Establishment and characterization of an orthotopic patient-derived Group 3 medulloblastoma model for preclinical drug evaluation

Medulloblastomas comprise a heterogeneous group of tumours and can be subdivided into four molecular subgroups (WNT, SHH, Group 3 and Group 4) with distinct prognosis, biological behaviour and implications for targeted therapies. Few experimental models exist of the aggressive and poorly characterized Group 3 tumours. In order to establish a reproducible transplantable Group 3 medulloblastoma model for preclinical therapeutic studies, we acquired a patient-derived tumour sphere culture and inoculated low-passage spheres into the cerebellums of NOD-scid mice. Mice developed symptoms of brain tumours with a latency of 17–18 weeks. Neurosphere cultures were re-established and serially transplanted for 3 generations, with a negative correlation between tumour latency and numbers of injected cells. Xenografts replicated the phenotype of the primary tumour, including high degree of clustering in DNA methylation analysis, high proliferation, expression of tumour markers, MYC amplification and elevated MYC expression, and sensitivity to the MYC inhibitor JQ1. Xenografts maintained maintained expression of tumour-derived VEGFA and stromal-derived COX-2. VEGFA, COX-2 and c-Myc are highly expressed in Group 3 compared to other medulloblastoma subgroups, suggesting that these molecules are relevant therapeutic targets in Group 3 medulloblastoma.


Results
Establishment of a transplantable cerebellar Group 3 medulloblastoma model. Initially, medulloblastoma tissue from MB-LU-181 was sub-grouped as Group 3 using the Illumina HumanMethylation450 array 22 . Low-passage tumour spheres derived from primary MB-LU-181 were orthotopically injected into NODscid mice and generated tumours with a latency of 17-18 weeks. Sphere cultures were re-established and serially transplanted for 3 generations, with a negative correlation between tumour latency and numbers of injected cells (Fig. 1a). Notably, inoculation of 1000 cells was enough to ensure 100% tumour penetrance within 10 weeks in third generation xenografts and one mouse developed medulloblastoma when ortothopically inoculated with only 10 MB-LU-181 cells (Fig. 1a).

Epigenetic, histologic and phenotypic profiling of transplanted medulloblastoma cells.
Genome-wide methylation assays of primary MB-LU-181 medulloblastoma and the corresponding neurospheres and xenografts resembled Group 3 medulloblastoma (Fig. 1b). As a control, primary medulloblastoma (MB-LU-187) and corresponding neurospheres clustered together and were classified as a Shh medulloblastoma (Fig. 1b). Using principal component analysis, the methylation of the primary samples, corresponding neurospheres and xenograft samples clustered together within subgroups, indicating a strong degree of stability between primary tumours, neurospheres and xenografts (Fig. 1c). Similarly, unsupervised hierarchal clustering also suggests a high degree of relationship between the primary tumour and corresponding neurosphere and xenografts (Fig. 1d). The most evident differences in genome-wide methylation were found between the Group 3 and Shh samples (Fig. 1c,d). The methylation pattern between the primary tumor and first neurosphere culture only exhibited minor changes, whereas, although similar, a more pronounced variation in the methylation pattern was observed between the primary tumor, the serial xenografts and third generation neurospheres (Fig. 1c,d). Copy number variant analysis of genome-wide methylation data showed MYC amplification in the primary tumor that was conserved in neurospheres cultures and xenograft transplantations ( Fig. 1e; Supplementary Fig. 1).
Xenografts were found to replicate medulloblastoma histology, with a homogenous dense mass of tumour cells displaying a high nuclei-to-cytoplasmic ratio (Fig. 2a) and high proliferation as determined by Ki-67 labelling ( Fig. 2c). In addition to the main tumour bulk, xenografted cells were sometimes seen lining the cerebellum and forming smaller secondary tumours (Fig. 2b). To determine if xenografts maintained the protein phenotype of the original tumour, we used immunofluorescent labelling of cryosections to evaluate the expression of a panel of lineage (nestin, nf-200 and GFAP) and putative tumour progenitor/stem cell markers (CD15, CD44 and CD133) in xenografts alongside primary tumour tissue (Fig. 2c). Xenografted cells replicated the phenotype of the original tumour, displaying high expression of nestin, nf-200 and CD133, and no expression of CD15, CD44, GFAP. We have also previously shown that expression of the neuronal progenitor marker CD24 22 is maintained in xenografts. Although medulloblastoma cells in both primary tumour and xenografts were GFAP-negative, xenografts were infiltrated by star shaped astrocytes that were interpreted as resident mouse astrocytes stained due to species cross-reactivity by the GFAP-antibody. The primary tumour was CD44-negative on tumour cells, but stained positive in clusters of CD45 + immune cells.
Both primary tumour and xenografts displayed high levels of MYC RNA (Fig. 3a) and MB-LU-181 neurospheres expressed high levels of c-Myc protein compared to DAOY medulloblastoma cells and human dermal fibroblasts cells (Fig. 3b,c). MB-LU-181 neurospheres were also 37-fold more sensitive to treatment with   (Fig. 3d). JQ1 treatment was accompanied by c-Myc and cyclin B protein down-regulation in MB-LU-181 cells (Fig. 3b,c).
Characterization of the microenvironment in medulloblastoma xenografts. The contribution of the microenvironment to tumour progression is well established, and the lack of human stroma consequently limits the experimental utility of culture-derived xenograft models. To some extent, host endothelial and immune cells may however replace functions of the human stroma. We characterized the vascular and inflammatory cell compartments of medulloblastoma xenografts by immunofluorescent staining of cryosections. Xenografts were devoid of human immune cells (hCD45 + , Fig. 2c) and human endothelial cells (hCD31 + , Fig. 4a), consistent with our observation that stromal cells do not survive following serial passaging in sphere-generating cell culturing conditions 21 . Instead, xenografts contained vast numbers of murine blood vessels (mCD31 + , Fig. 4b). CD31 + staining was entwined or lined with inducible nitric oxide synthase (iNOS) and cyclooxygenase-2 (COX-2) in both primary tumour and xenografts (representative images of CD31/COX-2 are shown in Fig. 4c).
NOD-scid mice lack mature B and T cells, whereas innate immunity is functional albeit low in activity. Occasional NK cells (NK1.1 + ) and granulocytes (Ly6G/Ly6C + ) could be detected in the normal compartments of mouse brains, but these cell types were not seen in close proximity to tumours (not shown). In contrast, xenografts were infiltrated with mouse myeloid cells (mCD45 + , F4/80 + , Fig. 4d-f) displaying markers indicative of a suppressive phenotype (iNOS − , not shown; COX-2 + , CD206 + , Fig. 4e,f). The same phenotype was observed in vessel-associated myeloid cells detected in the cerebellum of naïve NOD-scid (not shown). In comparison, only a subset of macrophages (CD68 + ) in the primary tumour displayed putative M2 markers (CD163 + or COX-2 + ) (Fig. 4e,f). CD8 + T cells were detected in or in close proximity to vessels in the primary tumour, but were rarely infiltrating cell dense tumour tissue (not shown).
We have previously described expression of COX-2 in primary medulloblastoma tissues and patient-derived cell cultures, and demonstrated that COX-2 inhibition reduces the growth of subcutaneous medulloblastoma xenografts 21,24 . In MB-LU-181 xenografts, COX-2 was predominantly stromal-associated (as exemplified in Fig. 4c,e), whereas the primary tumour MB-LU-181 also displayed areas of tumour cell-associated COX-2 staining in addition to staining on vessels and subsets of macrophages. To investigate the prerequisites for prostaglandin production in vivo, we also included microsomal prostaglandin E synthase-1 (mPGES-1), a downstream enzyme of COX-2, in the immune marker panel. mPGES-1 was detected in large clusters of the medulloblastoma cells of primary tumour and xenografts (Fig. 4g), but not in the COX-2 + stromal cell compartments.
The cytokine content of xenografts, alongside primary tumour and cultured spheres, was analysed on antibody-based multiarray platforms detecting 20 human and 10 mouse cytokines respectively. The primary tumour exhibited a wide range of both immune stimulatory and suppressive cytokines (Fig. 5a). Of the 10 analysed mouse-derived factors, 4 were detected in xenografts (Fig. 5a). IL-1β and KC/GRO were found in both xenografts and non-tumour bearing NOD-scid, whereas IL-6 and TNF-α were exclusively detected in xenograft tissue.
In cultured medulloblastoma spheres, only IL-8, IL-16 and VEGFA were detected. The secretion of these cytokines was maintained in the xenograft tissue. Expression of VEGFA in xenografted medulloblastoma cells was further confirmed with in situ hybridization (Fig. 5b).
Interestingly, we found that the tissue level of VEGFA in MB-LU-181 (depicted as Group 3 in graph) greatly exceeded VEGFA levels in 4 primary medulloblastomas of other molecular subgroups that were analysed in parallel (Fig. 5c). To investigate if the VEGFA observation has clinical relevance, we screened a dataset of 423 medulloblastomas for expression of VEGFA, and similarly found that VEGFA was preferentially expressed in Group 3 medulloblastomas compared to other subgroups (Anova p = 1.7e-17, Fig. 5d). We screened the same cohort for PTGS2 (COX-2) expression, and similarly found the highest expression in Group 3 medulloblastomas (Anova p = 1.2e-06). The expression of PTGES (mPGES-1) was similar in all subgroups (Anova p = 0.24) (Fig. 5e,f). Also, high levels of VEGFA were detected in both cells and supernatants from MB-LU-181 neurospheres compared to human dermal fibroblast (HDF) cells and cell lines from other medulloblastoma subgroups (Anova, p < 0.0001 for both cells and supernatants) (Fig. 5g,h).

Discussion
In this study, we describe the development and features of an orthotopic PDX model of Group 3 medulloblastoma (MB-LU-181), intended for future drug evaluation. The aggressive primary tumour displayed high-risk features, including overexpression of MYC and > 90% cell proliferation, and exhibited rapid clinical progression with < 220 days between clinical presentation and death of disease. Its derived xenografts largely recapitulated the epigenetic and phenotypic characteristics of medulloblastoma and maintained high proliferation rate, enabling serial tumour formation from a surprisingly low number of inoculated cells. In addition, xenografts maintained expression of VEGFA and stromal-derived COX-2. Both factors were overexpressed in Group 3 compared to other medulloblastoma subgroups, suggesting these factors to be relevant therapeutic targets for patients with Group 3 medulloblastoma.
To establish the PDX model, we inoculated low-passage tumour spheres into the cerebellum of NOD-scid mice. It has been suggested that NSG (NOD-scid Il2rg null ) mice should allow for higher tumour penetrance than NOD-scid mice, due to lack of NK cell activity 25 . However, we only occasionally detected NK cells in the brains of xenografted NOD-scid mice, suggesting that these cells are not actively involved in tumour rejection in this setting. Additional strategies previously used for generating orthotopic brain tumour PDX models include brain inoculation of fresh tumour material, either as a single-cell suspension or tissue chunks. All three methods have resulted in successful establishment of relevant brain tumour models 9,12,26 , although there is a lack of comparative studies indicating which method should be preferred. For logistic and financial reasons, freezing of tumour material prior to mouse inoculation is commonly required while awaiting histologic or molecular tumour classification. Here, we stored single-cell suspensions at − 80 °C for approximately six months before primary xenograft establishment, and briefly cultured thawed tumour cells to ensure viability and active proliferation.
MB-LU-181 xenografts replicated the phenotype of the primary tumour, including histology, expression and amplification of MYC and presence or absence of distinct tumour markers. The Group 3 medulloblastoma methylation pattern remained stable throughout serial culturing and xenografting. The highest degree of variation in the genome-wide methylation pattern was observed between the primary tumour and xenografts and corresponding neurosphere derived from the xenograft tumours. This was probably due to the transfer of human tumour cells to a mouse cellular niche, which necessitates adaption of the human cells to a mouse microenvironment. However, these findings are similar to the demonstration that medulloblastomas, including the Group 3 subtype, preserve their methylation subgroup pattern between primary and metastatic compartments 27 . The neurospheres isolated from xenografted tumours exhibited similar epigenetic and phenotypic characteristics compared to the primary tumour, and the neurospheres were highly sensitive to treatment with the MYC inhibitor JQ1.
Interestingly, in the xenografted tumours we observed medulloblastoma cells lining the cerebellum and forming secondary tumours distant from the primary tumour, corroborating reports of maintained migratory capacity of medulloblastoma cells following orthotopic injection 28 . Since our primary tumour sample was obtained at a time when the patient not yet displayed evidence of established metastases, our model may mimic the early events in metastatic dissemination that eventually led to fatal relapse in the patient. In the study by Dietl et al., the spread of xenografted Group 3 medulloblastoma cells was not mimicked by other transplanted models, suggesting that the migratory features of orthotopic Group 3 models reflects biological behaviour, rather than e.g. CSF spread during cell inoculation. Additional studies will be conducted to search for extracranial metastases in the MB-LU-181 model.
In a comprehensive comparison of soluble factors in tumour and xenograft tissue, a range of human proand anti-inflammatory factors was detected in the primary medulloblastoma, whereas only IL-8, IL-16 and VEGFA were seen in cultured cells and xenografts, suggesting that most cytokines in the primary tissue were stromal-derived. IL-8 has been attributed both immunosuppressive, pro-angiogenic and direct growth-promoting functions in brain tumours (as reviewed by Brat et al. 26 ). The lack of conserved IL-8 signalling between mice and human 29 indicates that the maintained IL-8 secretion in xenografts is not related to tumour/stroma interactions. Rather, IL-8 may play a role as a tumour-specific autocrine or paracrine factor that promotes medulloblastoma proliferation or survival, as previously demonstrated for other brain tumour types 30,31 . The role of IL-16 in brain tumours has not been elucidated, although occasional reports have described IL-16 secretion by myeloid cells in astrocytic tumours 32 . Our findings suggest that IL-16 also can be produced by tumour cells. Expression of IL-16 has been described in several peripheral cancers such as breast and prostate cancer 33,34 ; mechanistic studies are scarce, but it can be speculated that IL-16 contribute to inflammation at tumour sites, for instance by recruitment of leucocytes 33,35 .
A critical limit of PDX models is the lack of human stroma, and detailed knowledge about the microenvironment of each individual model is therefore crucial for evaluation of drug strategies affecting tumour-stromal interactions. In the current study, we initially expanded primary tumour cells in sphere-generating culturing conditions, in which non-neoplastic human cells are rapidly out-sorted 21 . Alternative approaches to generate PDX models include injection of fresh surgical specimen, which initially preserves the human stromal compartments. However, this strategy is not feasible for serial passaging, since human stromal cells of the primary xenograft are replaced by its murine counterparts after one or a few passages in vivo 11,36,37 .
Experimental data from us and others 11 show that the murine vasculature is sufficient to sustain growth of xenografted medulloblastoma cells in the absence of human endothelial cells. The high expression of VEGFA in xenografts and neurospheres indicates that it is one of several potential conserved signalling factors that could direct angiogenesis in this model. VEGF is a key mediator of the hypoxic response in tumours, and VEGFA is one of several angiogenic factors responsible for neovascularization and growth of medulloblastoma 38 . Our results from a gene expression analysis of 423 medulloblastomas further demonstrates enhanced expression of VEGFA in Group 3 compared to other medulloblastoma subgroups, indicating that this patient group in particular may benefit from VEGF-targeted therapies. So far, anti-VEGF compounds have shown limited clinical success in the treatment of adult and paediatric brain tumours, tentatively due to e.g. compensatory or parallel angiogenic mechanisms, skewed patient selection or the use of antibodies directed against both anti-and pro-angiogenic VEGF isoforms 39 . Nevertheless, targeting of angiogenic pathways remains a promising treatment approach for brain tumour patients in combination with other treatment strategies such as cytotoxic agents 40 . In this context, PDX models enable detailed in vivo studies of angiogenic signalling and treatment resistance 41 , as well as the subsequent design of more specific drug strategies that are optimal for distinct tumour subsets.
In addition to VEGF, the COX-2/mPGES-1/PGE 2 pathway has been investigated as a therapeutic target for a number of cancer forms, including paediatric brain tumours 24 . COX-2 is the rate-limiting enzyme for production SCIeNTIFIC REPORTS | 7:46366 | DOI: 10.1038/srep46366 of PGE 2 via the terminal enzyme mPGES-1. PGE 2 has been attributed a wide range of tumour-promoting functions, including increased survival, proliferation, invasiveness and chemo-resistance of tumour cells, as well as potent inhibition of anti-tumour immune effector functions 42,43 . We have previously shown that inhibition of COX-2 activity can reduce the growth of subcutaneous neuroblastoma and medulloblastoma PDXs 24,44 , and the enhanced expression of COX-2 in patients with Group 3 medulloblastoma suggest that therapeutic intervention may be particularly useful for this patient group. Several studies have demonstrated a direct link between COX-2 and VEGF signalling in tumour cells. PGE 2 may induce the expression of VEGF and thereby contribute to the angiogenic response and subsequent tumour evasion 45,46 . Conversely, VEGF may induce COX-2 expression in endothelial cells 47 . More recently, it was suggested that VEGF and COX-2 could act as independent regulators of angiogenesis, and that expression of COX-2 facilitates VEGF activity 48 . Consequently, COX-2 expression may partly be responsible for tumour resistance to VEGF-therapies, and combined inhibition of VEGF and COX-2 is a possible treatment strategy that may of particular interest for Group 3 medulloblastoma patients.
Interestingly, we found that distinct cell populations were responsible for the production of COX-2 and mPGES-1 in xenografts, where COX-2 was vessel-and macrophage-derived, while mPGES-1 detection was restricted to tumour cells. The primary tumour displayed the same staining patterns, although COX-2 was also seen on subsets of tumour cells. We have repeatedly observed distinct cellular origin of mPGES-1 and COX-2 in neuroblastoma 44 , experimental glioma 49 and here in medulloblastoma, and demonstrated that inhibition of COX-2 has a therapeutic effect in vivo even though COX-2 and mPGES-1 are expressed in different cell types 49 . These data suggests that intermediate metabolites are transferred between cell types, as previously described 50 , and the MB-LU-181 model will be useful to study such interactions. Still, it has to be confirmed that PGE 2 is indeed produced in xenografts.
Human medulloblastomas generally exhibit a quiescent or suppressed immune cell response compared to other paediatric brain tumour types, including few T cells and myeloid cells and a general lack of immune activation 51 . Immunodeficient mouse models may therefore better mimic the microenvironment of medulloblastomas than of brain tumour types with a more prominent immune activation. The most frequent immune cells detected in both primary tumour and xenografts were myeloid cells. Based on established markers of myeloid polarization, xenograft-infiltrating mouse myeloid cells displayed a suppressed/inactivated phenotype, reminiscent of myeloid cells detected in naïve NOD-scid. Even so, tumour-bearing mice displayed an up-regulation of IL-6 and TNF-α , which are key pro-inflammatory cytokines in macrophage activation 52 -suggesting that murine macrophages indeed have the functional capacity to respond to inflammatory stimuli (either the presence of tumour cells or tissue injury induced by inoculation). It is however unclear if the response is sufficient to have an antitumor effect.
It is also not clear at this point to what extent the murine macrophage response mimics the clinical situation. Cytokine profiling of the primary tumour identified human IL-6 and TNF-α -consistent with the corresponding murine factors found in the mouse tissue, and indicative of pro-inflammatory response -but also a range of other factors that could be derived from both innate and adaptive immune cells. The predominant phenotype of myeloid cells in the primary tumour could not be clearly determined in this study; tentative M2 marker CD163 were absent on most CD68 cells, and a cytokine profiling identified a range of factors that could be derived from pro-inflammatory (IL-1, IL-6, IL-12/23, IL-15, TNF-α ) and suppressive (IL-8, VEGFA) cells respectively 53 . Even so, studies of clinical samples and syngeneic mouse models show that Group 3 and Group 4 tumours are associated with low macrophage activity compared to other medulloblastoma subgroups 54,55 , and it would therefore be highly relevant to compare our results to matched clinical samples and subgroup-specific PDX models generated in the same experimental setting.
In summary, we describe a novel orthotopic MYC-driven PDX model of Group 3 medulloblastoma. The markedly poor outcome for patients with Group 3 medulloblastoma highlights the limits of current treatment protocols and the need for new therapeutic strategies. The PDX model described here represents a rare opportunity to investigate the biology of a particularly aggressive medulloblastoma, and evaluate novel drug strategies for a tumour that proved to be incurable by standard treatment.

Methods
Tumour sample used for xenografts. All experiments were performed in accordance with national regulations and were approved by the Local Ethical Review Board of Lund, Sweden (ETIK2008/642) and the Ethical Research Board of the Medical Faculty at Lund University, Lund, Sweden (serial number LU1028-03). All patients and/or their parents gave their informed consent prior to inclusion in the study. Tumour tissue was collected from a 4-year-old male medulloblastoma patient and coded as MB-LU-181. Tumour tissue from resected surgical material was split in four parts that were (a) snap-frozen and stored at − 80 °C, (b) fixed in NBF for routine H&E-staining, (c) frozen in liquid N 2 precooled isopentane (− 55 °C, VWR International AB, Lund, Sweden) for cryosectioning, and (d) used to establish a cell culture and subsequently xenografts, see below. Adjacent tumour tissue was sent for routine PAD, where the tumour was diagnosed as a medulloblastoma. MB-LU-181 was initially assigned a Group 3 affiliation by the methylation profile using Illumina Infinium 450k methylation array of snap-frozen tissue, as previously described 22 . Establishment of orthotopic medulloblastoma xenografts. All  Sweden). Following sphere formation, spheres were mechanically passaged and further propagated as previously described 21 . Cells were kept in culture for approximately one week before the first transplantation and for 2-3 weeks between in vivo passages, and were passaged approximately once a week.
Immediately prior to in vivo transplantation, spheres were dissociated, counted and kept on ice in Hank's Balanced Salt Solution (VWR). Initially, eight weeks old female NOD-scid mice (NOD/MrkBomTac-Prkdcscid) were anesthetized with isoflurane and mounted in a stereotactic frame (David Kopf Instruments). A 6 μ l cell suspension containing 20,000 low-passage tumour cells was orthotopically injected into the cerebellum (n = 2) at median/lateral + 1.0 mm, anterior/posterior − 2.0 mm, dorsal/ventral − 2.5 mm relative to lambda, with the needle at 30° using a 26 ga Hamilton syringe (Sigma Aldrich). The needle was left in situ for 1 min after lowering the needle and 2 min after the injection of the cell suspension, which was performed with a speed of 2 μ l per min. Neurological symptoms and weight loss indicative of brain tumour growth appeared after 17 and 18 weeks, respectively. Mice were then euthanized by carbon dioxide followed by cervical dislocation and cerebellar tissue was dissected into two parts, where one was used to re-establish sphere cultures for serial orthotopic transplantation (using the same protocol as described for the patient cell culture). Cell viability assay. Cell viability was measured using WST-1 (ROCHE, Mannheim, Germany) following the manufacturer's instructions. Briefly, cells were seed in 96-well plates (TPP ® , Trasadingen, Switzerland) and incubated overnight before the indicated concentrations of JQ1 were added. After 72 h, the WST-1 reagent was added and absorbance was measured. Three independent experiments with 8 replicates per conditions were performed. Data are expressed as relative survival compared to DMSO-treated cells, and IC 50 values were compared with unpaired t-test. The IC 50 was determined using non-linear regression analysis on effect-log concentration curves (GraphPad Prism).

Neurospheres, cell lines and in vitro
In vitro VEGFA measurement. VEGFA was measured in vitro using the Human VEGFA ELISA Kit (Thermo Scientific, Rockford, USA) in accordance with the manufacturer's instructions. Briefly, MB-LU-181 neurospheres were seeded in Corning Cell-Bind 6-well plate (Corning, VWR, Stockholm, Sweden). As a normal reference, HDF cells were seeded in 10 cm plates (Sarstedt, Nümbrecht, Germany) and incubated over night to allow the cells to attach. Then, the medium was replaced by OPTI-MEM and supernatants and pellets were collected after 72 h of incubation. Total protein content from pellets was obtained by lysis with RIPA buffer (Thermo Scientific, Rockford, USA). The protein concentration of pellets and supernatants was determined using the BCA assay kit (Thermo Scientific, Rockford, USA). The VEGFA expression was measured with ELISA and values were normalized to the total protein content. Three independent samples for each cell line and time point were obtained and processed. Values were compared with two-way ANOVA with Bonferroni as post-test.
Western blotting. Western blot analyses were used to assess protein expression of c-Myc and cyclin B1.
Histology and immunofluorescent labelling of tumour and xenograft sections. NBF-fixed MB-LU-181 tissue and derived xenografts were stained with routine HTX/eosin-staining and histology was evaluated by a trained pathologist. Immunofluorescent labeling of cryosections was performed as previously described 49 . The following primary antibodies were used: mouse anti-CD133/2 (5 μ g/ml, Milteney Biotec GmbH, Bergisch Gladbach, Germany); FITC-mouse anti-GFAP (5 μ g/ml), PE-mouse anti-human CD8 (diluted Genomic DNA from the primary MB-LU-181 tumour, subsequent generations of sphere cultures, serially transplanted mice was extracted from primary tumour tissue, dissected xenografts, and cultured spheres using GenElute ™ Mammalian Genomic DNA Miniprep Kit (Sigma-Aldrich, Sweden) according to the manufacturer's specifications. The concentration of DNA was measured with Nanodrop. Genome-wide methylation profiling was performed using the InfiniumEPIC BeadChip platform (Illumina, San Diego, California, USA) by the SNP&SEQ Technology Platform in Uppsala, Sweden (www.genotyping.se). DNA from another primary medulloblastoma (MB-LU-187) and its first-generation cultured sphere was also analysed.
Raw IDAT files were imported and analysed using the Bioconductor packages minfi 58 and RnBeads 59 . Methylation signals were preprocessed and normalized with the preprocessIllumina command and data quality was evaluated using the getQC feature. Normalized beta values were used to classify the samples using the MethPed algorithm 60 , to generate PCA plots, and to perform unsupervised hierarchical clustering. Copy number profiles are based on the Bioconductor package conumee and were generated using a publicly available methylation profiling tool developed and hosted by the DKFZ and University of Heidelberg (http://www.molecularneuropathology.org).