities associated with MB initiation, and our data suggest that disrupting the paracellular barrier property of the endothelium alone is not causal. Similarly, a recent study showed that modulation of Wnt-mediated BBB characteristics in orthotopic transplants of established mouse MB did not impact the tumor incidence or survival (Phoenix et al., 2016). We recognize that Ptch +/À vascular endothelial cells in the germ-line Ptch +/À model may exhibit elevated Hh pathway activity; however this is a positive regulator of BBB formation (Alvarez et al., 2011). Furthermore, our data showing dramatic acceleration of tumor formation upon deletion of Ndp from the Neurod2-Smo A1 MB model demonstrate that Norrin-dependent effects on tumorigenesis are independent of Ptch status in the endothelium. We also note that while Ptx treatment disrupted junctional protein expression in the cerebellar endothelium, this was not accompanied by induction of PLVAP, therefore we cannot rule out transcellular permeability as a possible feature of the pro-tumor stroma, or the possibility that the severity of BBB disruption is important.
Our work suggests that attenuation of Wnt signaling in endothelial cells and vascular invasion are endogenous factors regulating Ptch +/À lesion progression. These data are consistent with the finding that, although Norrin activity is not fully absent in Ptch +/À lesions (indicated by a largely intact BBB) Ndp expression is downregulated as a function of tumorigenesis. The exact mechanism by which endothelial cells deficient in Norrin/Fzd4 signalling could alter the Ptch +/À preneoplastic niche and affect GNPs remains to be determined. In Tie2Cre+;Fzd4 fl/fl mutants, we detected rare foci of disrupted morphology involving EGL-associated vessels, suggesting that surface vasculature adjacent to the EGL is particularly susceptible to Norrin-mediated instability. Thus, the interaction between Ptch +/À GNPs and endothelial cells lacking Norrin/Fzd4 pathway activation may provide a unique setting for lesion growth and vascular invasion. Interestingly, previous work analyzing the transcriptomes of Fzd4 KO versus Fzd4 WT endothelial cells isolated from the adult mouse retina and P16 cerebellum  showed that Fzd4-deficient endothelial cells exhibit transcript enrichment for genes known to promote both GNP growth (Igf1) and angiogenesis (Angpt2 and Esm1, both of which were upregulated in Ndp KO ;Ptch +/À versus Ptch +/À MB). However, we have not ruled out the possibility that inflammation is involved. The Tie2Cre driver used to delete Fzd4 is also expressed in hematopoietic lineages (Tang et al., 2010), which, combined with our observation of leukocyte accumulation in Ndp/Fzd4-deficient lesions, leaves open the prospect of an immune cell-mediated component within the pro-tumor stroma.

The permissive preneoplastic niche: A recurring feature of multistage tumor models
In tumors with defined stages of progression, such as pancreatic and breast cancer, the interaction between preneoplastic cells and their surrounding stroma is recognized as a critical modulator of tumor evolution (Coussens et al., 1996;Gullino, 1978;Hanahan and Folkman, 1996;Lin et al., 2006). Mouse models of these tumors have shown that the angiogenic switch, in particular, occurs prior to malignancy and may be a rate limiting step for tumor initiation (Giraudo et al., 2004;Lin et al., 2006;Smith-McCune et al., 1997). Here, we examined cerebellar lesions at the earliest stage that they are consistently detected in the Ptch +/À cerebellum, P14. At this stage, although we detected a modest increase in proliferation and accelerated Ptch LOH in Ndp KO ;Ptch +/À mutants, we did not find that disruption of Norrin/Fzd4 activity was associated with an increase in lesion volume. These data reflect previous studies in pancreatic and breast tumor models, where the angiogenic switch was not correlated with size in early-stage lesions, but was instead associated with, and required for, the transition from preneoplasia to malignancy (Folkman et al., 1989;Lin et al., 2006). Following this transition, transformed lesions will clearly have a growth advantage. The mechanism by which blood vessels in the pre-tumor microenvironment can promote transformation is not defined. Undoubtedly, further delineating the molecular events mediating crosstalk between preneoplastic cells and stromal elements will be important for understanding tumorigenesis in multiple cancers.

Materials and methods
Mice All experiments were approved by the University of Ottawa Animal Care Ethics Committee and adhered to the guidelines of the Canadian Council on Animal Care (CCAC). Ndp KO mice (RRID:MGI: 4414648), generated by disruption of the Ndp locus by a lacZ-containing cassette were obtained from Lexicon Pharmaceuticals (Junge et al., 2009) and maintained by interbreeding on a mixed background. Ndp is an X-linked gene, therefore Ndp KO males and females have Ndp À/y and Ndp À/À genotypes, respectively. Ndp À/y males were used for lacZ-based reporter expression analysis in Figure 1. Ndp À/À females are infertile, therefore the experimental cross to generate Ndp KO ;Ptch +/À compound mutants and littermate controls must be performed by crossing Ndp +/À females with Ptch +/À males. Thus, the Ndp KO and Ndp KO ;Ptch +/À genotypes are restricted to male mice carrying the Ndp À/y allele, whereas additional controls include both sexes. Ptch +/À (RRID:MGI:2177702), Tie2-Cre (RRID:IMSR_JAX:008863), Atoh1Cre (RRID:IMSR_JAX:011104), and Fzd4 fl/fl (RRID:MGI: 4412187), mice were obtained from Jackson Laboratories and maintained on a C57BL/6 background. Tie2Cre+;Fzd4 fl/fl females are infertile, therefore Tie2Cre+;Fzd4 fl/fl ;Ptch +/À compound mutants were generated by crossing Fzd4 fl/fl ;Ptch +/À females with Tie2Cre+;Fzd4 fl/fl males. Neurod2-Smo A1 mice (Hallahan et al., 2004) were maintained as homozygotes (RRID:MGI:3831004), crossed with Ndp +/À females, and tumors were monitored in Neurod2-Smo A1+/À ;Ndp -/y and Neurod2-Smo A1+/À ;Ndp +/y littermates. In every experiment, all compound mutants were compared to single mutant or wildtype controls from the same breeding cohort to ensure matched backgrounds. In Kaplan-Meier survival curve studies, mice were continually monitored and sacrificed upon display of advanced tumor symptoms or other adverse health effects as per CCAC endpoint guidelines. All animals were dissected to confirm presence or absence of medulloblastoma.

Antibodies
The following primary antibodies were used for immunostaining:

Tissue processing
For fixed tissue, pups younger than 14 days old were decapitated and brains were removed and placed directly into fixative. Animals 14 days and older were anesthetized and cardiac perfusion was performed using 10 ml of PBS then 10 ml of fixative, followed by dissection of the brain. Brains used for histological stains, Evans Blue visualization, in situ hybridization or immunostaining were then fixed in 4% paraformaldeyhyde (PFA) overnight at 4˚C, whereas brains used for X-gal staining were fixed in 2% PFA with 2 mM MgCl 2 and 1.25 mM EGTA (ethylene glycol tetraacetic acid) for 45 min at 4˚C. All tissues were then washed in PBS, cryoprotected in 30% sucrose/PBS overnight at 4˚C, and embedded in a 50:50 mixture of OCT:30% sucrose by freezing in chilled 2-methylbutane. For fresh frozen tissue, unfixed brains were dissected and embedded as described above. If used for immunostaining, cardiac perfusion was performed using 10 ml of cold PBS before dissection. For laser capture microdissection, tissue was immediately dissected and embedded without perfusion, for maximum RNA integrity.

Granule neuron progenitor (GNP) isolation
GNPs from the EGL or lesion-associated GNPs from P14 and P30 Ptch +/À mice were purified from cerebella by percoll gradient separation and pre-plating, as described previously (Oliver et al., 2005). Cells were then transferred to PDL-coated glass coverslips to proceed with immunostaining, or pelleted and immediately resuspended in lysis buffer for subsequent RNA extraction.
Immunostaining PFA-fixed brain tissue (immunostains in Figure 1, Figure 3 top row, Figures 7 and 8) Brains were sectioned sagittally in a Leica CM1850 cyrostat at 12 mm onto Superfrost Plus positively charged slides (Fisher Scientific), air dried for 1-2 hr, and stored with desiccant at À20˚C. Prior to immunostaining, antigen retrieval was performed in 10 mM sodium citrate buffer (pH 6) in a rice steamer. Slides were blocked with 10% normal serum in TBLS (tris-buffered saline containing 1% bovine serum albumin and 10 mM lysine) for 30 min at room temperature, and antibodies were diluted in TBLS and incubated overnight at 4˚C. Sections were incubated with Alexa fluor secondary antibodies (Molecular Probes) at 1:1000 for 1 hr at room temperature, and nuclei were stained with Hoescht before coverslipping with fluorescence mounting medium (Dako S3023). For co-immunostains, both primaries and both secondaries were incubated simultaneously. The following modifications were performed for peroxidase-based immunohistochemistry (IHC): Before blocking, endogenous peroxidases were quenched by incubating slides in 0.3% hydrogen peroxide in PBS for 15 min at room temperature. Biotinylated secondary antibodies were used at 1:200 (Dako), followed by incubation with avidin-biotin peroxidase (Vector Laboratories, PK-4000) for 40 min at room temperature, color development with 2.5% diaminobenzidine, and a brief counterstain with hematoxylin before mounting in 50:50 glycerol:PBS.
Acetone-fixed brain sections (immunostains in Figure 3B, Figure  Plus positively charged slides (Fisher Scientific), air dried for 1-2 hr, and slides were dipped in acetone for 20 s prior to storage at À80 until staining. Before staining, slides were fixed in ice cold acetone for 10 min, washed in ice cold 70% ethanol for 5 min followed by several washes in room temperature PBS, and blocked with 10% normal serum in PBS. Primary antibodies were diluted in 3% normal serum and incubated overnight at 4˚C, followed by secondary detection and mounting as described above. For the Pax6/CD31/EdU triple stain, (Figure 8-figure supplement 1), the EdU was detected immediately following Pax6/CD31 immunostaining, using the Molecular Probes Click-iT EdU Alexa Fluor 647 Imaging Kit according to manufacturer's instructions.

GNP immunostaining (Figure 2)
Once purified, GNPs were allowed to sit on PDL-coated glass coverslips for 2 hr in culture media, followed by the addition of 10 mM of anti-Fzd4 or anti-KLH antibodies to the media for 15 min, washing in PBS, fixation in 4% PFA, and secondary detection and mounting as described above.
X-gal staining ( Figure 1) Slides with 12 mm cryosections of 2% PFA-fixed brains were air-dried for 1 hr, placed in X-gal reaction buffer (1 mg/ml X-gal with 5 mM potassium ferrocyanide, 5 mM potassium ferricyanide, 2 mM MgCl 2 , 0.02% IGEPAL, 0.01% sodium deoxycholate in 0.1 M phosphate buffer), incubated overnight at 37˚C, then washed and mounted with 50:50 glycerol:PBS. For co-X-gal/IHC stains, IHC was performed as described above, immediately after incubation in X-gal reaction buffer. For each X-gal stain or X-gal/IHC co-stain, at least 4 separate sections from n = 3 cerebella were examined.
In situ hybridization ( Figure 3) Digoxigenin (DIG)-labeled antisense RNA riboprobes were prepared by in vitrotranscription from linearized plasmids containing complete or partial cDNA sequences of the following mouse genes: Atoh1 (a gift from Dr. Carol Schuurmans), Mycn, and Ccnd1. ISH was performed as previously described (Jensen and Wallace, 1997), with the following modifications: Slides were incubated 1 to 5 hr in staining buffer containing NBT and BCIP, and slides were mounted in 50:50 glycerol:PBS. For each probe, at least 3 separate sections from n = 3 tumors were examined.

Image acquisition
Brightfield images were visualized using an Axioplan microscope and captured with an Axiocam HRc camera. Fluorescent images were visualized using an AxioImager M1 microscope and captured with an Axiocam HRm camera, or a Zeiss AxioImager M2 microscope and Axiocam MRm camera. For the analysis of CD31/PH3 expression in Figure 5, CD31/Lef1 expression in Figure 6-figure supplement 2, and Pax6/CD31/EdU expression in Figure 8-figure supplement 1, images were captured using an LSM 780 confocal. Retinal whole-mount images were captured using an LSM 510 confocal. (All equipment from Carl Zeiss Inc.). Whole brain images of were captured using a Canon PowerShot SD1400 IS digital camera. Images were processed using Photoshop CS6 (Adobe).

Fzd4 blocking antibody, pertussis toxin (Ptx) and EdU administration
Anti-Fzd4 and anti-KLH control antibodies were generated as described (Paes et al., 2011), and functionally tested in vivo before use (Figure 7-figure supplement 1). Antibodies were diluted in PBS to 3 mg/ml immediately prior to administration, and injected intraperitoneally at a dose of 30 mg/kg. Ptch +/À animals in lesion studies were injected once at P7, whereas Ptch +/À animals in the Kaplan-Meier survival study were injected at P7 followed by a booster dose at P16, an age when the phenotypic effects of a P7 injection are still present (Paes et al., 2011). Ptx was diluted in PBS to 5 mg/ml and pups received intraperitoneal injections of 120 ng Ptx or PBS vehicle control at P7, P9, P11 and P13. Animals were injected intraperitoneally with a dose of 10 mg/kg EdU 4 hr prior to sacrifice.

EGL/lesion measurements and analysis of immunostains
For EGL/lesion measurements, sagittal 12 mm serial sections of cerebella were collected and examined along the mediolateral axis at intervals 144 mm apart, by hematoxylin and eosin (H and E) or cresyl violet staining, and blinded images at 5x magnification were captured. To quantify EGL thickness, three images (at identical locations at the cerebellar vermis between lobules VII and VIII) from n = 3 animals per genotype were measured. To quantify lesion number or volume, lesions were carefully followed continuously along the entire mediolateral axis, and scored as an individual lesion only if they remained spatially separate from all other lesions in every section. Lesion volume (mm 3 ) was calculated by measuring the 2-D area (mm 2 ) of each lesion section using ImageJ, multiplying it by the thickness separating each section from its neighbor (0.144 mm) to obtain the volume of each slice, and adding the individual slice volumes to obtain a total volume. To assess the vascularized, leaky vessel or CD45 accumulation status of the lesions, sagittal sections between 100 and 250 mm apart were examined along the entire mediolateral axis for Evans blue accumulation or stained with markers (anti-ColIV, laminin, CD31, PLVAP or CD45) to sample the entire lesion. To quantify lesions for PH3, cleaved caspase-3, NeuN, CD31, laminin or PLVAP immunostaining, or to examine PDGFRb immunostaining, 3 to 4 blinded sections from each lesion at 10x or 20x magnification were analyzed. Using ImageJ, the number of PH3+ cells per unit area or percent area stained for caspase-3, NeuN, CD31, laminin or PLVAP were determined. Quantification involving lesion vasculature included all lesion-associated vessels, whether they were 1) on the outer surface of the cerebellum, 2) deeper into the cerebellar folds but still meningeal, or 3) growing into the 'lesion proper.' To quantify the number of PH3+ or Lef1+ cells per endothelial area, 3 sections per lesion were first imaged at 20x magnification by epifluorescence to measure endothelial cell area via tracing of CD31+ vessels, and identify potential double labelled candidates. Sections were then examined by confocal microscopy to confirm double-labelled cells.

Evans Blue injections and visualization
A 2% wt per volume Evans Blue (Sigma) solution in 0.9% saline was administered by intraperitoneal injection at 4 ml per gram of body weight and allowed to circulate overnight. Following perfusion and fixation in 4% PFA as described above, whole brains were photographed and 12 mm cryosections were then visualized under the far red fluorescence filter. The same or immediately adjacent section was H and E-or anti-ColIV-stained to provide a matched image.

Laser capture microdissection
Fresh frozen cerebella were sectioned at 10 mm onto Superfrost microscope slides (Fisher Scientific) and placed immediately on dry ice before storage at À80˚C for no more than 5 days before microdissection. Sections were stained and dehydrated by passing through RNase-free coplin jars with solutions made in DEPC-treated distilled water, as follows: 30 s in 75% ethanol, 30 s in distilled water, 2 min in 1% toluidine blue in distilled water, 30 s in distilled water, 30 s in 75% ethanol, 30 s in 95% ethanol, 1 min in 100% ethanol (2 times), and 5 min in xylene. All staining solutions except xylene contained RNase inhibitor (Sigma R7397). Slides were immediately microdissected using the Arcturus XT laser capture microdissection system (Life Technologies) in infrared mode, according to the manufacturer's instructions. Six to 10 sections from each lesion were captured onto CapSure macro LCM caps (Life Technologies), transferred immediately to RLT plus lysis buffer (Qiagen) with b-mercaptoethanol, briefly vortexed, and stored on dry ice until RNA extraction.

RNA purification and quantitative RT-PCR
For microdissected lesion tissue, total RNA was extracted using the RNeasy Plus Micro kit (Qiagen) with genomic DNA eliminator columns, and amplified complementary DNA (cDNA) was prepared with the Ovation Pico WTA System V2 (NuGEN) according to the manufacturer's instructions. RNA integrity was assessed by an Agilent 2100 Bioanalyzer (Agilent Technologies) from slide scrapes. For all other samples, total RNA was extracted from freshly isolated/dissected GNPs, cerebellar tumor (with careful preservation of clean margins) or retina tissue using the RNeasy Mini Kit (Qiagen). Firststrand cDNA was synthesized using the QuantiTect Reverse Transcription Kit (Qiagen), with and without reverse transcriptase to assess genomic contamination during downstream RT-PCR. For all samples, target gene mRNA levels were determined by quantitative RT-PCR (qRT-PCR) using iQ SYBR Green Supermix (Bio-Rad) and a MyiQ iCycler (Bio-Rad). Primer pairs were designed using PRIMER-blast (http://www.ncbi.nlm.nih.gov/), and are as follows: Gapdh Primers were optimized using a 5-point standard curve of 2-fold diluted composite cDNA from relevant tissue and deemed acceptable with an R 2 > 0.95, a percent efficiency between 90-110%, a sharp single point melt curve, positive controls with Ct values > 10 cycle difference compared to no RT control samples, and expected amplicon size by agarose gel electrophoresis. All samples were run in triplicate, normalized to Gapdh, and quantified relative to the reference tissue indicated. During qRT-PCR for microdissected lesions, the expression of genes known to be highly expressed in lesions (Gli1, Mycn) were assessed in parallel to Ptch, along with microdissected tumor samples known to have Ptch LOH, or microdissected EGL known to have high levels of Ptch expression. Ptch LOH status was assigned by quantification relative to microdissected tumors with known LOH.

Microarray analysis of mouse tissue
Total RNA extracted from MB or GNPs was analyzed with the MouseWG-6 v2.0 Expression Bead-Chip array platform. Illumina BeadStudio outputs were analyzed and annotated with R packages limma (Ritchie et al., 2015) and illuminaMousev2.db version 1.26.0. Data were processed with neqc function (Shi et al., 2010). The hierarchical clustering and the principal component analysis plots were prepared from the 1500 most variable probes across all samples in terms of interquartile range, after data processing (background correction, normalization, and log-transformation), and filtering out probes annotated as Bad and No Match. Significant changing probes were detected with the linear modeling approach and empirical Bayes statistics of limma (Smyth, 2004). Gene Ontology (GO) cellular components enrichment was investigated using DAVID (Huang et al., 2009) for probes with adjusted P value below 0.05 and higher fold changes (>1.0) between Ndp KO ;Ptch +/À and Ptch +/À tumors. The tables in Figure 3 display the GO terms with P values<0.05. In an analogous manner we analyzed samples from purified P6 GNPs. The principal component analysis plot was prepared from the 1,500 most variable probes as above.

Human tumor samples and expression analysis
NDP expression profiles were determined across three independent cohorts with the R2 database analysis tool (http://r2.amc.nl) using publically available datasets from Heidelberg (Remke et al., 2011), Toronto  and Boston (Cho et al., 2011). The following probes were used for analysis: Toronto (Affymetrix Exon 1.0 T Probe Accession: 4006280), Heidelberg (Agilent 4 x 44 k Probe Accession: A_23_P73609) and Boston (Affymetrix u133a Probe Accession: 206022_at). P values represent ANOVA across the four subgroups. Survival analysis in Figure 1 was performed on the MAGIC dataset of clinically annotated SHH tumors (Affymetrix Human Gene 1.1 ST, Probe Accession: 8172220) Vanner et al., 2014). Log 2 transformed expression values of NDP were then ranked as either the bottom 10th percentile of expression versus the top 10th percentile. Survival was calculated using the log-rank method in the R-Statistical Environment (v3.1.3) using packages survival (v2.37-7) and ggplot2 (v1.0.0).

Statistics
Sample sizes are as reported in figures and figure captions. Mice were randomly assigned to either aFzd4 or Ptx treatment groups. Statistics were determined by GraphPad Prism 6 or SPSS software. The logrank test was used to generate Kaplan Meier survival curves (Figures 1, 2, 7 and 8). Lesion number (Figures 4 and 7), vessel or laminin density ( Figures 5 and 6), number of endothelial PH3+ cells ( Figure 5), qRT-PCR (Figures 1 -and 3) and EGL thickness (Figure 4-figure supplement 1) were analyzed by one-way ANOVA with Tukey post-hoc comparisons. Comparison of lesion volumes (Figures 1, 4 and 7), immunostain quantification (Figure 8), and tumor latency (Figures 2, 7 and 8) were analyzed by a two-tailed, unpaired Student's t-test. The proportion of vascularized, leaky or CD45+ lesions (Figures 5 and 6) and tumor incidence (Figures 2, 7 and 8) were analyzed by the hypergeometric test. The number of endothelial Lef1+ nuclei ( Figure 6-figure supplement 2) was assessed by a one-way ANOVA and Fischer's LSD post hoc comparison. A Levene's test for homogeneity of variance and normality tests were used to verify parameters for parametric analysis. Microarray analyses were analyzed as described above.