DNA Microarrays Reveal Relationship of Ewing Family Tumors to Both Endothelial and Fetal Neural Crest-Derived Cells and Define Novel Targets

The histogenetic origin of the Ewing family tumors (EFTs) stem cell has been a matter of debate since the original description by Ewing in 1921 (1). Whereas Ewing originally described the tumor as “diffuse endothelioma of the bone,” more recent experimental evidence suggests that EFT cells are capable of neuroectodermal differentiation (2, 3). EFTs are defined by expression of EWS-ETS fusion proteins, which are involved in the transformation process (4). Detection of the corresponding chimeric RNAs by reverse transcription-polymerase chain reaction (RT-PCR) can be used for molecular diagnosis of EFTs (5). In the huge majority of EFTs, reciprocal translocations of the EWS gene on 22q12 with FLI1 or ERG located on 11q24 or 21q22 are detectable. In contrast to FLI1 and ERG (both members of the ETS family), EWS is a ubiquitously expressed gene. The products of EWS-ETS translocations are aberrant transcription factors that are capable of transforming normal fibroblasts in vitro. In addition to the deregulated expression of ETS family members due to translocation to the EWS promoter, the fusion transcripts have a higher transactivating potential than wild-type ETS components (reviewed in ref. 6). Wild-type FLI1 plays a role during vascular development and in lymphoid function (reviewed in ref. 7). Interestingly, the avian FLI1 homologue is specifically expressed in a subset of neural crest cells giving rise to mesenchyme (8).

Although more than half of the patients with localized manifestation of the disease can be cured with current multimodality regimens, cytotoxic chemotherapy/radiotherapy has still not yielded satisfactory results for patients with the most advanced form of the disease (6). New therapeutic strategies are required for the treatment of these patients (9, 10). EWS-ETS fusion proteins are highly specific for EFT and represent candidate therapeutic targets, e.g., for the induction of EFT-specific T-cell responses (11). However, whether peptides corresponding to the fusion region of EWS-FLI1 can be recognized by cytotoxic T lymphocytes (CTLs) on the surface of tumor cells and whether these CTLs are able to kill the relevant tumor cells in vivo remain to be determined. In addition, such peptides are only useful for the vaccination of patients with HLA haplotypes that can bind and present these peptides. Other, yet unknown, tumor-specific or tumor-associated antigens from EFT cells can act as targets for CTLs or other targeted therapies.

Recently, DNA microarrays have been introduced for molecular discrimination between members of the “small round blue cell tumor” family (EFTs, neuroblastomas, rhabdomyosarcomas, and lymphomas; refs. 12, 13, 14). We used this technology for the identification of EFT-specific gene expression profiles to find new therapeutic targets and to get new insights into the biology of EFTs. We identified a set of genes with high expression in EFTs compared with a wide spectrum of normal tissues. The identified EFT-specific gene expression profile molecularly supports the concept of a neural crest and endothelial origin of the EFT stem cell. In addition, these genes can be used as diagnostic targets and for the development of new treatment strategies.

HEK293 cells (15) and A673 cells (16) were purchased from American Type Culture Collection (Manassas, VA). SBVGA1 and SBVGA3 cells are interleukin-2-transfected and mock-transfected derivatives of the A673 cell line. We showed previously that the gene expression profile of these cell lines resembles that of parental A673 cell lines (17). Cell lines SH-SY5Y (18), SIMA (19), CHP-126 (neuroblastoma; ref. 20), and SK-N-MC (EFT, initially described as neuroblastoma; ref. 18) were purchased from the Deutsche Sammlung für Zellkulturen und Mikroorganismen (Braunschweig, Germany). Bone marrow (BM) cells were obtained by BM aspiration from healthy volunteers. Epstein-Barr virus (strain B95.8)-immortalized B cells (lymphoblastoid cell lines) were established from peripheral blood mononuclear cells using standard methods. Primary EFT samples were a kind gift from C. Poremba and K-L. Schäfer (Düsseldorf, Germany). Primary neuroblastoma samples were kindly given to us by F. Berthold (Cologne, Germany).

For stable and transient transfection of HEK293 cells with EWS-FLI1, the cDNA for EWS-FLI1 type I was cloned into the eukaryotic expression vector pIRES2-EGFP (BD Clontech, Palo Alto, CA). Cells transfected with empty pIRES2-EGFP vector (mock) served as control. For transient transfection, standard diethylaminoethyl dextran transfection (21) was used. For stable expression of EWS-FLI1, cells were transfected using FuGENE 6 (Roche, Mannheim, Germany) and selected with 400 μg/mL G418. Expression of EWS-FLI1 in transfected HEK293 cells was proven by RT-PCR and Western blotting.

DNA microarray analysis was performed as described previously (22). For the identification of EFT-specific markers, we used customized microarrays (EOS-Hu01) containing 35,356 oligonucleotide probe sets for the interrogation of a total of 25,194 gene clusters. Data were used after γ-distribution normalization. In addition to 11 EFT samples, samples from 133 normal tissues were analyzed on the same microarray (EOS-Hu01). Primary image analysis was performed by using Microarray Suite 4.01 (Affymetrix, Santa Clara, CA) for HG-U95A microarrays and by using Microarray Suite 5.0 for EOS-Hu01 and HG-U133A microarrays. Images were scaled to an average hybridization intensity of 200 (HG-U95A microarrays) or 500 (HG-U133A microarrays). Data from A673 cells and derivative cell lines (SBVGA1 and SBVGA3) were matched and analyzed together with data from the published Gene Expression Atlas (23).

Cluster analysis and visualization was performed with Gene Cluster and TreeView (24).

For each sample (native tumors or cell lines), the presence or absence of gene expression was calculated on the basis of average fluorescence intensities of the representing probe sets. Genes with differential expression between EFT and BM were identified based on the following three criteria: (a) a minimal difference of mean average intensities of >400, (b) a minimal fold change of mean average intensities of >3, and (c) a Student’s t test P of <0.01. To identify genes specifically up-regulated in EFTs compared with normal tissues [normal body atlas (NBA)], the following strategy was used: First, for each probe set, we calculated the median and the 75th percentile of all EFT samples. Then we calculated the 10th, 85th, and 95th percentile of all NBA samples. We used the 10th percentile of NBA samples as background and subtracted those values from all genes in all samples. Finally, we calculated the ratio of EFT values (median and 75th percentile) to NBA values (85th and 95th percentiles) and sorted the data set on the highest value of any of these ratios to identify genes up-regulated in EFT compared with NBA.

All native EFT samples and EFT cell lines showed expression of EWS-FLI1 fusion transcripts corresponding to translocation type I by RT-PCR.

EWS-FLI1 chimeric transcripts were detected by RT-PCR using the primer combination EWS-forward (5′-tcctacagccaagctccaagtc-3′) and FLI1-reverse (5′-actccccgttggtcccctcc-3′) as described previously (5). For validation of microarray data by RT-PCR, the following primers were used: β-actin (ACTB), 5′-gtccaccttccagcagatgt-3′ (forward) and 5′-caccttcaccgttccagttt-3′ (reverse); adrenergic β1 receptor (ADRB1), 5′-ggggaagggagaagcattag-3′ (forward) and 5′-ggtttgccctacacaaggaa-3′ (reverse); BCL11B, 5′-tctttcatggggagagaagg-3′ (forward) and 5′-ttgccttgccagtacaaatg-3′ (reverse); DKK2, 5′-ttacttaaatcccatctgcagtc-3′ (forward) and 5′-tgttccattttcatttcaccaa-3′ (reverse); EGR2, 5′-ggtcgccttgtgtgatgtag-3′ (forward) and 5′-caaacaaatcagctccggta-3′ (reverse); EZH2, 5′-tgcaaatcattcggtaaatcc-3′ (forward) and 5′-aaggcagctgtttcagagga-3′ (reverse); GDF10, 5′-gccagcctgagtacctgaag-3′ (forward) and 5′-tgcagtggactttgaaagga-3′ (reverse); JAK1, 5′-tgtaaggagctggctgacct-3′ (forward) and 5′-cacctgctcccctgtattgt-3′ (reverse); LECT1, 5′-tatcttgggcatggtgtgaa-3′ (forward) and 5′-aagatgcaagcaagggaaga-3′ (reverse); LIPI, 5′-tccgagaatagagaccattctga-3′ (forward) and 5′-gctctctggtggttgcattt-3′ (reverse); NPY1R, 5′-ctgtgtgacttgtggcgtct-3′ (forward) and 5′-tgtccgcccttttaaaatca-3′ (reverse); NPY5R, 5′-aaaaagagtgggcctcaggt-3′ (forward) and 5′-gggacccctggtatgaactta-3′ (reverse), PCDH8, 5′-tgatttcaattgcggcttg-3′ (forward) and 5′-ggcaaggcaaaatttctcaa-3′ (reverse); PCDH11X, 5′-acccagaaaacaggcagatg-3′ (forward) and 5′-gctgtcgggcttgaaagtag-3′ (reverse); and OLFM1, 5′-taaagaggcgaggcaatgac-3′ (forward) and 5′-cacactctgaccatcgcttc-3′ (reverse). Quantitative RT-PCR (40 cycles) was performed on a Mx3000P system (Stratagene, La Jolla, CA) using SYBR Green as fluorochrome according to manufacturer’s instructions. β-Actin (ACTB) was used for normalization. For a comparative analysis, values from one neuroblastoma sample were arbitrarily set as 1, and relative quantities were calculated using the Mx3000P system software.

Expression of cyclin D1 in tumor cell lines was determined by fluorescence-activated cell-sorting (FACS) analysis using a cyclin D1 monoclonal antibody/isotype reagent set (Becton Dickinson, Heidelberg, Germany) in combination with cell cycle analysis according to the manufacturer’s instructions. Cell cycle analysis was performed by using a propidium iodide staining kit (Sigma, Taufkirchen, Germany) according to manufacturer’s instructions.

The histogenetic origin of EFT is still unclear, but bone and BM are the predominant localizations of EFTs. Therefore, we used normal BM as a control for the comparative analysis of genes up- or down-regulated in EFTs. We analyzed EFT and BM samples using customized microarrays with an optimized design (EOS-Hu01). As expected, we could easily discriminate EFTs from normal BM by hierarchical cluster analysis using differentially expressed genes (data not shown). In addition to newly identified genes with high expression in EFTs, we found several genes known to be highly expressed in EFTs, e.g., cholecystokinin (CCK; ref. 25), neuropeptide Y receptor 1 (NPY1R; ref. 26), six transmembrane epithelial antigen of the prostate (STEAP; ref. 27), CD99 (28), and D-type cyclins (29). In addition, many genes were not detected in EFTs or were underrepresented in EFTs compared with BM. It remains unclear whether the absence of expression in EFTs reflects “down-regulation” or only tissue-specific expression in normal BM and absence of expression in other cells. However, among the genes “down-regulated” in EFTs compared with BM, we found genes that are known to be absent in EFTs and have been considered for diagnostic purposes (e.g., CD24; ref. 30). A large number of hematopoietic markers were expressed at low levels or were absent in EFTs, e.g., CD16 or neutrophil elastase (ELA2). This most likely represents specific expression of these genes in hematopoietic tissues.

A large number of genes up-regulated in EFTs in comparison with BM are known to be expressed in neuronal cells or during neuroectodermal differentiation. This finding supports the concept of a neuroectodermal origin of EFTs. Thus, we analyzed the gene expression profile of EFTs in comparison with 133 samples from normal tissues (NBA) on EOS-Hu01 microarrays. We identified 38 probe sets corresponding to 37 genes that are up-regulated in EFT compared with all other tissues (Table 1). Twenty-one of these genes have been described before to be expressed in EFTs (Table 1). According to the Stanford Online Universal Resource for Clones and Expressed Sequence Tags (SOURCE),⁴ 13 of these 21 genes have the highest abundance of expression in EFT (Table 1). Based on our microarray expression data, three categories of up-regulated genes could be distinguished: (a) genes with exclusive expression in EFTs that are not expressed at detectable levels in any normal tissue (Table 1, numbers 4, 5, 6, and 27), (b) genes with expression in EFTs and a lower expression level in many or all other tissues (Table 1, numbers 19–22 and 28–35), (c) genes expressed in EFTs and a restricted number of normal tissues (Table 1, numbers 1–3, 7–18, 23–26, and 36–38). Examples for these categories are shown in Fig. 1.

The latter category is of primary interest for the question of histogenetic origin of EFTs. Based on information from SOURCE, a substantial proportion (20 of 37) of EFT-“restricted” genes are expressed in fetal tissues (Table 1). Moreover, several of these genes are expressed in adult and fetal brain. Thirty three of 37 genes (89%) were expressed in neuronal tissues or during neuronal differentiation. For 28 of 37 genes (75%), brain or the peripheral neuronal system ranks among the top 10 tissues expressing these genes. Six of these 37 genes (16%) have been found in neuronal tissues with the highest abundance of all normal tissues (Table 1). These data demonstrate a clear neuroectodermal signature in the transcriptome of EFTs. Striking evidence for this hypothesis was also obtained by using hierarchical clustering of EFTs and all normal samples using expression levels of EFT marker genes as data points (Fig. 2,A). All EFT samples cluster together in one single branch of the cluster tree. However, brain samples (both adult and fetal) cluster together in one separate cluster that is the next neighbor to the EFT cluster. EFT and brain samples are included in one cluster separated from all other tissues. From this analysis, it is evident that the brain is the tissue most resembling EFTs at the transcriptional level. Similar results were obtained by analysis of a published HG-U95A Gene Expression Atlas (23) with the A673 EFT cell line and several derivatives of this cell line (17) that have been analyzed on Affymetrix HG-U95A microarrays. For this analysis, we excluded samples from malignant tissues because we observed that otherwise all tumor cell lines (including EFT cell lines) cluster together separated from normal tissues (data not shown). In this case, cluster analysis again showed similarity between EFTs and samples of neuronal origin (Fig. 2 B). Remarkably, there is also evidence for a high similarity between EFT cells and human umbilical vein endothelial cells (HUVECs). For control, we performed the same analysis with samples from Burkitt’s lymphoma (BL) that are included in the same Gene Expression Atlas. Despite the high heterogeneity of these BL cell lines (e.g., with regard to Epstein-Barr virus status), all BL samples cluster together in direct proximity to samples from blood and hematopoietic stem cells. This demonstrates the resemblance of gene expression profiles from the tissue of origin in tumor cells. Neuronal tissues as well as HUVECs were clearly separated from BL. These results suggest that the observed similarity between EFTs and both neuronal and endothelial tissues are specific for EFT histogenesis and are not a nonspecific effect of cultivation in vitro. Using Self Organizing Maps, we obtained comparable results indicating the highest similarity between fetal brain and EFTs (data not shown).

The discrimination of EFTs from other small round blue cell tumors by histologic methods is often difficult, and EFT-specific gene expression profiles may be helpful for molecular confirmation of EFTs. To prove the diagnostic value of the newly identified EFT signature gene expression profile, we analyzed samples from additional EFT patients that were not included in the initial analysis and neuroblastoma patients on commercially available Affymetrix HG-U133A microarrays. As shown in Fig. 3,A, samples from all EFT patients gave strong signals for the huge majority of EFT-specific genes. In contrast, samples from neuroblastomas express these EFT-associated genes in significantly lower amounts. Based on Affymetrix presence/absence calls, neuroblastomas express only approximately 40% of EFT-associated genes (data not shown). In addition, we analyzed a set of cell lines that were initially established as neuroblastomas (SH-SY5Y, SIMA, CHP-126, and SK-N-MC). Notably, in the current literature, SK-N-MC cells are described as neuroblastoma (31) as well as Askin tumor (32), a member of the EFTs (33). In contrast to SH-SY5Y, SIMA, and CHP-126, SK-N-MC expresses all of the EFT signature genes, clearly indicating that this cell line does not represent a neuroblastoma but a EFT. In addition, these data show that EFT-associated genes that have been identified by use of customized EOS-Hu01 microarrays can be used for diagnostic purposes by using a commercially available microarray (Affymetrix HG-U133A) with a different design. The diagnostic value of the identified genes was further validated by conventional RT-PCR and quantitative RT-PCR (QRT-PCR). As shown in Fig. 4,A, SK-N-MC cells express all of the analyzed targets, whereas the neuroblastoma cell line SH-SY5Y showed no expression of these genes. Moreover, native EWS-FLI1-positive EFTs could clearly be discriminated from neuroblastomas and an EWS-FLI1-negative peripheral neuroectodermal tumor (tumor X in Fig. 4,B and C) by RT-PCR. In this analysis, the primers used with specificity for LIPI always amplified two products (Fig. 4,B). We sequenced both polymerase chain reaction (PCR) products and found that both correspond to the expected target LIPI, probably representing different splice variants (data not shown). Fig. 4 C shows additional results obtained by QRT-PCR. Using the indicated five markers, a clear discrimination between EWS-FLI1 positive EFT and other tumors was evident, underscoring the diagnostic value of the newly identified EFT-specific genes.

It was shown in the past that expression of EWS-FLI1 in normal fibroblasts induces neuroectodermal differentiation (34). Therefore, it seems possible that the observed similarity of gene expression profiles between EFTs and neuronal tissues does not reflect the histogenetic origin of EFT but is a direct or indirect consequence of expression of the fusion protein. To test this hypothesis, we transfected HEK293 cells with an expression plasmid encoding EWS-FLI1. Expression of the transgene was confirmed by RT-PCR (Fig. 5) and Western blotting (data not shown). The overall expression of EFT signature genes increased slightly in transfected HEK293 cells. However, expression of EFT signature genes does not increase significantly in transfected cells (Fig. 5,A). This indicates that EFT signature gene expression is not a direct consequence of EWS-FLI1 expression but rather a consequence of its neuroectodermal origin. The only EFT-associated gene significantly up-regulated in EWS-FLI-transfected HEK293 cells was cyclin D1. Expression of cyclin D1 was demonstrated at the protein level in EFT cell lines by FACS analysis (Fig. 5 B). CCND1 has been used as a therapeutic target for the biochemical and immunologic treatment of cancer. The high expression of CCND1 in EFTs, together with the fact that CCND1 is up-regulated after EWS-FLI1 expression in HEK293 cells, suggests that CCND1 might play a major role during EFT pathogenesis and that inhibition of CCND1 might be of therapeutic interest. Preliminary results suggest that EFT cells have a high sensitivity for treatment with 2-cyclopenten-1-one, an inhibitor of CCND1 expression (35), and for treatment with roscovitine, an inhibitor of downstream cyclin-dependent kinases (CDKs; ref. 36), suggesting that inhibition of CCND1 and downstream CDKs might be a promising treatment strategy for this disease (data not shown).

High-density DNA microarrays are a powerful tool for the en bloc analysis of the complete gene expression profile of normal and malignant cells. In the present study, we used this technology for the analysis of Ewing tumors and identified genes with high and specific expression in EFTs compared with normal tissues. Our data clearly show that EFT is molecularly related to neuronal and endothelial tissues. In addition, the identified genes can be used for molecular diagnosis and for the development of new EFT-specific therapeutic regimes.

Twenty-one of 37 genes with highest overexpression in EFTs compared with our NBA have been previously shown to be expressed in EFTs (Table 1). Based on published data, 13 of these genes have higher expression levels in EFTs than in any other tissues tested. This information indicates that the results of our microarray analysis are highly compatible with data that have been obtained with independent techniques such as serial analysis of gene expression. Moreover, expression of these EFT signature genes identified by customized arrays can be reproduced with commercially available microarrays (Fig. 3), which may be useful as a diagnostic tool for the discrimination of EFTs from other tumors with similar morphology. In addition, our RT-PCR analysis and QRT-PCR results indicate that discrimination between EFTs and neuroblastomas is possible on the basis of a few PCRs using the signature genes as targets.

Recent data from the analysis of c-myc-transfected B cells showed that the phenotype of a tumor cell is not necessarily a consequence of the histogenetic origin of the tumor cell (37). In contrast, one or a few master genes can drive the gene expression program of a cell, yielding a phenotype of unrelated histogenesis. It was previously shown that transfection with EWS-ETS imposes a neuroectodermal differentiation on normal fibroblasts (34). However, our data, obtained by DNA microarray analysis of EWS-FLI1-transfected HEK293 cells, showed that most EFT-specific markers are not significantly up-regulated in these transfected cells. Nevertheless, cluster analysis of EFTs and normal tissues, using the signature genes as data points, showed a close relationship between EFTs and neuronal tissues. It was shown that the phenotype of EWS-FLI1-transfected cells depends on the cellular background (38). Additional cofactors may be required for the induction of EFT signature genes by EWS-FLI1. Cells of neuronal differentiation seem to express such cofactors, as indicated by the observation that neuroblastoma cells acquire features of EFT after transfection with EWS-FLI1 (39). We conclude that the phenotype of EFT is not solely a consequence of EWS-FLI1 expression but reflects the histogenetic origin and developmental plasticity of the EFT stem cell.

Remarkably, several genes up-regulated in EFTs are expressed in fetal tissues (Table 1). This is in agreement with a close relationship of EFTs to other embryonal tumors.

Ewing initially described the tumor as endothelioma (1), a mesenchymal tumor. Interestingly, some of the EFT-specific genes are expressed in blood vessels (Fig. 2,A), and there are also strong similarities in gene expression profiles from A673 parental and derivative EFT cell lines and endothelial cells (HUVECs; Fig. 2 B). FLI1 is specifically expressed in cells from the neural crest during mesenchyme development (8) and has a function during vasculature development (7). This may reflect that the EFT stem cell is a cell that is arrested at a developmental stage that represents an early event during mesenchyme development from the neuroectodermal germ layer. An alternative explanation would be that ectopic expression of FLI1 (i.e., EWS-FLI1) in a neural crest-derived pluripotent cell imposes mixed expression of mesenchymal and neuronal markers. Interestingly, one of the EFT-specific genes (SCGF) was shown to be involved in the differentiation of endothelial cells from BM-derived stem cells (40). The developmental plasticity of at least a few rare neuronal stem cells allows transdifferentiation into several seemingly unrelated tissues (reviewed in ref. 41). Vice versa, BM-derived stem cells can differentiate into neuronal cells on rare occasions (42). Taken together, these data suggest that the EFT stem cell is a neuronal crest-derived stem cell at transition to mesenchymal endothelial development, residing in the BM.

Some of the genes with high expression in EFTs compared with normal tissues may be involved in the pathogenesis and/or progression of EFT. EZH2, a gene with proposed function in gene silencing, has been shown to be involved in progression of different cancer types (43, 44). The up-regulation of other genes (e.g., JAK1, DKK2, and BCL11B) may be responsible for altered signal transduction cascades in EFTs.

Interestingly, EFTs express several genes involved in metabolism/signaling of retinoids (Table 1; data not shown). CYP26B1 (P450RAI2), one of the genes with the highest overexpression in EFTs compared with BM (data not shown) and normal tissues (Table 1; Fig. 2), is involved in the inactivation of retinoic acid (45). P450RAI2 can metabolize retinoic acid (RA) to more polar (and seemingly inactive) derivatives. It was described that EFTs are relatively insensitive to the induction of differentiation by RA (46). However, in a recent publication, it was shown that SK-N-MC cells (shown in the present study to be an EFT cell line) are the most sensitive cells among several “neuroblastoma” tumor cell lines (31). We found that inhibition of P450 activity increases the antiproliferative and differentiation-inductive activity of RA in P450RAI2-positive but not P450RAI2-negative tumor cell lines (data not shown). This observation might suggest that expression of P450RAI2 is not only a mechanism for RA inactivation but also indicative for a general RA sensitivity of P450RAI2-expressing cells. Selective inhibition of P450RAI2 in combination with RA treatment may therefore be an interesting therapeutic strategy for the treatment of EFTs.

Another highly attractive target is LIPI. The corresponding sequence has been cloned from EFTs, and available data (SOURCE) on expression of this target in other malignant and normal tissues indicate that this gene is expressed in all other tissues (except testis) at very low or undetectable levels. Interestingly, the corresponding sequence was identified as candidate cancer-testis antigen (47). Recently, the corresponding protein has been identified as membrane-associated phospholipase A1 β (48). The highly specific expression of this gene in EFTs (and, based on the literature, some other tumor cells) may allow the development of specific and selective treatment strategies for EFTs, e.g., the generation of specific cytotoxic T cells with specificity for corresponding peptide antigens.

Another EFT signature target, CCND1, has been successfully used as immunologic target for allogeneic T cells (49). CCND1 is highly expressed in EFTs and is up-regulated in HEK293 cells as a consequence of EWS-FLI1 transfection and may be directly involved in the uncontrolled proliferation of EFT cells. Inhibition of cell cycle-regulatory cyclins and CDKs is currently under investigation as a tumor therapy in clinical trials (36). Whether or not such inhibitors of cyclin D1 or downstream CDKs may also be useful for the treatment of EFT patients remains to be determined.⁵

We thank S. Bergmann and I. Volkmer for grateful technical assistance. We thank C. Poremba, K-L. Schäfer, F. Berthold, and the Kompetenznetz Pädiatrische Onkologie und Hämatologie for supporting us with tumor samples and G. Richter for critical reading of the manuscript.