c-MYC mRNA levels are very low in MYCN amplified tumors (Figure 1), which is due to high MYCN protein repressing c-MYC mRNA expression 20. Previous quantitative PCR analyses in a cohort of 117 neuroblastoma patients revealed that mRNA levels of MYCN are significantly lower in stage 4-NA than in stage 4s-NA (p = 0.008) and localized-NA neuroblastomas (stages 1, 2, 3; p = 0.03) 14. To test whether this lower expression of MYCN in stage 4-NA tumors is due to elevated c-MYC activity that represses MYCN expression, we analyzed c-MYC and MYCN mRNA levels in a cohort of 251 primary neuroblastoma tumors using a customized 11K oligonucleotide microarray (other MYC gene family members were not differently expressed (data not shown)). Although c-MYC mRNA levels were not significantly higher in stage 4-NA (n = 52) than in localized-NA tumors (n = 138), we found an inverse correlation of MYCN and c-MYC expression between stage 4s-NA (n = 30) and stage 4-NA tumors. Stage 4-NA tumors showed lower expression of MYCN and higher expression of c-MYC, whereas stage 4s-NA tumors showed lower expression of c-MYC and higher expression of MYCN (Figure 1; p = 0.008 for c-MYC, p = 0.07 for MYCN).

Because increased activity of MYCN in stage 4s-NA or c-MYC in stage 4-NA tumors should both result in high expression of shared target genes compared to localized-NA neuroblastomas, we analyzed known direct MYCN/c-MYC target genes, namely MDM2 21, DKC1 22, and PTMA 23, in neuroblastoma subtypes. As expected, the highest expression of all three transcripts was observed in MYCN amplified tumors (Figure 1; p < 0.001 for all three transcripts, n = 31). MDM2 mRNA levels were higher in stage 4-NA (p = 0.005) and stage 4s-NA (p = 0.03) than in localized-NA tumors (the expression range of MDM2 is large because of two MYCN amplified tumors with non-syntenic co-amplification of MDM2 (data not shown)). Similarly, DKC1 and PTMA expression was higher in stage 4-NA (p < 0.001 for DKC1, p = 0.02 for PTMA) and in stage 4s-NA (p = 0.03 for DKC1, p = 0.007 for PTMA) than in localized-NA tumors. These results suggest an increased MYCN/c-MYC activity also in stage 4s-NA (MYCN) and in stage 4-NA (predominantly c-MYC) compared to localized-NA tumors. However, higher DKC1 mRNA levels in stage 4-NA tumors and higher PTMA mRNA levels in stage 4s-NA tumors also suggest differential regulation of MYCN/c-MYC target genes in these subtypes. To further analyze MYCN/c-MYC activity as well as differential regulation of MYCN/c-MYC target genes in neuroblastoma subtypes, we thought to define a comprehensive set of target genes directly regulated by MYCN and/or c-MYC in neuroblastoma cells.

To identify MYCN/c-MYC-regulated genes in neuroblastoma cells, we employed the experimental system SH-EP^MYCN, which stably expresses a tetracycline-regulated MYCN transgene 23. Exponentially growing SH-EP^MYCN cells cultured with tetracycline express c-MYC but almost no MYCN protein (Figure 2a). Induction of MYCN by removing tetracycline from the medium is associated with a rapid reduction of c-MYC at the mRNA and protein levels. c-MYC reduction occurs prior to the full expression of ectopically induced MYCN protein (Figure 2a). Accordingly, mRNA levels of direct MYCN/c-MYC targets, such as PTMA and DKC1, initially decline before accumulating MYCN protein leads to the re-induction of these genes. Similar profiles were observed with direct MYCN target genes, such as MDM2 and MCM7 (Additional data file 1).

We used SH-EP^MYCN cells for a global search of MYCN and c-MYC target genes in neuroblastoma cells using a customized neuroblastoma oligonucleotide microarray (11K, Agilent) that was enriched with probes for genes differentially expressed in neuroblastoma subtypes and for direct MYCN/c-MYC target genes 1424. Gene expression profiles of SH-EP^MYCN cells at 2, 4, 8, 12, 24, and 48 hours after targeted MYCN expression were generated. Self-organizing maps (SOMs) were used to capture the predominant pattern of gene expression. This analysis yielded 504 clusters (best matching units (BMUs)) consisting, on average, of 20 clones per cluster (Additional data file 1). We searched for clusters with characteristic gene expression profiles of direct MYCN/c-MYC target genes. In addition, known c-MYC target genes from a public database 25 and known MYCN target genes from a literature search were mapped to the 504 clusters (Additional data file 2). A significant enrichment of known MYCN/c-MYC targets was found in 6 clusters (clusters 140, 168, 195, 280, 308, and 336; p < 0.05, adjusted for multiple testing), consisting of 167 genes. The genes in these six clusters were induced by MYCN and c-MYC in SH-EP^MYCN cells. Based on their average gene expression profiles, we grouped the clusters into two subgroups, I and II. Subgroup I genes (clusters 140, 168, and 195) were expressed at equal levels in SH-EP^MYCN cells expressing endogenous c-MYC (2 hours) and in those fully expressing ectopic MYCN (24 and 48 hours), despite the fact that the maximum protein level of MYCN was significantly higher than that of endogenous c-MYC (Figure 2a; Additional data file 1). This indicates that subgroup I genes are regulated by MYCN, and also suggests that they are less responsive to MYCN than to c-MYC in SH-EP^MYCN cells. The mRNA levels of subgroup II genes (clusters 280, 308, and 336) were highest in SH-EP^MYCN cells fully expressing ectopic MYCN and followed the combined absolute c-MYC and MYCN protein levels during the time course experiment. We also found clusters with MYCN and c-MYC repressed genes (for example, subgroup III; Additional data file 1). However, enrichment of known MYCN/c-MYC repressed genes from the literature/database in defined clusters was not found using our statistical cut-off (after adjustment for multiple testing, no cluster showed p < 0.05). This was at least partly due to the fact that in SH-EP^MYCN cells, some genes were repressed by MYCN but not by c-MYC (subgroup IV). In addition, c-MYC repressed genes from different experimental systems compiled in the c-MYC target gene database were not necessarily repressed by MYCN and/or c-MYC in SH-EP^MYCN cells.

Therefore, we focused on genes for further validation that were induced by both MYCN and c-MYC proteins in SH-EP^MYCN cells and grouped into subgroup I and II. We extracted all available promoters from the genes represented on the array and scanned for canonical E-boxes (CACGTG) and for the 12 bp MYCN position-weight matrix 26 within -2 kb and +2 kb of the transcriptional start site. We ranked all 504 clusters according to the relative number of putative MYCN/c-MYC binding sites in each cluster. All clusters from subgroups I and II were among the 15 top-ranked clusters with enrichment of predicted MYCN/c-MYC binding sites (data not shown).

To further validate target gene regulation by MYCN/c-MYC in neuroblastoma cells, we performed ChIP-chip using a 244K oligonucleotide promoter microarray (Agilent). We analyzed the binding of MYCN and c-MYC to the promoters of the 147 subgroup I and II genes that were represented on the 244K promoter microarray. We used five neuroblastoma cell lines that either preferentially express high levels of MYCN (SH-EP^MYCN, IMR5/75 (approximately 75 copies of MYCN), and Kelly (approximately 100-120 copies of MYCN)) or c-MYC (SJ-NB12 and SY5Y). Additionally, as an intermediate type, parental SH-EP cells were analyzed, which preferentially express c-MYC, but also MYCN at low level 2023. ChIP-chip experiments were performed with a monoclonal antibody against human MYCN and a polyclonal antibody against human c-MYC for each of the neuroblastoma cell lines. A cut-off for positive binding was defined as >4-fold enrichment for one probe together with >2-fold enrichment for at least one of the two neighboring probes compared to input control. In addition, we manually inspected each of the MYCN and c-MYC-binding profiles from the 147 genes. Seven genes were excluded from the analysis because the probe sets for the genes mapped within the genes but outside the target gene promoter regions (all profiles for Kelly and SJ-NB12 cell lines are given in Additional data files 3 and 4, respectively; MYCN- and c-MYC-binding results are given in Additional data files 5-7). We also performed PCR-based ChIP for selected candidate genes (n = 13; Additional data file 8), which all showed analogous results to ChIP-chip (data not shown). Almost all 140 target gene promoters showed binding of MYCN and/or c-MYC in the six analyzed neuroblastoma cell lines as measured by ChIP-chip (Figure 2b). Intriguingly, hierarchical clustering of neuroblastoma cell lines according to the MYCN/c-MYC-binding pattern clearly separated MYCN- and c-MYC-expressing neuroblastoma cell lines. Differentiation between MYCN and c-MYC binding was mainly achieved through the monoclonal anti-MYCN antibody. The polyclonal antibody against c-MYC also gave positive binding signals for a large set of target gene promoters in neuroblastoma cell lines with high MYCN that lack detectable c-MYC expression (SH-EP^MYCN, IMR5/75 and Kelly). This was most likely due to unspecific binding of the polyclonal c-MYC antibody to MYCN in these cells. Nevertheless, the lack of binding of MYCN to a large set of target gene promoters in the c-MYC-expressing cells, SJ-NB12 and SY5Y, and the positive binding of c-MYC to almost all of these target gene promoters in these cells allowed the distinction between MYCN and c-MYC. Taken together, these results indicate that the genes from subgroups I and II represent a core set of target genes directly regulated by either MYCN or c-MYC in neuroblastoma cells dependent on which MYC protein is expressed.

To determine transcriptional activity of MYCN/c-MYC proteins in primary neuroblastomas (n = 251), we analyzed differential expression of subgroup I and II genes in neuroblastoma subtypes using the Global test as proposed by Goeman et al. 27. Almost all these genes (154 of 167; 92%) showed highest expression in MYCN amplified tumors, suggesting that regulation of these genes by MYCN is similar in neuroblastoma cell lines and tumors. Compared to localized-NA tumors (stages 1, 2, 3), expression of subgroup I and II genes was significantly associated with stage 4s-NA (p = 0.002), stage 4-NA (p < 0.001) and MYCN amplified tumors (p < 0.001). Global test results further indicated that an increasing number of MYCN/c-MYC target genes was induced from stage 4s-NA through stage 4-NA to MYCN amplified tumors (Additional data files 9-11). To further illustrate this, we grouped each of the 154 genes into one of four classes based on pair-wise comparisons (Mann-Whitney test, cut-off p < 0.05). These were, compared to localized-NA tumors: overexpressed in MYCN amplified and in stage 4s-NA tumors (class 1); overexpressed in MYCN amplified, stage 4-NA and stage 4s-NA tumors (class 2); overexpressed in MYCN amplified tumors (class 3); overexpressed in MYCN amplified and stage 4-NA tumors (class 4) (Figure 3). Compared to localized-NA tumors, 25 (16%) of the 154 MYCN/c-MYC target genes, including CCT4, FBL, MDM2, NCL, NPM1, PTMA, and TP53, were expressed at higher levels in stage 4s-NA tumors (Table 1). Eighty-eight (57%) of the 154 MYCN/c-MYC target genes, including 21 of those overexpressed also in stage 4s-NA tumors, were expressed at higher levels in stage 4-NA than in localized-NA tumors (Table 1, class 2; Additional data file 5). Accordingly, stage 4-NA tumors shared overexpression of 68 of 154 direct MYCN/c-MYC target genes (44%), including AHCY, RUVBL1, PHB, CDK4, and MRPL3, with MYCN amplified tumors. Together, this indicates that besides MYCN amplified tumors, stage 4-NA tumors, and to a lesser extent stage 4s-NA tumors, also show higher MYCN/c-MYC activity compared to localized-NA tumors. In line with this, we also found lower mRNA levels of an increasing number of MYCN/c-MYC repressed genes from stage 4s-NA (10 out of 68 (15%) in vitro validated repressed genes that are also lower in MYCN amplified tumors) through stage 4-NA (34 out of 68 (50%)) to MYCN amplified tumors (68 out of 102 in vitro validated repressed genes had the lowest expression levels in MYCN amplified tumors (67%)). Based on the relative expression of MYCN and c-MYC in neuroblastoma subtypes, we propose that elevated MYCN activity in stage 4s-NA tumors induces only a restricted set of MYCN/c-MYC target genes, whereas elevated c-MYC activity in stage 4-NA tumors induces a larger set of MYCN/c-MYC target genes.

Having shown that MYCN/c-MYC target gene activation is also associated with distinct neuroblastoma subtypes, we wanted to test whether MYCN/c-MYC activity as determined by the expression levels of their target genes is associated with overall survival and improves outcome prediction independent of known risk markers. We used the Global test to test the influence of each of the 504 experimentally defined gene clusters on overall survival directly, without the intermediary of single gene testing. The p-values for each cluster were adjusted for multiple testing and ranked according to their association with overall survival. Table 2 gives the association with overall survival of the six MYCN/c-MYC target gene clusters and the rank in relation to all other clusters. In a separate analysis, we determined the association with overall survival for each of the 504 experimental gene clusters adjusted for amplified MYCN, stage 4 versus stages 1, 2, 3, and 4s, and age at diagnosis ≥1.5 years (Table 2). These well-established risk markers highly correlated with poor outcome in univariate analyses (p < 0.001 for each of these three markers). As expected, the Global test without adjustment for co-variables indicated that all MYCN/c-MYC target gene clusters were significantly associated with poor overall survival (p < 0.001). Intriguingly, all six MYCN/c-MYC target gene clusters remained significantly associated with overall survival after adjusting for amplified MYCN, stage 4 versus stages 1, 2, 3, and 4s, and age at diagnosis ≥1.5 years. Of note, two of the MYCN/c-MYC target gene clusters (clusters 168 and 140, both from subgroup I showing a higher responsiveness to c-MYC than to MYCN in SH-EP^MYCN) revealed the strongest association with overall survival of all 504 clusters after adjusting for co-variables (Table 2). Figure 4 shows the association with overall survival for each gene from cluster 168 with and without adjustment for co-variables. Most of the genes within this cluster, such as AHCY, ARD1A, CDK4, HSPD1, PHB, RUVBL1, and TRAP1, remained associated with overall survival after adjustment for co-variables. A less significant association with overall survival was observed for clusters with MYCN/c-MYC repressed genes: clusters 454, 482, 484, and 486 were associated with poor overall survival without adjustment for co-variables in the Global test (p < 0.001, adjusted for multiple testing), but they showed no significant association with poor overall survival when adjusting for the co-variables amplified MYCN, stage 4 versus stages 1, 2, 3, and 4s, and age at diagnosis ≥1.5 years. We also asked whether direct MYCN/c-MYC target genes as defined by our analyses are represented in previously published gene expression-based classifiers that distinguish low-risk from high-risk neuroblastomas independent of other risk markers. Gene lists from these studies hardly overlapped, making interpretation difficult. The overlap with our MYCN/c-MYC target gene list was defined by using the gene names as common identifiers. Indeed, different genes defined by our study as direct MYCN/c-MYC target genes were represented in the gene expression classifier gene lists: from the 44 genes overexpressed in high-risk neuroblastomas independent of other markers described by Schramm et al. 28, we identified 10 genes directly regulated by MYCN/c-MYC (DDX21, SCL25A3, EIFA4A2, NME1, NME2, TKT, LDHA, LDHB, HSPD1, HSPCB); from the 20 genes overexpressed in high-risk neuroblastomas independent of other markers described by Ohira et al. 29, we identified 5 genes directly regulated by MYCN/c-MYC (EEF1G, AHCY, TP53, ENO1, TKT); and from the 66 genes overexpressed in high-risk neuroblastomas independent of other markers described by Oberthuer et al. 24, we identified 7 genes directly regulated by MYCN/c-MYC (PRDX4, MRPL3, SNRPE, FBL, LOC200916, PAICS, AHCY; Figure 5). Together, these results show that MYCN/c-MYC activity as determined by the expression status of a subset of MYCN/c-MYC target genes is significantly associated with poor overall survival independent of other established markers and is a consistent element of gene expression-based neuroblastoma risk classification systems.

All patients from this study (n = 251) were enrolled in the German Neuroblastoma Trials NB90-NB2004 with informed consent and diagnosed between 1989 and 2004 (patient characteristics are in Additional data files 2 and 12). Tumor samples were collected prior to any cytoreductive treatment. The only criterion for patient selection was availability of sufficient amounts of tumor material. Tumor specimens were checked for at least 60% tumor content.

Gene expression profiles from the tumors were generated as dye-flipped dual-color replicates using customized 11K oligonucleotide microarrays as previously described 24. The 11K Agilent microarray was constructed in our laboratory based on extensive neuroblastoma transcriptome information from different whole-genome analyses from primary tumors and neuroblastoma cell lines. These also include comparative transcriptome analysis of MYCN amplified versus not amplified tumors as well as of neuroblastoma cell lines with variable/conditional MYCN/c-MYC expression that allowed the enrichment with MYCN/c-MYC-regulated genes 1424 (unpublished data). The reference for each tumor RNA was an RNA pool of 100 neuroblastoma tumor samples. Data normalization and quality control is described in Additional data file 2. All raw and normalized microarray data are available at the ArrayExpress database (Accession: E-TABM-38) 36.

The SH-EP^MYCN cell line, previously also denoted as TET21N 23, expressing a MYCN transgene under the control of a tetracycline-repressible element was used to generate gene expression profiles from different time points after MYCN induction showing variable MYCN and c-MYC levels. RNA isolation from SH-EP^MYCN cells was performed as previously described 14. Gene expression profiles were generated as dye-flipped dual-color replicates using the same customized 11K oligonucleotide microarray platform used for the tumor samples. The reference for RNA from SH-EP^MYCN cells after MYCN induction was RNA from SH-EP^MYCN cells cultured in parallel that lack MYCN expression. Gene expression profiles from SH-EP^MYCN cells with variable MYCN and c-MYC levels were taken for a SOM analysis (Additional data file 2). Protein expression was assessed by immunoblotting using 50 μg of total cell lysates from the cell line experiments as previously described 37. Blots were probed with antibodies directed against MYCN (SantaCruz, sc-53993, Santa Cruz, CA, USA) and c-MYC (SantaCruz, sc-764, Santa Cruz, CA, USA).

Chromatin immunoprecipitation was performed as described previously 3839 using 10 μg of MYCN (SantaCruz, sc-53993), c-MYC (SantaCruz, sc-764) 4041 and normal mouse IgG (SantaCruz, sc-2025) antibodies and Dynabeads ProteinG (Invitrogen, Carlsbad, CA, USA). Eluted and purified MYCN-ChIP-DNA (1 μl) of IMR5/75 and SH-EP^MYCN was used as a template in PCR reactions running for 35 cycles. The primer sequences are given in Additional data file 8. In addition, ChIP-DNA templates from SH-EP^MYCN, SH-EP, Kelly, IMR5/75, SJNB-12 and SY5Y cells using MYCN and c-MYC antibodies were amplified for DNA microarray analysis (Agilent Human Promoter ChIP-chip Set 244K) using the WGA (Sigma-Aldrich, St. Louis, MO, USA) method 42. DNA labeling, array hybridization and measurement were performed according to Agilent mammalian ChIP-chip protocols. For the visualization of ChIP-chip results, the cureos package v0.2 for R was used (available upon request). The in silico promoter analysis for the identification of putative MYC binding sites (canonical and non-canonical E-boxes) is described in Additional data file 2.

Differential gene expression of MYCN/c-MYC and their target genes in neuroblastoma tumors was evaluated for stage 4s-NA, stage 4-NA and MYCN amplified using localized-NA tumors (stages 1, 2, 3) as reference using Goeman's Global test and the Wilcoxon rank sum test. A result was judged as 'statistically significant' at a p-value of 0.05 or smaller. Differential expression of MYCN was evaluated in two partially overlapping cohorts, one measured by quantitative PCR 14 and the other by oligo microarray (the overlap was 101 patients). To test the association of MYCN in vitro clusters with overall survival (death due to neuroblastoma disease), Goeman's Global test was used 27. To evaluate the influence of gene expression on outcome independent of established markers, the Global test was adjusted for the following co-variables: genomic MYCN status, stage of the disease (stage 4 versus stages 1, 2, 3, and 4s), and age at diagnosis (≥1.5 years versus <1.5 years). Because of multiple testing of probably dependent gene clusters, p-values were adjusted according to Benjamini and Yekutieli 43 to control the false discovery rate of 5%.