Main

Rhabdoid tumors (RT) are highly malignant neoplasms first described in the kidney of young children. RTs also arise in a variety of extra-renal sites including soft tissues and the central nervous system, where they are often called atypical-teratoid/RT.1, 2, 3 RTs at all sites have a common genetic abnormality, the mutation or deletion of the SMARCB1/hSNF5/INI-1 gene located at chromosome 22q11.3, 4, 5 This gene, which will be referred to as SMARCB1, encodes a component of the SWI/SNF chromatin remodeling complex that has an important role in transcriptional regulation (reviewed in Imbalzano and Jones6). This has resulted in considerable scientific attention to RTs in recent years, as they represent a potent model for the investigation of this important chromatin remodeling complex. Clinically, over 70% of children with RT present with non-localized disease, and chemotherapy alone is rarely curative.1 With an overall survival of 23%, new therapeutic options are needed. The overall goals of this study are to identify genetic pathways that will clarify the nature of RTs and to identify therapeutic targets.

MATERIALS AND METHODS

Patient Samples

Frozen tissue samples were obtained from the Renal Tumor Bank of the Children's Oncology Group. Specimens showing <80% tumor cellularity on frozen sections were excluded. A total of 53 tumors were analyzed: 10 RT, 12 cellular mesoblastic nephromas (CMN), 16 clear cell sarcomas of the kidney (CCSK), 15 favorable histology blastemal predominant Wilms tumors (WT). These samples were previously examined for the purpose of developing diagnostic signatures.7 Relatively equal numbers of tumors from each category were analyzed so that no single group biased the statistical analysis. All RT demonstrated loss of nuclear staining for BAF47, the protein encoded by SMARCB1,8 and all cellular CMNs demonstrated the ETV6-NTRK3 fusion transcript characteristic of this entity.9 Two fetal kidneys (16–18-weeks gestation) were analyzed for comparison.

Gene Expression Analysis

RNA was extracted and hybridized to Affymetrix U133A arrays (Affymetrix, Santa Clara, CA, USA), scanned, and subjected to quality control parameters and normalization as described earlier.7 The gene expression of RT was compared with the other tumor types using two-sample t-tests. Affymetrix NetAffx Analysis Center was used to categorize genes and to establish the chromosomal location of differentially expressed genes. Two bioinformatics tools were used to analyze the data, Protein ANalysis THrough Evolutionary Relationships (PANTHER), and Gene Set Enrichment Analysis (GSEA). PANTHER classifies genes by their functions using published experimental evidence, and these are grouped into families and subfamilies of shared function. The detection of significant over- or under-represented functional pathways in a preselected gene list is determined by the binomial test. The Bonferroni-corrected P-values are calculated to adjust for multiple testing.10 GSEA contains gene sets that relate to common biological function, including those from Gene Ontology groups as well curated gene lists from different pathways and publications. GSEA first ranks the expression of each gene based on its correlation with one of two phenotypes being compared (in this case RT vs non-RT). It then identifies this rank position within the independent gene set being queried. From this ranking it calculates an enrichment score that reflects the degree to which genes within the independent gene set are over represented within those genes most highly correlating with one of the two phenotypes. The normalized enrichment score (NES) takes into account the number of genes within the independent gene set. Permutation on the phenotype class labels is used to obtain the null distribution of NES and the nominal P-value. The false discovery rate (FDR) is computed by comparing the tails of the observed and null distributions for the NES (http://www.broadinstitute.org/gsea).11

Real-Time Quantitative RT–PCR

TaqMan Gold and the ABI Prism 7700 Sequence Detection System (Applied Biosystems, Foster City, CA, USA) were used. RT–PCR cycle parameters were 48°C for 30 min, 95°C for 15 min, followed by 40 cycles at 95°C for 15 s, and 59°C for 1 min. Sequences of probes and primers are shown in Table 1. Each threshold cycle (CT) was determined and the CT for the housekeeping gene (β-actin or GAPDH) was subtracted from this for normalization (dCT).

Table 1 Reagents utilized for verification of gene expression
Full size table

Immunohistochemistry

Formalin-fixed-paraffin-embedded tissue from six of each RT, CCSK, CMN, and WT were tested using the monoclonal antibodies provided in Table 1. This was visualized by a streptavidin–biotin system (Vectastain Elite ABC kit, Vector Laboratories) followed by ImmunoPure Metal Enhanced DAB Substrate (Thermo Scientific, Rockford, IL, USA) and counter-stained with hematoxylin.

Immunoblotting

Three frozen CCSKs, CMNs, RTs, and WTs were homogenized in sample buffer [0.125 mol/l Tris–HCl (pH, 6.8), 2% SDS, 10% glycerol, 0.001% bromophenol blue, and 5% β-mercaptoethanol]. Equal amounts of protein lysate were run on polyacrylamide SDS gels and transferred to a nitrocellulose membrane (neural cell adhesion molecule (NCAM), CDH2, LGALS1) or PVDF (Immobilon-P) membrane (TFRC, CMYC). Membranes were blocked and incubated overnight at 4°C in primary antibody (provided in Table 1). The washed membranes were incubated with alkaline phosphatase antimouse IgG (Vector Laboratories), and color was developed with a 5-bromo-4-chloro-3-indolyl phosphate/nitroblue tetrazolium alkaline phosphatase substrate kit IV detection kit (Vector Laboratories).

RESULTS

Overall Differential Gene Expression in RTs

To establish a broad list of probesets containing the majority of genes involved in the pathogenesis of RT, the expression of RT was compared with all the other tumor types combined (‘non-RT’), resulting in 2921 probesets with P-value <0. 001 and FDR of ≤1% (Supplementary Table 1, available at the journal's website). The raw data from this study have been deposited in GEO, accession number GSE11482. To define a more restrictive list of probesets more uniquely expressed in RT, the gene expression of RT was compared with each of the other tumor types separately. Comparing RT to each of these clinically, pathologically, and genetically different tumor categories minimizes erroneous conclusions that may result from a single comparison. Genes present in each one of the resulting four comparisons (RT vs CCSK, RT vs CMN, RT vs WT, and RT vs all non-RT) with a P-value of <0.001 in each are provided in Supplementary Table 2 (426 genes). Of these, 114 genes showed fold change >2 or <0.5 in the comparison between RT and all non-RT and are provided in Table 2, arranged in recurring functional groups indicated by GO and PANTHER. The chromosomal location of all probesets in Supplementary Table 1 was analyzed, and no chromosomal arm was over-represented, including probesets on chromosome 22. The 766 genes differentially expressed between RT and non-RT (P<0.0001) were analyzed in PANTHER, and those groups within the biologic process category over-represented with a Bonferoni corrected P-value<0.05 are listed in Supplementary Table 3.

Table 2 Genes with a P-value of <0.001 in each comparison of RT vs non-RT, RT vs CMN, RT vs CCSK, RT vs WT, and with a fold change of >2 or <0.5 in the comparison between RT and non-RT
Full size table

Decreased Expression of SMARCB1 and Associated Genes in Rhabdoid Tumors

SMARCB1 was within the top five genes most significantly differentially expressed in RT, and 12 of the top 114 genes in Table 2 have been shown to be concordantly expressed with SMARCB1 in prior gene expression analyses. These include genes previously demonstrated to be concordantly differentially expressed after induction of SMARCB1 within RT cell lines (ATP1B1, PTN, SPOCK1, DOCK4, SERPINE2), and after inactivation of SMARCB1 within murine embryonic fibroblasts (NQO1, PLOD1, PTP4A2, PTPRK).12, 13, 14 In addition, Pomeroy et al compared RT with four pediatric central nervous system tumors and provided the top 100 genes characterizing each tumor type. Of the resulting 500 genes, 99 were also found in our Supplementary Table 1 and were concordantly regulated (designated by an asterisk); two genes, COL5A2 and RSUI, were identified in Table 2.15 Finally, SPP1 was shown to be significantly upregulated in human RT.16 The gene expression patterns of SMARCB1 and selected SMARCB1-associated genes are illustrated in Figure 1. The downregulation of SMARCB1, PTN, and PTPRK, and the upregulation of NQO1 mRNA were confirmed by QRT–PCR (Figure 2). Absence of expression of the SMARCB1 protein was established in all RTs tested by immunohistochemistry, with strong nuclear positivity identified in all CCSK, CMN, and WT tested. In contrast, antibodies against NQO1 showed strong positivity in all RTs and pale to no staining for the remaining tumor types (Supplementary Figure 1).

Figure 1
figure 1

Patterns of gene expression within the different groups of pediatric renal tumors. The log expression levels (low to high) are plotted on the y axis. The x axis reflects an arbitrary tumor number, grouping the different tumor types starting with the congenital mesoblastic nephromas (infantile fibrosarcomas) in black, followed by clear cell sarcoma of the kidney in green, rhabdoid tumors in red, and Wilms tumors in blue.

Full size image
Figure 2
figure 2

Validation of expression of SMARCB1, FYN, PTN, NQO1, PTPRK, TFRC, and CCND1 using quantitative reverse transcriptase polymerase chain reaction. RNA from seven examples of each tumor type was analyzed in duplicate using the primers in Table 1. Expression changes are determined by subtracting the CT value of the housekeeping gene (β-actin or GAPDH) from the CT of the gene of interest (dCT). The average dCT from each individual tumor was subtracted from the average tumor RT dCT value for each gene (ddCT); these values were averaged for each tumor type and the error bars represent the standard deviation of the averages.

Full size image

Decreased Expression of Genes Associated with Neural Development

Strikingly, 28 (25%) of the genes in Table 2 are involved with neural development and all were sharply downregulated. This includes SMARCB1 itself and four genes directly associated with SMARCB1 that have previously been shown to be important in neural development (PTN, DOCK4, SPOCK1, and PTPRK). These findings are supported by analysis in PANTHER, in which the Neurogenesis group was within the top five most enriched groups of biological processes (Supplementary Table 3). To obtain more information regarding specific developmental pathways, GSEA was performed using selected groups in Gene Ontology that pertain to early development pathways.11 Significant enrichment was identified in non-RT compared with RT (consistent with downregulation in RT) within the following groups: GO:0022008 Neurogenesis (NES −1.59, nominal P=0.006, FDR 2%), GO:0048666 Neuron Development (NES −1.53, P=0.008, FDR 4%), GO:0030182 Neuron Differentiation (NES −1.48, P=0.008, FDR 5%), GO:0051960 Regulation of Nervous System Development (NES −1.44, P=0.009, FDR 5%). Many genes are redundant within these GO groups, therefore, a common gene list containing all genes in all the above groups was developed and is illustrated in order of rank within GSEA in Supplementary Figure 2a.

Of particular interest was the highly significant enrichment of the two available GO groups containing genes pertaining to neural crest development in non-RT (downregulation in RT). This includes GO:0014033 Neural Crest Development (NES −1.65, P<0.001, FDR 2%), and GO:0001755 Neural crest migration (NES −1.60, P<0.001, FDR 2%). A GSEA heatmap of all genes in both lists in rank order is illustrated in Supplementary Figure 3a. There was minimal overlap between the enriched genes of the neural development groups and those of the neural crest group. In contrast, gene sets within other early developmental GO categories (including GO:0045445 Myoblastic Differentiation, GO:0051216 Cartilage Development, GO:0001501 Skeletal development, GO:0048513 Organ development, GO:0042692 Muscle Cell Differentiation) did not show significant enrichment (defined as nominal P value<0.05, FDR<20%) in RT or non-RT. Finally, a recently published comprehensive list of genes involved in renal development was entered into GSEA and did not show significant enrichment.17 Of note, CCSKs demonstrated increased expression of genes involved with nervous system development (but not neural crest development), as we have reported earlier.18 Therefore, all the above analyses of RT vs non-RT and fetal kidney were repeated excluding CCSKs, and all groups retained their significance (illustrated in Supplementary Figures 2b and 3b). RNA expression patterns of selected neural developmental genes from Table 2 are shown in Figure 1. Decreased expression of FYN, PTN, and PTPRK was verified by QRT–PCR (Figure 2). Western analysis of key neural regulators CDH2 (N-cadherin) and NCAM (CD56) proteins were entirely negative in all RT tested, with expression in 2/3 CCSKs and in all three WT tested; CMNs showed no to low expression for both markers (Figure 3a).

Figure 3
figure 3

Protein expression in pediatric renal tumors. (a) Immunoblotting of CDH2, NCAM, LGALS1, MYC-C, and TFRC. Protein lysates from three examples of each tumor type were analyzed using antibodies provided in Table 1. Decreased expression of neural markers CDH2 and NCAM is identified in RT. Increased expression of oncogene LGALS1 is identified in 2/3 RT tested, as well as in two CMNs. Increased expression of MYC-C and one of its key targets (TFRC) is identified in RT. (b) Nestin expression in RT. Nestin mRNA was not differentially expressed in the pediatric renal tumors; however Nestin protein cytoplasmic and nuclear expression were identified in all RT tested using immunohistochemistry.

Full size image

Several genes in Table 2 are involved in Wnt signaling (PTN, DOCK4, CDH2, LEF1, ENC1, FSTL1) or in Notch signaling (HES1, NOTCH2), and all were downregulated. Both Wnt and Notch pathways are known to be critical to neural development. Significant negative enrichment in RT for the GO:0007219 Notch signaling pathway was identified (NES −1.48, P=0.02, FDR 5%, Supplementary Figures 3c and d). However, independent gene sets available within GSEA that contain targets of Wnt signaling showed no significant enrichment of Wnt targets. When the 84 genes with human homologues on the HU133A array from the Stanford gene set of known Wnt targets was entered into GSEA and analyzed (http://www.stanford.edu/~musse/Wntwindow.html), there was likewise no significant enrichment in RT. Immunohistochemistry for β-catenin (CTNNB1) showed no nuclear staining and focal, pale cytoplasmic staining in two out of six tumors, the remaining RTs were entirely negative (data not shown).

Expression of Neural Stem Cell Markers in Rhabdoid Tumor

The striking downregulation of neural developmental markers suggests that RTs may arise within neural or neural crest stem cells followed by developmental arrest. We therefore investigated the mRNA expression of markers of known neural and neural crest stem cells, including MSI1 (Musashi 1), CD133 (Prominen 1), FOXD3, ID3, SOX10, SNAI2, and SNAI1. Of these, only CD133 was found in Supplementary Table 1 and this was downregulated (fold change 0.3, P=0.0002). As these transcription factors may be regulated at very low levels, or translationally regulated, those neural crest and neural crest stem cell markers for which robust commercially available antibodies were available were analyzed, including SOX10, ID3, Musashi, and CD133. All were negative in RT with appropriate positive control staining. Of interest, antibodies directed against nestin, a marker of primitive neural as well as mesenchymal stem cells, demonstrated strong cytoplasmic positivity and scattered nuclear positivity in all RTs, with appropriate negative controls (Figure 3b). Embryonic stem cell markers OCT4, NANOG, and SOX2 did not show differential mRNA expression or upregulation in RT.

Expression of MYC-C and its Targets

MYC-C is known to directly interact with SMARCB1,19, 20 and was significantly upregulated in RT (fold change 2.37; P value 1.4 × 10−7; Supplementary Table 1; Figure 1). To validate the potential role for MYC-C upregulation in RT, the independently curated C2 gene sets within GSEA that contain MYC-C targets were analyzed. Significant enrichment was documented using gene sets Schumacher_MYC_UP (NES 1.56, P=0.002, FDR 1%) MYC_TARGETS (NES 1.44, P=0.03, FDR 5%), and MYC_ONCOGENIC SIGNATURES (NES 1.6, P=0.01, FDR 0.1%).21, 22, 23 A combined list containing all genes in each list was analyzed and illustrated in Supplementary Figure 4. Within Table 2, 13/114 (11%) genes are known targets of MYC-C and are coordinately expressed in RT (including two genes known to be downregulated by MYC-C). Finally, key cell-cycle genes regulated by MYC-C show appropriate expression in RT, including CCND1, CCND2, CCNE1, CDKN1A, CDKN1C, and CDKN2B (see below). Confirmation of expression of MYC-C and TFRC (a MYC-C target) proteins was performed and illustrated in Figure 3a; confirmation of mRNA expression of TFRC using QRT–PCR was also performed (Figure 2).

Cell-Cycle Gene Expression

Numerous studies have shown that re-expression of SMARCB1 in RT-derived cell lines results in upregulation of CDKN1A (p21CIP/WAF1) and CDKN2A (p16INK4a) and downregulation of CCND1 (cyclin D1), CCNE1 (cyclin E), CCNA1 (cyclin A), with secondary phosphorylated RB protein, cellular senescence, and apoptosis.24, 25, 26 Most of these findings are confirmed in the human RTs analyzed in our study, with some exceptions. Of the cyclin-dependent kinase inhibitors, both CDKN1A and CDKN2A are expressed at very low levels in all renal tumor types tested, including RT. CDKN1C (p57kip2) mRNA was also significantly downregulated in RT (fold change 0.4; P=2.15 × 10−6) (Supplementary Table 1; Figure 1). Immunohistochemistry for CDKN1A and CDKN1B showed staining of fewer than 1% of tumor cells in all renal tumors tested (data not shown). Of the cyclin-dependent kinases, CCND2 and CCNE mRNAs were not differentially expressed but were upregulated in all tumors, and antibodies to both cyclin E and cyclin D2 confirmed strong nuclear staining within the majority of RT cells (data not shown). Immunostaining for p53 was positive in fewer than 1% of RT nuclei. In contrast with prior studies, CCND1 mRNA was downregulated in RT in our study (FC 0.4, P=1.2 × 10–5, Figure 1). This was confirmed by quantitative RT–PCR (Figure 2) and immunohistochemistry showed protein expression in fewer than 1% of RT cells in the face of strong nuclear positivity in CCSKs, WT, and CMN (Supplementary Figure 1). (Of note, cyclin D2 and cyclin E are activated by CMYC,27 whereas CMYC has been shown to repress CCND1,28 CDKN1A, and CDKN2B.21)

Differential Expression of Genes Associated with Tumor Suppression, Invasion, and Metastasis

Many of the top 114 genes in Table 2 are recognized oncogenes yet were downregulated in RT. Conversely, 10 genes associated with tumor suppression, invasion, or metastasis demonstrated regulation in RT concordant with their activity and therefore represent potential therapeutic targets. SPP1 (osteopontin), MMP12, NCOA3, TFRC, RSU1, and ZNF217 are associated with tumor invasion and metastasis and were upregulated; SELENBP, COL18A1 (endostatin), PTPRK, and DOCK4 have tumor suppression functions and were downregulated (Table 2; Figures 1 and 4). Of these, PTPRK and DOCK4 have been previously directly associated with SMARCB1 expression.12, 13 Additional potential therapeutic targets found in Supplementary Table 1 include LGALS1 (fold change 1.7, P=1.07 × 10−5) and SERPINF1 (pigmented epithelium-derived factor (PEDF)) (fold change 2.3, P=3.4 × 10−11). These were not included in the top gene list because of their high expression in CMN as well as RT (Figure 4). Both LGALS1 and SPP1 were significantly upregulated in previous gene expression analysis studies.15, 16 Differential expression of TFRC and PTPRK were validated by QRT–PCR (Figure 2). Expression of SPP1 (osteopontin), PEDF, and TFRC proteins in RT was demonstrated by immunohistochemistry or immunoblotting (Figures 3 and 4).

Figure 4
figure 4

Expression of selected genes involved in tumor invasion and metastasis. (a) Pattern of RNA expression of key genes. The expression levels (low to high) are plotted on the y axis. The x axis groups the different tumor types starting with the congenital mesoblastic nephromas (infantile fibrosarcomas) in black, followed by clear cell sarcoma of the kidney in green, rhabdoid tumors in red, and Wilms tumors in blue. (b) Protein expression of SPP1 and PEDF (SERPINF1). The majority of cells within all RT show moderate positivity for SPP1. Note the striking positivity of non-neoplastic renal tubules for SPP1. Clusters of cells within RT show strong positivity for PEDF.

Full size image

Expression of Genes Associated with Transcription Regulation

In addition to MYC-C, a number of genes involved in transcription regulation and signal transduction were upregulated in RT. Of particular note, CBX6 was highly upregulated (Figure 1; Table 2). CBX6 is a member of the polycomb family, which is responsible for transcriptional repression mediated by the specific H3K27 histone methylation. Our data show loss of expression of GRB10 and CDKN1C in RT, both of which are epigenetically regulated by the specific H3K27 trimethylation mediated by the polycomb group.29, 30 RTs also showed loss of expression of multiple HOX genes, long known to be repressed by polycomb group proteins.31 (In Supplementary Table 1, HOXA10, HOXA11, HOXC4, HOXD3, HOXD4, HOXD9, HOXD10, HOXD11 are all downregulated.) In contrast to the polycomb group, the trithorax family is responsible for transcriptional activation mediated by the specific H3K4 methylation. MLL1, the predominant member of the trithorax family in humans, is known to achieve transcriptional activation by interacting directly with SMARCB1.32 Of further interest in this regard is the newly described role of ZNF217 as an organizer of repressive histones. ZNF217 (an oncogene significantly upregulated in RT, Table 2) has been show to demethylate H3K4me3 and to methylate H3K27 through interaction with EZH2, a member of the polycomb repressive complex 2.33 Therefore, the transcriptional repression seen in RT may be associated with histone 3 modifications.

To investigate this hypothesis, we relied on the availability of whole genomic surveys of methylation markers within embryonic stem cells (ESC). Zhao et al34 classified the H3K4me3 and H3K27me3 markers associated with over 17 000 genes in ESC and established three groups of genes: Group 1 with neither H3K4 nor H3K27me3 modification, Group 2 with only H3K4me3 modification, and Group 3 with both H3K4 and H3K27me3 (bivalent). We placed the gene list from each group in GSEA to determine whether their expression was enriched in RT or non-RT. The expression of genes in Groups 1 and 2 did not show significant enrichment in either RT or non-RT. However, the 1141 genes in Group 3 (those bivalently modified in ESC) showed significant enrichment in non-RT, indicating decreased expression in RT (NES −1.5, P=0.01, FDR 5%, Supplementary Figure 5a). When the other pediatric renal tumors were similarly analyzed (leaving RT out), no significant enrichment was identified in either CMN or WT. In contrast, CCSK showed significant positive enrichment for Group 3 genes. Therefore, we again analyzed Group 3 genes comparing RT to non-RT leaving out the CCSKs, and significant enrichment non-RT was retained (P<0.001, Supplementary Figure 5b). Of note, 11% of genes in Supplementary Table 1 are Group 3 genes, and 90% of these genes show decreased expression in RT. Zhao et al34 also demonstrated the top GO categories differentially represented through PANTHER in Group 3 to be genes important for developmental processes, ectoderm development, neurogenesis, transcription regulation, and signal transduction. Within RT, with the exception of transcription regulation, using PANTHER we found the same top categories at the top of our list (Supplementary Table 3). Finally, we performed a similar in silico analysis of 512 targets of Polycomb Repressive Complex 2 in ESC.35 GSEA analysis again revealed significant enrichment in non-RT (NES −1.4, P=0.05, FDR 9%).

DISCUSSION

RT are rare tumors that are virtually confined to infancy and are characterized by loss of the SMARCB1/hSNF5/INI-1 gene.4, 36 SMARCB1 is one of over 10 non-catalytic subunits of the highly evolutionarily conserved SWI/SNF ATP-dependent chromatin-remodeling complex expressed in all normal cells at all stages of development.6 SMARCB1 is recognized as a tumor suppressor gene because of its biallelic involvement in the development of RT, reviewed by Biegel et al.36 The SWI–SNF complex appears to relieve repressive chromatin structures by disrupting the DNA–histone interaction within nucleosomes, allowing the transcriptional machinery to access its targets more effectively, facilitating transcriptional activation and repression (previously reviewed by Stojanova and Penn37 and Roberts and Orkin38). The current hypothesis is that SMARCB1, through its ability to bind to several proteins, serves to recruit the SWI/SNF complex to specific target sites thereby mediating transcription regulation.38, 39 However, precisely how SMARCB1 accomplishes this is largely unknown. Proteins that have been demonstrated by others to physically interact with SMARCB1 or to be direct targets of SMARCB1 include MLL, MYC-C, PTN, HESR1, ATP1B1, and FZD7.12, 20, 32 Our gene expression analysis demonstrates key roles for these genes and others in the development of human RT, as discussed below.

SMARCB1 Loss Results in Repression of Genes Critical to Neural and Neural Crest Development

Genes involved with neural and neural crest development were sharply downregulated in RT. A number of recent studies have shown that SWI/SNF itself is required for normal vertebrate neurogenesis.40, 41 Pleiotrophin, PTN, has been shown to directly interact with SMARCB112 and to regulate the balance between differentiation and proliferation within neural stem cells by inhibiting proliferation and enhancing differentiation.42 The Notch signaling pathway is also critical for controlling the induction of neural development and for maintaining the neural progenitor population.43 Within the Notch pathway, HESR1 (HEY1) is a known direct target of SMARCB1,12 and recruitment of SMARCB1 to the HES1 and HES5 promoters has been demonstrated. NOTCH2 and HES1 (the ligand of HESR1) were both downregulated in RT (Table 2) and GSEA analysis confirmed the downregulation of the Notch signaling pathway in RT. Finally, a large number of important downstream transcription factors involved with neural and neural crest development were likewise sharply downregulated, including NCAM (regulated by HES1), CDH2 (neural-cadherin), FYN (a downstream target of PTN), and SOX11, which induces differentiation within neural stem cells and regulates various aspects of neural crest development.44 In summary, RTs show striking repression of neural differentiation, suggesting the possibility that RTs may arise in progenitor cells in which SMARCB1 loss results in prevention of neural development. This is supported by neural differentiation that develops within cell lines derived from both CNS and extra-CNS soft tissue RTs when SMARCB1 is re-introduced.45

Neural development begins within the embryonic ectoderm from which the neural plate differentiates. The neural plate folds to form the neural tube which then differentiates into the structures of the central nervous system. Cells of the neural crest arise at the border between the neural plate and the ectoderm for the entire length of the neuraxis (previously reviewed by Barembaum and Bronner-Fraser46). After induction, the neural crest cells migrate as undifferentiated precursor cells to various parts of the embryo where they differentiate into many cell types including cells of sensory neurons and glia of the peripheral nervous system, bone, cartilage, and melanocytes. Therefore, the ability to show divergent differentiation, the ability to migrate over considerable distances, and presence in both CNS and extra-CNS sites are features shared by both neural crest stem cells and RTs. Our study reveals significant downregulation of both neural and neural crest developmental genes. To address the hypothesis that RT develop within neural or neural crest stem cells with arrested development, we investigated the mRNA and protein expression of key neural and neural crest stem cell markers.46, 47, 48, 49, 50 Low to no expression of early neural and neural crest stem cell markers SNAIL1, SLUG, FOXD3, SOX10, ID3, CD133, Musashi was identified. Embryonic stem cell markers OCT4, NANOG, SOX2 were also not upregulated in RT. In contrast, strong expression of nestin was identified in RT. Nestin has been shown to be abundantly expressed in embryonic stem cell-derived progenitor cells that have the potential to develop into neuroectodermal, endodermal, and mesodermal lineages.51 Our data therefore suggest that RTs do not arise in neural or neural crest stem cells per se, but instead may arise in early progenitor cells after the differentiation trigger (and after loss of ESC markers) in cells destined to become neural and/or neural crest stem cells. Loss of SMARCB1 at this critical period of time may suppress neural and/or neural crest stem cell development.

Transcriptional Repression may Play a Role in RT Development

Clues toward the mechanisms involved in this developmental repression are provided by evidence that SMARCB1 loss may result in alterations in the activity of the trithorax and polycomb families, which are responsible for maintaining the reciprocal transcriptional states of key developmental regulators.52 MLL1, the key member of the trithorax group in humans, is a ubiquitously expressed nuclear protein that has been shown to interact directly with SMARCB1 through the MLL SET domain, resulting in remodeling of chromatin and transcriptional activation.32 MLL1 mediates transcriptional activation by catalyzing the methylation of histone H3 lysine 4 (H3K4me).53 MLL1 has also been shown to be required for neurogenesis.54 In contrast, the polycomb group (Pcg) mediates transcriptional repression through catalyzing the methylation of histone H3 lysine 27 (H3K27me).31 These critical histone 3 modifications have been comprehensively mapped across the genome in both human and mouse embryonic stem cells using varied experimental approaches.34, 55, 56 These studies, which show striking concordance, show that the majority of genes in ESC are marked with H3K4 alone (50–58%) or in combination with H3K27 (bivalent, 10–17%), whereas 28–33% show neither marking. The majority of genes with H3K4 methylation alone are associated with basic cellular functions such as housekeeping genes, proliferation, protein, and DNA metabolism, etc. In contrast, genes bivalently modified were enriched in GO developmental functions, particularly those involving neurogenesis, ectoderm differentiation, transcriptional regulation, and signal transduction. These bivalently marked genes overlapped considerably with genes demonstrated to be polycomb group targets.35 Indeed, bivalent genes appear to be dominated by the H3K27 repressive markers and to be downregulated in ESC. The presence of the H3K4 methylation in these bivalently modified histones may serve to enable a sharp increase in expression of lineage-specific transcription markers after the trigger for differentiation.34, 35, 55, 56

Our study shows that the same developmental pathways and genes that are marked by bivalently modified histones and dowregulated within ESC are likewise significantly downregulated in RT. These include the processes of neurogenesis, signal transduction, and ectoderm differentiation (Supplementary Table 3). Further, genes bivalently modified in ESC were among those that most strongly differentiated RT from non-RT by GSEA (Supplementary Figure 5). Analyses of independent sets of genes known to be polycomb targets35 likewise demonstrated the same findings. These analyses suggest that RT may arise within early stem cells whose bivalently modified histones remain dominated by the polycomb group histone markings. At the time of differentiation, this repression would generally be overcome by SWI/SNF, but in the absence of SMARCB1 this does not occur. Indeed, there is growing evidence that SWI/SNF may have an integral role in the balance between trithorax and polycomb group activity through the direct interaction between SMARCB1 and MLL1. This is best seen in a series of experiments that focused on the colocalized p16/INK4a and p14/ARF tumor suppressor genes. P16 is known to be repressed because of polycomb group-mediated H3K27 methylation, whereas the nearby p14 is not.57 Re-expression of SMARCB1 within RT cell lines has been shown to result in induction of p16 but not p14.24, 58 Kia et al59 performed chromatin immunoprecipitation demonstrating that SMARCB1 bound strongly to the p16 promoter, but not to the p14 promoter. Further, in the absence of SMARCB1, polycomb group silencers (PRC1, PRC2) were present in large amounts on the p16 promoter, but in low amounts on the p14 promoter. Re-expression of SMARCB1 resulted in the following: (1) reduced binding of PRC1 and PRC2 and increased MLL1 binding at the p16 promoter; (2) reduced H3K27 and increased H3K4 methylation at the p16 promoter; and (3) activated p16. On the basis of these CDKN2A studies, Kia et al concluded that polycomb group silencers are evicted by SWI/SNF, thereby allowing for chromatin modification. In the absence of SMARCB1, this did not occur. Our data suggest that this mechanism may occur more widely within bivalently modified genes in development. Indeed, a broader role of the SWI/SNF chromatin remodeling complex in development is suggested by studies showing that SMARCB1 is required for hepatocyte and adipocyte development.60, 61

It should be noted that there are a number of different histone modifications in addition to H3 methylation, which impact on gene expression. Furthermore, many histone modifications interact with one another, creating extreme complexity. To add to this potential complexity within RT is the presence of MYC-C activation. MYC-C is a transcription factor, which has long been known to drive a large number of diverse biological activities during development and oncogenesis. In recent years, it has become apparent that MYC-C may modulate transcription, in part at least, through H3 and H4 acetylation.62 A physical interaction between SMARCB1 and MYC-C proteins has been documented by several laboratories and the SNF complex acts directly to repress MYC-C during differentiation.19, 20, 27, 63 Finally, MYC-C has been shown to be critical regulator of neural crest formation.64 It is clear that further investigation is needed to clarify the interface between the two broad processes involved in chromatin remodeling, namely histone modification and the ATP-dependent nucleosomal remodeling mediated by SWI/SNF.

RT Show Decreased Cyclin-Dependent Kinase Inhibition

Numerous publications have implicated multiple members of the SWI/SNF complex in cell-cycle control. Several laboratories have demonstrated downregulation of cyclin-dependent kinase inhibitors CDKN1A and/or CDKN2A(p16/INK4a), and their upregulation on reintroduction of SMARCB1, accompanied by growth arrest.24, 26, 59 Mutation analysis of these genes in RT is not available in the literature. These observations would predict upregulation of MYC, downregulation of both CDKN1A and CDKN2A, and upregulation of cyclins A and E within RT, all of which are confirmed in our study. We also show downregulation of CDKN1C (p57Kip2), a major regulator of embryonic growth, that has recently been shown to be a downstream target of SMARCB1.65 Of interest, both CDKN1C and CDKN2A are regulated in part through the polycomb group by methylation of H3K27.30 Therefore, our data support that of other studies that indicate a proliferative advantage of RT through loss of cylin-dependent kinase inhibition directly linked to SMARCB1 loss.

Several previous studies based primarily on RT cell lines have suggested that cyclin D1 is a direct target of SMARCB1 and may be a possible therapeutic target.25, 66, 67 It was therefore surprising that neither our data nor gene expression after SMARCB1 induction in RT-derived cell lines12 supports these findings. Instead, our study shows decreased expression of CCND1 at both the RNA and protein levels. A likely mechanism for downregulation of CCND1 in RT is upregulation of MYC-C, which is known to repress CCND1 expression at its promoter.28

Potential Therapeutic Targets

RTs are highly malignant and lethal tumors, and the ultimate goal of this study is to improve their clinical outcome. Analysis of gene expression shows differential expression of a number of genes known to cause tumor suppression or tumor progression. In particular, two putative tumor suppressor genes have been shown earlier to be directly associated with SMARCB1 and are downregulated in RTs, PTPRK, and DOCK4. PTPRK (protein-tyrosine phosphatase receptor-κ) dephosphorylates EGFR, the prototypic receptor protein tyrosine kinase, and thereby regulates growth and survival.68 DOCK4 promotes β-catenin stability and is disrupted during tumorigenesis through mutation and deletion in several human cancers.69, 70 In addition, two known therapeutic targets differentially expressed in RT include SPPI and COL18A1. SPP1 (osteopontin) is a secreted phosphoprotein involved in immunity, angiogenesis, cell migration, and cell survival. Overexpression of SPP1 is associated with aggressiveness and metastasis in many different tumor types and is an independent predictor of behavior in melanomas, a tumor derived from the neural crest.71 Inhibition of SPP1 expression can reverse this phenotype.72 Previous reports have demonstrated elevated SPP1 plasma levels and expression levels in RT.73 SPP1 is cleaved into biologically active fragments by MMP12.74 Our study shows both SPP1 and MMP12 to be upregulated in RT. C-terminal cleavage of COL18A1 (downregulated in RT) results in the protein endostatin, a potent endogenous inhibitor of angiogenesis, migration, and invasion (previously reviewed by Abdollahi et al75). In clinical trials, recombinant endostatin inhibited the growth of a variety of tumors while exhibiting no apparent toxic side effects.76

Other genes involved in tumorigenesis or tumor progression were differentially expressed in RT. PEDF (or SERPINF1) has neuroprotective and antiangiogenic effects, suppresses cell-cycle progression, and supports neural stem cell self-renewal.77 Notably, PEDF protein is broadly expressed in the nervous system and circulates in plasma and cerebrospinal fluid and therefore may be useful as a diagnostic marker. LGALS1 (upregulated in RT in both our study as well as that of Pomeroy et al15) also has a well-documented role in tumorigenesis, metastasis, and invasion.78

Conclusions

The data presented suggest that RT arise within an early progenitor population during a critical developmental window. In these cells, loss of SMARCB1 results in (1) repression of neural development, (2) transcriptional repression that may be mediated by dysregulation of, or loss of interaction with, the trithorax/polycomb groups, (3) silencing of cyclin-dependent kinase inhibitors, and (4) differential expression of a number of prominent genes that are known to promote malignant behavior and represent potential therapeutic targets.