ARTICLE

Received 25 Mar 2015 | Accepted 8 Oct 2015 | Published 10 Nov 2015 DOI: 10.1038/ncomms9839 OPEN Cerebrospinal fluid-derived circulating tumour DNA better represents the genomic alterations of brain tumours than plasma

Leticia De Mattos-Arruda1,2,3, Regina Mayor1, Charlotte K.Y. Ng2, Britta Weigelt2, Francisco Martı´nez-Ricarte3,4, Davis Torrejon1, Mafalda Oliveira1, Alexandra Arias1, Carolina Raventos1, Jiabin Tang5, Elena Guerini-Rocco2, Elena Martı´nez-Sa´ez4, Sergio Lois4, Oscar Marı´n4, Xavier de la Cruz4,6, Salvatore Piscuoglio2, Russel Towers7, Ana Vivancos1, Vicente Peg4, Santiago Ramon y Cajal3,4, Joan Carles1, Jordi Rodon1, Marı´a Gonza´lez-Cao8, Josep Tabernero1,3, Enriqueta Felip1,3, Joan Sahuquillo3,4, Michael F. Berger5,9, Javier Cortes1,3, Jorge S. Reis-Filho2 & Joan Seoane1,3,6

Cell-free circulating tumour DNA (ctDNA) in plasma has been shown to be informative of the genomic alterations present in tumours and has been used to monitor tumour progression and response to treatments. However, patients with brain tumours do not present with or present with low amounts of ctDNA in plasma precluding the genomic characterization of brain cancer through plasma ctDNA. Here we show that ctDNA derived from central nervous system tumours is more abundantly present in the cerebrospinal fluid (CSF) than in plasma. Massively parallel sequencing of CSF ctDNA more comprehensively characterizes the genomic alterations of brain tumours than plasma, allowing the identification of actionable brain tumour somatic mutations. We show that CSF ctDNA levels longitudinally fluctuate in time and follow the changes in brain tumour burden providing biomarkers to monitor brain malignancies. Moreover, CSF ctDNA is shown to facilitate and complement the diagnosis of leptomeningeal carcinomatosis.

1 Vall d’Hebron Institute of Oncology, Vall d’Hebron University Hospital, P. Vall d’Hebron 119-129, 08035 Barcelona, Spain. 2 Department of Pathology, Memorial Sloan Kettering Cancer Center, 1275 York Avenue, New York, New York 10065, USA. 3 Universitat Auto`noma de Barcelona, 08193 Barcelona, Spain. 4 Vall d’Hebron Institute of Research, Vall d’Hebron University Hospital, Ps Vall d’Hebron 119-129, 08035 Barcelona, Spain. 5 Center for Molecular Oncology, Memorial Sloan Kettering Cancer Center, 1275 York Avenue, New York, NY 10065, USA. 6 Institucio´ Catalana de Recerca i Estudis Avanc¸ats (ICREA), Barcelona, Spain. 7 Department of Surgery, Memorial Sloan Kettering Cancer Center, 1275 York Avenue, New York, NY 10065, USA. 8 Quiro´n Dexeus University Hospital, 08028 Barcelona, Spain. 9 Human Oncology and Pathogenesis Program, Memorial Sloan Kettering Cancer Center, 1275 York Avenue, New York, NY 10065, USA. Correspondence and requests for materials should be addressed to J.S. (email: [email protected]).

NATURE COMMUNICATIONS | 6:8839 | DOI: 10.1038/ncomms9839 | www.nature.com/naturecommunications 1 & 2015 Macmillan Publishers Limited. All rights reserved. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9839

he genomic characterization of tumours is crucial for the In the case of samples from the autopsy material of patients optimal diagnosis and treatment of cancer. Given the BMBC2, BMBC3, BMBC4 and BMBC6, we had enough number Treported spatial and temporal intratumour heterogeneity, of specimens to infer phylogenetic trees representing the genomic repeated biopsies are required for an adequate characterization of subclonal diversity and be able to identify trunk ubiquitous the somatic genetic alterations found in human cancers1,2. This genetic mutations. Interestingly, trunk mutations were always approach has important limitations, particularly in the case of identified in the CSF ctDNA (Fig. 1b). brain malignancies3, due to the restricted and invasive access for In addition, we sequenced the DNA concomitantly extracted sampling tumour material and the challenges to recapitulate the from the CSF and plasma in an expansion cohort of 11 patients tumour clonal diversity through the analysis of a small fragment (2 Medullos, 5 BMLCs, 4 BMBCs) with CNS restricted disease of the tumour. Recent work has shown that cell-free circulating and barely any visceral tumour burden to facilitate the tumour DNA (ctDNA) in the plasma could be used to comparison of the contribution of the brain tumour DNA into characterize and monitor tumours4–7. ctDNA analysis of the CSF or plasma ctDNA. In all cases, CSF ctDNA was detected patients with brain tumours, however, has revealed either and harboured gene mutations that were either absent or detected absence or very low levels of tumour DNA in plasma7. with lower MAFs in plasma ctDNA (Supplementary Fig. 1). The cerebrospinal fluid (CSF) is in intimate contact with tumour cells in central nervous system (CNS) cancers and, ctDNA from CSF performs better than plasma. We next sought recently, ctDNA has been shown to be present in the CSF of 8,9 to determine whether CSF ctDNA would be more representative patients with brain tumours . The aim of our work was to of the brain lesions than plasma ctDNA. To this end we divided determine whether the analysis of CSF ctDNA could be useful for the patients into two groups depending on the amount of the characterization and monitoring of brain tumours in extracranial tumour burden (Supplementary Table 4). comparison with plasma ctDNA. We applied hybridization Importantly, in patients with a CNS restricted disease (Fig. 1a, capture-based massively parallel targeted sequencing and/or Supplementary Fig. 1), the MAFs in all samples of CSF ctDNA exome sequencing coupled with droplet digital PCR (ddPCR) to were significantly higher than in plasma (Supplementary Fig. 3) synchronous CSF and plasma-derived ctDNA, and tumour and, moreover, the sensitivity for somatic mutations of the CNS tissue deposits from patients with glioblastoma (GBM), was also significantly higher in CSF ctDNA than plasma ctDNA medulloblastoma (Medullo), and brain metastases from lung (Fig. 2, Supplementary Table 5). Some mutations were detected in cancer (BMLC) and from breast cancer (BMBC, six of them the CSF or plasma but not in the brain tumour specimen (Fig. 1). subjected to warm autopsies) including breast cancer patients These could be potential false positives or mutations not present with clinical features suggestive of leptomeningeal carcinomatosis in the sequenced tumour fragment but present in another region (LC). In this study, we show that ctDNA derived from central of the brain tumour. In patients with abundant visceral disease nervous system tumours is more abundantly present in the CSF (Fig. 1b), the MAFs of the gene mutations in the CSF and plasma than in plasma. CSF ctDNA can be used to detect brain tumour ctDNA were comparable (Supplementary Fig. 3). private mutations and to longitudinally monitor the changes in brain tumour burden. In addition, we provided evidence that the analysis of CSF ctDNA may complement the diagnosis of LC. CSF ctDNA recapitulates the private mutations from CNS lesions. We have recently observed that, in the context of disseminated disease, brain metastasis might exhibit private gene mutations Results 11 CSF ctDNA is representative of brain tumours. To study and different from the ones present in the rest of the tumour lesions . compare the ctDNA present in the CSF with plasma ctDNA, we We next investigated how CSF and plasma ctDNA might sequenced DNA obtained from tumour samples, germline DNA recapitulate the private mutations from CNS lesions in (peripheral blood lymphocytes), plasma and CSF of a cohort of 12 metastatic patients. To answer this question, we analysed the patients (4 GBM, 6 BMBCs, 2 BMLCs; Supplementary Table 1). warm autopsy materials of a patient with Li Fraumeni syndrome In all cases, except BMBCs, CSF was obtained at the same time and a diagnosis of both HER2-positive metastatic breast cancer than plasma through lumbar puncture or cerebral shunts nor- and esthesioneuroblastoma (BMBC3). Two sets of tumours were mally obtaining 1–2 ml of CSF. Tumours and fluids from all six present: the breast cancer-derived brain metastasis and, cases of BMBCs were obtained through warm autopsy and the independently, the meningeal implants and liver metastases CSF was collected from the cisterna magna. We performed tar- (Supplementary Fig. 4). The gene mutations of the brain geted capture massively parallel sequencing and, in all cases, metastasis were not present in the extracranial tumours and, somatic single-nucleotide variants (SNVs), insertion/deletions moreover, we identified three private gene mutations (PIK3CB (indels) and copy-number alterations (CNA) were identified in M819L, PIK3CB Q818H, AHNAK2 L5292V) exclusively present CSF ctDNA and plasma ctDNA, and validated in the brain in the meningeal lesion. The gene mutations with the highest tumour tissue from the respective patients (Fig 1a,b, MAFs of the brain metastasis and the private mutations in the Supplementary Figs 1 and 2, Supplementary Tables 2, 3, meningeal lesions were present in the CSF ctDNA and not in the Supplementary Data 1, 2, 3). The number of genomic alterations plasma ctDNA (Fig. 1b, see boxed mutations) indicating that identified through targeted capture sequencing varied from case brain private mutations are more represented in the ctDNA from to case being more abundant in BMBCs and less abundant in CSF than plasma. GBM cases due to the nature of the genes selected for targeted sequencing. A low rate of mutation capture was observed in the CSF ctDNA is longitudinally modulated throughout treatments. CSF ctDNA from GBM patients indicating that further work is To address whether the amount of ctDNA present in the CSF required in order to optimize the detection of ctDNA in GBM could fluctuate with time and be representative of the brain cases. CSF ctDNA was identified in all cases while plasma ctDNA tumour progression, we obtained concomitantly CSF and plasma was only detected in patients with abundant visceral disease. This from six patients (GBM and metastatic breast and lung cancer is in agreement with previous reports4. Our methodology exhibits patients with brain metastasis) at sequential time points a detection limit of 2% mutant allelic frequency (MAF)10 and (Supplementary Table 1, Fig. 3). In all cases, there was a minimal patients with low tumour burden present evidence of plasma or absent extracranial disease. Brain lesions were identified ctDNA with MAFs below 2% (ref. 4). using magnetic resonance imaging and brain tumour burden

2 NATURE COMMUNICATIONS | 6:8839 | DOI: 10.1038/ncomms9839 | www.nature.com/naturecommunications & 2015 Macmillan Publishers Limited. All rights reserved. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9839 ARTICLE

a b AHNAK2 (D1792G) BMBC2 BMBC4 GBM1 Bone implant MACF1 (H1935Y) PIK3CA (E979Q) MYOD1 (R121H) Gene Amino Brain CSF Plasma acid tumour ctDNA ctDNA NEB (L3015*) Pericardiumimplant NBEAL (E1863G) Liver

HRNR (Q1999L) Rightlung TP53 R273C MST1L (Q546H) ATRX R808* LAMA5 Normal tissue ZFHX3 (D115Y) IDH1 R132H (W2550L) Normal tissue Leftlung Meninges PPP2R1A (I326fs) PTEN (Y240*) BRCA2 MGAM (S540L) ESR1 (Y537N) (R2991H) GATA3 (G144V) Meninges MXRA5 (R2177H) MAP3K1(R54W) HER2 (R2235C) Peri-pancreaticLN GBM2 Left lung CDC73 (A418V) CARD11 (R1133H) Amino Brain CSF Plasma SUFU (G263S) TSC1 (Q844R) Gene acid tumour ctDNA ctDNA PIK3C2G (T622I) EPPK1 (D2378Y) EPHB1 M857I KMT2C (V1163I) TERT Promoter GRIN2A (K1189M) Para-tracheal PTEN D162V Liver KMT2D (S5498C) LN LAMA5 (S2832C) Amino Men. Men. Men. Men. Men. P. Liver Liver Liver B. CSF Plasma Amino Lung Lung P.-T. P.-P. CSF Plasma Gene Lung Gene Men. Liver GBM3 acid (1) (2) (3) (4) (5) imp. (1) (2) (3) imp. ctDNA ctDNA acid (1) (2) LN LN ctDNA ctDNA GATA3 G144V Amino Brain CSF Plasma PTEN Y240* Gene acid tumour ctDNA ctDNA ESR1 Y537N CDC73 A418V MXRA5 R2177H HERC2 R2235C TERT Promoter TENM1 L927F NBEAL2 E1863G LAMA5 W2550L PPP2R1A I326fs BRCA2 R2991H MGAM S540L GBM4 NEB L3015* CARD11 R1133H PIK3CA E970Q AKT3 Q425H TSC1 Q844R Amino Brain CSF Plasma PRKD1 R246T Gene AHNAK2 D1792G acid tumour ctDNA ctDNA PIK3C2G L778V GRIN2A K1189M MACF1 H1935Y TERT Promoter TENM1 R2602Q MYOD1 R121H PIK3CG V223I WDFY3 L313fs PTCH1 W337* ARID1B Q170H MST1L Q546H ASXL1 P1427fs EPPK1 D2378Y KMT2D S5498C ZFHX3 D115Y BMBC1 HRNR Q1999L Amino CSF Plasma SUFU G263S BMBC6 Gene BM PIK3C2G T622I acid ctDNA ctDNA BRCA1 (splicing) Brain SPEN E1305G Liver MAP3K1 T457fs LAMA1 G403R POLE E318K KMT2C V1163I Normal tissue TP53 R248Q BLM C685Y ARID5B E815K LAMA5 S2832C ESR1 (Y537S) ARID5B E587K DOCK11 L958P TBX3 (K137fs) HIST1H3B (D78N) HECW1 R797Q PLEC R1350H TP53 (P177fs) JAK2 (T557fs) MDN1 G1699E NEB E6109D MAP2K4 (E372X) U2AF1 (P176R) MAP3K1 R54W NCOR1 (R507X) ASXL2 (A1269I) FLT3 (T820S) TNFAIP3 (W113C) BMBC5 BMBC3 MET (N1081H) PIK3C2G (S231F) Amino BM BM BM CSF Plasma MAP3K1 (G607fs) Gastric Gene DIS3 (L237R) acid (1) (2) (3) ctDNA ctDNA KMT2D (R2966fs) ERBB4 (V851L) AHNAK2 (P2390R) TP53 splicing RB1 (I324fs) Liver FGFR2 D758H NOTCH4 Q421R Normal tissue Amino BM BM BM Liver Liver Liver Liver Liver Liver CSF Plasma MDN1 (C2694S) Genev Gastric PIK3CA K111T TP53 germline acid (1) (2) (3) (1) (2) (3) (4) (5) (6) ctDNA ctDNA INPP4B Q498H TBX3 K137fs TP53 P177fs ROS1 V1898G MAP2K4 E372X FAT1 V3009G PIK3CB (M819L) Meninges NCOR1 R507X INPP4A E805G PIK3CB (Q818H) AHNAK2 (L5292V) ASXL2 A1269I NTRK1 K703T FLT3 T820S IL7R L340R Normal tissue MET N1081H DNMT3A S82P TP53 germline Brain MAP3K1 G607fs NBN E510A MLL R2344T BRAF L232R ESR1 Y537S ROS1 (P1152Q) CHEK2 K255T Amino Liver Liver Liver CSF Plasma U2AF1 P176R Gene BM Men. LN PAK7 (S407*) HIST1H3B D78N PIK3CA C378W acid (1) (2) (3) ctDNA ctDNA MSH5 (Q292E) RB1 I324fs TNFAIP3 W113C LAMA5 (Q1100H) JAK2 T557fs KMT2D R2966fs CUBN (S1887C) AHNAK2 P2390R PIK3C2G S231F KMT2D (G1916S) BRCA1 Splicing BMLC3 PAK7 S407* XBP1(.) MSH5 Q292E DIS3 L237R Amino CSF Plasma FH (S11W) ERBB4 V851L Gene BM PIK3CB M819L acid ctDNA ctDNA FOXA1 (S384F) PIK3CB Q818H RAD54L (R320Q) TP53 Splicing AHNAK2 L5292V MLL2 Splicing ROS1 P1152Q CARD11 K911N MDN1 C2694S BMLC1 CDK12 K756R LAMA5 Q1100H MAF > 50% Trunk NOTCH2 C347F CUBN S1887C Amino CSF Plasma Gene BM MAF 20–50% MLL3 K4229M XBP1 . acid ctDNA ctDNA Internal branch NOTCH2 Splicing KMT2D G1916S EGFR L858R MAF 5–20% PTEN P248fs FH S11W FBXW7 S678* MAF 1–5% Terminal branch FOXA1 S384F TET2 V1213L GREM1 P36L No mutation RAD54L R320Q NFE2L2 E79K 1 mutation (SNV or indel)

Figure 1 | CSF ctDNA better captures the genomic alterations in patients with brain tumours than plasma ctDNA. (a,b) Analysis of CSF ctDNA, plasma ctDNA and primary brain tumour or metastatic lesions collected simultaneously. Heatmap of the non-silent genetic alterations from each of the twelve cases is shown and phylogenetic trees of the autopsied patients with brain metastasis from breast cancer (BMBC) are represented. Colour key for mutant allelic frequencies (MAFs) is shown. (a) Patients with restricted central nervous system (CNS) disease, glioblastoma (GBM), BMBC and brain metastasis from lung cancer (BMLC). (b) Patients with CNS and non-CNS disease. BM, brain metastasis; LN, lymph node; Men, meninges; P. Imp, pericardium implant; PT, para-tracheal; PP, peri-pancreatic.

Restricted CNS disease CNS and non-CNS disease Importantly, MAFs of CSF ctDNA decreased with surgical resection and/or responses to systemic therapy and increased CSF Plasma CSF Plasma ctDNA ctDNA ctDNA ctDNA with tumour progression (Fig. 3). The MAFs were modulated 100 100 P = 0.0006 over time and followed the same trend as the variation in brain tumour burden. These results indicated that CSF may be a useful 80 80 biomarker to monitor tumour progression and response to 60 58% 60 60.5% 55.5% treatment. 40 40 Sensitivity % Sensitivity % 20 0% 20 CSF ctDNA complements the diagnosis of LC. The identifica- tion of CSF ctDNA led us to the hypothesis that cell-free DNA in 0 0 7 Patients 5 Patients the CSF could be used as a diagnostic tool for LC. The diagnosis of LC relies on the detection of malignant cells in the CSF of Figure 2 | Sensitivity analysis of CSF ctDNA and plasma ctDNA. patients with clinical symptoms. Diagnosis of LC is not trivial and Sensitivity was inferred based on gene mutations detected in central its misdiagnosis has important clinical implications. To define nervous system (CNS) tumours, which were either identified in CSF or whether the analysis of CSF ctDNA can be employed to enhance plasma ctDNA (Supplementary Table 5). Data were pooled and the mean the sensitivity of the detection of LC by cytopathologic analysis of with standard deviation error bars is shown. A Mann–Whitney test was CSF, we performed standard of care cytopathologic analysis and used for the analysis and P value is shown. CSF ctDNA sequencing in the same samples obtained from three breast cancer patients with clinical signs and symptoms sugges- tive of LC. was quantified using computer aided planimetric analysis Importantly, there were discrepancies between the cytology (Supplementary Table 6). The tumour somatic genomic altera- and our CSF ctDNA analysis (Fig. 4). In BMBC2, although three tions, previously identified in the tumours by exome sequencing, cytopathologic analyses yielded negative results, we detected were determined in the CSF-derived DNA of the patients through ctDNA with MAFs ranging from 20 to 50% in the two CSF ddPCR (Fig. 3). As expected, the MAFs in all samples of CSF samples that were available (Fig. 4). Given that LC was confirmed ctDNA were higher than in plasma (Supplementary Table 7). at the autopsy of BMBC2, our results indicated that the CSF

NATURE COMMUNICATIONS | 6:8839 | DOI: 10.1038/ncomms9839 | www.nature.com/naturecommunications 3 & 2015 Macmillan Publishers Limited. All rights reserved. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9839

GBM1 GBM2 GBM3 ddPCR ddPCR ddPCR 4 100 60 CSF CSF CSF 3 80 40 2 60 40 20 1 20 EGFR (R108K) IDH1(R132H) EGFR (C620S) 0 TP53 (R114C) 0 0 FTH1 (I146T) 4 PTEN (D162V) OR51D1 (R135C) 60 ANK2 (K2337X) 100 Plasma 3 Plasma 80 Plasma 40 2 60 40 20 1 MAF (%) MAF (%) NA MAF (%) 20 0 0 0

) MAF (%) Brain tumour burden 100 Brain tumour burden )Brain tumour MAF (%) burden ) MAF (%) 3 60

10 3 3 80 8 40 60 6 40 4 20 20 2 Volume (cm

0 Volume (cm Volume (cm 0 0

PD PD 3 (Palliative care) 8 (RT/TMZ→TMZ) 5 (Palliative care) 12 (RT + TMZ + targeted therapy →TMZ) // * * BS * * * BS BS

BMBC1 BMLC1 BMLC2 ddPCR ddPCR ddPCR 100 CSF 0.3 CSF 30 CSF 80 0.2 20 60 40 POLE (E318K) 0.1 10 20 CD9 (W22L) ADAMTS12 (T982K) 0 ARID5B (E572K) 0 0 EGFR (L858R) AHRR (G353C) 100 PCDH1 (S190C) 0.3 30 80 Plasma Plasma Plasma 60 0.2 20 40 MAF (%) 0.1 10 MAF (%)

MAF (%)20 MAF (%) 0 0 0 8 )25 Brain tumour MAF (%) burden Brain tumour burden ) MAF (%) 3 20 3 6 15 4 10 5 2

Volume (cm 0 Volume (cm

CNS disease CNS disease free free PD PD PD PD PD // 33 (Systemic therapy) RT 5 (Palliative care) // Erlotinib + RT 11 (Gefitinib + targeted therapy) Follow up 7 (Follow up) ** BMS * * NSCLC_S * *

Autopsy BMS BMS Figure 3 | Dynamic changes in CSF ctDNA recapitulate the treatment courses of patients with brain tumours. Longitudinal monitoring of patients with GBM and brain metastases through CSF and plasma ctDNA and the analysis of brain tumour burden. Gene mutations were measured by ddPCR. Tumour volumes were calculated using computer aided planimetric analysis. Timelines reflect the most relevant clinical information for each patient. BS, brain surgery; BMS, brain metastasis surgery; CNS, central nervous system; NSCLC_S, non-small cell lung cancer surgery; PD, progressive disease; RT, radiotherapy; TMZ, temozolomide. Asterisk and arrow indicate time of magnetic resonance imaging and surgical procedure, respectively. Grey boxes indicate therapy or follow up, and their duration is provided in months. ctDNA analysis detected disease at a level not detectable by brain tumour genomic alterations than plasma and putative cytopathologic analysis. In BMBC1, one of the cytopathologic actionable gene mutations and CNA (that is, EGFR, PTEN, ESR1, analysis was discordant with the presence of CSF ctDNA, while in IDH1, ERBB2, FGFR2) can be identified. We observed that CSF BMBC4 the results of the cytopathologic analysis and the CSF ctDNA has a significantly higher sensitivity than plasma for CNS ctDNA were in agreement. In both cases, BMBC1 and BMBC4, genomic alterations and can be used to detect brain tumour LC was confirmed at the autopsy. In summary, our results build a private mutations and to monitor brain tumour progression. In proof-of-concept that opens the possibility to use CSF ctDNA to addition, we provided evidence that the analysis of CSF ctDNA complement the diagnosis of LC. Of note, in the case of patients may complement the diagnosis of LC. with brain metastasis and clinical signs suggestive of LC, the One of the hallmarks of GBM is the fact that all tumours analysis of CSF ctDNA can be misleading since it will be difficult relapse. Once diagnosed, the GBM tumour is surgically resected to discern whether the ctDNA in the CSF is originated from the and then the patient receives radio- and chemotherapy LC or the brain metastasis. Further studies will be needed to treatments. Even when the surgical resection is complete, the consolidate this methodology for LC diagnosis. tumour invariably relapses. Importantly, the relapsed tumour tends to evolve under treatment and present different genomic alterations than the primary tumour12. Surgical procedures Discussion (resection and biopsies) are seldom indicated in relapsed GBM In this study, we identified and characterized ctDNA in the CSF limiting its genomic characterization and precluding the of patients with brain lesions and compared it with plasma treatment of the relapsed GBM based on genomic information. ctDNA. We showed that CSF ctDNA is more representative of CSF ctDNA provides a minimally invasive method to assess the

4 NATURE COMMUNICATIONS | 6:8839 | DOI: 10.1038/ncomms9839 | www.nature.com/naturecommunications & 2015 Macmillan Publishers Limited. All rights reserved. NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9839 ARTICLE

BMBC1 the local institutional review board (IRB)/ethics committee of both hospitals. Cytopathology – ++ VHIO Warm Autopsy Program and the Prospective Translational Program were 80 approved by the IRB of Vall d’Hebron University Hospital (Barcelona, Spain). POLE (E318K) Informed consent was obtained from all patients. 60 ARID5B (E572K) PCDH1 (S190C) 40 DNA extraction. The diagnosis of each metastatic lesion was confirmed on review 6 MAF (%) of routine hematoxylin and eosin-stained slides . Ten 8-mm thick sections from 20 representative fresh frozen metastasis biopsies/resections were cut, stained with 0 nuclear fast red and microdissected with a needle under a stereomicroscope to 123 ensure 480% of tumour cell content, as previously described16. DNA from microdissected tumour samples was extracted using DNeasy Blood and Tissue BMBC2 Kit (Qiagen, USA), and germline DNA from peripheral blood lymphocytes Cytopathology – – – (‘buffy coat’) was extracted using the QIAamp DNA Mini Kit (Qiagen) according 80 to manufacturer’s instructions. CSF-derived and plasma-derived circulating cell-free DNA was extracted with the QIAamp Circulating Nucleic Acid Kit PTEN (Y240X) 60 (Qiagen, Valencia, CA, USA), as previously described6. DNA was quantified MRPS33 (P94T) using the Qubit Fluorometer (Invitrogen). 40 ESR1 (Y537N) NA MAF (%) 20 Targeted capture massively parallel sequencing. DNA samples from CNS 0 tumours (primary brain tumours or CNS metastases) of 23 cases, non-CNS 123 metastases, CSF and plasma samples as well as germline DNA were subjected to BMBC4 targeted capture massively parallel sequencing at the Memorial Sloan Kettering Cancer Center Integrated Genomics Operation (iGO), using the Integrated Cytopathology + Mutation Profiling of Actionable Cancer Targets (MSK-IMPACT) platform17 80 targeting all exons of 341 cancer genes harbouring actionable mutations. For four additional cases with breast cancer and brain metastases (BMBC1-4) were analysed 60 AKT3 (Q425H) with a customized breast cancer panel, targeting all exons of 254 genes recurrently mutated in breast cancer and/or related to DNA repair (Supplementary Data 1) 40 CDC73 (A418V) HERC2 (R2235C) was also performed. For these four cases, of the 595 genes captured, 107 genes were MAF (%) 20 common to both targeted capture platforms (that is, 488 unique genes), and were employed for validation. By applying the methods described above to each targeted 0 capture platform independently, the validation rate of somatic mutations (SNVs 1 and indels) affecting the exons of 107 genes present in both platforms was 496% (Supplementary Table 3). Figure 4 | Analysis of CSF ctDNA as a diagnostic tool for leptomeningeal Targeted sequencing was performed as previously described6,17,18. In brief, carcinomatosis in three metastatic breast cancer patients. The results of 20–450ng of DNA was used to prepare barcoded sequence libraries (New England serial clinical cytopathology analyses are shown in the upper part of the Biolabs, Kapa Biosystems), which were pooled at equimolar concentrations for hybridization exon capture (Nimblegen SeqCap). graph. In the lower part, mutant allelic frequencies (MAFs) measured by Paired-end 100-bp reads were generated on the Illumina HiSeq2000 (San Diego, ddPCR in the same CSF samples are depicted. NA, not available. CA), and reads were aligned to the reference human genome hg19 using the Burrows-Wheeler Aligner19. Local realignment, duplicate removal and base quality recalibration were performed using the Genome Analysis Toolkit20. Somatic SNVs were called using MuTect21, and small insertions and deletions (indels) were called genomic alterations of the relapsed tumour helping to select the using Strelka22, VarScan 2 (ref. 23) and SomaticIndelDetector17. All candidate optimal treatment dictated by the molecular characteristics of mutations were reviewed manually using the Integrative Genomics Viewer24. the brain cancer. Somatic mutations with allelic fractions of o1% and/or supported by o2 reads On the other hand, patients with brain metastasis exhibit a were disregarded. The mean sequence coverage of each target exon was subjected to a loess normalization to adjust for bias in nucleotide composition (G þ C) and dismal prognosis and are usually recalcitrant to treatments. It is compared with the diploid normal sample. Gene copy-number profiles were known that, most likely due to the special environment of the generated using circular binary segmentation17. brain, the genomic alterations of brain metastasis differ from the ones of the visceral malignancies and primary tumours11–15. The identification of the brain metastasis-specific genomic alterations Exome sequencing of tumour DNA and normal DNA. DNA (500ng) extracted from brain tumour and germline samples from GBM1, GBM2 and GBM3, BMBC1, through CSF ctDNA might facilitate the design of tailored BMLC1 and BMLC2 cases were subjected to exome sequencing. An average of treatments to target brain metastasis hopefully increasing the 100 million 100-bp paired-end reads were generated for each sample, equivalent clinical response of these deadly lesions. to an average depth of 260 Â (range of 190–315 Â ). Exome sequencing was performed using the Nextera Rapid Capture Exome kit (37 Mb; Illumina) on an In a context where the oncology field expects that therapeutic 25 approaches will be dictated and guided by the genomic features of Illumina HiSeq 2000 instrument using a validated protocol and according to the manufacturer’s recommendations (Macrogen). tumours, the presence of CSF ctDNA will be fundamental to the correct molecular diagnosis and treatment of brain tumours. Altogether, our results indicate that CSF ctDNA can be exploited ddPCR and quantification of circulating tumour-specific DNA. ddPCR as a ‘liquid biopsy’ of brain tumours opening a novel avenue of of plasma and CSF were performed using the QX200 Droplet Digital PCR 26 research in CNS circulating biomarkers with an important impact system (Bio-Rad) according to manufacturer’s protocols and the literature . TaqMan-based quantitative PCR assays were designed to specifically detect point in the future characterization, diagnosis, prognosis and clinical mutations and corresponding wild-type alleles as selected by exome sequencing of managing of brain cancer. primary brain tumours or brain metastases. Primer sequences are provided in Supplementary Table 8. Ten nanograms of genomic DNA extracted from tumour tissue and germline DNA from peripheral blood lymphocytes was used for digital Methods PCR analysis. In some cases, lower amounts of DNA (for example, 1–5 ng) were Patients. Breast cancer patients with brain metastasis were enrolled as part of the used, due to CSF and plasma DNA yield limitations. Vall d’Hebron Institute of Oncology (VHIO) Warm Autopsy Program. Patients with breast cancer and lung cancer with brain metastasis, and GBM and medulloblastoma were enrolled as part of VHIO Prospective Translational The phylogenetic tree generation. Phylogenetic trees were constructed using the Program, which studies plasma and CSF-derived biomarkers. Patients with lung maximum parsimony method. The trunks of the trees were rooted by a germline cancer with brain metastasis were enrolled as part a collaborative effort with DNA sequence that did not have any of the somatic mutations. Trunk, branch and Dexeus University Hospital (Barcelona, Spain) and the research was approved by sub-branches lengths are proportional to the number of mutations.

NATURE COMMUNICATIONS | 6:8839 | DOI: 10.1038/ncomms9839 | www.nature.com/naturecommunications 5 & 2015 Macmillan Publishers Limited. All rights reserved. ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms9839

Statistical analysis. A Mann–Whitney test was performed for statistical analysis. 21. Cibulskis, K. et al. Sensitive detection of somatic point mutations in Data in graphs are presented as means±s.d. impure and heterogeneous cancer samples. Nat. Biotechnol. 31, 213–219 (2013). 22. Saunders, C. T. et al. Strelka: accurate somatic small-variant calling References from sequenced tumor-normal sample pairs. Bioinformatics 28, 1811–1817 1. Gerlinger, M. et al. Intratumor heterogeneity and branched evolution revealed (2012). by multiregion sequencing. N. Engl. J. Med. 366, 883–892 (2012). 23. Koboldt, D. C. et al. VarScan 2: somatic mutation and copy number 2. Sottoriva, A. et al. Intratumor heterogeneity in human glioblastoma reflects alteration discovery in cancer by exome sequencing. Genome Res. 22, 568–576 cancer evolutionary dynamics. Proc. Natl Acad. Sci. USA 110, 4009–4014 (2012). (2013). 24. Robinson, J. T. et al. Integrative genomics viewer. Nat. Biotechnol. 29, 24–26 3. Omuro, A. & DeAngelis, L. M. Glioblastoma and other malignant gliomas: a (2011). clinical review. JAMA 310, 1842–1850 (2013). 25. Lamble, S. et al. Improved workflows for high throughput library preparation 4. Murtaza, M. et al. Non-invasive analysis of acquired resistance to cancer using the transposome-based Nextera system. BMC Biotechnol. 13, 104 (2013). therapy by sequencing of plasma DNA. Nature 497, 108–112 (2013). 26. Forshew, T. et al. Noninvasive identification and monitoring of cancer 5. Dawson, S. J. et al. Analysis of circulating tumor DNA to monitor metastatic mutations by targeted deep sequencing of plasma DNA. Sci. Transl. Med. 4, breast cancer. N. Engl. J. Med. 368, 1199–1209 (2013). 136ra68 (2012). 6. De Mattos-Arruda, L. et al. Capturing intra-tumor genetic heterogeneity by de novo mutation profiling of circulating cell-free tumor DNA: a proof-of- principle. Ann. Oncol. 25, 1729–1735 (2014). Acknowledgements 7. Bettegowda, C. et al. Detection of circulating tumor DNA in early- and We thank all the patients and their families that participated in this study. The authors late-stage human malignancies. Sci. Transl. Med. 6, 224ra24 (2014). acknowledge Fundacio´n Rafael del Pino for financial support (for L. De Mattos-Arruda) 8. Segal, M. B. Extracellular and cerebrospinal fluids. J. Inherit. Metab. Dis. 16, and the ERC grant (Glioma), Asociacio´n Espan˜ola contra el Ca´ncer (AECC), the Josef 617–638 (1993). Steiner Foundation, FIS PI13/02661 and the FERO and Cellex foundation. 9. Pan, W., Gu, W., Nagpal, S., Gephart, M. H. & Quake, S. R. Brain tumor mutations detected in cerebral spinal fluid. Clin. Chem. 61, 514–522 (2015). Author contributions 10. Cheng, D. T. et al. Memorial sloan kettering-integrated mutation profiling J.S. designed the study. L.D.M.-A., F.M.-R., D.T., M.O., J.R., J.Ca., M.G.-C., E.F., J.Ta., of actionable cancer targets (MSK-IMPACT): a hybridization capture-based E.F., J.Sa., J.C. contributed with patients. L.D.M.-A., R.M., C.K.Y.N., B.W., A.A., C.R., next-generation sequencing clinical assay for solid tumor molecular oncology. J.T., E.G.-R, E.M.-R., S.P., R.T., V.P., S.J.C. developed methods. L.D.M.-A., R.M., J. Mol. Diagn. 17, 251–264 (2015). C.K.Y.N. generated data. S.L., O.M., X.C., A.V., J.S.R.-F., J.Ta., M.F.B. contributed 11. Brastianos, P. K. et al. Genomic characterization of brain metastasis reveals sequencing data. L.D.M.-A., C.K.Y.N., B.W., M.F.B., J.S.R.-F., J.S. analysed sequencing branched evolution and potential therapeutic targets. Cancer Discov. 5, 1–14 data. L.D.M.-A., J.S.R.-F, J.C., J.S. interpreted the data. J.S wrote the paper with assistance (2015). of L.D.M.-A. and the other authors. All authors approved the final manuscript. 12. Johnson, B. E. et al. Mutational analysis reveals the origin and therapy-driven evolution of recurrent glioma. Science 343, 189–193 (2014). 13. Ding, L. et al. Genome remodelling in a basal-like breast cancer metastasis and Additional information xenograft. Nature 464, 999–1005 (2010). Accession codes: Whole-exome and targeted capture massively parallel sequencing have 14. Saunus, J. M. et al. Integrated genomic and transcriptomic analysis of human been deposited in the Sequence Read Archive (SRA) under accession code SRP049647. brain metastases identifies alterations of potential clinical significance. J. Pathol. 237, 363–378 (2015). Supplementary Information accompanies this paper at http://www.nature.com/ 15. Paik, P. K. et al. Next-generation sequencing of stage IV squamous cell lung naturecommunications cancers reveals an association of PI3K aberrations and evidence of clonal heterogeneity in patients with brain metastases. Cancer Discov. 5, 610–621 Competing financial interests: The authors declare no competing financial interests. (2015). Reprints and permission information is available online at http://npg.nature.com/ 16. Hernandez, L. et al. Genomic and mutational profiling of ductal carcinomas reprintsandpermissions/ in situ and matched adjacent invasive breast cancers reveals intra-tumour genetic heterogeneity and clonal selection. J. Pathol. 227, 42–52 (2012). How to cite this article: De Mattos-Arruda, L. et al. Cerebrospinal fluid-derived 17. Won, H. H., Scott, S. N., Brannon, A. R., Shah, R. H. & Berger, M. F. Detecting circulating tumour DNA better represents the genomic alterations of brain tumours than somatic genetic alterations in tumor specimens by exon capture and massively plasma. Nat. Commun. 6:8839 doi: 10.1038/ncomms9839 (2015). parallel sequencing. J. Vis. Exp. 80, e50710 (2013). 18. Natrajan, R. et al. Characterization of the genomic features and expressed fusion genes in micropapillary carcinomas of the breast. J. Pathol. 232, 553–565 This work is licensed under a Creative Commons Attribution 4.0 (2014). International License. The images or other third party material in this 19. Li, H. & Durbin, R. Fast and accurate short read alignment with article are included in the article’s Creative Commons license, unless indicated otherwise Burrows-Wheeler transform. Bioinformatics 25, 1754–1760 (2009). in the credit line; if the material is not included under the Creative Commons license, 20. DePristo, M. A. et al. A framework for variation discovery and genotyping users will need to obtain permission from the license holder to reproduce the material. using next-generation DNA sequencing data. Nat. Genet. 43, 491–498 (2011). To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/

6 NATURE COMMUNICATIONS | 6:8839 | DOI: 10.1038/ncomms9839 | www.nature.com/naturecommunications & 2015 Macmillan Publishers Limited. All rights reserved.