Neuroblastoma (NB) is a pediatric tumor originating from primitive sympathetic nervous cells, and is characterized by a variable clinical course and genetic heterogeneity. Three major genetic subtypes of NB are recognized: favorable near triploid tumors with a pattern of predominantly numerical gains and losses (Subtype 1) and 2 unfavorable subtypes with either 11q-loss and single copy MYCN (Subtype 2A), or with MYCN amplification and 1p-deletion (Subtype 2B).1, 2 The latter 2 metastatic subtypes almost invariably present with partial 17q gain, whereas tumors belonging to Subtype 1 usually present with whole chromosome 17 gain. The high incidence of 17q gain and the finding that this chromosomal defect is the strongest independent prognostic genetic marker in NB2, 3 has urged us to perform further investigations aimed at understanding the underlying molecular pathology. Previous low resolution mapping studies revealed no clustering of 17q translocation breakpoints, thus excluding the implication of a fusion gene or activation of a single oncogene. Consequently, it has been hypothesized that 17q overrepresentation confers an oncogenic effect through copy number gain of 1 or more critical dosage sensitive genes.3, 4, 5, 6 Given the ubiquitous occurrence of 17q gain in unfavorable tumors, identification of genes and pathways implicated by this genetic alteration is important as they may serve as targets for new therapies.
Recently, arrayCGH has been successfully applied to study genomic imbalances. Several studies have reported on the genome wide profiling of DNA copy number alterations in NB (for review see Ref.7). These studies have provided useful new information on the content and delineation of recurrent losses, gains and amplification of chromosomal segments. ArrayCGH also allows mapping of unbalanced translocation breakpoints (either presumed or inferred from available karyotypes). This was illustrated by high resolution mapping of 17q breakpoints in 6 NB cell lines and 138 primary tumors.8, 9
The primary aim of our study was to search for small interstitial regions of 17q gain in order to facilitate the identification or mapping of genes on 17q involved in NB pathogenesis. Although no such interstitial gains were detected, we could pinpoint 2 regions on 17q with putatively critical dosage sensitive genes through a combination of high resolution arrayCGH mapping of 17q breakpoints and positional gene enrichment analysis.
Material and methods
Primary NB tumors and cell lines
Primary NB tumor samples were collected before therapy at the Ghent University Hospital (Ghent, Belgium), Centre Léon Bérard (Lyon, France), Children's Cancer Research Institute (Vienna, Austria) and at the University Children's Hospital of Essen (Essen, Germany). In our study, 69 primary tumors were analyzed and staged according to the international neuroblastoma staging system (INSS), including 12 Stage 1, 9 Stage 2, 13 Stage 3, 29 Stage 4 and 6 Stage 4s tumors. Informed consent from all patients' relatives was obtained and the study was approved by the ethical committee of the Ghent University Hospital, Ghent, Belgium. In addition, 28 NB cell lines were included in this analysis. An overview of the most common genetic alterations of the examined primary tumors and cell lines is listed in Supplementary Table I and II in Michels et al.10
ArrayCGH analysis was performed according to the method described in Michels et al.10 Briefly, tumor DNA or cell line DNA and control DNA (500 ng) were differentially labeled with Cy3 and Cy5, and after blocking with Cot1-DNA hybridized to an in house produced 1 Mb resolution BAC array, supplemented with tiling path clones for chromosome 17 (resolution ∼130 kb). After hybridization, slides were scanned using a GMS 418 Array Scanner (Genetic MicroSystems, Woburn, USA). The scanned images were processed with Imagene 5.5 software (BioDiscovery, El Segundo, USA) and data analyzed using arrayCGHbase (http://medgen.ugent.be/arrayCGHbase/).11 Reporters were included in the analysis if all of the following criteria were met: signal to noise ratio > 3, standard deviation of the log2 ratio between triplicates < 0.2 and > 1 informative replicate. Circular binary segmentation analysis (implemented in arrayCGHbase) was used to delineate the boundaries of regional chromosome losses or gains.
Positional gene enrichment analysis
To delineate 17q regions containing dosage sensitive genes, we applied positional gene enrichment analysis (http://medgen.ugent.be/pge) (window width, 5 Mb; step size, 1 Mb; with correction for multiple testing of significance). This procedure enables the identification of chromosomal regions that are significantly overrepresented in custom gene lists through repeated Fisher's Exact test analysis of consecutive genome segments (unpublished data). By applying this method to the genes overexpressed in NB compared to normal fetal neuroblasts,12 we pinpoint so-called gene overexpression clusters.
Fluorescence in situ hybridization
For all examined primary tumors and cell lines, fluorescence in situ hybridization (FISH) was used to detect the most common genetic aberrations, i.e., MYCN amplification and 1p deletion. Probe labeling and FISH were performed as described earlier.13
To assess the accuracy of arrayCGH for breakpoint position identification, we randomly selected 11 NB cell lines with overrepresentation of distal 17q and performed dual-color FISH analysis using a combination of BAC RP11-678P16 (17p12), and the most distal normal clone or most proximal gained clone in the cell line under investigation.
Cox linear regression analysis
Data from 2 published NB microarray studies were downloaded.14, 15 Before importing the data in R, probe annotations and gene symbol names were updated using MatchMiner16 (http://discover.nci.nih.gov/matchminer). Cox linear regression analysis was performed for the expression data of the genes within the candidate regions using the coxph function in the R package survival.
Screening of 69 NB tumors and 28 NB cell lines yielded neither small interstitial gains nor amplifications, whereas gains of large distal 17q segments were detected in 25/28 (89%) of the NB cell lines and in 5/27 (18.5%) and 24/42 (57%) of low- and high-stage primary tumors, respectively. The absence of interstitial gains or amplifications is in keeping with the fact that these 17q gains are generally the result of unbalanced translocations leading to extra copies of the entire chromosome segment distal to the 17q breakpoint. Whole chromosome 17 gain was found in 33/69 investigated tumors, mostly Subtype 1 cases (age at diagnosis < 1 year, localized tumor stage, near triploid DNA content, and within the context of predominantly numerical chromosome changes). Two tumors showed whole chromosome 17 gain as well as partial 17q gain (these tumors were not included in the previous category of whole or partial chromosome 17 gain). Kaplan–Meier overall survival analysis confirmed the known association between partial 17q gain and adverse prognosis (p < 0.05).
Loss of the short arm of chromosome 17p was found in 12/28 (43%) NB cell lines and in 2/69 tumors (3%). Furthermore, 7 cases (6 cell lines and 1 primary tumor) showing loss of 17p also showed loss of the proximal part of 17q, almost always in association with gain of the region immediately distal to the deleted region. Seven tumors showed a normal chromosome 17 arrayCGH profile. Representative examples of 17(q) gain detected by arrayCGH are given in Figure 1. The detailed arrayCGH karyotypes are described in Michels et al.10 and raw data are available at arrayCGHbase (http://medgen.ugent.be/arrayCGHbase/- experiments, project Michels et al.).
The genomic position of 52 different chromosome 17q breakpoint positions were mapped at the single BAC level. An overview of all identified breakpoint positions is represented in Figure 2. In 7 of the 11 randomly selected NB cell lines, delineation of breakpoint positions as determined by arrayCGH was confirmed by FISH analysis. Moreover, in 2 of the 7 cell lines, the selected BAC clones were breakpoint overlapping (Fig. 3).
Two of the 28 NB cell lines and 7 of the 69 primary tumors presented with more than 1 breakpoint leading to a region of superimposed 17q gain, which is in keeping with the observation that some cases contain more than 1 unbalanced 17q translocation, often with different partner chromosomes.5 If one takes into account only the single regions of gain, the breakpoints leading to the largest and smallest gained segment are mapped at position 18.2 Mb (17p11.2) and 44.3 Mb (17q21.32), respectively (Ensembl v38) (Fig. 2). As previously indicated by G-banding, FISH and chromosomal CGH analysis, our results confirm that breakpoints are widely spread across the proximal long arm of chromosome 17. When combining our breakpoint mapping results with those from 2 recent high-resolution copy number profiling studies,8, 9 an almost contiguous breakpoint region of 24.6 Mb (17p11.2-q22) emerges, totalling 150 consecutive breakpoints (Supplemental Fig. 1).
As mentioned earlier, in the samples presenting with a single region of gain, none of the breakpoints in our study mapped distal to chromosome band 17q21.32. On the basis of these findings and given the predominant or unique role of unbalanced translocations leading to 17q gain, we thus assumed that the genes critically affected in NB through gene dosage effect should be located in the 17q21.32-17qter or in the 17q24.1-17qter segment (for tumors with superimposed gain). To pinpoint the putative dosage sensitive genes, we performed positional gene enrichment analysis (PGE) to search for dosage sensitive genes located across chromosome 17. By mapping the genes overexpressed in NB compared to normal neuroblasts (Supplemental File 1), we could demarcate 2 highly significantly enriched regions (p < 0.01, multiple testing corrected) on the long arm of chromosome 17 (Supplemental File 2). A first 3.61 Mb region encompasses the most distal 17q translocation breakpoint at 44.3 Mb (in samples with single regions of gain) and contains the following 11 known genes: KPNB1 and NFE2L1 (centromeric to the most distal breakpoint), and CALCOCO2, SNF8, SPOP, SLC35B1, MYST2, LRRC59, NME1, NME2 and UTP18 (distal to the most distal breakpoint). All details of the expression data are described in De Preter et al.12 and all expression microarray data are available from ArrayExpress http://www.ebi.ac.uk/arrayexpress/, under accession number E-MEXP-669.
Upon prognostic evaluation of all overexpressed genes in 2 large microarray studies using Cox linear regression, the KPNB1 gene was shown to be prognostic in both studies, while NME1 and NME2 each had prognostic power in 1 of the 2 studies, respectively (p < 0.05, multiple testing corrected).14, 15 This indicates that these genes are not only overexpressed but are also linked to patient survival and hence tumor aggressiveness. This makes them excellent candidates for further study in the search for dosage sensitive 17q genes in NB. The second region on chromosome 17q that shows significant enrichment for overexpressed genes is located around coordinate 60 Mb (17q24.1) and encompasses the superimposed gain breakpoint region containing the following genes: PECAM1, DDX5, PRKCA, PSMD12, AMZ2 and PRKAR1A. Concise information on each of these genes can be found in Supplemental File 4.
Chromosome 17q gain in NB has been recognized as one of the most frequently occurring genomic imbalances as demonstrated by FISH and chromosomal CGH analyses.13, 17, 20 Such 17q gains are typically present in aggressive tumors, both with and without MYCN amplification, and are associated with a poor prognosis.2, 3, 21 Despite this striking association, little progress has been made toward the identification of the culprit 17q oncogenes that contribute to NB pathogenesis through a presumed gene dosage effect. One of the first aims of our study was the identification of small duplications or amplifications (<5 Mb) that might have been overlooked by previous low resolution chromosomal CGH investigations. To this purpose, we used a chromosome 17 tiling path array for the analysis of 28 NB cell lines and 69 primary tumors. This screening, however, yielded no small interstitial 17q gains. Therefore, unbalanced chromosome 17 translocations seem to be the only mechanism contributing to 17q gain and consequently, mapping of breakpoints alone will not contribute to the further delineation of a critical region for 17q gain.
In the NB cell lines, a high frequency of 17p deletions was observed (43% vs. 3% in primary tumors). This finding is probably because of the fact that most of the NB cell lines are established from primary tumors that are refractory to therapy. Recently, it has been shown that inactivation of the p53 pathway is common in chemoresistant NBs.22 Therefore, we could hypothesize that the high frequency of 17p deletions in the cell lines is probably therapy induced and leads to p53 loss. The same high frequency of 17p deletions in NB cell lines was also noted by Schleiermacher et al.5 who used LOH analysis.
In addition, the frequency of copy number alterations (CNA) in our study is much higher in cell lines than in primary tumors (mean 13 CNA in cell lines and 9 CNA in primary tumors).18 Moreover, some copy number changes are confined to cell lines only such as deletions of 6q.10 These findings could also be due to in vitro culturing conditions.
The chromosome 17 tiling arrays enabled us to map the translocation breakpoint intervals at a high resolution level. The breakpoints were found to be scattered over a 21 Mb region, in cases with single regions of gain, all mapping to the proximal part of 17q (17q11.2-17q21.32, 22.58-44.3 Mb). Our results are consistent with 2 recent high-resolution copy number profiling studies, in which 44 of 55 of the 17q breakpoints were mapped to a 18 Mb interval located between 27 and 45 Mb,9 and 41 of the 43 breakpoints in a 21 Mb interval between 26.5 and 47.5 Mb.8 Taken altogether, an almost contiguous breakpoint region becomes apparent, overlapping with a region containing many intrachromosomal segmental duplications and remaining sequence gaps (Supplemental Fig. 1). This indicates that this region might contain unstable sequences, that form secondary structures, predisposing it to be involved in unbalanced translocations.23
In most cases, the arrayCGH profile was suggestive for a single 17q breakpoint (leading to a single region of gain). The resulting copy number gain for 17q invariably implicated a large segment, encompassing at least the distal part from 17q21.32 (44.3 Mb) to 17qter. This region comprises at least 34.5 Mb and 391 known genes. These results are in keeping with the arrayCGH results obtained in another high resolution study9 in which the most distal breakpoint (in all cases with a single region of gain, with exception of case 6) was defined at coordinate 45 Mb.
In our study, as well as in previous studies,5, 9, 27 some tumors and cell lines showed in addition to a large region of gain, a second smaller region of superimposed gain more distally located around coordinate 60 Mb (17q24.1).
In view of the consistent gain of the 17q21.32-qter segment and to a lesser extent the superimposed gain of 17q24.1-qter and assuming that these gains almost exclusively result from unbalanced translocations, we hypothesize that 1 or more critical genes sensitive to a gene dosage effect contributing to NB pathogenesis are located telomeric to the most distally located breakpoints at 17q21.32 and 17q24.1. To test both hypotheses, we performed positional gene enrichment analysis using available NB and normal fetal neuroblast transcriptome information, to search for overexpressed genes in NB cells that map to chromosome 17. Intriguingly, the 2 chromosomal regions with nonrandom enrichment of overexpressed genes coincide with the 2 proposed critical 17q regions delineated through our copy number analysis and thus providing, albeit indirect, evidence for our hypothesis. Furthermore, Cox linear regression analysis on 2 large microarray studies supports the fact that 3 of this genes residing in this region has prognostic significance14, 15 and therefore makes them excellent candidates for further study in the search for dosage sensitive 17q genes in NB. The karyopherin (importin) beta 1 gene (KPNB1) encodes for a protein that is involved in nuclear protein import, either in association with an adaptor protein or as an autonomous nuclear transporter receptor.25 Recently, the KPNB1 gene was found to be involved in the nuclear import of receptor tyrosine kinases such as ERBB2 and EGFR, both known to be implicated in cancer.26, 27 The human metastasis suppressor genes NME1(NM23-H1) and NME2(NM23-H2) are members of the gene family encoding nucleoside diphosphate kinases.28 The NME genes have been shown to play an important role in cellular proliferation, differentiation and tumor metastasis.29, 30
Reduced expression levels of NME1 and NME2 are usually associated with the increased metastatic potential of several cancer types.31 In NB, however, overexpression of NME is associated with an advanced stage and poor prognosis.32 Furthermore, somatic mutations in the NME genes have been described in NB, providing a possible role for the NME genes in NB.
The identification of a second more distal region on 17q24.1 by PGE analysis is relevant in the light of the observation of superimposed gain of a smaller segment in 2 cell lines and 3 tumors in our study. A similar observation was made by Schleiermacher et al.,5 Lastowska et al.24 and Stallings et al.9
As genomic analyses only are not likely to uncover a small critical region amenable to further study of selected candidate genes, the combination of breakpoint mapping together with information on overexpressed 17q genes in NB now for the first time yields a relatively small list of putative candidate genes and thus opens the way to further functional studies to finally identify the culprit NB 17q oncogenes. Until the gene dosage hypothesis is experimentally proven and the actual causal 17q genes have been identified, other hypotheses, however, also remain valid. One of these hypotheses constitutes an imbalance for genes located on both sides of the breakpoint as originally suggested by Bown et al.3 This hypothesis was also put forward by Chen et al.,33 in the light of the finding of overexpression of the WSB1 gene, located at 17q11, in favorable NB with whole chromosome 17 gain (also confirmed in our recent study by comparing the expression profiles of NB and normal fetal neuroblasts12).
Although the 17q gain enigma in NB remains to be resolved, the integrative analysis of genome and transcriptome profiles, as illustrated in our study, is a promising approach to identify candidate dosage sensitive genes. Future studies should aim at functional assessment of the dosage sensitivity, amongst others by knock-down of the candidate genes.
We thank the Wellcome Trust Sanger Institute (Hinxton, Cambridge, UK) for providing us with the 1 Mb BAC probes and tiling path chromosome 17 clones, The authors thank Dr. S. Ragsdale and Dr. J. Lunec for providing cell line material of respectively the NB-14 and SK-N-BE(2c) cell line. Prof. Jo Vandesompele and Prof. Nadine Van Roy are postdoctoral researchers of the Fund for Scientific Research (FWO), Flanders. Dr. Katleen De Preter is a postdoctoral researcher of the Flemish Institute for the Promotion of Scientific Technological Research in Industry (IWT). This text presents research results of the Belgian program of Interuniversity Poles of Attraction initiated by the Belgian State, Prime Minister's Office, Science Policy Programming and the European 6th framework programme EETpipeline.