Acta Neuropathologica https://doi.org/10.1007/s00401-018-1845-8

ORIGINAL PAPER

Genomic analysis reveals secondary glioblastoma after radiotherapy in a subset of recurrent medulloblastomas

Ji Hoon Phi1,2 · Ae Kyung Park3 · Semin Lee4,5 · Seung Ah Choi1,2 · In‑Pyo Baek6 · Pora Kim7 · Eun‑Hye Kim8 · Hee Chul Park9 · Byung Chul Kim9 · Jong Bhak4,5 · Sung‑Hye Park10 · Ji Yeoun Lee1,2,11 · Kyu‑Chang Wang1,2 · Dong‑Seok Kim12 · Kyu Won Shim12 · Se Hoon Kim13 · Chae‑Yong Kim14 · Seung‑Ki Kim1,2

Received: 15 December 2017 / Revised: 2 April 2018 / Accepted: 2 April 2018 © Springer-Verlag GmbH Germany, part of Springer Nature 2018

Abstract Despite great advances in understanding of molecular pathogenesis and achievement of a high cure rate in medulloblastoma, recurrent medulloblastomas are still dismal. Additionally, misidentifcation of secondary malignancies due to histological ambiguity leads to misdiagnosis and eventually to inappropriate treatment. Nevertheless, the genomic characteristics of recurrent medulloblastomas are poorly understood, largely due to a lack of matched primary and recurrent tumor tissues. We performed a genomic analysis of recurrent tumors from 17 pediatric medulloblastoma patients. Whole transcriptome sequencing revealed that a subset of recurrent tumors initially diagnosed as locally recurrent medulloblastomas are secondary glioblastomas after radiotherapy, showing high similarity to the non-G-CIMP proneural subtype of glioblastoma. Further analysis, including whole exome sequencing, revealed missense mutations or complex gene fusion events in PDGFRA with augmented expression in the secondary glioblastomas after radiotherapy, implicating PDGFRA as a putative driver in the development of secondary glioblastomas after treatment exposure. This result provides insight into the possible application of PDGFRA-targeted therapy in these second malignancies. Furthermore, genomic alterations of TP53 including 17p loss or germline/somatic mutations were also found in most of the secondary glioblastomas after radiotherapy, indicating a crucial role of TP53 alteration in the process. On the other hand, analysis of recurrent medulloblastomas revealed that the most prevalent alterations are the loss of 17p region including TP53 and gain of 7q region containing EZH2 which already exist in primary tumors. The 7q gain events are frequently accompanied by high expression levels of EZH2 in both primary and recurrent medulloblastomas, which provides a clue to a new therapeutic target to prevent recurrence. Considering the fact that it is often challenging to diferentiate between recurrent medulloblastomas and secondary glioblastomas after radiotherapy, our fndings have major clinical implications both for correct diagnosis and for potential therapeutic interventions in these devastating diseases.

Keywords Medulloblastoma · Recurrence · Secondary glioblastoma after radiotherapy · Genomic analysis

Introduction

Medulloblastoma (MB) is the most common malignant brain tumor in children. Decades of advances in surgery, adjuvant therapies, and patient care have led to a high cure rate of MB Ji Hoon Phi, Ae Kyung Park and Semin Lee contributed equally to which was invariably fatal in Harvey Cushing’s time. Clini- this article. cal trials are ongoing for reducing radiation dose to decrease Electronic supplementary material The online version of this late toxic efects in survivors. Molecular subgrouping of MB article (https​://doi.org/10.1007/s0040​1-018-1845-8) contains has been well-established by a series of genomic studies supplementary material, which is available to authorized users. [27]. There are four main subgroups of MB, namely, WNT, SHH, Group 3, and Group 4, which exhibit distinct gene * Seung‑Ki Kim [email protected] expression patterns, ages of onset, clinical behavior, and prognoses. More recently, genome-wide DNA methylation Extended author information available on the last page of the article

Vol.:(0123456789)1 3 Acta Neuropathologica profling has been recognized as a more robust and accurate revealed a misdiagnosis of secondary GBM after RT. PDG- method for not only molecular subgrouping of MB but also FRA is a frequently altered gene in GBM [4, 36]. CNS tumor classifcation [17]. Patients with WNT MB have This study demonstrates the power of genomic analyses a > 90% chance of survival [8, 37]. In contrast, patients with for diagnosis of brain tumors, especially during recurrence. Group 3 tumors and MYC amplifcation face a far worse Integrated diagnosis, combining histology and molecular prognosis (< 50% survival) [39]. Molecular targeted therapy characteristics, is recommended for recurrent MBs. Fur- is also attempted for the treatment of refractory MBs with thermore, altered PDGFRA signaling pathways can be tar- altered SHH signaling pathways [42, 43]. geted by receptor tyrosine kinase inhibitors, and outcomes Notwithstanding the advancements in MB treatment and of patients with secondary GBM after RT with PDGFRA research, the situation regarding recurrent MB looks difer- mutations can be improved with appropriate diagnosis and ent. Currently, at least a quarter of all MB patients sufer therapies. from tumor recurrence, and the prognosis is dismal. There is no efective therapy for recurrent MB if standard treatments have already been delivered [21]. The genomic and molecu- Materials and methods lar characteristics of recurrent MB are poorly understood, largely because of the lack of tumor tissues. It is known Patients and samples that recurrent MBs frequently acquire increased histological anaplasia and chromosomal imbalance [16]. A recent study Specimens were collected from 17 patients from three cent- has shown that the emergence of combined MYC and p53 ers, namely, Seoul National Children’s Hospital (SNUCH), pathway defects during recurrence leads to extremely poor Severance Hospital, and Seoul National University Bundang outcomes [13]. Another study has shown that recurrent MBs Hospital (SNUBH), with appropriate consent from institu- are derived through clonal selection from existing clones tional review boards. The primary tumor in all the patients in primary tumors and harbor a remarkable number of new was MB. Details of patient characteristics and samples are genetic events [24]. Interestingly, despite the accumulation described in Table S1. The biospecimens were composed of of genetic variations, the molecular subgroup of a MB is 14 tumors paired with patient-matched primary MB samples highly stable during recurrence [40]. Therefore, progressing (11 pairs of primary–recurrent MB and 3 pairs of primary genomic changes coexist with stable molecular subgroups MB–secondary GBM after RT); one pair of locally recur- during MB recurrence. To overcome the treatment resistance rent-distantly metastasized MB without a primary tumor of recurrent MBs, it would be desirable to fnd and target sample; one pair of secondary GBM after RT, from two emergent genomic aberrations in recurrent tumors. independent operations, without a primary MB sample; and A confounding problem in the management of MB recur- one unpaired secondary GBM after RT without a primary rence is the presence of secondary malignancies [15]. Whole MB sample. Normal DNA from peripheral whole blood cells neuro-axis radiation is the mainstay of treatment for MBs. (WBCs) was available for 12 patients. For DNA and RNA Therefore, secondary malignancy, especially malignant extraction from frozen tissues, DNeasy Blood and Tissue gliomas, can develop in a small percentage of survivors [5]. Kit (Qiagen) and mirVana miRNA Isolation Kit (Invitro- Clinically, it is often challenging to diferentiate between the gen) were used, respectively. Extraction of DNA and RNA two entities based solely on imaging results. Furthermore, from FFPE tissues was carried out using RecoverAll Total matched primary and recurrent tumor tissues are uncommon Nucleic Acid Isolation Kit (Thermo Fisher Scientifc). because surgery is usually not indicated for recurrent MBs. Even pathological examinations often provide misleading RNA sequencing and analysis information due to histological ambiguity, such as high degrees of anaplasia during recurrence. However, misdiag- mRNA sequencing was performed using a TruSeq RNA nosis of secondary glioblastoma (GBM) after radiotherapy Library Preparation Kit and TruSeq RNA Access Kit (Illu- (RT) as recurrent MB can result in grim outcomes because mina) and sequenced on an Illumina HiSeq 2000 or 2500 the treatment protocols and regimens are diferent. platform. We used an in-house script to trim low-quality Genomic and epigenomic analyses have shown great reads [proportion of ambiguous nucleotide (N ratio) > 0.1 potential not only in stratifying diseases but also in correct- or proportion of nucleotide with low Phred quality score ing misdiagnoses [49]. In the genomic analyses of primary (< 20) > 0.4]. The filtered sequence data were analyzed and recurrent MBs, we found that a subset of recurrent using STAR [7] for alignment, Cufinks v2.2.1 1 [47] for tumors contains mutations or gene fusion events in PDG- assembly, and a known set of reference transcripts from FRA, which is not characteristic of MB. The gene expression Ensembl v72 for expression estimation. The details of RNA pattern of these recurrent tumors co-segregated with that of sequencing and alignments are summarized in Table S2. GBM rather than MB. Re-evaluation of histological images Gene fusions were identifed by searching for the spanning

1 3 Acta Neuropathologica reads and split reads by ChimeraScan [14], deFuse [22], and for which matched normal data were available. The anno- FusionMap [11] using default settings. To detect robust can- tation of known cancer genes in the segments of SCNAs didates of gene fusions accompanied by change in expres- was performed using the Catalogue of Somatic Mutations sion, we used several selection criteria: detection with in Cancer (COSMIC) [10]. more than two software programs, high changes in expres- sion in recurrent tumors compared with primary tumors (|log2FC| ≥ 1) in either of two genes involved in fusion event, Results and high expression levels in either of the two genes in at least one tumor of the paired samples (maximum expression Samples of FPKM in either of the two genes across tumor pairs ≥ 50). To investigate genomic changes in recurrent MB, tumor Whole exome sequencing and analysis tissues of 17 patients were collected from three centers (Table S1), including 11 pairs of patient-matched pri- Whole exome sequencing (WES) was performed using the mary–recurrent MB (D2–D9, L1–L3), three pairs of primary 71 Mb SureSelect V4 plus UTR Kit or the 51 Mb SureSelect MB–secondary GBM after RT (L4–L6), one pair of locally V5 Kit (Agilent Technologies) according to the manufactur- recurrent and distantly metastasized MB without its primary er’s instructions. DNA libraries were constructed according MB sample (D1), and one pair of secondary GBM after to the protocol provided by the manufacturer and WES was RT obtained from separate operations without its primary performed using an Illumina HiSeq 2000 or 2500 platform MB sample (G1). The fnal sample was a single secondary to generate 101 bp paired-end reads. We used cutadapt [20] GBM after RT without its primary MB sample (G2). All and sickle (https​://githu​b.com/najos​hi/sickl​e) to remove initial diagnoses were based on pathological examinations adapter sequences and low-quality sequence reads. Bur- of the original specimens. Three pairs of primary MB–sec- rows–Wheeler aligner [18] was used to align the sequencing ondary GBM after RT (L4–L6) were initially diagnosed as reads onto the human reference genome (hg19). We used a primary–recurrent MBs. However, genomic analysis in this Genome Analysis ToolKit (GATK) (http://www.broad​insti​ study revealed that the recurrent tumors were not locally tute.org/gatk) for local realignment and score recalibration recurrent MBs but newly developed secondary GBMs after and to flter the sequence data. Picard (http://broad​insti​tute. RT. Other GBM tumors (G1 and G2), which occurred in githu​b.io/picar​d) and Samtools (http://samto​ols.sourc​eforg​ two MB patients and were originally confrmed as second- e.net/) were also used for basic processing and management ary GBMs after RT based on their pathologies, were used of the sequencing data. The details of WES and alignments as positive controls. are summarized in Table S3. Somatic single nucleotide variants (SNVs) and short insertions and deletions (indels) Subgroup afliation of recurrent medulloblastoma were analyzed using MuTect [6] and SomaticIndelDetec- and secondary glioblastoma after RT tor in GATK, respectively, with matched normal samples as controls. For comparisons between normal and tumor sam- We frst examined subgroup afliation of tumor samples ples without any available patient-matched normal DNA, using unsupervised hierarchical clustering and partitioning we generated variants from 12 normal WBC samples using around medoid (PAM) methods based on RNA-seq expres- GATK with UnifedGenotyper, VariantRecalibrator (hap- sion profles of 17 marker genes from a CodeSet [30] exclud- map, 1000g, dbsnp) and ApplyRecalibration options, and ing WNT group-specifc genes (Fig. 1a). The subgroups of we identifed intersections between the normal variants and primary MBs were well diferentiated by both methods, each tumor sample using BEDtools (http://sourc​eforg​e.net/ however, the clustering of one primary MB sample, L6.1, proje​cts/bedto​ols/) with the intersect option. SnpEf (http:// was not consistent. In addition, a negative silhouette width snpef​ .sourc​eforg​e.net/) was used to select somatic variants was observed in the PAM analysis (Fig. 1a, right) even located in coding sequences and predict their functional though the expression level of Group 3-specifc genes was consequences, such as silent or non-silent variants. Ampli- considerably high in L6.1. This ambiguous result might be con sequencing was also performed using an Ion AmpliSeq ascribed to the low RNA quality of the sample (the RNA Comprehensive Cancer Panel (Life Technologies) accord- Integrity Number = 2.8). Therefore, the L6.1 sample was ing to the manufacturer’s protocols to validate the missense assigned to Group 3 according to the high expression level of mutations in PDGFRA detected by WES. Sequencing was Group 3-specifc genes. All 11 primary–recurrent MB pairs carried out on an Ion Proton sequencer (Life Technolo- showed the same subgroup afliation regardless of recur- gies) with the Ion PI chip. Somatic copy number alterations rent location, distant (D2–D9) or local (L1–L3), which is (SCNAs) were detected based on WES data using Control- consistent with a previous report [40]. Even the subgroup FREEC (v10.4) with default options [2] in 12 tumor samples afliation of a pair of locally recurred (3rd operation) (D1.1)

1 3 Acta Neuropathologica

and distantly metastasized (4th operation) (D1.2) MBs (GSE30074) [33] and 34 adult GBMs (GSE58399) [32]. remained unchanged. However, the expression patterns of Hierarchical clustering and PAM analysis with 22 genes three secondary GBM tumors (L4–L6) initially misdiag- of a CodeSet (Fig. 1c) and PCA plot with top 3000 genes nosed as recurrent MBs clearly showed strong similarity to with the high expression variability (Fig. 1d) clearly showed those of pathologically confrmed secondary GBMs after that the secondary GBMs were grouped together with adult RT (G1 and G2). In the principal component analysis (PCA) GBMs, not with pediatric MBs. Furthermore, we found that plot using the 3000 most variably expressed genes, all the most of the secondary GBM tumors were most similar to secondary GBMs after RT were segregated into one cluster the proneural subtype GBM when considering the expres- (Fig. 1b). We further confrmed the entity of the secondary sion correlations calculated between each secondary GBM GBMs after RT based on microarray expression profles that sample and 163 TCGA GBM tumor samples based on previ- were available for two patients, L4 and L5, combined with ously reported GBM subtype-specifc genes [48] (Fig. 1e). two public datasets containing 30 pediatric primary MBs We compared the time interval between the frst diagnosis

1 3 Acta Neuropathologica

◂Fig. 1 Subgroup afliation of recurrent medulloblastomas and sec- rate in primary MBs was 0.27 per megabase compared ondary glioblastomas after radiotherapy based on expression profles. with patient-matched genomic DNA (Table S5), which is a Heatmap with unsupervised hierarchical clustering and silhou- ette plot of partitioning around medoid (PAM) of 33 samples based lower than the previously reported rate of 0.35 non-silent on RNA-seq expression profles of 17 discriminatory genes from a mutations per megabase [38]. To further select the damag- CodeSet [30] excluding WNT group-specifc genes. Red arrow indi- ing mutations, we excluded known polymorphic variants cates the sample with disagreement of subgroup clustering between (allele frequency ≥ 0.1%) from the non-silent mutations hierarchical clustering and PAM method. b PCA plot of top 3000 genes with high RNA-seq expression variability (IQR) of 33 samples based on the 1000 Genomes Project, Exome Sequencing of a. c Heatmap with unsupervised hierarchical clustering of 22 genes Project (ESP6500), and Exome Aggregation Consortium of a CodeSet [30] and silhouette plot of PAM of 68 samples includ- (ExAC). The details of the damaging mutations are listed ing two pairs of primary medulloblastoma (MB) and secondary glio- in Table S6. Consistent with a previous report [24], the blastoma (GBM) after radiotherapy (RT) (L4 and L5), 30 pediatric primary MBs (GSE30074), and 34 GBM tumors (GSE58399) based number of damaging mutations increased in recurrent on microarray (Afymetrix Human Gene 1.0 ST) expression profles. MBs, and only a part of these mutations was shared in Among the samples in the dataset of GSE58399, we used only 34 primary and recurrent MBs (Fig. 2a). In contrast, only samples designated “tissue” and “Stage: GBM (IV)” in the informa- one damaging mutation overlapped between the primary tion variables of “Sample_title” and “Sample_characteristics_ch1.1”, respectively. Red arrow indicates the sample with disagreement MBs and the secondary GBMs after RT, indicating that of subgroup clustering between hierarchical clustering and PAM the secondary GBMs after RT did not originate from method. d PCA plot of top 3000 genes with high expression varia- primary MBs. In the secondary GBMs after RT, non- bility (IQR) of 68 samples of c. e Box plots of Pearson’s correlation silent mutations in PDGFRA were observed in all fve coefcients calculated between ­log (FPKM + 1) values of each sam- 2 patients (Fig. 2b), including eight missense mutations ple of secondary GBM after RT and ­log2 (normalized counts + 1) val- ues from RNA-seq data of 163 TCGA GBM samples (n = 39, 28, 42, and one splice region mutation (Table S6). This implies and 54 for proneural, neural, classical, and mesenchymal subtypes, that genomic alterations in PDGFRA might be one of the respectively) based on 763 subtype-specifc genes [48]. Using a simi- key drivers of secondary GBM after RT. Most of the mis- lar procedure applied in Sandmann et al. [44], the similarity between PDGFRA each sample of the secondary GBMs after RT and the TCGA sam- sense mutations in occurred in extracellular ples was investigated by Pearson’s correlation coefcients based on immunoglobulin-like domains (L5, G1, and G2), and one the expression profles of 763 subtype-specifc genes that were com- other missense mutation was observed in the intracellular monly available in both datasets. f Recurrence free survival analysis tyrosine kinase domain, i.e., the D842V substitution (L4) with Kaplan–Meier plot and generalized Wilcoxon test. Comparisons of median time to recurrence were made between secondary GBMs (Fig. 2c), which is the most frequently found mutation in after RT and recurrent MBs (left) or across subgroups of MB (right) PDGFRA and is related to imatinib resistance. Two mis- sense mutations in PDGFRA found in tumors from two patients (L4.2 and L5.2) were confrmed by deep sequenc- of MB and the advent of recurrence. Median time to recur- ing (×1000) using Ion AmpliSeq Comprehensive Cancer rence was signifcantly longer for secondary GBM after RT Panel (Table S7). The mutations in PDGFRA were accom- (median 9.0 years) compared to recurrent MBs (median panied by signifcantly increased mRNA expression in 3.0 years) (generalized Wilcoxon p = 0.0034) (Fig. 1f, left). secondary GBMs after RT compared with primary MBs When the comparison was made across subgroups of MB, no (Fig. 2d). Additionally, TP53 and several genes involved statistical signifcance was obtained (Fig. 1f, right), which in response to ionizing radiation or DNA repair such as is not congruent to the previous report that Group 4 MBs ATM, DNMT3A, and UIMC1 were detected as mutated recurred signifcantly later than Group 3 and SHH tumors genes in the secondary GBMs after RT, indicating failure [40]. of protective mechanisms against postoperative therapeu- tic radiation (Fig. 2b). Somatic mutations in recurrent medulloblastoma In the analysis of ten patient-matched primary–recur- and secondary glioblastoma after RT rent MBs, no specifc common mutated gene was recog- nized. However, in recurrent MBs, new damaging mutations To identify the genomic alterations that drive recurrent were identifed in the genes involved in cancer pathways MBs and secondary GBMs after RT, somatic SNVs and (PIK3CA, PIK3CG, SOS2, HDAC1, E2F2, TGFBR2, PTEN, indels were called on the exomes of tumor samples. Vari- and MTOR) or stem cell maintenance (LIF, NOTCH2, and ous comparisons were made based on the availability of DDX6) (Fig. 2b). In addition, the mutated genes included the samples (Table S4). For the tumor samples without the genes, CELSR3, ABCA13, and, WDFY3, which were matched normal DNA, comparisons were made between previously detected in recurrent human MBs [24]. Other the tumor and multiple variants calls generated from 12 mutated genes involved in neurogenesis (ASPM and FAT4) normal WBC DNA samples (designated combWBC calls). were found in two recurrent MBs, respectively. In compari- Coding mutation rates and the number of mutations are sons between WBC DNAs and primary MBs, we detected presented in Table S5. The median non-silent mutation several mutated genes such as TP53, PTCH1, SMO, LDB1,

1 3 Acta Neuropathologica

Fig. 2 Somatic SNV/indels in secondary glioblastomas after radio- ing mutations in secondary GBMs after RT, recurrent MBs, and pri- therapy, recurrent medulloblastomas, and primary medulloblastomas. mary MBs. Asterisks indicate the genes identifed in previous studies a Venn diagrams of damaging mutations detected in primary, recur- in recurrent MBs (asterisk) [24] or primary MBs (double asterisks) rent medulloblastomas (MBs), and secondary glioblastomas (GBMs) [38]. c Lolliplot of missense mutations detected in secondary GBM after radiotherapy (RT). b Representative mutated genes with damag- tumors. d Expression levels of PDGFRA

PRKACA​, and KDM6A that were previously reported by Somatic copy number alterations in recurrent exome sequencing of a large cohort of primary MBs [38]. medulloblastoma and secondary glioblastoma The damaging mutations in TP53 and PTCH1 were observed in two patients, respectively, and some mutations including We investigated SCNAs occurring in primary/recurrent TP53 were retained in recurrent MBs, while others were lost. MBs and secondary GBMs after RT using WES data from 12 patients (D1–D4, D6–D8, L1, L4–L6, and G1) with available matched normal data (Table S8). The SCNAs

1 3 Acta Neuropathologica

Fig. 3 Somatic copy number alterations in primary medulloblas- observed cytogenetic gains and losses with suspected driver genes. tomas, recurrent medulloblastomas, and secondary glioblastomas Focal aberrations were defned as 5 Mb or less in size. c Expression after radiotherapy. a Copy number change in 12 patients whose nor- level of TP53, PTEN, and EZH2 in FPKM of RNA-seq data. Aster- mal DNA samples were available. Red and blue indicate copy num- isk denotes congruency between copy number change and expression ber gain and loss, respectively. b Overview of the two most frequently level observed in recurrent MBs were largely concordant hand, in the locally recurrent-distant metastasized MB pair with those in primary MBs (Fig. 3a). However, SCNAs (D1) and the secondary GBM after RT tumor pair (G1), observed in primary MBs were rarely shared by their new SCNAs occurred in latter samples while many of ini- respective patient-matched secondary GBMs after RT, tial SCNAs remained. confrming the distinct origin of the tumors. On the other

1 3 Acta Neuropathologica

The most frequent SCNAs in tumors were deletions number loss of 10q (D6, L1, L4, and G1.1) (Fig. 3c). The in the 17p telomeric region that includes TP53, which PTEN locus has been previously reported as the most fre- was observed in ten out of 12 patients (Fig. 3b). In pri- quently homozygously deleted region in MBs [29]. The two mary–recurrent MB pairs, the 17p deletion was observed most frequent copy number gain events were detected in the in both primary and recurrent tumors. However, the appear- 7q and 17q regions. The gain of the 7q region that includes ance of the 17p deletion in secondary GBMs after RT (L4 EZH2 as a putative driver gene was observed in four patients and L5) was a new observation. These fndings imply that with recurrent MB, and the expression of EZH2 was consid- TP53 loss plays a central role not only in the recurrence of erably elevated in three patients with distant metastasis (D1, MBs but also in the novel occurrence of secondary GBMs D2, and D4) (Fig. 3c). High-level expression of EZH2 was after RT. The expression level of TP53 was largely consist- also observed in one patient with distant metastasis without ent with the SCNA result, showing considerably low expres- SCNA information (D5). The gain of the 17q region was not sion levels of TP53 (FPKM value ranges from 0.92 to 8.92) consistently observed in primary and recurrent MBs; this in fve patients (D1, D3, D4, D8, and L1) (Fig. 3c). The region was newly detected (D7) or disappeared in recurrent next most frequent copy number loss was detected in the MB (D3). wide 10q region that includes tumor suppressor PTEN as a suspected driver gene, and the loss was also associated with low expression of PTEN in the tumors containing a copy

Fig. 4 Complex gene fusion events in the PDGFRA gene region in site of exon 1 of KIT and a cryptic splice site of intron 1 of LNX1 chromosome 4 in a secondary glioblastoma tumor after radiother- (NM_001126328). b Read depth coverage plot of RNA-seq data of apy. a Schematic diagram of gene fusion events with aligned mul- LNX1. c Expression of PDGFRA–LNX1 fusion transcript, PDGFRA tiple RNA-seq reads mapped on fusion junctions. The fusion gene and LNX1 in a primary medulloblastoma (L6.1) and a secondary glio- KIT–LNX1 (NM_001126328) was spliced using a normal splice blastoma after radiotherapy (L6.2), confrmed by RT-qPCR

1 3 Acta Neuropathologica

Intrachromosomal gene fusion in the PDGFRA the abrupt change of coverage depth of exon 7 of LNX1 region in a secondary glioblastoma (NM_032622) after the break point is also consistent with the presence of the fusion transcript of LNX1 (NM_032622)- A total of 20 most putative fusion genes with high expres- FIP1L1. The PDGFRA-LNX1 fusion cDNA was success- sion changes were detected in fve tumor samples (Table S9). fully amplifed in the secondary GBM after RT (L6.2) but Among them, complex intrachromosomal gene fusion events not in the primary MB (L6.1) (Fig. 4c). were observed in chromosome 4 in one secondary glioblas- toma after RT (L6.2) involving four neighboring genes, Pathological confrmation of secondary namely, FIP1L1, LNX1, and two receptor tyrosine kinases, glioblastomas after radiotherapy PDGFRA and KIT (Fig. 4a). We identifed three diferent fusion transcripts and multiple RNA-seq reads that spanned The original pathology slides of the presumed secondary the fusion junctions in which LNX1 was involved in all three GBMs after RT (L4–L6) were re-evaluated by a patholo- events: LNX1 (NM_032622)–FIP1L1, PDGFRA–LNX1 gist (S.P.). Morphological features of secondary GBMs (NM_032622), and KIT–LNX1 (NM_001126328). The after RT (L4.2, L5.2 and L6.2) showed polymorphic expression of all four genes was considerably elevated nuclei, long cytoplasmic processes, microvascular prolif- (Table S9). Furthermore, in the fusion with KIT, a cryptic eration and necrosis which were diferent from the features splice site in intron 1 of LNX1 (NM_001126328) was used of MBs (L4.1) (Fig. 5a). However, the anaplastic/large cell for splicing with exon 1 of KIT, and the depth coverage plot features observed in the secondary GBMs after RT mim- of the RNA-seq data showed high transcription levels of the icked anaplastic/large cell variant of MBs with large atypi- intronic sequence of intron 1 of LNX1 (Fig. 4b). In addition, cal nuclei. Therefore, the presence of edematous stroma

Fig. 5 Pathological confrmation of previously misdiagnosed second- L6.2), and two pathologically confrmed secondary GBMs after RT ary glioblastomas after radiotherapy. a Hematoxylin and eosin (H&E) (G1.1 and G2.1) as positive controls. b Olig2 immunohistochemis- (×200) slides of a primary medulloblastoma (MB) (L4.1) as a nega- try of MB (L4.1) and secondary GBMs after RT (L4.2, L5.2, G1.1, tive control, three secondary glioblastomas (GBMs) after radiother- and G2.1) and ultrastructural imaging (uranyl acetate and lead citrate apy (RT) initially misdiagnosed as recurrent MBs (L4.2, L5.2, and staining; ×12,300) of secondary GBM after RT (L6.2)

1 3 Acta Neuropathologica Expression Expression (FPKM) 75.7 79.9 98.9 88.9 26.1 21.8 112.2 128.1 179.7 203.7 46.9 54.9 36.9 27.8 32.8 41.1 29.0 17.8 40.4 26.0 39.0 41.6 13.3 23.0 EZH2 7q EZH2 region Gain Gain Gain Gain Neutral Neutral Gain Gain – – Neutral Neutral Neutral Neutral Neutral Neutral – – Gain Gain – – – – Expression Expression (FPKM) 8.2 6.0 17.4 15.0 7.7 8.9 8.7 7.5 162.9 19.5 33.7 39.8 32.6 36.3 0.9 1.8 38.8 52.9 5.9 6.0 31.3 44.7 147.8 91.7 17p TP53 region Loss Loss Neutral Neutral Loss Loss Loss Loss – – Loss Loss Loss Loss Loss Loss – – Loss Loss – – – – (mis - sense) (splice acceptor) TP53 Mutation Somatic Somatic Expression Expression (FPKM) 0.7 0.6 12.4 4.2 0.1 0.5 0.5 0.6 3.0 11.0 5.3 13.5 8.2 33.7 4.9 13.3 14.0 3.9 5.0 2.3 7.8 5.4 18.3 0.0 PDGFRA Muta - tion/gene fusion - Character istics Local Meta Primary Meta Primary Meta Primary Meta Primary Meta Primary Meta Primary Meta Primary Meta Primary Meta Primary Local Primary Local Primary Local Tumor Tumor label D1.1 D1.2 D2.1 D2.2 D3.1 D3.2 D4.1 D4.2 D5.1 D5.2 D6.1 D6.2 D7.1 D7.2 D8.1 D8.2 D9.1 D9.2 L1.1 L1.2 L2.1 L2.2 L3.1 L3.2 - Time inter Time (year) val primary to recurrence 3.4 2.5 6.8 3.0 3.0 1.3 3.6 1.0 1.6 3.9 4.0 2.0 Subgroup Group 4 Group Group 4 Group Group 4 Group Group 4 Group Group 4 Group SHH SHH SHH SHH SHH Group 3 Group Group 3 Group M stage – 3 0 0 0 0 1 0 0 3 0 3 Age 9 9 9 14 16 15 12 8 9 9 15 9 Gender M M M F F F M M F F M M Patient Patient label D1 D2 D3 D4 D5 D6 D7 D8 D9 L1 L2 L3 - metas - tasis recur rence Distant Local Clinical characteristics and recurrent aberrationsClinical characteristics genomic identifed in recurrent and secondary after medulloblastomas radiotherapy glioblastomas MB 1 Table Tumor Recurrent

1 3 Acta Neuropathologica

was also observed. Among various immunomarkers, Olig2 was the most valuable marker to distinguish GBMs from MBs because only gliomas are always positive for this Expression Expression (FPKM) 26.3 21.7 24.7 29.5 24.6 4.0 16.1 27.3 31.3 antibody (Fig. 5b). In the electron microscopy study, sec- ondary GBMs after RT had cytoplasm which was abundant with glial intermediate flaments (L6.2) (Fig. 5b). In addi- EZH2 7q EZH2 region Neutral Neutral Neutral Neutral Neutral Neutral Neutral Neutral – tion, all secondary GBMs after RT were negative for the IDH1 mutation-specifc antibody. The MIB1 (Ki67) index ranged from 12 to 48%. Expression Expression (FPKM) 28.9 23.0 29.2 20.7 18.7 51.4 6.3 19.1 21.0

Discussion 17p TP53 region Neutral Loss Neutral Loss Neutral Neutral Loss Loss – In this study, we investigated the genomic changes in recurrent MBs. A relatively low mutation rate is charac- (mis - sense) (mis - sense) (mis - sense) TP53 Mutation Germline Somatic Somatic teristic of pediatric brain tumors such as atypical teratoid/ rhabdoid tumor and MBs [3]. However, in recurrent MBs, the mutation rate increased several fold with the accumula- Expression Expression (FPKM) 13.2 86.2 10.9 1562.3 3.1 148.3 2306.1 151.6 504.6 tion of non-silent mutations (Table S5). Therapy-induced DNA damage and clonal selection of mutated tumor cells are considered to be the underlying mechanisms for the fusion PDGFRA Muta - tion/gene fusion Missense Missense Gene Missense Missense increased mutation rate in recurrent MBs [24]. Recurrent - MB is refractory to most chemotherapeutic regimens cur- rently available [21]. The accelerated genomic changes Character istics Primary GBM Primary GBM Primary GBM GBM GBM GBM may contribute to the failure of treatments. Nearly all patients with recurrent MBs exhibited genomic altera- tions of TP53 including somatic mutations and/or loss of TP53 Tumor Tumor label L4.1 L4.2 L5.1 L5.2 L6.1 L6.2 G1.1 G1.2 G2.1 the 17p region containing (Table 1), indicating the - crucial role of TP53 in MB recurrence. A recent study reported that deleterious alterations in TP53 or TP53 path- way genes were primarily responsible for the recurrence of Time inter Time (year) val primary to recurrence 7.3 10.0 4.3 9.0 10.0 SHH MBs in both humans and mice [24]. Gain of 7q with high expression of EZH2, was also prevalent in recurrent MBs, particularly enriched in Group 4 (Fig. 3b). It has Subgroup SHH SHH Group 3 Group – – been noticed that the high expression of EZH2, encod- ing a component of the polycomb repressive complex and histone H3K27 methyltransferase, is associated with gain M stage 0 0 0 2 0 of chromosome 7, which is observed specifcally in Group 3 and Group 4 and implicated in deregulated chromatin modifcation in these groups of MB [28, 41]. Age 1.9 5 8 8 5 Secondary malignancies account for 11.8% of the late death of childhood MB patients [26]. Brain tumors, mostly gliomas, constitute one-third of all secondary malignan- Gender F F F M M cies [26]. Ionizing radiation can cause DNA cleavage and can induce transformation and oncogenesis [46]. All clini- cal protocols for childhood MB include treatment with Patient Patient label L4 L5 L6 G1 G2 ionizing radiation along the entire craniospinal axis to eliminate microscopically disseminated tumor cells. The median latency from initial radiation therapy to the diag- GBM after RT

Secondary nosis of secondary glioma is 9–11 years [9, 50], which is consistent with our data (median 9 years) (Table 1). In fact, a longer time to recurrence is an important clue to

1 Table (continued) Tumor suspect secondary malignancy in childhood brain tumor

1 3 Acta Neuropathologica

Fig. 6 Deletion/insertion ratio in mutations newly occurred in each primary medulloblastoma, recurrent medulloblastoma, and secondary glio- blastoma after radiotherapy. a Box plots of deletion/insertion ratio of each tumor. b Barplots of deletion/insertion ratio of each tumor. *p < 0.05 survivors. However, one patient (D3) developed very late tumors [12, 17, 23, 45]. All fve secondary GBMs after recurrence of MB (6.8 years) and another (L6) devel- RT exhibited alterations in the PDGFRA locus, missense oped secondary GBM earlier than usual (4.3 years) in our mutations in four patients and a PDGFRA-LNX1 gene study. Therefore, molecular and genomic profling can be fusion in one patient. PDGFRA expression was augmented a great aid in correct diagnosis. in all cases. PDGFRA is a receptor tyrosine kinase and one A recent study reported an excess of deletions relative of the principal upstream regulators of the RAS/MEK/ to insertions as one of the mutational signatures of ioniz- PI3K signaling pathway. PDGFRA alterations such as gene ing radiation in secondary malignancies [1]. Accordingly, amplifcation and oncogenic mutations are prominent in we examined whether the deletion/insertion ratio increased pediatric high-grade gliomas [19, 34, 35]. Secondary in newly occurred mutations in recurrent MBs and second- GBMs after RT share these characteristics with pediatric ary GBMs after RT (Fig. 6). We observed that the dele- malignant gliomas, i.e., overexpression and alterations of tion/insertion ratio was signifcantly increased in recur- the PDGFRA gene. Frequent TP53 mutations and/or loss of rent MBs compared with primary MBs (Wilcoxon rank the 17p telomeric region containing TP53 were observed sum test p = 0.015) but not in secondary GBMs after RT in four out of fve secondary GBMs after RT (Table 1), (Fig. 6a). Consistently, in pairwise comparisons, the dele- implying that TP53 alteration is critical in the development tion/insertion ratio was elevated in most of the recurrent of secondary GBMs after RT. In one secondary tumor after MBs (Fig. 6b). However, the increase of deletion/insertion RT, L6.2, neither TP53 alteration nor PDGFRA missense ratio was not obvious in two secondary GBMs after RT, L4 mutation was observed, but complex gene fusion events and L5. In conjunction with treatment exposure, other etio- involving the PDGFRA region occurred. Furthermore, the logical factors such as genetic predisposition contribute to expression profle of this tumor, L6.2, showed similarity the development of secondary malignancies in pediatric to neural and mesenchymal subtype GBMs in addition to cancers [25]. Indeed, a germline TP53 mutation was found proneural subtype (Fig. 1e). Clinical data also revealed in one patient, L5. that the secondary malignancy after RT occurred consid- Brain tumors are very heterogeneous diseases and many erably earlier (4.3 years) in this patient (L6) than in other entities mimic each other in clinical and histopathological patients (L4, L5, G1, and G2) (7.3–10.0 years) (Table 1). features. Gene expression patterns, chromosomal aberra- Taken together, in the patient L6, the secondary GBM after tions, DNA methylation profling, and specifc mutations RT might have occurred by a diferent mechanism. Nev- can provide valuable information for diagnoses of brain ertheless, the occurrence of the GBM involved PDGFRA

1 3 Acta Neuropathologica alteration. On the other hand, no secondary tumor after mutations in papillary craniopharyngiomas. Nat Genet 46:161– RT harbored an IDH1 mutation, which is congruent to 165. https​://doi.org/10.1038/ng.2868 4. Brennan CW, Verhaak RG, McKenna A, Campos B, Noushmehr a previous investigation of pediatric high-grade gliomas H, Salama SR, Zheng S, Chakravarty D, Sanborn JZ, Berman SH [34]. In adult GBMs, IDH1 mutations are mostly found in et al (2013) The somatic genomic landscape of glioblastoma. Cell the glioma-CpG island methylator phenotype (G-CIMP), 155:462–477. https​://doi.org/10.1016/j.cell.2013.09.034 a subset of the proneural subtype [31]. Therefore, all the 5. Christopherson KM, Rotondo RL, Bradley JA, Pincus DW, Wynn TT, Fort JA, Morris CG, Mendenhall NP, Marcus RB Jr, Indeli- secondary GBMs after RT in this study are considered to cato DJ (2014) Late toxicity following craniospinal radiation for be similar to non-G-CIMP proneural subtype of GBM. early-stage medulloblastoma. Acta Oncol 53:471–480. https://doi.​ In this study, we explored the genomic landscape of org/10.3109/02841​86X.2013.86259​6 recurrent MB. Recurrent MBs have higher rates of genomic 6. Cibulskis K, Lawrence MS, Carter SL, Sivachenko A, Jafe D, TP53 EZH2 Sougnez C, Gabriel S, Meyerson M, Lander ES, Getz G (2013) alterations. alteration and overexpression are Sensitive detection of somatic point mutations in impure and het- observed in recurrent MBs. Notably, we found unexpected erogeneous cancer samples. Nat Biotechnol 31:213–219. https​:// secondary gliomas in a subset of presumed recurrent MBs. doi.org/10.1038/nbt.2514 However, the high proportion of the secondary tumor 7. Dobin A, Davis CA, Schlesinger F, Drenkow J, Zaleski C, Jha S, Batut P, Chaisson M, Gingeras TR (2013) STAR: ultrafast uni- observed in our study (29.4%, 5 out of 17 patients) does versal RNA-seq aligner. Bioinformatics 29:15–21. https​://doi. not refect the actual proportion in the population of MB org/10.1093/bioin​forma​tics/bts63​5 patients which should be determined in a controlled trial 8. Ellison DW, Onilude OE, Lindsey JC, Lusher ME, Weston CL, cohort. Conclusively, high-throughput sequencing can be a Taylor RE, Pearson AD, Cliford SC, United Kingdom Chil- dren’s Cancer Study Group Brain Tumour Consortium (2005) useful supplement to histological diagnoses for recurrent beta-Catenin status predicts a favorable outcome in childhood MBs. medulloblastoma: the United Kingdom Children’s Cancer Study Accession codes Group Brain Tumour Committee. Journal of clinical oncology RNA sequencing datasets have been deposited to : ofcial journal of the American Society of Clinical Oncology 23:7951–7957. https​://doi.org/10.1200/jco.2005.01.5479 Sequence Read Archive (SRA) with bioproject number 9. Elsamadicy AA, Babu R, Kirkpatrick JP, Adamson DC (2015) PRJNA418677. Whole exome sequencing datasets have been Radiation-induced malignant gliomas: a current review. deposited to SRA with bioproject number PRJNA418669. World Neurosurg 83:530–542. https​://doi.org/10.1016/j. wneu.2014.12.009 Acknowledgements This study was supported by a grant from the 10. Forbes SA, Tang G, Bindal N, Bamford S, Dawson E, Cole C, National R&D Program for Cancer Control, Ministry for Health and Kok CY, Jia M, Ewing R, Menzies A et al (2010) COSMIC (the Welfare, Republic of Korea (1420020). This work was also supported Catalogue of Somatic Mutations in Cancer): a resource to inves- by the 2016 Research Fund (1.160052.01) of Ulsan National Institute tigate acquired mutations in human cancer. Nucleic Acids Res of Science and Technology (UNIST) and the Collaborative Genome 38:D652–D657. https​://doi.org/10.1093/nar/gkp99​5 Program for Fostering New Post-Genome Industry of the National 11. Ge H, Liu K, Juan T, Fang F, Newman M, Hoeck W (2011) Research Foundation (NRF) funded by the Ministry of Science and FusionMap: detecting fusion genes from next-generation sequenc- ICT (MSIT) (NRF-2017M3C9A5031004). ing data at base-pair resolution. Bioinformatics 27:1922–1928. https​://doi.org/10.1093/bioin​forma​tics/btr31​0 12. Gessi M, Giangaspero F, Pietsch T (2003) Atypical teratoid/rhab- Compliance with ethical standards doid tumors and choroid plexus tumors: when genetics “surprise” pathology. Brain Pathol 13:409–414 Conflict of interest The authors declare no potential conficts of inter- 13. Hill RM, Kuijper S, Lindsey JC, Petrie K, Schwalbe EC, Barker est. K, Boult JK, Williamson D, Ahmad Z, Hallsworth A et al (2015) Combined MYC and P53 defects emerge at medulloblastoma relapse and defne rapidly progressive, therapeutically targetable disease. Cancer Cell 27:72–84. https​://doi.org/10.1016/j.ccell​ References .2014.11.002 14. Iyer MK, Chinnaiyan AM, Maher CA (2011) ChimeraScan: a tool for identifying chimeric transcription in sequencing data. Bioin- 1. Behjati S, Gundem G, Wedge DC, Roberts ND, Tarpey PS, Cooke formatics 27:2903–2904. https://doi.org/10.1093/bioin​ forma​ tics/​ SL, Van Loo P, Alexandrov LB, Ramakrishna M, Davies H et al btr46​7 (2016) Mutational signatures of ionizing radiation in second 15. Jakacki RI, Burger PC, Zhou T, Holmes EJ, Kocak M, Onar A, malignancies. Nat Commun 7:12605. https​://doi.org/10.1038/ Goldwein J, Mehta M, Packer RJ, Tarbell N et al (2012) Outcome ncomm​s1260​5 of children with metastatic medulloblastoma treated with carbo- 2. Boeva V, Popova T, Bleakley K, Chiche P, Cappo J, Schleier- platin during craniospinal radiotherapy: a Children’s Oncology macher G, Janoueix-Lerosey I, Delattre O, Barillot E (2012) Group Phase I/II study. J Clin Oncol 30:2648–2653. https​://doi. Control-FREEC: a tool for assessing copy number and allelic org/10.1200/JCO.2011.40.2792 content using next-generation sequencing data. Bioinformatics 16. Korshunov A, Benner A, Remke M, Lichter P, von Deimling 28:423–425. https​://doi.org/10.1093/bioin​forma​tics/btr67​0 A, Pfster S (2008) Accumulation of genomic aberrations dur- 3. Brastianos PK, Taylor-Weiner A, Manley PE, Jones RT, Dias- ing clinical progression of medulloblastoma. Acta Neuropathol Santagata D, Thorner AR, Lawrence MS, Rodriguez FJ, Bernardo 116:383–390. https​://doi.org/10.1007/s0040​1-008-0422-y LA, Schubert L et al (2014) Exome sequencing identifes BRAF 17. Korshunov A, Chavez L, Northcott PA, Sharma T, Ryzhova M, Jones DTW, von Deimling A, Pfster SM, Kool M (2017)

1 3 Acta Neuropathologica

DNA-methylation profling discloses signifcant advantages over of glioblastoma. PLoS One 9:e103327. https://doi.org/10.1371/​ NanoString method for molecular classifcation of medulloblas- journ​al.pone.01033​27 toma. Acta Neuropathol 134:965–967. https​://doi.org/10.1007/ 33. Park AK, Lee SJ, Phi JH, Wang KC, Kim DG, Cho BK, Haberler s0040​1-017-1776-9 C, Fattet S, Dufour C, Puget S et al (2012) Prognostic classifca- 18. Li H, Durbin R (2009) Fast and accurate short read alignment with tion of pediatric medulloblastoma based on chromosome 17p Burrows–Wheeler transform. Bioinformatics 25:1754–1760. https​ loss, expression of MYCC and MYCN, and Wnt pathway acti- ://doi.org/10.1093/bioin​forma​tics/btp32​4 vation. Neuro-oncology 14:203–214. https​://doi.org/10.1093/ 19. MacDonald TJ, Aguilera D, Kramm CM (2011) Treatment of neuon​c/nor19​6 high-grade glioma in children and adolescents. Neuro-oncology 34. Paugh BS, Qu C, Jones C, Liu Z, Adamowicz-Brice M, Zhang 13:1049–1058. https​://doi.org/10.1093/neuon​c/nor09​2 J, Bax DA, Coyle B, Barrow J, Hargrave D et al (2010) Inte- 20. Martin M (2011) Cutadapt removes adapter sequences from grated molecular genetic profling of pediatric high-grade glio- highthroughput sequencing reads. EMBnet J 17:10–12 mas reveals key diferences with the adult disease. J Clin Oncol 21. Massimino M, Gandola L, Spreafco F, Biassoni V, Luksch R, 28:3061–3068. https​://doi.org/10.1200/JCO.2009.26.7252 Collini P, Solero CN, Simonetti F, Pignoli E, Cefalo G et al 35. Paugh BS, Zhu X, Qu C, Endersby R, Diaz AK, Zhang J, Bax (2009) No salvage using high-dose chemotherapy plus/minus DA, Carvalho D, Reis RM, Onar-Thomas A et al (2013) Novel reirradiation for relapsing previously irradiated medulloblas- oncogenic PDGFRA mutations in pediatric high-grade gliomas. toma. Int J Radiat Oncol Biol Phys 73:1358–1363. https​://doi. Cancer Res 73:6219–6229. https​://doi.org/10.1158/0008-5472. org/10.1016/j.ijrob​p.2008.06.1930 CAN-13-1491 22. McPherson A, Hormozdiari F, Zayed A, Giuliany R, Ha G, 36. Phillips JJ, Aranda D, Ellison DW, Judkins AR, Croul SE, Brat DJ, Sun MG, Grifth M, Heravi Moussavi A, Senz J, Melnyk N Ligon KL, Horbinski C, Venneti S, Zadeh G et al (2013) PDGFRA et al (2011) deFuse: an algorithm for gene fusion discovery in amplifcation is common in pediatric and adult high-grade astro- tumor RNA-Seq data. PLoS Comput Biol 7:e1001138. https​:// cytomas and identifes a poor prognostic group in IDH1 mutant doi.org/10.1371/journ​al.pcbi.10011​38 glioblastoma. Brain Pathol 23:565–573. https​://doi.org/10.1111/ 23. Michaud K, de Tayrac M, D’Astous M, Duval C, Paquet C, bpa.12043​ Samassekou O, Gould PV, Saikali S (2016) Contribution of 37. Pietsch T, Schmidt R, Remke M, Korshunov A, Hovestadt V, Jones 1p, 19q, 9p and 10q automated analysis by FISH to the diag- DT, Felsberg J, Kaulich K, Goschzik T, Kool M et al (2014) Prog- nosis and prognosis of oligodendroglial tumors according to nostic signifcance of clinical, histopathological, and molecular WHO 2016 guidelines. PLoS One 11:e0168728. https​://doi. characteristics of medulloblastomas in the prospective HIT2000 org/10.1371/journ​al.pone.01687​28 multicenter clinical trial cohort. Acta Neuropathol 128:137–149. 24. Morrissy AS, Garzia L, Shih DJ, Zuyderduyn S, Huang X, https​://doi.org/10.1007/s0040​1-014-1276-0 Skowron P, Remke M, Cavalli FM, Ramaswamy V, Lindsay 38. Pugh TJ, Weeraratne SD, Archer TC, Pomeranz Krummel DA, PE et al (2016) Divergent clonal selection dominates medul- Auclair D, Bochicchio J, Carneiro MO, Carter SL, Cibulskis K, loblastoma at recurrence. Nature 529:351–357. https​://doi. Erlich RL et al (2012) Medulloblastoma exome sequencing uncov- org/10.1038/natur​e1647​8 ers subtype-specifc somatic mutations. Nature 488:106–110. 25. Ng AK, Kenney LB, Gilbert ES, Travis LB (2010) Second- https​://doi.org/10.1038/natur​e1132​9 ary malignancies across the age spectrum. Semin Radiat Oncol 39. Ramaswamy V, Remke M, Boufet E, Bailey S, Cliford SC, Doz 20:67–78. https​://doi.org/10.1016/j.semra​donc.2009.09.002 F, Kool M, Dufour C, Vassal G, Milde T et al (2016) Risk strati- 26. Ning MS, Perkins SM, Dewees T, Shinohara ET (2015) Evi- fcation of childhood medulloblastoma in the molecular era: the dence of high mortality in long term survivors of childhood current consensus. Acta Neuropathol 131:821–831. https​://doi. medulloblastoma. J Neurooncol 122:321–327. https​://doi. org/10.1007/s0040​1-016-1569-6 org/10.1007/s1106​0-014-1712-y 40. Ramaswamy V, Remke M, Boufet E, Faria CC, Perreault S, Cho 27. Northcott PA, Dubuc AM, Pfster S, Taylor MD (2012) Molec- YJ, Shih DJ, Luu B, Dubuc AM, Northcott PA et al (2013) Recur- ular subgroups of medulloblastoma. Expert Rev Neurother rence patterns across medulloblastoma subgroups: an integrated 12:871–884. https​://doi.org/10.1586/ern.12.66 clinical and molecular analysis. Lancet Oncol 14:1200–1207. 28. Northcott PA, Jones DT, Kool M, Robinson GW, Gilbertson RJ, https​://doi.org/10.1016/S1470​-2045(13)70449​-2 Cho YJ, Pomeroy SL, Korshunov A, Lichter P, Taylor MD et al 41. Robinson G, Parker M, Kranenburg TA, Lu C, Chen X, Ding (2012) Medulloblastomics: the end of the beginning. Nat Rev L, Phoenix TN, Hedlund E, Wei L, Zhu X et al (2012) Novel Cancer 12:818–834. https​://doi.org/10.1038/nrc34​10 mutations target distinct subgroups of medulloblastoma. Nature 29. Northcott PA, Shih DJ, Peacock J, Garzia L, Morrissy AS, 488:43–48. https​://doi.org/10.1038/natur​e1121​3 Zichner T, Stutz AM, Korshunov A, Reimand J, Schumacher 42. Robinson GW, Orr BA, Wu G, Gururangan S, Lin T, Qad- SE et al (2012) Subgroup-specifc structural variation across doumi I, Packer RJ, Goldman S, Prados MD, Desjardins A et al 1,000 medulloblastoma genomes. Nature 488:49–56. https://doi.​ (2015) vismodegib exerts targeted efficacy against recurrent org/10.1038/natur​e1132​7 sonic Hedgehog-subgroup medulloblastoma: results from phase 30. Northcott PA, Shih DJ, Remke M, Cho YJ, Kool M, Hawkins C, II pediatric brain tumor consortium studies PBTC-025B and Eberhart CG, Dubuc A, Guettouche T, Cardentey Y et al (2012) PBTC-032. J Clin Oncol 33:2646–2654. https​://doi.org/10.1200/ Rapid, reliable, and reproducible molecular sub-grouping of JCO.2014.60.1591 clinical medulloblastoma samples. Acta Neuropathol 123:615– 43. Rudin CM, Hann CL, Laterra J, Yauch RL, Callahan CA, Fu L, 626. https​://doi.org/10.1007/s0040​1-011-0899-7 Holcomb T, Stinson J, Gould SE, Coleman B et al (2009) Treat- 31. Noushmehr H, Weisenberger DJ, Diefes K, Phillips HS, Pujara ment of medulloblastoma with hedgehog pathway inhibitor GDC- K, Berman BP, Pan F, Pelloski CE, Sulman EP, Bhat KP et al 0449. N Engl J Med 361:1173–1178. https​://doi.org/10.1056/ (2010) Identifcation of a CpG island methylator phenotype that NEJMo​a0902​903 defnes a distinct subgroup of glioma. Cancer Cell 17:510–522. 44. Sandmann T, Bourgon R, Garcia J, Li C, Cloughesy T, Chinot https​://doi.org/10.1016/j.ccr.2010.03.017 OL, Wick W, Nishikawa R, Mason W, Henriksson R et al (2015) 32. Oh YT, Cho HJ, Kim J, Lee JH, Rho K, Seo YJ, Choi YS, Jung Patients with proneural glioblastoma may derive overall survival HJ, Song HS, Kong DS et al (2014) Translational validation of beneft from the addition of bevacizumab to frst-line radiother- personalized treatment strategy based on genetic characteristics apy and temozolomide: retrospective analysis of the AVAglio

1 3 Acta Neuropathologica

Trial. J Clin Oncol 33:2735–2744. https​://doi.org/10.1200/ 48. Verhaak RG, Hoadley KA, Purdom E, Wang V, Qi Y, Wilker- JCO.2015.61.5005 son MD, Miller CR, Ding L, Golub T, Mesirov JP et al (2010) 45. Spence T, Sin-Chan P, Picard D, Barszczyk M, Hoss K, Lu M, Integrated genomic analysis identifes clinically relevant sub- Kim SK, Ra YS, Nakamura H, Fangusaro J et al (2014) CNS- types of glioblastoma characterized by abnormalities in PDG- PNETs with C19MC amplification and/or LIN28 expression FRA, IDH1, EGFR, and NF1. Cancer Cell 17:98–110. https://doi.​ comprise a distinct histogenetic diagnostic and therapeutic entity. org/10.1016/j.ccr.2009.12.020 Acta Neuropathol 128:291–303. https​://doi.org/10.1007/s0040​ 49. Worst BC, van Tilburg CM, Balasubramanian GP, Fiesel P, Witt 1-014-1291-1 R, Freitag A, Boudalil M, Previti C, Wolf S, Schmidt S et al 46. Suzuki K, Ojima M, Kodama S, Watanabe M (2003) Radiation- (2016) Next-generation personalised medicine for high-risk pae- induced DNA damage and delayed induced genomic instability. diatric cancer patients—the INFORM pilot study. Eur J Cancer Oncogene 22:6988–6993. https://doi.org/10.1038/sj.onc.12068​ 81​ 65:91–101. https​://doi.org/10.1016/j.ejca.2016.06.009 47. Trapnell C, Williams BA, Pertea G, Mortazavi A, Kwan G, van 50. Yamanaka R, Hayano A, Kanayama T (2016) Radiation-induced Baren MJ, Salzberg SL, Wold BJ, Pachter L (2010) Transcript gliomas: a comprehensive review and meta-analysis. Neurosurg assembly and quantifcation by RNA-Seq reveals unannotated Rev. https​://doi.org/10.1007/s1014​3-016-0786-8 transcripts and isoform switching during cell diferentiation. Nat Biotechnol 28:511–515. https​://doi.org/10.1038/nbt.1621

Afliations

Ji Hoon Phi1,2 · Ae Kyung Park3 · Semin Lee4,5 · Seung Ah Choi1,2 · In‑Pyo Baek6 · Pora Kim7 · Eun‑Hye Kim8 · Hee Chul Park9 · Byung Chul Kim9 · Jong Bhak4,5 · Sung‑Hye Park10 · Ji Yeoun Lee1,2,11 · Kyu‑Chang Wang1,2 · Dong‑Seok Kim12 · Kyu Won Shim12 · Se Hoon Kim13 · Chae‑Yong Kim14 · Seung‑Ki Kim1,2

1 Division of Pediatric Neurosurgery, Pediatric Clinical 9 Clinomics Inc., Ulsan 44919, Republic of Korea Neuroscience Center, Seoul National University Children’s 10 Department of Pathology, Seoul National University College Hospital, Seoul 03080, Republic of Korea of Medicine, Seoul 03080, Republic of Korea 2 Department of Neurosurgery, Seoul National University 11 Department of Anatomy, Seoul National University College Hospital, Seoul National University College of Medicine, of Medicine, Seoul 03080, Republic of Korea Seoul 03080, Republic of Korea 12 Department of Pediatric Neurosurgery, Severance Children’s 3 College of Pharmacy and Research Institute of Life Hospital, Yonsei University College of Medicine, Brain and Pharmaceutical Sciences, Sunchon National University, Korea 21 Project for Medical Science, Seoul 03722, Suncheon 57922, Republic of Korea Republic of Korea 4 Department of Biomedical Engineering, School of Life 13 Department of Pathology, Severance Hospital, Sciences, Ulsan National Institute of Science and Technology Yonsei University College of Medicine, Seoul 03722, (UNIST), Ulsan 44919, Republic of Korea Republic of Korea 5 Korean Genomics Industrialization and Commercialization 14 Department of Neurosurgery, Seoul National University Center (KOGIC), Ulsan 44919, Republic of Korea Bundang Hospital, Seongnam 13620, Republic of Korea 6 TheragenEtex Bio Institute, Suwon 16229, Republic of Korea 7 School of Biomedical Informatics, The University of Texas Health Science Center at Houston, Houston, TX 77030, USA 8 Gerotech Inc., Ulsan 44919, Republic of Korea

1 3