Research Article Integrative Genomics Identifies Distinct Molecular Classes of Neuroblastoma and Shows That Multiple Genes Are Targeted by Regional Alterations in DNA Copy Number Qun Wang,1 Sharon Diskin,1,4 Eric Rappaport,1 Edward Attiyeh,1 Yael Mosse,1 Daniel Shue,1 Eric Seiser,1 Jayanti Jagannathan,1 Suzanne Shusterman,5 Manisha Bansal,1 Deepa Khazi,1 Cynthia Winter,1 Erin Okawa,1 Gregory Grant,4 Avital Cnaan,2 Huaqing Zhao,2 Nai-Kong Cheung,6 William Gerald,6 Wendy London,7 Katherine K. Matthay,8 Garrett M. Brodeur,1 and John M. Maris1,3 Divisions of 1Oncology and 2Biostatistics, Children’s Hospital of Philadelphia and Department of Pediatrics and 3Abramson Family Cancer Research Institute, University of Pennsylvania School of Medicine; 4Center for Bioinformatics, University of Pennsylvania, Philadelphia, Pennsylvania; 5Department of Pediatric Oncology, Dana-Farber Cancer Institute and Department of Pediatrics, Harvard Medical School, Boston, Massachusetts; 6Memorial Sloan-Kettering Cancer Center, New York, New York; 7Department of Statistics, University of Florida and Children’s Oncology Group, Gainesville, Florida; and 8University of California at San Francisco School of Medicine, San Francisco, California Abstract also differentially expressed in the independent validation data set, providing a prioritized list of candidate neuro- Neuroblastoma is remarkable for its clinical heterogeneity blastoma suppressor genes. Taken together, these data are and is characterized by genomic alterations that are strongly consistent with the hypotheses that the neuroblastoma tran- correlated with tumor behavior. The specific genes that scriptome is a sensitive marker of underlying tumor biology influence neuroblastoma biology and are targeted by genomic and that chromosomal deletion events in this cancer likely alterations remain largely unknown. We quantified mRNA target multiple genes through alteration in mRNA dosage. expression in a highly annotated series of 101 prospectively Lead positional candidates for neuroblastoma suppressor collected diagnostic neuroblastoma primary tumors using an genes can be inferred from these data, but the potential oligonucleotide-based microarray. Genomic copy number multiplicity of transcripts involved has significant implica- status at the prognostically relevant loci 1p36, 2p24 (MYCN), tions for ongoing gene discovery strategies. 11q23, and 17q23 was determined by PCR and was aberrant in (Cancer Res 2006; 66(12): 6050-62) 26, 20, 40, and 38 cases, respectively. In addition, 72 diagnostic neuroblastoma primary tumors assayed in a different labora- tory were used as an independent validation set. Unsupervised Introduction hierarchical clustering showed that gene expression was Neuroblastoma is an important childhood cancer, as it accounts highly correlated with genomic alterations and clinical for f15% of all pediatric oncology deaths (1). The clinical hallmark markers of tumor behavior. The vast majority of samples of neuroblastoma is heterogeneity, with the likelihood of tumor MYCN with amplification and 1p36 loss of heterozygosity progression varying widely according to age and disease burden at (LOH) clustered together on a terminal node of the sample diagnosis. A large body of data support the hypothesis that the dendrogram, whereas the majority of samples with 11q clinical behavior of human neuroblastoma may be reliably deletion clustered separately and both of these were largely predicted based on the analysis of a panel of prognostic variables distinct from the copy number neutral group of tumors. Genes (reviewed in ref. 2). The Children’s Oncology Group (COG) currently involved in neurodevelopment were broadly overrepresented stratifies patients into low-, intermediate-, or high-risk categories in the more benign tumors, whereas genes involved in RNA based on analysis of well-defined prognostic factors, including processing and cellular proliferation were highly represented patient age at diagnosis (3), International Neuroblastoma Staging in the most malignant cases. By combining transcriptomic System (INSS) stage (4), tumor histopathology (5), DNA index (6), and genomic data, we showed that LOH at 1p and 11q was and MYCN amplification status (7, 8). This stratification system is associated with significantly decreased expression of 122 currently used to assign therapeutic intensity, mandating that (61%) and 88 (27%) of the genes mapping to 1p35-36 and the algorithm be as precise as possible. Clinical experience with all of 11q, respectively, suggesting that multiple genes may be this system suggests that the algorithm is useful, but misclassifi- targeted by LOH events. A total of 71of the 1p35-36genes were cations almost certainly occur resulting in overtreatment or undertreatment (2). Additional tumor-specific prognostic markers may be required to achieve maximal predictive power. A large number of genomic aberrations have been defined in Note: Supplementary data for this article are available at Cancer Research Online (http://cancerres.aacrjournals.org/). neuroblastoma, and the pattern of these somatically acquired Presented in part at the 96th Annual Meeting of the AACR, Anaheim, CA, April 16- changes correlate with tumor behavior (9). Amplification of 20, 2005 and the 11th Conference of the Advances in Neuroblastoma, Genoa, Italy, June 16-19, 2004. the MYCN oncogene at 2p24 provides a paradigm for the clinical Requests for reprints: John M. Maris, Division of Oncology, Children’s Hospital of utility of a tumor-specific DNA rearrangement. MYCN amplification Philadelphia, 34th Street and Civic Center Boulevard, Philadelphia, PA 19104-4399. is present in f20% of newly diagnosed tumors and remains a Phone: 215-590-2821; Fax: 215-590-3770; E-mail: [email protected]. I2006 American Association for Cancer Research. powerful biomarker for an aggressive phenotype and poor survival doi:10.1158/0008-5472.CAN-05-4618 probability even in light of modern dose-intensive therapies Cancer Res 2006; 66: (12). June 15, 2006 6050 www.aacrjournals.org Downloaded from cancerres.aacrjournals.org on September 27, 2021. © 2006 American Association for Cancer Research. Integrative Genomics of Neuroblastoma (10, 11). Hemizygous deletions and/or loss of heterozygosity (LOH) assessed using Affymetrix GeneChip Test 3 arrays. Statistical modeling of have been defined at multiple genomic regions, including probe set behavior was conducted using Probe Profiler (Corimbia, Berkeley, chromosomal arms 1p, 3p, 4p, 11q, 14q, 16p, and 19q at frequencies CA) as described by the manufacturer. The software implements a model- between 20% and 45% of cases using a variety of techniques based approach for data extraction from probe sets and incorporates quality-control features to detect and correct for the contribution of (2, 12). Likewise, unbalanced gain of 17q material may be the most nonspecific cross-hybridization effects. The model weights probe pairs common genomic aberration in neuroblastoma primary tumors, as based on consistency of performance. Normalization to remove chip- it is present in 50% to 75% of cases. Perhaps most importantly, to-chip variation is accomplished by subtracting the second percentile these events are nonrandom and are clinically relevant, with LOH intensity value and dividing by the interquartile range. A quantitative at 1p36 and 11q being independently prognostic for disease expression score (e-score) is then calculated for each probe set. outcome (13). Despite the clear clinical relevance of these genomic Data from this experiment, including sample annotations, have been aberrations, no definitive neuroblastoma suppressor gene has been deposited in Gene Expression Omnibus database and are available for free identified. at http://www.ncbi.nlm.nih.gov/geo/. Genome-scale profiling technology allows for detailed reassess- Unsupervised clustering. Unsupervised two-way agglomerative hierar- ment of the relationship between DNA aberrations with regional chical clustering with Pearson’s correlation coefficient as a distance metric gene expression (14, 15). Microarray studies in neuroblastoma to was done using GeneSpring 7.0 (Silicon Genetics, Redwood City, CA). Probe sets with low average expression or low variance were excluded in date have mainly focused on defining prognostic signatures or an effort to minimize noise. Specifically, probe sets exhibiting an average identifying genes associated with oncogenic MYCN (16–18). In e-score <50 or an estimated SD <75 were excluded, leaving 2,177 probe sets addition, McArdle et al. correlated global gene expression profiles remaining for cluster analysis. with 11q deletions in a limited number of samples but showed Identification of enriched gene sets. Gene Ontology (GO) annotation the power of this combined approach for gene discovery (19). We was used to characterize the functional relationship of differentially therefore sought to determine regional and global gene expression expressed gene classes between the varying sample classes (23). EASE differences between a large number of samples with and without version 2.0 was selected to examine the GO categories of molecular clinically relevant chromosomal aberrations in primary neuroblas- function, biological process, and cellular component (24). EASE functions tomas. These data were used to further our understanding of the to determine the overrepresentation of GO terms within a given gene list. genetic basis for neuroblastoma heterogeneity and to prioritize For each of the three GO categories, EASE establishes