Identification of candidate cancer-causing genes in mouse brain tumors by retroviral tagging

Abstract

Murine retroviruses may cause malignant tumors in mice by insertional mutagenesis of host genes. The use of retroviral tagging as a means of identifying cancer-causing genes has, however, almost entirely been restricted to hematopoietic tumors. The aim of this study was to develop a system allowing for the retroviral tagging of candidate genes in malignant brain tumors. Mouse gliomas were induced by a recombinant Moloney murine leukemia virus encoding platelet-derived growth factor (PDGF) B-chain. The underlying idea was that tumors evolve through a combination of PDGF-mediated autocrine growth stimulation and insertional mutagenesis of genes that cooperate with PDGF in gliomagenesis. Common insertion sites (loci that were tagged in more than one tumor) were identified by cloning and sequencing retroviral flanking segments, followed by blast searches of mouse genome databases. A number of candidate brain tumor loci (Btls) were identified. Several of these Btls correspond to known tumor-causing genes; these findings strongly support the underlying idea of our experimental approach. Other Btls harbor genes with a hitherto unproven role in transformation or oncogenesis. Our findings indicate that retroviral tagging with a growth factor-encoding virus may be a powerful means of identifying candidate tumor-causing genes in nonhematopoietic tumors.

Retroviral insertional mutagenesis in mice causes a high incidence of tumors, and proviral tagging is a powerful method to identify cancer-related genes in hematopoietic tumors (1–3). With the exception of mouse mammary tumor virus-induced carcinomas (4), retroviral tagging has, however, not yet been applicable to solid tumors. The aim of this study was to tag genes in malignant brain tumors by using an approach for retroviral tagging, whereby gliomas were induced in mice with a recombinant Moloney murine leukemia virus (MMLV) encoding the platelet-derived growth factor (PDGF) B-chain (MMLV/PDGFB) (5). The model is based on the idea that segregation of fully malignant clones is a consequence of retroviral insertional mutagenesis of genes that cooperate with PDGF in gliomagenesis. The rationale for using PDGF as an initiating oncogene is based on the widely accepted notion that an autocrine activation of PDGF receptors is an important event in the genesis of human glioblastoma (6, 7). This notion is based on the finding that PDGF and PDGF receptors are commonly coexpressed in glioma. Moreover, the PDGF α-receptor gene is amplified in a subset of glioblastoma (reviewed in ref. 8), suggesting that activation of PDGF receptor signaling confers a selective growth advantage in tumor growth.

Our study yielded a list of 66 potential brain tumor loci. Several of these correspond to known tumor genes in cancer-causing pathways, whereas others encode proteins with previously unknown involvement in malignancies or PDGF signaling.

Materials and Methods

Tumor Induction, Histological Analysis, and DNA Isolation. The generation of MMLV/PDGFB-induced mouse brain tumors and preparation of genomic DNA have been described in ref. 5. In the study, we included mice from several injection series by using wild type C57BL/6 mice (purchased from Charles River Breeding Laboratories), TSG-p53^–/– and TSG-p53^+/+ mice (129/Sv × C57BL/6 background) (obtained from Taconic Farms), Ink4a-Arf^–/–, heterozygotes and wild-type littermates (129/Sv × C57BL/6 background) (received as a gift from Manuel Serrano, Cold Spring Harbor Laboratory, Cold Spring Harbor, NY). All experiments were performed in compliance with the local animal ethics committee (decision numbers C158/98, C154/1, and C47/2). Of the mice that received MMLV/PDGFB injections, 127 were wild-type C57BL/6, 8 were Ink4a/Arf ^+/+, 25 were Trp53^+/+, 19 were Trp53^–/–, 5 were Ink4a/Arf ^+/–, and 24 were Ink4a/Arf^–/–. Only brain specimens with a macrosopically and/or microscopically confirmed tumor tissue were included. One of the wild-type tumors was serially transplanted s.c. in nu/nu mice. These xenografts were included in the study.

Inverse PCR (IPCR). Host genomic sequences flanking the 5′ LTR of the integrated provirus were cloned by using IPCR as reported in ref. 9. A slight difference in DNA sequence made it possible to discriminate helper virus from PDGFB-containing virus in the IPCR. Sequences for the primers used in the first round of amplification were 5′-OS (5′-AATCTGCCGTCATCGACTTCG), 3′-O (5′-TTAGAGGAGGGATATGTGGTT), and 3′-OB (5′-TGTCCGAGGGGTACGTGGCTT). Primers used in the nested PCR were 5′IS (5′-CCCACCACCATCACTTTCAAA) and 3′-I (5′-GCGCGTCTTGTCTGCTGCAG). 5′-IS and 5′-OS were directed to a specific SupF tag in the retroviral construct (5). Primers 5′-OS and 3′-O were used for detection of helper virus, and 5′-OS together with 3′-OB were used for detection of PDGFB virus. Primers 5′-IS and 3′-I were used for both viral types. PCRs were performed as described in ref. 9.

Cloning and Sequencing of IPCR Fragments. IPCR products were cloned into pGEM-T Easy vectors (Promega). The products were sequenced from both directions with universal M13–21/T7 and SP6 primers, and the sequencing was performed on ABI Model 377 and 3700 DNA Sequencers (Applied Biosystems) by following the manufacturer's protocol (Big Dye Terminator Cycle Sequencing Ready Reaction Kit protocol, Applied Biosystems).

Sequence Analysis. From a total of 108 tumors, we retrieved 647 provirally tagged sequences (PTSs). To identify candidate genes, we carried out blast searches by using the ensembl mouse genome sequence database (Mouse Genome Sequencing Consortium, Mouse Assembly, v.19.30.1, 16 Dec. 2003). In addition, the Celera Discovery System search platform was used as a backup when the information obtained from the public database did not suffice. Most PTSs could be assigned a chromosomal location with an E value of 10^–25 or less. Precise location and transcriptional direction of the 647 proviruses were determined (the complete list is provided in Table 1, which is published as supporting information on the PNAS web site). A few shorter fragments with higher E values but with 100% sequence identity were also included. One percent of the fragments showed no significant similarity with the sequences in the databases and were excluded from further analysis. Seventy-three percent of the PTSs were located within 10 kb upstream or downstream of an annotated gene or EST sequence.

Results and Discussion

Newborn mice were injected intracerebrally with MMLV/PDGFB together with replication competent helper virus (MMLV). Sequences flanking the 5′ end of inserted provirus were cloned from 108 individual tumors (75 from wild-type mice, including serial transplants from 1 tumor, 14 from Trp53^–/– mice, 18 from Ink4a/Arf ^–/– mice, and 1 from an Ink4a/Arf^+/– mouse). Of the cloned PCR fragments, 647 were informative and unambiguously mapped individual integration sites (see Table 1). A common integration site (CIS) was defined as a site where two, three, or more than three insertions were located within a maximum distance of 30, 50, and 100 kb, respectively. Additionally, two or more insertions within the same annotated gene was also considered a CIS. These criteria have been set according to previously published statistical analysis (2). The inclusion criteria probably overestimate the number of relevant CISs, given that a recent study showed that murine leukemia virus integration is not an entirely random process but favors transcription start regions (10). The inclusion criteria were met by 170 of the 647 insertion sites. These sites were grouped into 66 CISs, which were denoted brain tumor locus (Btl)-1 to Btl-66 (Table 2, which is published as supporting information on the PNAS web site). One CIS was targeted six times (Btl-47) (Fig. 1A), five were tagged five times [Btl-6, Btl-8, Btl-18 (Fig. 2A), Btl-21, and Btl-40], and one was targeted four times (Btl-24). Of the remaining CISs, 17 were targeted three times and 42 were targeted twice. Several of the Btls were tagged both in wild-type mice and Trp53-null and/or Ink4a/Arf-null mice. The Btls were grouped according to the structure and/or function of the protein products of targeted genes (Table 2).

Fig. 1.

Schematic illustration of provirally tagged genes. Data were obtained from National Center for Biotechnology Information Build 32 (adapted from the ensembl Mouse Genome Browser and the University of California, Santa Cruz, Genome Browser). Arrows denote position and transcriptional orientation of the proviral genomes. (A) Btl-47. Proviral integrations tagging D8Wsu151e (TAX1BP2 mo) in six different tumors. (B and C) A total of six proviral integrations were found downstream of the Rhesus blood group family members b and c (Rhbg and Rhcg). Chr., chromosome.

Fig. 2.

Schematic illustration of the proviral tagging of the Nfi gene family. Data were obtained as described in Fig. 1. Arrows denote position and transcriptional orientation of the proviral genomes. (A) Five proviral integrations within 9 kb found within or near the transcriptional start site of Nfix. (B) Single proviral integrations in the three other members of the NFI family (Nfia, Nfib, and Nfic). Chr., chromosome.

Several of the tagged loci harbor genes known to be involved in neoplastic transformation and oncogenesis. These genes include Ddr1 (Btl-25), Trp53 (Btl-1), Fancc (Btl-60), Rad51l1 (Btl-58), Eef1a1 (Btl-50), Gli (Btl-52c), and Fos (Btl-59). An important aspect of these integrations is that they strongly support the underlying idea of the study and provide proof of concept of the model of retrovirally induced gliomagenesis. Whereas the case for these genes in neoplasia is rather strong, the majority of insertions tag genes whose involvement in transformation and oncogenesis is previously unknown. The implication of some of these findings are discussed below.

A number of the Btls were found to harbor genes that are involved in the regulation of the composition of the cell-surface proteoglycans. Sdc3 (Btl-26) encodes the heparan sulfate proteoglycan syndecan 3, whereas Sulf2 (Btl-13) is a recently discovered gene that encodes an endosulfatase that removes sulfate groups from heparin and heparan sulfate proteoglycan (11). An important role of this class of molecules is to modulate the cellular responsiveness to growth factors, such as PDGF (12), fibroblast growth factor (13), and Wnt (14). Interestingly, among the heparan sulfate proteoglycans, syndecan 3 is specifically expressed by oligodendrocyte precursors and down-regulated as the cells differentiate (15). Cspg4 (Btl-49) encodes a chondroitin sulfate proteoglycan (NG2), which is coexpressed with PDGF α-receptors on oligodendrocyte precursors and presumed to modulate their PDGF responsiveness (16). Furthermore, the level of NG2 expression has been shown to increase with tumor progression in human glioma (17). Taken together, these findings highlight the role of heparan sulfate and chondroitin sulfate proteoglycans in growth factor signaling in glial development and gliomagenesis.

Several members of the Ras family of signaling molecules were tagged. This family contains important effectors in signal transduction pathways, which are involved in various cellular processes, including growth regulation, differentiation, membrane motility, cell locomotion, vesicular transport, and endocytosis (18). Insertions were found in one Ras GTPase, Rab5c (Btl-55); two guanine nucleotide exchange factor (GEF) genes [the mouse orthologs of RacGEF, PREX1 (Btl-32), and the unknown GEF domain-containing 2810441C07Rik gene (Btl-38)]; and three genes encoding factors with GTPase-activating domains [the p190RhoGAP gene (Btl-4), the mouse ortholog of RAP1GAP (Btl-12), and the mouse ortholog of the RhoGAP domainencoding ARHGAP9 (Btl-52a)]. The p190RhoGAP gene is located on human chromosome 19q13.3 within a region that is commonly deleted in human glioma and has been implicated as a tumor suppressor (19). This view was recently supported by experimental data that revealed a tumor-suppressing effect of the GAP domain of p190RhoGAP in a mouse glioma model (9). Rap1 has been found to be a negative regulator of the Ras signaling pathway in astrocytes (20). The tagging of RAP1GAP may be interpreted in this context. P-Rex1 is a recently identified activator of Rac and is activated by PtdIns(3,4,5)P3 and β-γ subunits of G proteins (21). P-Rex1 has been coupled to the PDGF receptor-signaling pathway; cells expressing P-Rex1 become sensitized to PDGF-induced membrane ruffling activity (21), an event known to be induced by activated Rac and linked to cell locomotion (22). Interestingly, Rab5 was recently found to be involved in PDGF-induced actin remodeling, particularly circular membrane ruffling that has been linked to 3D cell migration (23). This finding may suggest a role in glioma cell invasion.

Two of the Btls that harbor genes encoding membrane proteins are particularly intriguing. Sppl2b (Btl-51) and Sppl3 (Btl-39) belong to the class of presenilin-like proteases and are structurally related to signal peptide peptidase (24). Signal peptide peptidase is an integral membrane enzyme that catalyzes the degradation of hydrophobic leader sequences after they have been cleaved off from the preproprotein. This process may release biologically active oligopeptides. The function of Sppls is not known, but their close relationship with Spp suggests that they are involved in the processing of the intramembrane portion of integral membrane proteins. The fact that two family members of SPPL were tagged in the mouse glioma model may give a clue to their biological function and help the search for substrates and products of their catalytic activity.

The mouse and human genomes harbor three genes of the Rhesus family of ammonium ion transporters. Two of these were tagged, each in three independent tumors, namely, Rhbg (Btl-33) (25) and Rhcg (Btl-44) (26) (Fig. 1 B and C). The Rh gene family is conserved in evolution and fulfills important functions in yeast and bacteria (27). The role of the Rh family in mammals has not been fully elucidated. RhBG and RhCG are expressed in human tissues that are known to be associated with ammonium transport and excretion. The expression in tissues, such as brain, is more intriguing and may indicate other cellular activities related to ammonium transport, such as signal transduction (27). The finding of three integrations in each of Rhbg and Rhcg in mouse brain tumors suggests an unexpected role in neoplastic cell growth.

A striking result was that the entire family of nuclear factor I (NFI) DNA-binding protein genes was tagged. We found five integrations in the Nfix locus (Btl-18) and one in each of Nfia, Nfib, and Nfic (Table 1) (Fig. 2). The NFI proteins bind to a consensus sequence TTGGC(N5)GCCAA that is present in the promoter regions of a number of cellular and viral genes (28) but not in the MMLV promoter. Expression analyses and gene-targeting experiments have indicated important roles for the NFI family in development. Thus, Nfia^–/– mice suffer from severe brain defects and die perinatally (29). There is some evidence that Nfix may have a role in growth control; overexpression of the protein in mink lung epithelial cells prevents them from transforming growth factor-β-induced growth arrest (30). To our knowledge, however, an oncogenic or transforming activity of the NFI proteins has not been established.

Twelve of the 66 Btls harbor genes that have been tagged in previously reported studies on retrovirally induced leukemia and lymphoma (Tables 1 and 2). These genes include Sulf2, Sox5, Ncor2 (Btl-40b) Fos, Trp53, and Btg2 (Btl-3b). Not surprisingly, our findings only partially overlap previously reported tagged genes. Even though there are considerable similarities and overlaps in the cancer-causing pathways among different tumor types, cell-specific events are to be expected. The Sox family of transcription factors may be taken as a possible example. In leukemia and lymphoma, insertions in Sox4 are very common (2), whereas the present study revealed multiple integrations in Sox5 and Sox10 and only a single integration in Sox4. The present list of candidate genes includes several genes and gene families that have not previously been implicated in oncogenesis. Striking examples are the Rh and Sppl families.

The present results demonstrate the use of an acutely transforming, oncogene-carrying retrovirus for tagging candidate genes in a solid tumor. The model is based on the assumption that the tagged genes cooperate with the initiating oncogene (sis/PDGFB) in the development of the tumor. Interestingly, a recent discussion on the pathogenesis of the T cell leukemias occurring in two patients after gene therapy of X-linked severe combined immunodeficiency (31) has a bearing on the model. In these leukemia cells, insertional mutagenesis of the LMO2 oncogene may cooperate with the retrovirally transduced gene, encoding the γ-chain of cytokine receptors. Further studies will show whether the idea of using a growth factor-encoding retrovirus as both a tumor initiator and as a genetic tag can be generalized and broadly applied to other tumor systems.