Genome-wide Array CGH Analysis of Murine Neuroblastoma Reveals Distinct Genomic Aberrations which Parallel those in Human Tumors

INTRODUCTION

Neuroblastoma, a tumor of sympathetic peripheral nervous tissue, is the third most common tumor of childhood, accounting for 15% of cancer-related deaths (1) . Amplification of the MYCN gene is the only consistent genetic mutation described to date in neuroblastoma, occurring in ∼30% of children with this disease and correlating with advanced stage and poor outcome (2) . Numerous regions of loss of heterozygosity have been identified in tumors (reviewed in Ref. 1 ). CGH 5 has validated loss of chromosome regions 1p36, 3p, 4p, 9p, 11q, and 14q and frequent gains of chromosomes 17 or 17q and 18q (3) . CGH has narrowed the regions of interest on these chromosomes to <25 Mb in some cases and implicated additional gains of chromosomes 1, 2, 6, 7, 12, and 13 (4, 5, 6, 7, 8, 9, 10, 11) . The best characterized regions of copy number abnormalities include loss of 1p36 in 19% of cases (12) and gain of 17q, which is also the most frequent genetic event, occurring in ≤83% of all tumors (3) and >90% of high-grade tumors (13) . Loss of 1p36, gain of 17q, and amplification of MYCN show significant interaction and correlate with advanced tumors and poor outcome (3 , 14) .

Overexpression of the human MYCN oncogene driven by the rat tyrosine hydroxylase promoter causes tumors in transgenic mice, recapitulating several histological and pathological aspects of clinical neuroblastoma (15) . Tumors from these mice localize to paraspinous ganglia and adrenal glands, stain for neuronal markers, show ultrastructural features of neurons (neurosecretory granules and synapses), and share with the human tumors a propensity to grow along radicular nerves. Previous genome-wide screens in these mice revealed a range of genetic aberrations. Chromosome CGH showed losses of mouse chromosomes 5, 9, 16, and X and gains of chromosomes 3, 11, and 17 (15) . A low-resolution AI screen showed frequent imbalance on chromosomes 1, 3, 10, 11, 14, and 18 (16) .

Array CGH based on hybridization of tumor DNA to genomic BACs provides a way to analyze genomic changes in tumor tissue with high sensitivity, detecting single-copy number aberrations in tissues in the presence of ≤60% contamination by normal tissue (17, 18, 19, 20) . Here, we used this technology to refine the genetic characterization of 40 murine neuroblastoma tumors, using an array comprised of 1056 BAC clones providing 5-Mb average resolution across the genome, and sub-Mb resolution in regions identified in previous genetic screens of transgenic tumor models (20) . We have validated specific observations from our array CGH screen using FISH and through AI analyses. These studies revealed a wide range of genetic copy number changes in murine tumors, including localized gains of a segment of mouse chromosome 11 orthologous to human chromosome 17q, which narrows the region of clinical interest in human neuroblastoma to 18.5 Mb. In addition, this study identified genetic changes in other regions implicated in human disease and other animal models, suggesting a conserved interaction among the mouse orthologues of human chromosomes 3p, 4p, and 11q, which interact in advanced cases of human neuroblastoma.

MATERIALS AND METHODS

Tumor Generation and DNA Isolation.

Tumors were dissected from a cross of transgenic C57BL/6J to wild-type Mus musculus castaneus Ei mice. All mice were obtained from The Jackson Laboratory (Bar Harbor, ME). Genomic DNA was isolated from frozen primary neuroblastomas by overnight digestion [50 mm Tris (pH 8.0), 1 mm EDTA (pH 8.0), 0.5% SDS, and 500 μg/ml Proteinase K; Roche, Indianapolis, IN)] followed by phenol-chloroform extraction and ethanol precipitation. DNA concentrations were quantified using a fluorometer (Amersham, Piscataway, NJ).

Array Preparation.

BACs described previously (20) , and additional BACs isolated from the RPCI-23 mouse BAC library (21) , were organized into 96-well plates. Each BAC contained a known sequence-tagged site marker, gene, or BAC end sequence, which was used to place it on the assembled mouse genome sequence using batch BLAT (22) . All genomic coordinates described in this study are based on the February 2002 freeze of the assembled mouse genome available through the UCSC Genome Browser. 6 A detailed list of the BAC clones used for these arrays, their associated markers, and genomic coordinates are also available online (see supplementary data). Single colonies were isolated from each BAC clone on LB plates containing 20 μg/ml chloramphenicol. A single colony per clone was arrayed into 96-well format, and DNA was prepared with automated instrumentation (Autogen960; Holliston, MA) using the standard BAC preparation protocol, typically yielding ∼200 ng of BAC DNA, which was resuspended to a final concentration of ∼5–10 ng/μl. Two microliters of BAC DNA were used for amplification by DOP-PCR (23) . Printing onto 3D-Link activated slides (Motorola Life Sciences, Pasadena, CA) was performed as described. Clones were printed in quadruplicate (two spatially separated duplicates). Two arrays were printed per slide, each 12 mm × 12 mm in size.

Array Hybridization.

Array CGH was performed essentially as described (20) . Briefly, 1 μg of DpnII-digested tumor DNA and reference DNA (derived from the spleen, brain, or heart of the same animal as the tumor) was labeled with Cy3 or Cy5, respectively (Amersham, Piscataway, NJ) using the Bioprime random-primed labeling kit (Invitrogen, Carlsbad, CA). Unincorporated nucleotides were removed using Sephadex G-50 columns (Amersham, Piscataway, NJ), and tumor and reference samples were combined, precipitated with 50 μg of mouse Cot-1 DNA, and resuspended in 60 μl of hybridization solution (10% dextran sulfate, 50% formamide, 2 × SSC, 4% SDS, and 10 μg/μl yeast tRNA). Preannealed DNA samples were then applied to arrays fitted with 2.4 cm² EasiSeal gaskets (Continental Lab Products, San Diego, CA). Slides were placed in individual, prewarmed hybridization chambers (plastic slide holders) containing Whatmann paper soaked with 750 μl of 2 × SSC, 0.1% SDS to prevent evaporation and hybridized for a minimum of 36 h at 37°C on a slowly rocking platform. After hybridization, slides were washed, counterstained with 4′,6-diamidino-2-phenylindole, and imaged using a custom-built CCD camera (18) . TIFF images were segmented, and spot statistics were calculated using custom software (24) .

Analysis of array CGH profiles and determination of gain and loss thresholds were performed as described (20) . Genomic profiles of log₂ fluorescence ratios from each array CGH analysis were normalized to a median 0, where the median was computed over all clones on the array with the exception of the localized regions of high-density clone coverage on chromosomes 4, 6, 9, and 16. Clones mapping to these regions contained about half of the clones present on the array and were excluded because localized changes in these regions would violate the assumption for normalization to the median that either: (a) the majority of the clones do not change, or (b) the gain and loss occur at the same frequency (25) . A mixture of three Gaussian distributions (corresponding to the normal, gained, and deleted portions of the tumor genome) was fit to a histogram of the ratios for each tumor using a hybrid least-squares/maximum likelihood method using the Solver tool in Microsoft Excel. Only 2 of the 40 tumors (5%) showed SDs of the central peak > 0.15. Clones with mean ratios falling 3 SD above or below the mean of the central (i.e. normal) distribution were scored as gained or lost, respectively.

Statistical Analysis.

Clones scored present in <70% of tumors were filtered out for whole-chromosome analysis. This filtering removed ∼13% of clones, leaving 845 with confirmed map positions on the February 2002 genome assembly. For whole-chromosome association analysis, a chromosome was scored as gained or lost if: (a) 90% of all clones on the chromosome had a log₂ ratio greater or less than 0, respectively; and (b) the median of the log₂ ratios of all clones on that chromosome was greater or less than 0.1 or −0.1, respectively. The main requirement that ≥90% of clones had to be greater or less than 0 may be justified by the following calculation: assuming that all of the clones on a given chromosome are at the same copy number and there are >10 clones/chromosome, we wanted to test a null hypothesis of diploid copy number or that the probability of log₂ ratio being greater or less than 0 is 0.5. Using a normal approximation for comparing two binomial proportions (26) , we calculated the probability of observing ≥90% of the clones being greater or less than 0 under the null hypothesis to be <0.005. Because we test 20 chromosomes in 40 samples, we expect to observe four false positives or fewer in the whole data set (<0.005 × 20 × 40). The requirement of minimum median aberration of log₂ 0.1 is an ad-hoc one to make this procedure even more conservative and allows detection of changes present in only a small but non-negligible subset of genetically nonhomogenous cells (as the decrease of magnitude of log₂ ratios with the proportion of cells with a given aberration has been investigated previously; Ref. 20 ), e.g., assuming 20% admixture of the normal cells, the log₂ ratio of 0.1 corresponds to ∼20% of the tumor cells gaining a copy of the chromosome. We then manually examined the chromosomes assigned “change” status on which not all clones were above or <0 and eliminated three cases where a small contiguous part of chromosome was not aberrant, i.e., the log₂ ratios in that region randomly fluctuated ∼0. These cases had passed 90% criterion because of the uneven coverage of the chromosomes on the array. We further analyzed only those chromosomes that were gained or lost in at least four samples. Thus, chromosomes 2, 4, 7, 10, 12, 13, and 19 were excluded from further analysis. All but three (15, 18, and X) of the remaining 13 chromosomes showed gains or losses exclusively. On 15, 18, and X, we recoded the less frequent events as “no change” (these events were singular except for X, which was gained in two cases and lost in seven). Remaining values were then coded as “1” for change and “0” for no change.

We then performed two-dimensional clustering of sample profiles (based on individual clones) and chromosomes (based on scores for whole-chromosome change). The chromosomes were clustered using agglomerative hierarchical average clustering with a binary metric in the “MVA” package in the R environment (27) , which can be freely downloaded from the Internet. 7 The chromosomes were coded with 0s and 1s where “1” indicated that chromosome was gained or lost in a given sample and “0” otherwise. The dissimilarity between a pair of chromosomes was defined using a binary metric, i.e., as proportion of samples in which only one of the chromosomes showed change among the samples in which at least one did. The samples were then ordered with agglomerative hierarchical average clustering using (1-pearson) correlation as a metric. Here, the log₂ ratios for the clones mapped to the chromosomes in the clustering diagram were used to compute the distance matrix for the samples (28) .

Interactions were assessed by performing one-sided Fisher exact tests for concordance on pairs of chromosomes using their binary codes for “change” versus “no change.” Interaction between chromosomes 1 and 3 was not adjusted for multiple comparisons, as a previous study had suggested that they acted together. For all other pairs of chromosomes, scores were adjusted for multiple comparisons. The correction was done using a minP multiple testing procedure (as described in Ref. 29 ), which provides strong control of the family-wise Type I error rate. The permutation procedure was performed by randomly shuffling the chromosomal scores across samples and recomputing the Fisher exact P for each pair of chromosomes. For three-way interactions, a similar test was done, in which chromosome pairs were recoded as “1” if both changed and “0” otherwise, and each pair was then compared with the remaining chromosomes. If the chromosomal interactions we observed were simply caused by global chromosome instability, then the likelihood that a specific chromosome would show gain or loss should be identical for all chromosomes. Alternatively, if such chromosomal interactions were the result of specific selection, then the likelihood that a specific chromosome would show gain or loss should be different for specific chromosomes. To address whether the chromosomes that changed in more than four tumors had a homogeneous rate of change, we applied a test for homogeneity of binomial samples to such chromosomes (27) . We obtained a P < 0.0002, thereby refuting the hypothesis of global instability in the absence of selection.

AI Analysis.

Tumor DNA from 32 mice included in the array CGH screen was screened for AI using SSLP microsatellite markers across chromosomes 1, 3, 10, 11, 14, and 18, with additional control markers on chromosomes 2 and 13 (see supplementary data). PCR was run as described (16) . Intensity of PCR bands was quantified using an ABI 377 fluorescence-based DNA sequencer (UCSF Genomics Core Facility) and analyzed using Genescan software (PE Biosystems, Foster City, CA). Ratios of band intensities for the two alleles of each marker were calculated and compared with the average ratio of 10 spleen samples for the same marker. Ratios of single markers in tumors with either <0.65 or >1.5 times the 10-spleen average were scored as imbalanced (30 , 31) .

FISH Analysis.

BAC DNA (Research Genetics, Carlsbad, CA) from clones RP23-101G23, RP23-426C22 (chromosome 4), RP23-16A3 (chromosome 18), RP23-18M23, and RP23-301L12 (controls) was prepped using the Plasmid Mini kit (Roche, Indianapolis, IN). Probes were prepared from BAC DNA using the Vysis nick translation kit according to the manufacturer’s protocol (Vysis, Inc., Downers Grove, IL). The NMYC transgene was detected using the Vysis human NMYC cDNA probe labeled with digoxigenin and detected with fluorescein antidigoxigenin (Roche, Indianapolis, IN). Probes were hybridized to primary fibroblast cultures and sections prepared from frozen normal and tumor tissues. Primary fibroblasts were isolated from freshly cut mouse tails and cultured and harvested using standard cytogenetic techniques. Frozen section slides were incubated in 75 mm KCl for 20 min, then Carnoy solution (3:1 methanol:acetic acid) for another 20 min, and dried at room temperature. The frozen section slides, and slides derived from the fibroblast cultures were then heated for 10 min at 90°C, incubated in Pepsin at 37°C for 30–60 s, washed in 70% ethanol for 30 s, and incubated in 2 × SSC at 37°C for 60 min. Slides were dried using an ethanol series (70%, 85%, 1 min each, and 100% for 10 min) followed by jet air drying. Slides were then incubated in 70% formamide at 70°C for 1 min and dried again through the same ethanol series (70%, 85%, 100%, and 1 min each) and jet dried. Probes were then denatured, added to slides, and incubated under a sealed coverslip for ≥12 h at 37°C. After hybridization, coverslips were removed, and slides were washed in 0.4 × SSC at 70°C for 2 min, rinsed in room temperature with 2 × SSC/0.1% NP40, jet air dried, and counterstained. Cells were viewed using an epifluorescence microscope equipped with Texas red/FITC/4′,6-diamidino-2-phenylindole filters.

RESULTS

Array CGH in Tumors from TH-MYCN Transgenic Mice.

Mice transgenic for MYCN were back-crossed serially to nontransgenic mice of strain C57BL/6J. Starting at the F7 back-cross into C57BL/6J, transgenic females were crossed to castaneus males, because in our hands, castaneus females are poor mothers. Twenty-three tumors described previously (16) were included in this analysis, with back-cross ranging from F7 to F10. An additional 17 animals added to this study were derived from back-crossed mice ranging from F10 to F15. Mice were euthanized when tumors were visible or when mice were symptomatic (32) . The histological characteristics of tumors in these animals were identical to those described previously.

Individual tumor hybridizations illustrated a range of genetic aberrations; combinations of whole chromosome gains and losses, localized gains and losses, and two tumors showing no identifiable genetic changes. Fig. 1A ⇓ shows a tumor with whole-chromosome gains of chromosome 1 and 3. The higher ratios shown by clones on chromosome 3 relative to clones on chromosome 1 suggest that a single gain of chromosome 3 is more frequent in cells within the tumor population or that cells tended to have more copies of chromosome 3 relative to chromosome 1. Fig. 1B ⇓ shows a tumor with gains of chromosomes 1, 3, 6, 11, and 17, localized gain on chromosome 8, loss of chromosomes 5, 9, and 16, and localized loss of distal chromosome 4. Fig. 1C ⇓ shows localized gain on chromosome 8, amplification on distal chromosome 18, and an interstitial deletion on chromosome 16.

We used FISH to verify loss of distal chromosome 4 in the tumor analyzed in Fig. 1B ⇓ using a BAC probe located within the region of loss observed by array CGH (Fig. 2A) ⇓ . Similarly, Fig. 2B ⇓ confirms the amplification on chromosome 18 using a BAC corresponding to the peak of the amplicon in the tumor profiled in Fig. 1C ⇓ . An ongoing modifier analysis suggested to us that distal chromosome 18 was the site of TH-MYCN transgene integration. 8 Two-color FISH using a probe for the TH-MYCN transgene and the probe from Fig. 2B ⇓ verified association of these two markers in interphase nuclei (Fig. 2C) ⇓ and metaphase spreads (Fig. 2D) ⇓ in fibroblasts from mice transgenic for TH-MYCN. To further define the linkage between the MYCN transgene and amplicon on chromosome 18, we performed two-color FISH against the tumor analyzed in Fig. 1C ⇓ . This analysis revealed coamplification of MYCN and distal chromosome 18, as show in Fig. 2E ⇓ .

A summary of array CGH analysis of 40 tumors is shown in Fig. 2F ⇓ . Gains and losses for each arrayed clone were determined as described previously (20) , revealing whole-chromosome gains of chromosomes 1 (25%), 3 (41%), 11 (29%), 14 (20%), 17 (15%), and 18 (17%). Whole chromosome losses were observed for chromosomes 5 (11%), 9 (10%), and 16 (14%). We also observed lower frequency gain of chromosomes 6 (12%), 8 (11%), 2 (6%), and 12 (7%), and losses of 15 (6%) and 19 (5%) were also observed (Table 1) ⇓ . Frequencies of chromosomal change were significantly nonhomogeneous (see “Materials and Methods”), suggesting differentially acting selective forces. Selected clones on chromosomes 1, 11, and 18 showed recurrent subchromosomal gain (Table 2) ⇓ .

AI Analysis Validates Observations by Array CGH.

We had reported previously an AI screen using 46 markers across the 20 mouse chromosomes in tumors from this model (16) . This analysis revealed imbalance on chromosomes 1, 3, 10, 11, 14, and 18 and showed an absolute association between imbalance of chromosomes 1 and 10 and a relative association between imbalance of chromosomes 1, 3, and 10. Most of the changes were imbalances of whole chromosomes.

In the current study, we further screened 96 markers for AI across chromosomes 1, 3, 10, 11, 14, and 18 (with control markers on chromosomes 2 and 13) on 32 tumors (see “Materials and Methods” for details). This study confirmed significant imbalance of markers on chromosomes 1 (highest frequency of imbalance 28%), 3 (38%), 11 (22%), 14 (31%), and 18 (41%). Average frequencies for all markers across each chromosome were: chromosome 1 (19%), 3 (29%), 11 (22%), 14 (32%), and 18 (40%; See supplementary data). The AI analysis revealed whole-chromosome copy number changes in most cases. All of these changes corresponded to chromosome gains according to array CGH (Fig. 1D) ⇓ . Consecutive, localized markers showing highest levels of AI on chromosomes 1 and 18 fell within 10 Mb of clones most frequently gained in the array CGH screen (data not shown).

Tumors Showing Partial Gain of Chromosome 11 Delineate a Minimal 15-Mb Region Orthologous to Human Chromosome 17q.

Array CGH showed whole-chromosome gain of chromosome 11 in ∼30% of tumors. Selected tumors gained only the distal half of this chromosome, which is orthologous to the region of human chromosome arm 17q implicated in 83% of human neuroblastoma (3) . Fig. 3 ⇓ illustrates the profile of chromosome 11 in three informative tumors. Horizontal dotted lines in each profile represent the cutoff of 3 SDs from the central distribution of all clones in each tumor, the threshold used to identify gain and loss of single clones. Vertical solid lines indicate the minimal gained region. This 15-Mb region of distal chromosome 11 corresponds to a segment of human chromosome 17q from 62.2–80.7 Mb. 6 Human orthology was determined by homology of genes flanking the region in mouse, Psmc5 on the proximal end (PSMC5 in human), and MCT4 on the distal end (SLC16A3 in human). Importantly, regions immediately proximal to this position in mice correspond to a more proximal region of chromosome 17q, outside of the minimal region implicated in human disease (Fig. 3D) ⇓ .

Cluster Analysis of Array CGH Profiles.

Using hierarchical clustering algorithms (28) , we clustered the array CGH profiles from 40 tumors to reveal recurrent patterns of whole-chromosome copy number changes (Fig. 4A) ⇓ . We confirmed an interaction observed previously between gains of chromosomes 1 and 3 (P ≤ 0.006). We also identified novel significant pair wise interactions between losses of chromosomes 5, 9, and 16 (P ≤ 0.001 for each possible pair) and of chromosome 15 with all elements of this pair (with either 9 or 16, P ≤ 0.003; with 5, P ≤0.02). Gain of chromosome 11 also interacts with this set (P ≤ 0.026, ≤ 0.051, and ≤0.137 for chromosomes 9, 5, and 16, respectively), as well as with gain of chromosome 1 (P ≤ 0.108). Ps were adjusted for multiple comparisons as described (29) , and all comparisons with an adjusted P < 0.5 are shown in Table 3 ⇓ . Two-way interactions are represented graphically in Fig. 4B ⇓ . To assess three chromosome interactions, we treated each significant pair wise interaction as a single event and compared the pair with other individual chromosomes. A highly significant complex emerged of all combinations of losses of chromosomes 5, 9, and 16 and gain of chromosome 11 (P ≤ 0.001 for comparisons involving each chromosome) and 15 with all pairs in this group (P ≤ 0.007). A separate complex, involving the gain of combinations of chromosomes 1, 3, and 6, showed less statistical significance after multiple test correction (P ≤ 0.1; see Table 4 ⇓ for all interactions with Ps < 0.5). Interestingly, gain of chromosome 11 participates in both complexes. Attempts to cluster genetic profile with latency and location of tumor (abdominal, thoracic, and neck) demonstrated no significant association. Histological analysis of tumors revealed a narrow range of morphology and did not show significant variability to associate with genetic profiles.

DISCUSSION

Array CGH Analysis of Murine Neuroblastoma Confirms Regions of AI.

Although neuroblastoma accounts for 15% of cancer-related deaths in childhood, relatively little is known about the genetic complexity of this disease. We therefore used the best characterized genetic mutation in this malignancy to model neuroblastoma in transgenic mice. Because targeted expression of MYCN to the peripheral neural crest leads to neuroblastoma, we hypothesize that mice transgenic for TH-MYCN provide a powerful tool to identify spontaneously arising mutations that contribute to the pathogenesis of these tumors. In addition, with the completed sequences of the human and mouse genomes (33, 34, 35) , analysis of mouse tumors may narrow large regions of change in human tumors through analysis of orthology between the two genomes.

Our array CGH analysis confirmed all of the changes observed through AI screening of a subset of the same tumors, with the exception of a localized imbalance on chromosome 10 reported previously (16) . Analysis of 14 markers along chromosome 10 showed no significant imbalance apart from the D10MIT44 marker used in our original screen, which showed a unique localized imbalance in a significant number of tumors and an absolute association with imbalance of chromosome 1. This marker has been subsequently remapped to chromosome 1, 9 thereby explaining the obligatory association of imbalances on chromosomes 1 and 10 in this tumor set. One other marker on chromosome 10, D10Mit136, showed AI in 22% of samples, although markers <5 Mb on either side showed AI frequencies of <4%. Markers on chromosomes 1, 3, 11, 14, and 18 demonstrated significant imbalance, in agreement with the previous study and with trends observed in the array CGH analysis. In addition, the array CGH analysis confirmed previous observations using chromosome CGH of a different set of 21 tumors, showing similar trends of losses of mouse chromosomes 5, 9, 16, and X and gains of chromosomes 3, 11, and 17 (15) .

Patterns of Localized Genetic Change in Murine Neuroblastoma Correlate with Changes Observed in Human Tumors and Other Mouse Models.

Amplification of MYCN is specifically associated with loss of 1p36 and gain of 17q in human neuroblastoma and is inversely associated with loss of 11q (12, 13, 14 , 36) . Our study of the TH-MYCN model has allowed us to further refine the locus of genes on human chromosome 17q likely to contribute to the genesis of neuroblastoma. The mouse orthologue of human chromosome 17 is entirely contained on mouse chromosome 11, 10 which was gained in 30% of murine tumors. Furthermore, a 15-Mb region of chromosome 11 was gained with even higher frequency and corresponds to a region from 62.2 to 80.7 Mb on human chromosome 17q (Fig. 3) ⇓ . Human tumors show frequent translocation of 17q23.1-qter, which corresponds to 56–80 Mb on the human physical map (13) . The human–mouse orthology thus narrows this region to the distal 18.5 Mb of chromosome 17q. Roughly 220 RefSeq genes map to this region in mice. 6 The 10-Mb peak of single clones showing the highest ratios in all three tumors includes ∼37 RefSeq genes, including Tlk 2, a serine-threonine kinase involved in regulation of chromatin assembly (37) . Importantly, Tlk 2 has been identified as the closest gene distal to the breakpoint of a constitutional translocation in a patient with human neuroblastoma (38) . This region also contains several cell adhesion genes, including Pecam, involved in angiogenesis (39) ; three G-proteins, Rgs9, Gna13, and Axin2, which interacts with the Mapt and Wnt3 genes located proximally in the region (40 , 41) ; and protein kinase C (Prkca). All of these genes are potential oncogenes in this region.

The mouse orthologue of human 1p36 includes the distal end of chromosome 4. We observed FISH-verified loss of a large portion of distal chromosome 4 in one tumor (Fig. 1B) ⇓ . In addition, one BAC, mapping to a region containing the human p73 gene as verified by PCR, was lost in 13% of tumors, although this could not be verified by FISH (data not shown). Further investigation of this locus will help clarify its potential role in progression of MYCN-induced murine neuroblastoma.

Murine tumors showed simultaneous losses of chromosomes 5, 9, and 16 (P ≤ 0.001). These three chromosomes share orthology with portions of human chromosome arms 3p, 4p, and 11q. Combined loss of human chromosomes 3p and 4p is common in stage 4 tumors, which also have 11q loss (P = 1.10⁻⁶ and 0.016, respectively; Refs. 3 and 9 ), suggesting that a multichromosome interaction is conserved between species. Although loss of 11q reportedly occurs infrequently in human tumors with MYCN amplification, the smallest region of overlap in human tumors, from 11q14 to 11q23 (42) , shares orthology with mouse chromosome 9. In addition, the smallest region of overlap from human chromosome 3pter-24 10 includes regions of mouse chromosomes 9 and 16. Gain of mouse chromosome 11, which parallels the most frequently observed genetic event in human neuroblastoma (7 , 13) , was significantly associated with loss of chromosomes 5, 9, and 16 (P ≤ 0.001).

Losses of chromosomes 9 and 16 have also been observed in the RIP-Tag-induced mouse islet carcinoma model, suggesting a role for common tumor suppressors in distinct mouse models (20 , 31) . Peaks of frequent deletion in murine neuroblastoma localized to within 1 Mb of the most common regions of loss on chromosomes 9 and 16 in RIP-Tag-initiated tumors, regions that were represented by very high density of clones on the array. These regions, from 101 to 102 Mb on chromosome 9 and from 36 to 37 Mb on chromosome 16, both map to human chromosome 3q. The region on chromosome 9 includes the gene Nck1, which has been shown to interact with the TrkB gene (43) . Expression of TrkB is associated with poor prognosis in human neuroblastoma (44) . The region on chromosome 16 includes the mouse homologues of EAF1, an ELL-associated factor implicated in leukemia (45) , and HCLS1, a gene down-regulated in ovarian cancers (46) . More proximal regions of chromosome 16 map to chromosome 16p in humans, implicated in familial neuroblastoma (47) .

Mouse chromosomes 1 and 3 share orthology with human chromosome arm 1q, which is frequently gained in advanced human neuroblastoma (48) . Similarly, mouse chromosome 6, gained in 12% of tumors, has significant orthology with human chromosomes 7 and 12, gained in 50 and 15% of human cases, respectively (3) . Human neuroblastomas also show gains of 18q in 30% of cases (3) . This chromosome arm shares orthology with mouse chromosomes 1 and 18, both of which were gained in our study.

FISH validation of patterns seen by array CGH confirmed that the region of gain on chromosome 18 represents the site of integration of the TH-MYCN transgene (Fig. 2) ⇓ . The frequent gain of chromosome 18 in this model is consistent with data published previously demonstrating that doubling the gene dosage for the MYCN transgene leads to more aggressive malignancy (15 , 49) . We observed gains of chromosome 12, which contains the endogenous mouse Nmyc gene, in only 5% of samples. Amplification of endogenous Nmyc may not be necessary for tumorigenesis, however, given the presence of the overexpressed MYCN transgene.

Murine tumors initiated by the TH-MYCN transgene display a wide range of genetic aberrations with similarity to human neuroblastoma. Although the similarities between murine and human neuroblastoma suggest mutation in orthologous genes, specific validation to clarify the relevance of murine neuroblastoma to human disease awaits localization and identification of specific genes mutated in tumors from both species. Focus on chromosomes showing orthologous aberration in human tumors, particularly when paired with higher resolution human mouse genome orthology maps, should better characterize genes in candidate loci. Additional studies of subchromosomal gains and losses, and the interactions of loci on multiple chromosomes, should facilitate the identification of genetic pathways undiscovered previously in the progression of clinical disease.