Neuroblastoma (NB) is a rare but often highly aggressive tumour in children. Despite intensive gene copy number and mRNA expression studies thus far only two genes, namely MYCN 1and PHOX2B 234 have been found to be directly implicated in NB development. In order to provide clues for more effective therapies, insights into the molecular pathogenesis of this tumour are urgently needed. Analyses of recurrent patterns of somatically acquired DNA copy number alterations resulted in the delineation of three major genetic subgroups with predictive tumour behaviour (subtype 1, 2A and 2B), of which subtype 2A NB represents an aggressive subgroup of metastatic NB 56 characterised by loss of 11q, gain of 17q and a normal MYCN copy number status 6.

Deletion of the long arm of chromosome 11 is found in 15–22% of sporadic NB 5678910 and has also been described in constitutional cases of NB 1112, suggesting the presence of one or more tumour suppressor gene(s) on chromosome 11. Functional evidence for a NB tumour suppressor gene at 11q came from microcell mediated chromosome transfer (MMCT) experiments, in which transfer of an intact chromosome 11 in a NB cell line with 11q loss resulted in a more differentiated phenotype 13. Array comparative genomic hybridisation (array CGH) analysis of these MMCT hybrids revealed 11q25 as a plausible location for a NB differentiation gene 14. Extensive microsatellite heterozygosity mapping studies however point at various critical regions of loss, located at 11q23.3 7 and within the chromosomal region 11q14-11q23 15. Despite these mapping efforts, no genes with proven tumour suppressor activity in NB have been identified thus far.

In this study, we used an integrated approach combining high resolution copy number profiling and gene expression meta-analysis. This approach identified CADM1 as a strong candidate 11q tumour suppressor gene with prognostic power, which possibly exerts its effect through haplo-insufficiency.

In an attempt to more accurately map 11q losses and to identify potential homozygous losses that were previously missed using low resolution analyses, we performed high resolution BAC array CGH copy number profiling on 100 primary NB tumours (16 stage 1, 12 stage 2, 21 stage 3, 40 stage 4 and 11 stage 4S tumours) and 29 cell lines using a 1 Mb BAC array supplemented with a chromosome 11q tiling path (position 63–72 Mb (11q13) and 84–134 Mb (11q14.1-qter)). Three patterns of chromosome 11 loss could be observed. Twenty-four (24%) tumours showed loss of the entire chromosome 11. These tumours consisted of predominantly low stage tumours with an overall pattern of numerical imbalances. In addition, also 3 cell lines (10%) showed whole chromosome 11 loss. Seventeen tumours (17%) and 16 cell lines (55%) showed partial 11q loss. Most deletions were large encompassing the distal end of 11q, in keeping with data from literature 678910153435. In cell lines CLB-GA, SMS-KAN, and NB-5, small interstitial losses were detected (respectively minimal 42.90, 30.84, and 4.76 Mb in size). To our knowledge this is the first report of interstitial 11q loss in SMS-KAN and NB-5 cell lines. No homozygous 11q losses were observed.

Two common minimal regions of loss (shortest region of overlap, SRO) at 11q23 could be delineated (Figure 1). The telomeric SRO (SRO1, 3.54–3.72 Mb, 11q23.2-q23.3) is delineated by the breakpoints in N206 (centromeric border) and SMS-KAN (telomeric border) and contains 31 genes. More centromeric, a second SRO (SRO2, 1.65–1.79 Mb, 11q23.1-q23.2) is delineated by the breakpoints in STA-NB3 (centromeric border) and NB-5 (telomeric border) and contains 12 genes. Recent evidence indicates that also miRNAs can function as tumour suppressor genes. According to the miRBase, 3637 (August 2007, version 10.0) no miRNAs were however located within the defined SROs. The finding that the most distal SRO partially overlaps with the SRO defined by Guo et al. 7 and both SROs are positioned within the critical region defined by Maris et al. 15 validates our mapping efforts. Nevertheless, it is important to note that SRO mapping should always be interpreted with care as individual observations can dramatically reduce the size or location of the SRO. This is especially the case when including cell line data, in which alterations could be induced during culturing. In this respect, it should therefore be noted that NB cell lines have been proven to be very stable over time and that, although rare, interstitial 11q loss has also been described in primary NB tumours 671538, further underscoring the relevance of our defined SRO regions.

To prioritize potential NB 11q candidate genes within the two critical regions, we performed a meta-analysis on publicly available gene expression data of 692 tumours from seven independent studies 20212223242526. An overview of the obtained results is given in Additional file 2. In a first analysis we determined which genes in our SROs have prognostic power using Cox regression analysis. In total, 24 genes were prognostically significant in one or more studies (see Additional file 2), of which four (CADM1, NCAM1, CD3E, ZNF259) were significant in three or more studies. In the four studies for which DNA copy number data were available, we determined in a second analysis which genes could discriminate between the various NB subgroups (subtype 1, 2A, 2B) (Kruskal-Wallis test). This multi-group analysis was specifically applied instead of a binary comparison between 11q normal versus 11q deleted samples in order to avoid combining tumours with a distinct biology and prognosis. Thirteen genes had discriminative power, of which five were significant in at least two studies (CADM1, NCAM1, BACE1, ZNF259, KIAA0999). Genes that were prognostic significant in more than three independent studies and had subgroup discriminating power in more than two independent studies were considered as strong candidate genes (CADM1, NCAM1 and ZNF259).

ZNF259 is a zinc finger protein involved in normal cell cycle progression 39. ZNF259 has been suggested as a modifier gene in spinal muscular atrophy 40, but no direct link with carcinogenesis has been established thus far. NCAM1 is a neural cell adhesion molecule with an important role in tumour migration and metastasis. In a rat model it was demonstrated that NCAM transfected glioma cells show a decreased invasion compared to low NCAM expressing control cells 41. Similarly, loss of NCAM in a transgenic mouse model of pancreatic β cell carcinogenesis caused the formation of lymph node metastasis, whereas NCAM overexpression prevented tumour metastasis 42. In several tumour types, including colon carcinoma 43, gastrointestinal neoplasia 44 and astrocytic tumours 45, downregulation of NCAM expression and an associated poor survival has been described. In chemoresistant NB cells an enhanced invasive capacity was observed due to downregulation of NCAM adhesion receptors 46. CADM1 encodes a cellular adhesion molecule with a role in synaptic formation of neural cells 47. This gene, also known as IGSF4 (immunoglobulin superfamily, member 4) or TSLC1 (tumour suppressor in lung cancer-1), is a bona fide tumour suppressor gene in non-small lung cancer and was identified by functional complementation of a frequently deleted region in nude mice 48. Reduced or lost expression has been described in a variety of human cancers and in vivo suppression of tumorigenesis after restoration of expression has been demonstrated in esophageal cancer, cervical cancer and prostate cancer 49505152.

To select the most promising candidate gene within the three remaining candidate genes, we integrated the gene expression profile of normal NB precursor cells (see Additional file 2). The transcriptome information of these normal fetal neuroblasts represents a unique resource for identifying genes contributing to NB development and the malignant phenotype 21. Remarkably CADM1 was the only gene in our candidate list that was differentially expressed in NB versus neuroblasts using the Rank Product algorithm. A significantly lower expression of CADM1 was observed when comparing unfavourable NB to neuroblasts 21. This is in agreement with the Kruskall-Wallis data showing CADM1 downregulation in unfavourable NB compared to favourable subtype 1 NB. Consequently, we consider CADM1 to be a compelling candidate NB tumour suppressor gene for further analysis.

In several cancer types, two-hit inactivation of CADM1 occurs by LOH and hypermethylation of the promoter region, whereas CADM1 mutations are rare. Indeed, it was shown that only two CADM1 sequence variants could be picked up in a panel of 31 NB cell lines 53, while in a study of 25 NB with proven CADM1 allelic loss no mutations were found in the remaining allele 54. Epigenetic modification has been described for several genes in NB 5556. To study if the remaining allele is subject to methylation and resulting epigenetic silencing, we measured CADM1 mRNA expression in a panel of 22 NB cell lines before and after exposure to the demethylating agent 5-aza-2'-deoxycytidine (DAC). Four out of 13 cell lines with loss at the CADM1 locus showed a more than two fold upregulation of CADM1 mRNA expression after DAC treatment. This was also the case for two of the nine cell lines with normal CADM1 copy number. Representative examples of mRNA expression before and after treatment with DAC are given in Figure 2. As the maximal observed expression upregulation was 4.4 fold in cell line CHP-134, these data were inconclusive to predict the involvement of methylation associated CADM1 silencing in NB. Consequently, we performed bisulphite sequencing of twenty promoter region CpG dinucleotides in a panel of 31 NB cell lines (including the cell lines that were studied in the DAC experiment). Note, that the investigated CpG sites also cover the 93 bp consensus CpG region described by Kuramochi et al. 48. Methylation of all studied CpG sites was identified in 12.5, 20 and 11% of analysed clones in the cell lines NGP, SH-EP and SK-N-BE(2c). It should be noted that NGP was also present in the panel of cell lines treated with DAC, where a more than two fold upregulation was observed after DAC treatment. However, in all other cell lines only sporadic CpG methylation was detected at low frequency (< 30%). Based on this evidence and taken in account that cell lines typically show a higher extent of CpG island methylation compared to primary NB, epigenetic CADM1 inactivation probably does not play a major role in NB. The latter could explain the observed limited fold changes in the DAC re-expression data.

While CADM1 forms an excellent positional and functional candidate tumour suppressor gene for NB, we can not exclude a role for the 2 other identified candidates suppressor genes ZNF259 and NCAM1. Based on our own methylation and previously published mutation data, CADM1 is most likely not inactivated through a two-hit Knudson mechanism. However, novel concepts in tumour suppressor genetics such as haploinsufficiency currently challenge the two-hit paradigm and are gaining importance in cancer genetics. Hence, an intriguing possibility is sensitivity of CADM1 to gene dosage. In this respect, the observed reduced expression of CADM1 in unfavourable NB as compared to normal neuroblasts is interesting. CADM1 belongs to the family of immunoglobulin cell adhesion molecules and encodes a transmembrane glycoprotein involved in cell-cell interactions. As disruption or altering of cell-cell interactions is an important feature of tumour invasion and metastasis, it could therefore be hypothesised that one working CADM1 allele is insufficient to accomplish the normal gene function. This would make NB cells more aggressive and amenable for migration, invasion and metastasis formation. Furthermore, although the biological cascade of the CADM1 protein is not yet elucidated, it is interesting to note that EPB41L3 – an established binding partner of CADM1 57 – has been described as higher expressed in mass screening NB tumours versus aggressive NB 58 and favourable NB versus unfavourable NB 21. In 4/5 data sets, for which expression data on CADM1 and EPB41L3 were available, also a significant positive correlation between CADM1 and EPB41L3 mRNA expression was found (p < 0.05). Further research should therefore be performed to elucidate the involvement of CADM1 in NB.

Tumorigenesis is a multistep process characterised by a multitude of genetic and epigenetic alterations. In view of the limited success of classical genomic approaches towards the identification of NB tumour suppressor genes, we integrated high resolution array CGH with transcriptome analysis and identified CADM1 as a candidate 11q tumour suppressor gene with prognostic power. CADM1 does not seem to be inactivated through a two-hit mechanism in NB. However, due to the frequent involvement of CADM1 in human tumorigenesis and its presumed role in invasion, a hallmark of malignancy, further studies should be performed to confirm CADM1 involvement in NB and elucidate its mode of action.

The authors declare that they have no competing interests.

EM carried out the genomic studies and drafted the manuscript. JH performed the methylation study. KDP carried out the meta-analysis of neuroblastoma mRNA expression data. GL, AS, AE and BB provided tumour material. JV and FS designed and coordinated the study, and edited the manuscript. All authors have reviewed the manuscript and FS, JV and ADP were the final editors of the manuscript.