Modeling medulloblastoma in vivo and with human cerebellar organoids

Medulloblastoma (MB) is the most common malignant brain tumor in children and among the subtypes, Group 3 MB has the worst outcome. Here, we perform an in vivo, patient-specific screen leading to the identification of Otx2 and c-MYC as strong Group 3 MB inducers. We validated our findings in human cerebellar organoids where Otx2/c-MYC give rise to MB-like organoids harboring a DNA methylation signature that clusters with human Group 3 tumors. Furthermore, we show that SMARCA4 is able to reduce Otx2/c-MYC tumorigenic activity in vivo and in human cerebellar organoids while SMARCA4 T910M, a mutant form found in human MB patients, inhibits the wild-type protein function. Finally, treatment with Tazemetostat, a EZH2-specific inhibitor, reduces Otx2/c-MYC tumorigenesis in ex vivo culture and human cerebellar organoids. In conclusion, human cerebellar organoids can be efficiently used to understand the role of genes found altered in cancer patients and represent a reliable tool for developing personalized therapies.

M edulloblastoma (MB) is one of the most aggressive brain tumors affecting children and stands as a cause of a high percentage of morbidity and mortality among cancer patients 1 . MB is a biologically and clinically heterogeneous tumor, including several subgroups 2 . Among the groups, patients with Group 3 MB (characterized by c-MYC upregulation), have the worst outcome with~50% of the tumors metastatic at the time of diagnosis. The currently available therapy for MB consists of maximal safe resection, craniospinal radiation (for children ≥ 3 years old) and chemotherapy. Therefore, developing humanized mouse model of Group 3 medulloblastoma would be of paramount importance for the identification and testing of new drugs for pediatric patients, tailored on the genetic condition of the patient itself. Recently, several studies have utilized nextgeneration sequencing technologies to map the genomic landscape of MB and to identify novel driver mutations [3][4][5][6][7][8] . A secondgeneration medulloblastoma subgrouping of Group 3/4 has led to the identification of eight subtypes with major clinicopathological and molecular features 9 . Group II, III, and V are at high clinical risk (5 years overall survival 41-58 months in retrospective series) and enriched for c-MYC amplification. Notably, the function and tumorigenicity of some oncogenes, such as c-MYC, have been tested ex vivo and also in vivo where its overexpression together with either Gfi1 overexpression or p53 loss of function, is sufficient to promote MB in mice [10][11][12] . The same gene combinations have been electroporated in utero to examine the susceptibility of distinct cerebellar progenitors to transformation 13 . Furthermore, p53 has been found to interact with c-Myc family in mouse models and human patients of Group 3 MB, suggesting their contribution to MB relapse 14 .
In this study, we performed an in vivo screen by postnatal transfection of mouse cerebella to identify novel driver genes combinations able to induce Group 3 MB. Recently, human neuroepithelial stem cells (derived from iPSC) have been used to model MB. In particular, iPSC from patients with Gorlin syndrome have been differentiated in neuroepithelial stem cells, and it has been shown that these are able to induce MB upon orthotopic implantation in mice 15 . On the other hand, human forebrain organoids have been already used to test the function of putative genes involved in glioma [16][17][18] , but cerebellar organoids have never been used to model human MB. Notably, human iPSC robustly differentiate into cerebellar progenitors and generate human cerebellar organoids that partially recapitulate the complexity of human cerebellum 19,20 . Here, we exploited human cerebellar organoids to develop reliable MB organoids and to validate our screen in human cells.

Results
In vivo transfection of cerebellar cells. In order to uncover novel gene combinations that could be responsible for Group 3 MB development, we have reproduced in mice the genetic alterations found in human patients. In a first set of experiments, we tested our ability to target and transfect cerebellar cells in vivo. A mix of Venus-coding plasmid and jetPEI transfection reagent (Polyplus-transfection) were stereotaxically injected in vivo into newborn (P0) CD1 wild-type mice 21 . To obtain stable expression of Venus, we used PiggyBac system 22 that allows multiple insertions of Venus under control of CAG promoter. As shown in Fig. 1a-c, at 3, 7, and 23 days post injection (d.p.i.) we detected Venus-positive cells in external granule layer (EGL), internal granule layer (IGL), and in deep cerebellar nuclei (DCN). Interestingly, at 7 d.p.i. Venus is expressed in Sox9 (Fig. 1d) and Sox2 ( Supplementary Fig. 1A) positive glial cells, in Olig2-positive oligodendrocytes (Fig. 1e), Barhl1-positive granule neuron progenitors ( Supplementary Fig. 1B), few calbindin positive Purkinje cells ( Supplementary Fig. 1C), and NeuNpositive neurons ( Supplementary Fig. 1D) in IGL. Therefore, with this strategy we have been able to perform stable transfection of glial cells, neurons, and progenitors in postnatal mouse cerebellum.
Otx2 and c-MYC as novel MB driver genes. To identify novel putative oncogenes/oncosuppressors combinations, we analyzed the list of patient-specific mutations identified in previous exome sequencing, microarray, and CNVs data of Group 3 MB [3][4][5][6][7][8] . We decided to focus on all the genetic alterations present in Group 3 MB patients with a frequency higher than 2% or genes that show differential expression (higher than 16-folds) compared with normal cerebellum 23 (see the Methods section). Based on this analysis, we created a list of gene combinations ( Fig. 1f; Supplementary Fig. 1E) to be tested in mice for their ability to induce MB. To recapitulate the human gene amplification or overexpression, we used the PiggyBac system, which allows multiple insertions of the selected putative oncogene. On the other hand, we used CRISPR/Cas9-mediated loss-of-function approach to remove the selected putative oncosuppressors (Methods). Cas9 technology has been already used to model MB 21 and to screen genes involved in tumor growth and metastasis in mice 24,25 .
Several gene combinations did not give rise to tumors but only to the formation of big clusters of cells with weak Venus expression 3 months post injection ( Supplementary Fig. 1F). Since we never observed these cell clusters in control experiments (injection of Venus alone), we speculate that these could be dead or senescent cells due to either oncogenes expression or oncosuppressors deletion. None of the gene combinations, where putative oncosuppressors were silenced with Cas9 technology, led to tumor formation. This might be due to inefficient gene deletion or because of missense, nonsense, and frameshift mutations present in human patients are not efficiently recapitulated by our strategy. Among all the tested combinations, we observed reduced mice survival with Gfi1 + c-MYC (GM) and Otx2 + c-MYC (OM) genes overexpression (Fig. 1g) and formation of brain tumors (Fig. 1f, h, 2a, b). The Gfi1 + c-MYC overexpression in postnatal cerebellar progenitors has been previously described as able to generate Group 3 MB in mice 11,26 , therefore validating the efficiency of our method. As shown in Fig. 2a Fig. 2C, D). Notably, the tumors are NPR3 positive (Fig. 2e, f) and GFAP negative (Fig. 2g, h) such as human Group 3 MB 27,28 . In fact, NPR3 is a specific marker that is expressed in human Group 3 MB and is not present in the other MB subgroups 28 , suggesting that our model could recapitulate human tumors. The histopathological and immunophenotypical analysis confirmed that tumors are Synaptophysin-positive MB and with cytoplasmic b-catenin, therefore, non-Wnt MB (Supplementary Fig. 2E-I). Notably, within the tumors, we observed Sox9/Venus double-positive cells, putative glial cells ( Supplementary Fig. 3A, B), but also Venus-negative-Olig2positive cells that could be infiltrated oligodendrocytes (Supplementary Fig. 3C, D). Furthermore, the tumors seemed to be Barhl1 negative, a marker for granule neuron progenitors and granule neurons ( Supplementary Fig. 3E, F). Interestingly, in one mouse injected with OM we observed sacral metastasis (Venus and pH3 positive, Fig. 2i-l), suggesting that our new Group 3 mouse models can recapitulate the malignant phenotype of human MB. Interestingly, OTX2 is overexpressed/amplified in human Group 3 MB and has been already found to be required for the tumorigenesis of MB cell lines 29 Here, we have demonstrated that Otx2 is able to induce MB in vivo from mouse cerebellar progenitors that have not been manipulated outside the mouse brain. Our data indicate that concomitant overexpression of Otx2 and c-MYC originate MB in vivo. The two genes alone are not able to induce MB, suggesting that cooperation between the two transcription factors is required for tumor formation. Notably, the combination of Gfi1 and Otx2 overexpression does not induce MB (Fig. 1f), therefore suggesting that c-MYC alteration is required to transform cerebellar cells.
SMARCA4 blocks Otx2/c-MYC-induced Group 3 MB. To identify a druggable signaling pathway representing a possible new therapeutic alternative to block Group 3 MB, we analyzed the sequencing data of human patients harboring OTX2 and c-MYC overexpression. Among all the putative oncosuppressors altered in Group 3 MB, SMARCA4 is the most frequently mutated (harboring missense mutations in helicase domains), highlighting its possible role in MB formation. Notably, in Group 3 MB patients (that have high levels of OTX2 and c-MYC), there are variable levels of SMARCA4 expression that are lower than in human neural stem cells ( Supplementary Fig. 4A, B). Furthermore, endogenous Smarca4 expression is downregulated in our OM-induced MB mouse models as compared with the GM tumors ( Fig. 3a-c). Based on this evidence, we speculate that our mouse model could represent patients expressing high levels of OTX2/c-MYC and low levels of SMARCA4. Therefore, we expressed high levels of SMARCA4 with the intent of inhibiting OM-induced MB generation. Interestingly, 20/20 mice injected with both OM and SMARCA4 overexpression vectors were still alive after 3 months and none of them developed MB (Fig. 3d). On the other hand, SMARCA4 overexpression partially reduced the incidence of MB in GM animals and 3/20 mice developed MB with no influence on their survival (Fig. 3e). These results suggest that different Group 3 MB tumors might show different sensitivity to SMARCA4 deregulation.
To better characterize the effects of SMARCA4, we analyzed the cerebellum of mice injected with OM with and without SMARCA4 overexpression, 10 days after injection. As shown in Fig. 3f, OM overexpression led to formation of Ki67/Venuspositive cells clusters with few β3-tubulin and GFAP-positive cells (Fig. 3g, h). Notably, when SMARCA4 was co-overexpressed, there was no formation of cell clusters and we identified only few Ki67/Venus double-positive cells (Fig. 3i). Furthermore, several Venus-positive cells expressed β3-tubulin and GFAP (Fig. 3j, k). We observed that Otx2 and c-MYC proteins are still present in transfected cerebellum with SMARCA4 co-overexpression (Supplementary Fig. 4C-Q), suggesting that SMARCA4 does not affect their expression, but Otx2/c-MYC oncogenic-mediated activity. Finally, we tested if SMARCA4 overexpression could lead to abnormal levels of other proteins part of the SWI/SNF complex and be responsible for the phenotype observed. As shown in Supplementary Fig. 4R, SMARCA4 overexpression in human cerebellar progenitors that have a gene expression pattern resembling hindbrain fate (iPSC-derived AF22 cells) 32,33 seems to have no effect on the SWI/SNF complex proteins expression.
Human cerebellar organoids as a reliable model for Group 3 MB. To generate an organoid-derived model of human Group 3 MB, we tested the function of GM and OM combinations in human cerebellar organoids. Notably, Muguruma et al. 19,20 clearly demonstrated that hESC/iPSC aggregates are steered to differentiate into cerebellar progenitors and neurons. In fact, they were able to induce progenitors to self-form neuroepithelial structures that mimic early cerebellar plate, composed by cerebellar progenitors, cerebellar neurons (interneurons, Purkinje cells, and granule neurons) and glial cells. We recapitulated Muguruma protocol and we electroporated PiggyBac vector expressing Venus into human cerebellar organoids (Fig. 4a). As shown in Fig. 4b and Supplementary Fig. 5A, B, we were able to electroporate human cerebellar organoids at day 35 of differentiation, when almost all cerebellar progenitors are present 20 and to deliver DNA into intact 3D human cerebellar organoids ( Supplementary Fig. 5B). In order to confirm whether the electroporation procedure induced any change in the transcriptome of the cerebellar organoids, we performed an analysis of the differentially expressed genes (DEGs). As seen in Supplementary Data 1, the number of DEGs is very small (upregulated in treated n = 26; downregulated in treated n = 7; log fold change > 1, count per million average expression > 3, Benjamini-Hochberg adjusted the p-value ≤ 0.05). Moreover, the functional analysis in those small gene subsets indicated biological terms enriched in extracellular matrix/cell adhesion and in protein folding (for details see Supplementary Data 1). We then proceeded to validate our patient-specific screen results in the human organoids by electroporating them at day 35 with GM and OM. As shown in Fig. 4c, d, we observed the formation of small buds of Venuspositive cells in GM and OM electroporated organoids after 25 days, but not in control organoids electroporated with Venus only (Fig. 4b). Notably, the GM and OM organoids showed increased number of PCNA-positive cells within the Venuspositive cells compared with control organoids (Fig. 4e, g; Supplementary Fig. 5C, D). Furthermore, OM organoids (within the Venus-positive cells) had less β3-tubulin-positive cells compared with control cells (Fig. 4f, g). Taken together, these data suggest that GM and OM combinations induce overproliferation of human cerebellar progenitors and impair their differentiation (Fig. 4g). Notably, we observed an increase in Sox9-positive cells that could indicate an increase in glial cells or early cerebellar progenitors, while we did not observe any change in the number of Skor2-positive Purkinje cell precursors ( Fig. 4g; Supplementary  Fig. 5E-H). To understand how GM and OM combinations modify the proliferation of cerebellar progenitors, we analyzed the organoids at an earlier time point (5 days after electroporation). As shown in Fig. 4h, at this stage we observed an increase in PCNA-positive cells possibly due to the mitotic activity of GM Fig. 1 In vivo transfection of mouse cerebellum with different gene combinations. a-c Confocal images of DAPI staining and GFP immunofluorescence (Venus) of sagittal brain section of CD1 mouse 3 (a), 7 (b), 23 (c) days after transfection with pPBase and pPBVenus at P0. Arrows point to Venus-positive cells. d Confocal images of DAPI staining and immunofluorescence for GFP (Venus) and Sox9 of sagittal brain section of CD1 mouse 7 days after transfection with pPBase and pPBVenus at P0. e Confocal images of DAPI staining and immunofluorescence for GFP (Venus) and Olig2 of sagittal brain section of CD1 mouse 7 days after transfection with pPBase and pPBVenus at P0. The white square in (d, e) marks the region shown at higher magnification. f List of combinations of putative oncogenes and putative oncosuppressors transfected in CD1 mice at P0. Putative oncogenes are reported in blue, putative oncosuppressors are reported in black. The number of mice displaying abnormal clusters and tumors are reported. g Kaplan-Meier survival curve of mice injected at P0 with different gene combinations. h DAPI staining and GFP immunofluorescence (Venus) of sagittal brain section of CD1 mouse 3 months after transfection with pPBase + pPBMyc + pPBGfi1 + pPBVenus at P0. Scale bars 250 µm (a), 500 µm (b, c), 100 µm in (d, e), 1 mm in (h). and OM combinations. Since our previous data in mice suggest that SMARCA4 has an effect in reducing OM tumorigenesis, we tested its effects in human cerebellar organoids. As shown in Supplementary Fig. 5I, J, SMARCA4 overexpression is able to reduce OM-induced proliferation in human cerebellar organoids, confirming that SMARCA4 might have a role in reducing OM-induced Group 3 MB. Notably, SMARCA4 overexpression was not able to reduce GM-induced proliferation in human cerebellar organoids, confirming our findings in mice ( Supplementary  Fig. 5K). To test whether organoid-derived model of human Group 3 MB is able to induce cancer in vivo, we injected them in Foxn1 nu (nude) mice (Fig. 5a). We observed the formation of MB    35 . The methylation data files from our organoids were also run through the brain tumor classifier (https://www.molecularneuropathology. org/mnp) 34 . All samples were classified in the Methylation class family Medulloblastoma G3 and G4 (subclass Group 3) with a score > 0.3 (Supplementary Data 3). Of note, by exploiting the Medulloblastoma classifier Group 3/4 v1.0 that takes into account the new consensus on the second-generation molecular subgrouping of medulloblastoma 9 , GM samples were classified as Subtype II, high-risk G3 tumors, while OM samples as Subtype IV, standard-risk G3 tumors. These data highlight the differences obtained by manipulating distinct genes in our models. Therefore, our organoid-based model for Group 3 MB could be used as a reliable tool for human MB modeling.
To better characterize the mechanisms underlying SMARCA4 in MB development, we have overexpressed OM and SMARCA4 in human cerebellar progenitors (AF22 cells) and we checked the expression of several downstream genes of OTX2, c-MYC, and SMARCA4 ( Fig. 6a, b). Among the different analyzed genes, we found that OM affects CDKN2B and CRABP1 expression and that SMARCA4 is able to rescue these effects (Fig. 6c). Notably, SMARCA4 loss of function does not influence CDKN2B and CRABP1 modulation by OM (Fig. 6d, e). In line with our results, p15 INK4b protein (encoded by CDKN2B gene) is a well-known oncosuppressor whose modulation by c-MYC and SMARCA4 has already been suggested in cell lines [36][37][38][39] . CRABP1 protein binds to retinoic acid and helps to transport it into cells. Its role in human carcinogenesis is poorly understood, but high CRABP1 expression levels are associated with poor patient prognosis, high tumor grade in breast cancer 40 and CRABP1 protein modulates cell cycle progression and apoptosis induction in mouse and human cell lines 41 . In Group 3 MB patients, SMARCA4 gene presents several missense mutations (12 out of 131 patients) and the most common SMARCA4 mutation T910M (3 out of 12) has been already characterized in mouse fibroblast, human cell lines, and SMARCA4-deficient cell lines [42][43][44] , but not in cerebellar progenitors. Furthermore, SMARCA4 T910M protein has been found normally incorporated into the BAF complex, but its ATPase activity is highly compromised [42][43][44] . Since SMARCA4 T910M mutation is mainly present in heterozygosity, to analyze its function, we co-overexpressed (at comparable levels, Fig. 6b) the wild-type and mutated form with OM. Interestingly, SMARCA4 T910M co-overexpression blocks the SMARCA4 wild-type effects on CDKN2B and CRABP1 gene expression (Fig. 6c). To further characterize SMARCA4 T910M functions, we analyzed its effects in vivo. As shown in Fig. 6f, g, SMARCA4 wt is able to block OM-induced MB while SMARCA4 T910M overexpression is able to counteract SMARCA4 wt effects. We further analyzed SMARCA4 T910M effects in human cerebellar organoids and as shown in Fig. 6h, it is able to block the SMARCA4 wild-type effects also in this system. Taken together, these data suggest that in patients harboring heterozygotes SMARCA4 missense mutations, SMARCA4 T910M represses SMARCA4 wild-type functions and acts as a dominant-negative.
EZH2 inhibition reduces Group 3 MB growth. SMARCA4 has an antagonistic relationship with histone methyltransferase EZH2 (part of the PRC2 complex), which indeed is essential for survival and growth of SMARCA4-deficient cancer cells 45,46 . Therefore, in search of novel molecules to be used to treat OM MB, we focused on a nonspecific inhibitor of EZH2, 3-deazaneplanocin A (DZNep). Treatment with DZNep induces significant antitumor activity in various cancer types, corresponding to inhibition of PRC2 and removal of H3K27me3 marks [47][48][49] . First, we performed ex vivo culture of cancer cells from mouse tumors arisen upon OM injection. Ezh2 inhibition blocked OM-induced cancer cells growth, DZNep reduced the size of the ex vivo cultures of OM-derived tumor spheroids (Supplementary Fig. 9A-C). To Fig. 2 In vivo transfection of cerebellar cells with Gfi1/c-MYC and Otx2/c-MYC induces Group 3 MB. a DAPI staining and GFP immunofluorescence of CD1 mouse brain section 3 months after transfection with pPBase + pPBMyc + pPBGfi1 + pPBVenus at P0. b DAPI and GFP immunofluorescence of CD1 mouse brain section 1 month after transfection with pPBase + pPBMyc + pPBOtx2 + pPBVenus at P0. c Hematoxylin and Flag (Gfi1) immunohistochemistry of CD1 mouse tumors after transfection with pPBase + pPBMyc + pPBGfi1 + pPBVenus at P0. d DAPI and Otx2 immunofluorescence of CD1 mouse brain section 1 month after transfection with pPBase + pPBMyc + pPBOtx2 + pPBVenus at P0. e, f Hematoxylin and NPR3 immunohistochemistry of CD1 mice tumors after transfection with pPBase + pPBMyc + pPBGfi1 + pPBVenus at P0 (e) and with pPBase + pPBMyc + pPBOtx2 + pPBVenus at P0 (f). g, h GFP and GFAP immunofluorescence of tumors in CD1 mice after transfection with pPBase + pPBMyc + pPBGfi1 + pPBVenus at P0 (g) and with pPBase + pPBMyc + pPBOtx2 + pPBVenus at P0 (h). i In vivo bioluminescent imaging of the upper body of a CD1 mouse transfected with pPBase + pPBMyc + pPBOtx2 + pPBVenus at P0, with a plasmid encoding firefly luciferase (pPBLuc). Imaging was performed weekly from the sixth week after transfection. j In vivo bioluminescent imaging of the lumbosacral region of a CD1 mouse 8 weeks after transfection with pPBase + pPBMyc + pPBOtx2 + pPBVenus + pPBLuc at P0. The luminescence signal is expressed in photons/sec/mm2. k DAPI and immunofluorescence for pH3 of transversal section of the sacral portion of the spine of a CD1 mouse 8 weeks after transfection with pPBase + pPBMyc + pPBOtx2 + pPBVenus + pPBLuc at P0. l DAPI and immunofluorescence for GFP and pH3 of sacral transversal section of the spine of a CD1 mouse 8 weeks after transfection with pPBase + pPBMyc + pPBOtx2 + pPBVenus + pPBLuc at P0. l is the higher magnification of the white square in (k). Scale bars 1 mm in (a, b), 100 µm in (c-h), 500 µm in (k).
compare DZNep effects with other treatments, we used also Panobinostat (LBH589) a histone deacetylase (HDAC) inhibitor that inhibits murine and human MYC-driven MB in vitro and in vivo 50,51 . As shown in Supplementary Fig. 9A Otx2 + c-Myc + Smarca4 (10 days)

j' K'
Otx2 + c-Myc (10 days Supplementary Fig. 10A, B, 24h after DZNep treatment almost 90% the cells inside tumor spheroids were dead or in apoptosis. Furthermore, we observed an increase in Cleaved Caspase-3-positive cells with DZNep treatment compared with control ( Supplementary Fig. 10C, D) and a decrease in the number of pH3-positive cells ( Supplementary Fig. 10E, F). Since DZNep is not a specific inhibitor only for EZH2, we tested (in mouse OM-derived tumor spheroids) other two inhibitors that more specifically repress EZH2 functions, Tazemetostat (EPZ-6438) and GSK-126 52,53 . As shown in Fig. 7a, b upon 24 h of treatment with Tazemetostat and GSK-126, we did not observe increased apoptosis or differences in cell cycle compared with control. On the other hand, upon 3 days of Tazemetostat treatment, we observed an increased number of cells either dead or in late apoptosis (Fig. 7c, d), reduced number of cells in cell cycle compared with control (Fig. 7e, f). We further tested Tazemetostat and GSK-126 treatments in OM-derived MB produced in human organoids. As shown in Fig. 7g, h, we observed an increase in the number of Cleaved Caspase3 and a decrease in PCNA-positive cells in organoids treated for 5 days with Tazemetostat compared with control. In conclusion, the analysis performed in our MB models suggests that EZH2 inhibition with Tazemetostat might represent a suitable therapeutical strategy for Group 3 MB patients with high levels of OTX2 and c-MYC.

Discussion
Human Group 3 MB is one of the most aggressive MB subgroup and is characterized by high c-MYC levels. To recapitulate Group 3 MB tumors, several research groups have overexpressed c-Myc together with either Gfi1 or dominant-negative p53 [10][11][12]26 . Those modifications were sufficient to promote Group 3 MB in mice in vivo and in isolated CD133-positive progenitors from postnatal murine cerebellum. Moreover, drug screens have been performed in Group 3 MB mouse models and patient-derived xenografts (PDX) that led to the identification of molecules able to increase mouse survival 51,54 . However, Group 3 and MB, in general, represents a very heterogeneous family of tumors that requires different therapeutic approaches to obtain reliable results and lower the side effects. Up to now, no patient-specific MB models exist and specific therapeutic strategies for the different Group 3 MB patients are lacking. We performed an in vivo screen to identify new cancer driver genes starting from published genomewide analyses of MB patients. Our analysis led to the identification of Otx2/c-MYC (OM) as a novel driver gene combination required for tumorigenesis. OTX2 is highly expressed in non-SHH MB subgroups, and OTX2 locus is amplified and/or overexpressed in a subset of Group 3 and Group 4 tumors. Previously, it has been shown that OTX2 is required for human MB cell lines growth, but its capacity of inducing MB from cerebellar cells in vivo has never been tested. Furthermore, OTX2 downstream genes NRL and CRX are required for the MB formation in mice using Group 3 PDX 55 . Here, we demonstrated that Otx2 overexpression is necessary for c-MYC tumorigenesis in mouse cerebellar cells in vivo. To confirm the relevance of our data for human tumorigenesis, we used human cerebellar organoids. We generated human iPSC-derived cancer organoids with GM and OM overexpression mimicking human Group 3 MB genetic alteration. Indeed, the use of DNA methylation signature 34 in combination with Group 3-specific markers analysis indicates that our organoid-based MB model recapitulates several features of human Group 3 MB. Of note, exploiting the MB classifier group 3/4 9 , the GM cancer organoids were classified as Subtype II, high-risk G3 tumors, while OM cancer organoids as Subtype IV, standard-risk G3 tumors. These data highlight the differences obtained by manipulating distinct genes in our models. Therefore, our organoid-based model for Group 3 MB can be used as reliable tool to generate novel patient-specific cancer models. Moreover, our data show that OM oncogenic properties are inhibited by the chromatin modifier SMARCA4, a gene also found mutated (putative loss of function) in Group 3 MB that has been already described as involved in cerebellum development and SHH MB 56,57 . Since SMARCA4 is not able to block the oncogenic role of GM combination in Group 3 MB formation, we speculate that patients with Group 3 tumors should be treated differently, depending on their mutations. As a core subunit of the SWI/SNF chromatin remodeling complex, Smarca4 contains an ATPase domain that provides the enzymatic activity required to remodel chromatin structures and regulate transcription 58,59 . Mutations that inactivate SWI/SNF subunits are found in~20% of human cancers, suggesting that a functional complex might be required to prevent tumor formation in several tissues 58,59 . SMARCA4 gene presents several missense mutations in Group 3 MB patients, and the most common SMARCA4 mutation is the T910M. SMARCA4 T910M protein has been found incorporated normally into the BAF complex, but leads to highly compromised ATPase activity 42,43 . Notably, SMARCA4 is involved in decatenating newly replicated sister chromatids, a step required for proper chromosome segregation during mitosis 42 . Failure in this process could lead to aneuploidy that is common in MB and ranges from the partial gain or loss of single chromosomes to full tetraploidy 3,42 . However, it has been published that the relative rate of tetraploidy of SMARCA4 mutant MB was similar to that of wild-type SMARCA4 MB 3 . Therefore, SMARCA4 could have also other mechanisms by which it is involved in MB development. Since SMARCA4 T910M mutation is mainly present in heterozygosity in human Group 3 MB, we analyzed its function in cooverexpression with wild-type SMARCA4 in human neuroephitelial cells, in vivo and in human cerebellar organoids. Our data show that SMARCA4 T910M is able to reduce SMARCA4 wildtype functions on CDKN2B and CRABP1 expression and restores the Otx2/c-MYC oncogenic functions. Based on this evidence, we speculate that mutant SMARCA4 T910M could act as dominantnegative and block the wild-type form. Supporting this hypothesis, some other SMARCA4 mutants act as dominant-negative in Fig. 3 SMARCA4 represses Otx2/c-MYC induced Group 3 medulloblastoma. a DAPI staining and Smarca4 immunofluorescence of sagittal brain tumor section of CD1 mouse 3 months after transfection with pPBase + pPBMyc + pPBGfi1 + pPBVenus at P0. b DAPI staining and Smarca4 immunofluorescence of sagittal brain tumor section of CD1 mouse 1 month after transfection with pPBase + pPBMyc + pPBOtx2 + pPBVenus at P0. c Western blot of brain tumors of CD1 mice after transfection with pPBase + pPBMyc + pPBGfi1 + pPBVenus (GM) and pPBase + pPBMyc + pPBOtx2 + pPBVenus (OM). d Kaplan-Meier survival curve of mice injected at P0 with Otx2 + c-Myc and Otx2 + c-Myc + Smarca4. e Kaplan-Meier survival curve of mice injected at P0 with Gfi1 + c-Myc and Gfi1 + c-Myc + Smarca4. f-h Confocal images of GFP (Venus) and Ki67 (f), β3-tubulin (g), GFAP (h) immunofluorescence of transfected cell clusters in CD1 mouse 10 days after transfection with pPBase + pPBMyc + pPBOtx2 + pPBVenus at P0. The white squares in (f, g, h) mark the region shown at higher magnification in (f', g', h'). i-k Confocal images of GFP (Venus) and Ki67 (i), β3-tubulin (j), GFAP (k) immunofluorescence of transfected cell clusters in CD1 mouse 10 days after transfection with pPBase + pPBMyc + pPBOtx2 + pPBSmarca4 + pPBVenus at P0. The white squares in (i, j, k) mark the region shown at higher magnification in (i', j', k'). Arrows point to double-positive cells. Scale bars 100 µm.
human cell lines 60 , but their function in vivo (during MB development) has not been investigated. Finally, a recent study clarified that SMARCA4 T910M contributes to the targeting of SWI/SNF complexes on chromatin in ovarian carcinoma cells and that could influence CDKN2B expression 44 . This evidence suggests that SMARCA4 T910M might play similar roles in MB and ovarian carcinoma development. Interestingly, SMARCA4 has been previously found to be partially required for SHH MB development, suggesting different roles of SMARCA4 depending on the MB subgroups and on the genetic alteration that are involved in tumor generation 57 . The SWI/SNF complex has an antagonistic relationship with polycomb repressive complex 2 (PRC2) in several tumors 45,46 . Indeed, histone methyltransferase EZH2 (part of the PRC2 complex) is essential for survival and growth of SMARCA4-deficient cancer cells 45 67 . Based on our data (Tazemetostat effects on mouse and human MB), we speculate that Tazemetostat could be a valuable treatment in Group 3 MB patients with high OTX2/c-MYC levels and low levels of SMARCA4 or SMARCA4 mutations. This would suggest to recruit patients affected by a specific MB subtype during some ongoing or future clinical trials. The described results prove the usefulness of human organoids to investigate molecular mechanisms underlying cancer development. More importantly, the creation of specific MB subgroup organoids demonstrates how the system can be used to model MB taking into consideration its molecular stratification. All together our data will pave the road to use organoids-based models for a more tailored drug screen and therapy. A single-nucleotide mutation (C2729T) was introduced in the human SMARCA4 sequence to generate the missense mutation T910M using a one-step PCR method. pPB CAG SMARCA4-Venus and pPB CAG Smarca4 T910M-Venus plasmids encoding for wild-type and mutant SMARCA4-Venus fusion proteins were generated by assembling Venus-coding sequence in frame with wild-type or mutant SMARCA4 coding sequences. The SMARCA4-Venus and SMARCA4 T910M-Venus constructs were cloned into into pPB CAG backbone without a IRES-GFP cassette. The PiggyBac donor plasmid pPB CAG Otx2 was generated by PCR amplification of Otx2 coding sequence from pPB TetO-Otx2 UBC-rtTA and cloned into into pPB CAG backbone without a IRES-GFP cassette. Three different double-stranded oligonucleotides coding for human SMARCA4 3′-UTR shRNA (target sequence shRNA1: 5′-GCTGTAGGACTGTTTGTGA-3′; target sequence shRNA2: 5′-CGGGTAGCAGCAGATGTAG-3′; target sequence shRNA3: 5′-TTGGGGAACACACGATACC-3′) and control shRNA (target sequence: 5′-ACTACCGTTGTTATAGGTG-3′) were cloned downstream of the U6 promoter into the pSCV2 plasmid according to the pSilencer instructions from Ambion.
Immunohistochemistry stainings were performed on rehydrated paraffin sections. Antigen retrieval was performed by incubating slices for 30 min in retrieval solution (10 mM sodium citrate, 0.5% Tween-20 (v/v), pH 6.0) at 98°C. Primary antibodies were incubated overnight at 4°C, and secondary antibodies for 1 h at room temperature in antibody solution. ABC solution was used 2 h at room temperature (Vectastain Elite ABC Kit Standard PK-6100). The sections were incubated with the substrate at room temperature until suitable staining was observed (DAB Peroxidase Substrate Kit, SK-4100). Nuclei were counterstained with hematoxylin.
The used antibodies are listed below: Imaging. Images were acquired with a Zeiss Axio Imager M2 (Axiocam MRc, Axiocam MRm), and for confocal imaging with either Leica TCS Sp5 or X-Light V2 confocal Imager optical. Images were processed using ImageJ software. Figures were prepared using Adobe Photoshop (Adobe).
Genes identification and Smarca4 expression levels in G3 MB. Genes that show differential expression (higher than 16-fold) compared with normal cerebellum have been identify using the online tool: https://hgserver1.amc.nl/cgi-bin/r2/main.cgi? &dscope&=&MB500&option&=&about_dscope# Smarca4 expression levels in human G3 MB tumors were obtained from the Pfister-223-MAS5.0-u133p2 data set. Smarca4 expression levels in wild-type hESderived neural stem cells were obtained from the data sets GSE9921 and GSE7178. Smarca4 expression levels in human normal adult cerebellum were obtained from the data set GSE3526. All data were normalized to the average Smarca4 expression level in either the GSE9921 or the GSE3526 data sets. Smarca4 mutational status in each G3 MB patient was identified using the somatic variants data collected in the same online tool.
Genomic DNA extraction. Organoids were lysed in lysis buffer (20 mM EDTA, 10 mM Tris, 200 mM NaCl, 0.2% Triton X-100, 100 µg/ml Proteinase K, pH 8.0) for 1 h at 37°C. Genomic DNA was extracted with phenol-chloroform and precipitated with isopropanol. Fig. 5 Cerebellar organoids electroporation with Otx2/c-MYC induces Group 3 MB in vivo. a Schematic representation of in vivo injection of modified cerebellar organoids. b DAPI staining and GFP immunofluorescence (Venus) of the sagittal brain section of nude mouse 1 month after injection of human cerebellar organoids electroporated with pPBase + pPBMyc + pPBOtx2 + pPBVenus. c Confocal images of GFP (Venus) and PCNA immunofluorescence of tumors in nude mouse 1 month after injection of human cerebellar organoids electroporated with pPBase + pPBMyc + pPBOtx2 + pPBVenus. d NPR3 immunohistochemistry and Hematoxylin staining of tumor in nude mouse 1 month after injection of human cerebellar organoids electroporated with pPBase + pPBMyc + pPBOtx2 + pPBVenus. e MDS (multidimensional scaling) analysis performed on the 1000 most variable probes of the whole-genome DNA methylation data shows a close similarity between organoids and group 3 MBs. Color legend of the MDS plot as follows: OM Organoids (Organoids_OM, black); GM Organoids (Organoids_GM, gray); WNT, Wingless MB (blue); SHH-A, Sonic Hedgehog MB-adulthood and childhood (red); SHH-B Sonic Hedgehog MB infant (dark red); G3, Group 3 MB (yellow); G4, Group 4 MB (green). f Hierarchical clustering and heatmap of beta values relative to the 39 high-quality CpG islands better discriminating MB subgroup in the Hovestadt set (Hovestadt et al.) 35 . The heatmap shows normalized methylation levels in organoid samples and MB samples. Clusters were obtained by means of Ward's minimum variance method, using the Euclidean distance. Color legend: OM Organoids (Organoids_OM, black); GM Organoids (Organoids_ GM, gray); WNT, Wingless MB (blue); SHH-A, Sonic Hedgehog MB-adulthood and childhood (red); SHH-B Sonic Hedgehog MB infant (dark red); G3, Group 3 MB (yellow); G4, Group 4 MB (green). Scale bars 1 mm in (b), 100 µm in (c, d). In vivo bioluminescence imaging. Mice were intraperitoneally administered 150 mg/kg D-Luciferin (Santa Cruz Biotechnology) 10 min before imaging. Animals were anesthetized with 2% isoflurane, and bioluminescent signal was captured using the In-vivo Xtreme system (Bruker). Mice were imaged weekly starting from 6 weeks after the transfection.
Organoids maintenance, modification, and injection. Human iPS cells (iPSC, ATCC-DYS0100) were maintained in self renewal on a layer of geltrex (Gibco, A14133-01), in E8 Basal Medium (Gibco, A15169-01) supplemented with E8 supplement (50 × ). All cells were mycoplasma free. iPSC were dissociated with EDTA (Invitrogen) 0.5 mM, pH 8.0, for 3 min incubation, to maintain cell clusters. Cerebellar organoids were cultured as described by Muguruma et al. 20 Ishida et al. 19 , and were electroporated at 35 days of differentiation protocol with 16. Histopathological evaluation. Two Otx2 + c-MYC mouse tumors were diagnosed by neuropathologists Francesca Gianno and Felice Giangaspero. In addition to standard hematoxylin and eosin staining, immunostaining was done on formalinfixed paraffin-embedded tumors after dewaxing and rehydrating slides.
Tumor spheroids for flow cytometry were harvested after drug treatment, and dissociated with 0.25% Trypsin-EDTA (Gibco) to a single-cell suspension. Cells for analysis of apoptosis were incubated for 30 min in culture medium to allow recovery from dissociation, then resuspended in PBS and stained with LIVE/DEAD Fixable Near-IR Stain (Invitrogen) for 20 min. Subsequently, cells were resuspended in annexin-binding buffer (10 mM HEPES, 140 mM NaCl, 2.5 mM CaCl 2 , pH 7.4) and stained for 15 min with Annexin V, Alexa Fluor 647 conjugate (Invitrogen). Cells for DNA content analysis were fixed with ice-cold 70% v/v ethanol at −20°C for at least 20 min. Fixed cells were stained with 10 µg/ml propidium iodide (Invitrogen) and treated with 100 µg/ml PureLink RNase A (Invitrogen) for 30 min at 37°C. Cells were analyzed using a FACSCanto (BD Biosciences) and data analysis was performed using FlowJo software (LLC).
Survival analysis. Survival analysis was performed calculating the lifespan in days of every mice injected with each specific combination of plasmids. Mice which were killed at the experimental endpoint or died due to undetermined causes during the study were censored in the analysis. Data were displayed using the Kaplan-Meier format, and statistical significance of the results was tested using the Log-rank (Mantel-Cox) test.
Statistical analysis. Organoids quantitative analysis: the data are presented as mean + s.e.m., for each condition 5-11 organoids were examined and at least 110-200 cells were quantified. Organoids electroporated with pPBase + pPBVenus (encoding for Venus) were considered as the positive control, and were used to set the parameters for cells count on ImageJ software. Data were compared using a unpaired Student's t test, two tails.
Tumor Spheroids data are presented as mean + s.d. of two biologically independent ex vivo experiments. In total, 120 spheroids were counted for each condition. After the drug treatment, brightfield and fluorescence images of the tumor spheroids were acquired, and tumor spheroid area was quantified using ImageJ. Data were compared using a unpaired Student's t test, two tails with Welch's correction.
DNA methylation profiling. DNA methylation profiling was performed, according to protocols approved by the institutional review board with written consent obtained from the patients' parents. Tumor areas with highest tumor cell content ( ≥ 70%) were selected for DNA extraction. Samples were analyzed using Illumina Infinium HumanMethylationEPIC BeadChip (EPIC) arrays according to the manufacturer's instructions, on Illumina iScan Platform. In detail, 250 ng or 500 ng of DNA was used as input material for fresh-frozen or formalin-fixed paraffin-embedded tissues, respectively. Generated methylation data were compared with the Heidelberg brain tumor classifier 34 (http://molecularneuropathology.org) to assign a subgroup score for the tumor compared with 91 different brain tumor entities. All tumors had a score of at least 0.8 in the reported methylation class.
EPIC BeadChip data were analyzed by means of R (V .3.4.3), using different packages: ChAMP pipeline (V.2.9.9) 71 for quality checks and filters, to calculate methylation levels and functionally annotate probes at the gene level. Multidimensional scaling (MDS) on the cohort samples was performed using cmdscale function, with Euclidean distance. Heatmap depicting normalized beta values was created by means of pheatmap function, using Ward's minimum variance method 72 , and Euclidean distance to cluster samples and probes. Lowquality CpG islands among the 48 ones identified from Hovestadt et al. 35 were removed from the analysis. Finally, bootstrap analyses were carried out using pvclust package 73 .
Gene expression analysis of electroporated organoids. Two biological replicates for each group (Untreated or Electroporated) were used. The total organoids RNAs were extracted using Trizol reagent (Invitrogen), subjected to DNase-I (Ambion) treatment and RNAs were depleted of ribosomal RNA. Sequencing libraries for whole transcriptome analysis were prepared using Stranded mRNA-Seq Library Preparation Kit. RNA-seq was performed on an Illumina HiSeq 2500 Sequencer using standard conditions at the Next-Generation Sequence Facility of University of Trento (CIBIO). The obtained reads were 75 bp long, paired ends, and 30 Millions on average for each sample. Quality control analysis was performed using FastQC (www. bioinformatics.babraham.ac.uk/projects/fastqc/). All the sample sequence reads were mapped with STAR aligner (v2.5.3) using recommend parameters. To provide an estimate of gene expression and compute differential gene expression, the reads were proportionally assigned to the human gene transcripts (ENSEMBL HG38), based on the mappings using HT-SEQ count (http://wwwhuber.embl.de/users/anders/HTSeq. Differential gene expression analysis was performed using the gene raw counts, within the R/Bioconductor edgeR package. The differential gene expression pipeline within the edgeR package was customized to estimate the dispersion parameter for each library using the biological group dispersion and identify DE genes between treated versus the control samples. log2(fold change) ≥ 1 and baseMean > 3 CPM were considered for differentially regulated genes and the p-value was adjusted for multiple testing using the Benjamini-Hochberg correction with a false discovery rate (FDR) ≤ 0.05. Differentially expressed gene lists obtained from low-level procedures were analyzed for functional associations. Data were analyzed through DAVID Bioinformatics Resources v6.8 using the suggested standard parameters.
Reporting summary. Further information on research design is available in the Nature Research Reporting Summary linked to this article.

Data availability
DNA Methylation Raw Data and Classifier Results. The data sets generated during and/ or analyzed during the current study are available in GEO: GSE128218. Token: sxefoyqubnivpkh.

Primary antibodies
Host    h Quantification of cerebellar organoids cells at day 41, electroporated at day 35 with pPBase + pPBMyc + pPBOtx2 + pPBVenus (OM), and treated for 5 days with either DMSO or Tazemetostat or GSK-126. Percentage of PCNA-positive cells between Venus-positive cells. a n = 2 biologically independent experiments. (b, c, e) n = 5 biologically independent experiments. g, h n = 5-11 biologically independent organoids. Error bars in (a, b, c, e) represent standard deviation, error bars in (g, h) represent standard error of the mean. Paired Student's t test, two tails. *p-value < 0.05, **p-value < 0.01.