Modeling Gliomas Using Two Recombinases

Gliomas form the most common and a heterogeneous group of primary brain tumors. The most malignant form of glioma, glioblastoma, is one of the most lethal cancer types with a median survival of less than 15 months (1, 2). Gliomas display enormous intratumoral heterogeneity that imposes severe limitations for development of therapeutic approaches (3–5). Better understandings of cellular and molecular components embedded in heterogeneous tumors are essential to improve personalized treatment. Therefore, a pressing need for neuro-oncology is the ability to accurately and effectively interrogate cellular and molecular candidates in the complex tumor tissues under controlled circumstances.

Genetically engineered mouse models (GEMM) serve as first-line model systems to transform our understanding of tumor development, and to provide preclinical platforms for evaluating the efficacy of antitumor drugs (6, 7). In these models, a sufficient, but usually minimal, number of cancer-related genes is delivered to cell(s)-of-origin of tumors to initiate tumorigenesis in mice. Consequently, tumors arise de novo and coevolve with native microenvironment in the organ, thereby resulting in the formation of tumors unlike the canonical transplantation models with established tumor cells, which usually exhibit rapid formation of tumors regardless of their microenvironment (8, 9). In fact, GEMMs for gliomas recapitulate key aspects of the human disease, including genetic, temporal, and histopathologic features (10–13), which shed light on the importance of cell of origin, progression, heterogeneity, and the tumor stroma that cannot be easily studied in human cancers (14–16).

From the technical perspective, site-specific recombination techniques such as Cre/loxP have been powerful methods centrally used for the studies of gene function, modeling diseases, and the labeling and manipulation of specific cell populations in vivo (17). Traditionally, GEMMs for cancer research have employed Cre-mediated recombination, which limits further genetic manipulation in these model systems. Although alternative recombination systems such as FLP, PhiC31, Dre, and Vika have been identified, and engineered to efficiently process catalytic reactions in the mammalian cells, GEMMs with these alternatives have at present limited availability (18–21).

Here, we describe the development of new transgenic animals, GFAP-FLPo mice, for modeling gliomas using lentiviral vectors. This GFAP-FLPo mouse strain expresses FLPo recombinase exclusively in GFAP-positive cells in the brain, which allows us to restrict the initial sources of cell of origin of gliomas with the delivery of a combination of Hras^G12V and short hairpin RNA (shRNA) against Trp53. Furthermore, stage- and tumor-specific activation of Cre recombinase combined with multifluorescent protein expression system visualizes morphologically heterogeneous tumor populations in gliomas initiated even with strong oncogenic events. This model will provide more flexible choice of strategies to study basic principles of the complex glioma ecosystem.

All mice were maintained under pathogen-free conditions at the Salk Institute (La Jolla, CA), and all procedures performed in this study were approved by the Institutional Animal Care and Use Committee. DNA microinjection method was employed at the Salk Transgenic Core to achieve random integration of GFAP-FLPo transgene into the mouse genome to generate GFAP-FLPo transgenic mice. For the construction of GFAP-FLPo vector, GFAP-cre vector (Addgene catalog no. 24704; ref. 22) was used as an initial source of 15-kb GFAP promoter cassette and Cre sequence was replaced by FLPo-poly(A). This Gfap promoter cassette contains all components for mouse Gfap gene expression including regulatory elements, exons, introns, and 3′ UTR, while the expression of endogenous Gfap gene expression is disrupted by a point mutation at start codon sites (ATG to TTG; Fig. 1A). FLPo-bGHpoly(A) fragment was constructed on pLV-GFP vector, and a NotI/SalI-digested fragment containing FLPo-bGHpoly(A) was then ligated into the NotI/SalI-digested GFAP-cre vector (into the first exon of Gfap gene). The construction was verified by restriction enzyme digestion and sanger-sequencing method. GFAP-FLPo-bGHpoly(A) vector was then linearized with SfiI digestion and purified for microinjection. Obtained GFAP-FLPo mice were backcrossed into a C57BL/6J (Jackson laboratory) background for at least 8 generations. GFAP-FLPo transgenic line #62 will be available from The Jackson Laboratory as JAX Stock No. 033116. FRT-reporter RC::FLTG (23), FRT-STOP-FRT-KrasG12D (24), and Rosa26-CAG-Brainbow2.1/Confetti (25, 26) mice were obtained from the Jackson Laboratory (Stock No. 026932, 008653, and 013731, respectively). Tamoxifen (T5648, Sigma) was dissolved in sunflower seed oil (S5007, Sigma) and injected intraperitoneally into mice for 3 days at a concentration of 100 mg/kg body weight.

PCR primers to amplify the conjugation part of GFAP-FLPo transgene were Gfap Fw: CGGAGACGCATCACCTCTG, FLPo Rv: GATCAGCTTCTTCAGGCTG. The expected size, 570 bp of amplicon, was determined by agarose gel electrophoresis.

Thick (100–200 μm) brain sections were prepared with vibratome and stored in tissue freezing media (25% Glycerol, 30% Ethylene glycol, 1.38 g/L NaH₂PO₄, and 5.48 g/L Na₂HPO₄). The primary antibodies used in this study as follows; anti-RFP/tdTomato (MBS448092; MyBioSource), anti-GFAP (Z0334; Dako), anti-NeuN (MAB377, Millipore Sigma), anti-Olig2 (AB9610, Millipore Sigma), and anti-Sox2 (14-9811-80; eBioscience). The secondary antibodies were as follows (all from Invitrogen): Alexa Fluor 568 anti-goat IgG, Alexa Fluor 647 anti-rat IgG, Alexa Fluor 647 anti-rabbit IgG, and Alexa Fluor 647 anti-mouse IgG. The nucleus was stained with DAPI.

Brainbow2.1 tissues with 200–1 mm thickness were prepared with vibratome and then cleared with tissue-clearing method, ScaleS (27). Briefly, processed brain tissues were transferred into either ScaleSQ(0) or Triton X-100–containing ScaleSQ(5) and incubated at 37°C for 4 hours, and then transferred into S4(0) to be incubated at room temperature for overnight. The images were acquired with laser-scanning microscopy, LSM 810 or 880 with Airyscan (Zeiss) equipped with the laser lines (405, 458, 488, 514, 561, 594, and 633 nm). Separation of brainbow2.1 fluorophores was obtained by using a 458-nm laser line for mCFP (458–498 nm barrier filter), a 488-nm laser line for nGFP (491–518 nm barrier filter), a 514-nm laser line for EYFP (518–562 nm barrier filter), and a 561-nm laser line for RFP (588–635 nm barrier filter). The obtained images were processed with Imaris (Bitplane) or Zen (Zeiss) software.

VSV-pseudotyped third-generation lentiviruses were produced by Polyethylenimine-MAX (Polyscience) or Lipofectamine 2000 (Thermo Fisher Scientific)-based transfection of 293T cells with a transfer plasmid, packaging plasmids (GAG/POL, RSV-Rev), and an envelope plasmid (VSV-G). Sodium butyrate (2 μmol/L) was added into medium on the occasion of transfection and medium change to increase the viral production (28). Supernatant was collected 48 and 72 hours posttransfection, and lentiviruses were concentrated by the ultracentrifugation. Biological titer of lentiviruses was evaluated with 293T based on fluorescence expression. 293T cells were cultured in DMEM (Corning, 10-013-CV) supplemented with 10% heat-inactivated FBS. FBS was evaluated and selected on basis of the assays measuring 293T proliferation.

Lentiviruses (1 μL, 4 × 10⁸–10⁹ IFU/mL, 0.1 μL/30 seconds–1 minute) were stereotaxically injected into GFAP-FLPo mice at the age of 6–16 weeks old under anesthesia. The following coordinates were used (in mm posterior, lateral, and dorsal to the bregma): hippocampus (HP; 2.0, 1.5, 2.3) and cortex (CTX; 1, 1, 0.5).

Statistical analyses were performed using GraphPad Prism software. The Mann–Whitney U, Student t, and Fisher exact tests were used for statistical evaluations. Log-rank tests were used to evaluate statistical differences in Kaplan–Meier analyses. Data are represented as mean ± SEM or ± SD; P < 0.05 were considered to be significant.

GFAP, a class-III intermediate filament, is a cell-specific marker expressed in astrocytes and neural stem cells (NSC) that have experimentally been shown to be one of the cell of origin of gliomas in the adult mouse brain (10, 29). Detailed in vivo tracing experiments have shown that 15 kb of entire mouse Gfap gene cassette restricts transgene expression to GFAP-expressing cells in the central nervous system more effectively than 2 kb of conventional mouse/human GFAP-5′-flanking region containing cis-acting regulatory elements (30, 31). To direct the expression of an optimized FLPo, specifically to GFAP-expressing cells, we employed mouse Gfap gene cassette with a disrupted start codon of endogenous Gfap gene (22, 32). FLPo recombinase was integrated into first exon of GFAP cassette (Fig. 1A). After the confirmation of FLPo recombination activity in vitro, GFAP-FLPo founder lines were established by pronuclear injection of linearized GFAP-FLPo construct, and GFAP-FLPo transgene carried by 4 GFAP-FLPo founder lines (#32, #41, #62, and #71) was confirmed by genotyping PCR analysis to detect the junction region of FLPo cDNA and GFAP gene cassette (Supplementary Fig. S1A). To confirm in vivo expression of FLPo recombinase, we next analyzed FLPo mRNA expression levels in the brain, kidney, liver, lung, and spleen from 4 GFAP-FLPo lines and WT animals. While no FLPo mRNA was detected in any tissues from #71 mice, FLPo expression was detected predominantly in the brains from #32, #41, and #62 founder strains. We noticed relatively higher expression of FLPo in the nonbrain tissues such as the spleen from #32 and #41 mice unlike endogenous Gfap gene expression (Fig. 1B; Supplementary Fig. S1B–S1D). These data suggest that insertional site of transgene construct may affect level and tissue specificity of transgene expression and we, therefore, employed #62 mice for further analysis.

To investigate FLP-mediated recombination and its cell-type specificity in the mouse brain, we next crossed GFAP-FLPo mice with reporter mice that express tdTomato after FLP-mediated deletion of a FRT-flanked transcriptional stop cassette (RC::FLTG; Fig. 1C; ref. 23). We first examined whether tdTomato-expressing cells are GFAP-positive in the various brain regions such as cortex, subgranular zone of the hippocampal dentate gyrus, corpus callosum, and thalamus (Fig. 1C). GFAP protein is not expressed in the cell body such as the nucleus, whereas the expression of tdTomato can be observed both in the cell body and the dendrites. We, therefore, examined fluorescently labeled cells in thick brain sections (100–200 μm) using confocal microscopy (Fig. 1D–F). Immunostaining experiments revealed that tdTomato-expressing cells were all GFAP-positive in the brains of 3-week-old animals (Fig. 1D and G; Supplementary Fig. S1E and S1F, Fig. 2A and Table 1). In addition, tdTomato-positive cells in the cortex are all negative for neuronal marker NeuN (Fig. 1E and G) and oligodendrocyte lineage marker olig2 (Fig. 1F and G). We also performed the quantitative analysis of FLPo-mediated recombination in GFAP-positive cells in a region-specific manner (Table 1). These data suggest that FLPo expression is tightly controlled and induced in specific GFAP-positive populations such as astrocytes in GFAP-FLPo animals.

GFAP-positive NSCs reside in the subgranular zone (SGZ) of the hippocampal dentate gyrus, to continuously generate neurons in the adult mouse forebrain (33). We observed that 7.3% of GFAP-positive cells with radial glia cell like morphology was labeled with tdTomato in the SGZ of 3-week-old GFAP-FLPo reporter animals, while the FLPo-mediated recombination event is strongly biased toward GFAP-positive astrocytes in the HP (Supplementary Fig. S2A; Table 1; non-SGZ: 69.6%, SGZ-radial glia like: 7.3% in the HP). Radial glia like tdTomato-positive cells in the SGZ are positive for a stem cell marker Sox2, although astrocytes are also positive for Sox2 in the mouse brain (Supplementary Fig. S2B and Table 1). To investigate whether FLPo-expressing cells in the SGZ display NSC-like properties to produce non-GFAP–positive neuronal lineage progeny, we performed long-term lineage tracing experiments (Fig. 1C). While tdTomato-positive cells were all positive for GFAP and negative for NeuN in the SGZ of 3-week-old animals, GFAP-negative/tdTomato-positive cells were increasingly observed in the 8-week-old and 35-week-old animals (Fig. 2A–C). Notably, the tdTomato-positive cells gave rise to NeuN-positive cells during 8 weeks and 35 weeks in the SGZ, but not in the cortex (Fig. 2B–D). Thus, these data suggest that limited FLPo expression is induced in NSCs in the SGZ of GFAP-FLPo animals.

We have previously established a novel mouse model of malignant gliomas using lentiviral vectors (10, 11). To evaluate whether lentiviral vectors deliver genetic components specifically into GFAP-positive cells in a FLP-dependent manner, we injected CAG-FRT-RFP-FRT-GFP into GFAP-FLPo animals. One week after a lentiviral injection into the cortex of GFAP-FLPo animals, GFP-positive cells were all positive for GFAP and exhibited a sign of hypertrophy that is a key feature of reactive astrocyte seen after acute and focal brain injury (Fig. 3A; ref. 34). To establish FLP-inducible glioma models, we generated a new construct backbone, Frt-pTomo, which harbors FLP-mediated transgene expression platform, 2A peptide-based multigene expression system, and shRNA-mediated gene depletion (Fig. 3B). As we have successfully modeled gliomas by employing a combination of Hras^G12V and p53 suppression in GFAP-Cre animals (11), we generated cytomegalovirus immediate-early promoter (CMV)-FRT-RFP-FRT-GFP-2A-Hras^G12V (Flag-tagged)-U6-shp53 and cotranfected it with either Cre or FLPo recombinase in 293T cells (Supplementary Fig. S3A). FLP-mediated recombination was confirmed by the presence of GFP and Flag-tagged Hras^G12V only after the transduction of FLPo. To induce gliomas with FLP/FRT system, we next directly transduced GFAP-positive cells by injecting Frt-pTomo lentiviral vectors into the CTX or HP of GFAP-FLPo mouse brains. All transduced animals developed tumors when injected either in the CTX or HP (Fig. 3C and D; Supplementary Fig. S3B and S3C). No such tumors were found in animals injected with Hras^G12V alone (Fig. 3C and D), and in GFAP-FLPo; FRT-stop-FRT-Kras^G12D mice during 6-month observation (Supplementary Fig. S3D). Although GFAP-FLPo model required higher virus titer (4 × 10⁵ IFU/mouse) than GFAP-cre model (1 × 10⁵ IFU/mouse), the survival time frame of animals with gliomas was similar to that with tumor generated in GFAP-cre mice (Fig. 3C; Supplementary Fig. S3E). Histologic analysis of established tumors provided highly malignant features seen in high-grade gliomas such as necrosis, cellular pleomorphisms, multinuclear giant cells, and highly infiltrative properties along blood vessels, (Fig. 3E and F). We had previously shown psudopalisading necrosis in mice transduced with pTomo lentiviral vectors (10).

To demonstrate the feasibility of dual recombinase system using Cre and FLPo, we employed CAG-Brainbow2.1 cassette that stochastically expresses either one of four fluorescent proteins after the temporal induction of Cre recombinase activity (Fig. 4A; refs. 25, 26). Tumor-specific induction of multifluorescent proteins distinguishes adjacent glioma cells with different fluorescent proteins and therefore visualizes morphologically heterogeneous features of tumor cells at in situ 3D tissue context (35). We developed a Frt-pTomo vector, which possesses a fusion of a mutated estrogen receptor and Cre (namely Cre^ERT2) to induce Cre activity specifically in tumor cells after the treatment of tamoxifen. To test the accessibility of active form of tamoxifen (4-Hydroxytamoxifen) in FLP-induced glioma tissue, Cre^ERT2-mediated recombination, and leakiness of the Cre^ERT2 system, we first infected GFAP-FLPo; CAG-Brainbow2.1 animals with Frt-pTomo vectors expressing Cre^ERT2, Hras^G12V, and shp53 to induce gliomas. Although there was no evidence of fluorescent proteins expression in the normal brain of GFAP-FLPo; CAG-Brainbow2.1 animal, we did detect small amounts of leakiness of recombination activity in the pathologic area even in the absence of tamoxifen (Supplementary Fig. S4A). High-dose tamoxifen treatment 4 weeks after virus infection induced nearly equal amount of 4 fluorescent protein expression (Supplementary Fig. S4A). At 1, 4, and 7 weeks postinjection into the CTX or HP, we treated animals with tamoxifen to induce expression of fluorescent proteins and collected tissues with 1-week incubation time (Fig. 4A–C; Supplementary Fig. S4B and S4C). With this technique, we observed at least two morphologically distinct cell populations, small cells (16.70 μm ± SEM 1.373, N = 30) that dominated tissues at early timepoints and elongated cells (105.8 μm ± SEM 5.659, N = 24) that frequently showed infiltrative phenotype along blood vessels (Fig. 4B–D). These cells are consistently observed even with the injection of viruses into the HP (Supplementary Fig. S4B and S4C). Although gliomas are initiated synchronously by lentiviruses with defined oncogenic insults most likely into limited number of astrocytes, the established tumor possesses heterogeneous phenotypes.

The development of model systems that facilitates our ability to investigate gliomas at the molecular and cellular level, is an essential process to combat such an incurable disease. Among cancer models such as cell lines, organoids, model organisms, and patient-derived xenografts, GEMMs are specifically equipped with the capacity to study progression and drug response under the influence of native microenvironment (8, 9, 36). In this study, we describe the development of a novel glioma model using FLP/FRT recombination system. This glioma model should enable additional manipulation of mouse genome mediated by Cre in any cell type at any time in vivo, and therefore, accelerate the precise understanding of cellular and molecular mechanisms in the developmental processes of gliomas. In fact, our observation reveals that tamoxifen treatment sufficiently induces Cre^ERT2-mediated recombination in glioma cells at any stages of gliomagenesis. Thus, various genetic strategies using Cre recombinase, (i) temporal and sequential induction of oncogenic events rather than delivering those to cell of origin at the same time, (ii) temporal depletion of genes in established tumors to perform a proof-of-concept study for therapeutic feasibility, (iii) specific depletion of genes in stroma such as immune cells and endothelial cells that are abundant in the tumor tissues, and (iv) lineage tracing of tumor cell/stroma subpopulations during gliomagenesis and drug treatment, could be employed with this model.

Genetic, epigenetic, and microenvironmental factors collectively determine the properties of glioma cells, and any alterations or changes in those regulatory components theoretically provide the opportunity to develop heterogeneous phenotypes in glioma tissues (37–39). In fact, recent longitudinal and multiregional analyses of clinical biopsies support the presence of multidirectional transitioning of glioblastoma subtypes and phenotypic association of glioblastoma subtypes with their surrounding microenvironment (40, 41). Furthermore, various experimental evidences provide mechanistic insights of highly plastic nature of glioma cells partly regulated by developmental programs, miRNA, and inflammation (42–45). Genetics and chemical-based manipulation of factors that are highly associated with heterogeneous phenotypes using model systems would be further required for the precise understandings of tumor heterogeneity and progression.

Although a combination of mutations in Ras and Tp53 is not common for human gliomas, delivering such oncogenic insults to GFAP-positive cell of origins constantly develops mouse gliomas, which exhibit many features of human malignant gliomas as described previously (11, 46). One of advantages of lentivirus-based animal model is rapid and flexible choice of strategies to manipulate cancer-related genes to initiate gliomas. The development of FLP/FRT-mediated mouse models initiated with different sets of the relevant driver genes will allow for the detailed analysis of tumor-specific molecular and cellular mechanisms, which should provide key insights into intertumoral heterogeneity. Furthermore, even with strong oncogenic drivers for tumor initiation, established tumors exhibit histologically and morphologically heterogeneous phenotypes. These evidences imply that tumor-intrinsic and/or extrinsic changes may be additionally required for glioma cells to shape the trajectory toward malignant gliomas and also to develop heterogeneous phenotypes. Further investigations using transcriptomic analysis to capture gene expression dynamics of single cells from different stages, and in situ analysis to characterize tumor-microenvironment interaction would provide the mechanistic insights of cause of both of inter- and intratumoral heterogeneity underlying therapeutic failure in human gliomas.

Recently, CRISPR (clustered regularly interspaced short palindromic repeats)/Cas9 system has been applied for modeling of human cancers in mouse (7, 47). As the generation of GEMMs and breeding animals are time-consuming processes, use of CRISPR/CAS9 system to directly mutate cancer-related genes in vivo provides more flexible and rapid strategies to interrogate candidate genes. However, the variety of insertions and deletions (indels) might be created in the initial pool of cell of origins, which potentially can result in the development of established tumors with experimentally generated heterogeneity. Careful recognition of advantage and disadvantage of model systems is required for further application, and lineage tracing strategies to tag the cell(s)-of-origin at single-cell level might be useful to interpret the data (48).

In conclusion, our mouse model expands the utility of GEMMs in neuro-oncology, and will enable a more rapid and complete understandings of the molecular and cellular regulators of all aspects of gliomagenesis.

No potential conflicts of interest were disclosed.

Conception and design: T. Hara, I.M. Verma

Development of methodology: T. Hara

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): T. Hara

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): T. Hara, I.M. Verma

Writing, review, and/or revision of the manuscript: T. Hara, I.M. Verma

We thank Dinorah Friedmann-Morvinski, Yasushi Soda, Eugene Ke, Gerald Pao, Junko Ogawa, Xiaoyan Zhu, Nina Tonnu, and Mark Schmitt for helpful discussions and supports. This work was supported, in part, by a Grant-in-Aid for JSPS Fellows from the Japan Society for the Promotion of Science, a grant from the NIH-NCI (5R01CA195613), Cancer Center Support (5P30CA014195), the H.N. and Frances C. Berger Foundation, and Leona M. and Harry B. Helmsley Charitable Trust grant 2017-PG-MED001. This work was also supported by the Transgenic Core Facility of the Salk Institute, with funding from NIH-NCI CCSG: 5P30CA014195, and by the Waitt Advanced Biophotonics Core Facility of the Salk Institute, with funding from NIH-NCI CCSG: 5P30CA014195 and the Waitt Foundation.