PRC2-Mediated Transcriptomic Alterations at the Embryonic Stage Govern Tumorigenesis and Clinical Outcome in MYCN-Driven Neuroblastoma

Neuroblastoma is an embryonal childhood malignancy that originates in sympathoadrenal progenitors derived from migrating neural crest stem cells. Thus, this disease occurs predominantly in the adrenal medulla and the sympathetic ganglia (1). Unlike in adult tumors, somatic mutations are rare in neuroblastoma and other childhood cancers (2, 3). Besides, intensive genetic analyses have identified diverse genetic variations in neuroblastoma, including MYCN amplification, aberrant copy number alterations (17q gain, 11q loss, etc.), single nucleotide variants (ALK, ATRX, etc.), chromothripsis, TERT rearrangements (3–5), and the disruption of the let-7 miRNA family (6). However, the precise mechanisms regulating causative genomic alterations and epigenetic deregulation in human neuroblastoma are poorly understood. Because neuroblastoma is a developmental disorder, its tumorigenic events may be accompanied by normal developmental programs of the sympathoadrenal cell lineage. Therefore, we may not be able to elucidate the mechanisms of neuroblastoma tumorigenesis without investigating the early stage of this disease, particularly during embryogenesis. In terms of causative events, the exogenous expression of several genes, such as MYCN (7), LIN28B (8), and mutant ALK (9), in genetically engineered mouse models, results in the development of neuroblastoma that mimics the clinical presentation of this disease (10). However, the oncogenic roles of these changes in early pathogenesis, especially during the embryonic stages, remain poorly understood, partly due to the lack of a proper experimental method that selectively isolates transformed tumor cells obtained during the early stage of tumorigenesis.

To address the abovementioned challenges, we utilized the MYCN transgenic (TH-MYCN) mouse neuroblastoma model (7), wherein the expression of human MYCN by the rat tyrosine hydroxylase promoter results in neuroblastoma formation that is observable in the postnatal sympathetic ganglia. Spheroid culture is a powerful in vitro experimental tool to isolate proliferative cells, such as cancer cells, and this system provides experimental opportunities to address cellular and molecular mechanisms in cancer (11). We and others have reported spheroid culture conditions for neuroblastoma cells from TH-MYCN mice or human neuroblastoma patients; however, these methods are only applicable to the culture of later-stage neuroblastoma cells (11, 12).

In this study, we report a spheroid culture condition to enrich early-stage tumor cells from TH-MYCN mice and transcriptomic, epigenomic, and genomic analyses of the early pathogenesis in neuroblastoma. Using this model, we identified the cellular and molecular events on embryonic day 13.5 (E13.5) in TH-MYCN mice, including the upregulation of MYC targets and deregulation of polycomb repressive complex 2 (PRC2) targets, which were essential for neuroblastoma tumorigenesis and strongly impacted later-stage human neuroblastoma.

TH-MYCN mice (7) on a 129^+Ter/SvJcl mice background (CLEA Japan, Inc.) were maintained in a pathogen-free, temperature-controlled environment with a 12-hour light/dark cycle and fed mouse feeder pellets and water ad libitum at our animal facility. All animal experiments were approved by the Animal Care and Use Committee of the Nagoya University Graduate School of Medicine (Nagoya, Japan).

The components of RA (+) and RA (−) media are summarized in Supplementary Fig. S1A. Homemade chick embryo extract was prepared as described previously (13) and described in detail in Supplementary Materials and Methods. For primary culture, the embryonic superior mesenteric ganglion (SMG) was enzymatically digested with TrypLE (12563011, Thermo Fisher Scientific) containing 0.25 mg/mL Collagenase Type IV (C5138, Sigma-Aldrich) for 15 minutes at 37°C. The postnatal SMG was enzymatically digested first with 2.5 mg/mL Collagenase Type IV for 20 minutes at 37°C, followed by TrypLE for 20 minutes at 37°C. These digested tissues were further dissociated by pipetting with fire-polished Pasteur pipettes in quenching solution [L-15 Medium (11415064, Thermo Fisher Scientific) containing 1% BSA Fraction V Solution (15260037, Thermo Fisher Scientific), 10 mmol/L HEPES (15630106, Thermo Fisher Scientific), and penicillin–streptomycin (15140148, Thermo Fisher Scientific)] containing 0.1 mg/mL DNase I (D4527, Sigma-Aldrich) and 10 mmol/L MgCl₂ (20908-65, Nacalai Tesque). The dissociated cells were filtered by passing them through 35-μm cell strainers (352235, Corning). Single cells were cultured on either low-attachment PrimeSurface dishes (Sumitomo Bakelite) or nontreated culture dishes (Iwaki) at 37°C in a humidified incubator containing 5% CO₂.

The medium was changed every 3 to 4 days. When the spheres reached 200–300 μm in diameter, they were passaged by digesting them with StemPro Accutase (A1110501, Thermo Fisher Scientific) and subsequently dissociated by pipetting them with fine-tipped pipettes. To evaluate sphere sizes, single cells were plated at low density (1–2.5 cells/μL) and allowed to grow clonally. An inverted microscope (IX81, Olympus) was used to capture images. The sphere sizes were measured, and the sphere numbers were counted manually using the cellSens software (Olympus).

Single-cell suspensions of E13.5 and 3-week-old TH-MYCN^+/− spheres were prepared as described earlier. Single-cell suspensions from TH-MYCN^+/− tumor tissues were isolated as previously described (14). The dorsal flanks of wild-type mice (7–8 weeks of age, male and female mixed evenly) were subcutaneously transplanted with 1 × 10⁵ cells in 50% Matrigel (356234, Corning). The mice were monitored for tumor growth every week, and tumor size was measured using digital calipers. The tumor volumes were calculated with the following formula: volume (mm³) = (length × (width)²)/2.

Array comparative genomic hybridization (arrayCGH), methyl-CpG binding domain (MBD) protein-enriched genome sequencing (MBD-seq), and microarray analysis are described in detail in Supplementary Materials and Methods. A gene ontology analysis was conducted against gene ontology database with Bonferroni correction using the PANTHER overrepresentation test on the PANTHER classification system (15). A gene set enrichment analysis was conducted using GSEA software with default settings (16, 17). Publicly available data for human neuroblastoma analysis were analyzed using R. Gene expression matrix and annotation information of 498 neuroblastoma samples obtained from GSE49710 and GSE62564. A principle component analysis (Bioconductor: pcaMethods) was used to calculate the first principle component for a set of genes, which was considered a signature score. Graphs were drawn using the beeswarm package and survival package.

Chromatin immunoprecipitation for H3K27me3 was performed as described previously (18) and described in detail in Supplementary Materials and Methods. The genomic regions with respect to transcription start site (TSS) and transcription termination site (TTS; regions from TSS-3K to TTS+3K) for each gene were partitioned into nonoverlapping subregions (bins) of 300 bps, and the raw reads were assigned to each bin. The number of reads was further quantile-normalized to adjust sample variations. The average H3K27me3 peak profiles were calculated and plotted using ngs.plot package.

Knockdown of Ezh2 using short hairpin RNA and Ezh2 inhibitor treatment are described in Supplementary Materials and Methods.

qPCR, Western blotting, immunoprecipitation, overexpression of MYCN, hematoxylin and eosin (H&E) staining, immunofluorescence analysis, IHC, in situ hybridization, and statistical analysis are described in Supplementary Materials and Methods.

Data obtained in this study have been deposited in NCBI's Gene Expression Omnibus (Edgar et al., 2002) and are accessible through GEO Series accession number GSE87784 and GSE89741.

The postnatal sympathetic ganglia in TH-MYCN mice, f.e., the SMG (Supplementary Fig. S1B), contain undifferentiated/proliferating neuroblasts positive for Phox2b (a sympathetic lineage marker), Ki67 (a proliferation marker), and N-Myc (Fig. 1A; Supplementary Fig. S1C; refs. 19, 20). The initial step of this study consisted of establishing suitable spheroid culture conditions that enrich and expand these early-stage neuroblasts in vitro. We attempted to use a culture condition that was used to maintain derivatives of neural crest stem cells, such as sympathetic and enteric neural progenitors (21). The medium contained retinoic acid (RA), which is known to induce neuroblastoma differentiation and is used to treat high-risk patients in the clinic (22). Thus, we prepared medium containing RA [RA (+) medium] and RA-free medium [RA (−) medium; Supplementary Fig. S1A].

Cells from the SMG were cultured in either RA (+) or RA (−) medium to evaluate their sphere-forming ability. Primary spheres were formed in both conditions (Fig. 1B). Spheres in RA (+) medium exhibited neurite-like structures, an indication of differentiation, whereas those in RA (−) medium exhibited a distinct round shape (Fig. 1B). The mean sphere size that emerged in RA (−) medium was significantly larger than those that developed in RA (+) medium, and RA (−) spheres were maintained even after several passages (Fig. 1B and C; Supplementary Fig. S1D). We did not observe differences in the expression levels of the human MYCN transgene between the two media (Fig. 1D). Neuroblasts positive for Phox2b were successfully isolated from TH-MYCN^+/− SMG in both media, but the population of neuroblasts positive for Ki67 was significantly lower in RA (+) spheres than in RA (−) spheres (Fig. 1E). We further performed a microarray analysis of RA (+) and RA (−) spheres and identified significantly differentially expressed genes (Supplementary Fig. S1E and S1F; Supplementary Table S1). The data collectively indicated that treating undifferentiated neuroblasts from TH-MYCN mice with RA accelerates transcriptomic changes that promote cell differentiation; thus, RA (−) medium is suitable for maintaining proliferative neuroblasts in vitro.

Sphere formation was observed in primary culture and subsequent passages from the SMG of 3-week-old TH-MYCN^+/− mice but not that from WT mice in RA (−) medium (Fig. 2A). We performed a microarray analysis to investigate the transcriptomic differences among 3-week-old WT SMG, TH-MYCN^+/− SMG, and TH-MYCN^+/− spheres. A principal component analysis revealed that the PC1 component (85.6% contribution) clearly separated three sample groups, wherein TH-MYCN^+/− primary and passage spheres were clustered together (Fig. 2B). This clear separation was also evidenced by an unsupervised hierarchical clustering analysis according to the differentially expressed probes among the four groups (Fig. 2C; Supplementary Table S2). It clustered the four sample groups except for one TH-MYCN^+/− SMG, which was clustered with WT SMG, likely due to the lower number of neuroblasts in the original SMG tissue (Fig. 2C). In addition, the clustering grouped genes into three major groups, groups A (2,049 genes), B (3,713 genes), and C (195 genes; Fig. 2C; Supplementary Fig. S2A). The genes in group A were upregulated in the TH-MYCN^+/− SMG and spheres, including genes that positively regulate neuroblastoma development, such as Lin28b (8), Alk (9), Bmi1 (23), Mybl2 (24), Lmo3 (25), Bdnf (26), Aurka (27), FACT (facilitates chromatin transcription, composed of Supt16 and Ssrp1; ref. 28), Mdk (20), and Neurod1 (29). The genes in group B were strongly downregulated in TH-MYCN^+/− spheres, including genes that are negatively associated with neuroblastoma development, such as Ntrk1/2/3 (26), Ngfr (30), Casz1 (30), and Clu (30). The genes in group C were only upregulated in the TH-MYCN^+/− SMG, including several immune-related genes, such as Ccl3/4, Cxcl1, Ifna11, Il1a/b, and Tlr1; this pattern was likely due to infiltrating immune cells, such as myeloid cells (31) and T cells (32). A gene ontology analysis showed enrichment in cell proliferation–related GO terms in group A, extracellular molecule-related GO terms in group B, and cytokine activity in group C (Supplementary Fig. S2B; Supplementary Table S2), which were consistent with the functions of the genes listed above. Together, these results suggest that undifferentiated neuroblasts from the 3-week-old TH-MYCN SMG were selectively enriched and almost stably maintained as spheres in vitro, while expressing molecules characteristic of neuroblastoma. Thus, spheroid culture in RA (−) medium is an ideal experimental tool to address the molecular mechanisms of early neuroblastoma tumorigenesis.

We next investigated the initial timing of MYCN-driven tumorigenesis and the critical molecular events in its early pathogenesis in TH-MYCN mice. We examined the transverse plane of the E13.5 SMG region, where Phox2b-positive sympathoadrenal progenitors were clustered around the dorsal aorta in both WT and TH-MYCN^+/− mice (Fig. 3A). A specific antisense RNA probe for human MYCN mRNA (Supplementary Fig. S3A) detected human MYCN mRNA expression in a small subset of neuroblasts in the E13.5 TH-MYCN^+/− SMG region (Fig. 3B; Supplementary Fig. S3B). The number of MYCN-positive cells increased on postnatal day 0 SMG and further increased in 2-week-old TH-MYCN^+/− SMG (Fig. 3B).

We next investigated the sphere-forming ability of E13.5 WT and TH-MYCN^+/− neuroblasts. Because normal sympathoadrenal progenitors can proliferate at E13.5, spheres formed from both E13.5 WT and TH-MYCN^+/− SMG, and these spheres did not differ in cellular morphology and mean sphere size (Fig. 3C and D). However, E13.5 TH-MYCN^+/− spheres, but not E13.5 WT spheres, were maintained after several passages (Fig. 3E). All cells in E13.5 WT primary and TH-MYCN^+/− passaged spheres were positive for Phox2b, and thus they are sympathoadrenal lineage (Fig. 3F). All TH-MYCN^+/− sphere cells were positive for N-Myc, and the ratio of Ki-67 positive cells were higher than that of E13.5 WT sphere cells, suggesting that MYCN-positive transformed cells were selected by subculture. Accordingly, human MYCN expression was increased by passages (Supplementary Fig. S3C). WT primary spheres formed from the E13.5 and postnatal day 0 SMG, but these spheres could not be passaged (Fig. 3G). In contrast, TH-MYCN^+/− spheres formed from all different stages and were maintained even after several passages (Fig. 3G). To evaluate the tumorigenicity of these spheres, E13.5 and 3-week-old TH-MYCN^+/− spheres were subcutaneously transplanted into WT mice. Allografts of TH-MYCN^+/− tumor cells (1 × 10⁵ cells) developed into subcutaneous tumors within 4 weeks of transplantation (Fig. 3H). Surprisingly, E13.5 and 3-week-old sphere cells (1 × 10⁵ cells) also formed subcutaneous tumors that progressed slower than tumor cell allografts (Fig. 3H). All transplanted tumors were undifferentiated neuroblastoma, and we did not observe obvious histologic differences among these tumors (Supplementary Fig. S3D). These results collectively demonstrate that MYCN-driven tumorigenesis is observable on as early as E13.5 in the TH-MYCN SMG; thus, critical molecular events in the early pathogenesis of neuroblastoma should be captured by analyzing E13.5 TH-MYCN spheres.

We compared the transcriptomes of E13.5 WT and TH-MYCN^+/− spheres using a microarray analysis. Genes positively associated with neuroblastoma, such as Neurod1, Lmo3, Bdnf, Bmi1, Lin28b, Mybl2, Lmo1, and FACT (Ssrp1 and Supt16), were significantly upregulated, and genes negatively associated with neuroblastoma, such as Ngfr, Clu, and Ntrk3, were significantly downregulated in E13.5 TH-MYCN^+/− spheres (Fig. 4A). In addition, most of the well-defined 51 MYC core targets (28, 33) were significantly upregulated in E13.5 TH-MYCN^+/− spheres (Fig. 4A). A GO analysis revealed the enrichment of cell proliferation-related GO terms in upregulated genes and the enrichment of extracellular compartment- and development-related GO terms in downregulated genes (Fig. 4B; Supplementary Table S3). To reveal the upstream regulatory mechanisms, we focused on well-defined molecular target gene sets created by Ben-Porath and colleagues, including targets of MYC, PRC2, and NOS (NANOG, OCT4, and SOX2), which are associated with human embryonic stem cell identity and certain types of cancers (34). A gene set enrichment analysis revealed the significant enrichment of MYC target gene sets in upregulated genes (Fig. 4C; Supplementary Table S4). Remarkably, PRC2 target gene sets were significantly enriched in downregulated gene sets (Fig. 4C; Supplementary Table S4). The OCT4 target gene set was also enriched in downregulated genes, but the NANOG and SOX2 target gene sets were not enriched (Supplementary Table S4). The absolute values of the normalized enrichment scores of PRC2 target gene sets (approximately 1.7–1.8) were higher than those of the MYC target gene set (approximately 1.4) and OCT4 target gene set (approximately 1.6). These results suggest that PRC2 target genes more significantly contribute to the differential expression patterns between E13.5 WT and TH-MYCN^+/− spheres. Indeed, the majority of differentially expressed PRC2 targets were significantly downregulated in E13.5 TH-MYCN^+/− spheres (Fig. 4D; Supplementary Table S5). The PRC2 major components Ezh2, Eed, and Suz12 were slightly upregulated, whereas the Ezh2 family gene Ezh1 was downregulated in E13.5 TH-MYCN^+/− spheres, suggesting that Ezh2 is the enzyme primarily responsible for PRC2 in this cellular context (Supplementary Fig. S4A). The expressions of PRC2 components and a set of differentially expressed genes were validated by qPCR (Supplementary Fig. S4B). But, Ezh2 protein only showed a tendency to increase (Supplementary Fig. S4C). These results suggest that the differential expression of PRC2 components may not directly contribute to the differential expression of PRC2 target genes. On the other hand, exogenous expression of MYCN in E13.5 WT spheres resulted in increase in Ezh2 expression (Supplementary Fig. S4D), suggesting that N-Myc promotes the transcription of Ezh2.

We performed chromatin immunoprecipitation sequencing (ChIP-seq) for the tri-methylation of Histone H3 at lysine 27 (H3K27me3), which is a histone mark modified by PRC2 and results in gene silencing (35). H3K27me3 was strongly enriched at the promoter region of PRC2 targets compared with non-PRC2 targets (Fig 4E). Importantly, the H3K27me3 enrichment was increased in E13.5 TH-MYCN^+/− spheres and further elevated in TH-MYCN^+/− tumor spheres, suggesting that the PRC2 target genes were transcriptionally suppressed by Ezh2-mediated H3K27me3 modification. Physical association between N-Myc and Ezh2, and the transcriptional suppression of PRC2 target genes by the complex were previously reported (36). Indeed, N-Myc, Ezh2, and Suz12 proteins were physically associated in E13.5 TH-MYCN^+/− spheres evidenced by coimmunoprecipitation (Fig. 4F), suggesting that N-Myc regulates the function of PRC2 for example by recruiting PRC2 on certain genomic regions.

We also performed array comparative genomic hybridization. As previously reported (37), segmental chromosomal gains or losses were identified in TH-MYCN^+/− tumor tissues and spheres; however, they were not observed in E13.5 and 3-week-old TH-MYCN^+/− spheres (Supplementary Table S6), suggesting that copy number alterations were not a major cause of neuroblastoma tumorigenesis in TH-MYCN mice. DNA methylation of promoter CpG islands is a gene-silencing mechanism, and we assumed that promoter-associated DNA methylations might affect the differential expression pattern between E13.5 WT and TH-MYCN^+/− spheres. Thus, we performed methyl-CpG binding domain protein-enriched genome sequencing (MBD-seq). We considered ± 500-bp regions of TSSs as promoter regions and identified 23 and 73 specifically promoter-methylated genes in E13.5 WT and TH-MYCN^+/− spheres, respectively (Supplementary Fig. S5A and S5B). We then assessed differences in the expression levels of promoter-methylated genes between E13.5 WT and TH-MYCN^+/− spheres; however, the expression levels of most genes remained unchanged (Supplementary Fig. S5C–S5E). Thus, we concluded that the DNA methylation of promoter CpG islands did not contribute to transcriptomic changes. Collectively, these results highlight the contribution of PRC2 in the transcriptome to early neuroblastoma pathogenesis in TH-MYCN mice.

We next investigated the role of Ezh2 in early tumorigenesis because it is the enzyme responsible for the function of PRC2 (35). We used lentivirus-mediated shRNAs targeting Ezh2 to impair Ezh2 function. Four of 5 different shRNAs decreased the Ezh2 protein levels by over 50% in 3 days after infection (Fig. 5A). The knockdown of Ezh2 by two different shRNAs (#4 and #5) drastically suppressed E13.5 TH-MYCN^+/− sphere formation (Fig. 5B). We next used a potent and selective Ezh2 inhibitor, EPZ-6438, which inhibits the histone methyltransferase activities of EZH2, leading to a decrease in H3K27me3 (38, 39). Treatment with EPZ-6438 resulted in dose-dependent decreases in the H3K27me3 levels (Fig. 5C). Accordingly, the mean sphere sizes were significantly and dose-dependently decreased in EPZ-6438-treated E13.5 TH-MYCN^+/− spheres (Fig. 5D). The average cell growth IC₅₀ values of EPZ-6438 treatment were approximately 1–2 μmol/L for the three different sphere types, that is, E13.5 TH-MYCN^+/−, E13.5 TH-MYCN homozygote (TH-MYCN^+/+), and 3-week-old TH-MYCN^+/− spheres (Supplementary Fig. S6). The expressions of PRC2 targets including cyclin-dependent kinase inhibitors (40) and Ezh2 targets in neuroblastoma (30) were broadly downregulated in E13.5 TH-MYCN^+/− spheres, and recovered at the comparable levels to that in E13.5 WT spheres by EPZ-6438 treatment (Fig. 5E). Together, these results demonstrate that the function of Ezh2 maintains the cell identity of transformed neuroblastoma cells in vitro by epigenetically repressing the expression of its target genes.

IHC staining showed that Ezh2 was expressed in sympathoadrenal progenitors in both the E13.5 WT and TH-MYCN^+/− SMG (Fig. 6A). The Ezh2 expression levels in these cells were comparable with those in surrounding cells, suggesting that Ezh2 is widely and similarly expressed in different types of cells in the E13.5 transverse plane around the dorsal aorta (Fig. 6A). In contrast, Ezh2 expression was minimal in differentiated ganglion cells both in the 3-week-old WT and TH-MYCN^+/− SMG, but highly expressed in undifferentiated neuroblasts from the TH-MYCN^+/− SMG (Fig. 6A). Higher Ezh2 protein levels in the 3-week-old TH-MYCN^+/− SMG were also observed by Western blotting (Fig. 6B).

We next evaluated the potential of Ezh2 as a target for neuroblastoma treatment. TH-MYCN^+/+ mice were used in this experiment because tumor formation is stable; all mice exhibited enlarged tumors in the SMG area at three weeks of age and died due to the tumor at 7 to 8 weeks of age (Fig. 6C). We used EPZ-6438 because its efficacy has already been demonstrated in certain types of cancers, including rhabdoid tumors (38) and non-Hodgkin lymphoma (39), and it is currently being evaluated in a phase I/II clinical trial (ClinicalTrials.gov Identifier: NCT01897571). We followed the treatment procedures described by Knutson and colleagues (39) and selected 300 mg/kg per day as a suitable dose for this study. EPZ-6438 drastically reduced H3K27me3 levels in the tumor mass (Fig. 6D). Accordingly, we observed significant in situ tumor growth suppression in TH-MYCN^+/+ mice by the treatment (Fig. 6E; Supplementary Fig. S7A). However, tumor cells were observed without obvious histologic changes (Supplementary Fig. S7B), although the drug inhibited the function of Ezh2 also evidenced by H3K27me3 staining (Fig. 6F; Supplementary Fig. S7C), suggesting that the treatment did not lead to complete regression. Accordingly, cessation of the EPZ-6438 treatment led to tumor regrowth and resulted in no obvious extension of overall survival (Supplementary Fig. S7D). We did not observe weight loss and other visible adverse effects. To properly assess the benefit of EPZ-6438 to the survival, treatment regimens such as the dose, drug delivery system, combination therapies, and treatment schedule should be considered in future. Collectively, these results clearly demonstrated that Ezh2 is highly expressed in MYCN-transformed neuroblastoma in vivo and is a potential target for neuroblastoma treatment.

We also examined the early transcriptomic changes found in TH-MYCN mice in human neuroblastomas. To this end, we utilized publicly available datasets (gene expression of 498 neuroblastoma samples), and calculated the signature scores based on the expression levels of a set of genes using a principle component analysis, and used the first principle component values as the signature scores. The signature scores for genes that were significantly (over 2-fold) upregulated (191 genes) and downregulated (1,563 genes) between E13.5 TH-MYCN^+/− and WT spheres were calculated, and these changes represented the transcriptomic changes that occur during early neuroblastoma tumorigenesis. In addition, the signature scores for MYC core targets (51 genes) and PRC2 targets (654 genes) were also calculated. The signature scores of up- and downregulated genes in E13.5 TH-MYCN^+/− spheres positively (r = 0.76) and negatively (r = −0.56) correlated with the signature score of MYC core targets, respectively (Fig. 7A). Importantly, the signature score of PRC2 targets was strongly and negatively associated with the signature score of MYC core targets (r = −0.84; Fig. 7A). These correlations were stronger than those with the expression levels of MYCN (Fig. 7A). Moreover, these signature scores were strongly associated with several clinical statuses, including the MYCN status, RISK factors, and INSS staging categories (Fig. 7B). We next investigated the ability of these scores to predict patient prognosis. Specifically, a high MYC target signature was associated with poor prognosis (P = 2.8 × 10⁻²⁶), and the prognostic power of this signature was stronger than that of MYCN expression (P = 9.0 × 10⁻⁶), which is consistent with the findings of previous studies (Fig. 7C; refs. 28, 41). Notably, the signature scores of up- and downregulated genes in E13.5 TH-MYCN^+/− spheres as well as the signature score of PRC2 targets were also able to predict patient prognosis (Fig. 7C). The expression levels of PRC2 components, that is, EZH2, EED and SUZ12, were not strongly associated with the signature score for MYC targets and MYCN expression (Supplementary Fig. S8A), and did not predict patient prognosis (Supplementary Fig. S8B). These results indicate that the transcriptomic characteristics, including the deregulation of PRC2 targets, during early neuroblastoma pathogenesis in TH-MYCN mice are sustained and are critical determinants of the malignant phenotype of neuroblastoma.

The targeted expression of oncogenes in sympathoadrenal lineage cells in several mouse models results in neuroblastoma formation (7–9). However, the detailed molecular mechanisms of the cellular context-dependent oncogenic events have not yet been investigated. We found that tumorigenic cells were observable as early as E13.5 in TH-MYCN mice. This cellular event was accompanied by transcriptomic alterations, that is, the up- and downregulation of gene sets, the activation of MYC targets, and the deregulation of PRC2 targets. Because PRC2 inhibition suppressed the growth of neuroblastoma cells in vitro and markedly reduced tumor mass in vivo, PRC2 should be an essential player in neuroblastoma tumorigenesis in TH-MYCN mice. Notably, the activation of MYC targets and deregulation of PRC2 targets predicted the prognosis of human neuroblastoma. Overall, our study revealed cellular events that initiate tumorigenesis at the mid-gestation period in TH-MYCN mice and molecular events in which early-stage transcriptional alterations, including the deregulation of MYC and PRC2 targets, govern tumorigenesis and the malignant phenotype of human neuroblastoma.

Changes in the DNA methylome are a characteristic epigenetic deregulation in neuroblastoma. Notably, the presence of a CpG island methylator phenotype is associated with MYCN amplification and high-risk neuroblastoma (42, 43), and a recent study suggested the cooccurrence of DNA hypermethylation and an increase in H3K27me3 marks on the same genomic locus in human neuroblastoma (43). Furthermore, PRC2 has been reported to physically interact with DNA methyltransferases to directly regulate DNA methylation (44). A physical interaction has also been identified between MYCN and PRC2 in neuroblastoma (36). We also observed it in early neuroblastoma (Fig. 4F). Although these lines of evidence suggest a link between DNA methylation and PRC2 function, we did not observe a strong association among differences in mRNA expression, DNA methylation, and H3K27me3 modifications in E13.5 TH-MYCN spheres (Supplementary Fig. S5D and S5E). Therefore, our results particularly highlight the essential role of PRC2 and H3K27me3 among various types of epigenetic regulations during early tumorigenesis in TH-MYCN mice, although the axis of MYCN–PRC2–DNA methylation in neuroblastoma development warrants further investigation.

The MYCN and MYC target signatures are strongly associated with the malignant phenotype of neuroblastoma (28); however, directly targeting these transcription factors with small-molecule inhibitors is challenging, as widely discussed in the literature (45). In this study, we clarified that the specific inhibition of the histone methyltransferase activity of Ezh2 suppressed tumor growth in TH-MYCN mice, and the MYC target signature and PRC2 target signature strongly correlated in human neuroblastoma. Therefore, the disruption of PRC2 function, such as the inhibition of Ezh2 by EPZ-6438, may provide a promising option for the treatment of human neuroblastoma.

No potential conflicts of interest were disclosed.

Conception and design: S. Tsubota, S. Kishida, K. Kadomatsu

Development of methodology: S. Tsubota, D. Cao

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): S. Tsubota, M. Ohira, S. Yamashita, T. Ushijima

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): S. Tsubota, T. Shimamura, S. Yamashita

Writing, review, and/or revision of the manuscript: S. Tsubota, S. Kishida, D. Cao, K. Kadomatsu

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): M. Ohira, S. Kiyonari

Study supervision: K. Kadomatsu

We thank Dr. Hideki Enomoto in Kobe University Graduate School of Medicine for providing anti-Phox2b antibody and the experimental procedure of spheroid culture method, and Ayaka Hatano for her technical assistance.

S. Tsubota was supported by JSPS KAKENHI grant JP14J00157. S. Kishida was supported by JSPS KAKENHI grant JP24590377. T. Ushijima and K. Kadomatsu were supported by a grant for the Practical Research for Innovative Cancer Control from Japan Agency for Medical Research and Development (16ck0106011h0003). K. Kadomatsu was supported by JSPS KAKENHI grant JP15k15079 and CREST, JST (15656320).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.