A Novel Method for Rapid Molecular Subgrouping of Medulloblastoma

Medulloblastoma is the most common malignant brain tumor of childhood (1–3). Over the last decade, integrative genomic studies have radically changed the understanding of the biology underlying medulloblastoma pathogenesis, revealing a considerably more heterogeneous disease than previously thought. These studies have identified four principal molecular subgroups, named Wingless (WNT), Sonic Hedgehog (SHH), group 3, and group 4, with distinctive clinicopathologic and molecular features that proved to be significantly better correlated with prognosis than classical stratification (4–12).

The WNT and SHH subgroups are characterized by activating mutations that affect critical regulators of the corresponding signaling pathways, whereas group 3 and group 4 have a low incidence of recurring mutations but are characterized by recurrent, partly overlapping, chromosomal alterations. The distinct genetic features, demographics, and metastatic pattern of these molecular subgroups are associated with diverse clinical outcomes. Patients with WNT medulloblastoma have an excellent prognosis with current therapy schemes (5-year event-free survival greater than 90%) and are currently considered for controlled reduction of treatment (https://clinicaltrials.gov). The prognosis of SHH-activated medulloblastomas is largely dependent on patient's age and specific genetic features where children with TP53 mutated SHH tumors have poorer outcome. Subgroup-driven clinical trials are currently being conducted aimed at the estimation of the efficacy of SHH pathway inhibitors (e.g., vismodegib) at diagnosis or in recurrent or refractory SHH-activated tumors (https://clinicaltrials.gov). Patients with group 3 tumors have the most unfavorable prognosis, especially when associated with MYC amplification, while group 4 has an intermediate prognosis. Specific targeted therapeutic strategies for group 3 and group 4 are currently not available and are consequently desperately needed.

These molecular entities have become increasingly important to refine risk stratification, and will be of significant help to improve treatment adjusted to the different risk categories but also to design clinical studies. The revised 2016 World Health Organization Classification of Brain Tumors has incorporated to the histologic classification of medulloblastoma the genetically defined entities WNT, SHH (-TP53-mutant; -TP53-wild-type) and non-WNT/non-SHH, where group 3 and group 4 are provisionally included as one entity (13, 14).

Genome-wide methylation array and/or gene expression profiling are currently considered the gold standard for the classification of patients with medulloblastoma (11, 14, 15). These are robust microarray and multigene-based classification methods, however, the implementation of these approaches in daily clinical practice is challenging for the majority of centers treating patients with brain tumors. IHC-based classification has also been used as an alternative subgrouping method, principally to detect WNT and SHH subgroups, however, it has proven to be difficult to standardize in the clinical setting, in part, due to the lack of specificity (14).

Here, we have developed and validated a robust clinically applicable approach for molecular subgrouping of WNT, SHH and non-WNT/non-SHH tumors based on a reduced panel of highly specific epigenetic markers. Similarly, we have developed an epigenetic classifier specific for group 3 and group 4 tumors. The characteristics of the epigenetic markers that we have selected enable the analysis of single DNA samples obtained from fresh-frozen or formalin-fixed paraffin-embedded (FFPE) biopsies of primary, metastasis or relapse compartments, using diverse DNA detection methodologies that can be implemented in daily clinical practice by most centers treating patients with brain tumors.

In this study, we have included 122 DNA samples from frozen (n = 102) and FFPE (n = 20) biopsies of clinically annotated primary medulloblastoma obtained from Hospital Sant Joan de Déu (Barcelona, Spain) and Sick Children's Hospital of Toronto, Canada (Supplementary Table S1). All patients included in the study were younger than or 18 years of age at the moment of diagnosis. Molecular subgroup affiliation as determined in recently published genomic studies was available for all cases included in this study (9, 11). Fresh-frozen samples from posterior fossa non-medulloblastoma tumors [atypical teratoid/rhabdoid tumor (AT/RT), embryonal tumor with multilayered rosettes (ETMR)], and a normal pediatric cerebellum sample were also included in the study. This study was approved by the Institutional Review Boards. Written informed consent was obtained from patients/guardians before collection of samples.

We analyzed genome methylation data of 1,455 specimens comprising primary medulloblastoma tumors (n = 913, including three datasets of 106 fresh-frozen, 169 FFPE, and 638 frozen samples of patients ≤18 years) and nonmedulloblastoma tumors (n = 457) and normal human tissues and cells (n = 85). Illumina Infinium HumanMethylation450 BeadChip (HM450K) datasets included in this study have been previously reported (Supplementary Table S2). For comparison between the HM450k and the Illumina Methylation EPIC BeadChip 850K array platform, we used a publicly available dataset analyzed with both platforms (ref. 16; Supplementary Table S2; Supplementary Methods). Normalized data are publicly available at the NCBI Gene Expression Omnibus (GEO) data repository (Supplementary Table S2). To enable a robust and comprehensive analysis, RAW methylation data (two color iDAT files) were included in the study. iDAT files of the abovementioned datasets and subgroup affiliation data, have been kindly provided by research groups of the International Cancer Genome Consortium (ICGC) PedBrain Tumor Research Project, The Canadian Paediatric Cancer Genome Consortium (CPCGC), McGill Integrated Cancer Research Program (Montreal), Dana-Farber Cancer Institute (Boston), Scripps Research Institute (La Jolla), and the Friedman Brain Institute at Mount Sinai Hospital (New York).

Microarray raw data (iDAT files) of samples (tumors, normal samples) were analyzed by minfi package available through Bioconductor (17). To exclude technical biases, we used an optimized pipeline with several filters, as described previously (18, 19). From the initial dataset of 485,512 sites (excluding probes detecting SNPs), we removed those with poor detection P values (P > 0.01) and those with sex-specific DNA methylation (n = 6,926). The remaining sites (n = 478,129) were used for downstream analyses. Single cytosine methylation values (β values) were calculated as the ratio of the methylated signal intensity to the sum of methylated and unmethylated signals.

DNA methylation changes were analyzed as previously described (18, 19). Standard deviation (SD) density plots were used to define the most variable CpGs. Unsupervised analyses were performed by principal component analysis and hierarchical clustering using the Euclidean metric, as described previously (refs. 18, 19; Supplementary Methods).

For the identification of CpGs with the most significant differences among subgroups, we used the following criteria: (i) the minimum absolute difference between DNA methylation values of CpG sites within each subgroup, less than 0.1 SD, (ii) the maximal bimodal methylation values between subgroups, and (iii) a subgroup-specific methylation profile in comparison with the rest of the medulloblastoma subgroups.

The misclassification rate for any given combinations of the selected CpGs was assessed using a Lineal Discriminant Analysis (LDA) function by MASS package (R software), as reported previously (Supplementary Methods; ref. 20). Among the combinations that showed a good performance, we selected one with the best compromise between the error rate and number of variables. We trained the LDA model using the DNA methylation values of the selected combination of CpG sites on the training cohort (n = 106 medulloblastoma cases). By applying this model to the different test cohorts, we calculated the misclassification rate of our methylation signatures. Other class prediction methods such as Support Vector Machine (SVM) were used to further evaluate the capacity of the combination of CpGs to accurately classify samples (Supplementary Methods).

To identify a reduced set of differentially methylated CpG sites that could enable the classification of medulloblastoma into the four principal subgroups, we reanalyzed previously published genome-wide DNA methylation array data including primary medulloblastoma tumors (n = 106) as well as diverse non-medulloblastoma tumors (n = 457) and normal human tissues and cells (n = 85; Supplementary Table S2). Molecular subgroup affiliation as determined in recently published array studies was available for all the medulloblastoma tumors included in this study (Supplementary Table S2).

The training cohort comprised 106 fresh-frozen medulloblastoma tumors, including 18 WNT, 21 SHH, 29 group 3, and 38 group 4, collected within the ICGC PedBrain Tumor Project (12). Unsupervised hierarchical clustering and principal component analysis (PCA) using all CpG sites within a SD >0.3 (n = 5,904 most variable CpGs, 1.2%) clearly identified the presence of four subgroups (Supplementary Fig. S1A–S1C). Consistent with the previously published study with this dataset, the WNT and SHH groups were clearly distinct, whereas group 3 and group 4 were closely related (Supplementary Fig. S1B and S1C; ref. 12).

We next focused our analysis on reducing the dimension and complexity of our initial set of 5,904 CpG sites with variable methylation to select a more manageable group of CpG sites allowing us to classify patients with similar accuracy. We aimed to identify CpG sites with a consistent methylation profile within each subgroup (SD < 0.1), but clearly divergent from the other subgroups (maximum difference from mean methylation level of other groups). This approach enabled us to identify CpG sites with distinctive methylation signature for the WNT and SHH subgroups, whereas the heterogeneous and overlapping methylation patterns of group 3 and group 4 impeded the identification of consistent methylation profiles across each of these two entities. group 3 and group 4 were thus included within a single group named non-WNT/non-SHH, similar to the recently revised 2016 WHO classification for medulloblastoma (13). We applied our selection criteria and analyzed the classification capacity of any given combination of the eligible CpG sites using a Lineal Discriminant Analysis (LDA) method. From the resulting combinations of CpGs, there were multiple with a very good performance. Among these we selected one with six CpG sites for subsequent analyses and subgroup prediction, as they showed a good compromise between error rate (0%) and number of variables (Table 1; Fig. 1A). The prediction capacity of the selected CpGs to accurately classify samples was evaluated using other class prediction methods such as Support Vector Machine (SVM; Supplementary Methods). We observed equivalent accurate prediction results, the SVM reliably assigned 100% of cases confirming the robust performance of the combination of the selected six CpGs (data not shown).

The identified cytosines were located in the gene body of LHX6 (cg25542041), USP40 (cg12925355), CHTF18 (cg10333416), KIAA1549 (cg01268345), 5′-UTR of AKAP6 (cg18849583), and a long intergenic nonprotein coding RNA 2178 (cg02227036), the majority affecting enhancer (Enh, EnhBiv) sites (ref. 21; Table 1). The varying methylation levels of the six cytosines were not translated into gene expression changes, suggesting that they may not have functional impact, but could represent a stable molecular mark of the cellular origin of each subgroup, as described previously (20).

PCA analysis of the training cohort using the selected panel of six CpG sites (named ^EpiWNT-SHH) showed a similar clustering to those obtained using the 5,904 most variable CpGs (Fig. 2). All samples could be assigned to a subgroup with an excellent degree of concordance (100%) with previously assigned methylation subgroup classification (Fig. 2).

As the HM450K array has been discontinued and replaced by the Illumina Methylation EPIC BeadChip 850K array, we performed a direct comparison using a publicly available dataset including five medulloblastoma cases analyzed with both platforms (Supplementary Methods; ref. 16). The HM450K and the EPIC850K platforms showed comparable DNA methylation profiles for the CpGs included in the ^EpiWNT-SHH classifier demonstrating the equivalent performance of the two platforms (Supplementary Fig. S2A and S2C).

The specificity of the panel ^EpiWNT-SHH was investigated using diverse publicly available HM450K datasets and was found to be highly specific for WNT, SHH, and non-WNT/non-SHH medulloblastoma tumors; none of the other tumors or normal human tissues and cells showed a similar methylation pattern (Supplementary Fig. S3A). The specificity was also evaluated by bisulfite pyrosequencing and direct bisulfite-sequencing (BSP) in a cohort of samples including one normal cerebellum and other pediatric brain tumors from the posterior fossa which could histologically mimic medulloblastoma, such as AT/RT (atypical teratoid/rhabdoid tumor) and ETMR (embryonal tumor with multilayered rosettes). None of the samples showed a methylation pattern equivalent to any of the molecular subgroups of medulloblastoma (data not shown).

Given that tumor sample material is sometimes scarce (small biopsies), we investigated the effect of DNA input amounts on bisulfite conversion assay performance and the impact on the resulting methylation pattern using low-input titration series ranging from 100 ng to 10 ng of DNA from fresh-frozen and FFPE of a medulloblastoma tumor (SHH subgroup) included in the study cohort (HSJD_09). Successful PCR amplification was obtained using input amounts of bisulfite converted DNA as low as 10 ng for fresh-frozen and FFPE tumor DNA samples. Direct bisulfite-sequencing and bisulfite pyrosequencing showed an excellent agreement with the expected methylation pattern at all the DNA input quantities (Supplementary Fig. S4). These results revealed that the epigenetic biomarkers included in the classifier perform well and are appropriate for molecular subgrouping even when sample material is limited.

Recent studies have shown that the HM450K array is suitable for the analysis of DNA from FFPE material (12). We tested panel ^EpiWNT-SHH on a DNA methylation dataset of 169 FFPE medulloblastoma samples for which subgroup affiliation was available (15 WNT, 39 SHH, 42 group 3, and 73 group 4; ref. 12). PCA using FFPE methylation data exclusively of the six CpG sites of panel ^EpiWNT-SHH reproduced clusters equivalent to those obtained with the training cohort (Supplementary Fig. S5A). The LDA-based classifier using FFPE methylation data confirmed the high prediction capacity of ^EpiWNT-SHH, allowing the correct classification of WNT, SHH, and non-WNT/non-SHH tumor subgroups with an excellent concordance with previously reported subgroup affiliation of matching samples [100% agreement (95% confidence interval (CI), 97.8%–100%); k = 1 (95% CI, 1–1; Supplementary Fig. S5A)]. These results show that our methylation signature is independent of the type of material and can thus be applied to DNA methylation data obtained from FFPE specimens.

Next, we tested panel ^EpiWNT-SHH on an independent DNA methylation dataset of recently published 638 frozen medulloblastoma samples (including 52 WNT, 151 SHH, 131 group 3, and 304 group 4 tumors of patients ≤ 18 years of age; ref. 15). Three cases did not reach our quality criteria and were thus excluded. By applying the LDA classifier on the DNA methylation data of the ^EpiWNT-SHH CpGs, we were able to classify all cases (635 cases) with an excellent concordance [99.4% agreement (95% CI, 98.4%–99.8%); k = 0.99 (95% CI, 0.97–1)] with the previously reported molecular subgrouping (15). The four discordant cases, previously reported as WNT and three SHH, were classified as non-WNT/non-SHH tumors by the ^EpiWNT-SHH panel. However, these cases showed unspecific methylation patterns with our classifier, which did not clearly fit to any of the medulloblastoma subgroups (Supplementary Fig. S5B).

We next sought to validate panel ^EpiWNT-SHH using DNA from frozen (n = 102) and FFPE (n = 20) samples by using alternative methods such as bisulfite pyrosequencing and direct bisulfite-sequencing (BSP) (Supplementary Fig. S6). Molecular subgroup affiliation was available as determined in recently published array studies (12). The classification of samples into WNT, SHH or non-WNT/non-SHH subgroups by pyrosequencing and applying the LDA prediction model showed an excellent match with subgroup affiliation for frozen [100% agreement (95% CI, 96.4%–100%); k = 1 (95% CI, 1–1)] and FFPE [100% agreement (95% CI, 83.2%–100%); k = 1 (95% CI, 1–1)] samples. This cohort included a medulloblastoma tumor associated with the Gorlin syndrome (germline mutation in the Shh receptor PTCH), which was correctly classified as a SHH subgroup tumor.

A portion of the samples (98 frozen and 20 FFPE) were also analyzed by BSP. The subgroup-specific bimodal profile [either highly methylated (mean subgroup β-value >0.75) or greatly unmethylated (mean subgroup β-values <0.15)] of the methylation signature permitted to obtain sequences with peaks that clearly detected the methylated or unmethylated status of each of the CpG sites biomarkers (Supplementary Fig. S6). This enabled a rapid and easy analysis of the epigenetic marks, allowing accurate molecular subgrouping for both frozen [100% agreement (95% CI, 96.3%–100%); k = 1 (95% CI, 1–1)] and FFPE [100% agreement (95% CI, 83.2%–100%); k = 1 (95% CI, 1–1)] DNA samples. These results show that our simplified methylation signature has a good discriminatory capacity that allows for similar performance using fresh-frozen and FFPE samples analyzed using diverse DNA methylation detection methods.

Finally, we analyzed the stability of our epigenetic markers over the course of the clinical evolution of tumors by comparing samples of four patients with medulloblastoma (two group 3, one group 4, and one SHH) obtained from primary and/or metastasis at the moment of diagnosis and at relapse. In all cases, the methylation profile for the subgroup affiliation was maintained at metastasis as in the primary tumor, and at relapse/progression as at diagnosis suggesting that the epigenetic markers remain stable across tumor compartments (primary and metastasis) and are not influenced by treatment.

We next focused our analysis on the methylation profiles of group 3 and group 4 tumors. We proceeded to analyze the methylation data of group 3 and group 4 samples included in the training cohort (n = 67) using the most variable CpG sites (SD > 0.3; 2,612 CpGs). Similarly to previous reports, PCA analysis at k = 2 showed two adjacent subgroups with a small fraction of tumors located at the boundary between group 3 and group 4 (Supplementary Fig. S7) (14). PCA k = 3 outlined three subgroups, group 3 (22/67, 33%), group 4 (27/67, 40%), and a third mixed group 3–4 including seven group 3 and 11 group 4 tumors (18/67, 27%; Supplementary Fig. S7). We proceeded to analyze the methylation profiles of the clearly defined group 3 and group 4 tumors, excluding the mixed group 3–4. By applying our selection criteria (SD < 0.1 within each group and the largest delta between the two groups) and analyzing the classification rate of various combinations of CpGs using the LDA-based classification method, we were able to select five differentially methylated CpG probes that showed a consistently methylated profile for group 3 and unmethylated for group 4 (Fig. 1B). These cytosines were located in intronic regions of RPTOR (cg09929238 and cg08129331) and RIMS2 (cg12565585) genes, as well as the 3′-UTR region of VPS37B (cg13548946) and an intergenic region in chromosome 12 (cg05679609; Table 1). These methylation changes were not found to be associated with significant variations of the expression of targeted genes.

Similar to panel ^EpiWNT-SHH, the methylation profile of the five CpG probes was found to be highly specific for group 3 and group 4 when compared with WNT and SHH subgroup medulloblastoma tumors and diverse tumors and normal human tissues (Supplementary Fig. S3B).

The five CpG probes (named ^EpiG3-G4) were tested using DNA methylation data of all group 3 (n = 42) and group 4 tumors (n = 73) included in the FFPE medulloblastoma cohort (12). PCA analysis using panel ^EpiG3-G4 CpGs showed two closely related, albeit not overlapping groups of samples. By applying the LDA classifier model, we were able to classify all samples and assign the correct subgroup to 112 of the 115 samples [97.4% agreement (95% CI, 92.6%–99.5%); k = 0.94 (95% CI, 0.88–1; Supplementary Fig. S8A)]; three cases with an intermediate methylation profile were found to be switched from group 3 to group 4 (12).

Next, we tested the ^EpiG3-G4 panel on the group 3 (n = 131) and group 4 (n = 304) cases included in the independent frozen cohort of medulloblastoma cases (15). Using the same LDA-based analysis procedure, we were able to discriminate correctly 401 of 435 samples [92.2% agreement (95% CI, 89.1%–94.5%); k = 0.82 (95% CI, 0.76–0.88; Supplementary Fig. S8B; ref. 15)]. We observed discordant subgroup assignment in 34 cases, mainly with intermediate methylation profiles, as compared with previously reported subgroup affiliation data (15).

The performance of the ^EpiG3–G4 panel was then tested on the EPIC850K array. The EPIC array stably reproduced with a comparable performance the results from the HM450K platform by equally classifying the medulloblastoma cases as G3 and G4 (Supplementary Fig. S2B and S2C).

We proceeded to validate the panel ^EpiG3-G4 on DNA from frozen (n = 51) and FFPE (n = 7) samples classified as non-WNT/non-SHH by panel ^EpiWNT-SHH. By applying bisulfite pyrosequencing analysis and LDA prediction model, we classified the frozen tumors with a good concordance [94.1% agreement (95% CI, 83.8%–98.8%); k = 0.88 (95% CI, 0.75–1)] with previous subgroup classification data. The FFPE samples showed a more heterogeneous methylation profile that allowed for correct classification of five of seven cases.

Taken together, we have identified a reduced methylation signature of five CpG sites that permits an accurate classification of strictly defined group 3 and group 4 medulloblastoma tumors, and has shown a good discriminatory capacity that allows for assignment of a large proportion of intermediate group 3 and group 4 medulloblastoma cases.

In this study, we have developed a clinically applicable epigenetic classifier for the classification of the consensus medulloblastoma subgroups WNT, SHH, and non-WNT/non-SHH based on a reduced set of six biomarkers. This epigenetic classifier, named ^EpiWNT-SHH, is characterized by a bimodal subgroup-specific methylation pattern that allows for accurate and rapid molecular classification of single DNA samples from frozen and FFPE medulloblastoma tissue samples using molecular approaches that can be implemented in the routine clinical practice in the majority of centers treating patients with medulloblastoma in developed countries (Fig. 3).

To develop the epigenetic classifier, we used cohorts of medulloblastoma samples that have been characterized in previous studies using robust microarray and multigene-based classification methods (Supplementary Table S2). This allowed for a direct comparison between the subgroup classifications of the different methods. Our epigenetic classifier ^EpiWNT-SHH recapitulated with an excellent match (>99% concordance) the subgroup classification of the genome-wide DNA methylation microarray and gene signature profiling methods currently considered the gold standard for the molecular classification of medulloblastoma patients.

The establishment of the molecular subgroup of medulloblastoma tumors has become of critical importance for the inclusion/exclusion criteria of the new generation of clinical trials. This requires reliable classification approaches that can be reproducibly applied to the different types of material available in the clinical setting such as small biopsies, frozen or FFPE material, from either primary tumor, metastasic, or relapse sites. The methylation profile of our biomarkers remained stable in metastasis as in the primary tumor, and at recurrence as at diagnosis, indicating a strong degree of stability of the methylation signature of panel ^EpiWNT-SHH. These observations are consistent with previous studies showing the maintenance of subgroup affiliation at recurrence and metastasis, demonstrating that medulloblastoma subgroups remain stable and are not affected by treatment (22–24). However, our conclusions are based on a small cohort of tumors and therefore warrant further investigation and validation in larger cohorts of primary/metastatic paired samples.

The design of subgroup-driven clinical trials is critical to improve treatment intensity adjusted to appropriate risk categories. Ongoing clinical trials are currently evaluating the efficacy of new therapies that target subgroup-specific genetic alterations such as SMO (i.e., vismodegib) inhibitors for SHH-driven medulloblastoma tumors. Patients with WNT medulloblastoma (<16 years of age) have excellent survival rates and are currently being recruited in clinical studies aimed at assessing the benefit of a controlled reduction of chemo- or (radio)therapy to limit long-term neurotoxicities and other adverse side-effects (https://clinicaltrials.gov; refs. 25–28). In contrast, patients with group 3 medulloblastoma represent a high-risk group with scarce therapeutic options for whom new strategies are urgently needed.

The new generation of clinical trials have provisionally included group 3 and group 4 tumors in a single entity termed non-WNT/non-SHH, because they comprise a small set of tumors exhibiting overlapping (epi)genetic features that current molecular subgrouping approaches are unable to clearly separate (13, 14, 26). We have developed an epigenetic classifier, named ^EpiG3-G4, specific for group 3 and group 4 tumors based on five CpGs markers that showed a high classification capacity and enabled accurate (100%) classification of clearly defined group 3 and group 4 medulloblastomas. Similar to previous microarray and multi-gene based classification strategies, panel ^EpiG3-G4 showed a more limited discrimination capacity for cases with an intermediate methylation pattern, difficult to affiliate to either of the two subgroups. Nevertheless, panel ^EpiG3-G4 reliably classified approximately 95% of all group 3 and group 4 patients. The genetics and biology underlying group 3 and group 4 subgroups is still greatly unknown. These tumors lack recurrent mutations but show common chromosomal alterations such as isochromosome 17q. Candidate genetic targets have yet to be identified for potential subgroup-specific therapies. Further investigation is desperately needed to better understand the molecular basis underlying the pathogenesis of these tumors. In this context, ^EpiG3-G4 represents a rapid and reproducible approach that allows for group 3 and group 4 classification applicable to DNA from frozen or FFPE samples that may prove to be useful for research purposes (Fig. 3).

In conclusion, we have developed an epigenetic classifier that allows for simple and accurate classification of medulloblastoma tumors into clinically relevant consensus subgroups WNT, SHH, and non-WNT/non-SHH (13, 14). In addition, we propose a similar approach for the specific classification of group 3 and group 4 medulloblastoma tumors. The proposed strategies allow for classification of single DNA samples from biopsies both frozen as well as FFPE of primary, metastasis, or relapse specimens, using diverse DNA methylation detection methods. Our results show that the proposed strategy is robust, accurate, and cost-effective, making it adequate for the clinical application of molecular subgrouping of medulloblastoma in most centers treating patients with brain tumors.

No potential conflicts of interest were disclosed.

Conception and design: S. Gómez, C. Lavarino

Development of methodology: S. Gómez, M. Suñol, M. Kulis, J.I. Martin-Subero, C. Lavarino

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Gómez, A. Garrido-Garcia, L. Garcia-Gerique, I. Lemos, M. Suñol, M. Kulis, A.M. Carcaboso, M.W. Kieran, N. Jabado, S. Dracheva, A. Kozlenkov, V. Ramaswamy, V. Hovestadt, P. Johann, O. Cruz, M.D. Taylor, J. Mora, C. Lavarino

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Gómez, S. Pérez-Jaume, N. Jabado, V. Ramaswamy, P. Johann, D.T.W. Jones, S.M. Pfister, J.I. Martin-Subero, J. Mora, C. Lavarino

Writing, review, and/or revision of the manuscript: S. Gómez, C. de Torres, S. Pérez-Jaume, M.W. Kieran, N. Jabado, D.T.W. Jones, S.M. Pfister, A.M. La Madrid, O. Cruz, M.D. Taylor, J.I. Martin-Subero, J. Mora, C. Lavarino

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): S. Gómez, A. Garrido-Garcia, I. Lemos, B. Luu, J.I. Martin-Subero, J. Mora, C. Lavarino

Study supervision: C. Lavarino

This study was funded by Associations of parents and families that support the Developmental Tumor Biology Laboratory, Hospital Sant Joan de Déu of Barcelona, Spain. The authors acknowledge G. Garcia-Castellví for fundraising and the “Biobanc de l'Hospital Infantil Sant Joan de Déu per la Investigació” integrated in the Spanish Biobank Network of ISCIII for the sample and data procurement. The authors also thank Dr. C. Hawkins, Dr. S. Ryall, Dr. L.C. Laurent, Dr. J. Loring, and Dr. K. Nazor for their assistance in obtaining DNA methylation data, Dr. G. Clot and Dr. G. Castellano for statistical and bioinformatics advice, and M.J. Nagel, Dr. D. Monk, and Dr. M. Sanchez-Delgado for precious technical support.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.