Mouse medulloblastoma driven by CRISPR activation of cellular Myc

MYC-driven Group 3 (G3) medulloblastoma (MB) is the most aggressive of four molecular subgroups classified by transcriptome, genomic landscape and clinical outcomes. Mouse models that recapitulate human G3 MB all rely on retroviral vector-induced Myc expression driven by viral regulatory elements (Retro-Myc tumors). We used nuclease-deficient CRISPR/dCas9-based gene activation with combinatorial single guide RNAs (sgRNAs) to enforce transcription of endogenous Myc in Trp53-null neurospheres that were orthotopically transplanted into the brains of naïve animals. Three combined sgRNAs linked to dCas9-VP160 induced cellular Myc expression and large cell anaplastic MBs (CRISPR-Myc tumors) which recapitulated the molecular characteristics of mouse and human G3 MBs. The BET inhibitor JQ1 suppressed MYC expression in a human G3 MB cell line (HD-MB03) and CRISPR-Myc, but not in Retro-Myc MBs. This G3 MB mouse model in which Myc expression is regulated by its own promoter will facilitate pre-clinical studies with drugs that regulate Myc transcription.

domains such as VP16, p300, VPR or KRAB to epigenetically activate or suppress gene transcription [18][19][20][21][22] . To our knowledge, the application of dCas9 to enforce the expression of oncogenic drivers to induce tumor development in vivo has not been addressed. Here, we demonstrate the ability of the CRISPR-dCas9-VP160 system to modulate endogenous Myc expression in Trp53-null neurosphere cells to generate an orthotopic mouse model of G3 MB amenable to BETi treatment.

Results
Design of CRISPR activation of endogenous Myc. Using previous H3K27acetylation chromatin immunoprecipitation sequencing (ChIP-Seq) and ATAC-Seq data from purified mouse G3 Retro-Myc MBs 8 , we located a ~1.2 Kb open chromatin region corresponding to the cellular Myc dual P1 and P2 promoter region ( Supplementary Fig. S1) to which we designed a series of CRISPR guide RNAs. To facilitate gene activation, we fused sequences encoding 4X or 10X tandem repeats of the transactivation domain of Herpes simplex virus protein VP16 (VP64 or VP160, respectively) to the C-terminus of nuclease-deficient dCas9 (D10A, H840A) and fused these to T2A-GFP in a lentivirus backbone or transposon vector 23 (Fig. 1a). Alternatively, we used sequences encoding a group of transcription activator-like effector (TALE) polypeptides fused to VP64 and T2A-GFP 24 (Fig. 1b). CRISPR and TALE design software 8,25 pinpointed 13 sgRNAs (sgRNA-M1 to M13) and 8 TALE binding motifs (TALE-TF-1 to -8) within a ~1.2 Kb segment upstream of the initiator ATG of the cellular Myc gene. These sgRNA and TALE sequences were compared against the whole mouse genome using the NCBI BLAST nucleotide program to rule out adventitiously targeted loci. Both design strategies recognized three overlapping target loci designated sgRNA-M5, -M7, and -M9 and TALE-TF-2, -4 and -8 (Fig. 1c).
The retro-Myc MB model was generated by enforced high titer vector-mediated-Myc expression in Trp53 −/− ; Cdkn2c −/− GNPs which were orthotopically transplanted into the cortices of recipient nude mice, as previously described 7 . Retro-Myc MBs are polyclonal tumors containing multiple, randomly integrated proviral DNA sequences that, in aggregate, encode high levels of Myc RNA and protein 8 . To identify TALE and sgRNAs that activate cellular Myc expression to levels approaching those found in mouse G3 Retro-Myc MBs, we cloned a 0.665 Kb minimal promoter that includes two TATA box sequences and TALE-TF binding motifs upstream of the reporter Luciferase gene in a pGL3 expression vector 26 . NIH3T3 cells were co-transfected with or without Myc promoter-reporter plasmids and TALE-TF vectors. Compared to the empty vector, TALE-TFs-1, 2, 3, 4 and 8 upregulated reporter activity more than 2-fold with TALE-TFs-2, 4 and 8 being most effective (Fig. 1d). Combined use of TALE-TFs -2, -4 and -8 induced luciferase activity 14-fold (Fig. 1d). In turn, NIH3T3 cells infected with a lentivirus vector encoding both dCas9-VP64 and GFP were used to generate stable cell lines that were subsequently infected with lentiviruses expressing U6 promoter-driven sgRNAs co-expressing RFP. Immunoblotting of double-positive GFP + /RFP + cells selected by fluorescence-activated cell sorting (FACS) with antibodies to Myc and dCas9 showed that sgRNA-M5, -M7 and -M9, closely corresponding to TALE-TF-2, -4 and -8, induced endogenous Myc expression (Fig. 1e). Myc levels induced by sgRNA-M9 mimicked those achieved in G3 tumors from Retro-Myc MB cells (Fig. 1e). No activation was detected with sgRNA-M2, M3 or M11. Because three similar sequences in the Myc promoter targeted by both TALE-TF and sgRNAs led to significant Myc expression, subsequent experiments were continued using CRISPR alone.

Activation of cellular Myc in Trp53-null GNPs generates G3 MBs.
To determine if cells expressing trans-activated Myc generated MBs, we transiently transfected Trp53-null GNPs grown as neurospheres with a pT2K Tol2 transposon vector 11 co-encoding dCas9-VP160 and mCherry together with a non-integrating vector encoding the Tol2 transposase (T2TP). Neurospheres were co-transfected with a third pT2K vector specifying cyan fluorescent protein (CFP) and sgRNA-M5, -M7 and -M9, either individually or in combination. Pooled mCherry and CFP double-positive neurospheres were isolated by FACS. Individual sgRNA-M5, -M7, and -M9 guides induced Myc RNA expression (Supplementary Fig. 2a and b), while the combination of the three sgR-NAs increased Myc protein to levels similar to those in a mouse G3 MB Retro-Myc tumor (Fig. 2a). Hence, we cloned sgRNA-M5, -M7 and -M9 into the U6-guide RNA cassette, and co-expressed these guides in a single vector backbone. FACS sorted mCherry + /CFP + cells (total 1 × 10 6 per mouse) expressing dCas9-VP160 and sgRNA-M5, -M7 and -M9 using either three independent vectors or one vector were orthotopically transplanted into the cerebral cortices of 6 to 8-week-old recipient nude mice 8 . As a negative control, an equal number of cells expressing dCas9-VP160 without sgRNAs were implanted. All mice implanted with Trp53-null neurospheres expressing dCas9-VP160 with the combination of three sgRNA-M5, -M7 and -M9 developed (CRISPR-Myc) tumors, whereas none were generated in the negative control group during 6-months of observation (Fig. 2b). In several cases, tumors metastasized to the spinal cord, consistent with the aggressive nature of Myc-driven G3 MBs. The presence of all three sgRNAs in the tumors was confirmed by genomic PCR and nucleotide sequencing. Therefore, the co-recruitment of dCas9-VP160 to sgRNA-M5, -M7 and -M9 binding sites was sufficient to induce levels of endogenous Myc expression compatible with tumor development in vivo.
When compared to MBs arising in mouse and human WNT, SHH and two Retro-Myc G3 MB mouse models, the CRISPR-Myc tumors shared cardinal properties with Retro-Myc G3 tumors, including large cell anaplastic features (Fig. 2c) and canonical histological markers (Fig. 2d). In vitro, tumor cells grew as neurospheres, could be expanded for multiple passages, and stably expressed the mCherry and CFP fluorescence markers that were fused to the dCas9-VP160 and sgRNA vectors (Fig. 2e). We performed microarray analyses and compared the CRISPR-Myc tumors to other previously derived mouse G3 MBs initiated through independent approaches 7,11 , as well as to Shh and Wnt 27 mouse MBs 27,28 . Principal component analysis (PCA) revealed that CRISPR-Myc tumors clustered together and were most similar to Retro-Myc MBs, while clearly different from Shh and Wnt tumors (Fig. 3a). As expected, a subset of genes that typify three subgroups of MB were differentially expressed among individual subgroups (Fig. 3b). Elevated Myc expression itself, higher in Retro-Myc than in CRISPR-Myc MBs, as well as that of Npr3, a hallmark of human G3 MBs, were restricted to both mouse G3 Myc MB models. Conversely, lower levels of Shh-associated genes, Mycn, Gli1, Atoh1, Sfrp1 and Boc, were detected in Retro-and CRISPR-Myc MBs. However, Cd44 levels were higher in Retro-Myc MBs than in CRISPR-Myc MBs, as were those of Prom1 and Lgr5, two markers of stem-like cells (Fig. 3b). Although the latter differences in gene expression between the two Myc MB models correlate with higher overall levels of viral versus cellular Myc expression induced in these tumors, the levels of CRISPR-Myc were sufficient to reproducibly yield G3 MBs.
Validation of sgRNAs specific binding to the Myc promoter upon CRISPR activation. To validate the binding specificity of dCas9 to the pre-designed segments directed by sgRNAs, we performed ChIP, as previously described 29 , using an antibody against dCas9 for analysis of Trp53-null neurospheres expressing dCas9-VP64 and sgRNA-M5, -M7 and -M9. PCR analysis of the ChIP showed substantial enrichment of dCas9 protein at the M5 genomic segment (Fig. 4a). In agreement, ChIP-Seq peak analysis clearly demonstrated dCas9 binding at each of the three intended target sites (Fig. 4b). By analyzing peaks across the whole genome, no binding sites other than the on-target sequences were detected with sgRNA-M5 and -M9. However, for sgRNA-M7, although most reads were on-target, we resolved a number of off-target binding sites with near perfect alignment to proximal protospacer adjacent motifs (PAMs) but with distal mismatches (Fig. 4c). Off-target binding sites resided on non-coding intronic regions of other genes (e.g. Gprc5c and Cystm1) except for the promoter of Smyd5, a histone methyltransferase for H4K20 in heterochromatic regions that maintain embryonic stem (ES) cell lineage specification 30 (Supplementary Fig. 3).

BET bromodomain inhibitor JQ1 suppresses MYC expression in human and mouse MB cells.
Brd4, a member of the bromodomain and extra-terminal (BET) protein family that reads acetylated lysines was previously shown to activate MYC transcription in acute myeloid leukemia (AML) 31 . Conversely the small molecule Brd4 inhibitor JQ1 effectively inhibited MYC expression in this leukemia but was unable to do so in models in which MYC was ectopically expressed 31 . A previous study in a G3 MB line (MB002) showed that JQ1 suppresses stem cell signaling and inhibits tumor cell self-renewal and development 32,33 . The human (HD-MB03) and mouse Retro-Myc (#19251) and CRISPR-Myc (#3578 and #7444) MB tumor cell lines were treated with 1 and 10 μM of JQ1 for 24 or 72 hours. Treatment with 10 μM of JQ1 for 72 hrs inhibited MYC, BRD2, and BRD4 protein expression levels in human and CRISPR-Myc MB cell lines without affecting expression of these proteins in the Retro-Myc MB cell line (Fig. 5a,b). In turn, tumorsphere formation of human MB and CRISPR-Myc versus Retro-Myc MBs was inhibited (Fig. 5c,d) and associated with increased apoptosis (Fig. 5e).

Discussion
Among the known molecularly characterized MB subgroups, the highest levels of MYC expression and gene amplification typify aggressive G3 tumors 34,35 . Mouse models of MB that recapitulate the cardinal features of human G3 tumors were previously generated in Trp53-deficient GNPs by retroviral gene transfer 7,9 in which insertion of multiple Myc-encoding proviruses induced polyclonal populations of tumor cells that in aggregate expressed sufficiently high levels of Myc protein to yield G3 tumors 8 . The major limitations of past models have been that retrovirally-driven gene expression bypasses normal regulation of Myc by its endogenous regulatory elements; that tumor latency in individual experiments is highly variable; and that the contribution of proviral insertional mutagenesis is difficult to exclude. By using the CRISPR activation system 23 to induce endogenous Myc expression in Trp53-null GNPs, we have now generated aggressive large cell anaplastic MBs driven by the cellular Myc gene that exhibit the canonical histological and molecular properties of the G3 subgroup.
Optimization of CRISPR-driven gene activation was previously undertaken by fusing dCas9 with different effectors, including concatemerized VP16 and p300 core domains 19,23,36 . Most of these models utilize sgRNAs targeting regions proximal to the transcriptional starting site (TSS), while the p300 core domain also works on distal enhancers 37 . By systematically comparing a series of more than 20 effectors with known transcriptional activity, the hybrid of VP64-p65-Rta (VPR) was found to exhibit the most potency for activating endogenous genes 38 and was applied to promote the differentiation of human induced pluripotent stem (iPS) cells to neurons (iNeurons) by targeting Neurogenin2 (NGN2) or the Neurogenic Differentiation Factor 1 (NEUROD1) 38 . Alternatively, gene activation by protein-based TALE-TFs targeting the Oct4 distal enhancer in conjunction with enforced expression of Myc, Klf4, and Sox2 reprogrammed mouse embryonic fibroblast (MEF) cells into iPS cells in 3-4 weeks 39 . Following on such studies, we successfully piloted the use of both CRISPR/dCas9-VP160 and TALE-VP64 in activating cellular Myc and found that three particular and similar segments within the Myc promoter facilitated its induction by either method. We exploited CRISPR/dCas9 for subsequent experiments, acknowledging the greater convenience of the CRISPR system particularly in broader applications such as unbiased large-scale genome-wide screening 23,40 , regardless of the models used or phenotypic outputs. Although the CRISPR system has enabled highly efficient genome editing in a wide variety of organisms 41 , unwanted off-target effects may present problems. Several studies have used high-fidelity Cas9 (SpCas9-HF) 42 , mutant scaffold RNA structures 43 and truncated sgRNA with complementarity lengths of 17 or 18 nucleotides other than 20 nucleotides 44 , to avoid such complications. In our study, we used genome-wide ChIP-Seq to identify potential off-target binding sites. For sgRNA-M5 and M9, we did not detect any off-target binding sites. However, for sgRNA-M7, we identified 10 off-target lower affinity binding sites, many in introns, compared to on-target controls. Open chromatin from ATAC-Seq analysis showed that not all off-target binding sites had H3K27ac marks with the exception of Smyd5, implying that transcriptional activation need not follow from sgRNA binding.
Recent analysis using a larger cohort of patients subdivided MBs into multiple specific subtypes within each of the four major MB subgroups according to clinical outcome, gene methylation, transcriptome and genomic landscapes 3,45,46 . Many more potential driver genes were identified within these specific subgroups which might be validated using the CRISPR activation system as described here. A potentially interesting target in G3 MB is Gfi1, the activation of which in conjunction with Myc bypasses the requirement for Trp53 loss of function in generating MBs 8,10 . However, our preliminary attempts to target this gene have not as yet yielded useful information.  Unlike Retro-Myc tumors, the CRISPR-Myc MBs retain a heightened sensitivity to the BRD4 inhibitor JQ1, providing proof of principle that this new model might also provide a substrate for pre-clinical identification of small, potentially therapeutic molecules that regulate Myc transcription.

Methods
Animal husbandry. CD-1 nu/nu mice (Charles River Laboratories) were used as recipients for cortical transplants. Trp53-null mice in a C57BL/6 background were utilized to generate GNPs, as previously described 7,8 . All animal experiments were approved by, and conducted in accordance with St. Jude Children's Research Hospital Animal Care and Use Committee guidelines, as required by the United States Animal Welfare Act and the National Institutes of Health's policy to ensure proper care and use of laboratory animals for research.
The dCas9 contains two point mutations: D10A and H840A, which disrupt the nuclease activity of Cas9. We generated the sgRNA cloning vector Lenti-U6-sgRNAs-RFP by inserting a U6-sgRNA cassette into pCDH-CMV-MCS-EF1-RFP backbone (System Biosciences, CD512B-1) after cleavage by SnaBI and EcoRI. All target specific sgRNAs were predicted by online software (http://crispr.mit.edu/). Oligomers containing 20 bp of selected target sequence (Supplementary Table S1) were cloned into the sgRNA cloning vector at BsmBI sites. To generate TALE-TF-VP64 plasmids targeting different segments of the Myc promoter, sequences of TALEs predicted by online software (http://crispr.mit.edu/) were compared against the whole mouse genome using the NCBI nucleotide BLAST program to rule out alternative binding loci. TALE monomers were assembled using a commercial tool kit (Addgene, #000000019) as described previously 25 . Eighteen assembled monomers were inserted into one of four backbone plasmids, each of which included an N-terminal signal specific for thymidine and one of four base-specific C-terminal TALEs to generate a 20-nucleotide recognition sequence. These were fused in frame at the N-terminus with three Flag tags and a nuclear localization sequence, and at the C-terminus to sequences encoding the trans-activator module VP64. Correctly assembled TALEs cDNAs were verified by nucleotide sequencing analysis; transfection of cDNAs into human 293 T cells using Lipofectamine 2000, followed by immunoblotting of cell lysates with antibodies to the FLAG epitope (M2; Sigma-Aldrich), confirmed the expression of proteins of the expected masses (~135 kDa).
Luciferase reporter assay. The mouse minimal Myc promoter that includes a 0.665 Kb fragment upstream of the initiating ATG codon was cloned into the pGL3 vector (Promega). For transfection, 1 × 10 5 NIH3T3 cells were plated into 12-well plates overnight. A mixture of 0.5 μg of Myc reporter and 1 μg of different TALE-TF-VP64 plasmids were used for each transfection as described previously 8 . Two days after transfection, luciferase reporter assays were carried out following the manufacturer's instructions (Promega). Histopathology and immunohistochemistry. For histopathology, samples were formalin-fixed, paraffin-embedded, and sectioned at 5 μm thickness. For each sample, a section was stained with hematoxylin and eosin. For immunohistochemistry, tumor sections were subjected to heat antigen retrieval with 10 mM citrate buffer for 30 minutes. Sections were blocked for 30 minutes in 10% normal goat serum in phosphate buffered saline (PBS). Sections were incubated with specific antibodies: Ki67 (1:1000, #556003, BD Pharmingen), GFAP (1:200, Chemicon), Synaptophysin (1:100, #RM-9111-S, Thermo Fisher Scientific, Rockland, IL) and cleaved Caspase-3 (1:100, #CP229C, BioCare Medical, Concord, CA). Sections were then incubated for 60 minutes with biotinylated goat anti-rabbit IgG (Vector labs) at 1:200 dilutions. Blocker D, Streptavidin-HRP and AEC detection kit (Abcam) were used according to the manufacturer's instructions. Representative images of each sample/ stain combination were captured and analyzed using Axiovision software (Carl Zeiss Microscopy).

Lentivirus infection and orthotopic transplantation.
High titer lentivirus stocks were generated in 293 T cells as previously described 7 . GNP-derived neurospheres were infected with Lenti-dCas9-VP64-T2A-GFP and Lenti-U6-sgRNAs-RFP, sorted by flow cytometry, expanded in culture, harvested, and suspended in Matrigel (BD Bioscience) before orthotopic transplantation of 1 × 10 6 cells into the cortices of CD-1 nu/nu recipient mice, as previously described 8 . Animals were examined daily for symptoms of sickness (doming of the head, ataxia and reduced activity). Moribund mice were humanely sacrificed, tumor cells were harvested and RNA was extracted for further analysis. Tumors, brains, and spines of mice were reviewed by histopathology to evaluate whether the tumors were G3 MBs with large cell/anaplastic and metastatic features.
Quantitative real-time PCR. Total RNA was extracted from tumor cells using Trizol (Thermo Fisher Scientific) and reverse transcribed into cDNA using the Multiscribe reverse transcription reagents according to the manufacturer's protocol (Applied Biosystems). cDNA amplification was performed with an ABI 7900 Real-Time PCR system and FAM-Labeled Taqman primer/probes specific to the genes of interest (Applied Biosystems). Threshold cycle (CT) values from triplicate measurements were averaged and normalized to those obtained from the internal control gene, GAPDH. Relative gene expression was determined by the 2 − ΔΔCT method 49  Affymetrix microarray analysis. RNA from tumor cells was subjected to hybridization using Affymetrix Mouse Gene 2.1 ST Array. The data were normalized and log2-transformed using RMA as implemented in Partek Genomic Suite 6.6 (St Louis, MO) 50 . The resulting matrix was joined to a RMA matrix of previously published mouse models from GSE65888 (arrayed on Affymetrix Mouse Genome 430 2.0 Arrays) 11 . The highest expressing probeset for each gene was used and the two array datasets were joined by gene symbol in STATA MP/14.2 yielding 17870 genes. Next using Partek 6.6, the data was corrected for the chip effect and visualized by PCA. Select genes of interest were after z-score normalized, hierarchically clustered, and presented in a "heat map" format.
ChIP-Seq and data analysis. ChIP-Seq was performed exactly as described previously 8 . Chromatin was subjected to ChIP using rabbit polyclonal antibody directed against dCas9 (Clontech, #632607). To validate the dCas9 binding on their target sites, a primer pair spanning the target sites (M5) was used. PCR was performed using immunoprecipitated DNA and whole cell extract DNA as a control input. DNA was purified using phenol:chloroform and sequenced 51 on the Illumina HiSeq platform; high-quality reads were aligned to the reference mouse genome (mm9) to visualize the sgRNA-mediated dCas9-DNA binding sites.
We used BWA (version 0.5.9-r26-dev, default parameter) to align the reads to the mouse (UCSC mm9) genome and then marked duplicated reads with Picard (version 1.65), with only non-duplicated reads kept by Samtools (parameter "−q 1 −F 1024" version 0.1.18). To control the quality of the data and estimate the fragment size, the non-duplicated version of SPP (version 1.11) was used to draw cross-correlation with support of R (version 2.14.0) 52 . Upon manually inspecting the cross-correlation plot generated by SPP, the best fragment size estimates (the smallest fragment size estimated by SPP in all of our cases) were used to extend each read and to generate bigwig files to view on the Integrative Genomics Viewer (IGV) (version 2.3.47). We scaled the bigwig files to 15 M reads so that the heights of peaks between samples are roughly comparable. To find the off-targets, MACS2 (version 2.0.10.20131216) was used to call peaks (−nomodel with fragment size estimated above) and low mapability region (50% of the peak mapability from wgEncodeCrgMapabilityAlign50mer < 0.5) were excluded. PatMaN was used to match the sequences summarized with extended reads counts by bedtools (v2.24.0) 53 . ATAC-Seq. ATAC-Seq was performed with 200,000 mouse ES cells and 75,000 MEF cells as described 54 . In brief, after preparation of nuclei, the nuclei were mixed with transposase for the digestion reaction. DNA was purified, amplified by PCR, and sequenced on a Hi-Seq 4000. Adapters were trimmed from total reads, which were then mapped to the mm9 mouse genome version. Mitochondrial DNA reads and duplicates were removed. Bigwig files were finally visualized using IGV.
Additional statistical analyses. The Kaplan-Meier method was used to generate mouse survival curves.
Statistical analyses were performed in the GraphPad Prism software version 6.0. P-values were calculated by an unpaired two-tailed t-test from three independent experiments. Data access. Accession numbers for the microarray and ChIP-Seq data reported in this paper are GEO: GSE102037 and GSE102096.