Although neuroblastoma is characterized by numerous recurrent, large-scale chromosomal imbalances, the genes targeted by such imbalances have remained elusive. We have applied whole-genome oligonucleotide array comparative genomic hybridization (median probe spacing 6 kb) to 56 neuroblastoma tumors and cell lines to identify genes involved with disease pathogenesis. This set of tumors was selected for having either 11q loss or MYCN amplification, abnormalities that define the two most common genetic subtypes of metastatic neuroblastoma. Our analyses have permitted us to map large-scale chromosomal imbalances and high-level amplifications at exon-level resolution and to identify novel microdeletions and duplications. Chromosomal breakpoints (n = 467) generating imbalances >2 Mb were mapped to intervals ranging between 6 and 50 kb in size, providing substantial information on each abnormality. For example, breakpoints leading to large-scale hemizygous loss of chromosome 11q were highly clustered and preferentially associated with segmental duplications. High-level amplifications of MYCN were extremely complex, often resulting in a series of discontinuous regions of amplification. Imbalances (n = 540) <2 Mb long were also detected. Although the majority (78%) of these imbalances mapped to segmentally duplicated regions and primarily reflect constitutional copy number polymorphisms, many subtle imbalances were detected that are likely somatically acquired alterations and include genes involved with tumorigenesis, apoptosis, or neural cell differentiation. The most frequent microdeletion involved the PTPRD locus, indicating a possible tumor suppressor function for this gene. (Cancer Res 2006; 66(7): 3673-80)
- molecular cytogenetics
- pediatric cancers
- mapping and cloning of cancer genes
Neuroblastomas are derived from primitive cells of the sympathetic nervous system and account for ∼15% of all childhood cancer deaths (see ref. 1 for review). These tumors are particularly noted for extensive heterogeneity in clinical behavior, ranging from spontaneous regression to aggressive clinical course and death from disease. Patient age, tumor stage, and several different genetic abnormalities are important factors that influence clinical outcome. Loss of 1p and 11q, gain of 17q, and amplification of the MYCN oncogene are particularly strong genetic indicators of poor disease outcome ( 2– 5). Two of these abnormalities, loss of 11q and MYCN amplification, form the basis for dividing advanced-stage neuroblastomas into genetic subtypes due to their rather striking inverse distribution in tumors ( 6, 7). Many other recurrent partial chromosomal imbalances, including loss of 3p, 4p, 9p, and 14q and gain of 1q, 2p, 7q, and 11p, have been identified by metaphase comparative genomic hybridization (CGH) or allotyping studies ( 5– 7). Although many of these chromosomal imbalances undoubtedly play major roles in neuroblastoma pathogenesis, the targeted genes and genetic pathways have remained elusive.
Higher-resolution array CGH analyses hold great promise for identifying neuroblastoma tumor suppressor and oncogenes by narrowing down the critical regions that harbor the genes. CGH analysis of neuroblastoma using either BAC ( 8) or cDNA ( 9) microarrays have been used to detect more subtle chromosomal imbalances and to narrow the intervals containing genes important for neuroblastoma pathogenesis. Mosse et al. ( 8) used arrays containing ∼4,100 BAC clones (∼1-Mb average spacing) to analyze 42 neuroblastoma cell lines by CGH analysis. This study provided detailed mapping of shortest regions of overlap for many recurrent genomic imbalances and identification of novel abnormalities that were too subtle to detect by metaphase CGH. Although homozygous deletion of tumor suppressor gene regions have been noted at high frequency in several different tumor types ( 10), the array CGH study of Mosse et al. ( 8) identified only a single region of homozygous deletion, a 9.3-Mb region on chromosome 3p. Recently, we ( 11) showed the utility of oligonucleotide array CGH for mapping unbalanced chromosomal breakpoints and identifying subtle chromosomal imbalances at resolutions ranging from <1 kb (focused fine tiling arrays) to ∼25 kb (whole-genome arrays). The arrays contain up to 390,000 repeat masked oligonucleotides. Here, we report on the first major application of this oligonucleotide array CGH technology for high-resolution mapping of unbalanced chromosome breakpoints and identification of subtle chromosomal imbalances in a large set of human neuroblastoma primary tumors and cell lines. These results allow a global view of genomic instability at exon-level resolution and the identification of gene loci that are recurrently deleted.
Materials and Methods
DNA samples from primary tumors were obtained from several sources, including the Children's Oncology Group Tumor Bank (Philadelphia, PA), a tumor bank at Our Lady's Hospital for Sick Children (Dublin, Ireland), and the Children's Hospital at Westmead (Sydney, New South Wales, Australia). Tumors used in this study were selected based on either having loss of 11q material or MYCN amplification, which define the two most common genetic subtypes of stage IV disease. The abnormalities identified in this study are therefore not representative of the entire neuroblastoma population. Samples from the Our Lady's Hospital for Sick Children and Children's Hospital at Westmead were characterized previously for large-scale chromosomal imbalances by metaphase CGH analysis ( 12– 15), whereas samples from the Children's Oncology Group have been analyzed for loss of 11q by allotyping with several different polymorphic loci ( 6). MYCN amplification status in each sample was also assessed by either fluorescence in situ hybridization, Southern blotting, or quantitative PCR. Oligonucleotide array CGH results for three tumors (31, 37, and 47) and two cell lines (SK-N-AS and Kelly/N206) have been reported previously ( 11).
Oligonucleotide array CGH was carried out by NimbleGen Systems, Inc., at their facility in Iceland. Methods of DNA labeling, array construction, and hybridization as well as methods for array normalization and data analysis have been described in detail by Selzer et al. ( 11). Briefly, the whole-genome array incorporates 390,000 oligonucleotides (median probe spacing of 6 kb) from unique sequence regions that are of variable length to achieve a melting temperature of 76°C. Total genomic DNA from each tumor is directly labeled by random primer extension labeling (Cy3) without an intervening whole-genome amplification step ( 11). A matched reference DNA from the cancer patient was used as the reference DNA (labeled with Cy5) in a limited number of instances (tumors 178, 439, and 1389). Nonmatched reference DNAs were from a pool of six male individuals (Promega, Madison, WI). Data were analyzed using both unaveraged probes and by window averaging (25- and 50-kb window sizes) as described previously ( 11). In all instances, DNA sequence coordinates are from National Center for Biotechnology Information Build 35.
Methods for validating oligonucleotide array CGH results using real-time quantitative PCR have been described in detail by Selzer et al. ( 11). The single-copy loci GAPDH or ACTB were used for normalization. Primer sequences for each sequence-tagged site marker used in Fig. 4 can be obtained using the University of California at Santa Cruz Genome Browser tool. 6
Analysis of breakpoints leading to large-scale chromosomal imbalances. CGH analysis of 49 primary neuroblastoma tumors, 1 matched metastatic tumor, and 6 cell lines (SK-N-AS, Kelly/N206, SK-N-BE, IMR32, NGP, and NB69) was carried out using a whole-genome oligonucleotide microarray with a median probe spacing of 6 kb. A total of 467 independent chromosome breakpoints were identified that led to the generation of 416 partial chromosome imbalances >2 Mb long. Each breakpoint could be mapped to intervals ranging in size from 6 to 50 kb (breakpoint positions obtained primarily by 25-kb window averaging are listed in Supplementary Table S1). The distribution of the breakpoints was bimodal, with tumors tending to have either very few breakpoints (<5) or a large number (>15). Sixty-four whole chromosome gains or losses were also detected so that the total number of imbalances >2 Mb was 480.
There was a strong correlation between large-scale chromosomal imbalances (>10 Mb) detected by oligonucleotide array CGH and prior metaphase CGH analyses. However, oligonucleotide array CGH detected many imbalances <10 Mb that were not detected by metaphase CGH. There was also substantial consistency between loss of 11q detected by allotyping studies and oligonucleotide array CGH analysis, except for tumors 178, 151, and 1389. These three tumors were determined to have 11q loss by allotyping but not by oligonucleotide array CGH. A likely explanation for the inconsistency is that the event leading to loss of 11q heterozygosity was a copy neutral change, perhaps resulting from gene conversion or mitotic recombination (i.e., loss of one set of alleles and reduplication of the remaining set).
Chromosome 11q breakpoint positions ranged from coordinates 65.5 to 114.2 Mb among the 32 tumors or cell lines that had hemizygous loss of 11q. Fifty-six percent of these breakpoints clustered within a 4-Mb region between coordinates 68 and 72 Mb. Smaller clusters of breakpoints mapped between coordinates 80 and 85 Mb (19%) and to a narrow 260-kb interval between 98.775 and 99.03 Mb (9%) containing the CNTN5 locus. The most distal breakpoint at 114.2 Mb occurred in the MYCN amplified Kelly/N206 cell line. This abnormality may reflect a random, secondary alteration, so that the region 98 Mb to qter more likely represents the candidate tumor suppressor gene(s) region on 11q.
Ten of the 32 (31%) chromosome 11q breakpoint intervals leading to 11q loss contained segmental duplications (50-kb intervals assigned by 25-kb window averaging). From Armengol et al. ( 16), we were able to ascertain that the expected frequency of segmental duplications occurring within randomly placed 50-kb intervals on chromosome 11 is only 15%, allowing us to conclude that our observed frequency is significantly higher than would be expected by random chance. Five of the 11q breakpoints mapped within a narrow 225-kb interval (71037500-71262500 bp) covered by segmental duplications, further supporting the concept that 11q breakpoints are nonrandomly associated with segmentally duplicated regions.
Fifty breakpoints leading to gain of chromosome 17q were identified, with some tumors having multiple regions of 17q gain. Eighty percent of 17q breakpoints mapped to an 18-Mb interval between coordinates 27 and 45 Mb. Sixteen of 50 (32%) chromosome 17q breakpoint intervals mapped to segmentally duplicated regions, which is only slightly greater than would be expected (31%) based on the overall frequency of segmental duplications on this chromosome.
Twenty-seven breakpoints leading to loss of chromosome 3p could be identified, some of which involved interstitial loss of 3p material rather than terminal deletion. Seventy percent of 3p breakpoints mapped to a 20-Mb interval between coordinates 48 and 68 Mb. Clustering of 3p breakpoints was less evident than for either 17q or 11q, and none of the 27 breakpoints leading to 3p loss contained segmental duplications.
Analysis of the remaining 358 breakpoints leading to imbalances >2 Mb indicated that 59 (16%) were associated with segmental duplications, which is not significantly different from what would be expected by random chance (expected = 18% based on ref. 16).
Analysis of high-level amplification. Eighteen tumors or cell lines had high-level amplification of the MYCN locus at 2p24 (log2 ratio ≥ 2.0). High-level amplification also occurred at 18 chromosome 2 regions that were discontinuous with the MYCN locus and located at varying distances from MYCN (Supplementary Table S2). The level of amplification and number of distinct sites amplified in a tumor were highly variable. Some cases, such as the cell line SK-N-BE, possessed a single, contiguous region of amplification encompassing the MYCN gene ( Fig. 1A ). The majority of tumors, however, possessed amplifications of the MYCN region and other high-level amplifications on chromosome 2 that were noncontiguous with the MYCN amplicon. The additional amplification events were sometimes in close proximity to the MYCN locus ( Fig. 1B), and in some instances, many megabases distant from this locus ( Fig. 1C). Low-level gain of chromosome 2 regions that were interspersed between regions of high-level amplification were common. The most extreme example of moderate-level gains along the entire length of chromosome 2 occurred in tumor 1068, which had 36 distinct genomic regions with log2 ratios exceeding 0.7 ( Fig. 1D).
Three of the tumors possessing MYCN amplification also had high-level amplification of other chromosomal regions, including 11q13 (CCND1) in tumor 86 and 12q24 (MDM2) in tumor 6 and NGP. Both CCND1 and MDM2 amplification have been noted previously in neuroblastoma ( 17, 18). There were also additional sites of high level of amplification on chromosomes 11 and 12 that were discontinuous with CCND1 and MDM2 (Supplementary Table S2).
Analysis of breakpoints leading to small-scale chromosomal imbalances. Imbalances (n = 540) that were <2 Mb long, ranging from 0.025 to 1.3 Mb in size, were detected. It is unlikely that these subtle imbalances are technical artifact because no imbalances were observed using the same analysis criteria in three experiments where normal DNA from the same individual was used as both test and reference DNA ( 11).
The recent discovery of large-scale, widespread copy number polymorphisms (CNP) at segmentally duplicated loci ( 19, 20), however, suggests that many of these smaller-scale imbalances represent constitutional copy number differences between test and reference genomes. This possibility was tested by examining the sequences present at each imbalance with the University of California at Santa Cruz Genome Browser tool. 6 We ascertained that 383 of these variants mapped to regions containing segmental duplications and therefore might reflect CNPs. In addition, 38 imbalances, which did not overlap segmentally duplicated regions, corresponded to known CNPs listed in the Database of Genomic Variants. 7 Therefore, 421 (78%) of the imbalances <2 Mb long are potentially normal constitutional variants. The use of a matched constitutional DNA sample as the reference for three tumors strongly suppressed the appearance of subtle variants in those tumors, further indicating that the majority of subtle imbalances represent germ-line copy number differences between the patient DNA and the reference DNA (pooled from six normal males).
Not all imbalances at segmentally duplicated loci, however, were suppressed by the use of a matched constitutional reference DNA, indicating that some imbalances are somatic events. In several instances, we found that large-scale chromosomal gains or losses that include duplicated loci could lead to the “appearance” of imbalance in other regions that possess homology to the segmental duplications. For example, tumor 1359 had gain of a large region on chromosome 7q (69.96-107.76 Mb) that included a small region that is duplicated on chromosome 1p between 83.337 and 83.662 Mb. As shown in Fig. 2 , gain of the large segment of 7q resulted in the appearance of gain over a small 325-kb region of chromosome 1p (83.337-83.662 Mb) and over a 500-kb region on 7q (75.762-76.262 Mb) that has homology to the 1p region and other sites on 7q. Although we cannot rule out the possibility that copy number polymorphic differences contributed to the imbalance, CNPs do not have to be invoked to explain these effects. These types of effects were noted for several other genomic regions.
The remaining 119 imbalances that were <2 Mb long did not map to segmentally duplicated regions and were not listed as CNPs in the Database of Genomic Variants. Fifty-nine of these 119 imbalances mapped to either intergenic regions or predicted gene sequences of unknown function. The significance of these imbalances remains to be determined. Sixty imbalances mapped to regions containing characterized genes, many of which play roles in tumorigenesis, apoptosis, or neural cell differentiation ( Table 1 ). The most frequent region of recurrent microdeletion involved the PTPRD gene region at chromosome 9p23 in three cell lines (NGP, Kelly, and SK-N-AS) and three tumors (31-MAR, 430, and 47). The deletions included either PTPRD coding exons (NGP, metastatic tumor 31-MAR, and tumor 430) or exons from the 5′ untranslated region (UTR; tumor 47, Kelly, and SK-N-AS). We also found one tumor (1173) with an unbalanced chromosome 9p breakpoint that disrupted the 5′ UTR of PTPRD. Chromosome 9p deletions, along with a deletion map of the PTPRD locus, are illustrated in Fig. 3 .
The PTPRD deletion in 31-MAR was clearly somatically acquired because 31-MAR represents a metastatic bone marrow aspirate ( Fig. 3A) and the deletion was not present in material from the primary tumor ( Fig. 3B). Real-time quantitative PCR analysis of primary tumor DNA and a matched constitutional DNA from case 430 also confirmed that the deletion was somatically acquired in the tumor from this patient ( Fig. 4A ).
In the Kelly/N206 cell line, two different shifts in log2 ratio were detected, indicating that part of the PTPRD region is homozygously deleted. As illustrated in Fig. 4B, real-time quantitative PCR analysis with primers from the putative homozygously deleted region generated primer product that was significantly lower than would have been expected for only hemizygous loss, thus confirming homozygous loss. The SK-N-AS cell line had large-scale loss of the majority of 9p as well as a further decrease in the log2 ratio for a small region of PTPRD, also indicative of homozygous loss. Real-time quantitative PCR analysis of SK-N-AS with primers from the undeleted region, the region of hemizygous loss, and the putative homozygously deleted region was consistent with homozygous deletion in a subpopulation of cells ( Fig. 4C).
The exon-level resolution mapping of unbalanced chromosomal breakpoints accomplished in this study helps narrow down the regions harboring potential neuroblastoma oncogenes or tumor suppressor genes. At the same time, these experiments provide information on the sequences found in proximity to the breakpoints, which can facilitate our understanding of the events that generate recurrent chromosomal imbalances. Breakpoints leading to 11q loss in neuroblastoma cells were more clustered than breakpoints on other chromosomes and contained segmental duplications at a higher than expected frequency, indicating that the duplications play a role in generating 11q breakpoints. Thus, it seems likely that the mechanism generating 11q loss in many instances is quite different from the mechanisms generating other chromosomal imbalances. In this regard, we note that loss of 11q occurs primarily through a common chromosomal mechanism, an unbalanced t(11;17) (ref. 15), whereas the mechanisms leading to many other recurrent abnormalities, except for loss of 1p which is due to a t(1;17), are more variable. The segmental duplications found at the 11q breakpoints, however, are not duplicated within the defined 17q breakpoint intervals, indicating that they have not caused inappropriate homologous recombination to occur. The precise molecular mechanisms generating this translocation require further studies.
The detection of a complex series of highly amplified regions that are discontinuous with MYCN and interspersed with regions of lower-level gains is consistent with earlier studies by Shiloh et al. ( 21). The oligonucleotide array CGH analyses carried out in our study, however, provide substantially higher-resolution mapping of amplicon boundaries than what has heretofore been carried out.
The extremely high-resolution achievable by oligonucleotide array CGH analysis, although a major strength, is also a weakness because of the present difficulties associated with distinguishing pathologically important imbalances from those that are simply due to the high level of genomic instability that is characteristic of most tumors. The imbalances detected at segmentally duplicated loci often reflect constitutional copy number differences between tumor and reference DNA sample from different individuals as observed in several studies ( 19, 20). Our results indicate that the use of a matched constitutional DNA sample from the patient as the reference DNA sample does not completely eliminate the appearance of small imbalances at segmentally duplicated loci. In some instances, the appearance of subtle imbalances at segmentally duplicated regions is caused by the large-scale losses or gains of regions containing segmental duplications that are homologous to the subtle region where imbalance is detected. These intrachromosomal/interchromosomal “long distance” effects are easily traceable and should be taken into account when interpreting high-resolution oligonucleotide array CGH data. Although use of a matched constitutional reference DNA sample suppresses the appearance of the majority of CNP variants, matched samples are not always available for rare pediatric tumors. The development of a comprehensive CNP database should be very helpful for distinguishing between normal variants and somatic abnormalities found in some tumors.
We had an expectation of finding homozygous deletions for the 11q region in neuroblastoma given that homozygous deletions have been identified at high frequency in cell lines from many different tumor types (e.g., ref. 10). However, consistent with the BAC array CGH study of Mosse et al. ( 8), we have found that recurrent homozygous deletions are exceedingly rare in neuroblastoma. Apart from a previously reported 67-kb homozygous deletion of an intronic region of CNTN5 ( 11), no homozygous loss on chromosome 11q could be detected in any of the 32 tumors and cell lines that possessed large-scale hemizygous loss of this chromosomal region. Interestingly, the breakpoint resulting in large-scale hemizygous loss of 11q disrupted the CNTN5 locus in two primary tumors (1100 and 23) and one cell line (NGP). CNTN5 is a glycosylphosphatidylinositol-anchored neuronal membrane protein that mediates cell surface interactions during nervous system development. Further studies are required to determine if this region is simply a hotspot for breakage or whether disruption of this gene plays a role in neuroblastoma pathogenesis.
The lack of homozygous deletions on chromosome 11q could mean that a tumorigenic effect is being rendered through haploinsufficiency of one or more genes on 11q. In this regard, there are two genes on chromosome 11q involved with genomic stability (ATM and H2XA) that may have a tumorigenic effect through a haploinsufficient mechanism (see ref. 22 for review). A simple model for the development of the 11q− neuroblastoma subtype would involve simultaneous loss of 11q and gain of 17q through an unbalanced t(11;17). Perhaps hemizygous loss of critical genes involved with maintenance of genomic integrity on 11q, such as ATM and H2XA, coupled with gain of antiapoptotic sequences, such BIRC5 and miR-21, on 17q, leads to an early-stage tumor. Loss of 11q and gain of 17q were the sole imbalances detected in one tumor, which supports the notion that these are early events. In addition, expression microarray analysis of 11q− neuroblastomas has shown that rare low-stage 11q− tumors sometimes have global gene expression profiles that are much more similar to low-stage hyperdiploid tumors than to high-stage 11q− tumors, an indication that loss of 11q and gain of 17q are insufficient for the establishment of metastasis ( 13). It is tempting to speculate that some of the microdeletions and duplications listed in Table 1 contribute to the establishment of metastatic potential in neuroblastoma; however, the possibility that many of them are merely a consequence of global genomic instability cannot be ruled out. Additional studies are required to confirm any role that these subtle abnormalities might play in neuroblastoma pathogenesis.
One of the more important events for the development of metastatic neuroblastoma might be loss of PTPRD, which occurred in ∼12% of our samples. PTPRD seemed to be homozygously deleted in two neuroblastoma cell lines and is also commonly homozygously deleted in lung cancer ( 23– 25), suggesting that it plays a fundamental role in cancer development. It was recently determined that two mRNA isoforms of PTPRD exist, one containing a long 5′ UTR (L form) and a second without the 5′ UTR (S form; ref. 24). This gene cooperates with the MIM gene to induce cytoskeletal changes ( 26). It has been suggested that deletions occurring in the PTPRD locus could represent some kind of region-specific hotspot for rearrangement ( 24), given that large-scale inversions and duplications nonrandomly occur in this region in lymphocytes of BRCA2 mutation carriers ( 27). However, neither neuroblastoma or lung cancer patients are likely to have constitutional mutations in BRCA2, and the large-scale duplications and inversions on chromosome 9 are quite different from the subtle somatically acquired deletions. A role for PTPRD in cancer seems plausible, given that other protein tyrosine phosphatase receptors play important roles in tumorigenesis ( 28– 30). Ascertaining the possible functional effects that deletions of the 5′ UTR might have on either transcription or translation of PTPRD should shed considerable light on this issue.
Grant support: Children's Medical and Research Foundation (Dublin, Ireland).
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.
Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/).
- Received November 18, 2005.
- Revision received January 11, 2006.
- Accepted February 1, 2006.
- ©2006 American Association for Cancer Research.