Somatic genome editing with the RCAS/TVA-CRISPR/Cas9 system for precision tumor modeling

It has been gradually established that the vast majority of human tumors are extraordinarily heterogeneous at a genetic level. To accurately recapitulate this complexity, it is now evident that in vivo animal models of cancers will require to recreate not just a handful of simple genetic alterations, but possibly dozens and increasingly intricate. Here, we have combined the RCAS/TVA system with the CRISPR/Cas9 genome editing tools for precise modeling of human tumors. We show that somatic deletion in neural stem cells (NSCs) of a variety of known tumor suppressor genes (Trp53, Cdkn2a and Pten), in combination with the expression of an oncogene driver, leads to high-grade glioma formation. Moreover, by simultaneous delivery of pairs of guide RNAs (gRNAs) we generated different gene fusions, either by chromosomal deletion (Bcan-Ntrk1) or by chromosomal translocation (Myb-Qk), and we show that they have transforming potential in vitro and in vivo. Lastly, using homology-directed-repair (HDR), we also produced tumors carrying the Braf V600E mutation, frequently identified in a variety of subtypes of gliomas. In summary, we have developed an extremely powerful and versatile mouse model for in vivo somatic genome editing, that will elicit the generation of more accurate cancer models particularly appropriate for pre-clinical testing.


Introduction 35
A decade of studies has underlined the complexity of the genetic events that characterize 36 the genomic landscapes of common forms of human cancer 1 . While a few cancer genes are 37 mutated at high frequencies (>20%), the greatest number of cancer genes in most patients appear 38 at intermediate frequencies (2-20%) or lower 2 . Strikingly, the functional significance of the vast 39 majority of these alterations still remains elusive. A current high priority in cancer research is to 40 functionally validate candidate genetic alterations to distinguish those that are significant for 41 cancer progression and treatment response. In order to do so, it is essential to develop flexible 42 genetically engineered mouse models that can speed up the functional identification of cancer 43 driver genes among the large number of passenger alterations 3 . 44 The growing level of sophistication of the genome engineering technologies has made it 45 possible to target almost any candidate gene in the in vivo setting. The CRISPR (Clustered 46 Regularly Interspaced Short Palindromic repeats) -Cas (CRISPR-associated), the most powerful 47 genome editing system so far, has revolutionized research in many fields, including cancer 48 animal modelling, by allowing precise manipulation of the genome of individual cells. Its 49 applications span from the inactivation of tumor suppressor genes, to the generation of somatic 50 point mutations and more complex genomic rearrangements such as gene fusion events. 51 A possibly significant limitation of CRISPR-based in vivo somatic genome editing is the 52 requirement to concurrently deliver the RNA guides and the Cas9 enzyme to the specific tissue 53 of interest. To deal with this issue, various groups recently generated transgenic mice expressing 54 Cas9 in a Cre-or tetracycline-dependent manner 4-6 . The combination of somatic genome 55 editing with the vast collection of currently available genetically engineered mouse models will 56 provide the chance to introduce defined genetic lesions into specific cell types, leading to the 57 development of more accurate tumor models. 58 The RCAS/TVA based approach uses replication-competent avian leukosis virus (ALV) 59 splice-acceptor (RCAS) vectors to target gene expression to specific cell types in transgenic 60 mice. In these mice, cell type-specific gene promoter drive expression of TVA, the cell surface 61 receptor for the virus. The RCAS/TVA system has been successfully used in different mouse 62 models to deliver genes or shRNAs of interest into a plethora of cell types: neural stem cells, 63 astrocytes, hepatocytes and pancreatic acinar cells, among many others 7 .
Here we describe a series of new mouse models that combine the genome editing 65 capability of the CRISPR/Cas9 system with the somatic gene delivery of the RCAS/TVA 66 approach to generate precision tumor modeling. To prove the efficacy of such a powerful system, 67 we produced a number of in vivo and ex vivo models of glioma with tailored genetic alterations. 68 The gliomas are a large group of brain tumors and within gliomas the glioblastoma 69 (GBM) is the most frequent form of the disease and overall the most common and lethal primary 70 central nervous system (CNS) tumor in adults. A series of large-scale genomic analysis has 71 underlined the complexity of the genetic events that characterize the glioma genome. However, 72 so far, we have been able to study only a minority of these genetic alterations due to the lack of 73 appropriate tumor models. 74 We took advantage of the previously developed Rosa26-LSL-Cas9 (LSL-Cas9) knockin 75 mouse strain 4 and combined it with the Nestin-tv-a (Ntv-a) and the GFAP-tv-a (Gtv-a) 76 transgenic mice that express the TVA receptor under the control of the rat nestin and human 77 GFAP promoter, respectively 8,9 . Moreover, we have further crossed those strains with either the 78 Nestin-Cre (Nes-Cre) or hGFAP-Cre transgenic lines 10,11 , to allow for Cas9 expression in neural 79 stem/progenitor cells and in astrocytes, or with a mouse strain carrying a tamoxifen-activated 80 recombinase, hUBC-CreERT2 12 , for ubiquitous and inducible Cas9 expression. 81 Through in vivo delivery of RCAS plasmids that carry guided RNAs (gRNAs) for a series 82 of tumor suppressor genes known to be frequently deleted in GBM (Trp53, Cdkn2a and Pten), in 83 combination with the expression of the Platelet Derived Growth Factor Subunit B (PDGFB), we 84 show that we can efficiently generate high-grade gliomas in mice that express both Cas9 and 85 TVA in the Nestin or GFAP positive cells. Moreover, by simultaneous ex vivo transduction into 86 neural stem cells (NSCs) of RCAS plasmids expressing pairs of gRNAs we generated either 87 chromosomal deletion (Bcan-Ntrk1 gene fusion) or chromosomal translocation (Myb-Qk gene 88 fusion) and we show that they lead to glioma formation when transplanted in 89 immunocompromised mice. We further generated Braf mutant gliomas by inducing a homology-90 directed repair-mediated BRAF V600E mutation. 91 Lastly, by ex vivo treatment of some of these tumor models we demonstrate their utility 92 for pre-clinical testing of targeted therapies. 93 expression of the Cas9-P2A-EGFP throughout the brain of adult mice and pups ( Fig. 1a-b and 125 Supplementary Fig. 1b). Also of note it is the co-localization of NESTIN and GFAP with EGFP 126 in the area of the sub-ventricular zone ( Fig. 1a- Gtv-a pups has been previously shown to be sufficient to induce gliomas with variable 153 penetrance (from 40% to 75%), but only a small fraction of the injected mice (25%) presented 154 high-grade tumor features, such as pseudopalisades necrosis and microvascular proliferation [17][18][19] . Moreover, RCAS-PDGFB injection into Ntv-a and Gtv-a adult mice resulted in very low 156 tumor penetrance (approximately 15-20%) and long latency (over 100 days) 13 . Co-injections 157 RCAS-PDGFB and RCAS-TSG-gRNA (either one of Trp53, Cdkn2a or Pten gRNAs) into Ntv-158 a; Nes-Cre; LSL-Cas9 and Gtv-a; hGFAP-Cre; LSL-Cas9 pups resulted in a shortened tumor 159 latency and increased total tumor incidence as compared to the co-injections of RCAS-PDGFB 160 and RCAS-gRNA non-targeting control (Ctrl) (Fig. 2c and Supplementary Fig. 2). Most 161 importantly, the vast majority (80-100%) of the RCAS-PDGFB/RCAS-TSG-gRNA injected 162 mice showed histological features of high-grade gliomas (Fig. 2c-d). We also generated RCAS 163 plasmids expressing the gRNA and the PDGFB in the same constructs (hU6-gRNA-PGK-Puro-164 T2A-PDGFB) (Fig. 2a). When injected into Ntv-a; Nes-Cre; LSL-Cas9 pups, the RCAS-Cdkn2a-165 gRNA-PDGFB bicistronic vector was able to induce high-grade tumor formation with full 166 penetrance and very short latency (approximately 40 days) ( Fig. 2c and Supplementary Fig. 2). 167 Analogously to what was previously reported for the RCAS-PDGFB, injection in adult 168 mice showed a considerably reduced tumor incidence. Actually, in our 120 days' experimental 169 timeframe, we didn't observe any tumors in the mice co-injected with the RCAS-PDGFB and hUBC-CreERT2 +/T tamoxifen-treated mice, which could signify inefficient production of these 215 cells in the bone marrow (Fig. 3a). Despite this reduction, we did not detect significant 216 differences in neither the number nor the percentage of granulocytes in whole blood cell counts 217 that is highly expressed in the brain, while NTRK1, that codes for the TrkA kinase, is almost 249 undetectable in the adult brain ( Supplementary Fig. 4b). The mouse homologues, Bcan and 250 Ntrk1, located on Chr3, have a similar gene structure and expression pattern to their human 251 counterparts ( Supplementary Fig. 4a-b). Hence, we argued that the Bcan-Ntrk1 fusion would be 252 an appropriate genomic alteration to be studied with the RCAS/tv-a/Cas9 system. 253 In order to generate the Bcan-Ntrk1 gene fusion we designed gRNAs in the introns 13 254 and 10 of Bcan and Ntrk1, respectively (Fig. 4a). The pair of gRNAs was subsequently cloned 255 into an RCAS plasmid containing both a hU6 and mU6 promoters (hU6-gRNA-mU6-gRNA- As shown in figure 4f, the Bcan-Ntrk1 tumorspheres were exquisitely sensitive to Trk inhibition, 296 while no effect was observed on the control p53-null TVA-Cas9 NSCs. Entrectinib led to a 297 significant reduction of tumor cells growth associated with an increase of the number of 298 apoptotic cells, detected as sub-G1 population in a propidium iodide staining ( Fig. 4f-g). 299 Overall these data indicate that the RCAS/TVA-CRISPR/Cas9 system is a very powerful 300 model to study the role of gene fusions in tumorigenesis and as possible therapeutic targets. 301 302

Generation of the Myb-Qk chromosomal translocation 303
The human BCAN-NTRK1 and mouse Bcan-Ntrk1 fusions are generated by a small 304 chromosomal deletion of approximately 200Kb. To test whether the RCAS/tv-a/Cas9 system was 305 also suited for inter-chromosomal translocations, we decided to model the MYB-QKI gene 306 fusion, a recently identified putative driver of a subtype of pediatric low-grade gliomas (PLGG), 307 known as angiocentric gliomas 32 . Although MYB and QKI are both located on Chr6 in human, 308 the mouse homologues Myb and Qk are located on different chromosomes, Chr10 and Chr17, 309 respectively. 310 MYB encodes for a transcription factor that is a key regulator of hematopoietic cell 311 proliferation and deregulated MYB activity has been observed in variety of human cancers. QKI 312 is a tumor suppressor gene that encodes for a RNA-binding protein, QUAKING, that plays a role 313 in the development of the CNS, among other organs. Several MYB-QKI gene fusions have been 314 described in angiocentric gliomas, all of them involved the same QKI 3' region (exon 5 to 8) 315 fused to different MYB exons (1-9, 1-11 or 1-15). Here, we focused on the most frequent MYB 316 (exon1-9) -QKI (exon 5 to 8) fusion event. 317 To generate the mouse Myb (exon1-9) -Qk (exon 5 to 8) fusion, we designed gRNAs in 318 the intron 4 for Myb and 9 for Qk (Fig. 5a), and we cloned them into the RCAS-gRNA-pair. 319 Genomic PCR and sequencing from the NIH-3T3 TVA-Cas9 (data not shown) and also from 320 p53-null TVA-Cas9 NSCs infected with the RCAS-Qk-gRNA-Myb-gRNA, confirmed the 321 generation of the Myb-Qk fusion (Fig. 5b). By RT-PCR we also observed the expression of the 322 Myb-Qk transcript (Fig. 5c). 323 In human and mouse normal adult brain, Myb mRNA expression is almost undetectable 324 ( Supplementary Fig. 5b). The MYB-QKI fusion has been shown to functionally activate the MYB 325 promoter and to possibly contribute to an autoregulatory feedback loop 32 . Indeed, when we 326 measured Myb expression in cells expressing the Myb-Qk fusion we observed an increase of Myb 327 mRNA as compared to control cells (Fig. 5d). We also observed increased mRNA levels of a 328 series of genes (Erbb2, Cdk6 and Slc9a31) that have been shown to be upregulated by the MYB-329 QKI fusion 32 (Fig. 5d). 330 We then tested their transforming potential in vitro by plating the cells in soft-agar, and 331 we observed that the p53-null NSCs expressing the Myb-Qk fusion, but not the p53-null NSCs 332 infected with the Ctrl gRNA cells, were able to form colonies. 333 334

Modeling BRAF V600E mutation by homology directed repair (HDR) 335
One of the known applications of the CRISPR/Cas9 system is to induce point mutations 336 through Homologous Recombination (HR). Delivery of a gRNA with either double-stranded 337 DNA (dsDNA) or single-stranded DNA (ssDNA) repair templates, containing a desired modified 338 sequence together with variable length upstream and downstream homology arms, has been used 339 to recreate oncogenic driver mutations 4 .
Activating mutations in the BRAF kinase gene (V600E) have been identified in various 341 types of pediatric gliomas (Pilocytic astrocytomas (<10%), pleomorphic xanthoastrocytomas 342 (WHO grades II and III; 50%-65% cases), gangliogliomas (20%-75% cases)) and also adult 343 high-grade gliomas (5%) 33 . 344 To model a missense gain-of-function Braf mutation we used the strategy previously 345  Fig. 6a). This latter mutation, although 357 undesired, is a conservative mutation from an aspartate to an asparagine residue and it's not 358 expected to have any functional consequence on BRAF activity. 359 When transplanted intracranially into NOD/SCID mice, the p53-null Braf V637E NSCs 360 induce tumor formation in 100% of the injected mice (6/6), with an average survival of 66 ± 11.5 361 days. Histopathological examination of the tumors evidenced a number of features characteristic 362 of high-grade gliomas: nuclear atypia, high number of mitotic figures and necrotic areas (Fig. 6b  363 and Supplementary Fig. 6b). Moreover, we observed clusters of tumor cells infiltrating the 364 normal brain parenchyma, with the vast majority of these cells surrounding tumor vessels 365 ( Supplementary Fig. 6b), resembling the vascular co-option observed both in primary and 366 metastatic brain tumors. It is also to note the presence of some giant cells ( Supplementary Fig.  367 6c) and areas of the tumors with epithelioid morphology (Supplementary Fig. 6d). 368 Immunohistochemical analysis of the BRAF mutant tumors revealed high percentage of 369 Ki67 positive cells, positivity for both OLIG2 and NESTIN and elevated MAPK kinase activity, 370 as evidenced by pERK IHC (Fig. 6b). Additionally, we were able to confirm the expression of the BRAF V637E mutation using an antibody specifically designed to recognize the human 372 BRAF V600E mutant. 373 To validate that the BRAF mutant tumors were dependent on an active BRAF signaling 374 pathway, we isolated tumorspheres from two of those tumors and we treated them in vitro with 375 Dabrafenib, a specific BRAF inhibitor that is currently in clinical trials for BRAF V600E mutant 376 melanomas. Both tumorspheres lines carried the Braf V637E mutation as confirmed by PCR and 377 Sanger sequencing (Supplementary Fig. 6e). As shown in figure 6d, Dabrafenib treatment 378 induced growth reduction in both Braf V637E tumors, but not in p53-null NSCs. Moreover, western 379 blot analysis showed a reduction of MAPK kinase signaling pathway after exposure to 380 Dabrafenib in Braf V637E tumor cells but not in control cells (Fig. 6e). 381 In summary, we have presented a platform that will be useful to study not only the role of 382 tumor suppressor genes and genomic rearrangements but also of potential oncogenic mutations. Immunoblotting 586 Cell pellets were lysed with JS lysis buffer (50mM HPES, 150mM NaCl, 1% Glycerol, 587 1% Triton X-100, 1.5mM MgCl 2 , 5mM EGTA) and protein concentrations were determined by 588 DC protein assay kit (Biorad). Proteins were separated on house-made SDS-PAGE gels and 589 transferred to nitrocellulose membrane (Amersham). Membranes were incubated in blocking 590 buffer (5% milk 0.1% Tween, 10 mM Tris at pH 7.6, 100 mM NaCl) and then with primary 591 antibody either 1 hour at room temperature or overnight at 4°C according to the antibody.  Supplementary Fig. 3a). 626 Complete blood counts were carried out using the Abacus Junior Vet (Diatron). Cells 627 were isolated from spleen and brain by mechanical disruption. Red Blood Cells were lysed using 628 the red blood cell lysis buffer (Sigma-Aldrich). All cells were stained with CD45-PerCP 629

Statistical analysis 702
Data in bar graphs are presented as mean and SD, except otherwise indicated. Results 703 were analyzed by unpaired two-tailed Student's t-tests using the R programming language 60 . 704 Kaplan-Meier survival curve were produced using the "survminer" R package and P values were 705 generated using the Log-Rank statistic. Box-plots were made with the "ggplot2" R package. 706 Drug dose response curves were produced with GraphPad Prism.