The assembly of eukaryotic DNA into nucleosomes and derived higher order structures constitutes a barrier for transcription, replication and repair. A number of chromatin remodeling complexes, as well as histone acetylation, were shown to facilitate gene activation. To investigate the function of two closely related mammalian SWI/SNF complexes in vivo, we inactivated the murine SNF5/INI1 gene, a common subunit of these two complexes. Mice lacking SNF5 protein stop developing at the peri‐implantation stage, showing that the SWI/SNF complex is essential for early development and viability of early embryonic cells. Furthermore, heterozygous mice develop nervous system and soft tissue sarcomas. In these tumors the wild‐type allele was lost, providing further evidence that SNF5 functions as a tumor suppressor gene in certain cell types.
A number of chromatin remodeling complexes have been described in recent years. These complexes were shown to loosen the nucleosomal structure and facilitate access of transcription factors to DNA (Kingston and Narlikar, 1999). Biochemically, several of these complexes have a fairly similar activity, suggesting possible overlapping functions. It is obvious that a genetic approach is essential to unravel their specific function in vivo. The SWI/SNF multi‐protein complex is a prototype of one class of these complexes. The SWI/SNF complexes are evolutionarily conserved and implicated in transcriptional activation of a considerable number of genes through chromatin remodeling (Carlson and Laurent, 1994; Peterson, 1996). However, recent evidence suggests that they may also play a role in transcriptional repression (Holstege et al., 1998). Two related multisubunit SWI/SNF complexes have been characterized in mammalian cells. These complexes share most of their subunits but can be distinguished by their SWI2/SNF2‐related ATPase/helicase catalytic subunit, which is either Brm (SNF2α) or Brg1 (SNF2β) (Wang et al., 1996). Even though both subunits exhibit intrinsic chromatin remodeling activity in vitro, full activity is achieved only when supplemented with BAF170/BAF155 and hSNF5 (INI1), which are conserved subunits of all SWI/SNF complexes (Phelan et al., 1999). Transfection studies in mammalian cell lines, and genetic studies in Drosophila and mice, implicate the catalytic subunit of the SWI/SNF complexes in the regulation of cell cycle progression, possibly by collaborating with pRb in the control of E2F activity (Dunaief et al., 1994; Strober et al., 1996; Muchardt et al., 1998; Reyes et al., 1998; Staehling‐Hampton et al., 1999; Zhang et al., 2000). Moreover, hSNF5 was shown to be mutated in early childhood malignant rhabdoid tumors (MRT) and in various malignancies of the CNS (Versteege et al., 1998; Sevenet et al., 1999a).
We have recently shown that the gene encoding murine SNF2α helicase can be inactivated without impairing development. In these SNF2α−/− mice, increased expression of the SNF2β protein appears to compensate for the loss of SNF2α protein (Reyes et al., 1998). To investigate whether SWI/SNF complexes are indeed essential for development and to explore the role of SNF5 in oncogenesis, we undertook the inactivation of the SNF5 gene.
Results and Discussion
We generated a loss‐of‐function mutation of the SNF5 gene in embryonic stem (ES) cells by replacing exons 1 and 2 with a lacZ reporter gene, encoding β‐galactosidase with a nuclear localization signal (Figure 1A). Western blot analysis of whole cell extracts using a polyclonal antibody raised against the full‐length human SNF5 did not reveal any truncated SNF5 or chimeric LacZ–SNF5 protein in these ES cells, excluding the presence of an interfering protein. We obtained chimeric mice originating from two different heterozygous ES cell clones. These mice transmitted the SNF5–lacZ mutation to F1 offspring, which developed normally.
Analysis of β‐galactosidase activity in heterozygous adult animals revealed that SNF5 was expressed ubiquitously in all organs. Similarly, SNF5 was expressed in all embryonic cells during post‐implantation development (data not shown). To investigate whether the SNF5 gene was activated at earlier time points, we monitored β‐galactosidase activity in fertilized oocytes and in early cleavage stage embryos. Crosses of SNF5+/lacZ males with wild‐type females showed that the onset of SNF5 zygotic expression occurs at the four‐cell stage and is maintained, at least until the blastocyst stage, both in the inner cell mass (ICM) and the trophectoderm (Figure 2A). In contrast, crosses of wild‐type males with heterozygous females revealed the presence of maternal stores of protein in fertilized oocytes and in two‐cell embryos (Figure 2B). The widespread expression of SNF5 in pre‐implantation embryos raised the possibility that the SWI/SNF chromatin remodeling activity may be essential for early development.
To explore this issue we bred SNF5+/LacZ animals. No nullizygous animals were born from these heterozygous intercrosses, indicating that the absence of SNF5 causes embryonic lethality (Table 1A). To determine the developmental stage at which SNF5−/− mice die, embryos from heterozygous intercrosses were isolated at various gestational times and genotyped by PCR. Analysis of embryos from 6.5 and 7.5 days p.c. showed no nullizygous embryos, but 28% of decidua were empty (only 2% detected in crosses between wild‐type and SNF5 heterozygous mice), which might represent resorption sites of homozygous mutant embryos. To test whether SNF5−/− embryos survive to the blastocyst stage, embryos were isolated at 3.5 days p.c. and genotyped. Nullizygous blastocysts were identified in a Mendelian distribution (Table 1B). They did not show any developmental delay and were morphologically indistinguishable from wild‐type blastocysts (not shown). Thus, SNF5−/− embryos develop successfully to the blastocyst stage, but die shortly after implantation, before day E6.5. To determine whether the developmental defect in SNF5−/− mice was due to a proliferation or survival defect following the blastocyst stage, 3.5‐day embryos derived from heterozygous intercrosses were placed in culture and monitored for outgrowth potential. Roughly one‐third of SNF5‐null embryos failed to emerge from the zona pellucida, while the others emerged but the trophectoderm did not spread and the ICM stopped growing after 1 day in culture (Figure 3A). Enzymatic removal of the zona pellucida prior to plating gave identical results. As shown in Figure 3A, in wild‐type and heterozygous embryos the trophoblasts migrated out of the compact blastocysts, attached to the substrate and formed polyploid giant cells. The ICM increased in size and by day 4 formed a discernible node. These features were not observed in any of the mutant blastocysts. The failure of SNF5‐null embryos to hatch or attach in vitro indicates that the trophectoderm is defective. By day 4 in culture, both the ICM and trophectoderm of mutant‐derived cells invariably died. We explored further the cause of SNF5−/− cell death by TUNEL staining, which detects extensive DNA fragmentation resulting from apoptotic cell death. SNF5−/− blastocysts cultured for 96 h displayed widespread TUNEL signal, while apoptotic cells could not be detected in wild‐type and heterozygous blastocysts after 4 days in culture (Figure 3A). These results indicate that SNF5 is essential for the viability of early embryonic cells in culture and its loss results in apoptotic cell death. At the present time we cannot exclude the possibility that SNF5 is required at even earlier times. The survival of mutant embryos to early post‐implantation stages might be due to the presence of maternal stores of SNF5.
It has recently been reported that constitutive mutations of SNF5 predispose to early childhood cancer (rhabdoid predisposition syndrome—RPS) (Sevenet et al., 1999b). Therefore, we monitored SNF5 heterozygous mice for the potential development of tumors. Careful examination of these animals revealed susceptibility to early onset of cancer. In a population of 124 mice, 32% (40 animals) developed tumors by the age of 15 months, the earliest tumors being detected at 4 months of age (see survival curve in Supplementary data, available at EMBO reports Online). The percentage is most likely underestimated since additional heterozygous mice displayed severe wasting symptoms but no macroscopic lesions were detected. Tumor susceptibility was independent of sex or genetic background (C57BL/6 × 129/Sv or inbred 129/Sv). Wild‐type littermates (104 mice) did not develop tumors within the 15 month observation period.
Tumors were detected in different locations, but intra‐cranial (brain, cerebellum, beneath the optic chiasma; 30%) and paravertebral sites (around the dorsal root ganglia or spinal nerve in the spinal cord; 27%) were very common. We observed additional macroscopic subcutaneous tumors in less common sites (see Supplementary data).
Histological analysis was performed on 37 tumors. In spite of the different locations of the tumors, their microscopic morphology was relatively uniform. They appeared as solid sheets of undifferentiated cells, separated by scanty stroma. The lesions usually consisted of a mixture of polygonal or spindle cells, with large clear pleomorphic nuclei, conspicuous nucleoli and eosinophilic cytoplasm. Mitotic figures were scanty. Giant multinucleated cells were often present. In ∼30% of the cases we observed a variable proportion of cells that showed a typical rhabdoid phenotype, with large intracytoplasmic eosinophilic inclusions that displace the nucleus at the periphery of the cell (Figure 4A; Wick et al., 1995). All of the tumors that we analyzed were positive for vimentin staining (19/19) (Figure 4B). Some of the lesions tested showed immunoreactivity for PS100 (14 of 31), NGF‐R (p75) (11 of 25) and GFAP (3 of 25), indicating that they could could originate from neural crest‐derived cells (see Supplementary data). In summary, on the basis of their morphology and immunophenotype, the tumors were classified as undifferentiated sarcomas with variable rhabdoid features.
To determine whether tumor formation in the SNF5 heterozygous mutant mice was dependent upon loss of the remaining wild‐type SNF5 allele, as postulated by Knudson's two‐hit model (Knudson, 1971), we performed Southern blot analysis on a panel of 10 tumor DNAs (four of them are shown in Figure 4C). Even though the tumor samples still contained some normal cells, the Southern blot showed a substantial loss of the wild‐type SNF5 allele, while surrounding tissues retained both the mutant and wild‐type alleles at a 1:1 ratio as expected. These data provide direct evidence that loss of SNF5 function contributes to tumor formation in the mouse.
The phenotype of the mouse model and human MRT syndrome exhibit a degree of similarity, although we noted some differences with respect to tumor incidence and spectrum. Unlike human patients, where SNF5‐associated cancer occurs mainly in very young children, SNF5+/LacZ mice developed tumors as adults. Furthermore, human kidneys are a frequent site for MRT (Weeks et al., 1989), yet we found no intra‐renal macroscopic tumors in our mice. These discrepancies may be explained by species‐dependent differences in susceptibility to SNF5 loss of heterozygousity (LOH), or differences in the growth control pathways of mouse and human cells in particular tissues.
The critical requirement for SNF5 for cell viability in early development may seem contradictory to the tumorigenesis associated with SNF5 loss. However, it is likely that SNF5‐associated chromatin remodeling complexes participate in multiple transcriptional events and therefore affect numerous cellular processes. SNF5 inactivation might result in imbalanced transcriptional regulation and give rise to different effects depending on the cell type. Thus, it is conceivable that in some cell lineages SNF5 disruption is lethal, while in others it impairs differentiation or promotes cell growth. Consistently, it has been shown previously that embryonic carcinoma F9 cells are not viable upon inactivation of Brg1 (Sumi‐Ichinose et al., 1997), whereas some human‐derived tumor cell lines (C33 and SW13) proliferate efficiently in the absence of Brg1 and hbrm (Muchardt and Yaniv, 1993). We cannot formally exclude the possibility that SNF5 might have additional functions beyond its role in the SWI/SNF complexes. However, since inactivation of Brg1, the major helicase/ATPase subunit of SWI/SNF complex present in early development (LeGouy et al., 1998), also results in very early developmental arrest (Bultman et al., 2000), it is likely that the early lethality of SNF5‐null embryos is caused by the loss of functional SWI/SNF complexes. Despite the presence of additional chromatin remodeling complexes in mammalian cells, our results strongly suggest an essential and non‐redundant function for SWI/SNF in embryonic development. Compared with knockouts of other transcriptional coactivators, the SNF5 nullizygous embryos show a similar peri‐implantation lethality as embryos deficient for the RNA polymerase II coactivator subunit Srb7 (Tudor et al., 1999). In contrast, disruption of the transcriptional histone acetylases CBP/p300 or Gcn5l2 results in later lethality (10.5 days p.c.) (Goodman and Smolik, 2000; Xu et al., 2000).
The molecular mechanism underlying tumor development upon SNF5 deficiency is unknown at present. Several studies imply a role for Brm and Brg1 in Rb‐mediated transcriptional repression, in particular, in the repression of genes involved in G1–S progression such as Cycin E and Cyclin A (Trouche et al., 1997; Zhang et al., 2000). However, it should be emphasized that in both human and mouse, the spectrum of tumors arising in Rb and SNF5 heterozygotes is different. Furthermore, Rb−/− embryos die much later than SNF5−/− embryos (Jacks et al., 1992). Considering that p107 and p130, two closely related pocket proteins, display partial functional redundancy with Rb (Mulligan and Jacks, 1998), we might speculate that the SWI/SNF complex could cooperate with Rb relatives as well, and that mice nullizygous for Rb, p107 and p130 would die at peri‐implantation, as observed with SNF5−/− embryos. Nevertheless, the SWI/SNF complexes could be involved in additional growth control checkpoints. Alternatively, cell death resulting from SNF5 inactivation may create a bias in the tumor spectrum obtained in heterozygous animals. Our studies of the SNF5 gene provide a novel example of a dual role for a certain number of tumor suppressor genes. These genes are essential for development and organogenesis; they can be inactivated in certain cell types and cause cancer.
Generation of the SNF5‐targeted allele.
A 15 kb genomic fragment spanning the promoter region and sequence encoding the N‐terminus of murine SNF5 was isolated from a 129/Sv mouse genomic library, using a human SNF5 cDNA probe. Genomic fragments of 3.5 kb and 4.5 kb flanking the first two exons of SNF5 were used to generate a targeting construct in which the first two exons were replaced by a NLS‐lacZ cassette. Briefly, a NcoI site was generated at the initiation codon of SNF5 to allow insertion of the NLS‐lacZ cassette in frame. A 7 kb SalI fragment corresponding to the 3.5 kb SacI–NcoI 5′ flanking region of SNF5 fused to the lacZ reporter gene was inserted upstream of the PGK‐neoR cassette into a modified pPNT vector containing the 4.5 kb XbaI 3′ flanking sequence (Figure 1A). CK35 ES cells (Kress et al., 1998) were electroporated with the NotI‐linearized vector and clones were selected in 300 μg/ml G418 and 2 μM gancyclovir. Among 200 ES cell colonies, three correctly targeted clones were identified by Southern blot analysis, using 5′ and 3′ external probes located outside the homologous regions of the targeting vector (Figure 1B). Two independent targeted clones were injected into C57BL/6 host blastocysts, and chimeric males were mated to 129Sv and C57BL/6 females to generate mixed background and inbred lines carrying the mutation.
Genotyping of mouse tails and embryos.
DNA from either mouse tails or post‐implantation embryos (E6.5 or E7.5) was directly precipitated with isopropanol after proteinase K treatment. For pre‐implantation embryos (E3.5), DNA was prepared by incubation of individual blastocysts in 5 μl of proteinase K buffer (10 mM Tris pH 8.4, 50 mM KCl, 0.01% gelatin + 300 μg/ml proteinase K) for 1 h at 55°C followed by incubation at 95°C for 10 min. The entire DNA isolate was then used directly for PCR. A three‐way PCR was performed with a sense primer within the SNF5 promoter (5′‐AAGGAGCCCAGTAGTGACAC‐3′), and two reverse primers in the SNF5 first exon (5′‐GCCGATCATGTAGAACTCCC‐3′) and in the lacZ gene (5′‐AAGCGCCATTCGCCATTCAG‐3′), generating a wild‐type 220 bp product and a mutant 440 bp product, respectively.
Blastocyst outgrowth study.
Blastocysts generated from heterozygous intercrosses were collected by uterine flush at 3.5 days p.c. and individually cultured for 5 days in gelatin‐coated 96‐well round bottom plates, in 100 μl Dulbecco's modified Eagle's medium supplemented with 15% fetal calf serum, 100 μM β‐mercaptoethanol and penicillin (100 IU/ml)/streptomycin (100 μg/ml) (Gibco‐BRL). After photography, cells were scraped off and collected for PCR genotyping as described for 3.5 days p.c. embryos.
Embryos were collected at different gestation stages, fixed for 10 min in 4% paraformaldehyde and washed extensively in phosphate‐buffered saline. Whole‐mount X‐gal staining was performed overnight as described (Thepot et al., 2000).
Histology and immunohistochemistry.
Tumors were fixed in 10% neutral‐buffered formalin, embedded in paraffin and processed routinely for histology. For immunohistochemical study, standard indirect peroxidase or peroxidase–anti‐peroxidase methods with amplification by the EnVision+ system (Dako) were used on paraffin sections. Primary antibodies were the following: EPOS monoclonal mouse antibodies against vimentin (Sigma; clone vim 13.2); desmin (Dako; clone 33); neuron‐specific enolase (Dako; clone H14); smooth muscle actin (DAKO; clone 1A4); epithelial membrane antigen (clone E29). Rabbit polyclonal antibodies were: multi‐cytokeratins (NCL‐CKp; Novocastra); PS100 (Dako); myosin (Biogenex); GFAP (Dako); p75 nerve growth factor receptor (DAKO).
Analysis of SNF5 loss of heterozygosity.
SNF5 LOH analysis was performed on DNA isolated from fresh tumor samples. The genotype of the tumor DNA was determined by Southern blotting of PstI‐digested DNA and hybridization with the genomic 5′ external probe (Figure 1A).
Supplementary data for this paper are available at EMBO reports Online.
We thank C. Kress for the generous gift of the CK35 ES cell line; C. Coffinier for help and advice throughout the work; O. Delattre and T. Magnusson for sharing unpublished information; Huot Khun for histotechnical assistance; X. Sastre for help with the interpretation of histological data; and B. Bourachot for stimulating discussions. We are grateful to J. Weitzman and S. Schaper for critical reading of the manuscript and friendship. This work was supported by grants from the TMR Marie Curie Research Training, Association for International Cancer Research, Association de la Recherche contre le Cancer and Ligue Nationale contre le Cancer.
- Copyright © 2000 European Molecular Biology Organization