Chromosomal Localization of DNA Amplifications in Neuroblastoma Tumors Using cDNA Microarray Comparative Genomic Hybridization¹

Patient Tissue Accrual and Polymerase Chain Reaction (PCR) of MYCN Copy Number

Primary tumors were all clinically matched stage IV neuroblastomas, collected from patients at the Hospital for Sick Children. All samples were anonymous and were collected under the guidelines of the Hospital for Sick Children Research Ethics Board. The tumors were flash frozen directly after surgical removal and stored in liquid nitrogen. Genomic DNA from the tumor was extracted following standard protocols [12]. The amplification status of the MYCN gene was determined by standard quantitative PCR analyses as previously described [13]. Briefly, samples P1126, P1071, P1079, and P1008 were found to have approximately 100 or more copies of MYCN. Sample P1061 showed intermediate 25 to 50 copies of MYCN, and samples D3025, D3060, D3077-95, and D3077-97 were observed with less than 25 copies. Samples P1126, P1079, and P1071 were derived at different time points (pretreatment, posttreatment, and posttreatment inguinal lymph node metastasis, respectively) from the same patient. Table 1 lists the clinical features of the patient cohort.

Chromosome CGH and Fluorescence In Situ Hybridization (FISH)

Established protocols for CGH on human metaphase chromosome spreads were followed [3]. Image capture and analysis were performed using the Vysis PathVysion and Karyotype software, respectively (Vysis, Downers Grove, IL). Telomeric and centromeric regions were excluded in the analyses due to the presence of highly repetitive genomic sequences at these sites. Ten metaphases were analyzed to create the final CGH profile with 99% confidence intervals. Where possible, FISH assessment of MYCN copy number in cytogenetic preparations from neuroblastoma primary tumors was performed as previously described [13]. A minimum of 100 interphase nuclei was evaluated in each cytogenetic sample for assessing the MYCN copy number.

cDNA Array CGH

Microarrays used in this study [Human 19k2 microarray, Clinical Genomics Centre (CGC), Toronto, Ontario, Canada] contain a total of 19,200 features (Research Genetics, Invitrogen, Huntsville, AL) including 18,980 human cDNAs and 220 positive and negative control features. The features of the 19k2 microarray are arranged on two glass slides (parts A and B), each with 32 subgrids spotted over an approximate area of 18x36 mm². All features are placed in duplicate, for a total of 38,400 spots. A complete list of the cDNA collection used for these arrays and protocols used for array construction can be found at the University Health Network CGC web site (http://www.microarrays.ca). It should be noted that some genes that have been reported to be part of the MYCN amplicon, such as N-cym [14] and DDX1 [15], are not presently on the current configuration of the 19k2 microarray. Sequence verification studies being performed by the CGC allowed us to identify by BLAST analysis a small percentage of cDNAs contaminated with mitochondrial DNA (mtDNA). Identification of such sequences has allowed us to make appropriate changes and filters for our software suite (see below).

The procedure for cDNA array CGH is a modified version of the protocol published by Pollack et al. [11], and reviewed in Ref. [16]. Each microarray slide (part A and B) was blocked under a glass coverslip for 1 hour at 37°C with blocking solution (3% BSA, 4x SSC, 0.1% Tween-20) before hybridization with the denatured probe. Four micrograms each of high molecular weight tumor and normal genomic DNA was separately digested with EcoRI for 2 hours, purified (Qiaquick PCR kit), vacuum dried, and resuspended in 25 µl water. Random primer labeling was performed using the Bioprime Labeling kit (Invitrogen, Carlsbad, CA) according to manufacturer's instructions, with modifications to directly incorporate Cy3- and Cy5-dCTP (Amersham, Piscataway, NJ). The Cy3- and Cy5-labeled products were suppressed with 30 µg Cot-1 DNA (Invitrogen), 100 µg yeast tRNA (Invitrogen), and 20 µg poly(dAdT) (Sigma-Aldrich, St. Louis, MO), loaded onto a microcon 30 filter (Amicon, Bedford, MA), and centrifuged for 20 minutes at 5000xg. The sample was recovered with 60 µl hybridization buffer (3.4x SSC and 0.3% SDS), and denatured at 100°C for 90 seconds. The probe was immediately chilled on ice, and allowed to preanneal at 37°C for 0.5 to 1 hour. The probe was divided and added to each microarray slide (part A and B) under a glass coverslip and sealed with rubber cement. Hybridization was at 37°C for 16 to 20 hours in a moist chamber humidified with hybridization buffer. The microarrays were washed at 65°C for 5 minutes in 2x SSC, 0.03% SDS, followed by successive washes in 1x and 0.2x SSC at room temperature (5 minutes each). After air drying, the slides were scanned using an Axon GenePix 4000A confocal scanner and fluorescence intensities were quantified with GenePix Pro 3.0 software (Axon Instruments, Union City, CA). Replicate and reverse-labeling experiments were performed for selected samples. Control genomic DNAs used in these studies included three cell lines with established levels of high copy number MYCN amplifications, including the retinoblastoma cell line Y79, and the neuroblastoma cell lines IMR32 and NUB-7 (American Tissue Culture Collection, Rockville, MD).

Bioinformatics and Analyses

Custom software scripts were developed to update the chromosome localizations for the cDNAs of the 19k2 microarray. However, any gene list having GenBank accession and I.M.A.G.E. clone [17] identifiers can be updated with this approach. The algorithm involved parsing NCBI UniGene and MapViewer databases to identify up-to-date cytoband localizations. Briefly, the UniGene database (build 147) was retrieved (ftp://ftp.ncbi.nih.gov/) and used to update UniGene identifiers in the gene list based on accession number. Subsequently, the accession and updated UniGene identifiers for each cDNA were cross-referenced to sequence mapping data and additional cytoband information in the MapViewer and UniGene databases, respectively. The information updates for each cDNA in the gene list include the chromosomal distance in megabases, the corresponding cytoband (based on ISCN 800 level band resolution size measurements; [18]), and a qualitative assessment score of mapping precision based on the concordance of the results gathered from the databases. Updated gene lists for these microarrays, as well as a more detailed description of the update algorithm, can be found at the laboratory's web site (http://www.utoronto.ca/cancyto/). Table 2 depicts the genetic representation of the 19k2 microarrays.

To facilitate analysis of results from multiple 19k2 microarray experiments, custom software was also developed. The software suite is currently comprised of four PC DOS-based components: 1) normalize, 2) project, 3) clusterp, and 4) profiler. The normalize software normalizes quantified Cy5 and Cy3 fluorescence intensities, with or without background subtraction, across the entire array as well as across individual subgrids. Additionally, preceding normalization, filters are applied to remove spots below user-specified thresholds for fluorescence intensity, foreground-to-background ratio, and/or spot diameter size. After normalization and filtering, duplicate spots from the microarray are averaged together. Using project, two or more normalized replicates of an experiment, including reverse-labelings, can be grouped together as a project. Additionally, project also groups together results from parts A and B of a 19k2 microarray experiment. The clusterp software uses selected project files to generate output that can be read by the Eisen Cluster software package (http://rana.lbl.gov) for two-dimensional hierarchical clustering [19,20]. Clustering was performed on genes showing hybridization in ≥80% of samples using an uncentered correlation similarity metric. Finally, profiler uses the updated chromosome localization information to generate composite chromosome plots of normalized fluorescence intensity ratios from selected projects, generating images analogous to karyotype profiles from chromosome CGH. Note that only results for cDNA sequences with known cytoband and megabase position are plotted. To make microarray CGH methods more generally accessible to the research community, this software suite and the algorithms for automated cytoband and megabase localization of microarray cDNA features are freely available online (http://www.utoronto.ca/cancyto/). All experimental results are also available online as tab-delimited text files (http://www.utoronto.ca/cancyto/NB2003/).

Optimization and Performance of cDNA Array CGH

The cDNA array CGH technique was optimized initially by comparing genomic DNA from IMR32, a metastatic neuroblastoma cell line with known copy number changes, and characterized by chromosome CGH (Figure 1A), against a normal reference. Replicate cDNA array CGH experiments (n=4) including a reverse-labeling hybridization demonstrated good reproducibility and the expected MYCN amplification of sequences mapping to the 2p24 chromosomal region. The distribution of normalized fluorescence intensity ratios for all hybridization features (n=72,139 in four hybridizations) was 0.94±0.25 (mean±1 SD), establishing a 99% confidence interval limit of 0.19 to 1.69. (Figure 1B) shows the composite CGH genomic profile of IMR32 by cDNA array CGH.

To semiquantitate the chromosomal distributions and relative levels of copy number imbalance by CGH using both cDNA and metaphase targets, the MYCN normalized fluorescence intensity ratios was derived from a series of 12 neuroblastomas and cell lines against copy number measurements determined by standard quantitative PCR and Southern blot methods. Southern blot analyses of the neuroblastoma cell lines IMR32 and NUB-7 [21], and the retinoblastoma cell line Y79 [22], have previously shown MYCN to have 50, 100, and 50 copies, respectively. By cDNA array CGH, the normalized fluorescence intensity ratios for MYCN spots were 6.5, 11.2, and 5.2, for IMR32, NUB-7, and Y79, respectively (not shown). Similarly, MYCN normalized fluorescence intensity ratios for the patient samples were approximately 10% of the corresponding PCR measurement. Overall, there was good correlation between the cDNA array CGH normalized fluorescence intensity ratios, and the copy number determined by PCR or Southern blot (R²=0.71). This analysis showed that with this sensitivity of detection, deletions or copy number gains below the level of 10 to 20 copies cannot be determined with certainty. Most likely, this sensitivity limit arises because of the small size of (intronless) cDNA targets on the arrays leading to basal background hybridization signals that mask the small signals arising from lower copy number imbalances. In contrast, multiple analyses (including use of dye switches) demonstrated that there was an excellent dynamic range at higher MYCN copy number with good reproducibility of copy number estimates, and with ratios that were consistent with high copy number MYCN amplification.

cDNA Array CGH of Neuroblastoma Tissues

In the next series of experiments, amplification analysis using IMR32 metaphase chromosome CGH was compared to the profiles obtained using cDNA array CGH. Chromosome CGH showed two distinct regions of amplification on chromosome 2p (Figure 1A), with the most distal centering on MYCN at 2p24, and the second more proximal location of gene amplification was at 2p15. The analysis of IMR32 by cDNA array CGH also identified these two amplification loci, and more clearly resolved the putative genes subject to amplification. At the 2p24 locus, the MYCN amplification was clearly identified by microarray CGH. In addition, the gene NSE1 was found to be coamplified within 1 Mb telomeric to MYCN in IMR32. Interestingly another gene, DNMT3A, which is situated nearly 10 Mb centromeric to MYCN, was also observed to be amplified. It is noteworthy that CGH applied to metaphase preparations was unable to differentiate the size of the 2p24-p23 amplicon in IMR32 with such precision.

The ratio analysis of the MEIS1 sequences on the cDNA array clearly identified this gene in the cytoband 2p15 to be subject to amplification (Figure 1B). The MEIS1 amplification was recently reported in IMR32 by others using different methodologies [23,24]. The normalized fluorescence intensity ratios of 5.24 and 6.85 from two distinct cDNA sequences on the microarray representing MEIS1 would predict between 50 and 70 copies. Furthermore, the analyses also indicated that a currently unknown sequence (GenBank accession N68286) within 15 kb telomeric to MEIS1 was also coamplified. These novel findings concerning the fine structure of the amplicons mapping to 2p in IMR32 have not been previously reported.

Microarray CGH was performed using genomic DNA derived from nine newly diagnosed neuroblastoma tumor samples. Of the nine studied, one (P1071) was subjected to both chromosome and cDNA array CGH, and MYCN amplification was also confirmed by interphase FISH (Figure 2, A–C). The cDNA array CGH resolved the 2p24 MYCN amplification boundaries to include the genes HPCAL1 and KIAA0188/LPIN1 that map to a 6-Mb flanking region telomeric of MYCN. The copy number gain of 22q shown by chromosome CGH was indicated to be due to amplification of the gene ST13 at 22q13 by cDNA array CGH (Figure 2, A and B, respectively). The same larger MYCN amplicon at 2p24 was also observed in samples from the same patient, P1126 and P1079, but not in any of the other patient samples. In samples P1008 and P1061, the only amplification was MYCN; however, sample P1008 showed additionally amplification of a flanking unknown sequence (accession T79037) which by BLAST sequence analysis is reported to be the gene NAG. The remaining DNA samples did not show detectable amplification of MYCN, or of any other genetic loci. A summary of the amplified genes and sequences identified by cDNA array CGH in neuroblastomas is given in Table 3.

Unsupervised two-dimensional hierarchical clustering analysis was then performed on the cDNA array CGH data as a class discovery procedure for categorizing DNA copy number relationships between the neuroblastomas. The idea of clustering gene expression to detect functional relationships is conceptually similar to analyzing DNA copy number relationships in a group of tumors to empirically delineate classes of sequences subject to amplification. Hierarchical clustering of similarly amplified sequences may suggest close linkage of genes on common amplicon structures. In support of this idea, unsupervised hierarchical clustering was able to provide clear class distinction based on copy number amplification in neuroblastoma. As shown in Figure 3, the majority of primary tumor DNA samples were shown to branch together within the dendogram, away from the cell line IMR32. Within the cluster of primary tumor samples, the MYCN low- or non-amplified samples clustered closely together (D3025, D3060, D3077-95, D3077-97). Samples P1008 and P1061 coclustered separately in this group, likely due to the high copy number of MYCN. Samples from the same patient (P1126, P1079) clustered together, and apart from the other patient samples, presumably due to the higher MYCN amplification and larger MYCN amplicon. The sample P1071 was derived from an undifferentiated neuroblastoma inguinal lymph node metastasis that has additional gene amplifications not present in DNA isolated from the more differentiated samples from this same patient (P1126, P1079).

In summary, hierarchical clustering permitted classification of this sample set of stage IV neuroblastoma patient tumors into three groups based on pattern of copy number changes: 1) no amplifications evident (D3025, D3060, D3077-95, D3077-97); 2) MYCN amplification as the only evident genomic copy number change, or a small MYCN amplicon (P1061, P1008); and 3) a large MYCN amplicon with additional gene amplifications (P1126, P1071, P1079). These findings are summarized in Figure 4.