Purpose: Neuroblastoma (NB) is a heterogeneous disease characterized by distinct clinical features and by the presence of typical copy-number alterations (CNAs). Given the strong association of these CNA profiles with prognosis, analysis of the CNA profile at diagnosis is mandatory. Therefore, we tested whether the analysis of circulating cell-free DNA (cfDNA) present in plasma samples of patients with NB could offer a valuable alternative to primary tumor DNA for CNA profiling.
Experimental Design: In 37 patients with NB, cfDNA analysis using shallow whole genome sequencing (sWGS) was compared with arrayCGH analysis of primary tumor tissue.
Results: Comparison of CNA profiles on cfDNA showed highly concordant patterns, particularly in high-stage patients. Numerical chromosome imbalances as well as large and focal structural aberrations including MYCN and LIN28B amplification and ATRX deletion could be readily detected with sWGS using a low input of cfDNA.
Conclusions: In conclusion, sWGS analysis on cfDNA offers a cost-effective, noninvasive, rapid, robust and sensitive alternative for tumor DNA copy-number profiling in most patients with NB. Clin Cancer Res; 23(20); 6305–14. ©2017 AACR.
Neuroblastoma (NB) is an aggressive childhood tumor with low mutational burden and characterized by typical copy-number alterations (CNAs) used to co-guide treatment decisions. Here, we show the feasibility of noninvasive detection of CNAs using shallow whole genome sequencing (sWGS) of cell-free DNA (cfDNA) from plasma from patients with NB with or without metastatic disease. Interestingly, using this approach, we also detected aberrations that were not observed in the tumor biopsy. Given the importance of establishing a CNA profile at diagnosis for therapeutic stratification and the fact that in many patients primary tumor material is often very limited or even unavailable, we propose to include this low-cost WGS approach in the genetic work-up of patients with NB at diagnosis. Moreover, the proposed method offers the perspective for noninvasive genomic testing at several time points over the treatment course without additional risk and burden for the patient.
Neuroblastoma (NB) is the most common extracranial solid tumor of childhood and is characterized by a remarkable biological and clinical heterogeneity ranging from spontaneous regression (stage 4s) to overt metastasis (1, 2). Recurrent somatic mutations are rare in NB, while DNA copy-number alterations (CNAs) are present in almost all cases (3–7). The distinct patterns of CNAs in favorable (whole chromosome imbalances only) versus highly aggressive NBs (segmental chromosomal gains and losses) have led to the mandatory inclusion of multi-locus assays for CNA profiling for the genetic work-up of new patients with NB at diagnosis (6, 7). To this purpose, either fluorescence in situ hybridization (FISH), multiplex ligation-dependent probe amplification (MLPA), arrayCGH, or SNParray was used (8, 9). However, all these methods depend on the availability of primary tumor tissue at diagnosis, which can be obtained only through an invasive procedure. Open surgical biopsies are preferred, but often diagnosis relies on tru-cut or fine-needle biopsies. However, these small biopsies only provide a genomic view of one particular small area of a primary tumor, while NB tumors are known to demonstrate intratumoral, spatial, and temporal heterogeneity (10–13).
During the last few years, it has become evident that circulating tumor cells or circulating free nucleic acids present in peripheral blood or plasma can be used as a valuable alternative to biopsy. These so-called liquid biopsies can be used for diagnosis, monitoring therapy response, or prediction of relapse and allow frequent disease monitoring without additional risk and burden for the patients (14–16). Especially targeted approaches on liquid biopsies have allowed the detection of specific mutations or methylation events, while whole genome studies so far are relatively scarce (17). In NB, liquid biopsies have already shown to be useful to detect MYCN amplification and ALK mutations using real-time quantitative PCR and digital droplet PCR (18, 19). Thus far, for NB only one single study compared primary NB tumor DNA and cfDNA using a classical copy-number profiling technique (i.e., arrayCGH) (20).
In this study, we report on the first successful application of shallow whole genome sequencing (sWGS) for the detection of CNAs in cfDNA of 37 patients with NB. CNA profiles generated from cfDNA isolated from as little as 200 μL plasma were compared with genomic profiles of the corresponding tumors determined by standard array technology.
Materials and Methods
The local ethical committee approved the study and written consent was obtained from all patients enrolled in this study or their representatives. The clinical characteristics of the patients are summarized in Table 1 and Supplementary Table S1. In total, plasma samples were obtained from 12 low-stage (including two stage 4s) and 25 patients with high-stage NB, including 13 MYCN-amplified cases.
|Case number .||INSS stage .||Age at diagnosis (months) .||Metastasis .||MYCN status .||TumorDNA genomic profile .||cfDNA genomic profile .|
|Case number .||INSS stage .||Age at diagnosis (months) .||Metastasis .||MYCN status .||TumorDNA genomic profile .||cfDNA genomic profile .|
MNA: MYCN amplification.
Blood sample collection, plasma preparation, and cfDNA extraction
Whole blood samples from all patients with NB were collected in EDTA vacutainer tubes (BD Biosciences) and kept at 4°C until further processing. Plasma was prepared within 24 hours after blood collection. Plasma was obtained after 1 × 10 min centrifugation at 1,600 rcf at 4°C, centrifuge acceleration, and deceleration was set to two, followed by a subsequent centrifugation step of 1 × 10 minutes at 16,000 rcf at 4°C. The plasma was stored at −80°C (21 samples) or was processed immediately for subsequent cfDNA extraction (16 samples). cfDNA was extracted using the QIAamp Circulating Nucleic Acid Kit (Qiagen). The Qiagen kit was used in conjunction with the QiaVac 24 plus vacuum manifold (Qiagen). Isolation of cfDNA was done starting from 1 to 3.5 mL of plasma, if less plasma was available, volumes were adjusted by adding of 1× PBS. For 21 of 37 samples the input of plasma was 200 μL. DNA extraction was performed according to the manufacturer's instructions. In case of <3.5 mL starting volume of plasma, all volumes of the ingredients in the kit were adjusted accordingly. DNA was eluted in 50 μL of AVE buffer (Qiagen). DNA concentration was measured using the Qubit high-sensitivity kit (Thermo Fisher Scientific). Size distribution of the cfDNAs was measured using the Fragment analyzer (Advanced Analytical) according to the manufacturer's instructions.
Shallow depth WGS of cfDNA
Shallow depth WGS (sWGS) of cfDNA was performed using an Ion Proton sequencer (Thermo Fisher Scientific), starting from 5 ng input of cfDNA using the protocol as previously described (21). The minimal number of reads per sample was set at 10 million (mean coverage of 0.4×). Four tumors (cases 5, 13, 15, and 16) were also analyzed using sWGS on the HiSeq 3000 (Illumina Inc.), starting from 200 ng input of tumor DNA according to the manufacturer's instructions. Using a bin size of 100 kb, the R-Bioconductor package QDNAseq (22) was applied to visualize the DNA copy-number profile and call genomic aberrations. Each genome profile (line view and chromosome view) was manually checked for aberrations. All profiles were visualized using the online tool Vivar (http://cmgg.be/vivar/; ref. 23).
ArrayCGH analysis of tumor tissue and DTC samples
For those patients with NB that lacked tumor biopsy material at diagnosis, disseminated NB cells (DTCs) present in the bone marrow were isolated using magnetic activated cell sorting as previously described (24).
DNA from primary tumors or DTCs was isolated using the DNeasy blood and tissue kit (Qiagen) according to the manufacturer's instructions. Tumor samples or DTCs were profiled on a custom-designed 180K array (Agilent Technologies) as previously described (25). Tumor DNA and control DNA (400 ng) were labeled using the random prime labeling (BioPrime ArrayCGH genomic labeling system, Thermo Fisher Scientific) with Cy3 and Cy5 dyes (Perkin Elmer), respectively. Processing was performed according to the manufacturer's instructions (Agilent Technologies). Slides were scanned with an Agilent scanner (G2505C, Agilent Technologies) and data extracted using the feature Extraction v10.1.1.1 software program (Agilent Technologies), profiles were visualized using the Vivar platform (http://cmgg.be/vivar/; ref. 23). ArrayCGH profiles were further processed with the “no waves” and circular binary segmentation (CBS) algorithm (26, 27).
Data mining and statistics
Data-mining and statistical analysis was performed in R statistical environment (R-version 3.2.1), including t-tests, Mann–Whitney tests, histograms (stats package), and boxplots (ggplot2 package). With the function normalmixEM (mixtools package), mixed model fitting on the histograms of the copy-number data was performed. From these fits, the mean and the standard deviation of the different distributions (for gains, normal copy-number, and deletions) was calculated, followed by z-score calculation, that is, (‘mean signal for normal copy number' – ‘mean signal for altered copy number')/(‘standard deviation of the signal for the alteration’).
With the samtools view function, the distribution of the read lengths was retrieved. After normalization of the read length distribution data, the histograms for the different samples were plotted and compared in R.
qPCR analysis of LIN28B
To validate the LIN28B amplification, qPCR was performed as described by De Preter and colleagues (28). For LIN28B, both primers were designed with Primer 3 version 0.4.0 (http://bioinfo.ut.ee/primer3-0.4.0) using the standard parameters of this software. The following primer sequences were used for LIN28B: forward primer 5′ AATACTGGGCATTACCTCCCAAC 3′ and reverse primer 5′ ATCAGTGTGGGCGGTACAAAA 3′.
Normalization was done using two reference genes, BCMA and SDC4. These genes are respectively located at 16p13 and 20q13, both chromosomal regions that rarely show genetic abnormalities in NB.
All reactions were performed in duplicate and NGP was used as a positive control as this cell line was shown to have a focal gain of LIN28B by arrayCGH.
Quality assessment of cfDNA from plasma
In this study, circulating cell-free DNA (cfDNA) was isolated from plasma of 37 patients with NB (Table 1). Plasma from patients with cancer showed significantly (P < 0.01) higher cfDNA yields ranging from 17 to 7,500 ng/mL plasma (median: 194 ng/mL) as compared with the cfDNA yields in the plasma of normal controls (#=8) and pregnant women (#=100; Fig. 1A). Interestingly, we could also show that sufficient cfDNA could be isolated for sWGS (i.e., 5 ng) starting from only 200 μL of plasma.
Setting the parameters for copy-number profile analysis from shallow WGS data
The most important parameter to be set for QDNAseq analysis is the bin size, which also determines the resolution for detection of CNAs (22). Therefore, we analyzed the copy-number profile of a case with three different bin sizes (i.e., 15 kb, 50 kb, and 100 kb) (Fig. 2). After model fitting on the histograms of the copy-number data, we calculated z-scores to determine how well deleted and gained regions are discriminated from regions without genomic aberrations. The z-score represents how many standard deviations the mean signal for aberrant regions deviates from the mean signal for nonaberrant regions. sWGS data analyzed at 100 kb bin size have z-scores that match with clear identification of gains and losses, comparable as in the arrayCGH copy-number data (z-scores around 3 for gains and 5 for losses).
Comparison of copy-number profiles obtained using sWGS on plasma cfDNA versus arrayCGH on primary tumor DNA
In total, cfDNA genomic profiles from plasma and corresponding primary tumor profiles from 37 patients with NB were established using sWGS and conventional oligonucleotide arrayCGH, respectively (Table 1). A detailed overview of all aberrations that were identified in the tumor tissue and in the corresponding cfDNA can be found in Supplementary Table S1 and Supplementary Fig. S2. For one case, tumor tissue was not available and DTCs were used instead. We confirmed a segmental aberrant profile in the cfDNA in 22 out of 26 cases presenting with segmental aberrations in the tumor DNA (Fig. 3A and B). Twenty-one of these 22 cases belong to the high-stage tumors, while one case was a low-stage tumor (case 12). In the cfDNA of the remaining four cases, we identified no copy-number changes. For two of these cases, it could be explained by the fact that peripheral blood was obtained 1 month after diagnosis and after therapy (both cases showed negative MIBG scans at that moment).
In addition, a typical profile with only whole chromosome imbalances was detected in three out of 10 cases presenting with only numerical aberrations in the tumor DNA (Fig. 3C). From the 10 cases presenting with only numerical aberrations in the tumor DNA, two cases belonged to the high-stage group, whereas eight cases were low-stage patients. One out of the three cases presenting with only numerical aberrations in the cfDNA belonged to the high-stage group. In the cfDNA of six of the other cases no copy-number changes could be identified, while in one case (case 32) we identified segmental aberrations at chromosome 11 (del(11)(p15.5p12),dup(11)(p12q14.1),del(11)(q14.1q25)) in the cfDNA in addition to the numerical aberrations that were also present in the tumor DNA. In one case (case 2), a flat profile for both the tumor and the cfDNA was observed.
Detection of amplifications and other focal aberrations in cfDNA
Given the importance of focal CNAs that target specific genes in relation to therapeutic stratification and/or tumor biology, we further zoomed in on focal copy-number amplifications/gains and deletions (aberrations <3 Mb). Amplifications, in particular those encompassing the MYCN locus, are known to infer a poor prognosis in patients with NB (31, 32). Accurate assessment of MYCN is thus of utmost importance in the genetic work-up of the primary tumor as it may directly impact treatment decisions. Importantly, in this study, all MYCN amplifications observed by arrayCGH in 13 tumor DNA samples were also detected in cfDNA. Furthermore, a chromosome 5p amplicon containing IRX1/IRX2 (5p15.33, ±2,430 kb) (case 30) as well as a focal gain of the CCDC148/PKP4 locus (2q24.1, ±240 kb; case 15) was detected both in the cfDNA and tumor tissue. Remarkably, in case 8, a LIN28B amplification was only present in the cfDNA while not detected in the tumor DNA, suggesting possible tumor heterogeneity (6q16.3-6q21, ±2.7 Mb).
Next, we investigated in more detail focal deletions targeting one or a small number of genes (<1 Mb). ATRX deletions are known to mark a specific subset of NB tumors with an alternative telomere lengthening (ALT) phenotype (5, 33, 34). In this series, a focal ATRX deletion (Xq21.1, ±800 kb) was detected in both the cfDNA and the tumor DNA of one patient (case 7). In addition, a focal AUTS2 deletion (7q11.22, ±820 kb) was detected in both DNA fractions of another case (case 14). In contrast, three other focal aberrations including a CDKN2A/B homozygous deletion (case 3) (9p21.3, ±190 kb), a PTPRD deletion (case 7) (9p24.1, ±320 kb), and a small 12q-deletion (USP15, MON2, PPM1H, and AVPR1A) (case 9) (12q14.1/12q14.2, ±850 kb) that were detected in the tumor DNA could not be detected in the cfDNA using 100 kb binning. Interestingly, the ∼190 kb CDKN2A/B focal homozygous deletion could be detected in the cfDNA sWGS data by using a smaller bin size (15 kb), while the other two focal deletions remained undetectable at this resolution.
Capturing spatial and temporal tumor heterogeneity with cfDNA sWGS
An interesting observation from our study was the finding that noninvasive cfDNA analysis in several cases provided additional information regarding the spatial and temporal tumor heterogeneity of the primary tumor (and possibly metastases) at diagnosis and relapse. Indeed, in 13 cases, additional genomic alterations were observed in the cfDNA compared with the tumor DNA. Remarkably, in some cases (like cases 7 and 13), these additional alterations were detected at lower copy-number level than the shared alterations (Fig. 4), suggesting that these alterations are only present in subclones. In contrast, in five tumors we detected aberrations in the tumor tissue that were not present in the cfDNA.
To rule out that these discordances can be attributed to the use of two different technologies (i.e., arrayCGH and sWGS), the copy-number profile of four tumors (cases 5, 13, 15, and 16) was examined with both technologies, and no differences were observed (Supplementary Fig. S1).
In addition to spatial heterogeneity, cfDNA also allows to noninvasively capture temporal heterogeneity. For one case (case 8), cfDNA was available both at diagnosis and at relapse (1 year after diagnosis; Fig. 5). The cfDNA at diagnosis showed some extra segmental as well as numerical aberrations in comparison with the tumor tissue, including an amplification of the LIN28B gene. At relapse, this amplicon was not present anymore in the cfDNA, but the profile at relapse contained additional segmental alterations such as del(4)(q31.3q35.2), dup(9)(q31.3q33.1), dup(10)(p15.3p11.21), and del(17)(q11.2q21.2). This observation was confirmed using independent qPCR based copy-number quantification (Supplementary Fig. S3).
We describe the feasibility to reliably detect copy-number aberrations using sWGS on cfDNA isolated from plasma of NB patients. While several studies already reported on the use of cfDNA for the detection of mutations (35, 36), copy-number analysis on liquid biopsies has been rarely attempted (17, 18, 37, 38). Shallow whole genome or low pass sequencing have been described for the detection of copy-number analysis in cfDNA in the context of prostate cancer, lung cancer, and ovarian cancer (39–42). In NB, however, this is the first report on the use of sWGS analysis on cfDNA and the comparison with arrayCGH profiles generated on the corresponding tumor tissue or disseminated tumor cells. Easier scoring and the apparently lack of wave patterns in the sWGS profiles of the cfDNA and four tumor samples is striking, although sWGS and arrayCGH profiles are not exactly comparable due to differences in resolution. We could successfully determine the copy-number profile on 25 out of 35 cfDNA NB samples isolated from blood taken at diagnosis. Although cfDNA profiling was successful in a minority of low-stage localized tumors (four out of nine), copy-number profiles could mainly be determined in the cfDNA of patients with high-stage, metastasizing tumors (18 out of 19), which is in line with the results of Chicard and colleagues (20). Most of the measures are very comparable between both studies; however, some quality measures are a little bit lower in our study, but this can be explained by the different sample-set composition, that is, in the Chicard study, less low-stage, nonmetastasized samples (stages 1, 2) (13/66 = 20%) were included compared with our study (10/37 = 27%). It is well appreciated that in patients with low-stage, nonmetastasizing tumors, lower amounts of (tumor derived) cfDNA are detected. This is supported by the lower cfDNA concentrations in blood of low-stage patients. This finding is in line with other studies (19, 20, 43). In case 6, sWGS of the cfDNA revealed a flat profile despite very high cfDNA concentration. This discordance can be explained by the fact that the peripheral blood stored in an EDTA tube was kept over weekend before plasma isolation, resulting in lysis of the white blood cells and leading to contamination of the cfDNA with constitutional DNA, as previously described (44). This and previous reports thus underline the importance of good cfDNA sample work-up protocols for successful downstream analyses.
Determination of MYCN amplification status is important because it is a biomarker strongly associated with unfavorable outcome of NB disease (31, 32). MYCN amplification detection has already been described in cfDNA, but most of these studies relied on real-time quantitative PCR or digital droplet PCR (18, 45–47). Our sWGS platform allowed the identification of MYCN amplification in all 13 cases with described MYCN amplification in the tumor cells. In addition, other focal copy-number aberrations associated with clinical outcome such as ATRX and CDKN2A/B deletion (5, 33, 34, 48) could be identified in the cfDNA. Another important prognostic biomarker is the presence of segmental CNAs being associated with poor survival outcome (6, 7). In 22 out of 24 cases, a copy-number profile with segmental aberrations was confirmed using sWGS on cfDNA. In three out of 11 cases, sWGS of cfDNA confirmed the presence of only numerical alterations. Interestingly, in one case with only numerical aberrations in the tumor sample, the cfDNA presented with additional segmental aberrations at chromosome 11 (case 32).
The relatively high occurrence of tumor heterogeneity is further demonstrated by the presence of alterations in the cfDNA that could not be detected in the tumor tissue in half of the cases (13 out of 25 cases where a nonflat profile was detected in the cfDNA). Detailed examination of these aberrations showed that some of these might only be present at the subclonal level. These findings underscore the problems associated with the availability of small biopsies, since they do not accurately represent the whole tumor. Furthermore, these results indicate that cfDNA analysis better reflects the spatial tumor heterogeneity present in a single patient as previously described (12, 13). However, alterations that were identified in the tissue DNA remained undetected in the cfDNA of five cases. Among others, we could not detect the presence of a small deletion encompassing the PTPRD gene. Optimization of the data-mining pipeline using z-score analysis or increasing read depth might help to detect such small CNAs. The applied sWGS procedure on cfDNA, as described in this project, uses the same workflow as our routinely performed NIPT (noninvasive prenatal testing) analysis for aneuploidy detection (21), which allows rapid and routine profiling of cfDNAs from patients with NB. While standard NIPT protocols start from 3.5 mL plasma, we found that reduced cfDNA input from plasma samples as little as 200 μL plasma were sufficient for successful analysis. The use of such small samples is particularly important for very young patients with NB, given the smaller blood volumes that can be sampled. Furthermore, compared with the previously reported array-based Oncoscan platform for which at least 20 to 50 ng of cfDNA is needed (20), sWGS only requires as little as 5 ng of cfDNA. Further advantages of sWGS profiling are the cost-effective procedure with a comparable price to arrayCGH and the fact that this sequencing-based method can be easily integrated in the workflow of contemporary molecular diagnostic laboratories, within a time frame of 21 calendar days.
In conclusion, we demonstrate the feasibility of sWGS to detect copy-number aberrations from a small input of plasma and cfDNA in a series of patients with NB representing all tumor stages. This is of special interest in clinical management of patients with NB, as these patients are often very young at diagnosis and biopsies are generally very small or cannot be performed. We therefore propose to include this low-cost WGS approach in the genetic work-up of patients with NB at diagnosis. For high-stage patients, cfDNA analysis could offer an alternative to analysis of scarce and small tumor biopsies; however, for most low-stage tumors, DNA analysis of the tumor biopsy will still be mandatory, but in these cases tumor availability is in general not a problem. Optimization and fine-tuning of the data-mining pipeline is warranted to allow detection of all present alterations including small alterations and homozygous deletions. Future plans will also be directed toward the identification of mutations in a selected panel of genes on cfDNA in patients with NB and toward the mapping of the temporal tumor heterogeneity. The sensitive detection of mutations, however, requires other techniques such as digital droplet PCR and duplex sequencing (49, 50). Indeed, a major advantage of liquid biopsies is that they are obtained noninvasively in the patient and therefore offers the possibility to do serial analyses: at diagnosis, during therapy, and at relapse (12, 13, 51, 52).
Disclosure of Potential Conflicts of Interest
No potential conflicts of interest were disclosed.
Conception and design: N. Van Roy, B. Menten, G. Laureys, B. De Wilde, F. Speleman, K. De Preter
Development of methodology: B. Menten, A. Dheedene, F. Speleman, K. De Preter
Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): N. Van Roy, C. Vandeputte, J. Van Dorpe, G. Laureys, M. Renard, T. Lammens, B. De Wilde, K. De Preter
Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): N. Van Roy, B. Menten, C. Vandeputte, T. Sante, F. Speleman, K. De Preter
Writing, review, and/or revision of the manuscript: N. Van Roy, M. Van Der Linden, B. Menten, C. Vandeputte, J. Van Dorpe, G. Laureys, T. Lammens, B. De Wilde, F. Speleman, K. De Preter
Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): N. Van Roy, M. Van Der Linden, B. Menten, A. Dheedene, C. Vandeputte, M. Renard, T. Sante, T. Lammens, F. Speleman, K. De Preter
Study supervision: N. Van Roy, B. Menten, G. Laureys, F. Speleman, K. De Preter
We are thankful for the technical assistance of Peter Degrave, Geert De Vos, Dimitri Broucke, Tamara Declercq, Isabel Brisard, Shalina Baute, Tine De Pretre, and Melek Yoruk.
The authors would like to thank the following funding agencies: the Belgian Foundation against Cancer (project 2015-146) to F. Speleman, the Flemish liga against cancer (B/14651/01 and STIVLK2016000601), Ghent University (BOF16/GOA/23) to F. Speleman, the Belgian Program of Interuniversity Poles of Attraction (IUAP Phase VII - P7/03) to F. Speleman, the Fund for Scientific Research Flanders (Research project G021415N) to K. De Preter and project 18B1716N to B. De Wilde, the European Union H2020 (OPTIMIZE-NB GOD9415N and TRANSCAN-ON THE TRAC GOD8815N) to F. Speleman. The Ghent University Hospital innovation project for molecular pathology (project number: KW/1694/PAN/001/001) to J. Van Dorpe and M. Van Der Linden, vzw Kinderkankerfonds, a nonprofit childhood cancer foundation under Belgian law to T. Lammens. Grant from Ghent University (Multidisciplinary Research Partnership “Bioinformatics: from nucleotides to networks,” 01MR0410) to B. Menten and A. Dheedene.
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.