Cross-species epigenetics identifies a critical role for VAV1 in SHH subgroup medulloblastoma maintenance

Abstract

The identification of key tumorigenic events in Sonic Hedgehog subgroup medulloblastomas (MB_SHH) will be essential for the development of individualized therapies and improved outcomes. However, beyond confirmation of characteristic SHH-pathway mutations, recent genome-wide sequencing studies have not revealed commonly-mutated genes with widespread relevance as potential therapeutic targets. We therefore examined any role for epigenetic DNA methylation events in MB_SHH using a cross-species approach to candidate identification, prioritization and validation. MB_SHH–associated DNA methylation events were first identified in 216 subgrouped human medulloblastomas (50 MB_SHH, 28 WNT_, 44 Group 3, 94 Group 4) and their conservation then assessed in tumors arising from four independent murine models of Shh medulloblastoma, alongside any role in tumorigenesis using functional assessments in mouse and human models. This strategy identified widespread regional CpG hypo-methylation of VAV1, leading to its elevated expression, as a conserved aberrant epigenetic event which characterizes the majority of MB_SHH tumors in both species, and is associated with a poor outcome in MB_SHH patients. Moreover, direct modulation of VAV1 in mouse and human models revealed a critical role in tumor maintenance, and its abrogation markedly reduced medulloblastoma growth. Further, Vav1 activity regulated granule neuron precursor (GNP) germinal zone exit and migration initiation in an ex vivo model of early post-natal cerebellar development. These findings establish VAV1 as a critical epigenetically-regulated oncogene with a key role in MB_SHH maintenance, and highlight its potential as a validated therapeutic target and prognostic biomarker for the improved therapy of medulloblastoma.

Introduction

Medulloblastoma, the most common malignant brain tumor of childhood, comprises four consensus molecular subtypes [Wnt/Wingless (WNT), Sonic Hedgehog (SHH), Group 3 and Group 4] characterized by distinct clinical, molecular and pathological features, and by different postulated cells of origin.^1-4 These advances are rapidly informing targeted therapeutic strategies and the design of individualized clinical trials aimed at increased cure rates and reduced long-term treatment side-effects.⁵

Targeted therapies for MB_SHH (~25% of patients) are currently most advanced. MB_SHH (~70% five-year survival) peak in infancy, but also occur throughout childhood and into adulthood.² Tumors are characterized by SHH signaling activation via mechanisms including pathway mutations (PTCH1 (~45% of MB_SHH), SUFU (~8%) and SMO (~14%)) and gene amplifications (GLI2 (~8%), MYCN (~13%)).⁶ Early clinical trials of SHH pathway inhibitors (targeting SMO) show promise,^{7, 8} but significant challenges are predicted in their application, with reported developmental toxicities, ⁹ acquired drug resistance, ¹⁰ and intrinsic insensitivity in tumors with pathway activation downstream of SMO (e.g. SUFU mutation, GLI2 and MYCN amplification).⁶ Thus, there is clear need to identify additional targetable genes and pathways in MB_SHH, to support additional or alternative targeted therapeutic strategies.

The identification of therapeutically exploitable ‘driver’ events from the many aberrations identified in contemporary genome-wide genetic and epigenetic investigations represents a major challenge. Recent genome-wide mutational screens in MB_SHH have revealed notable genetic diversity, however the most frequent novel coding mutations characterize small patient groups (e.g. MLL2 (~16% of MB_SHH) and DDX3X (~26%)) and most mutations are found singly in individual tumors.^{6, 11} Significantly, a prominent role for epigenetic alterations is also emerging.^{6, 12-14} Our recent DNA methylation profiling study of 230 medulloblastomas showed MB_SHH harbor specific DNA methylation events which characterize significant proportions of tumors and identify candidate disease driver genes.¹⁵ However, their functional relevance and contribution to tumor development is unknown.

Genetically-engineered mouse models, which give rise to medulloblastoma with constitutive SHH pathway activation, support granule neuron precursors (GNPs) of the developing cerebellum as their cells of origin^{4, 16-19} and provide opportunities to systematically interrogate epigenetic events in the mouse medulloblastoma genome. It may be reasoned that any conservation of aberrant DNA methylation events between MB_SHH and SHH-driven mouse medulloblastomas could identify events important in tumorigenesis and provide faithful model systems for their functional investigation.

Here we report a cross-species investigation of DNA methylation events in medulloblastomas from 230 human patients (216 with subgroup assigned (50 MB_SHH, 28 WNT_, 44 Group 3, 94 Group 4)) and four independent SHH mouse models. Using this approach, we identify regional hypo-methylation of VAV1, leading to its elevated expression, as a conserved and widespread epigenetic event in both species, affecting the majority of MB_SHH tumors. We further show (i) VAV1 status predicts poor treatment outcome within the MB_SHH subgroup and (ii) VAV1 plays a critical functional role in medulloblastoma maintenance, highlighting its clear potential as a biomarker and validated drug target in MB_SHH. Finally, we provide evidence that Vav1 is expressed in early cerebellar GNPs, where its activity regulates their development. The roles characterized for VAV1 in MB_SHH, and in cerebellar development, demonstrate the potential of comparative tumor epigenomics using genetically-engineered mice to identify, prioritize and functionally analyze critical molecular events in human tumorigenesis.

Results

Human MB_SHH tumors harbor distinct and characteristic DNA methylation events

We first sought to identify DNA methylation events which may contribute to human MB_SHH. We have previously shown the four major medulloblastoma molecular disease subgroups -WNT, SHH, Group3 and Group 4 - may be distinguished based on their DNA methylation patterns (Supplementary Figure S1A).¹⁵ Comparison of tumor DNA methylation patterns of subgrouped medulloblastomas (n=216) with the normal cerebellum (n=21) showed normal cerebella form a distinct group, which is most closely related to MB_SHH (Supplementary Figure S1B).

Statistical analysis identified 15 genes (associated with 18 CpG sites) whose methylation status significantly distinguished MB_SHH from other medulloblastomas, and from the normal cerebellum (Figure 1A). Both hypo- (n=6) and hyper-methylation (n=9) events were observed. In addition, the methylation status of RASSF1A and COL1A2 CpG sites, two of the most frequent hyper-methylation events reported in medulloblastoma,^{12, 20} was assessed and showed expected frequent and tumor-specific methylation which affected all tumor subgroups (Figure 1A). In summary, these investigations show MB_SHH harbor distinct DNA methylation patterns and identify a series of gene-specific DNA methylation events for further analysis.

Figure 1. Identification of DNA methylation events associated with human MB_SHH and their analysis in mouse tumors and cerebella identify VAV1/Vav1 hypo-methylation as a conserved cross-species epigenetic event in SHH-associated medulloblastomas.

A. Heatmap showing the most differentially methylated CpG sites between human primary MB_SHH tumors (n=50) and (i) non-MB_SHH tumors (n=166) and (ii) cerebellar samples (n=21). Heatmap shows methylation scores (green (unmethylated) to red (fully-methylated)). All CpG sites selected showed a statistically significant difference in average methylation status between groups (mean β difference >0.34 and adjusted p<0.05 following correction for multiple testing). B. and C. The DNA methylation status (methylation score) of orthologous regions in the mouse and human genomes (mean, based on all CpG sites) was calculated for each gene of interest by bisulfite sequencing of 12 spontaneous medulloblastomas arising from four independent Shh-associated mouse models (genotypes indicated) and 11 strain-matched normal cerebella, in comparison with 6 MB_SHH human tumors and 6 human cerebella. B. Heatmap shows methylation scores, from green (unmethylated) to red (fully-methylated) for individual mouse tumors and cerebella. ‘p’ indicates significance following Mann-Whitney tests. The difference between the average DNA methylation score for mouse cerebella, and the average methylation score for mouse tumor samples, is shown. C. Heatmap comparing average methylation scores in all analyzed tumors (T) and cerebella (C) from mice and humans. Conservation of overall methylation status between species (average methylation score difference ≤0.2) is indicated (Y, Yes; N, No).

Identification of conserved DNA methylation events in mouse models of SHH-driven medulloblastoma

We next assessed the DNA methylation status of regions orthologous to those encompassing the CpG sites identified in human MB_SHH tumors, in medulloblastomas from four independent SHH-associated mouse models, alongside strain-matched normal cerebella of different developmental stages. Eight orthologous regions surrounding CpG sites which showed MB_SHH-specific methylation, alongside regions surrounding RASSF1A and COL1A2, were selected for analysis by direct bisulfite sequencing (Figure1, Supplementary Table S1 and Figure S2A, B). The overall cerebellar methylation status was conserved in the majority of regions assessed (7/10; Figure 1C); cerebella from different mouse strains, including immature and mature mice, showed equivalent patterns (Figure 1B). Consistent methylation profiles were also observed across tumors from the different SHH-associated mouse models (Figure 1B).

Most strikingly, VAV1 showed widespread tumor-specific patterns of hypo-methylation across the regions examined which were of equivalent extent, magnitude and frequency in human MB_SHH and SHH mouse tumor models, compared to cerebellar controls. In all mouse tumors, the majority of CpG sites across the Vav1 region were hypo-methylated (Figure 1B, C; Supplementary Figure S2C), while in humans, VAV1 hypo-methylation encompassed all assessed CpG sites (Supplementary Figures S2B and S3A, B, C). Tal1, Dsc2, Spdef and Rassf1a also showed statistically significant methylation differences, but these represented methylation changes in individual samples or CpG sites and were not comparable to differences seen between human MB_SHH tumors and cerebella (Figure 1B, C). These findings identify regional VAV1 hypo-methylation as a frequent evolutionarily conserved epigenetic event in human MB_SHH, for further analysis.

VAV1 hypo-methylation characterizes the majority of MB_SHH and is associated with a poor outcome

VAV1 hypo-methylation was a frequent region-wide event in human MB_SHH (70% (36/50)), affecting only occasional tumors (2% (4/166)) within the other molecular subgroups (Figure 2A, B; Supplementary Figure S3C). In whole-cohort analysis (n=230), VAV1 hypo-methylation was significantly correlated with desmoplastic nodular (DN) pathology (p=2.3×10⁻¹¹, Fisher’s exact test), infant patients (<3 years at diagnosis; p=0.003)) and MYCN gene amplification (p=0.02) (Figure 2C), consistent with the established associations of these features with MB_SHH in this cohort ¹⁵.

Figure 2. VAV1 hypo-methylation in MB_SHH: Incidence, regional distribution, and association with clinical, pathological and molecular disease features.

A. VAV1 methylation levels in our cohort of 216 primary medulloblastomas with subgroup classification. Box plots show methylation levels (β-value, probe E9) in the MB_SHH subgroup (n=50) compared to non-MB_SHH tumors (n=166) and normal cerebella (CB; n=21). The cerebellar methylation distribution (mean-3SD; cut-off marked) was used to distinguish hypo-methylated tumors (β<0.53) for subsequent assessment of clinical, pathological and molecular associations. ‘p’ value, Kruskal-Wallis test. non-MB_SHH tumors are shown colored by subgroup (Group 3, yellow; WNT, blue; Group 4, green). B. Methylation levels at adjacent CpG sites to probe E9, following analysis by bisulfite sequencing in representative MB_SHH, non-MB_SHH tumors and normal cerebella samples showing region-wide hypomethylation of CpG sites in the majority of MB_SHH tumors. CpG sites are labelled by their position relative to the translational start site of VAV1. Black circles, ≥80% methylation; three-quarter black circles, ≥60%; half-black circles, ≥40%; quarter-black circles, ≥20% methylation; white circles, <20% methylation. C. Clinical, pathological and molecular features of VAV1 methylated and hypo-methylated tumors. ‘p’ values, X² or Fisher’s exact test, as appropriate. D. Kaplan-Meier survival curves show event-free survival (EFS) for VAV1 hypo-methylated and methylated tumors (β-value, probe E9) in a sub-cohort of 36 primary MB_SHH medulloblastomas comprising patients with available survival information, aged ≥3 and <16 years old at diagnosis, treated with standard upfront radiotherapy and chemotherapy ¹⁵. ‘p’ value, log-rank test.

Within the MB_SHH subgroup, VAV1 hypo-methylation was observed across tumors of all key clinical, pathological and molecular demographics, and did not associate with any particular medulloblastoma disease feature; it is however notable that all MYCN amplifications occurred in VAV1 hypo-methylated tumors (Figure 2C). Associations were assessed using the CpG site recognized by the VAV1_e9 probe which lies within the region conserved between humans and mice. Notably, VAV1 hypo-methylation was associated with a poor outcome amongst MB_SHH tumors (Figure 2D), in an analysis of 36 patients aged ≥3 years at diagnosis, who received standard upfront radiotherapy and chemotherapy and were derived predominantly from the PNET3 clinical trial.^{15, 21} The rarity of VAV1 hypo-methylation in non-MB_SHH precluded the analysis of any relationships to survival in these subgroups. Thus, VAV1 hypo-methylation occurs frequently within the MB_SHH subgroup and is associated with specific disease features and poor outcomes.

VAV1 hypo-methylation is associated with elevated expression in human and mouse SHH medulloblastomas

VAV1 showed a direct methylation-dependent relationship between methylation status and expression in human medulloblastoma cell lines treated with the DNA methyltransferase inhibitor, 5′-Aza-2′-deoxycytidine (5-azaCdR) (Figure 3A, B). All VAV1-methylated cell lines (D283Med, MEB-Med8a and D425Med) had low VAV1 expression which was upregulated following 5-azaCdR de-methylation. In contrast, VAV1 expression was higher in hypo-methylated cells (DAOY) and was unaffected by 5-azaCdR. Consistent with this, Vav1 expression was significantly increased in mouse tumors (all hypo-methylated) compared to normal adult cerebella (all methylated) (Figure 3C; Supplementary Figure S2C), and in VAV1 hypo-methylated human tumors compared to VAV1 methylated tumors or the normal human cerebellum (all methylated) (Figure 3D). Finally, VAV1 expression was highest in MB_SHH tumors (associated with VAV1 hypo-methylation) compared to the other medulloblastoma molecular subgroups (Figure 3E; data from published dataset²²). These demonstrated relationships between VAV1 methylation and expression are consistent with the hypo-methylation-dependent elevation of VAV1 expression in human and mouse SHH-subgroup medulloblastomas.

Figure 3. VAV1/Vav1 hypo-methylation is associated with elevated expression.

A. RT-PCR analysis of VAV1 mRNA expression in human medulloblastoma cell lines before (−) and after (+) treatment with the DNA methyltransferase inhibitor, 5 AzaCdr. Mean VAV1 expression (+SE) is shown based on three independent replicates. B. DNA methylation levels at individual CpG sites in the assessed VAV1 region in human cell lines following analysis by bisulfite sequencing shown alongside the methylation score. Black circles, ≥80% methylation; three-quarter black circles, ≥60%; half-black circles, ≥40%; quarter-black circles, ≥20%; white circles, <20% methylation. C. RT-PCR analysis of Vav1 mRNA expression in Vav1 hypo-methylated mouse tumors (n=9; 3 each of Smo/Smo (blue circles), Ptch1^+/− Cdkn2c^−/− (light-blue squares) and Cdkn2c^−/− p53^FL/−, Nestin-Cre+ (grey-blue triangles) mice) and strain-matched Vav1 methylated normal mouse cerebella (CB; n=9; Smo/Smo matched strain, dark pink diamonds; Cdkn2c^−/− strains, pink squares). D. RT-PCR analysis of VAV1 methylated (n=22) and hypo-methylated (n=12) human tumors and methylated human normal cerebella (CB; n=5). VAV1 methylation status was determined as described in Figure 2. E. VAV1 mRNA expression in a cohort of 255 primary human medulloblastomas previously described by ²² containing no WNT subgroup tumors. D and E. Molecular subgroup status is shown for human medulloblastomas (MB_SHH, red; WNT, blue; Group 3, yellow; Group 4, green; non-classified, grey), and ‘p’ values (C, t-test; D, E, ANOVA) are indicated.

VAV1 expression is pro-proliferative in human medulloblastoma cells

To investigate the functional consequences of VAV1 hypo-methylation and expression in human medulloblastoma, we modelled the effect of silencing VAV1 over-expression in hypo-methylated human medulloblastoma cells. VAV1 silencing using siRNA (60-80% knockdown, day 3; Figure 4A, B) significantly decreased proliferation (Figure 4C) and altered cell cycle distribution; a pronounced G₂ arrest and increased G₀/sub-G₁ fraction occurred following VAV1 knockdown (Figure 4D). These data support a pro-proliferative role for the elevated VAV1 expression observed in VAV1 hypo-methylated human medulloblastoma cells.

Figure 4. VAV1 expression promotes proliferation in VAV1 hypo-methylated human medulloblastoma cells.

A,B. Consistent reduction of VAV1 expression (60-70%) in VAV1 hypo-methylated DAOY medulloblastoma cells (Figure 3) at 3 days post-transfection with 3 anti-VAV1 siRNA duplexes, compared to non-silencing (NS) siRNA. A. Western blot shows representative VAV1 knockdown and graph shows the mean reduction in VAV1 expression (+SE; based on three independent replicates) caused by anti-VAV1 siRNA (duplex 1) relative to NS siRNA controls, following normalization to actin protein levels. B. RT-PCR analysis of VAV1 mRNA expression at 3 days post-transfection with anti-VAV1 siRNA (duplexes 1-3). Graph shows mean reduction in VAV1 expression (+SE; based on four independent replicates) caused by VAV1 siRNA relative to NS siRNA controls, following normalization to the expression of a control gene (TBP). C. Cell proliferation following transfection with anti-VAV1 siRNA or NS siRNA, determined by XTT assay. Graphs show log₂ number of cells at 1 to 5 days post-transfection relative to day 0 (before addition of siRNA). Results shown represent 4 independent replicates (+ SE). D. Fluorescence-activated cell sorting (FACS) analysis of cell cycle distribution at 3 days post-transfection (duplex 1). Graph shows mean (+SE) percentage of cells in each cell cycle phase, based on 4 independent replicates. Representative cell cycle distribution plots are shown for each siRNA. All ‘p’ values shown represent paired t-tests.

Dominant-negative Vav1 suppresses medulloblastoma maintenance in the Ptch1^+/− Cdkn2c^−/− mouse model

To address whether Vav1 was required for tumor growth in a mouse SHH medulloblastoma model, we purified tumor cells from spontaneously occurring Vav1 hypomethylated/overexpressing medulloblastomas arising in Ptch1^+/− Cdkn2c^−/− mice (Figures 1,3C, Supplementary Figure S2C) and infected them with retroviruses carrying GFP alone, wild-type Vav1/GFP or a dominant-negative form of Vav1 (Vav1DN/GFP ^{23, 24}), an isolated amino terminal fragment (residues 1-186), which acts in trans to inhibit the GEF function of native Vav1. Cells from each infected pool were transplanted in the flanks of recipient nude mice and tumor formation assessed.

Approximately two months post-implantation, tumors appeared on the flanks of all mice transplanted with tumor cells marked with GFP only or with Vav1/GFP viruses. These tumors harbored similar weights (Figure 5C). In contrast, tumors that developed after the transplant of tumor cells infected with Vav1DN-carrying viruses were smaller than those derived from GFP only-infected cells and only 6 out of 9 transplants induced secondary tumors (Figure 5A-C). Thus, enforced expression of a Vav1 dominant-negative mutant inhibits tumor maintenance in our model. Ectopic over-expression of wild-type Vav1 did not impact, suggesting interference with the endogenous Vav1 expression associated with its hypo-methylation is important for the inhibition of tumor growth.

Figure 5. Inhibition of SHH tumor maintenance by a dominant-negative form of Vav1.

A-D. In each experiment Ptch1^+/− Cdkn2c^−/−mouse medulloblastoma cells infected with a GFP only virus were transplanted into the left side of recipient nude mice, whereas tumor cells infected with Vav1 or Vav1DN viruses were transplanted on the right side. Parallel FACS analysis revealed that approximately 20-40% tumor cells were infected. A. A representative mouse transplanted with tumor cells infected with GFP- (left) or Vav1DN/GFP- (right) carrying retroviruses. A significantly larger tumor appeared on the left side from GFP alone-infected cells (white arrow). B. Surgically removed tumors from a mouse in which tumors developed on both sides. Scale bar, 5mm. C. Quantification of the weight ratio between Vav1- or Vav1DN-expressing tumors compared to control GFP alone-expressing tumors. *p<0.05. Black circles represent transplanted mice that did not develop tumors. D. Contribution of either Vav1-or Vav1DN-expressing cells to tumor maintenance. The GFP index is defined as the ratio of the percentage of GFP-positive tumor cells in the secondary tumors compared to the percentage of GFP-positive infected tumor cells before transplant. Note that 3 recipient mice did not develop secondary tumors from Vav1DN-infected tumor cells (black triangles). *p<0.05. E. Kaplan-Meier survival curve of mice following stereotactic cortical transplantation of GNPs, infected with Vav1/GFP (n=8) or Mycn /GFP GNPs (n=7) retroviruses.

To examine the contribution of retrovirally-infected cells to tumor development, we analyzed by FACS the percentage of infected GFP-positive tumor cells before transplant and in each secondary tumor. On average, we found a ~40% reduction of Vav1DN-infected GFP⁺ cells in secondary tumors compared to the initially infected pool (Figure 5D). In contrast, when tumor cells were infected with GFP-only retroviruses, the percentage of GFP positive cells increased by ~ 20% in the secondary tumors (Figure 5D). Consistent with the tumor weight, enforced wild-type Vav1 expression did not affect the percentage of GFP⁺ cells in the secondary flank tumors (Figure 5D).

In contrast, enforced wild–type Vav1 expression did not appear to contribute significantly to medulloblastoma initiation in the Ptch1^+/− Cdkn2c^−/− model. Cerebellar granule neuron progenitors (GNPs), the cell type of origin of MB_SHH, were enriched from the cerebella of postnatal (P) day 7, Ptch1^+/− Cdkn2c^−/− pups, then infected with retroviruses carrying GFP-alone or co-expressing wild-type Vav1. Mycn-carrying retroviruses were used as a positive control. ²⁵ Average infection efficiencies were comparable (30.86% for GFP, 35.34% for Vav1/GFP and 40.28% for Mycn/GFP). Tumors failed to develop more than 6 months after stereotactic transplantation of GNPs, infected with GFP only containing virus (n=7) or co-expressing wild-type Vav1 (n=8), into the cerebral cortex. GNPs infected with a Mycn containing retrovirus induced medulloblastomas within 2 months as previously shown (n=7/7²⁵) (Figure 5E). In summary, dominant-negative Vav1 suppressed medulloblastoma maintenance in our Ptch1^+/− Cdkn2c^−/− mouse model.

Vav1 activity regulates granule neuron precursor (GNP) germinal zone exit and migration initiation in normal cerebellar development

We next examined whether Vav1 activity had any effect on the development of GNPs and postmitotic cerebellar granule neurons (CGNs). GNPs are located near the cerebellar surface in a germinal zone (GZ) called the external germinal layer (EGL), whereas after cell cycle exit and neuronal differentiation, CGNs exit their GZ by migrating away from the EGL to cross the molecular layer (ML) and eventually reside within the internal granule layer (IGL).²⁶ GNP or CGN position with the developing cerebellar cortex thus represent an excellent model to examine neuronal progenitor maturation, as cells of the CGN lineage occupy stereotyped positions in the developing cerebellar cortex dependent upon their differentiation status.²⁶ We first found Vav1 was expressed at elevated levels in developing (P7) GNPs compared to cerebella from wild-type mice (Figure 6A). Then, we used electroporation of early post-natal mouse cerebellum followed by ex vivo culture of organotypic cerebellar slices as an assay to examine GNP germinal zone (GZ) exit and migration initiation.²⁷ Expression constructs for Vav1DN and the fluorescent nuclear reporter H2B-mCherry were co-electroporated into the cerebellar cortices of P7 mice and cerebellar slices were cultured ex vivo. While control GNPs remained within the EGL after 24 hours, cells expressing elevated Vav1DN entered the ML and IGL (Figure 6B, D), suggesting that inhibition of Vav1 activity is sufficient to induce precocious GZ exit.

Figure 6. Vav1 regulates GNP germinal zone exit and migration initiation in cerebellar development.

A. Vav1 is expressed in purified P7 GNPs from wild-type (WT) mice and in Ptch1^+/− tumors, relative to the normal cerebellum from one-month old mice. Western blot shows Vav1 and control protein (Hsc70) expression in independent samples. B E. P7 EGL was co-electroporated with the indicated expression constructs and H2B-mCherry. After 24 (B, D) or 48 (C, E) hours of ex vivo culture, the migration distance of H2B-labeled CGN from the pial layer (outer dashed line) was analyzed. EGL boundary is between zero and 45 μm (first bin in histograms); red overlay indicates the average migration distribution of control cells at each time point. B. Most control cells remain within the EGL (dashed lines) at 24 h, while Vav1DN over-expressing cells have prematurely entered the ML and IGL. D. Migration distance versus frequency histogram of control neurons expressing only H2B-mCherry (n=967 cells) or Vav1DN over-expressing (n=2119 cells) CGNs. Control vs. Vav1DN, p=1.08×10⁻⁶⁷ (Chi² test). C. While control cells expressing only H2B-mCherry entered the ML and IGL after 48 h, Vav1CA over-expressing cells remained in the EGL. E. Migration distance vs. frequency histogram of control (n=2252 cells) or Vav1CA over-expressing (n=2570 cells) CGNs. Control vs. Vav1CA, p= 0.0008 (Chi² test). Scale bar, 100 μm.

Given that Vav1 inhibition appeared to regulate GNP GZ exit and migration initiation, we next evaluated whether Vav1 activity could regulate GNP’s GZ exit using a gain-of-function approach. Expression vectors for a constitutively active Vav1 lacking the auto-inhibitory amino terminal domain (Vav1CA²⁸ ) were electroporated into P7 EGL. After 2 days of ex vivo culture, control CGNs initiated migration correctly and had entered the molecular layer and IGL while Vav1CA expressing CGNs remained within the EGL (Figure 6C, E), suggesting that activation of Vav-1 activity blocks GZ exit. Finally, we examined the morphology of Vav1DN and Vav1CA over expressing cells via time-lapse microscopy of in vitro CGN cultures. Whilst control neurons and Vav1DN expressing neurons elaborated the long neurites of mature CGNs, Vav1CA expressing retained the short processes which are the characteristic of immature CGNs (See Supplementary Movies 1-3).

Together, these data show Vav1 activity blocks GNP exit from the developing cerebellar germinal zone and indicate a role in normal cerebellar development.

Discussion

Our investigations of DNA methylation events in human MB_SHH and four independent SHH mouse medulloblastoma models have identified VAV1 hypo-methylation as a conserved and defining MB_SHH event which affects around 70% of human tumors, more common than any mutated gene (PTCH1 (~45%) and TERT (~30%) are most frequent),^{6, 29, 30} and which describes MB_SHH of all key clinical, pathological and molecular demographics. Its complete regional de-methylation (Supplementary Figure S3), coupled with an absence of significant copy number alterations or mutations in large cohorts [data from^{11, 22}] indicates hypomethylation is its major mechanism of alteration. VAV1 methylation status is intimately related to its expression; VAV1 expression is controlled by methylation in human cell lines and its hypo-methylation, in both mouse and human tumors, is significantly associated with increased expression. Consistent with this, the region of uniform hypo-methylation we have characterized overlaps with the minimal region shown in mutagenesis studies to be sufficient to drive VAV1 expression.³¹

Modelling VAV1 epigenetic silencing in mouse and human tumor models validated a role for VAV1 expression in SHH-associated tumor growth and maintenance. Introduction of dominant-negative Vav1 into tumor cells from established hypo-methylated Ptch1^+/− Cdkn2c^−/− mouse tumors significantly reduced tumor re-establishment and growth. In contrast, enforced expression of wild-type Vav1 did not further increase tumorigenic potential, indicating endogenous Vav1 levels in hypo-methylated cells are sufficient to promote tumorigenesis. Further, the impact of VAV1 silencing on proliferation and cell cycle arrest in VAV1 hypo-methylated human cells is consistent with VAV1-dependent effects in other cancer cell types, ^{24, 32} while the identification of Vav1 mutation as a SHH-cooperating event in a murine insertional mutagenesis screen ³³ provides further genetic evidence to substantiate its role in MB_SHH. Any role for VAV1 in tumor initiation is less clear; enforced Vav1 expression in engrafted Ptch1^+/− Cdkn2c^−/− GNPs did not cause tumor development, in contrast to Mycn controls. ²⁵ However, given Ptch1’s established role in initiation, ³⁴ these results do not preclude Vav1 involvement. Future strategies to examine any interaction between Ptch1 and Vav1 in spontaneous SHH-tumor formation, which account for endogenous Vav1 expression³⁵ should prove informative.

The role of VAV1 in the majority of MB_SHH tumors, together with the inhibition of tumor growth following its silencing, identify and validate VAV1 and its dependent pathways as important potential therapeutic targets. In particular, its homogeneous mechanism of activation and SHH-independence highlight potential applications for alternative or second-line / combination therapies with SHH pathway inhibitors. Consistent with its protumorigenic role, VAV1 hypo-methylation was associated with a poor outcome in clinical trial-derived MB_SHH patients,^{15, 21} supporting its utility as both a prognostic and predictive biomarker for use in stratified therapies. Multivariate survival analysis, validation in alternative cohorts, and investigations in all disease demographic groups, are now essential to confirm reproducibility of these initial findings prior to any clinical application.

Vav1 is expressed in GNPs [the proposed cells of origin for MB_SHH³⁶] during early cerebellar development. Our data show its activity blocks GNP exit from the developing cerebellar germinal zone, suggesting a mechanism by which disruption of these processes by Vav1 hypo-methylation and expression may contribute to medulloblastoma tumorigenesis. These findings further emphasize the intimate links between cerebellar development and medulloblastoma, and evidence that temporally/spatially restricted pathways in normal cerebellar development become constitutively activated during tumorigenesis.³⁶ Coupled with the established roles of SHH pathway activation in GNP proliferation ³⁴ and MB_SHH, our data also suggest common functions for Vav1 and SHH in both processes, and provide a basis for the investigation of any functional interaction. Interestingly, Vav3 plays roles in the migration of GNPs from the EGL to the IGL, and in the survival of GCNs in the IGL³⁷, implicating the involvement of multiple Vav family members in cerebellar development. Consideration of developmental toxicities will thus be essential in developing targeted therapies against VAV1, however that Vav1^−/− mice are viable and healthy apart from impaired T-cell development ³⁸ is encouraging.

How might VAV1 exert its function? VAV1 expression is normally confined to the hematopoietic system, where its best characterized function is as a GDP/GTP nucleotide exchange factor for the Rho/Rac family of GTPases. However, roles in multiple diverse processes including activation of the JNK, ERK, Ras, NF-kB, and NFAT pathways, cytoskeleton organization, migration and TP53-mediated apoptosis have also been described.^{32, 39, 40} Direct small molecule VAV1 inhibitors are not widely reported, however strategies against many of its candidate downstream pathways are in development.⁴¹ A detailed exploration of the pathways impacted by VAV1 in MB_SHH and normal cerebellar development will therefore now be essential to understand their mechanistic basis and to enable therapeutic targeting.

Finally, available evidence suggests an emerging widespread relevance for VAV1 in tumor development. In addition to its expression in hematopoiesis and hematological malignancies,⁴² VAV1 expression has been reported in solid tumors including neuroblastomas³² and pancreatic ductal adenocarcinomas, where expression was associated with a poor survival outcome, and with demethylation of the VAV1 promoter in tumor cell lines. ²⁴ In preliminary investigations of other pediatric brain tumor types, VAV1 hypo-methylation at the CpG residues involved in MB_SHH is also a feature of a proportion of posterior fossa ependymomas, high-grade gliomas and CNS-PNETs, as well as occasional non-MB_SHH medulloblastomas (Supplementary Figure S4; Figure 2), suggesting potential roles in their development for further investigation.

Materials and Methods

Patient samples, DNA methylation profiling and subgroup status

DNA methylation profiling and consensus clustering of molecular subgroups was performed as previously described¹⁵ on 230 primary medulloblastomas (of which 216 were confidently assigned to a subgroup), alongside 21 normal non-neoplastic cerebella (representing fetal, infant and adult samples⁴³), using the Illumina GoldenGate Cancer Panel I microarray. ⁴⁴ The DNA methylation status of 1505 CpG sites (807 genes) was represented as a β-score from zero (unmethylated) to one (methylated). Subgroup status was additionally assessed using mRNA signature,⁴⁵ immunohistochemistry (IHC) ⁴⁶ and mutational (CTNNB1) ⁴⁵ biomarker assays.

5-azaCdR treatment of human cell lines

Four human medulloblastoma cell lines (D283Med, MEB-Med8a, D425Med and DAOY) were grown in the presence or absence of the demethylating agent 5-azaCdR as previously described.¹² Cell lines were validated by confirmation of published karyotypes. ⁴⁷

Mouse tumor and tissue samples

Mouse medulloblastomas from Smo/Smo mice ¹⁸, Ptch1^+/− Cdkn2c^−/− mice ¹⁶, Cdkn2c^−/− p53^FL/−, Nestin-Cre⁺ mice irradiated at P7 ¹⁶ and Sufu^+/− p53 ^−/− mice ¹⁷ (all n=3) were assessed alongside control cerebella from strain-matched non-transgenic mice.

Percoll-enriched GNPs from post-natal P7 cerebellum and GNP-like tumor cells from spontaneously-arisen medulloblastomas were purified from Ptch1^+/− Cdkn2c^−/− mice. ¹⁶ P7 cerebellar GNPs were also purified from wild-type mice. CD-1 nu/nu mice (Charles River Laboratories, Wilmington, MA) were used for cortical or flank transplantation. Wild type C57BL/6 mice (Charles River Laboratories, Wilmington, MA) were used for ex vivo cerebellar slice assays. All animal experiments were approved by and conducted in accordance with St. Jude Children’s Research Hospital Animal Care and Use Committee guidelines.

Nucleic acid extraction

DNA and RNA were extracted from frozen tumors, tissues and cell lines using Trizol (Invitrogen, Carlsbad, CA, USA), and DNA from FFPE samples using a Qiagen DNeasy kit (Qiagen, Valencia, CA, USA), according to manufacturer’s instructions.

Identification of differentially methylated CpG sites

CpG sites showing significant differences between the SHH and other medulloblastoma subgroups (combined) were identified (Mann-Whitney U test, retaining probes significant (p<0.05) after Benjamini-Hochberg false discovery rate (FDR) correction). All CpG sites identified showed an absolute change in average (mean) β≥0.34 between these groups. Differences in CpG methylation between the SHH subgroup and normal cerebella were assessed equivalently.

Analysis of DNA methylation status by bisulfite sequencing

Bisulfite DNA treatment was carried out using the Epitect bisulfite kit (Qiagen Ltd. Valencia, CA, USA), and PCR using standard reaction conditions (Supplementary Table S2). Purified products were directly sequenced with a CEQ DTCS kit (Beckman Coulter, High Wycombe, UK) and analyzed on a CEQ 2000XL DNA analysis system (Beckman Coulter). The methylation status at each CpG residue was determined by assessment of relative peak intensities, as previously described.¹² The DNA methylation status (methylation score) of orthologous regions in the mouse and human genomes (mean, based on all sequenced CpG sites) was subsequently calculated for each region of interest.

cDNA synthesis and quantitative RT-PCR

cDNA was synthesized from total RNA using a Biorad iScript™ cDNA synthesis kit (Biorad, Irvine, CA, USA) according to manufacturer’s instructions. Measurements of VAV1 transcript levels in mouse and human samples were determined using SYBR Green on the ABI PRISM 7900HT Detection System (Applied Biosystems, Foster City, CA, USA), using the comparative Ct relative quantification method (Supplementary Table S2).

RNA interference

Cells were seeded in 6- or 96-well plates and allowed to adhere for 24 hours, then transfected with 40nM small interfering RNA (siRNA) duplexes to VAV1 (sense strands: duplex1 CGUCGAGGUCAAGCACAUUdTdT;⁴⁸ duplex2 GCCAUCAGCAUUAAAUAUAdTdT, duplex3 UCAAAUACAAGGAGAGGUUdTdT (DharmaconGE, Layfayette, CO, USA) or negative control siRNA duplexes (sense UUCUCCGAACGUGUCACGUdTdT (Qiagen)) using Lipofectamine™ RNAiMAX (Invitrogen), following the manufacturer’s instructions.

Growth assays

Cells were seeded at 1 ×10³ in 96-well plates and allowed to adhere for 24 hours. The number of viable cells at day 0 (prior to siRNA addition) and days 1-5 (after addition) was quantified using an XTT cell proliferation assay kit (Roche, Hertfordshire, UK) according to the manufacturer’s instructions. Experiments were performed in quadruplicate.

Flow cytometry

Adherent cells were harvested with trypsin 36 and 48 hours after siRNA transfection, fixed in ice-cold 70% ethanol/30% phosphate buffered saline (PBS) solution, then resuspended in 40 mg/ml propidium iodide/0.1 mg/ml RNAse A (Sigma-Aldrich, Gillingham, UK) at 37°C for 30 minutes, and assessed using a FACScan (Becton Dickinson Oxford, UK) and Modfit software (Verity software, Maine, USA). Experiments were performed in quadruplicate.

Retrovirus production

Plasmids used for retrovirus production were generated using the mouse stem cell virus LTR (MSCV) backbone expressing the green fluorescent protein (GFP) downstream of the internal ribosomal entry site (IRES) pMSCV-IRES-GFP. The full-length Vav1 cDNA (amino acids 1-845 (pMSCV-Vav1-IRES-GFP)) and the N-terminal region (amino acids 1-186 (pMSCV-Vav1DN-IRES-GFP)) were inserted downstream of the MSCV LTR in the pMSCV-IRES-GFP plasmid. The pMSCV-Mycn-IRES-GFP construct and generation of retroviral stocks were previously described. ²⁵

Mouse cerebellar and tumor cell purification, virus infection and orthotopic transplantations

GNP-enriched mouse cerebellar and medulloblastoma cells were purified by percoll gradient. ^{25, 49} Purified cells were plated on Matrigel-coated dishes and infected with retroviruses (GFP alone, Vav1-IRES-GFP, Vav1DN-IRES-GFP and Mycn-IRES-GFP). 2 days after infection, cells were harvested, re-suspended in Matrigel (BD Biosciences, San Jose, CA, USA) and transplanted into CD-1 nu/nu mice (2.0 × 10⁶ cells/injection) as previously described. ²⁵ GNPs infected with GFP alone, Mycn-IRES-GFP or Vav1-IRES-GFP retroviruses were transplanted into the cortices of recipient mice. Virally-infected tumor cells were transplanted in the flank of naïve recipient animals. Cortically-transplanted animals were sacrificed, when neurological symptoms were confirmed. Flank injections of virally-marked tumor cells were performed on both sides of each mouse; in the left side, 2.0×10⁶ tumor cells infected with GFP and on the other side, the same number of tumor cells infected with either Vav1/GFP or Vav1DN/GFP retroviruses. In parallel, a subset of infected cells was analyzed by FACS to calculate the efficiency of infection reflected by the percentage of GFP⁺ cells. Transplanted mouse recipients were monitored daily; tumors were harvested and weighed, and tumor cells purified by Percoll gradient and analyzed by FACS for GFP expression as previously described.³

Protein lysate preparation and immunoblotting

Whole-cell extracts were prepared from human cell lines, wild-type 1 month-old whole cerebella, Percoll-purified GNPs from P7 cerebella, and mouse Ptch^+/− medulloblastomas. 25-35 mg protein/sample was separated by gel electrophoresis and immunoblotted with defined antibodies as previously described.¹² Primary antibodies were VAV1 (C14; Santa Cruz, Santa Cruz, CA, USA), Actin (Sigma-Aldrich) and HSC70 (K19; Santa Cruz). Appropriate secondary antibodies with horse radish peroxidise conjugates (Dakocymation Ltd, Glostrup, Denmark) were bound to the primary antibodies, followed by incubation with enhanced chemiluminescence (ECL or Dura) substrate (Pierce, Rockford, IL, USA), visualization by autoradiography, and densitometry using a Fuji-Las (Fujifilm Corporation, Tokyo, Japan) camera and the AIDA image analyzer program (Raytek, Sheffield, UK).

Ex-vivo cerebellar electroporation, organotypic slice culture and imaging

Plasmids for ex-vivo cerebellar slice experiments were generated using the pCIG2 vector, which contains a CAG promoter for high-level expression in CGNs and GNPs. The dominant-negative form of Vav1 (Vav1DN), the constitutively active form of Vav1 (Vav1CA) or H2B-mCherry cDNAs were inserted downstream of the CAG promoter in the pCIG2 expression vector.

Ex-vivo cerebellar electroporation and organotypic slice culture was performed on P7 mouse cerebella as described previously. ²⁷ Migration distance measurements of fixed slices were performed by marking the centre of individual cell nuclei marked by mCherry-H2B using the pen function of SlideBook (Intelligent Imaging Innovations, Denver, CO, USA), exporting the cellular coordinates and calculating the distance between the center of each cell and the cerebellar surface using a custom written algorithm in Igor Pro (Wavemetrics Inc, Lake Oswego, OR, USA). All measurements were analyzed statistically using Microsoft Excel and graphed using Kaleidagraph v4.03 (Synergy software systems, Reading, PA, USA).

Statistical analysis

All array-based bioinformatic and statistical analyses were performed using R (v2.15.0)⁵⁰ and the stated statistical test. Survival analyses were performed using event-free survival (EFS) times. ¹⁵

Supplementary Material

1

Supplementary Table 1. Selection of differentially methylated human sequences and their mouse orthologs for investigation. Genes selected met the following criteria: (i) the human and mouse genomes harbored orthologous regions (Supplementary Figure S1A), (ii) methylation changes in human tumors affected multiple CpG sites in the region surrounding the array-associated probe site, and (iii) the average methylation levels at these additional sites were strongly correlated with the array site in human samples (r² >0.70; Supplementary Figure S1B). The probe sequence used to assay the human CpG site of interest on the Illumina GoldenGate Cancer Panel I methylation array was extended by 50bp proximally and distally using the reference sequences provided by Illumina. This extended sequence was then compared to the mouse assembled genomes using the Blastn algorithm (no filters) (www.ncbi.nlm.nih.gov/Blast.cgi). Orthologous regions of the mouse genome which mapped to the expected gene were identified for further analysis and suitable primers were designed to assess similar sized homologous regions in mouse and human DNA. The positions of the bisulfite sequencing PCR products relative to the transcriptional start site and the number of CpGs analyzed for both are shown. *There is no equivalent mouse transcript to the human transcript used for MEST.

2

Supplementary Table 2. Primers used for bisulfite and real time PCR. For bisulfite PCR, primers were designed to amplify regions (200-500bp) surrounding the CpG site assessed on the Illumina GoldenGate Cancer Panel I microarray or its orthologous position in the mouse genome. Primers were designed using the program Methprimer (www.urogene.org./methprimer). The annealing temperature for all primers was 55°C, cycling at this temperature was preceded by 5 cycles of ‘touchdown PCR’ from 60°C in increments of −1°C to increase the specificity of the products. The human TRIM29 PCR reactions required the addition of Betaine (Sigma-Aldrich, Gillingham, UK) to the PCR mix (final concentration 1M). For quantitative real time RT-PCR, primers were designed using Primer Express software (Applied Biosystems) to ensure that products overlapped intron/exon boundaries. Dissociation curves confirmed PCR product specificity. Expression was assessed relative to expression of a control gene (28S rRNA or TBP), using the comparative Ct method for relative quantification. Dilution series were carried out to confirm relative expression was consistent over a range of input DNA concentrations.

3

Supplementary Figure 1. Identification of DNA methylation events associated with MB_SHH. A. Clustering of DNA methylation patterns in 230 primary human medulloblastomas identifies four consensus subgroups which are highly significantly related to established SHH and WNT subgroup biomarkers (‘p’ values are shown). The 216 tumors which could be confidently assigned to a subgroup are shown ¹⁵. MB_SHH, red; Group 3, yellow; WNT, blue; Group 4, green. MB_SHH biomarker (SHH mRNA signature / GAB-1/YAP1 IHC) positivity, red. WNT subgroup biomarker (WNT mRNA signature / β-catenin IHC) positivity, blue. Data not available, grey. Magnitudes of the four defining metagenes (V1 to V4; highly expressed metagenes, red; lowly expressed, blue) and subgroup-specific DNA methylation events (unmethylated, green; fully-methylated, red) are shown. B. Principal component analysis loadings plot shows class designations for WNT (blue), MB_SHH (red), Group 3 (yellow) and Group 4 (green) tumors and normal cerebella (pink; n=21). Covariance spheroids with a 95% confidence interval have been plotted to demonstrate group memberships.

4

Supplementary Figure 2. Identification and assessment of DNA methylation alterations in orthologous regions of the mouse and human genomes. A. Examples of regions of homology between mouse and human genomes. Aligned sequences are shown for the regions surrounding TRIM29 (probe P833), CCKAR (probe E79), VAV1 (probe E9) and MEST (probe P4). The probe site interrogated by the Illumina array is boxed (dashed line); other conserved CpG sites are boxed (solid line) and non- conserved CpG sites are underlined. H, human; M, mouse. B. Correlation between the array-assessed β-value and the average methylation score determined by bisulfite sequencing for all CpG sites in the surrounding region (Table S1) for nine genomic regions, in six randomly-selected cerebella and six randomly-selected MB_SHH tumors. (RASSF1A; r², 0.92). C. Methylation levels at individual CpG sites in the assessed Vav1 region in mouse tumors and strain-specific cerebella following analysis by bisulfite sequencing, shown alongside the overall average methylation score for the region. CpG sites are labelled by their position relative to the translational start site of the Vav1 gene. Black circles, ≥80% methylation; three-quarter black circles, ≥60%; half-black circles, ≥40%; quarter-black circles, ≥20% methylation; white circles, <20% methylation.

5

Supplementary Figure 3. VAV1 hypo-methylation in human medulloblastomas: Regional CpG patterns, molecular and clinical significance. A. Representative examples of bisulfite sequencing on the reverse strand of the region surrounding VAV1 (probe E9). Assayed CpG sites (labelled by their position relative to the translational start site of the Vav1 gene) are underlined and calculated methylated peak height (G-green) relative to total peak height is given for each of the 10 CpG sites analyzed. The methylation score is the average of these values and is shown together with the β value for the CpG site (boxed) assessed on the Illumina array. B. Methylation levels at individual CpG sites in the assessed VAV1 region in representative human tumors (MB_SHH (n=25) and non-MB_SHH (n=8) tumors) and cerebella (n=7), following analysis by bisulfite sequencing and shown alongside the methylation score. Black circles, ≥80% methylation; three-quarter black circles, ≥60%; half-black circles, ≥40%; quarter-black circles, ≥20% methylation; white circles, <20% methylation. The CpG site assessed by probe E9 is shown, and the cut-off used to discriminate VAV1 hypo-methylated tumors (0.53; Figure 3) is marked. C. Box plots showing methylation levels (β-values) for VAV1 (probe P317) in the MB_SHH subgroup (n=50) compared to non- MB_SHH (n=166) tumors and cerebella (CB; n=21). The mean cerebellar methylation (-3SD; cut-off marked at 0.30) is shown to distinguish hypo-methylated tumors. Non-MB_SHH tumors are shown colored by subgroup (Group 3, yellow; WNT, blue; Group 4, green).

6

Supplementary Figure 4. VAV1 methylation patterns in multiple pediatric brain tumor types. Box plots show methylation patterns for probes VAV1 E9 (A) and VAV1 P317 (B) for posterior fossa ependymomas (dark green, n=45); supratentorial ependymomas (light green, n=16); pediatric high-grade gliomas (grey, pHGG, n=15) and CNS-PNETs (black, n=30). For comparison, methylation status of non-neoplastic cerebellar (pink, n=21) and cerebral (dark pink, n=16) cohorts are shown (data for the 16 cerebra are previously described ⁴³). Horizontal cut-off lines, corresponding to the mean -3SD cerebellar methylation (probe E9, β=0.53 and probe P317, β=0.30) are shown for comparison with Figures 2A and S3C. ‘p’ values from Kruskal-Wallis tests are shown.

Supp Movie 1

Supplementary movies. Time lapse images documenting the morphology of CGNs following electroporation with equal amounts of vector only (Movie 1); Vav1DN (Movie 2) and Vav1CA (Movie 3).