ARTICLE https://doi.org/10.1038/s41467-020-19175-0 OPEN Circulating tumour DNA from the cerebrospinal fluid allows the characterisation and monitoring of medulloblastoma Laura Escudero 1, Anna Llort 2, Alexandra Arias1, Ander Diaz-Navarro 3,4, Francisco Martínez-Ricarte2,5, Carlota Rubio-Perez 1, Regina Mayor1, Ginevra Caratù1, Elena Martínez-Sáez2, Élida Vázquez-Méndez2, Iván Lesende-Rodríguez6, Raquel Hladun2, Luis Gros2, Santiago Ramón y Cajal2, Maria A. Poca2,5, Xose S. Puente 3,4, Juan Sahuquillo 2,5, Soledad Gallego2,5 & Joan Seoane 1,4,5,7 ✉ 1234567890():,; The molecular characterisation of medulloblastoma, the most common paediatric brain tumour, is crucial for the correct management and treatment of this heterogenous disease. However, insufficient tissue sample, the presence of tumour heterogeneity, or disseminated disease can challenge its diagnosis and monitoring. Here, we report that the cerebrospinal fluid (CSF) circulating tumour DNA (ctDNA) recapitulates the genomic alterations of the tumour and facilitates subgrouping and risk stratification, providing valuable information about diagnosis and prognosis. CSF ctDNA also characterises the intra-tumour genomic heterogeneity identifying small subclones. ctDNA is abundant in the CSF but barely present in plasma and longitudinal analysis of CSF ctDNA allows the study of minimal residual disease, genomic evolution and the characterisation of tumours at recurrence. Ultimately, CSF ctDNA analysis could facilitate the clinical management of medulloblastoma patients and help the design of tailored therapeutic strategies, increasing treatment efficacy while reducing excessive treatment to prevent long-term secondary effects. 1 Vall d’Hebron Institute of Oncology (VHIO), Vall d’Hebron University Hospital, 08035 Barcelona, Spain. 2 Vall d’Hebron Institut de Recerca (VHIR), Vall d’Hebron University Hospital, 08035 Barcelona, Spain. 3 Dpto. de Bioquímica y Biología Molecular, IUOPA-Universidad de Oviedo, 33006 Oviedo, Spain. 4 CIBERONC, Barcelona, Spain. 5 Universitat Autònoma de Barcelona (UAB), 08193 Cerdanyola del Vallès, Spain. 6 Universidade de A Coruña (UDC), 15006 A Coruña, Spain. 7 Institució Catalana de Recerca i Estudis Avançats (ICREA), 08010 Barcelona, Spain. ✉email: jseoane@vhio.net NATURE COMMUNICATIONS | (2020)11:5376 | https://doi.org/10.1038/s41467-020-19175-0 | www.nature.com/naturecommunications 1 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-19175-0 M edulloblastoma (MB) is the most prevalent malignant alterations of the tumour and allows the subgroup classification brain tumour in childhood. This embryonal tumour of and risk stratification of MB patients. In addition, the study of the central nervous system (CNS) is a complex and follow-up CSF samples identifies minimal residual disease and evolving heterogeneous disease with a wide range of prognosis reveals the genomic evolution at relapse. Altogether, the analysis being in some cases an extremely aggressive malignancy. MB can of CSF ctDNA is a relatively noninvasive and reliable tool that be subdivided into four molecular consensus subgroups: WNT, could aid in the diagnosis, prognosis and monitoring of children SHH, group 3 and group 4. Further subtypes have been pro- with MB. posed1–5 and a consensus risk-classification scheme that stratify patients in four groups with a 5-year progression-free survival Results of 91% (favourable), 81% (standard), 42% (high risk) and 28% Classification and risk stratification of MB patients. In order to (very high risk) has been established4,6. assess whether the analysis of CSF ctDNA could be a tool for the Treatment for MB includes tumour resection, chemotherapy and, diagnosis and monitoring of MB patients, tumour, blood and CSF for most patients, craniospinal radiotherapy. Brain tumour surgical samples were collected from 13 patients (Fig. 1a). resection is currently the first treatment option and surgical speci- Hydrocephalus was detected through MRI in all patients at the mens are used for diagnosis and tumour characterisation. However, time of hospital admission and CSF drainage, through the lack of sufficient sample, tissue artifacts or excessive presence of implantation of a transitory external ventricular drain (EVD), was non-tumour tissue can create uncertainty to accurately characterise required to reduce ICP. Aiming to avoid the implantation of a and classify tumours. definitive CSF shunt, 12 of the 13 patients underwent endoscopic In addition to the interpatient tumour heterogeneity, intratu- third ventriculocisternostomy before the implantation of the EVD. mour heterogeneity has been described in MB and the efficacy of During this intervention, the CSF sample was obtained and used for single biopsies to representatively characterise this complexity can standard of care clinical measurements (cytology) and research be limited. The fact that spatial heterogeneity has been reported at purposes. In the following days, the patients underwent surgical the level of somatic mutations and copy number alterations resection of the posterior fossa tumour and additional CSF samples (CNAs) exemplifies the limitations of single biopsies7. were obtained. Before the administration of chemotherapy, three Moreover, MB tumours evolve with time highlighting the need patients required the implantation of an Ommaya reservoir for longitudinal samples to characterise tumours at different time (intraventricular catheter attached to a reservoir implanted under points during the progression of the disease. Despite current efforts, the scalp) allowing for periodical CSF samples to be collected. For around 30% of patients relapse, locally or through metastatic dis- the remaining 10 patients, follow-up samples were acquired semination to the leptomeninges, which is the major cause of through a lumbar puncture. mortality in children8,9. The comparison of primary and relapsed Whole-exome sequencing (WES) of matching tumour/ tumours has shown that specific molecular alterations can appear in normal DNA was performed and common MB germline and the relapse setting10,11. In addition, secondary radiotherapy-induced somatic driver mutations, focal copy number variants (CNVs), tumours can emerge in some patients12–14. To assess metastatic arm-level and whole chromosome gain/loss were identified dissemination, cerebrospinal fluid (CSF) cytology together with (Fig. 1b and Supplementary Table 1). The frequency of somatic brain and spinal MRI are performed and patients are classified mutations in coding regions was calculated resulting in a according to Chang M-staging system15. Unfortunately, cytology median 0.36 mutations per Mb [range: 0.08–1.21], similar to exhibits limited sensitivity and sometimes cannot provide accurate previous studies2,30. information. Patients were classified into previously defined MB subgroups Altogether, this indicates that there is an underlying and and subtypes4,5, and were assigned a predicted risk stratification ongoing need to improve the molecular characterisation and (5-year PFS) based on the molecular subgroup and the presence monitoring of MB. A relatively noninvasive method to char- of risk modifiers (amplification of MYC and MYCN, metastasis at acterise primary and relapsed tumours and monitor the evolution diagnosis, chromosome 13 loss, large cell anaplastic (LCA) of the disease is required to assist the clinical management of MB histology)4 (Fig. 1c). patients. In this regard, liquid biopsies and in particular cell-free The clinicopathological characteristics, molecular features, circulating tumour DNA (ctDNA) have opened an avenue of MB-subgroup classification and risk stratification of the 13 MB research16. Tumour cells release DNA fragments that circulate in paediatric patients at the time of enrolment (12 at the time of body fluids, and the analysis of ctDNA can provide information first-diagnostic and 1 at the diagnostic of relapse (MB8)) are about the molecular profile and clinical evolution of cancer17. shown in Fig. 1d. Our group and others have previously reported the detection of CSF ctDNA in adult brain malignancies18–23. The analysis of ctDNA in the CSF better reflects the brain tumour genomic ctDNA is more abundant in CSF than in plasma. We sought to alterations than in plasma, it can be used as a molecular diag- evaluate whether ctDNA could be detected in the CSF of the 13 MB nostic tool and the dynamic changes during patient monitoring patients and how it compared to plasma ctDNA and standard of recapitulate the disease course18,21–25. care CSF cytology. To facilitate ctDNA detection, somatic muta- Hydrocephalus is very common in paediatric patients with tions were selected from the WES of matched tumour/normal posterior fossa tumours like MB. In these cases and prior to samples based on variant allele frequency (VAF) and impact. surgical treatment of the tumour, CSF drainage is required to Mutations were validated by ddPCR in the tumour/normal DNA alleviate the intracranial pressure (ICP)26,27. In addition, CSF and then the pre-surgery CSF and plasma cell-free DNA (cfDNA) samples are routinely collected for cytologic analysis. However, samples were analysed. despite the access to CSF, the study of CSF ctDNA has not yet No cells were detected in the CSF from the cytologic analysis been explored in the paediatric setting except for a few studies for any of the cases. However, ctDNA was detected in the CSF focusing on paediatric diffuse midline gliomas through the ana- samples obtained before the surgical intervention for 76.9% lysis of H3 mutations in the CSF ctDNA by droplet-digital PCR (10/13) of patients. In contrast, ctDNA could not be detected in (ddPCR) or targeted Sanger sequencing28,29. In the current study, plasma, except for one patient albeit with a very low VAF (2.2%) we detect ctDNA in the CSF but barely in the plasma of patients (Fig. 2a). Altogether, these results demonstrate that CSF with MB. The analysis of CSF ctDNA reflects the genomic represents a better source for ctDNA than plasma in MB patients 2 NATURE COMMUNICATIONS | (2020)11:5376 | https://doi.org/10.1038/s41467-020-19175-0 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-19175-0 ARTICLE a b # M u ta tio n s 60 40 20 0 2 4 6 0 TP53 23% GLI2 23% CSF MYCN 23% PTCH1 15% BCOR 8% BLM 8% CTNNB1 8% KMT2D 8% MYC 8% Tumour Alterations PRDM6 8% PTEN 8% Missense SUFU 8% Nonsense Plasma Frameshift 17p 46% 7 31% Inframe insertion 10q 15% Germline 16q 15% Amplification 11 15% Gain Blood 2 8% Deletion Patients with medulloblastoma (n = 13) 6 8% 8 8% 9q 8% ddPCR 10 8% tumour/normal 14q 8% CSF/plasma 16 8% WES MB5 MB13 MB8 MB12 MB3 MB9 MB7 MB6 MB11 MB4 MB2 MB1 MB10 tumour/normal WES CSF1/CSF2 c Schwalbe et al., 2017 Risk stratification WNT SHH-Child, SHH-Infant Risk modifiers Favourable WES 13 MB Mutations Group3-LR, Group3-HR, Group4-LR, Group4-HR M+, R+, MB LCA, Chr13–, Standard Identification of MB CNVs classification MYC amp, molecular features Hovestadt et al., 2020 system MYCN amp High WNT Cytogenetics SHH (α, β, γ, δ) Group3 /Group4 (I-VIII) Very high d MB subgroup Demographics Clinical features Molecular features classification model Tumour Vol M R Treatment PFS Germline Mut/ Hovestadt Schwalbe Risk Age Sex H* Pathology Mutations CNVs Cytogenetics location (cm3) status status protocol (years) mutations Mb et al., 2020 et al., 2017 stratification EVD SIOPE MB1 5 M Vermis 63.7 LCA M– R+ 5.46 0.12 ETV high risk EVD SIOPE 0.36 MB2 13 F ETV Vermis 19.8 LCA M– R– high risk 5.31 7+ G3/4 III-VII G3/4-LR Favourable EVD SIOPE BLM MB3 14 F ETV Vermis 27.7 LCA M– R+ high risk 5.23 0.48 MYC G3/4 II,III,V G3-HR High F EVD SIOPE 0.48 7+ 16q– MB4 0 ETV Vermis 23.3 LCA M– R– high risk 4.96 G3/4 V G4-LR Favourable Cerebellar SIOPE MYCNSUFU 17p– 16q– MB5 13 M EVD 53.6 LCA M– R– 2.49 TP53 0.52 SHH-α SHH-child Very high Hemisphere high risk GLI2 2+7+10q–16– 6 M EVD SIOPE MB6 ETV Vermis 33.8 C M– R+ high risk 2.29 0.10 17p– SHH-α SHH-child Very high EVD 65.0 M– R– 3.27 0.20 7+ 17p– 11– MB7 6 M ETV Vermis C PNET 5 G3/4 IV,VI G3/4-LR Favourable SIOP/UKCCSG MB8 3 F EVD Vermis 41.4 D/N M– R– 3.06 1.21 PTCH1 17p– 8– 9q– SHH-α SHH-child Very high MYCN 6 ETV Vermis 4.4 LCA M– R– CXT+RT 0.44 TP53 MB9 6 F EVD LCA SIOPE 0.18 17p– 6– ETV Vermis 28.2 M– R– high risk 2.88 CTNNB1 WNT WNT Favourable* EVD Vermis 64.4 M– R– HIT-MED 2.51 MB10 1 F ETV D/N bloc SKK 0.08 EVD MB11 6 M ETV Vermis 35.3 C M– R– PNET 5 2.48 0.40 17p– SHH-α SHH-child Favourable Middle PTCH1 MB12 3 M EVD 21.0 C M+ NA COG 3.09 PTCH1 0.28 MYCNPTEN cerebellar 99703 BCOR 10q– SHH-α,β,γ SHH-child Very high ETV peduncle KMT2D GLI2 MB13 13 F EVD Cerebellar SIOPE TP53 0.85 GLI2 14q– 11– 10– SHH-child Very high 56.3 LCA M– R– 1.32 SHH-α ETV hemisphere high risk and highlight the higher sensitivity of the ctDNA analysis (Fig. 1d). These results suggest that high CSF ctDNA levels could compared to cytology. be an indicator of more aggressive tumours. Different levels of CSF ctDNA at diagnosis were observed across the 13 patients. Three of the patients exhibited very high levels (above 25% VAF) when compared to the rest of the patients WES of CSF ctDNA reflects the tumour genomic alterations. and were identified as outliers (Fig. 2b). To investigate these At the time of completing this study, disease progression was not differences, the levels of ctDNA (VAF %) and cfDNA (ng/ml) in observed for 10/13 patients and CSF ctDNA was not detected the CSF were compared with the tumour volume (cm3) but no during patient monitoring (Supplementary Table 2). In contrast, correlations were observed (Supplementary Fig. 1). Interestingly, cancer progression was observed for the remaining three patients the three patients with abundant CSF ctDNA were at very high (MB5, MB6 and MB8), that were considered at very high risk of risk of relapse and disease progression was observed for two of relapse (Fig. 1c.). A thorough analysis of the tumour/normal them (MB6 and MB8) at 27.5 and 5.3 months, respectively DNA and CSF cfDNA at diagnosis and during follow-up or NATURE COMMUNICATIONS | (2020)11:5376 | https://doi.org/10.1038/s41467-020-19175-0 | www.nature.com/naturecommunications 3 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-19175-0 Fig. 1 Project outline and characterisation of a cohort of 13 paediatric patients with MB. a Study schema indicating the patient samples obtained and the methodology used for the analysis. b Molecular features, including driver events in MB, identified from the WES characterisation of 13 MB patients. The number of mutations identified for each patient is indicated in the top and the proportion of patients with each alteration on the right. Distinct alterations are indicated in the colour legend. c Summary of the MB common molecular alterations investigated for the classification into subgroup and association of risk stratification. d Clinicopathological characteristics, molecular features, MB-subgroup classification and risk stratification (based on Schwalbe et al., 2017 5-year progression-free survival (PFS)) for the 13 MB patients. PFS indicated as years until the date of progression (and if not applicable, death) in red or last medical check-up in black. The frequency of somatic mutations in coding regions per Mb (Mut/Mb) was calculated based on a 50.4 Mb (MB1- MB12) and 17.7 Mb (MB13) WES panel. CNV deletion (log2 value < −1, blue), amplification (log2 value >1, red). For the cytogenetic analysis, the threshold for the percentage of arm-level or whole chromosome gain/loss was determined by the mean and the percentiles of the events from the cohort, greater than 75% percentile (>65.5%) in black and greater than the mean (>44.5%) in grey. The risk stratification for patient MB9 was considered Favourable* because WNT subgroup (CTNNB1 & 6-) predominated over SHH (17p-). The definition of the symbols and acronyms are as follows. Hydrocephalus (H): external ventricular drain (EVD), endoscopic third ventriculocisternostomy (ETV). Pathology: large cell anaplastic (LCA), classic (C), desmoplastic/nodular (D/N). Metastasis (M) status: positive (+), negative (−). Residual (R) disease status: >1.5 cm2 (+), <1.5 cm2 (−). Treatment protocol: chemotherapy (CXT), radiotherapy (RT). High risk (HR), Low-risk (LR). a VAF (%) by ddPCR b Tumour CSF1 Plasma1 100 MB12 BCOR:p.E518* ND ND 90 VAF (%) MB8 PTCH1:p.L1024Pfs ND 100 80 MB3 TRRAP:p.E3107K ND CSF ctDNA VAF (%) 80 70 MB4 XRCC5:p.K35N 0.4 ND MB5 RASGRF1:p.D57H 6.0 ND 60 60 MB13 BMP5:p.R321* ND 40 50 MB9 CTNNB1:p.S37C ND 2.2 20 40 MB1 RIOK2:p.S456L 7.0 ND 0 30 MB10 HSPG2:p.R3992H ND ND MB6 SRCAP:p.R2444* ND 20 MB7 YY1AP1:p.R665* 0.2 ND 10 MB2 PUS7:p.D535A 3.3 ND 0 MB11 NFIA:p.R322H 9.0 ND MB cohort Fig. 2 ctDNA is abundant in the CSF of paediatric patients with MB. a Heatmap representing the VAFs analysed by ddPCR from tumour DNA, and CSF and plasma ctDNA samples obtained before surgical resection for the 13 patients. Colour legend for mutant allelic frequencies (VAFs) is shown. VAFs values <10% are indicated. Not detected (ND). b Distribution of CSF ctDNA VAF. Outliers identified using ROUT method are highlighted with a grey circle and selected as High CSF ctDNA. relapse of these three patients, together with another predicted MB SHH subgroup, for which residual disease, metastatic dis- very high-risk patient (MB13), was undertaken. semination, LCA histology or MYCN amplification are risk In order to determine whether WES of ctDNA in the CSF modifiers indicators of very high risk4. Importantly, these results could reliably identify and reflect the genomic alterations of the showed that the CSF ctDNA analysis could be used as a diag- tumour, we performed WES of the tumour and CSF cfDNA. We nostic tool to classify and risk-stratify patients. found that 98.9% (88/89) of the mutations detected in the primary tumour sample with VAF > 5% could also be detected in CSF ctDNA monitoring identifies minimal residual disease. the matching CSF ctDNA. Importantly, there was a significant We next sought to address whether the WES of CSF ctDNA could correlation between the tumour VAFs and the ones obtained assist in the monitoring of the genomic tumour evolution of MB from the CSF ctDNA (R2 = 0.57, 0.96, 0.53 and 0.87) (Fig. 3a–d). patients and provide information concerning response to treat- This observation indicates that the CSF ctDNA recapitulated the ment or disease progression. Follow-up samples of CSF and intratumour heterogeneity present in the primary tumour. Thus, plasma were collected in order to determine whether the levels of the analysis of CSF ctDNA can provide information about the CSF ctDNA could fluctuate with time during the course of the subclonal genomic architecture of MB tumours. disease. The analysis of CNVs, arm-level or whole chromosome gain/ Patient MB6 underwent surgical resection and was diagnosed loss revealed a similar situation. The overall CNV landscape in as classic MB. Postoperative imaging identified residual disease. the CSF samples resembled the one found in the matching After first-line treatment completion, relapse was observed, and tumours (MB6, MB8 and MB13). Despite the restricted amount the patient was subsequently treated with second-line chemo- of cfDNA from MB5-CSF, key MB CNVs detected in the tumour and radiotherapy. Despite an initial reduction in tumour could also be detected in the CSF (Fig. 3e). size, sequential MRIs throughout 7 months showed that the leptomeningeal nodule was stable (no change in size was CSF ctDNA facilitates MB-subgroup and risk stratification. To observed) (Fig. 4a). When the follow-up CSF ctDNA was determine whether the molecular alterations identified from analysed, mutations present in the tumour and in the first CSF sequencing the CSF ctDNA could provide diagnostic and prog- were also identified. VAFs of the mutations found in the follow- nostic information, patients were classified into a MB-subgroup up CSF showed a substantial decrease as expected due to the and risk-stratified (Fig. 3e). MB common molecular alterations, decrease in tumour burden caused by the surgical resection including PTCH1 and TP53 mutation; MYCN and GLI2 ampli- (Fig. 4a–c and Supplementary Fig. 2). We were unable to detect fication; SUFU deletion and 17p loss were detected and facilitated additional mutations in follow-up CSF ctDNA; however, subgroup classification. The four patients were identified as we identified 3/5 mutations with a VAF < 1.5% that were 4 NATURE COMMUNICATIONS | (2020)11:5376 | https://doi.org/10.1038/s41467-020-19175-0 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-19175-0 ARTICLE a b c d MB5 MB6 MB8 MB13 100 100 100 100 CSF ctDNA (VAF %) CSF ctDNA (VAF %) CSF ctDNA (VAF %) CSF ctDNA (VAF %) 80 80 80 80 60 2 R = 0.57 60 60 60 40 p = 0.0181 40 40 40 n=9 R 2 = 0.96 2 R = 0.53 R 2 = 0.87 20 20 p = 0.0203 p < 0.0001 p < 0.0001 20 20 n=4 n = 61 n = 15 0 0 0 0 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100 Tumour (VAF %) Tumour (VAF %) Tumour (VAF %) Tumour (VAF %) e Demographics Clinical features Molecular features Classification model M R Germline Mut/ Hovestadt Schwalbe Risk Age Sex Pathology Cytogenetics stratification status status mutations Mb Mutations CNVs et al., 2020 et al., 2017 MYCNSUFU 17p– 16q– Tumour 0.51 2+7+10q–16– SHH-α SHH-child Very high MB5 GLI2 13 M LCA M– R– TP53 MYCN 17p– (33%) Very high CSF 0.45 SHH-α SHH-child GLI2 Tumour 0.23 17p– SHH-α SHH-child Very high MB6 6 M C M– R+ CSF 0.17 17p– SHH-α SHH-child Very high PTCH1 Tumour D/N M– R– 1.21 MYCN 17p– 8– 9q– SHH-α SHH-child Very high 3 F TP53 MB8 6 LCA M– R– CSF 2.17 PTCH1 MYCN 17p– 8– 9q– SHH-child Very high SHH-α TP53 OTX2 Tumour 0.85 14q– 11–10– SHH-α SHH-child Very high MB13 GLI2 13 F LCA M– R– TP53 CSF 0.85 GLI2 14q– 10q– 10– SHH-α SHH-child Very high Fig. 3 WES of CSF ctDNA reveals the genomic alterations of the tumour and identifies MB-subgroup. a–d Correlation of CSF ctDNA VAF with tumour VAF for each mutation identified in the tumour sample (VAF > 5%). Linear regression and Goodness-of-fit R2 indicated in each figure. n, represents the number of mutations detected in the primary tumour sample with VAF > 5%. e Clinicopathological characteristics and molecular alterations based on tumour and CSF ctDNA WES analysis. MB-subgroup classification and risk stratification (based on Schwalbe et al., 2017 5-year PFS). Frequency of somatic mutations in coding regions per Mb (Mut/Mb) based on a 17.7 Mb (MB5, MB6, MB13), 50.4 Mb (MB8-Tumour) and 51.6 Mb (MB8-CSF) WES panel. CNV: deletion (log2 value < −1, blue), amplification (log2 value > 1, red). For the cytogenetic analysis, the threshold for the percentage of arm-level or chromosome gain/loss was determined by the mean and the percentiles of the events from the cohort of 13 MB as indicated in Fig. 1: greater than 75% percentile (>65.5%) in black and greater than the mean (>44.5%) in grey. The definition of the symbols and acronyms are in accordance with Fig. 1 and are as follows. Pathology: large cell anaplastic (LCA), classic (C), desmoplastic/nodular (D/N). Metastasis (M) status: positive (+), negative (−). Residual (R) disease status: >1.5 cm2 (+), <1.5 cm2 (−). previously detected in both tumour and CSF1 (Fig. 4b). Among Analysis of CSF ctDNA reveals genomic evolution. In contrast these, a stop gain in SRCAP (SRCAP:p.R2444*) was validated to the previous cases outlined, patients MB5 and MB8 progressed by ddPCR (Fig. 4c and Supplementary Fig. 2). In addition, loss and died from their disease following relapse. of the chromosome 17p arm was identified in both tumour and Patient MB8 underwent surgery and was diagnosed as a CSF1 (Fig. 3e). The fact that a small amount of ctDNA was still desmoplastic/nodular (D/N) MB. Postoperative imaging identi- observed in the CSF sample obtained at the end of the fied residual disease <1.5 cm2 and treatment was initiated. Three treatment indicated that the small residual nodule that persisted years postdiagnosis, the patient relapsed and dissemination to in the patient could be detected through the analysis of CSF CSF was detected by MRI. After a second surgery, the histological ctDNA (Fig. 4b, c). appearances changed from D/N MB to LCA MB. Chemo- and A similar case was patient MB13. This patient underwent radiotherapy treatment was initiated but within 5 months, the tumour resection and was diagnosed with LCA MB. Postsurgery patient died (Fig. 5a). MRI determined residual disease <1.5 cm2 while, interestingly, The sequencing analysis of the tumour and CSF samples at CSF ctDNA was still detected with VAF 7.6% for BMP5:p.R321* relapse revealed common alterations including four predicted 13 days after surgery (Fig. 4d, g). Treatment was initiated and no drivers: PTCH1:p.L1024Pfs, TP53:p.N239Tfs, CDKN2B:p.G124D evidence of residual disease was observed during follow-up and KMT2A:p.W1199S (Fig. 5b). Mutations in PTCH1 and imaging performed two and a half months after surgery. The CDKN2B were validated using ddPCR in the tumour/normal mutational landscape and CNV profile of the first CSF collected DNA and the CSF and plasma cfDNA (Supplementary Fig. 2). highly resembled that of the tumour. The analysis of the follow- Importantly, PTCH1 oncogenic mutations are described as up CSF samples showed that 3/15 mutations were detected with actionable gene mutations for SMO-inhibitors vismodegig and a VAF < 1% (Fig. 4e, f). Among these, BMP5:p.R321* and MIB1: sonidegib (in clinical trials for recurrent or refractory MB showing p.M468I were validated by ddPCR (Fig. 4g and Supplementary encouraging results)31–34. It is also important to highlight the CNV Fig. 2). These results indicated that although no malignancy was landscape (Fig. 5c) where the presence of MYCN amplification observed by MRI, a residual disease was still present since the provided information about resistance to SMO inhibitors in MB35. CSF ctDNA was detected. Both cases, MB6 and MB13, In addition to the tumour-specific mutations that were also detected evidenced that CSF ctDNA could facilitate the identification of in the ctDNA, CSF-private mutations were identified such as the minimal residual disease in MB patients. driver mutation KMT2A:p.W1199S that was validated by ddPCR. NATURE COMMUNICATIONS | (2020)11:5376 | https://doi.org/10.1038/s41467-020-19175-0 | www.nature.com/naturecommunications 5 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-19175-0 a EVD and ETV Treatment (SIOPE high-risk) Treatment (CTX + RT) Tumour resection MB6 Day 6 Day 0 Day 0 Day 7 Day 841 Day 1057 Day 1064 Diagnostic (MRI) Residual (MRI) Relapse (MRI) Follow-up (MRI) Samples Samples CSF Blood CSF Blood (Plasma) (Plasma) 50 × 30 × 45 mm R >1.5 cm2 7 months b WES VAF (%) c ddPCR VAF (%) 50 VAF (%) Tumour CSF CSF CSF ctDNA VAF (%) 100 40 GJC3:p.T55I d. 6 d. 0 d. 1064 SRCAP:p.R2444* 80 SRCAP:p.R2444* 0.4 30 60 YARS:p.V394V 2.0 ND 40 20 PRX:p.EE1359–1360E– 5.8 1.5 20 10 ZFHX3:p.Q1741– 4.8 0.5 GJC3:p.T55I ND ND 0 2 0. 0.1 0 n=5 0 250 500 750 1000 1250 1500 Time (Days) d EVD and ETV Treatment (SIOPE High-risk) Tumour resection MB13 Day 3 Day –5 Day 0 Day 6 Day 78 Day 120 Diagnostic (MRI) Residual (MRI) Follow-up (MRI) Samples Samples CSF Blood CSF Blood (Plasma) (Plasma) 50 × 50 × 45 mm R <1.5 cm2 e WES VAF (%) f WES CNV Tumour CSF CSF Tumour CSF CSF d. 3 d. 0 d. 120 VAF (%) d. 3 d. 120 d. 0 TMC8:p.S315N 100 GLI2:AMP ND BMP5:p.R321* 0.4 80 GPX4:p.C16F ND 60 UGT1A6:p.P266L ND 40 g ddPCR VAF (%) NEB:p.T5897N ND 30 20 CSF ctDNA VAF (%) COL4A1:p.P102T ND LRP2:p.V4141I ND 0 MIB1:p.M468I TBCE:p.E101A ND 20 BMP5:p.R321* CTRC:p.T58T ND SHH:p.A376D ND WDR45:p.L65F ND 10 SYTL5 (.) ND HUWE1:p.V1577D 0.4 1.1 PHIP:p.V165F ND 0 MIB1:p.M468I 1.0 0 100 200 300 400 n = 15 Time (days) These CSF-specific mutations either revealed intratumour hetero- perimedullary leptomeningeal implant at cervical cord was geneity and were missed from the tumour biopsy analysed or identified. Thirty months after diagnosis, the patient relapsed. indicated the interlesion heterogeneity from the diffused leptome- Following tumour resection and chemotherapy, an acute tumour ningeal nodules detected by MRI (Fig. 5a–c and Supplementary progression was observed, and the patient died 9 months after Fig. 2). relapse (Fig. 5d). Similarly, patient MB5 underwent surgical resection and was Despite the limited amount of CSF cfDNA, mutations and diagnosed as LCA MB. Postoperative imaging identified residual most relevant genomic alterations identified in the tumour were disease <1.5 cm2. Treatment was initiated and during follow-up a also detected in the CSF ctDNA, including MYCN and GLI2 6 NATURE COMMUNICATIONS | (2020)11:5376 | https://doi.org/10.1038/s41467-020-19175-0 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-19175-0 ARTICLE Fig. 4 CSF ctDNA characterises the primary tumour and identifies minimal residual disease in patients that respond to treatment. Longitudinal monitoring of two paediatric patients with MB: a–c MB6 and d–g MB13. a, d Clinical course indicating intervention for hydrocephalus, tumour surgery, treatment protocol, radiologic images (MRI), tumour size or residual disease, and patient samples obtained. Yellow arrow within the MRI indicates the tumour at relapse. b, e, f WES analysis of brain tumour DNA and CSF cfDNA before surgery (CSF1) and during follow-up (CSF2). b, e Heatmap representing the mutations identified; VAF values <10% were annotated and blue arrows indicate the mutations selected for further validation. f Relevant MB CNV identified were indicated in red for amplification (log2 > 1) or blue for deletion (log2 < −1). c, g Graphical representation of the CSF ctDNA dynamics over time for the selected mutations using ddPCR. Definition of symbols and acronyms: not detected (ND), day (d), external ventricular drain (EVD), endoscopic third ventriculocisternostomy (ETV), residual (R) disease status: >1.5 cm2 (+), <1.5 cm2 (−), Society of Pediatric Oncology Europe (SIOPE) protocol, chemotherapy (CXT) and radiotherapy (RT). a Day 1115 b WES VAF (%) EVD and ETV Tumour resection Tumour resection Treatment (SIOP/UKCCSG) Treatment (CXT+RT) Tumour CSF MB8 d.1115 VAF (%) Day 0 GEMIN4:p.A537T 100 PTCH1:p.L1024Pfs 80 Exitus TP53:p.N239Tfs ACSL1:p.A74T 60 Day –6 Day 1 Day 1100 Day 1118 Day 1205 PRRC2C:p.G2448E CARS2:p.D354H 40 Diagnostic (MRI) Residual (MRI) Relapse (MRI) Day 1115 Residual (MRI) Follow-up (MRI) PLEKHA4:p.Q577H 20 Samples ABCB10:p.R464C SLC4A2:p.F1009L 0 BCAP31:p.V127L CDKN2B:p.G124D n = 61 CSF Blood IL17RC (.) ND (Plasma) HTT:p.F1939L ND KMT2A:p.W1199S ND PRKCD:p.D329N ND FAM205A:p.V1147 ND 46 × 40 × 46 mm R <1.5 cm2 PSD2:p.A412V 0.8 POU3F2:p.Q142- 2.3 BRI3BP:p.L16- ND c WES CNV MYC:p.E425A MYC:p.E425K ND ND Tumour CSF ARID1B:p.G321- ND 7.7 ERCC3:p.EE263-264E ND 5.3 d.1115 d.1115 MYCN:AMP n = 53 OTX2:AMP d Day 0 Day 910 e EVD Tumour resection Tumour resection WES VAF (%) Treatment (SIOPE high-risk) Treatment (CTX) MB5 Tumour CSF CSF d.0 d.0 d.1173 VAF (%) RPL266:p.V95G 1.1 100 Exitus ZNF423:p.H1186H 0.7 80 Day –3 Day 3 Day 879 Day 1069 Day 1173 HCN1:p.Q828* 8.9 ND 60 Diagnostic (MRI) Day 0 Residual (MRI) Relapse (MRI) Follow-up (MRI) DNAH6:p.A428T 0.6 Samples ADORA2A:p.I200I 7.8 40 Samples 0.2 GCNT2:p.T133M 6.7 ND 20 TCF4:p.C357C 0.2 0 POLA1:p.H707R ND ND ZFAT:p.R762C 5.1 0.4 MXRA5:p.T350T 0.8 ND CSF Blood CSF Blood ADD1:p.P718L ND ND (Plasma) (Plasma) OBSCN:p.S194Pfs ND ND ACTA2:p.Y190S ND ND MYH6:p.A1298A ND ND 57 × 47 × 40 mm R <1.5 cm2 CCT5:p.K150M ND ND GRM6:p.Y610* ND ND ND ND f WES CNV MYH7B:p.R288C MBL2 (.) ND ND Tumour CSF CSF RELN:p.M2993I ND ND d.0 d.0 d.1173 NRXN2:p.A1555V ND ND HPD:p.S54Cfs ND ND MYCN:AMP GLI2:AMP n = 21 SUFU:DEL Fig. 5 CSF ctDNA reveals the evolving genomic landscape at relapse of patients with MB. Longitudinal monitoring of two paediatric patients with MB that relapsed and died from the disease: a–c MB8 and d–f MB5. a, d Clinical course indicating intervention for hydrocephalus, tumour surgery, treatment protocol, radiologic images (MRI), tumour size or residual disease, and patient samples obtained. Yellow arrows within the MRIs indicate the tumours at relapse. b, e Heatmap representing mutations and VAFs (%) identified from the WES of tumour DNA, CSF1 and CSF2 cfDNA samples. VAFs <10% were annotated. For patient MB8, a representative list of mutations (with a high/moderate impact or identified as drivers/predicted drivers by the Cancer Genome Interpreter) is shown. The first heatmap is for mutations detected in both tumour and CSF (n = 61) and the second one for private mutations (n = 53). Blue arrows indicate the mutations selected for further validation. c, f Relevant MB CNV identified were indicated in red for amplification (log2 > 1) or blue for deletion (log2 < −1). Definition of symbols and acronyms: not detected (ND), day (d), external ventricular drain (EVD), endoscopic third ventriculocisternostomy (ETV), residual (R) disease status: >1.5 cm2 (+), <1.5 cm2 (−), chemotherapy (CXT) and radiotherapy (RT). amplification, and partial loss of 17p (Fig. 5e, f and Fig. 3e). and were not detected at relapse for the HCN1:p.Q828* mutation Surprisingly, the analysis of CSF ctDNA at relapse revealed a detected in the primary tumour. In contrast, the levels of ctDNA genomic transformation, with a distinct molecular profile not for alterations identified from the CSF during progressive disease shared with the primary nor with the first relapsed tumour postrelapse were stable or increased throughout 81 days until the (Fig. 5e and Supplementary Fig. 2). Levels of ctDNA decreased patient died (ADD1:p.P718L VAF: 41.6-52% and ACTA2:p.Y190S NATURE COMMUNICATIONS | (2020)11:5376 | https://doi.org/10.1038/s41467-020-19175-0 | www.nature.com/naturecommunications 7 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-19175-0 VAF: 13.2-15%) (Supplementary Fig. 2). These results indicate the characterisation of the intratumour heterogeneity of MB the presence of two independent tumours, one at diagnosis and a facilitating the identification of small aggressive clones that could different one at relapse. Analysis of germline variants detected by promote relapse or resistance to therapy. WES revealed that the patient carried a TP53:p.R282W mutation, During monitoring of the patients through CSF ctDNA, we indicating that he suffered from Li-Fraumeni syndrome. were able to identify minimal residual disease and assess tumour Importantly, our data highlights that the analysis of CSF ctDNA evolution. The cerebellum is involved in sensory-motor and can be used to monitor genomic tumour evolution. In this cognitive functions and MB survivors can suffer from treatment- particular case, CSF ctDNA was able to identify the extreme induced long-term sequelae as well as from the appearance of situation of two completely different tumours at diagnosis and secondary tumours36–38. These patients would benefit from the relapse. study of serial CSF ctDNA samples to assess tumour progression and response to treatment, in order to adjust the therapy intensity and duration to the specific status of the tumour at a specific time Discussion point. Moreover, CSF ctDNA can complement imaging techni- The diagnosis, the information about prognosis, as well as the ques for early detection of relapse, guiding the therapeutic identification of therapeutic targets in MB rely on the molecular strategy. Relapse is the major cause of death in MB. Biopsies are characterisation of the tumour. To date, in order to characterise rarely performed at relapse and the absence of data about the MB, tumour specimens are required. This provides limitations tumour characteristics may challenge the selection of effective since brain tumour surgical procedures may sometimes not be second-line therapeutic strategies. In addition, studies comparing feasible and, even when they are feasible, tumour fragments may primary versus relapse tumours have shown that other mutations provide limited information about the molecular characteristics can arise in the recurrent tumour and that secondary tumours of the tumour including its intratumour heterogeneity. Moreover, can be completely different from the primary tumour11–14. In this longitudinal biopsies are seldom performed challenging the study study, we have shown that WES of CSF ctDNA can characterise of the tumour genomic evolution as well as the characterisation of the disease at relapse, including the detection of CSF-private secondary tumours in the relapse setting. In the present study, we mutations that reveal the intratumour/interlesion heterogeneity. show that the analysis of CSF ctDNA allows the molecular In one patient with a Li-Fraumeni syndrome, CSF ctDNA was characterisation of MB and facilitates the clinical management of able to identify the emergence of a completely independent these paediatric patients. tumour at relapse. In the absence of the CSF ctDNA analysis and The genomic alterations identified in CSF ctDNA recapitulated since this patient was not suitable for biopsy of the relapsed those present in the tumour and hence the analysis of the CSF tumour, the change in the nature of the tumour would be ctDNA constitutes a less invasive alternative to discern the unknown and hence the patient may not have been accurately genomic characteristics of MB. Key molecular alterations managed. The analysis of CSF ctDNA at relapse will allow the including, but not limited to, TP53 and PTCH1 mutations; identification of changes in the presence of therapeutic targets MYCN and GLI2 amplifications; SUFU deletions and 17p loss during tumour progression in a way that the treatment strategy were detected in the CSF ctDNA and provided sufficient infor- could be adapted to the molecular characteristics of the tumour at mation for MB molecular diagnosis and risk stratification. Thus, each time point. CSF ctDNA allowed the classification of the tumours to the In this proof-of-concept study, we demonstrate that the ana- correct subtype and risk subgroup and provided crucial infor- lysis of CSF ctDNA allows the characterisation of MB and pro- mation about prognosis. vides an early and accurate molecular diagnosis, including Since most MB patients exhibit hydrocephalus26,27, they subtyping and risk stratification. It also allows patient monitoring, require CSF drainage prior to surgery. CSF is then available in a providing information about minimal residual disease, tumour noninvasive way and the CSF ctDNA can be analysed to provide evolution and the characterisation of the disease at relapse. information about the tumour even before surgery. From our Studies with larger cohorts are warranted in order to translate our cohort, all 13 patients were required to undergo CSF drainage findings into the clinical practice of paediatric patients with MB. days before surgery. The characterisation of ctDNA in the CSF This can ultimately lead to tailored therapeutic strategies, redu- can thus provide surgeons and oncologists with critical infor- cing unnecessary treatment that would lead to long-term sec- mation to evaluate the risk/benefit balances and guide the surgical ondary effects and selecting optimal, adjusted, individualised and chemo- and radiotherapy strategies. Knowing the patient therapies. CSF samples are routinely obtained as part of the diagnosis and prognosis prior to surgery could help identify those standard of care and with the implementation of a CSF-based test patients with good prognosis that would benefit from a con- in the clinical practice, valuable information could be obtained servative tumour resection and a mild chemo- and radiotherapy from each patient, bringing precision medicine closer to MB dosage. In addition, it would allow the identification of a subset of patients (Fig. 6). patients for which residual disease is a risk modifier and an indicator of shorter PFS, highlighting the importance of total Methods tumour resection4. Patient cohort, clinical data and samples. A total of 13 paediatric patients As it was shown in other CNS malignancies18,19,21,24, we diagnosed with medulloblastoma were included in the study. All patients were observed that CSF is a better source for ctDNA than plasma. diagnosed and treated at the Vall d’Hebron University Hospital (Barcelona, Spain). A written consent, indicating that the samples obtained as standard of care in the Interestingly, we observed that high levels of CSF ctDNA before clinical practice could be used for research, was obtained from the parents/legal surgical resection were detected in patients at very high risk of representant in agreement with the declaration of Helsinki. The study was relapse, and disease progression was observed for two of them. approved by the Vall d’Hebron University Hospital IRB. Future studies with larger cohorts will be required to investigate whether the abundance of CSF ctDNA could be a valuable Sample collection and DNA extraction. Tumour DNA was extracted from a prognosis indicator. fresh-frozen or paraffin-embedded section of the tumour biopsy using the QIAamp Importantly, CSF ctDNA recapitulated the subclonal genomic DNA mini kit (Qiagen) and the QIAamp DNA FFPE tissue kit (Qiagen), respec- tively. Germline DNA was extracted from peripheral blood cells using the QIAamp landscape of the tumour. A significant correlation was observed DNA mini blood kit (Qiagen). Peripheral blood was collected in K2EDTA- between the VAFs of the same gene mutations in the tumour and containing tubes (Vacutainer) and plasma was acquired following a 1600 × g the CSF. This indicates that the study of the CSF ctDNA allows centrifugation for 10 min. Both plasma and CSF samples were centrifuged at 8 NATURE COMMUNICATIONS | (2020)11:5376 | https://doi.org/10.1038/s41467-020-19175-0 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-19175-0 ARTICLE Timeline for precision medicine using liquid biopsies Clinical course of the disease CSF CSF CSF Tumour Patient Intervention for Surgical Patient Disease hospitalisation hydrocephalus resection monitoring progression CSF ctDNA CSF ctDNA CSF ctDNA Diagnosis Monitoring tumour progression Characterisation of Prognosis Response or resistance relapsed tumour Subgrouping to therapy Tumour evolution Risk stratification Early detection of relapse Therapeutic targets Minimal residual disease Therapeutic targets Fig. 6 Proposed strategy to improve the clinical management of paediatric patients with MB using liquid biopsies. At the time of patient hospitalisation, MRI allows the detection of hydrocephalus and the brain tumour. CSF drainage is required for patients with hydrocephalus and the CSF sample can be used for WES of ctDNA. The analysis of CSF ctDNA identifies mutations, CNVs, arm-level and chromosome gain/loss and the frequency of somatic mutations. The genomic characterisation of the cancer provides information about MB-subgroup and the risk stratification, that is critical for the diagnosis and prognosis. In addition, therapeutic targets can be identified from the analysis of CSF ctDNA. The information obtained from the CSF ctDNA can guide the surgical resection, identifying those patients for which residual disease is a risk modifier and confers worse prognosis. Analysis of follow-up CSF samples could complement the MRI to identify minimal residual disease or early detection of relapse. Improving the monitoring of tumour progression and identifying response or resistance to therapeutic targets. Importantly, if disease progression is identified, analysis of CSF ctDNA can characterise the relapsed tumour, providing information about the diagnosis, prognosis and potential therapeutic targets. This will be extremely critical for patients with disseminated disease so the therapeutic strategy can be adjusted. Moreover, the comparison of CSF ctDNA from primary with refractory disease would provide information about the emergence of secondary tumours. The study of tumour heterogeneity and cancer evolution would reveal mechanisms of resistance and aid in the discovery of therapeutic targets for MB. Altogether, the analysis of CSF ctDNA can aid in the clinical management of patients and provide information about relapse to advance research towards precision medicine. 3000 × g for 5 min and the supernatant were collected. cfDNA from plasma and Filtering was performed based on the following criteria. For SNVs, we CSF samples were extracted using the QIAamp Circulating Nucleid Acids kit focused on mutations affecting the protein coding sequence; a depth of coverage (Qiagen). Genomic and cfDNA were quantified using the Qubit fluorometer. ≥20 was required to compare the tumour with the ctDNA VAF; if VAF > 5% in any of the samples from the same case, it was considered as a mutation, and its presence in the other samples was studied. To determine whether a sample DNA sequencing and analysis of genomic alterations. Whole-exome DNA contained a mutation that was detected in another sample from the same case, a sequencing (WES) was performed using 2.5 μg of matching tumour and germline VAF greater or equal to 1% was established as threshold. For germline genomic DNA samples of the 12 patients. Library preparation and enrichment was mutations, we analysed coding regions with a mapping quality ≥50. To choose performed using SureSelect Human All Exon V5 (Agilent Technologies) and only heterozygous variants, we focused on 0.35 ≤ VAF ≥ 0.75, and mutations sequenced using Illumina paired-read sequencing platform HiSeq2500 (read with a VAF greater or equal to 0.001 in the control cohort of the gnomAD length: 2 × 100; coverage: >75× control, >100× tumour; 50.4 Mb). For one patient database were removed to ensure that polymorphism were not present in the (MB13), WES was performed using the SureSelect XT HS Human Focused Exome final set of mutations. For CNVs, only regions which included ≥10 probes were 17.7 Mb as explained below. considered. Regions with a log2ratio greater or equal to 0.5 were considered Given the limited amount of cfDNA available from the 13 patients, CSF-derived gains, and lower or equal to −0.5 were considered deletions. For a more cfDNA from four patients underwent DNA sequencing (6–200 ng). For one patient stringent analysis we used −1 ≤ LOG2RATIO ≥ 1 as a threshold to report (MB8), SureSelect Human All Exon V5 (Agilent Technologies) was used for the MB CNVs. CSF-derived cfDNA obtained at relapse (51.6 Mb). For the remaining 3 patients, longitudinal CSF samples were obtained before surgical intervention of the tumour biopsy and during progression or follow-up. Matching tumour and germline Classification of MB-subgroup. To identify mutations, CNVs and cytogenetic (200 ng), and follow-up CSF samples (6–200 ng) were analysed using the SureSelect features relevant to MB, we focused on the following alterations4,5. Mutations: XT HS Human Focused Exome 17.7 Mb (custom constitutional panel, Agilent BCOR, CTDNEP1, CTNNB1, DDX3X, GFI1, GFI1 act, GFI1B act, KBTBD4, KDM6A, technologies) for library enrichment and sequenced using Illumina platform KMT2C, KMT2D, PRDM6 act, PTCH1, SMARCA4, SMO, SUFU, TERT, TERT NextSeq with a read length of 150 bp. promoter, TP53, TP53 (GERMLINE, GL), ZMYM3. CNVs: GLI2 amp, MYC amp, To process the sequencing data, FASTQ files were aligned to the human MYCN amp, OTX2 amp, PTEN del, SUFU del. Cytogenetics: 1q+, 2+, 4−, 5+, 6−, reference genome (hg19/GRCh37 hs37d5) using BWA-MEM, and PCR duplicates 7+, 8−, 8+, 9p+, 9q−, 10−, 10q−, 11−, 13−, 13+, 14+, 14q−, 15−, 16−, 16q−, were removed. Alignment information is indicated in Supplementary Data 1. 17p−,17p+, 18+, 21−. i17q was excluded because it was not possible to calculate Somatic mutation calling was performed using Sidrón39 and annotated using the from the data. The number of events from each category were counted and when Variant Effect Predictor (VEP)40. Germline mutations were extracted using alterations belonged to distinct subgroups, the group with the most events was BCFTOOLS and filtered using common SNPs from dbSNP. Copy Number selected (Supplementary Table 1). Variants (CNVs) were detected using the program exome2cnv41. Normal samples For the identified germline mutations, those reported as pathogenic and were compared to each other to identify any region that could be altered only by associated with a hereditary cancer predisposing syndrome in ClinVar42 were the region’s own variability. A summary with the events was done to obtain the annotated (Supplementary Data 2). MB-subgroup information from the germline percentage of loss/gain of each chromosome. mutations was based on previous studies43. NATURE COMMUNICATIONS | (2020)11:5376 | https://doi.org/10.1038/s41467-020-19175-0 | www.nature.com/naturecommunications 9 ARTICLE NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-19175-0 The frequency of somatic mutations per Mb was calculated from the tumour 14. Phi, J. H. et al. Genomic analysis reveals secondary glioblastoma after and CSF somatic coding mutations (with VAF > 5%) divided by the WES panel size radiotherapy in a subset of recurrent medulloblastomas. Acta Neuropathol. (Supplementary Data 1, 3–7). The values obtained were compared with the 135, 939–953 (2018). frequency of somatic mutations observed for MB2,30. 15. Chang, C. H., Housepian, E. M. & Herbert, C. Jr. An operative staging system The proportion of arm-level chromosomal gain/loss was calculated for those and a megavoltage radiotherapeutic technic for cerebellar medulloblastomas. with log2 value >0.5 and <−0.5; and % chromosome arm loss >15%. For total Radiology 93, 1351–1359 (1969). chromosome gain/loss, both arms needed to be affected. To define a threshold, the 16. Siravegna, G. et al. How liquid biopsies can change clinical practice in proportion of chromosome gain/loss obtained from the 13 tumour samples were oncology. Ann. Oncol. 30, 1580–1590 (2019). plotted (n = 46) to determine the quartile distribution. Samples followed a normal 17. Diaz, L. A. Jr. & Bardelli, A. Liquid biopsies: genotyping circulating tumor distribution (KS normality test), with the following column stats: min: 18%; 25% DNA. J. Clin. Oncol. 32, 579–586 (2014). percentile: 29.5%; median 44.5%; 75% percentile: 65.5%, max: 100%. For the MB 18. De Mattos-Arruda, L. et al. Cerebrospinal fluid-derived circulating tumour classification subgroup, we included the arm-level or whole chromosome gain/loss DNA better represents the genomic alterations of brain tumours than plasma. within the 75% percentile [covered 65.5–100%, black) and above the median Nat. Commun. 6, 8839 (2015). [covered 44.5-65.5%, grey). 19. Wang, Y. et al. Detection of tumor-derived DNA in cerebrospinal fluid of patients with primary tumors of the brain and spinal cord. Proc. Natl Acad. ddPCR analysis. We used droplet-Digital PCR (ddPCR), a robust method with Sci. USA 112, 9704–9709 (2015). high sensitivity and precision, to determine the presence of ctDNA in the CSF. 20. Pentsova, E. I. et al. Evaluating cancer of the central nervous system through From the NGS analysis of paired tumour/normal DNA, mutations were selected next-generation sequencing of cerebrospinal fluid. J. Clin. Oncol. 34, based on the impact of the mutation, the VAFs and whether they were relevant to 2404–2415 (2016). MB pathogenesis or identified as driver mutations by the Cancer Genome Inter- 21. Seoane, J., De Mattos-Arruda, L., Le Rhun, E., Bardelli, A. & Weller, M. preter44. Custom Taqman SNP genotyping assays for ddPCR were designed for Cerebrospinal fluid cell-free tumour DNA as a liquid biopsy for primary brain each selected mutation and ddPCR was performed with 10 ng of tumour/normal tumours and central nervous system metastases. Ann. Oncol. 30, 211–218 genomic DNA, and 1–10 ng of CSF and plasma cfDNA, using the QX200 Droplet (2019). Digital PCR system (Bio-Rad) according to manufacturer’s protocols and the 22. Miller, A. M. et al. Tracking tumour evolution in glioma through liquid literature45. biopsies of cerebrospinal fluid. Nature 565, 654–658 (2019). 23. Mouliere, F. et al. Detection of cell-free DNA fragmentation and copy number Reporting summary. Further information on research design is available in the Nature alterations in cerebrospinal fluid from glioma patients. EMBO Mol. Med. Research Reporting Summary linked to this article. https://doi.org/10.15252/emmm.201809323 (2018). 24. Bobillo, S. et al. Cell free circulating tumor DNA in cerebrospinal fluid detects and monitors central nervous system involvement of B-cell lymphomas. Data availability Haematologica https://doi.org/10.3324/haematol.2019.241208 (2020). The DNA sequencing data have been deposited in the European Genome-phenome 25. Martinez-Ricarte, F. et al. Molecular diagnosis of diffuse gliomas through Archive (EGA) database under the accession code EGAS00001004651. The data sequencing of cell-free circulating tumor DNA from cerebrospinal fluid. Clin. underlying Figs. 1–5 and Supplementary Table 1 are provided as Supplementary Data 1–7. Cancer Res. 24, 2812–2819 (2018). All the other data supporting the findings of this study are available within the article and 26. Sainte-Rose, C. et al. Management of hydrocephalus in pediatric patients its supplementary information files and from the corresponding author upon reasonable with posterior fossa tumors: the role of endoscopic third ventriculostomy. request. A reporting summary for this article is available as a Supplementary J. Neurosurg. 95, 791–797 (2001). Information file. 27. Lee, M., Wisoff, J. H., Abbott, R., Freed, D. & Epstein, F. J. Management of hydrocephalus in children with medulloblastoma: prognostic factors for Received: 4 June 2020; Accepted: 30 September 2020; shunting. Pediatr. Neurosurg. 20, 240–247 (1994). 28. Huang, T. Y. et al. Detection of Histone H3 mutations in cerebrospinal fluid- derived tumor DNA from children with diffuse midline glioma. Acta Neuropathol. Commun. 5, 28 (2017). 29. Panditharatna, E. et al. Clinically relevant and minimally invasive tumor surveillance of pediatric diffuse midline gliomas using patient-derived liquid References biopsy. Clin. Cancer Res. 24, 5850–5859 (2018). 1. Taylor, M. D. et al. Molecular subgroups of medulloblastoma: the current 30. Grobner, S. N. et al. The landscape of genomic alterations across childhood consensus. Acta Neuropathol. 123, 465–472 (2012). cancers. Nature 555, 321–327 (2018). 2. Northcott, P. A. et al. The whole-genome landscape of medulloblastoma 31. Rudin, C. M. et al. Treatment of medulloblastoma with hedgehog pathway subtypes. Nature 547, 311–317 (2017). inhibitor GDC-0449. N. Engl. J. Med. 361, 1173–1178 (2009). 3. Cavalli, F. M. G. et al. Intertumoral heterogeneity within medulloblastoma 32. Li, Y., Song, Q., Day, B. W. & Phase, I. and phase II sonidegib and vismodegib subgroups. Cancer Cell 31, 737–754 (2017). e736. clinical trials for the treatment of paediatric and adult MB patients: a systemic 4. Schwalbe, E. C. et al. Novel molecular subgroups for clinical classification and review and meta-analysis. Acta Neuropathol. Commun. 7, 123 (2019). outcome prediction in childhood medulloblastoma: a cohort study. Lancet 33. Lou, E. et al. Complete and sustained response of adult medulloblastoma to Oncol. 18, 958–971 (2017). first-line sonic hedgehog inhibition with vismodegib. Cancer Biol. Ther. 17, 5. Hovestadt, V. et al. Medulloblastomics revisited: biological and clinical 1010–1016 (2016). insights from thousands of patients. Nat. Rev. Cancer 20, 42–56 (2020). 34. Kieran, M. W. et al. Phase I study of oral sonidegib (LDE225) in pediatric 6. Ramaswamy, V. et al. Risk stratification of childhood medulloblastoma in the brain and solid tumors and a phase II study in children and adults with molecular era: the current consensus. Acta Neuropathol. 131, 821–831 (2016). relapsed medulloblastoma. Neuro Oncol. 19, 1542–1552 (2017). 7. Morrissy, A. S. et al. Spatial heterogeneity in medulloblastoma. Nat. Genet. 49, 35. Kool, M. et al. Genome sequencing of SHH medulloblastoma predicts 780–788 (2017). genotype-related response to smoothened inhibition. Cancer Cell 25, 393–405 8. Koschmann, C., Bloom, K., Upadhyaya, S., Geyer, J. R. & Leary, S. E. Survival (2014). after relapse of medulloblastoma. J. Pediatr. Hematol. Oncol. 38, 269–273 36. Kopp, L. M., Gupta, P., Pelayo-Katsanis, L., Wittman, B. & Katsanis, E. Late (2016). effects in adult survivors of pediatric cancer: a guide for the primary care 9. Zapotocky, M. et al. Differential patterns of metastatic dissemination across physician. Am. J. Med. 125, 636–641 (2012). medulloblastoma subgroups. J. Neurosurg. Pediatr. 21, 145–152 (2018). 37. Ito, M. Cerebellar circuitry as a neuronal machine. Prog. Neurobiol. 78, 10. Ramaswamy, V. et al. Recurrence patterns across medulloblastoma subgroups: 272–303 (2006). an integrated clinical and molecular analysis. Lancet Oncol. 14, 1200–1207 38. Roussel, M. F. & Hatten, M. E. Cerebellum development and (2013). medulloblastoma. Curr. Top. Dev. Biol. 94, 235–282 (2011). 11. Hill, R. M. et al. Combined MYC and P53 defects emerge at medulloblastoma 39. Puente, X. S. et al. Non-coding recurrent mutations in chronic lymphocytic relapse and define rapidly progressive, therapeutically targetable disease. leukaemia. Nature 526, 519–524 (2015). Cancer Cell 27, 72–84 (2015). 40. McLaren, W. et al. The ensembl variant effect predictor. Genome Biol. 17, 122 12. Verma, V. et al. Metachronous medulloblastoma and glioblastoma: (2016). Implications for clinical and technical aspects of re-irradiation. Rep. Pr. Oncol. 41. Valdés-Mas, R., Bea, S., Puente, D. A., López-Otín, C. & Puente, X. S. Radiother. 21, 84–89 (2016). Estimation of copy number alterations from exome sequencing data. PLoS 13. Wang, Y. et al. Radiation-induced glioblastoma with rhabdoid characteristics ONE 7, e51422 (2012). following treatment for medulloblastoma: A case report and review of the 42. Landrum, M. J. et al. ClinVar: public archive of relationships among sequence literature. Mol. Clin. Oncol. 9, 415–418 (2018). variation and human phenotype. Nucleic Acids Res. 42, D980–D985 (2014). 10 NATURE COMMUNICATIONS | (2020)11:5376 | https://doi.org/10.1038/s41467-020-19175-0 | www.nature.com/naturecommunications NATURE COMMUNICATIONS | https://doi.org/10.1038/s41467-020-19175-0 ARTICLE 43. Waszak, S. M. et al. Spectrum and prevalence of genetic predisposition in Biomedicals, Northern Biologics, Roche/Glycart and Hoffmann la Roche. X.S.P. is a medulloblastoma: a retrospective genetic study and prospective validation in a cofounder and has ownership interest from DREAMgenics. The remaining authors clinical trial cohort. Lancet Oncol. 19, 785–798 (2018). declare no competing interests. 44. Tamborero, D. et al. Cancer Genome Interpreter annotates the biological and clinical relevance of tumor alterations. Genome Med. 10, 25 (2018). 45. Forshew, T. et al. Noninvasive identification and monitoring of cancer Additional information Supplementary information is available for this paper at https://doi.org/10.1038/s41467- mutations by targeted deep sequencing of plasma DNA. Sci. Transl. Med. 4, 020-19175-0. 136ra168 (2012). Correspondence and requests for materials should be addressed to J.S. Acknowledgements We would like to thank the patients at the Vall d’Hebron Hospital that were enrolled in the Peer review information Nature Communications thanks Moritz Gerstung and the study and their families. The study was undertaken with the support of the Fundación other, anonymous, reviewer(s) for their contribution to the peer review of this work. Peer Asociación Española contra el Cáncer (AECC), FERO (EDM), Ramón Areces Foundation, reviewer reports are available. Cellex Foundation, BBVA (CAIMI), the ISCIII, FIS (PI16/01278) and the Juan de la Cierva fellowship (L.E). X.S.P. is supported by Ministerio de Economía y Competitividad Reprints and permission information is available at http://www.nature.com/reprints (MINECO) SAF2013-45836-R and CIBERONC; A.D.N. is supported by the Department of Education of the Basque Government (grant number PRE_2017_1_0100). We thank Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in CERCA Programme/Generalitat de Catalunya for institutional support. published maps and institutional affiliations. Author contributions Open Access This article is licensed under a Creative Commons L.E. performed the experimental work, analysed and interpreted the experimental and Attribution 4.0 International License, which permits use, sharing, bioinformatics data, co-wrote the manuscript and designed the figures. A.A. and R.M. adaptation, distribution and reproduction in any medium or format, as long as you give contributed to collecting, processing and facilitating experimental work. A.L. and S.G. appropriate credit to the original author(s) and the source, provide a link to the Creative coordinated the clinical aspects of the study and provided clinical samples and information. A.D.N. and X.S.P. carried out the bioinformatics analysis and interpreted the data. G.C. and Commons license, and indicate if changes were made. The images or other third party I.L. contributed towards the WES. C.R.P. contributed to bioinformatics and statistical material in this article are included in the article’s Creative Commons license, unless analyses. E.M.S., S.R.C., E.V., F.M.R., M.A.P, J.S., R.H. and L.G. contributed to the clinical indicated otherwise in a credit line to the material. If material is not included in the aspects of the study and provided clinical samples. J.S. conceptualised and supervised the article’s Creative Commons license and your intended use is not permitted by statutory study, interpreted research data and co-wrote the manuscript. regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/ licenses/by/4.0/. Competing interests J.S. is a cofounder of Mosaic Biomedicals and has ownership interests from Mosaic Biomedicals and Northern Biologics. J.S. received grant/research support from Mosaic © The Author(s) 2020 NATURE COMMUNICATIONS | (2020)11:5376 | https://doi.org/10.1038/s41467-020-19175-0 | www.nature.com/naturecommunications 11