Supported by: FWO, Grant number: G. 0198.08; Stichting Koningin Wilhelmina Fonds (KWF), Stichting Kindergeneeskundig Kankeronderzoek (SKK), Kinderen Kankervrij (KiKa), European 6th Framework Programme EET pipeline, Stichting tegen Kanker, and the Swedish Childhood Cancer Foundation.
The cell cycle is regulated by cyclin dependent kinases (CDKs) that are activated by intermittently expressed cyclins. Each phase of the cell cycle is more or less controlled by specific CDK/cyclin complexes, though there is redundancy. The cyclin/CDK complexes are inhibited by cyclin dependent kinase inhibitors (CDKIs). Cell cycle deregulation is one of the hallmarks of cancer, and several human tumors are characterized by typical aberrations in cell cycle regulating genes such as deletions of the RB1 gene in retinoblastomas, translocations causing CCND1 (cyclin D1) over expression in mantle cell lymphoma, and CDKN2A (INK4A; p16) deletions in familial melanomas (Sherr, 1996; Sherr and Roberts, 2004; Santamaria et al., 2007).
Neuroblastomas are extra-cranial neuroendocrine tumors that arise from the sympathetic nervous system. The tumor is characterized by a wide diversity of clinical and histopathological presentations. It is one of the rare human malignancies known to demonstrate spontaneous regression with complete disappearance of undifferentiated disseminated tumors, as observed for Stage 4S neuroblastomas. On the other end of the spectrum are the Stage 4 neuroblastoma tumors, often with bone metastasis, that are therapy resistant and very aggressive, resulting in a poor prognosis. Though neuroblastomas have a low incidence, they are the second cause of cancer-related deaths in children. Neuroblastomas can be subdivided in three histological types, i.e., undifferentiated, poorly differentiated, and differentiating neuroblastoma. Prognosis is better in more differentiated tumors (van Noesel and Versteeg, 2004; Maris et al., 2007). Relatively few genes have been implicated in neuroblastoma pathogenesis. MYCN is amplified in 20–30% of neuroblastoma tumors, which strongly correlates with a bad prognosis. MYC oncogene family members are involved in cell growth through protein synthesis, transcriptional regulation of ribosomal RNA processing, cell adhesion, and tumor invasion (Boon et al., 2001; Adhikary and Eilers, 2005). MYCN is also a cofactor of the origin of replication complex and up-regulates expression of many of the proteins of this complex (Koppen et al., 2007; Cole and Cowling, 2008). Besides MYCN, only few other genes have been implicated in neuroblastoma pathogenesis. PHOX2B is mutated in less than 5% of neuroblastoma (van Limpt et al., 2004), while ALK mutations were recently identified in about 7% of neuroblastoma (Mosse et al., 2008). In contrast, chromosomal aberrations of larger regions are frequent in neuroblastoma and lead to recurrent imbalances such as gain of 17q, found in the majority of neuroblastoma, and deletions affecting the chromosome arms 1p and 11q. Homozygous deletions are exceptionally rare, and it has proved to be difficult to pinpoint single genes in the aberrant chromosomal regions that actually contribute to neuroblastoma.
Thus far, no prominent genetic defects have been identified in neuroblastoma that cause cell cycle deregulation, though a low frequency of deletions and amplifications of cell cycle regulating genes have been described (Van Roy et al., 1995; FAn et al., 1999; Molenaar et al., 2003; Caren et al., 2008). In this article, we show by array comparative genomic hybridization (CGH) and single nucleotide polymorphism (SNP)-based whole genome genotyping array analysis of a panel of primary neuroblastoma that about one-third of the tumors carry copy number aberrations of core cell cycle genes. Integrated analysis of genomic data and mRNA expression data revealed that the presence of these genetic aberrations correlated with high expression of E2F target genes that regulate the S and G2/M phase. Tumors with deregulation of E2F target genes have a poor prognosis.
MATERIALS AND METHODS
Patients and Cell Line Samples
The neuroblastic tumor panel used for Affymetrix Microarray analysis contains 88 primary neuroblastoma tumor samples. Ganglioneuroma and ganglioneuroblastoma were excluded from the series. For CGH and SNP array analysis, 82 neuroblastoma tumor samples were used. For six samples, the DNA quality was not good enough for high throughput analysis, or DNA was not available. All tumors analyzed by CGH and SNP array were also included in the mRNA analysis. All samples were derived from primary tumors of untreated patients. Material was obtained during surgery and immediately frozen in liquid nitrogen. The Affymetrix expression data from adult tumors and normal tissues were derived from the Expression Project for Oncology (ExpO) database from the International Genomics Consortium (http://www.intgen.org/expo.cfm).
High-molecular-weight DNA was extracted from tumor tissue using standard procedures. We used a custom 44K Agilent aCGH chip, enriched for critical regions of loss/gain for neuroblastoma (10 kb resolution), miRNAs/T-UCRs (five oligos/gene) and cancer gene census genes (five oligos/gene) (Agilent Technologies). A total of 150 ng of tumor and reference DNA was labeled with Cy3 and Cy5, respectively (BioPrime ArrayCGH Genomic Labeling System, Invitrogen). Further processing was done according to the manufacturer's guidelines. Features were extracted using the feature extraction v10.1.0.0.0 software program. Data were analyzed using the R2 web application (http://R2.humangenetics-amc.nl/). Circular binary segmentation was used for scoring the regions of gain, amplification, and deletion. The coordinates of the first probe sets outside of the gained, amplified, or deleted regions are given to indicate the borders of the genetic aberrations.
Affymetrix Expression Analysis
Total RNA of neuroblastoma tumors was extracted using Trizol reagent (Invitrogen, Carlsbad, CA) according to the manufacturer's protocol. RNA concentration and quality were determined using the RNA 6000 nano assay on the Agilent 2100 Bioanalyzer (Agilent Technologies). Fragmentation of cRNA, hybridization to hg-u133 plus 2.0 microarrays and scanning were performed according to the manufacturers protocol (Affymetrix Inc., Santa Barbara, CA). The expression data were normalized with the MAS5.0 algorithm within the GCOS program of Affymetrix Inc. Target intensity was set to 100. All data were analyzed using the R2 web application, which is freely available at (http://r2.amc.nl).
Tumor DNA was extracted as previously described (Caron et al., 1993), quantified with NanoDrop and the quality was determined by the Abs 260/280 and 230/260 ratio. SNP arrays were processed according to the manufacturer's recommendations with the Infinium II assay on Human370/660-quad arrays containing > 370.000/> 660,000 markers and run on the Illumina Beadstation (Swegene Centre for Integrative Biology, Lund University, SCIBLU, Sweden) according to the manufacturer's recommendations. Raw data were processed using Illumina's BeadStudio software suite (Genotyping module 3.0), producing report files containing normalized intensity data and SNP genotypes. Subsequently, log 2 Ratio and B-allele frequency data were imported into the R2 web application for detailed analysis and comparison with the CGH and expression data.
Aberrations of CCND1-CDK Complex Genes in Neuroblastoma
Cells have to pass the G1 entry checkpoint to irreversibly proceed through the G1 phase of the cell cycle. This checkpoint is tightly controlled by the D type cyclins that activate the kinase activity of CDK4 and CDK6. Activated CDK4/6 phosphorylates the RB1 protein on several residues, resulting in a release and activation of E2F transcription factors. The target genes that are subsequently transcribed by E2F control further progression of the cell cycle.
Deletions affecting the long arm of chromosome 11 mark a subgroup of near diploid tumors which, despite mostly being MYCN nonamplified, show a poor prognosis (Attiyeh et al., 2005). aCGH analysis showed that the CCND1 locus at chromosomal band 11q13 is proximal of these deletions and never lost. Instead, CCND1 is often gained. In our series, we identified copy number gain of the 11q region containing CCND1 in 10 out of 82 neuroblastomas (Fig. 1A and Supporting Information Fig. S1). Moreover, the CCND1 gene sporadically showed high level amplification, which is in line with our previous findings (Molenaar et al., 2003; Michels et al., 2007). In our cohort of 82 neuroblastomas, we identified two primary tumors with high level amplification of the 11q13 region including the CCND1 gene (Supporting Information Fig. S2). This implies a frequency of gain or amplification in the 11q13 region of 15%. The mean mRNA expression level of CCND1 is elevated in neuroblastomas, when compared with normal tissue and even to many adult malignancies (Supporting Information Fig. S3).
Deregulation of the G1 checkpoint can also occur by aberrations of CDK4 or CDK6, as many tumors show amplifications or activating mutations of these CDKs (Schmidt et al., 1994; Kanoe et al., 1998; An et al., 1999). In our series of 82 neuroblastomas, analyzed by array CGH, we identified one tumor with CDK4 amplification. In addition, the 12q14 region encompassing the CDK4 gene was gained in two out of 82 neuroblastomas. In one tumor, this gain occurred only in a small segment of 1.2 Mb including the CDK4 gene (Fig. 1B and Supporting Information Fig. S4). The mRNA expression of CDK4 was high compared with normal tissue and significantly (P = 1.6 × 10−6) correlated with a bad prognosis (Supporting Information Figs. S5 and S6). CGH analysis did not reveal any genetic aberrations of CDK6 (7q21), but CDK6 was highly expressed in neuroblastoma, and this expression relates to an undifferentiated tumor histology (Supporting Information Figs. S7 and S8).
The RB1 and E2F genes are not located in regions with frequent genetic aberrations in neuroblastoma. E2F6 maps to chromosome arm 2p, which is often gained in neuroblastoma, but E2F6 is not a key player in G1 cell cycle regulation (Trimarchi et al., 1998). The E2F transcription factors E2F1 and E2F3 had a moderate high expression compared with normal tissue types, and expression correlated (P = 4.9−5 and P = 2.0−4) with a poor prognosis (Supporting Information Figs. S9 and S10).
Copy Number Aberrations of CDK Inhibitors
The CCND1-CDK4/6 complex can be inhibited by the INK4 group of CDKIs. The best known inhibitor is CDKN2A which is located at chromosome band 9p21 at the same gene locus as p14ARF. CDKN2A has not been reported to harbor single nucleotide mutations in neuroblastoma but sporadic homozygous deletions of CDKN2A are found (Caren et al., 2008). In the 82 neuroblastoma analyzed by CGH, we identified seven tumors with small interstitial 9p21 deletions indicating hemizygous loss of CDKN2A (Fig. 1C and Supporting Information Fig. S11). To confirm these deletions, we used SNP array analysis, which showed that the CDKN2A aberrations affected one allele only. The previously described homozygous deletions of CDKN2A were therefore not found in our cohort of 82 primary neuroblastoma tumors. Analysis of Affymetrix mRNA gene expression data showed that CDKN2A expression is equal or lower compared with normal tissues. A striking exception was formed by the high CDKN2A expression in a neuroblastoma tumor with CDK4 amplification (Supporting Information Fig. S12). Interestingly, CDKN2A over-expression was also observed in glioblastomas with CDK4 amplifications, suggesting that CDK4 amplification leads to up-regulation of CDKN2A expression (Supporting Information Fig. S20). The CDKN2B (INK4B/p15) gene maps to 9p21, 30 kb distal from CDKN2A and is included in the deletions of CDKN2A in our series of 82 neuroblastoma tumors. The CDKN2C (INK4C/p18) gene is located at 1p32, which is outside the smallest region of overlap (SRO) for 1p deletions in this study. Still this region is frequently deleted in neuroblastoma, and the 1p32 region has been suggested to play a role in neuroblastoma development (Hiyama et al., 2001). In our cohort of neuroblastoma, loss of 1p32 correlated with decreased CDKN2C expression (P = 0.03, t-test), suggesting that copy number defects add to the decreased expression of CDKN2C. Finally, the CDKN2D (INK4D/p19) gene is located on 19p13, which did not show genetic aberrations in our panel of 82 neuroblastoma. Interestingly, the mRNA expression analysis of this gene shows a strong inverse correlation with MYCN expression (P = 1.2 × 10−6) (Supporting Information Fig. S13) suggesting an inhibitory effect of MYCN on this CDKI.
We conclude that in neuroblastoma the G1 checkpoint regulating genes CCND1 and CDK4 show occasional high level amplification. Moreover, the CCND1, CDK4, and CDKN2A, B, and C are located in regions that frequently show low copy-number gains or losses. The over-all frequency of genomic aberrations of G1 regulating genes in our series of neuroblastoma was 30% (Fig. 2A). In addition, we identified very high expression of the G1 regulating genes on mRNA level, which was most abundant for CCND1, CDK4, and CDK6.
No Numeric DNA Aberrations but Frequent mRNA Over-Expression of S/G2M Phase Regulatory Genes in Neuroblastoma
S phase progression is controlled by CDK2 which is activated by CCNE (cyclin E) and CCNA (cyclin A). Activated CDK2 causes a further phosphorylation of RB1 to maintain E2F transcriptional activity. Active CDK2 also phosphorylates nucleophosmin and components of the prereplication complex to regulate initiation of DNA replication (Okuda et al., 2000; Coverley et al., 2002; Woo and Poon, 2003). The activated CDK2 kinase complex is inhibited by the CDKN1 family of CDKIs: CDKN1A (Cip1/p21), CDKN1B (Kip1/p27), and CDKN1C (Cip2/p57) (Sherr and Roberts, 1999).
None of the CDK2, CCNA1/2, CCNE1/2, and CDKN1A/B/C genes showed gains, amplifications, or deletions in our 82 neuroblastoma cohort as assayed by array CGH. The CDK2 mRNA expression levels showed a strong correlation with poor prognosis, and this might be in line with the synthetic lethal relation with MYCN that we have reported recently (Supporting Information Fig. S14). From the S phase regulating cyclins, CCNA2 showed increased mRNA expression compared with normal tissue (Supporting Information Fig. S15). CCNA2 is a direct transcriptional target of E2F, and we previously showed that CCNA2 expression strongly decreased after silencing of CCND1 or CDK4 in neuroblastoma (Molenaar et al., 2008). Therefore, the high CCNA2 expression in neuroblastoma most likely reflects high E2F activity.
The CDK inhibitors CDKN1A (Cip1/p21) and CDKN1B (Kip1/p27) are extensively studied in neuroblastoma (Kawamata et al., 1996; Poluha et al., 1996; Bergmann et al., 2001; McKenzie et al., 2003; Nakamura et al., 2003). In our tumor panel, the mRNA expression levels of CDKN1A were not up or down-regulated compared with other tumors, and there was no correlation with clinical or genetic characteristics. Low CDKN1B mRNA expression levels were correlated (P = 1.8−5) with bad prognosis which is in line with previous findings (Supporting Information Fig. S16) (Bergmann et al., 2001). CDKN1B was previously found to be strongly up-regulated during induction of neuronal differentiation and down-regulated by MYCN in a cell line model (Nakamura et al., 2003). These findings were not reflected in our expression profiles of neuroblastomas. CDKN1B mRNA expression did not associate with differentiation or with MYCN expression. Also, the CDKN1C gene has been reported to be repressed by MYCN (Bell et al., 2007). However, also these regulatory relations were not supported by a significant correlation between MYCN amplification and CDKN1C expression in our series.
Progression from G2 to M phase is driven by CDC2 (CDK1) in complex with CCNB1 (cyclin B1) or CCNB2 (cyclin B2)]. Active CDC2 causes phosphorylation of a group of key proteins involved in chromatin condensation, nuclear membrane breakdown, and microtubule and actin reorganization (Lukas et al., 2004; Kops et al., 2005). CDC2 and the regulating B type cyclins did not show aberrations by CGH analysis of our neuroblastoma panel. The mRNA expression levels of these G2M regulating genes were increased compared with normal tissues and comparable with other tumor types (Supporting Information Figs. S17–S19). The expression pattern in the 88 neuroblastoma panel for both CDC2 and the B type Cyclins strongly relates to the expression of CCNA2 (Fig. 2A), which suggests a common regulatory mechanism. We conclude that CCNA2 and the G2 regulating CDC2 and B type cyclins all show increased expression and that these genes have strongly overlapping expression patterns.
Genomic Aberrations of G1 Regulating Genes Correlates with Over-Expression of S/G2M Regulating Genes and a Poor Prognosis
We subsequently correlated the expression data of the core cell cycle regulating genes (Fig. 3) with genetic aberrations and clinical characteristics in neuroblastoma. We first performed K-means clustering of the patients in two groups, based on Affymetrix mRNA expression of the core cell cycle regulating genes in 82 neuroblastomas. Clustering was performed 25 cycles of 10 rounds, and results were consistent in all 25 cycles. Genes were then clustered in a hierarchical tree. CDK1, CCNA2/B1/B2, and E2F1/2/3 showed a striking similarity in expression (Fig. 2A). These genes are all established E2F target genes and are strongly regulated after silencing of the G1 regulating genes CCND1 and CDK4 as we published previously (Molenaar et al., 2008). This suggests that their high expression in neuroblastoma is a direct result of E2F activation.
The clustering of tumors divides the samples in a group with high and a group with low expression of these E2F target genes (Fig. 2A). To relate these data with the genetic aberrations in G1 regulating genes, we added tracks summarizing the genetic defects described in this article. These tracks mark the samples with amplification, gain, or deletions of CCND1, CDK4, CDKN2A/B, and CDKN2C (Fig. 2A). Tumors in the group with high E2F activity showed a higher frequency of genetic aberrations detected in G1 regulating genes (Fig. 2A): 21 out of 51 tumors in the group with high expression of E2F target genes have genetic aberrations, compared with four out of 31 tumors in the group with low expression of E2F target genes (P value 7.0 × 10×3).
To relate this to known clinical and genetic characteristics, we added tracks for MYCN amplification, INSS stage, and outcome. All MYCN amplifications occurred in the cluster containing tumors with high expression of E2F target genes. There was a clear tendency for higher tumor stages in the cluster with high expression of E2F target genes. Most striking was the high frequency of patients with a bad outcome in the group with high E2F activity. This is visualized in Figure 2B, showing Kaplan Meier curves of patients with tumors with high expression of E2F target genes, compared with tumors with low expression of E2F target genes. Strikingly, weak expression of E2F target genes associates with an excellent prognosis.
In this article, we show that copy number aberrations in cell cycle genes in neuroblastoma frequently occur in the group of G1 regulating genes. The tumor-driving genes CCND1 and CDK4 show copy number gains or high level amplifications. Furthermore, three out of four INK4 classes of G1 cell cycle inhibitors are located in genomic regions that frequently showed allelic loss in neuroblastoma. Several of these events have been reported before. We and others have reported sporadic cases of CCND1 amplification (Molenaar et al., 2003; Michels et al., 2007). CDK4 amplifications in neuroblastoma have been identified before but only in established cell lines (Van Roy et al., 1995; FAn et al., 1999). A recent article described hemizygous and sporadically homozygous deletions of CDKN2A (Caren et al., 2008). We did not identify any homozygous deletions in CDKN2A in our series. The hemizygous deletions of CDKN2A/B and CDKN2C and the focal gains of small regions encompassing CCND1 and CDK4 are newly reported findings. Striking is the overall frequency of 30% of tumors with copy number aberrations affecting G1 regulating genes. Integrated analysis of oncogenic pathways by high throughput methods has previously been used to identify pathways with high overall activations in certain tumor types (Jones et al., 2008; Parsons et al., 2008). In neuroblastoma, the G1 cell cycle signal transduction route appears to be involved in a significant subset of the tumors. The frequency of genomic defects could be underestimated, as we did not include genetic aberrations like single nucleotide mutations. Moreover, although we did not analyze protein expression the mRNA expression analysis suggested an even higher frequency of over-expression of the cell cycle regulating genes CCND1, CDK4, and CDK6. It is currently unknown how over-expression of these genes is driven. Aberrant regulation by Micro RNAs (MIRs) or epigenetic events like methylation could be involved in deregulated expression of these genes. The 1p36 located MIR-34A, for instance, has binding sites in the 3′ UTR of the CCND1 and CDK6 mRNA, and this is a candidate tumor suppressor miRNA in neuroblastoma (Sun et al., 2008).
Clustering analysis showed that the genomic aberrations affecting G1 genes correlate with over-expression of E2F target genes involved in S and G2/M phase progression. Although these analyses all relate to mRNA expression, these findings suggests that the genetic aberrations in G1 regulating genes result in an increase in RB1 phosphorylation and a release of RB1 from the E2F transcription factor. As a result, E2F target genes become over-expressed (Fig. 3). We have recently confirmed that this chain of events actually occurs in neuroblastoma cell lines. Silencing of CCND1 and/or CDK4 resulted in decreased expression of the E2F target gene CCNA2 (Molenaar et al., 2008). The other E2F target genes are probably regulated in a similar way, as their expression patterns strongly correlate.
The cluster of tumors with high E2F activity includes all tumors with MYCN amplification, which could indicate a direct involvement of MYCN in cell cycle regulation in neuroblastoma. Most studies on cell cycle regulation by MYC family members have focused on MYC. MYC directly regulates CDK4 and CCND2 at the transcriptional level and inhibits CDKN1A and CDKN2B by direct transcriptional repression (Bouchard et al., 1999; Hermeking et al., 2000; Gartel et al., 2001). For MYCN, these direct transcriptional interactions have not been proven. We observed no correlation between MYCN amplification and expression of CDK4, CCND2, CDKN2B, or CDKN1A in our cohort. Other studies have indirectly shown a relation of MYCN with cell cycle regulation. Over-expression of MYCN in quiescent cells accelerates S-phase progression after mitogenic stimulation, while silencing of MYCN by RNAi causes a decrease of cells in S-phase (Wartiovaara et al., 2002; Woo et al., 2008). Direct MYCN transcriptional target genes in cell cycle regulation however have not been identified. As described above, we did observe an inverse correlation between MYCN expression and CDKN1C and CDKN2D. Whether these correlations reflect a direct or indirect regulation of CDKN1C and CDKN2D by MYCN remains to be studied.
A striking result of our analyses is the very poor prognosis for cases in the cluster with high E2F activity and frequent G1 deregulating events (Fig. 2B). Strong deregulation of G1 cell cycle genes in high risk tumors opens opportunities for targeted drug development. Preclinical and clinical studies with small molecule inhibitors targeting G1 regulating genes though, have been disappointing. Also targeted inhibition of G1 regulating genes in neuroblastoma did not cause tumor death but rather resulted in a cell cycle arrest and neuronal differentiation. As the G1 aberrations lead to high activity of S and G2M phase regulating genes, these might be better drug targets. Our data suggest that CDK2 and CDC2 inhibitors should be evaluated for efficacy in neuroblastoma models.