Sleeping Beauty Insertional Mutagenesis Reveals Important Genetic Drivers of Central Nervous System Embryonal Tumors

Embryonal tumors, including medulloblastoma and central nervous system primitive neuroectodermal tumors (CNS-PNET), represent the most common malignant pediatric brain tumors (1). For ease of historic comparison, CNS-PNET is used in this article according to the 2007 World Health Organization CNS tumor classification and includes CNS neuroblastomas, CNS ganglioneuroblastomas, medulloepitheliomas, and ependymoblastomas, although CNS-PNET no longer exists as an umbrella term (2). Medulloblastoma and CNS-PNET have similar histology: densely packed, small cells with hyperchromatic nuclei and little cytoplasm. Medulloblastomas are usually cerebellar, while CNS-PNETs occur predominantly in the cerebrum. Aggressive, multimodality treatments improve survival but produce lifelong side effects, and 5-year survival rates remain 60%–65% for medulloblastoma and 20%–40% for CNS-PNET (3).

Medulloblastoma and CNS-PNET are molecularly heterogeneous. Medulloblastoma includes four molecular subgroups: WNT, Sonic hedgehog (SHH), group 3, and group 4; WNT and SHH are associated with mutations activating those pathways, but groups 3 and 4 remain less defined (4). A genomic study by Picard and colleagues identified three distinct CNS-PNET subgroups: primitive-neural, oligo-neural, and mesenchymal (5). Using methylation- and gene expression–based analyses, Sturm and colleagues identified four molecular subgroups of CNS-PNET associated with gene fusions (6). While our understanding of the tumor biology has improved, a lack of animal models and targetable oncogenic drivers impede therapeutic development, particularly in group 3/4 medulloblastoma and CNS-PNET.

We used Sleeping Beauty (SB) transposon mutagenesis to identify novel medulloblastoma and CNS-PNET drivers. Transposition initiated in neural progenitor cells using Nestin-Cre was used alone, with Trp53^lsl-R270H/+, or with Pten deficiency to generate medulloblastomas and CNS-PNETs. These tumors resembled human medulloblastoma and CNS-PNET histologically and transcriptionally. Three candidate oncogenes, Arhgap36, Foxr2, and Megf10 were validated in vitro and in vivo and their mechanisms examined.

Animal studies were conducted using procedures approved and monitored by the Institutional Animal Care and Use Committee at the University of Minnesota (UofMN, Minneapolis, MN). Nestin-Cre mice (7) were bred to either T2/Onc (chromosome 1/15; ref. 8) or T2/Onc2 (chromosome 4; ref. 9) to generate Nestin-Cre:T2/Onc(2). Rosa26^lsl-SB11/+ (10) were bred to either Trp53^lsl-R270H/+ (11) or Pten^flox/flox (12) to generate Rosa26^lsl-SB11/+:Pten^flox/flox or Rosa26^lsl-SB11/+:Trp53^lsl-R270H/+. Nestin-Cre:T2/Onc(2) mice were bred to Rosa26^lsl-SB11/+:Pten^flox/flox or Rosa26^lsl-SB11/+:Trp53^lsl-R270H/+ to generate mice with and without CNS-restricted SB mutagenesis on wild-type (WT), Trp53^lsl-R270H/+, and Pten^flox/+ backgrounds. T2/Onc(2) excision PCR was performed as described with primers in Supplementary Table S1 (8).

Unstained tissue microarray (TMA) sections of formalin-fixed, paraffin-embedded (FFPE) human tumor specimens were obtained through the University of Minnesota Materials Procurement Network (Minneapolis, MN; 11 samples) and Johns Hopkins University (Baltimore, MD; 54 samples). FFPE tissue slides were stained with hematoxylin and eosin or IHC using standard methods. Immunoprecipitations were done (Active Motif #54001) with 500 μg total protein. Western blotting was done with whole-cell lysates as described previously (13) using antibodies in Supplementary Table S2.

Ligation-mediated PCR to identify transposon insertion sites was performed as described previously (13). Transposon insertion sites were annotated using TAPDANCE (14). Nonredundant insertion sites representing >0.1% of the mapped insertions from each tumor library were used to generate common insertion sites (CIS; P < 0.05).

Isolated tumor RNA (Qiagen, catalog no. 75114) was assessed for quality using capillary electrophoresis (RIN > 6.5, Agilent 2100 BioAnalyzer). Paired-end sequencing (30–40 million reads/sample) of TruSeq-prepared libraries was performed (Illumina HiSeq 2000). Raw FASTQ files are available at the NCBI Sequence Read Archive and linked to Gene Expression Omnibus SuperSeries (GSE122050). FASTQ files were mapped to the MM10 genome (T2/Onc and Rosa26^lsl-SB11/+ as additional chromosomes; ref. 15) using STAR-Fusion (https://github.com/STAR-Fusion/STAR-Fusion/wiki). Transcript FPKM values were computed using cuffquant and cuffnorm and adjusted by +0.1 (16).

To identify T2/Onc:genome fusions, we analyzed the chimeric.out.junction and chimeric.out.sam output files from STAR-Fusion to summarize the number of junction (one read contains the T2/Onc:genome junction) and bridge (one paired-end read maps to T2/Onc and the other to the genome) reads present within 1,000 bp regions. Fusions supported by ≥1 junction read or ≥3 bridging reads were retained for analysis. Manual detection of T2/Onc(2):Arhgap36 transcripts was done using 500 ng of purified RNA (Invitrogen, catalog no. 15596-018), reverse transcribed (Invitrogen, catalog no. 18080-051) and amplified using primers in Supplementary Table S1.

Gene cluster similarity was used for unsupervised, unbiased identification of similar gene clusters across transcriptional datasets. Transcriptional profile datasets were individually log-transformed, mean-centered, filtered for highly variant genes, and hierarchically clustered using average linkage and (1–Pearson correlation) as the distance metric. Gene clusters with node correlation and size >respective thresholds were retained. Cross-dataset cluster pairs were tested for enrichment of common gene members (Fisher exact test) to identify conserved transcriptional patterns.

For CNS-PNET expression analysis, cDNAs were synthesized (Applied Biosystems, catalog no. 4368814) and qRT-PCR was performed (Invitrogen, catalog no. 4369016). For Shh activation assays, purified cellular RNA (Ambion, catalog no. 12183025) was reverse transcribed (Invitrogen, catalog no. 11755050) and qRT-PCR was done in triplicate (Roche, catalog no. 4673492001). Shh activation was done as described previously (17). For 5′-rapid amplification of cDNA ends (5′-RACE; Ambion, catalog no. AM1700), tumor RNA was extracted from human medulloblastomas (Invitrogen, catalog no. 15596-018) and normal human brain RNA was purchased from BioChain (R1244039-50, R1244035-50, and R1234040-10). Subsequent detection of transcripts by RT-PCR was performed with 500 ng RNA (Invitrogen, catalog no. 18080-051). Primers and probes are listed in Supplementary Table S1.

Cell lines were maintained, authenticated, and tested for Mycoplasma as described in Supplementary Table S3. MTS (Promega, catalog no. G1111), soft agar assays, and transfections were done as described previously (13). Stable lines transfected with cDNAs [ARHGAP36 (Q6ZRI8-5), FOXR2 (Q6PJQ5-1), and Megf10 (Q6DIB5-1)] were cultured as polyclonal populations in puromycin. Transient transfection in HEK293Ts was done as per manufacturer's protocol (Invitrogen, catalog no. 11668019). CRISPR-KO clones were isolated as described previously (13). Briefly, Daoy cells were transfected with PiggyBAC transposase and a puromycin-selectable PiggyBAC transposon vector containing two FOXR2 guide RNAs (sequences in Supplementary Table S1) and Cas9. Isolated clones were sequenced to identify changes in FOXR2. Wound-healing assays were performed as described previously (18). Primary granule neuron precursors (GNP) were isolated from neonatal C57BL/6J and thymidine incorporation assays were performed as described previously (19).

NRG mice (Jackson, catalog no. 007799) were injected as described previously (20). Briefly, C17.2 cells were prepared in Hank's Balanced Salt Solution (HBSS), counted, and stored on ice prior to injection (2 × 10⁵ cells/2 μL injection). P0 mice were injected in the fourth ventricle (stereotactic coordinates: 1.5 mm anterior to Bregma, 1.5-mm deep). Successful injection was verified on P1 by luciferase imaging as described previously (20). Adult intracranial injections were performed as described previously (19). Female NU/J mice (Jackson, catalog no. 002019; 6- to 8-weeks old) were anesthetized (81 mg/kg ketamine, 13.8 mg/kg xylazine) and injected with 1 × 10⁶ cells/5 μL (prepared as above). For flank tumor assays, female NU/J mice (Jackson, catalog no. 002019; 6- to 8-weeks old) were injected with 1 × 10⁶ C17.2 cells (prepared as above) resuspended 1:1 in HBSS and Matrigel (Corning CB, catalog no. 40234C). Tumor volume = (l × w²)/2, l = length and w = width.

To identify genetic drivers of medulloblastoma and CNS-PNET, we targeted Nestin⁺ neural and glial precursor cells with SB mutagenesis on three genetic backgrounds: wild-type (WT), Pten heterozygous (Pten^flox/+), or Trp53 mutant (Trp53^lsl-R270H/+). Pten^flox/+ and Trp53^lsl-R270H/+ served as sensitizing backgrounds as they are mutated in human medulloblastoma and CNS-PNET (21, 22). IHC revealed SB expression throughout the developing brain, including cells within the granule layer, white matter, surrounding the fourth ventricle, subependymal midbrain, subventricular zone, and olfactory bulb (Supplementary Fig. S1A–S1G). Experimental cohorts harbored one of three transposon concatemers (Supplementary Fig. S1H). SB mutagenesis significantly reduced survival in combination with Trp53^lsl-R270H (Supplementary Fig. S1I–S1K). Upon necropsy we observed masses in the brain, testicles, bone, peripheral lymph nodes, and spleen (Supplementary Fig. S1I–S1K; Supplementary Table S4).

Histologic analysis of brain masses revealed the presence of infratentorial medulloblastoma and supratentorial CNS-PNET (22 medulloblastomas and 14 CNS-PNETs) with highest medulloblastoma frequency in Trp53^lsl-R270H/+ mice (Fig. 1A and B). The high-copy transposon (T2Onc2, chromosome 4) produced the highest proportion of medulloblastoma, while CNS-PNETs were equally derived from chromosome 4 and 15 concatemers (Supplementary Fig. S2A). Tumors expressed nuclear SB by IHC and showed transposon mobilization by PCR (Supplementary Fig. S2B and S2C).

SB-induced medulloblastomas originated in the cerebellum whereas CNS-PNETs occurred in the rostral portion of the brain, overwhelming the olfactory bulbs, lateral ventricle, and cortex (Fig. 1A, C, and D). Both tumor types resembled their human counterparts histologically (small round cells with high nuclear to cytoplasmic ratios, Homer Wright rosettes, vascularization, and mitotic figures), expressed diagnostic markers for human medulloblastoma and CNS-PNET (Synaptophysin, Ki67, and Nestin), and stained negatively for the astrocytic marker Gfap (Fig. 1C and D). Medulloblastomas (50%) and 100% of CNS-PNETs exhibited metastatic characteristics, including infiltration into the leptomeninges, parenchyma, brainstem, and cortex (Supplementary Table S4). Leptomeningeal spread is of interest due to its associated poor prognosis, prevalence (one-third of patients), and difficulty in modeling (5, 23).

We transcriptionally profiled 18 SB-induced medulloblastomas, revealing two clear subgroups that each matched human medulloblastoma subtypes (Fig. 2A; ref. 6). The first subgroup (N = 13) showed increased Gli1, canonically associated with human SHH medulloblastoma (24). The second subgroup (N = 5) showed increased Npr3 and Kcna1, markers for human group 3 and 4, respectively (24). More globally, the Gli1-overexpressing medulloblastoma gene set overlapped with highly expressed genes in human SHH medulloblastoma (N = 48; Fig. 2A; Supplementary Table S5; ref. 6). Similarly, the Npr3- and Kcna1-overexpressing mouse gene set overlapped with highly expressed genes in human group 3 and 4 medulloblastoma (N = 120). Mouse tumors did not exhibit Wnt signatures.

Transcriptional profiling comparing mouse CNS-PNETs (N = 5) with published human CNS-PNETs (N = 58) revealed two subgroups (Fig. 2B; ref. 6). A single cluster of covarying genes (N = 298) was significantly enriched in both human and mouse CNS-PNET and contained high levels of CNS neuroblastoma with FOXR2 activation (CNS NB-FOXR2)–associated genes, including MMP24, KCJN9, and CHGB (Fig. 2B; Supplementary Table S6). Consistent with CNS NB-FOXR2 activation classification, SB-induced CNS-PNETs showed significantly increased Olig1, Olig2, and Sox10 by qPCR and were Olig2⁺ by IHC (Supplementary Fig. S2D and S2E). The three remaining mouse CNS-PNETs had elevated expression of CNS EFT-CIC marker genes (Shc4, Argdib, and Pole) but no clearly corresponding human subgroup.

We performed ligation-mediated PCR on 22 medulloblastomas and 13 CNS-PNETs and identified 3.9 × 10⁵ and 15.5 × 10⁴ nonredundant insertions, respectively. TAPDANCE analysis (14) identified 13 medulloblastoma and 15 CNS-PNET DNA-CIS (D-CIS; Fig. 3A; Supplementary Table S7). We also identified RNA-CIS (R-CIS) in both tumor types, defined as transposon fusion transcripts present in both ≥10% of cases and ≥1 tumor (Fig. 3A; Supplementary Table S7). For several putative oncogenes, the presence of a T2/Onc(2) fusion transcript significantly increased expression (Fig. 3B). We identified genes previously implicated in medulloblastoma, including Gli1 and Pten; upregulation of GLI1 expression and PI3K pathway activation through PTEN loss are observed in human medulloblastoma (21). Predicted transposon-mediated driving effects on Gli1 expression were confirmed by IHC (Supplementary Fig. S3A). We also confirmed Pten reduction and a corresponding increase in pAkt with Pten insertions (Supplementary Fig. S3B). Arhgap36 was most frequently modified, with insertions identified in 14 (D-CIS) and 13 (R-CIS) medulloblastomas and two CNS-PNETs (R-CIS). Enox2, a tumor-associated NADH oxidase involved in the growth of several cancer cell lines (25), was also a D- and R-CIS in medulloblastoma (predicted SB oncogene).

We identified known and novel molecular changes in SB-induced CNS-PNETs. Pten was a predicted tumor suppressor gene (TSG) and loss of PTEN through 10q loss or mutation is observed in human CNS-PNET (22). Novel to this study, Nf1 was the most targeted CNS-PNET TSG. We identified several other predicted Ras effector gene alterations, including Eras overexpression and Erbb2ip and Rasa3 disruption (Supplementary Table S4). Tumors harboring these insertions exhibited increased pErk (Supplementary Fig. S3C; Supplementary Table S4) supporting Ras pathway activation. We also observed NF1 locus deletion in a subset of human CNS-PNETs (Supplementary Fig. S3D).

We next analyzed our CIS gene expression in published human medulloblastomas and CNS-PNETs (Fig. 3C–E; ref. 6). GRIA4 showed high expression in both SHH medulloblastoma and CNS NB-FOXR2. FOXR2 is elevated in a subset of WNT medulloblastoma and CNS NB-FOXR2. Interestingly, ARHGAP36 is highly expressed in group 3 and 4 medulloblastoma, with low expression in SHH (Supplementary Fig. S4A), while in the mouse, Arhgap36 insertions occurred in Shh and group 3/4 tumors (7/12 Shh and 4/5 group 3/4 tumors; Supplementary Table S4).

Transposon location and orientation implicate Arhgap36 as an oncogene, with insertions upstream of the locus or within intron 1 and significantly increasing gene expression (Fig. 4A and B). T2/Onc:Arhgap36 transcripts displayed precise fusion of the T2/Onc splice donor to the Arhgap36 exon 2 splice acceptor (Supplementary Fig. S4B), generating a 15 amino acid N-terminal truncation with translation from an in-frame ATG. Tumors with Arhgap36 insertions showed high levels of cytoplasmic Arhgap36 by IHC compared with tumors without Arhgap36 insertions, which displayed sparse nuclear expression similar to normal granule cells (Fig. 4C). Spatial and temporal analysis of ARHGAP36 expression in normal human and mouse cells within developing and mature cerebella showed nuclear localization throughout the molecular layer, Purkinje cell layer, and internal granule cell layer (Supplementary Fig. S4C). In two combined TMAs of human medulloblastoma, 37 of 65 (56%) and 8 of 65 (12%) expressed cytoplasmic and nuclear ARHGAP36 expression, respectively, by IHC (Fig. 4D). ARHGAP36 protein was expressed across all human subgroups, although increased ARHGAP36 transcript was only expressed in group 3/4 human medulloblastoma (Fig. 4E; Supplementary Fig. S4A; ref. 24). Overall and cytoplasmic ARHGAP36 expression correlated with accelerated mortality (Fig. 4F; Supplementary Fig. S4D).

We further investigated ARHGAP36 transcript profiles using 5′-RACE on human group 3/4 medulloblastoma samples, cell lines, and normal cerebellar cells. Several ARHGAP36 amplification products were identified (Supplementary Fig. 4E and F) predicting expression of canonical isoform 1(*), 5(**), and 3(***; ****). All three isoforms contain intact ARHGAP36 predicted functional domains, including an arginine-rich domain (ARR), nuclear localization sequence (NLS), and GTPase-activating protein (GAP) domain. Interestingly, the 5′ ends of *** and **** begin with intron 2 sequences splicing to exon 3 and an in-frame ATG located in exon 4. Target-specific RT-PCR revealed these ARHGAP36 sequences in additional tumor samples, while normal fetal cerebellum only expressed isoform 1 (Supplementary Fig. S4G).

To further characterize the role of ARHGAP36 in medulloblastoma, we overexpressed truncated ARHGAP36 (isoform 5) in the mouse neural progenitor cell line C17.2 (Supplementary Fig. 5A; ref. 26). Increased ARHGAP36 significantly enhanced soft agar colony formation but did not affect proliferation or collective cell migration rate (Fig. 5A; Supplementary Fig. S5B–5SC). C17.2 cells expressing ARHGAP36 formed tumors significantly faster in the flanks of NU/J mice than luciferase control cells (Fig. 5B). When injected orthotopically into adult NU/J mice, C17.2 cells localized to the granule layer of the cerebellum where ARHGAP36 expression drove leptomeningeal spread into the cerebrum and cerebellar tumor formation, reducing median survival from 99 to 71 days (Fig. 5C and D). In addition, increased ARHGAP36 expression in primary GNPs significantly increased their proliferation (Fig. 5E). As previously reported, ARHGAP36 strongly activated Shh signaling in C17.2 cells in a ligand-independent manner, providing a potential mechanism for ARHGAP36-driven tumorigenesis (Fig. 5F; refs. 17, 27). In addition, SB transposon insertions in the Arhgap36 locus were mutually exclusive with insertions in Gli1 and Gli2, Shh pathway activators (Fig. 5F).

Megf10 (predicted SB oncogene) was identified as an R-CIS, with fusion transcripts in three medulloblastomas significantly increasing expression (Fig. 3A and B). Megf10 is expressed throughout the developing CNS and is a positive regulator of Notch signaling (28). MEGF10 is upregulated in a subset of human CNS-PNET and medulloblastoma (15/26 medulloblastomas, 14/58 CNS-PNETs; Supplementary Fig. S5D). Megf10 expression in C17.2 cells significantly enhanced colony formation in soft agar, proliferation by MTS, and flank tumor formation (Fig. 5G–I). In addition, increased Megf10 expression in GNPs increased their proliferation by 1.7-fold, although not significantly possibly due to low sample size (Fig. 5E). Megf10 had no effect on C17.2 cell Notch signaling, however, by Western blot analysis (Supplementary Fig. S5E).

All Foxr2 transposon insertions were located upstream of the translation start site and drove increased Foxr2 expression, predicting an oncogenic role (Fig. 6A and B). The presence of a Foxr2 insertion significantly reduced median survival from 166.5 to 116.5 days (Supplementary Fig. S6A). FOXR2 overexpression in C17.2 cells significantly increased soft agar colony formation and collective cell migration without increasing proliferation (Fig. 6C–E; Supplementary Fig. S6B). C17.2 FOXR2 cells formed flank tumors significantly faster than C17.2 Luc controls (Fig. 6F). When injected orthotopically into adult NU/J mice, C17.2 FOXR2 cells migrated to the granule layer of the cerebellum and formed large, vascular tumors, reducing median survival from 99 to 43 days (Fig. 6G and H; Supplementary Fig. S6C). C17.2 FOXR2 cells injected orthotopically into neonatal NRG mice also significantly reduced median survival compared with C17.2 Luc and resulted in tumor formation (110 vs. 183 days; Fig. 6I). Importantly, FOXR2 overexpression drove increased proliferation in primary GNPs (Fig. 5E). Using the CRISPR/Cas9 system, we knocked out (KO) FOXR2 in Daoy, a human medulloblastoma cell line. Daoy clone no. 21 had a nonsense mutation in exon 1, resulting in FOXR2 protein loss, decreased proliferation, and decreased soft agar colony formation, all rescued by FOXR2 cDNA expression (Fig. 6J–L).

FOXR2 has many suggested oncogenic mechanisms, including interaction with C-MYC (29). We confirmed this interaction by CoIP in a human Schwann cell line, HSC1λ, and C17.2 cells stably expressing FOXR2 (Fig.7A; Supplementary Fig. 6D). To determine whether FOXR2 interacts with other forms of MYC, we transiently transfected HEK293T cells with V5-tagged C-MYC, L-MYC, and N-MYC. We observed reduced interaction of FOXR2 with N-MYC and minimal interaction with L-MYC (Fig. 7B). To determine whether FOXR2 stabilizes C-MYC, we treated cells with cycloheximide to inhibit translation. Almost all C-MYC protein was degraded in control HSC1λ cells, but with FOXR2, C-MYC levels were only reduced by half after 3 hours, indicating FOXR2 is stabilizing (Fig. 7C). Interestingly, C-MYC was highly stable in C17.2 cells regardless of FOXR2 expression, likely due to their immortalization by V-Myc (26), implying that FOXR2 transforms C17.2 cells through alternative mechanisms.

To further characterize the oncogenic mechanism of FOXR2, we synthesized FOXR2 cDNA constructs missing the following predicted domains: NLS (ΔNLS; ref. 30), MYC interaction (ΔMYC; ref. 29), low complexity regions (ΔLC1, ΔLC2, and ΔLC1/2; ref. 31), and forkhead box transcription factor (ΔTF; Fig. 7D; ref. 31). We stably expressed each mutant and performed soft agar assays in C17.2 and HSC1λ cells. Surprisingly, no single-deletion mutant completely ablated the colony formation promoting capacity of FOXR2 in either line, but loss of the Myc interaction and LC2 domains significantly reduced colony formation (Fig. 7D; Supplementary Fig. S6E). C17.2 FOXR2ΔMYC also had an intermediate phenotype in the flank (Fig. 6F). We verified that the FOXR2ΔMYC mutant did not bind C-Myc (Supplementary Fig. S6F). Given no single domain loss completely ablated colony formation but some did reduce it, we conclude FOXR2 has a multifaceted mechanism. We observed a slight change in the actin cytoskeleton of FOXR2-expressing cells, prompting assessment of focal adhesion kinase (FAK) activation. C17.2 cells expressing FOXR2 displayed increased Fak phosphorylation (Y397), resulting in increased (activating) phosphorylation at Src Y416 (Fig. 7E). This effect is Myc-independent (Supplementary Fig. S6G). Correspondingly, FOXR2 loss in Daoy cells resulted in decreased pFAK and pSRC rescuable by FOXR2 cDNA expression (Fig. 7E).

We used SB transposon mutagenesis to identify novel drivers of medulloblastoma and CNS-PNET. Over half of our D-CIS and several of our R-CIS were reported in previous SB medulloblastoma screens (23, 32–34), including Pten, Wac, Arid1b, Arhgap36, Foxr2, and Megf10, making them especially compelling candidates (Supplementary Table S8). Notably, several R-CIS are located on chromosomes 4 and 15, the locations of the T2/Onc2 and T2/Onc concatemers, respectively. Although local hopping may account for bias toward genes on these chromosomes, several are implicated in cancer, including Tle1 and Ptprd (35, 36). In addition, these concatemers have been previously used to identify R-CIS in osteosarcoma, and only 1 R-CIS gene was common to both studies (Cdkn2a; Supplementary Table S9; ref. 15). Our medulloblastoma R-CIS include several highly compelling targets, including Megf10. We identified a novel oncogenic role for Megf10 in neural progenitor cells, the mechanism for which warrants further study.

Ours is the first transposon screen to produce CNS-PNETs. We identified several genes with known roles in neural cancer not previously implicated in CNS-PNET, including Setd2, Ambra1, and Usp9x (37–39). Several Ras-associated genes were mutated in our screen, including Nf1, Eras, Pten, and Ras3, suggesting an importance of Ras pathway activation and cooperation with p53 loss in PNETagenesis. In addition, we identified NF1 loss in human CNS-PNETs. Activated RAS/MAPK signaling with p53 loss has been shown to drive CNS NB-FOXR2 formation in zebrafish (40), and somatic PTEN loss is associated with human CNS-PNET (22). Interestingly, we did not recover any CNS-PNETs on the Pten-deficient background, possibly indicating that p53 loss creates a permissive cell with subsequent Ras activation.

FOXR2 is a member of the forkhead-box (FOX) transcription factor family, which contribute to a wide variety of cellular processes (41). FOXR2 acts as an oncogene in several neural cancers including: malignant peripheral nerve sheath tumors, glioma, CNS-PNET, and medulloblastoma (6, 13, 34, 42). Interestingly, although FOXR2 has been shown to be upregulated in CNS NB-FOXR2 (6), we did not recover Foxr2 insertions in the SB-induced CNS-PNETs, including two tumors that resembled CNS NB-FOXR2 transcriptionally. Other insertions may mimic the CNS NB-FOXR2 phenotype; these two tumors exclusively harbored insertions in Epb4.1l1, Itpr1, Rbfox1, and Sphkap.

The mechanisms of FOXR2-driven tumorigenesis has proven diverse and elusive. FOXR2 can promote WNT signaling, activate SHH signaling, promote EMT, and affect cell cycle (34, 42–45). We examined each of these pathways in C17.2 cells and found no effect of FOXR2 on β-catenin localization, Axin2, Gli1, p21, or cyclin D1 mRNA expression, or N-cadherin, E-cadherin, or vimentin protein levels (Supplementary Fig. S7A–S7D). We found that FOXR2 binds and stabilizes C-MYC. Mouse models with C-MYC–driven tumors show an addiction to C-MYC expression, suggesting C-MYC is a strong therapeutic target in cancer (46). However, directly targeting C-MYC has been difficult. Therefore, targeting C-MYC interacting proteins, such as FOXR2, may prove useful for cancer therapy. We also found that FOXR2 promotes activation of the FAK/SRC signaling pathway. FAK activation is associated with poor prognosis and drug resistance in a variety of cancers and targeting FAK produces deleterious off-target effects (47). Interestingly, cotargeting of FAK and C-MYC was recently shown to have synergistic effects in ovarian cancer (48). The ability of FOXR2 to activate both of these pathways makes it an excellent candidate for targeted therapy. In addition, FOXR2 has minimal expression in adult tissues, making off-target toxicity risk low (13).

We identified ARHGAP36 as both a mouse and human medulloblastoma oncogene. ARHGAP36 expression in C17.2 cells promoted anchorage independent growth, tumor formation, leptomeningeal spread, and SHH activation. Current therapies targeting an upstream pathway member, Smoothened (SMO), have been met with resistance through SMO mutations (49). Because its interactions with PKA and SUFU are both downstream of SMO, ARHGAP36 poses a good target for treatment-resistant, SHH-driven medulloblastoma (17, 27). In addition, Arhgap36 was the most upregulated gene in mouse allografts propagated in the presence of a SMO antagonist (17). Interestingly, Arhgap36 insertions occurred in mouse Shh and group 3/4 tumors, and ARHGAP36 is expressed across all subgroups of human medulloblastoma, indicating ARHGAP36 may also have non-SHH protumorigenic effects.

We identified several candidate driver genes in medulloblastoma and CNS-PNET relevant to human cancer. To our knowledge, this is the first study to present a genetically induced CNS-PNET mouse model, providing an opportunity for studying this rare and aggressive tumor. We also present tumors that resemble group 3/4 medulloblastoma with high incidence of leptomeningeal spread, again providing a needed mouse model for these tumors. Interestingly, these diverse tumor types were driven with the same, Nestin-driven Cre recombinase, indicating that the cell of origin of non-Wnt medulloblastoma and CNS-PNET is Nestin⁺ or a close descendent. We used RNA-seq to identify CIS genes and subtype mouse SB-induced tumors based on human expression data. Arhgap36, our top CIS gene, was shown to transform a mouse neuroblast line. Foxr2 was identified as a proto-oncogene and shown to promote C-MYC stability and FAK pathway activation. Both of these genes offer promise as novel therapeutic targets in human medulloblastoma and warrant additional study. Further functional testing of additional CIS genes may reveal additional treatment options for embryonal tumors.

D.A. Largaespada is chairman of scientific advisory board at B-MoGen, chief scientific officer at Surrogen, reports receiving commercial research grant from Genentech, and has ownership interest (including stock, patents, etc.) in Surrogen, ImmuSoft, NeoClone, and B-MoGen. No potential conflicts of interest were disclosed by the other authors.

Conception and design: P.J. Beckmann, J.D. Larson, E.P. Rahrmann, P. Das, R.J. Wechsler-Reya, D.J. Odde, A.L. Sarver, D.A. Largaespada

Development of methodology: P.J. Beckmann, J.D. Larson, R.L. Williams, B.R. Tschida, P. Das, R.D. Krebs, M.M. Frees, A.E. Rizzardi, S.C. Schmechel, A.L. Sarver, D.A. Largaespada

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): P.J. Beckmann, J.D. Larson, A.T. Larsson, J.P. Ostergaard, S. Wagner, E.P. Rahrmann, G.A. Shamsan, G.M. Otto, R.L. Williams, J. Wang, C. Lee, P. Das, B.S. Moriarity, X. Wu, Q. Rosemarie, R.D. Krebs, A.M. Molan, A.M. Demer, M.M. Frees, A.E. Rizzardi, S.C. Schmechel, C.G. Eberhart, R.B. Jenkins, R.J. Wechsler-Reya

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): P.J. Beckmann, J.D. Larson, A.T. Larsson, J.P. Ostergaard, E.P. Rahrmann, G.A. Shamsan, J. Wang, C. Lee, A.M. Dubuc, D. Picard, F.J. Rodriguez, R.D. Krebs, S.C. Schmechel, R.B. Jenkins, R.J. Wechsler-Reya, D.J. Odde, A. Huang, M.D. Taylor, A.L. Sarver

Writing, review, and/or revision of the manuscript: P.J. Beckmann, J.D. Larson, A.T. Larsson, J.P. Ostergaard, G.A. Shamsan, J. Wang, C. Lee, B.R. Tschida, A.M. Dubuc, B.S. Moriarity, D. Picard, F.J. Rodriguez, Q. Rosemarie, R.D. Krebs, A.M. Molan, A.M. Demer, S.C. Schmechel, C.G. Eberhart, R.B. Jenkins, D.J. Odde, M.D. Taylor, A.L. Sarver, D.A. Largaespada

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): P.J. Beckmann, J.D. Larson, A.T. Larsson, J.P. Ostergaard, A.E. Rizzardi, A.L. Sarver

Study supervision: P.J. Beckmann, J.D. Larson, R.J. Wechsler-Reya, D.J. Odde, D.A. Largaespada

Other (cell Culture and other assays performed in the study): R.D. Krebs

This work was supported by UofMN Genomics Center, UofMN Biology Materials Procurements Network, Research Animal Resources, and University Imaging Centers that are supported by the NCI. Support for this research was provided by The American Cancer Society (Research Professor Award no. 123939 to D.A. Largaespada), the NIH (U54CA210190 to D.A. Largaespada and D.J. Odde; R01CA113636 to D.A. Largaespada; T32 T32GM113846 to P.J. Beckmann; R50-CA211249 to A.L. Sarver; T32 AI083196 to B.R. Tschida; T32CA009138 to J.D. Larson; and R01CA172986 to D.J. Odde), the Children's Cancer Research Fund, and the Hedberg Family Chair (all to D.A. Largaespada).

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.