Neuro-Oncology Neuro-Oncology Advance Access...

Clinical implementation of integrated whole-genome copy numberand mutation profiling for glioblastoma

Shakti H. Ramkissoon†, Wenya Linda Bi†, Steven E. Schumacher, Lori A. Ramkissoon, Sam Haidar, David Knoff,Adrian Dubuc, Loreal Brown, Margot Burns, Jane B. Cryan, Malak Abedalthagafi, Yun Jee Kang, Nikolaus Schultz,David A. Reardon, Eudocia Q. Lee, Mikael L. Rinne, Andrew D. Norden, Lakshmi Nayak, Sandra Ruland,Lisa M. Doherty, Debra C. LaFrankie, Margaret Horvath, Ayal A. Aizer, Andrea Russo, Nils D. Arvold,Elizabeth B. Claus, Ossama Al-Mefty, Mark D. Johnson, Alexandra J. Golby, Ian F. Dunn, E. Antonio Chiocca,Lorenzo Trippa, Sandro Santagata, Rebecca D. Folkerth, Philip Kantoff, Barrett J. Rollins, Neal I. Lindeman,Patrick Y. Wen, Azra H. Ligon*, Rameen Beroukhim*, Brian M. Alexander*, and Keith L. Ligon*

Center for Molecular Oncologic Pathology, Dana-Farber Cancer Institute, Boston, Massachusetts (S.H.R., L.A.R., S.H., D.K., Y.J.K., K.L.L.);Department of Medical Oncology, Dana-Farber Cancer Institute, Boston, Massachusetts (S.E.S., L.B., M.B., D.A.R., E.Q.L., M.L.R., A.D.N.,L.N., S.R., L.M.D., D.C.L., P.K., B.J.R., P.Y.W., R.B., K.L.L.); Department of Neurosurgery, Brigham and Women’s Hospital, Boston,Massachusetts (W.L.B., E.B.C, O.A.-M., M.D.J., A.J.G., I.F.D., E.A.C.); Department of Biostatistics and Computational Biology, Dana-FarberCancer Institute, Boston, Massachusetts (L.T.); Department of Radiation Oncology, Brigham and Women’s Hospital, Boston,Massachusetts (M.H., A.A.A., N.D.A., B.M.A.); Department of Radiation Oncology, Dana-Farber Cancer Institute, Boston, Massachusetts(M.H., A.A.A., N.D.A., B.M.A.); Harvard Radiation Oncology Program, Boston, Massachusetts (A.R.); Kravis Center for Molecular Oncology& Department of Epidemiology and Biostatistics, Memorial Sloan- Kettering Cancer Center, New York, New York (N.S.); Broad Institute,Cambridge, Massachusetts (R.B.); Department of Pathology, Boston Children’s Hospital, Boston, Massachusetts (K.L.L.); Departmentof Pathology, Brigham and Women’s Hospital, Boston, Massachusetts (S.H.R., A.D., J.B.C., M.A., S.S., R.D.F., N.I.L., A.H.L., K.L.L.)

*Co-corresponding Authors: Keith L. Ligon ([email protected]); Brian M. Alexander ([email protected]); RameenBeroukhim ([email protected]); Azra H. Ligon ([email protected])†These authors contributed equally to this manuscript.

Background. Multidimensional genotyping of formalin-fixed paraffin-embedded (FFPE) samples has the potential to improvediagnostics and clinical trials for brain tumors, but prospective use in the clinical setting is not yet routine. We report our experiencewith implementing a multiplexed copy number and mutation-testing program in a diagnostic laboratory certified by the ClinicalLaboratory Improvement Amendments.

Methods. We collected and analyzed clinical testing results from whole-genome array comparative genomic hybridization (Onco-Copy) of 420 brain tumors, including 148 glioblastomas. Mass spectrometry–based mutation genotyping (OncoMap, 471 muta-tions) was performed on 86 glioblastomas.

Results. OncoCopy was successful in 99% of samples for which sufficient DNA was obtained (n¼ 415). All clinically relevant loci forglioblastomas were detected, including amplifications (EGFR, PDGFRA, MET) and deletions (EGFRvIII, PTEN, 1p/19q). Glioblastomapatients ≤40 years old had distinct profiles compared with patients .40 years. OncoMap testing reliably identified mutations inIDH1, TP53, and PTEN. Seventy-seven glioblastoma patients enrolled on trials, of whom 51% participated in targeted therapeutictrials where multiplex data informed eligibility or outcomes. Data integration identified patients with complete tumor suppressorinactivation, albeit rarely (5% of patients) due to lack of whole-gene coverage in OncoMap.

Conclusions. Combined use of multiplexed copy number and mutation detection from FFPE samples in the clinical setting canefficiently replace singleton tests for clinical diagnosis and prognosis in most settings. Our results support incorporation ofthese assays into clinical trials as integral biomarkers and their potential to impact interpretation of results. Limited tumorsuppressor variant capture by targeted genotyping highlights the need for whole-gene sequencing in glioblastoma.

Keywords: array CGH, clinical trials, genomics, genotyping, glioblastoma.

Received 23 September 2014; accepted 16 January 2015# The Author(s) 2015. Published by Oxford University Press on behalf of the Society for Neuro-Oncology. All rights reserved.For permissions, please e-mail: [email protected].

Neuro-OncologyNeuro-Oncology 2015; 0, 1–12, doi:10.1093/neuonc/nov015

1 of 12

Neuro-Oncology Advance Access published March 9, 2015

Molecular biomarkers are increasingly used for many aspects ofclinical care based on their potential diagnostic, prognostic, andpredictive capacities. Multiplexed genomic data offer the possi-bility to obtain results for known biomarkers while simultane-ously generating information that may be useful for futureexploratory analyses or for clinical trial eligibility screening.The Cancer Genome Atlas (TCGA) and others have identifiedseveral pathways with recurrent aberrations in glioblastoma(GBM); however, routine incorporation of genomic precisionmedicine tools into the clinical environment of laboratories cer-tified by the Clinical Laboratory Improvement Amendments(CLIA) still presents significant challenges. While targeted mul-tiplexed assays are increasingly being adopted, integration ofresults from multiple tests on patients’ tumors is not widely im-plemented in the clinical or clinical trial research setting.1 – 3

For patients diagnosed with malignant glioma, copy numberaberrations such as EGFR amplification currently support tumorclassification and are frequently identified by fluorescence insitu hybridization (FISH). While whole-genome somatic copynumber analysis platforms, including array comparative geno-mic hybridization (aCGH) and single nucleotide polymorphism(SNP) arrays, have become routine tools to characterize cancergenomes in the research setting,4 – 7 reliable somatic copy num-ber analysis of formalin-fixed paraffin-embedded (FFPE) tissueshas been slow to emerge in routine clinical practice, largely dueto relatively poor DNA integrity isolated from clinical samples.8,9

Recent advances, however, now enable reliable and sensitiveaCGH testing of FFPE samples.10 Targeted multiplexed somaticmutation testing has been more rapidly implemented usingseveral platforms.11,12

Here, we report our results and experience with FFPE-basedaCGH (OncoCopy) and mass spectrometry mutational geno-typing assays13,14 (OncoMap) in a CLIA-laboratory setting atthe Dana-Farber/Brigham and Women’s Cancer Center (DF/BWCC). We demonstrate that these tests reliably provide criticaldiagnostic and prognostic data for use in clinical management,including selection of targeted therapies and enrollment intobiomarker-based clinical trials.

Methods

Patient Selection

Analysis of data generated from tumor specimens and clinicalvariables was conducted following approval from the DF/BWCCinstitutional review board. Genotyping data from clinical Onco-Copy reports were obtained from the medical record, while rawaCGH data were obtained from the Brigham and Women’s Hos-pital (BWH) Cytogenetics Laboratory. Somatic mutation profilingwas performed with consent for the DF/BWCC PROFILE clinical re-search study approved by the DF/BWCC institutional reviewboard. Tests were performed within the Cytogenetics (OncoCopy)and Molecular Diagnostics (OncoMap) Divisions of the BWH Cen-ter for Advanced Molecular Diagnostics, a CLIA-certified laborato-ry environment. All samples underwent central histopathologicreview by at least 2 board-certified neuropathologists (S.H.R.,S.S., R.D.F. or K.L.L) using World Health Organization criteria.Patient trial participation was assessed by retrospective reviewof medical records. Trial participation was scored as positive forany therapeutic trial at any time point in a patients’ care.

Molecularly informed trials were defined as those that had mo-lecular enrollment criteria (immunohistochemistry or genomicassay based) where OncoCopy or OncoMap results could havetheoretically contributed to enrollment or other decisions in thetrial.

Clinical Array Comparative Genomic Hybridization

Minimum sample requirements were established based on re-search assay performance and validation studies performed aspart of migration to the CLIA lab. Pathologists were instructedto submit 10×5 mm FFPE sections from a block with a mini-mum tumor size of �0.5 cm2 in cross section, .50% tumor nu-clei. Coring or microdissection was performed on circled regionsof the tumor when necessary to achieve these requirements.Post-extraction, the minimum amount of DNA required to per-form hybridization was set at 1.3 mg DNA by Nanodrop.

A minimum of 1.3 mg DNA (10×5 mm FFPE sections) wasobtained as part of clinical care. Due to the impact of necrosison aCGH signal detection, the total necrosis was required to be≤30% for GBM samples being submitted for aCGH withoutmacrodissection. The value of ≤30% was selected as a partic-ularly strict lower level limit because it allows for pathologistinterobserver variability in scoring tumor necrosis. For GBMwith .30% total necrosis, macrodissection (including coring)of viable tumor tissue was recommended to achieve this.

Patient and reference DNA (Promega) were fragmentedusing previously described fragmentation simulation methodsand hybridized to Agilent SurePrint G3 Human 1 Million featurearrays.10 Clinical analysis was performed using Agilent Work-bench software, and log ratios were normalized using the cen-tralization algorithm, with a threshold score of 6.0 and bin sizeof 10. Somatic copy number analyses of 8 consecutive probeswith mean log2 ratios .0.18 (gains) and ,–0.30 (losses) werecalled using the ADM2 algorithm (Algorithm Design Manual,2nd ed). Clinical reports contained only aberrations that affect-ed a previously annotated list of genes/regions with diagnostic,prognostic, and/or therapeutic significance in brain tumors(Supplementary Table S1, termed OncoCopy). The time fromtissue submission to data reporting averaged 2 weeks. Rawwhole-genome data files were retained and used forexploratory research analyses (eg, Genomic Identification ofSignificant Targets in Cancer [GISTIC]).

OncoMap

DNA was isolated from five to ten 5-mm FFPE sections contain-ing at least 50% tumor nuclei as previously described.13

Somatic mutations in tumor DNA were detected using the mul-tiplexed Sequenom-based assay OncoMap in the DF/BWCCCLIA-certified laboratory. The assay OncoMap v4 detects muta-tions in 471 different loci from 41 cancer genes (SupplementaryTable S2). The average time from tissue submission to report ofdata was 6 weeks, which includes 1 week for tissue acquisition/DNA preparation, 3 weeks for genotyping analysis and valida-tion, and 1–2 weeks for pathologist review. Data visualizationwas performed using the Memorial Sloan-Kettering CancerCenter cBioPortal for Cancer Genomics software.15

Ramkissoon et al.: Integrating aCGH and genotyping for neuro-oncology

2 of 12 Neuro-Oncology

http://neuro-oncology.oxfordjournals.org/lookup/suppl/doi:10.1093/neuonc/nov015/-/DC1



Fluorescence In situ Hybridization

FISH was performed on 5-mm FFPE tissue sections using methodsdescribed previously16 with the following probes: MYCN SpectrumGreen/CEP2 Spectrum Orange (Abbott Molecular), MYC SpectrumOrange/CEP8 (D8Z2) (Abbott Molecular), PDGFRA SpectrumGreen (RP11-231C18)/Zytolight CEN4 Spectrum Orange, EGFR(RP11-815K24) Spectrum Orange/CEP7 (D7Z1) Spectrum Green(Abbott Molecular), and MET (RP11-95I20) Spectrum Orange/CEP7 (D7Z1) Spectrum Green (Abbott Molecular).

Exploratory Research Analysis of aCGH Data

Circular binary segmentation was used to segment copy num-ber data for research aCGH analysis using parameters(a¼ 0.01, undo.splits¼ none, minimum width¼ 5).17 Seg-mented data were analyzed with GISTIC 2.018 to determinestatistically significant recurrent somatic copy number aberra-tions (SCNAs), after filtering germline copy number variations,using the following parameters: minimum segment size¼ 10,lesion amplitude threshold of 0.2 for aCGH cohort; focal versusbroad SCNA events were defined with a cutoff of 0.5× chromo-some arm length, and gene confidence level¼ 0.99. Resultswere compared with well-established data from TCGA.19 Seg-mented data were visually presented using Integrative Geno-mics Viewer 2.3 in heatmap format.20

A 2-tailed Fisher’s exact test was applied to identify peaksthat were differentially observed based on the patient’s age.For hypothesis testing, the 148 GBM patients were sequentiallypartitioned into 2 groups based on arbitrarily selected age cut-offs (35, 40, or 45 y), and copy number profiles were compared.The most significant differences were detected when thepatient cohort was split into a ≤40 year group and a .40year group. Copy number alterations were called when associ-ated with log2 copy number changes .0.2. Peaks with a Benja-mini–Hochberg false discovery rate q-value of ,0.2 werehighlighted between the 2 groups. Wilcoxon rank-sum analysiswas performed to compare frequency of broad, focal, and allSCNA events between patients younger versus older than 40years for aCGH and TCGA cohorts (Supplementary Table S3).

To compare the incidence of focal regions of amplificationand deletion between the cohorts of TCGA and DF/BWCCaCGH, genes representative of each peak from the analysesby TCGA were chosen (Supplementary Table S4A and B). Thefrequencies of amplification or deletion above a threshold of0.1 for TCGA data and 0.2 for aCGH data were compared forthese representative genes using Fisher’s 2-tailed exact testwith Bonferroni correction for multiple hypotheses. Only sub-jects .40 years of age were included in this comparison to min-imize age variation between younger subjects across TCGA andDF/BWCC aCGH populations.

Results

Clinical Performance Characteristics of OncoCopy Appliedto FFPE Brain Tumors

From the period 2012–2013, OncoCopy (whole-genome aCGH)was clinically requested on 469 primary brain tumors for diag-nostic, prognostic, or treatment-related indications (Fig. 1A).

OncoCopy is a routine clinical test requested by the patient’soncologist, pathologist, or other physicians, performed through

Fig. 1. (A) Summary of brain tumor samples submitted for OncoCopyfrom 2012–2013. (B) Demographics of GBM patients in DF/BWCCcohort compared with TCGA. (C) GBM-associated focal genomicevents identified in DF/BWCC adult GBM by OncoCopy compared withTCGA. (D) Summary of GBM-associated oncogenes and tumorsuppressor genes detected by OncoMap compared with TCGA.


Neuro-Oncology 3 of 12



the BWH Division of Cytogenetics and paid for by the patient’sinsurance company. Examples of frequent indications for test-ing in gliomas were EGFR amplification (diagnostic, trial eligibil-ity for glioblastoma), 1p/19q codeletion (diagnostic, prognostic,treatment for oligodendrogliomas), and BRAF duplication (diag-nostic of pilocytic astrocytoma). Following DNA isolation andimplementation of quality control measures, 49 samples(10.4%) yielded insufficient DNA. Insufficient DNA yields werehighest in GBM (17.7%, 33/186) and lowest in meningiomas(0/122); however, 85% of GBM that yielded insufficient DNAwere consult cases from outside hospitals where access to ad-ditional FFPE material was challenging. The remaining sampleswere from biopsies with limited amounts of tumor material.Importantly, although 18% of GBM samples were insufficientfor OncoCopy, the extracted DNA was used for OncoMap and/or MGMT promoter methylation analysis.

We successfully obtained genome-wide SCNA profiles withderivative log ratio spread noise measures of ,0.30 in 99% ofcases (415/420). Repeat review of histology for 5 GBM sampleswith derivative log ratio spread .0.30 revealed .30% necrotictissue. In total, 415 brain tumors, including 148 GBM, were an-alyzed and included during this pilot period (2012–2013). Pa-tient demographics for 148 and 86 GBM cases are reported inthis study by OncoCopy and OncoMap, respectively (Fig. 1B).

OncoCopy Reliably Detects Common Clinically RelevantAberrations in GBM

Data generated from whole-genome copy number analysis byaCGH comprehensively detect all known genomic gains or loss-es clinically relevant to GBM diagnoses and prognoses in a sin-gle assay, but also detect numerous SCNAs of unknownsignificance which we elected not to report.21 To capitalize onthe comprehensive nature of this test while reducing the com-plexity for clinical use, we utilized manufacturer recommendedsoftware (Agilent Cytogenomics) to restrict clinical reporting to41 SCNAs (termed OncoCopy) with known diagnostic or prog-nostic value to brain tumors, including those that are relevantto clinical trial enrollment or targeted therapies22 (Supplemen-tary Table S1).

We examined the collective OncoCopy genotyping resultsderived from individual adult GBM patients for 14 commonand relevant GBM aberrations and compared incidences forthese reported by the 2008 GBM dataset of TCGA.4 The mostsignificant amplification encompassed EGFR (7p11.2) and waspresent in 35% (52/148) of samples. The oncogenic EGFRvIIIvariant (deletion of exons 2–7) was detected in 11.5% (6/52)of EGFR-amplified GBM, which was comparable to rates previous-ly reported using aCGH and SNP-based platforms but lower thanthat for RNA sequencing or Nanostring assays.23 – 26 The mostsignificant deletion encompassed CDKN2A/B (9p21.3) and wasdetected in 39% (58/148) of samples. Differences between ourclinical results on FFPE material versus frozen tissue used byTCGA were statistically significant for only CDKN2A homozygousloss, which occurred at a lower incidence in our dataset(P¼ .004; Fig. 1C).4 Taken together, our analysis demonstratesthat OncoCopy data generated from FFPE GBM samples for clin-ical purposes provide reliable, reproducible results by identifyingdiagnostic and clinically relevant GBM copy number aberrations.

OncoCopy Identifies Loci With Clonal HeterogeneityResulting From Genomically Distinct Tumor CellSubpopulations

In a subset of patient samples, our clinical analysis found focal(117.3 Kb to 5.3 Mb) low-level gains (log2 ratio range 0.25–2.0)in MYCN, PDGFRA, EGFR, MYC, or MET. To determine whether theserepresent a subpopulation of amplified cells or low-level gain inthe majority of cells, we performed FISH with gene-specificprobes on 8 representative cases in which OncoCopy showedlow-level gains involving MYCN (n¼ 2), MYC (n¼ 2), PDGFRA(n¼ 2), MET (n¼ 2), or EGFR (n¼ 1). In all cases, we detected ge-nomic amplifications in subpopulations of tumor cells rangingfrom 5% to 35% (Fig. 2A and B). One tumor, GBM08, demon-strated at least 2 distinct genomic subpopulations detected byaCGH, corresponding to PDGFRA or MET amplification in differenttumor cell subpopulations, as has been previously described27,28

(Fig. 2A, bottom panel).

Exploratory Research Analysis Using Clinically ObtainedFFPE aCGH Data

To collectively analyze the ability of clinically generated FFPEaCGH data to detect known and novel SCNAs, we performedGISTIC analysis on the dataset generated from 148 GBMcases to identify recurrent SCNAs and compared our resultswith TCGA GBM data.4,29 Whole chromosome 7 gain and chro-mosome 10 loss were the most readily visualized SCNAs acrossour GBM patients (Fig. 3A). GISTIC 2.0 analysis18 also identified12 recurrent focal amplifications and 27 recurrent focal dele-tions (Fig. 3B, Supplementary Fig. S1). Significantly amplifiedregions contained known relevant GBM oncogenes, includingEGFR (7p11.2), CDK6 (7q21.2), KRAS (12p21.2), and MYCN(2p24.3) (Fig. 3B, Supplementary Fig. S1). Recurrent deletionsincluded GBM-relevant tumor suppressor genes, such asCDKN2A/B (9p21.3), PTEN (10q23.31), and PARK2 (6q27).

Distinct Copy Number Profiles in GBM Based on Ageat Diagnosis

Adult GBM tumors are commonly thought to be associated witha copy number “signature,” typified by polysomy 7 with or with-out EGFR amplification, CDKN2A/B single copy or homozygousloss, and monosomy 10 that leads to single copy PTEN loss.Based on observations during routine clinical care, we hypothe-sized that genomic profiles of young GBM patients were differentfrom those of GBM patients in the later decades of life. The SCNApatterns among younger GBM patients looked strikingly differentfrom patterns observed in older GBM patients (Fig. 4A). To quan-tify these differences, we compared the number of SCNAs persample for patient cohorts dichotomized at ages 35, 40, and45. We found significant differences in the number of arm-levelSCNAs per GBM between patients ≤40 years (n¼ 26) comparedwith .40 years of age (n¼ 122) (7.4 and 5.3 events per sample,respectively; P¼ .04 after multiple hypotheses correction); how-ever, we did not see similar differences among focal SCNAs. Wealso observed fewer arm-level events among TCGA GBM in olderpatients relative to younger patients, but this was not statisti-cally significant (P¼ .3). In our cohort of 148 GBM cases, only8/26 GBM patients≤40 years had an IDH1/2 mutation compared







with 2/122 patients .40 years, suggesting that the copy num-ber profiles for the vast majority of younger GBM patients are re-lated to other mechanisms than those proposed for IDH mutantgliomas.

These genome-wide differences were also associated withdifferences in individual chromosomes (Fig. 4B). Specifically,

chromosome 7 gain and chromosome 10 loss (containingEGFR and PTEN, respectively) each occurred in 75% of GBM inpatients older than 40 compared with only 25% of patients inthe younger cohort. Chromosome 19 gain was also observedalmost exclusively among older patients; no patients youngerthan 40 demonstrated gains in chromosome 19q, and only

Fig. 2. FISH validates genomically distinct tumor subclones. Representative images of 5 tumors (GBM02, 04, 05, 07, and 08) assessed by FISH forMYCN, EGFR, MET, MYC, and PDGFRA confirm the presence of tumor subpopulations that were identified as low-level gains by aCGH. (B) FISH analysisof low-level gains identified by aCGH in 8 GBM samples with tumor subpopulations.



1 patient demonstrated gain of 19p. Younger patients weremore likely to harbor losses of chromosomes 3, 4q, 5p, 16p,19q, and 21q and gains of 1q, 4, 9q, 10p, 11q, and 12p.

Distinct sets of recurrent focal SCNAs were also enriched inyounger or older patients (Fig. 4C). Older patients were enrichedfor amplifications of EGFR and deletions of CDKN2A/B. Youngerpatients were enriched for deletion of a region on chromosome11 encompassing CDKN1C and CEND1.

Computational Analysis of Clinical OncoCopy DataCompares Favorably With TCGA Analysis

We next compared copy number profiles amongst the DF/BWCC aCGH cohort with the population in TCGA. Due to the sig-nificant differences seen between younger and older patients,we restricted this comparison to adults older than 40 years.Globally, overlay of GISTIC peaks representing significantlyamplified and deleted focal chromosome regions demonstrat-ed high levels of similarity between aCGH and TCGA data(Fig. 5A). We also compared frequencies of events at eachpeak region between the 2 datasets. Only 4 of 22 amplificationpeaks and 2 of 39 deletion peaks exhibited significant differenc-es; all of these were enriched in the dataset from TCGA (Fig. 5B,Supplementary Table S4). These included amplifications of

MDM4 (1q32.1), CDK4 (12q14.1), 2 regions without knownGBM oncogenes: 1p36.21 and 19p13.3, deletions of PTEN(10q23.31), and a region of 10p13 without a known GBMtumor suppressor gene.

OncoMap Somatic Mutation Profiling Reliably CapturesClinically Relevant Mutations in GBM-AssociatedOncogenes

OncoMap testing was prospectively performed as anenterprise-level clinical research program at the DF/BWCCwith the goal of providing tumor genotyping data to requestingclinicians and consenting patients for clinical trial decision mak-ing. OncoMap testing was paid for by the DF/BWCC and provid-ed to patients at no cost. We analyzed OncoMap data from 86GBM patients and found recurrent mutations in IDH1 (4.7%),PIK3CA (3.5%), PIK3R1 (3.5%), and BRAF (2.3%) at rates similarto those reported in TCGA for the specific mutations targeted bythe assay (Fig. 1D). Mutations in TP53, PTEN, and RB1, the mostcommon GBM-associated tumor suppressor genes queried inthe OncoMap assay, were detected in 5.8%, 3.5%, and 1.2%of tumors in our study cohort, whereas TCGA reported frequen-cies of 34%, 31.9%, and 9.9%, respectively, using Sanger-basedwhole-gene sequencing. Comparison of OncoMap probes,

Fig. 3. Copy number aberrations across 148 GBM cases by aCGH. (A) Heat map demonstrating amplifications (red) and deletions (blue) among theDF/BWCC GBM cohort. (B) Focal GISTIC peaks with significance of q-value ,0.1 and associated genes from 148 GBM cases.




designed to capture “hot spots” in tumor suppressor genes,with actual TCGA 2008 mutation data showed that the expect-ed theoretical detection frequencies of OncoMap applied to the

cohort from TCGA would be 21% (8/38) for TP53, 20% (6/30) forPTEN, and 0% (0/9) of the RB1 mutations, while 100% (9/9) ofPIK3R1 (oncogene) mutations would have been detected.

Fig. 4. Array CGH reveals distinct genomic profiles between GBM patients ≤40 years of age and those .40 years. (A) Heat map of amplifications(red) and deletions (blue) demonstrates enrichment for chromosome 7 amplifications, CDKN2A/B deletions on chromosome 9, and chromosome10 deletions in GBM patients .40 years of age. (B) Recurrent arm-level events in GBM patients ≤40 years of age compared with patients .40 years.(C) GISTIC 2.0 peaks in GBM patients ≤40 years of age compared with patients .40 years.



Integrative OncoCopy and OncoMap Reporting CanInform Trial Inclusion and Exclusion Criteria Based onPathway Status of Tumor Suppressors or Oncogenes

In our cohort, comprehensive testing for both OncoCopy andOncoMap was performed in 37 GBM patients and we examinedthis group for ability of combined testing to provide unique re-sults not achieved with either test alone (Fig. 6). Such integra-tive analysis identified one tumor with concurrent RB1mutation and RB1 single copy loss, and one tumor with PTENmutation and loss of the second PTEN allele. Such events are

consistent with complete pathway inactivation. As anexample of the potential utility of this integrative informationin the clinical trial setting, a patient with combined PTEN alter-ations was considered “PTEN inactivated” and was molecularlyeligible for the pan –phosphatidylinositol-3 kinase inhibitorBKM120 (Novartis), as part of an open phase II clinical trialat DF/BWCC for recurrent GBM (Ivy Early Phase Clinical TrialsConsortium, NCT01339052). However, the patient did not sub-sequently enroll due to failure to meet other clinical eligibilityrequirements.

Fig. 5. GISTIC 2.0 analysis and comparision of DF/BWCC aCGH data with TCGA GBM SNP array data. (A) Significance (x-axis) of DF/BWCC deletions(dark blue) and amplifications (lavender) transposed onto TCGA data (light blue and red, respectively) across the genome (y-axis). (B) SCNA peakswith significantly different incidence between adult (.40 y) DF/BWCC patients and TCGA cohort, as defined by P-value ,.05 (2-tailed Fisher’s exacttest with Bonferroni correction).



To further determine the theoretical value of multiplex test-ing in our patient cohort, we retrospectively determined thenumber of patients in our study who were involved in clinicaltherapeutic trials at any point in their care at our institution.Of 198 GBM patients, 77 (39%) were enrolled into 41 distinctclinical trials. Among those 77 cases, 39 (51%) participated ina clinical trial of a targeted agent where eligibility criteria or trialresults could be critically informed by data generated by Onco-Copy or OncoMap. As expected due to their longer survival, re-current GBM patients seen in a tertiary care setting were morelikely to be enrolled in a clinical trial over the course of their dis-ease than newly diagnosed GBM patients: 53% of recurrent(17/32) and 36% (60/166) of newly diagnosed GBM patients ul-timately participated in one or more clinical trials. Patients with

recurrent and newly diagnosed GBM were equal in level of en-rollment to trials with molecular eligibility criteria (50% of bothcohorts).

DiscussionImplementation of both aCGH and somatic mutation detectiontechnologies in CLIA-certified clinical laboratories at DF/BWCChas enabled us to report both genome-wide SCNAs and muta-tions from FFPE primary brain tumors. We found that thesedata streams provide efficient and complementary tumor gen-otyping data useful for diagnostics/prognostics while also offer-ing genetic profiles that can be used in real time for clinical trialselection and decision making.

Fig. 6. Integrative OncoPrint of OncoCopy and OncoMap data for 37 GBM patients. The circles indicate patients with both copy number alterationsand a mutation in PTEN or RB1.



Gliomas, including GBM, are a disease in which copy numberalterations appear to be predominant drivers of tumorigenesisand for which whole-genome copy number analysis providesparticularly relevant diagnostic, prognostic, and therapeutic in-formation.4,19,30 In our clinical experience, the most usefulprognostic information based on copy number was related tosimultaneous detection of 1p/19q codeletion and EGFR ampli-fication. These copy number aberrations are relevant to prog-nosis in tumors histologically diagnosed as mixed gliomas,and emerging data suggest in fact that they are likely patho-gnomonic diagnostic features of oligodendroglioma and glio-blastoma, soon to be adopted by the field.31,32 To date, mostgenome-wide SCNA studies, such as those in TCGA, rely solelyon frozen tumor tissue, which essentially excludes this type oftesting from the routine clinical diagnostic pipeline, particularlyfor referral patients, who rarely have frozen tissue available. Ourresults demonstrate that clinical FFPE aCGH can replace target-ed FISH studies that were traditionally performed as standardof care, while providing extensive additional genomic data. Thefailure to extract sufficient DNA in GBM was higher than expect-ed based on our research experience with the assay; however,this was primarily explained by logistical factors of tissue acqui-sition from outside sites and failures in the early stages ofthe implementation where pathologists were not sufficientlytrained in the importance of avoiding necrosis in submitted ma-terial; these shortcomings have been readily improved as expe-rience with the testing has been gained.

Molecular heterogeneity in the form of subclonal tumor cellpopulations is well described in GBM and is hypothesized tohave clinical importance as a treatment resistance mecha-nism.27,33 While there are no specific criteria that can be uni-formly applied to definitively identify subclonal populationsbased on copy number aberrations in any individual tumor,our results do support that specific loci can be reliably identifiedas highly likely to represent subclonal events based on priorknowledge of consistent patterns of aberration involving specif-ic loci. Our experience suggests that low-level, focal gains af-fecting specific oncogenes (MYCN, PDGFRA, EGFR, MET, andMYC) using criteria in this study of ,10 Mb and log2 ratios be-tween 0.25 and 2.0 are sufficiently indicative of intratumoralheterogeneity that clinical reports should formally note thispossibility. Furthermore, it seems reasonable to make the sug-gestion in the report that targeted FISH may need to be per-formed if formally required for clinical decision making(eg, entry onto clinical trial).

While sensitivity of aCGH is important to consider, our find-ings demonstrate that aCGH can detect evidence of amplifica-tions occurring in as few as 5%–10% of cells (Fig. 2). However,we note that our EGFRvIII detection rate is lower than somepreviously reported incidences, such as RNA sequencing andNanostring assays, but is similar to levels reported in aCGH orSNP assays, including the cohort from TCGA.26 This is likelydue to the intrinsic dynamic range limitations of aCGH, whichcan be exceeded in the setting of high copy number amplifica-tions seen in GBM. Based on single cell sequencing of GBM, theEGFRvIII mutation typically exists in subclones of tumor cells,which may be masked by the more dominant wild-type epider-mal growth factor receptor–amplified cell population.28 Thisrange problem is not unique to the aCGH platform and is alsoa limitation of SNP-based copy number assays and SCNAs

calculated by quantification of reads in next-generation se-quencing. In our experience, these latter methods are currentlyless sensitive than aCGH for detection of the deleted exons inEGFRvIII. While aCGH offers a viable and rapid-turnaroundscreen for EGFRvIII and other copy number alterations in theclinical setting, we suggest that in situations where EGFRvIIIstatus is required for clinical trial entry, EGFR-amplified GBMthat lacks EGFRvIII should be reflexively submitted for an EGFR-vIII dedicated assay such as immunohistochemistry, reversetranscriptase PCR, or Nanostring.

Our study also revealed data suggestive of biologic under-pinnings of the disease. Patients ≤40 years old had a muchlower incidence of “classic” GBM copy number aberrations in-volving EGFR, PTEN, and CDKN2A, and higher frequencies of sev-eral other events, including amplifications of AKT3 and CCND2.These findings suggest that GBM arising in younger patients dif-fers from that which develops later in life and may require alter-native treatment strategies.

While SCNAs represent key driver events in GBM, mutations inoncogenes and/or tumor suppressor genes represent a comple-mentary level of molecular disruption contributing to the cancerphenotype.4,19 Incorporating clinical single nucleotide variationanalysis for brain tumors is important because it identifies diag-nostic and clinically relevant events, including mutations in IDH1/IDH2 (adult low-grade gliomas), BRAF (gangliogliomas, pleomor-phic xanthoastrocytomas), or INI1 (atypical teratoid rhabdoidtumors).11,13,34 – 36 Our results show that mass spectrometry–based methods reliably capture oncogenic mutations andother targeted somatic events in a clinical environment; however,they also highlight the need in GBM for rapid adoption of wholegene/exome sequencing given the low incidence of “hot spots”within the most common tumor suppressor genes involved inGBM. Integration of results from these technologies should great-ly increase the completeness of assessing tumor suppressorgenes and improve interpretation of responses in clinical trials.

While TCGA required extremely large infrastructure and staffinvestments, current technologies and methods now allowsimilar integrative genomics on a scale and timeframe feasiblefor patients in an academic laboratory setting. Genomically de-fined clinical trials increasingly require costly screening of largenumbers of patients using singleton tests such as FISH. Suchapproaches are particularly problematic as clinical trials in-crease in number and complexity. Furthermore, multi-armgenomically stratified trials of targeted agents are currentlybeing designed for GBM and other cancers where routineassessment of multiple biomarkers is essential to the trial de-sign.37,38 Such trials would be efficiently enabled and their costsreduced by incorporation of multiplexed genotyping approach-es described here. Indeed, with a 1- to 2-week turnaround timefor OncoCopy, copy number data are reported within a time-frame compatible with the time it takes to obtain final patho-logic diagnoses and transfer patient care from neurosurgery tothe oncology service. Improvements in the sample-processingpipeline, including reducing DNA preparation and genotypingtime or dedicating a pathologist review of molecular data,can further accelerate the turnaround in the future. OncoMapresults are complementary and often reveal trial-specific muta-tions in oncogenes (eg, PIK3CA, PIK3C2B, BRAF), for whichpatients would be eligible. The 6-week turnaround for largepanels, which are run as enterprise-wide batched results, was



less problematic for patients with newly diagnosed diseasegiven that few trials currently were available. However, thislonger time to results did represent a significant challenge forpatients presenting at recurrence from outside hospitals, whogenerally need to make decisions within 4 weeks or less. Thissuggests that laboratories may need to consider having sepa-rate testing for high-throughput, high-dimension, non-time-sensitive assays and lower-throughput, low-dimension, morerapid testing for clinical trial incorporation. As the number ofgenomics-based clinical trials continues to expand, we expectan increase in the percentage of patients with newly diagnosedand recurrent GBM enrolling in molecularly stratified trials.

FundingThis work was supported by grants R01CA170592 (K.L.L., P.Y.W.),K08NS087118-01 (S.H.R.), and P50CA165962 (K.L.L., P.Y.W., A.H.L.);the Ivy Early Phase Clinical Trials Consortium (K.L.L., P.Y.W., M.B., L.B.);the Sontag Foundation (K.L.L., R.B.); and the King Abdulaziz City for Sci-ence and Technology (KACST), Saudi Arabia (M.A.A.)

AcknowledgmentsWe would like to thank Marian Slaney and Sebastian Valentin in BWHNeuropathology for their expert help in managing clinical tissuesamples; Heather Homer in the Center for Molecular OncologicPathology for her work with the FISH assays; and Mark Listewnik inBWH Cytogenetics for his work in generating and collecting aCGH data.

Conflict of interest statement. None declared.

References1. So D, Joly Y. Commercial opportunities and ethical pitfalls in

personalized medicine: a myriad of reasons to revisit the myriadgenetics saga. Curr Pharmacogenomics Person Med. 2013;11(2):98–109.

2. McNamara MG, Sahebjam S, Mason WP. Emerging biomarkers inglioblastoma. Cancers. 2013;5(3):1103–1119.

3. Corless CL. Medicine. Personalized cancer diagnostics. Science.2011;334(6060):1217–1218.

4. Network TCGA. Comprehensive genomic characterization defineshuman glioblastoma genes and core pathways. Nature. 2008;455(7216):1061–1068.

5. Beroukhim R, Mermel CH, Porter D, et al. The landscape of somaticcopy-number alteration across human cancers. Nature. 2010;463(7283):899–905.

6. The Cancer Genome Atlas Research Network. Integrated genomicanalyses of ovarian carcinoma. Nature. 2011;474(7353):609–615.

7. Pinkel D, Segraves R, Sudar D, et al. High resolution analysis of DNAcopy number variation using comparative genomic hybridizationto microarrays. Nat Genet. 1998;20(2):207–211.

8. Nowak NJ, Miecznikowski J, Moore SR, et al. Challenges in arraycomparative genomic hybridization for the analysis of cancersamples. Genet Med. 2007;9(9):585–595.

9. Idbaih A, Criniere E, Ligon KL, et al. Array-based genomics inglioma research. Brain Pathol. 2010;20(1):28–38.

10. Craig JM, Vena N, Ramkissoon S, et al. DNA fragmentationsimulation method (FSM) and fragment size matchingimprove aCGH performance of FFPE tissues. PloS One. 2012;7(6):e38881.

11. Dias-Santagata D, Lam Q, Vernovsky K, et al. BRAF V600Emutations are common in pleomorphic xanthoastrocytoma:diagnostic and therapeutic implications. PloS One. 2011;6(3):e17948.

12. Chi AS, Batchelor TT, Dias-Santagata D, et al. Prospective,high-throughput molecular profiling of human gliomas.J Neurooncol. 2012;110(1):89–98.

13. MacConaill LE, Campbell CD, Kehoe SM, et al. Profiling criticalcancer gene mutations in clinical tumor samples. PloS One.2009;4(11):e7887.

14. MacConaill LE, Garcia E, Shivdasani P, et al. Prospectiveenterprise-level molecular genotyping of a cohort of cancerpatients. J Mol Diagn. 2014;16(6):660–672.

15. Cerami E, Gao J, Dogrusoz U, et al. The cBio cancer genomicsportal: an open platform for exploring multidimensional cancergenomics data. Cancer Discov. 2012;2(5):401–404.

16. Firestein R, Bass AJ, Kim SY, et al. CDK8 is a colorectal canceroncogene that regulates beta-catenin activity. Nature. 2008;455(7212):547–551.

17. Olshen AB, Venkatraman ES, Lucito R, et al. Circular binarysegmentation for the analysis of array-based DNA copy numberdata. Biostatistics. 2004;5(4):557–572.

18. Mermel CH, Schumacher SE, Hill B, et al. GISTIC2.0 facilitatessensitive and confident localization of the targets of focalsomatic copy-number alteration in human cancers. GenomeBiol. 2011;12(4):R41.

19. Brennan CW, Verhaak RG, McKenna A, et al. The somatic genomiclandscape of glioblastoma. Cell. 2013;155(2):462–477.

20. Thorvaldsdottir H, Robinson JT, Mesirov JP. Integrative GenomicsViewer (IGV): high-performance genomics data visualization andexploration. Brief Bioinform. 2013;14(2):178–192.

21. Pinkel D, Albertson DG. Array comparative genomic hybridizationand its applications in cancer. Nat Genet. 2005;37:(Suppl):S11–S17.

22. Louis DN, Wiestler OD, Cavenee WK. WHO Classification of Tumoursof the Central Nervous System. Lyon: IARC; 2007.

23. Shinojima N, Tada K, Shiraishi S, et al. Prognostic value ofepidermal growth factor receptor in patients with glioblastomamultiforme. Cancer Res. 2003;63(20):6962–6970.

24. Bachoo RM, Maher EA, Ligon KL, et al. Epidermal growth factorreceptor and Ink4a/Arf: convergent mechanisms governingterminal differentiation and transformation along the neuralstem cell to astrocyte axis. Cancer Cell. 2002;1(3):269–277.

25. Zhu H, Acquaviva J, Ramachandran P, et al. Oncogenic EGFRsignaling cooperates with loss of tumor suppressor genefunctions in gliomagenesis. Proc Natl Acad Sci U S A. 2009;106(8):2712–2716.

26. Kastenhuber ER, Huse JT, Berman SH, et al. Quantitativeassessment of intragenic receptor tyrosine kinase deletions inprimary glioblastomas: their prevalence and molecularcorrelates. Acta Neuropathol. 2014;127(5):747–759.

27. Snuderl M, Fazlollahi L, Le LP, et al. Mosaic amplification of multiplereceptor tyrosine kinase genes in glioblastoma. Cancer Cell. 2011;20(6):810–817.



28. Francis JM, Zhang CZ, Maire CL, et al. EGFR variant heterogeneity inglioblastoma resolved through single-nucleus sequencing. CancerDiscov. 2014;4(8):956–971.

29. Beroukhim R, Getz G, Nghiemphu L, et al. Assessing the significanceof chromosomal aberrations in cancer: methodology andapplication to glioma. Proc Natl Acad Sci U S A. 2007;104(50):20007–20012.

30. Ramkissoon LA, Horowitz PM, Craig JM, et al. Genomic analysis ofdiffuse pediatric low-grade gliomas identifies recurrent oncogenictruncating rearrangements in the transcription factor MYBL1. ProcNatl Acad Sci U S A. 2013;110(20):8188–8193.

31. Reuss DE, Sahm F, Schrimpf D, et al. ATRX and IDH1-R132Himmunohistochemistry with subsequent copy number analysisand IDH sequencing as a basis for an “integrated” diagnosticapproach for adult astrocytoma, oligodendroglioma andglioblastoma. Acta Neuropathol. 2015;129(1):133–146.

32. Louis DN, Perry A, Burger P, et al. International Society OfNeuropathology--Haarlem consensus guidelines for nervous systemtumor classification and grading. Brain Pathol. 2014;24(5):429–435.

33. Jun HJ, Bronson RT, Charest A. Inhibition of EGFR induces a c-METdriven stem cell population in glioblastoma. Stem Cells. 2014;32(2):338–348.

34. Balss J, Meyer J, Mueller W, et al. Analysis of the IDH1 codon 132mutation in brain tumors. Acta Neuropathologica. 2008;116(6):597–602.

35. Kannan K, Inagaki A, Silber J, et al. Whole-exome sequencingidentifies ATRX mutation as a key molecular determinant inlower-grade glioma. Oncotarget. 2012;3(10):1194–1203.

36. Cryan JB, Haidar S, Ramkissoon LA, et al. Clinical multiplexedexome sequencing distinguishes adult oligodendroglial neoplasmsfrom astrocytic and mixed lineage gliomas. Oncotarget. 2014;5(18):8083–8092.

37. Alexander BM, Galanis E, Yung WK, et al. Brain Malignancy SteeringCommittee clinical trials planning workshop: report from theTargeted Therapies Working Group. Neuro Oncol. 2015;17(2):180–188.

38. Conley BA, Doroshow JH. Molecular analysis for therapy choice:NCI MATCH. Semin Oncol. 2014;41(3):297–299.



Neuro-Oncology Neuro-Oncology Advance Access...

Documents

Transcript of Neuro-Oncology Neuro-Oncology Advance Access...