ª
Oncogene (2002) 21, 5504 – 5514 2002 Nature Publishing Group All rights reserved 0950 – 9232/02 $25.00 www.nature.com/onc
Modeling cancer in mice Laurie Jackson-Grusby*,1 1
Whitehead Institute for Biomedical Research, Nine Cambridge Center, Cambridge, Massachusetts, MA 02142, USA
The laboratory mouse is one of the most powerful tools for both gene discovery and validation in cancer genetics. Recent technological advances in engineering the mouse genome with chromosome translocations, latent alleles, and tissue-specific and temporally regulated mutations have provided more exacting models of human disease. The marriage of mouse tumor models with rapidly evolving methods to profile genetic and epigenetic alterations in tumors, and to finely map genetic modifier loci, will continue to provide insight into the key pathways leading to tumorigenesis. These discoveries hold great promise for identifying relevant drug targets for treating human cancer. Oncogene (2002) 21, 5504 – 5514. doi:10.1038/sj.onc. 1205603 Keywords: embryonic stem cell; gene targeting; mouse model; recombinase; inducible transgene; microarray
Introduction Yeast geneticists have long touted the awesome power of their toolbox for identifying genetic pathways important for a range of basic eukaryotic functions (Struhl, 1983). During the past twenty years, an equally impressive range of tools have become available to the mouse geneticist, which has perpetuated the status of the laboratory mouse as the key organism for modeling human genetic disease. Historically, inbred mouse strains that spontaneously develop certain types of tumors, or chemical carcinogens that caused defined lesions in vivo, were studied to understand disease etiology (see: http://tumor.informatics.jax.org, http:// potency.berkeley.edu/). The long tumor latency and the inability to link genotype to phenotype in these spontaneous and carcinogen-induced tumor models were overcome by creating transgenic tumor prone mice. Ground breaking experiments in several laboratories established the first transgenic mouse tumor models by over-expressing viral (Brinster et al., 1984; Hanahan, 1985) or cellular (Adams et al., 1985; Stewart et al., 1984) oncogenes in specific tissues. Germline inactivation of the prototype tumor suppressor genes Rb (Jacks et al., 1992) and Tp53 (Donehower et al., 1992) using gene targeting technology in mouse embryonic stem cells (ES cells) appeared to complete
*Correspondence: L Jackson-Grusby, E-mail:
[email protected]
the necessary tools to model the scope of mutations present in human tumors. Indeed, much of the conceptual framework for using mouse models to study tumor biology was established using these initial oncomice. Cancer, however, is a disease characterized by progressive accumulation of somatic mutations, which was not adequately mimicked using these pioneering technologies. For example, a transgene carrying an activated oncogene may be widely expressed in a given tissue, which does not reflect the sporadic oncogene activation usually seen in focal primary tumors. In addition, multi-copy transgene arrays may not be subject to the same mutational or regulatory mechanisms as the endogenous proto-oncogene. Germline mutations of either oncogenes or tumor suppressor genes may also lead to embryonic lethality that precludes their analysis in tumor models. Such problems can now be avoided by creating tissue-specific mutations that may be activated at defined periods of development (Gu et al., 1994; Lewandoski, 2001). Moreover, these improvements allow the dissection of cell-specific requirements for individual genes, necessary for unraveling complex issues such as tumor-stroma interactions (Hanahan and Weinberg, 2000). Technological advancements in mouse genome engineering have also been accompanied by major advances in molecular analyses of tumors, aided by recent functional genomics efforts. This review will cover the various strategies for engineering oncogenic mutant alleles and the genetic and molecular approaches used to define the pathways altered in tumor prone mice. Engineering mouse tumor models Tumorigenic mutations exert their effects through either dominant gain-of-function oncogene activation or recessive loss-of-function tumor suppressor gene inactivation. Experimental strategies for engineering temporally controlled tissue-specific gain and loss-offunction alleles depend upon the type of mutation being introduced, are outlined below, and are illustrated in each of the figures. Manipulation of the mouse genome is most carefully controlled by targeted recombination in ES cells (Thomas and Capecchi, 1987). Randomly integrated transgenes obtained by pronuclear injection may show variable patterns of expression, which can be avoided by creating a knockin allele that expresses the transgene under the control
Inducible tumor models L Jackson-Grusby
5505
of an endogenous gene, or by selecting a pre-defined favorable locus for integration (McCreath et al., 2000). In vitro differentiation of transgenic ES cell lines allows for screening of appropriate expression patterns for a given transgene before committing to its analysis in mice. Activation alleles Dominant gain-of-function alleles in mice are used to model effects of mutant oncogenes or dominantnegative tumor suppressor gene alleles. Three general strategies have been developed for ligand regulated transgene expression, including nuclear hormone receptor protein fusions, chemically inducible transcription factors, and ligand regulated site-specific recombinases (diagrammed in Figures 1, 2 and 4). An advantage of hormone receptor chimeras is that only one allele need be constructed, whereas the other systems involve expression of two or more components. Nuclear hormone fusion proteins are maintained in a latent inactive state that is released upon hormone binding, thereby activating the fusion protein (Mattioni et al., 1994). Although many in vitro studies describe transcription factor and oncogene fusions to the estrogen receptor (Reichmann et al., 1992), until recently there have been few successful reports from in vivo studies due to leaky background activity. Two groups have modified the estrogen receptor ligand binding domain to retain sensitivity to the synthetic ligand 4’hydroxy-tamoxifen (4OHT) while inactivating the binding to endogenous hormone to circumvent background issues (Indra et al., 1999; Littlewood et al., 1995). Pelengaris et al. (1999) have capitalized on the improved specificity of the tamoxifen-regulated domain by creating a fusion with c-myc. Transgene expression was directed to suprabasal keratinocytes using the involucrin promoter (Figure 1). Topical administration of 4OHT caused myc-dependent hyperproliferation and papilloma formation that entirely regressed upon withdrawal of the drug demonstrating dependence on sustained myc activity for tumor maintenance. Additional hormone receptor fusions to transcription factors and to site-specific recombinases are used in binary transgene activation paradigms described below. Small molecule regulated transcription factors have been developed using (1) nuclear hormone receptor fusions with progesterone (Wang et al., 1999) and ecdysone receptors (No et al., 1996), (2) chemically inducible dimerizers (Ho et al., 1996), and (3) prokaryotic repressors lacI (Cronin et al., 2001) and tetR (Furth et al., 1994) each modified with transcriptional activation domains. These bigenic approaches incorporate tissue specific expression of the latent transcription factor and a responder transgene that links a responsive promoter to the inducible gene of interest (Figure 2). Responder transgenes can be screened for desired expression patterns in vitro, targeted to favorable chromosomal locations, or engineered such that the endogenous promoter is replaced by the responsive promoter (Wutz et al.,
Figure 1 Inducible oncogenesis mediated by an estrogen receptor-c-myc (myc-ERT) fusion protein. Oncogene expression was targeted to keratinocytes using the involucrin promoter, causing papillomatosis with persistent induction by 4’hydroxy-tamoxifen (4OHT). Withdrawal of 4OHT resulted in loss of c-myc target gene expression and tumor regression (Pelengaris et al., 1999)
Figure 2 A mouse model for melanoma formation shows dependence on sustained ras signaling for tumor maintenance. Doxycyclin (DOX) induction of the tet-dependent transcription factor rtTA drives mutant H-ras expression in melanocytes, and leads to melanoma formation. These tumors regress upon withdrawal of the doxycyclin inducer (Chin et al., 1999). tetO, tet operator
2002). Generally the small molecule inducer activates gene expression, however the tet system has added flexibility due to engineered transcription factors that bind either in the presence (tTA) or in the absence (rtTA) of tetracyclines (Urlinger et al., 2000). The maturity of the tet system is reflected by its greater adoption in mice to date. Chin et al. (1999) have beautifully demonstrated the sustained requirement for ras signaling in melanomas by introducing Oncogene
Inducible tumor models L Jackson-Grusby
5506
Oncogene
two transgenes, a tet-regulated Ha-ras V12G allele and a tyrosinase promoter driving the tet-dependent transcription factor rtTA, into the background of Ink4a deficiency (Figure 2). Administration of the tetracycline analog doxycyclin induces melanoma formation, while removal of doxycyclin causes regression. Similar studies have shown sustained requirements for myc in a hematopoietic malignancy (Felsher and Bishop, 1999) and ras in lung adenocarcinomas (Fisher et al., 2001). The ability to repeatedly turn these transgenes on and off makes them ideally suited for studies on temporal requirements for oncogenes at distinct stages of tumorigenesis. One disadvantage of transcriptionally inducible transgenes is the non-physiologic levels expressed from these alleles, which among other effects may alter a gene’s cell cycle regulation. The use of inducible sitespecific recombination preserves the endogenous regulation of the mutated gene, yet provides a single switch for stably activating the oncogenic allele. The mutated allele is held silent by virtue of a transcriptional termination sequence embedded within an intron 5’ to the mutation. Excision of this terminating ‘stop sequence’ is catalysed by a site-specific recombination system, usually bacteriophage P1 derived Cre-loxP (Sauer and Henderson, 1988) or S. cerevisiae derived Flp-frt (O’Gorman et al., 1991). These recombinases recognize 34 base pair sequences comprised of 13 base pair inverted repeats separated by an 8 base pair spacer sequence (Figure 3). In order to target a sequence for deletion, a pair of recombination recognition sequences, either loxP or frt sites, are engineered to flank the sequence. An effective ‘stop sequence’ comprised of an antibiotic resistance gene with four poly adenylation signals has been developed by Soriano (1999), who engineered recombinase reporter alleles at the ROSA26 locus that activate histochemical markers upon site-specific recombination. Ligand regulated Cre (Indra et al., 1999) and Flp recombinases (Logie and Stewart, 1995) have been constructed as
fusions with estrogen receptor or under the control of a tet-responsive promoter (Hasan et al., 2001) to provide temporal control of recombination. Recombinase activation was first used to create an oncogenic mouse strain by Lakso et al. (1992) who expressed SV40 T antigen under the crystalline promoter upon Cre-mediated recombination resulting in lens tumor induction. Jackson et al. (2001) cleverly employed the recombinase activation strategy to induce lung adenocarcinomas with a mutated K-ras gene (Figure 4). The mutant, yet silent, K-ras allele was activated using an adenovirus vector to deliver Cre recombinase to the lung. Delivery of Cre using the adenoviral vector provided two clear advantages over transgenic delivery. First, tumor multiplicity could be controlled by limiting the multiplicity of infection. This model provides a closer approximation of human lung cancer, in which single tumors form and progress through increasing stages of malignancy, than other mouse models involving mutant ras overexpression throughout the lung. Second, the introduction of Cre via a single infection serves as the initiating event, which therefore synchronizes tumor progression. An experimental paradigm that most closely mimics somatic mutational events occurring in sporadic human tumors was recently described by Johnson et al. (2001). Instead of exploiting an inducible recombination system, these authors relied upon endogenous intrachromosomal recombination mechanisms to activate a latent K-ras mutant allele. To prime the chromosome for a mutagenic recombination event, they engineered two non-functional partial K-ras gene copies in tandem. Intra-chromosomal recombination between homologous sequences provided the initiating event, and yielded a single functional oncogenic K-ras allele (Figure 5). A striking 100% of these mice succumb to early onset lung adenocarcinomas, and of these roughly 30% developed thymic lymphomas and skin papillomas. At present it is unclear whether the tumor
Figure 3 Recognition sites for Cre and Flp recombinases. Both recombinases recognize unique 34 bp target sequences, loxP and frt sites, comprised of inverted repeats separated by an 8 bp spacer sequence. This core spacer defines the orientation of the target sequence. To direct recombinase-mediated deletions, the target sequences need to be placed in the same orientation flanking the sequence to be deleted (O’Gorman et al., 1991; Sauer and Henderson, 1988)
Figure 4 Cre recombinase-mediated activation of a K-ras G12D mutation. The G12D point mutation in the K-ras gene is held silent by an excisable cassette that terminates transcription within the K-ras locus. Expression of the point mutation is activated by Cre recombinase, which removes the ‘STOP’ cassette from the gene. Delivery of Cre using an adenoviral vector (AdCre) provides stochastic activation of the mutation in the target organ (Jackson et al., 2001)
Inducible tumor models L Jackson-Grusby
5507
Figure 6 Cre-mediated inactivation of the Apc tumor suppressor gene. A fully functional conditional Apc allele was engineered with loxP sites flanking exon 14 (Shibata et al., 1997). Delivery of Cre using an adenovirus vector (AdCre) resulted in deletion of the sequences flanked by lox sites that includes exon 14, and produced a non-functional Apc allele Figure 5 Activation of a K-ras mutation by intra-chromosomal recombination. A latent K-ras mutant allele comprised of two non-functional but overlapping gene segments was engineered. These direct repeats are a substrate for endogenous intra-chromosomal recombination events that serve to activate the allele (Johnson et al., 2001). Deletion events mediated through homologous repetitive sequences are frequent mutational events found in human tumors
spectrum observed in the latent K-ras mice reflects those tissues most sensitive to oncogenic ras, or alternatively represents cell types with highest rates of intrachromosomal recombination. Construction of additional latent oncogene alleles will undoubtedly follow this elegant study and may serve to address this question. Inactivation alleles Assessment of tumor suppressor function requires somatic inactivation of one or both alleles for a given gene according to the Knudson two-hit model. Germline inactivation of the first allele may be accomplished by standard mutagenesis through homologous recombination in ES cells (Thomas and Capecchi, 1987), while inactivation of the second allele occurs in vivo in cells that undergo loss of heterozygosity (LOH). Heterozygous tumor suppressor gene knockout mice are often valuable models for inherited human cancer syndromes (Donehower et al., 1992), however the expected outcome is not always observed due to species differences of unknown origin (Jacks et al., 1992). Pleiotropic effects of germline mutations and phenotypes that deviate from those caused by similar mutations in human patients can be minimized by using recombinase-mediated tissue-specific gene inactivation (Gu et al., 1994). Analogous to recombinase-mediated activation, this strategy relies upon excision of a critical region of the gene through site-specific recombination (Figure 6). Using the Cre-lox system for example, the engineered allele contains loxP sites embedded within the introns flanking a key exon. This allele is fully functional in the unrecombined state, however upon expression of Cre the interval containing the exon between the loxP sites is excised converting the allele to a non-functional mutant form. Importantly,
because the unrecombined allele is functional, homozygous mice can be analysed. Therefore, forced somatic inactivation of either one or both alleles can be achieved. Mutations in the APC tumor suppressor gene initiate a majority of human colon cancers, and mice heterozygous for Apc mutations also develop intestinal polyps along with other pathologies (discussed further below). In order to selectively inactivate Apc in the colon, Shibata et al. (1997) created a conditional allele of the Apc tumor suppressor gene by introducing loxP sites flanking exon 14. Heterozygous and homozygous mice carrying this allele failed to develop intestinal tumors as expected for a silent mutation. Deletion of Apc exon 14 was precisely induced by delivery of an adenovirus vector expressing Cre into the colon. The resulting conditional mutant mice developed polyps exclusively in the colon, validating the Cre-lox approach for tumor suppressor gene inactivation and providing a more exacting model of Apc-mediated colon cancer. NF2 tumor suppressor gene mutations cause tumors in both humans and mice, albeit with different cell-type predispositions to cancer. Human NF2 mutations lead to schwannomas whereas the conventional Nf2 knockout mouse develops primarily osteosarcomas (McClatchey et al., 1998). In order to determine the species-specific differences that result in these different tumor spectra, Giovannini et al. (2000) created a conditional allele of the mouse Nf2 gene using the CreloxP system. By expressing Cre specifically in schwann cells in an Nf2 conditional mouse, they were able to force the inactivation of both Nf2 alleles in the cell lineage affected in human patients. Amazingly, these mice recapitulate the human phenotype. These results indicate that the lack of schwannomas in the conventional Nf2 knockout resulted from a reduced rate of LOH in mouse schwann cells compared to other cell types, and not from inherent biological differences between the Nf2 pathway between mice and humans. This study offers hope for developing additional mouse models that better reflect the human disease phenotype for those tumor suppressor genes with mouse knockout phenotypes at variance with the human pathology. Oncogene
Inducible tumor models L Jackson-Grusby
5508
Chromosome engineering Karyotype analysis of many human cancers often reveals gross chromosomal abnormalities such as large deletions, inversions and translocations. Commonly observed chromosome breakpoints are usually associated with mutations that facilitate tumor growth, causing either tumor suppressor gene inactivation or activation of a novel oncogene produced by fusion of two normally disparate chromosomal segments (Rabbitts, 1994). Transgenic mice have played an important role in assessing the function of these fusion genes in cellular transformation in vivo (Adams et al., 1985; Rabbitts, 1994), however dominant lethality of these transgenes can limit the utility of standard pronuclear injection or gene targeting approaches. In addition, these strategies generally introduce only one of the two alleles caused by a conservative recombination event such as a reciprocal translocation. Consequently, the heterologous recombination systems described above have been employed in order to more faithfully reconstruct the chromosomal abnormalities found in human tumors (Ramirez-Solis et al., 1995; Smith et al., 1995). Interstitial deletions or inversions are engineered by sequentially introducing a pair of recombination target sequences (Figure 3) at disparate sites within a single chromosome by gene targeting in ES cells (RamirezSolis et al., 1995). Orientation of the recombination substrates determines whether the intervening region will be deleted or inverted. Transient expression of the recombinase initiates the chromosomal rearrangement, which occurs at a higher frequency when gene targeting has occurred on the same chromosome placing the recombination substrates in cis rather than on sister chromosomes in trans. Translocations can similarly be created, as illustrated in Figure 7, through insertion of recombination target sequences into non-homologous chromosomes to prime them for the mutational event. Recombinase expression activates the translocation event, which can be identified by positive selection in vitro using a marker gene reconstruction approach (Smith et al., 1995). However, somatic recombination induced translocations would more faithfully mimic human sporadic cancers caused by such translocations. Towards this goal, two groups have reported preliminary efforts to model human leukemias using site-specific recombination in vivo (Buchholz et al., 2000; Collins et al., 2000). Cre-mediated translocation events have been molecularly identified in mice by each group, however neither has reported mice that develop leukemias. This may be due to translocations induced in insensitive cell types, and awaits crosses to additional Cre transgenics for further characterization. Defining genetic pathways using oncomice Progression of all tumor types to malignancy is widely appreciated to result from a collection of genetic and Oncogene
Figure 7 Induced translocations caused by site-specific recombination. Using the Cre-loxP system, loxP sites are introduced into two non-homologous chromosomes. Expression of Cre can be performed in vitro to permit selection for the translocation event (Smith et al., 1995). Alternatively, somatic events can be induced by Cre expression in vivo (Buchholz et al., 2000; Collins et al., 2000)
epigenetic alterations each of which confer upon the cell a defined ability to sustain their growth and metastasize (Hanahan and Weinberg, 2000; Jones and Laird, 1999). Engineering precise mutations in the mouse genome provides an experimental handle to uncover the relative importance of each mutation in tumorigenesis. In addition, mouse cancer models have provided a successful platform for identifying mutations that either synergize with or suppress tumor formation in a predisposed strain. Genetic and genomic strategies to identify pathways that influence tumor progression in mice are outlined below. Mouse genetics Both forward and reverse genetics approaches have successfully identified collaborating genes in tumor formation. Multiple mutations can be combined by interbreeding to discern whether tumor latency, burden, or spectra are altered in compound mutant mice (Sinn et al., 1987). In the absence of a conditional knockout allele, effects of embryonic lethal mutations may be observed in heterozygotes due to haploinsufficiency (Laird et al., 1995). Alternatively, lethal mutations can be sequentially introduced into ES cells and analysed in chimeric mice (RobanusMaandag et al., 1998). Employing Cre-lox technology to avoid lethality of Brca1 mutations, Xu et al. (1999) created a mammary epithelium-specific conditional Brca1 mutation that showed evidence of genetic instability, p53 inactivation, and late onset mammary tumors. To test whether p53 was involved in tumor formation, they introduced a p53 null mutation into Brca1 conditional mice and observed accelerated tumor growth. Notably, BRCA1 mutations are the most prevalent mutations in familial breast and ovarian cancer, and many of these tumors also show
Inducible tumor models L Jackson-Grusby
5509
inactivation of P53, demonstrating the relevance of the mouse model to human cancer. Tissue-specific effects of mutations can be assessed by engineered somatic mosaicism or chimerism, which may be introduced into an animal by either viral infection or cell transplantation methods (in addition to the previously mentioned transgenic approaches). Holland et al. (1998) have described a novel method for restricting transgene expression within somatic mouse tissues by expressing TVA, the receptor for avian leukosis virus (ASLV), in transgenic mice. ASLV infection is limited to cells expressing TVA, so the celltype specificity of ASLV vector expression is determined by the pattern of TVA expression. Viral packaging cell lines expressing the oncogenic viruses are injected into the organ to initiate gene transfer. By expressing TVA in glial precursors this group demonstrated synergy between a viral delivered mutant EgfR and Ink4 deficiency in gliomagenesis (Holland et al., 1998). Chimerism achieved by bone marrow transplantation has been used by Coussens et al. (2000) to demonstrate the requirement for the matrix metalloproteinase MMP-9 specifically in inflammatory cells in a model of invasive squamous cell carcinoma. Significantly, MMP9 deficiency resulted in reduced tumor incidence in these mice, yet tumors that arose were highly malignant. This strategy is effective albeit limited to hematopoietic cell types. The observations made by Coussens et al. (2000) underscore the importance of examining effects of mutations in specific cell types both intrinsic and extrinsic to the tumor. Mosaic mice carrying combinations of conditional knockout alleles hold great promise as a general tool for uncovering paracrine signaling mechanisms, and for identifying epistatic relationships between mutations during tumor progression. This requires use of multiple recombinases to sequentially introduce individual mutations. However, the majority of the reported conditional alleles have been engineered to respond to Cre owing to its greater catalytic efficiency in vivo as compared to Flp. A Flp variant with enhanced thermostability, termed Flpe, was isolated by Buchholz et al. (1998) and shown to be roughly as effective as Cre in mammalian cells. The battery of recombination systems for use in mice can be expected to exceed Cre and Flp as preliminary reports from several groups have shown alternative recombination systems that can also function in mammalian cell culture (Diaz et al., 2001; Ringrose et al., 1997). Two forward genetic approaches, retroviral and chemical mutagenesis, have been used to elucidate genetic pathways in mouse tumor models. Insertional mutagenesis using mouse mammary tumor virus has revealed roles for Wnt, Fgf, and Notch family members in breast tumors (Gallahan and Callahan, 1997; Shackleford et al., 1993). Similarly, retroviral infection using Moloney leukemia virus has been exploited by Berns et al. (1999) to identify collaborating oncogenes in murine lymphomagenesis. By systematically mutagenizing mouse strains deficient for identified oncogenes, a process termed complemen-
tation tagging, these authors have been able to place genes into complementation groups that define signal transduction and gene regulation pathways altered in these tumors. Chemical mutagenesis performed by Dove and colleagues yielded a mouse strain predisposed to intestinal polyp formation, termed Min for multiple intestinal neoplasia (Moser et al., 1990). Identification of the causative point mutation in the Apc tumor suppressor gene (Su et al., 1992) revealed the Apc -Min mouse as an important model for Familial Adenomatous Polyposis, the corresponding cancer predisposition syndrome in humans, and also as a valuable tool for colon cancer research since nearly 80% of sporadic colon cancers carry APC mutations. In their genetic mapping efforts, the Dove group uncovered a genetic modifier of the Min mutation that reduced the burden of intestinal polyps. This Modifier of Min gene (Mom1) was genetically mapped and subsequently demonstrated to encode the phospholipase Pla2g2a (Cormier et al., 1997; MacPhee et al., 1995), an inhibitor of cyclooxygenase-2 (COX-2) in the prostaglandin synthetic pathway. Pharmacologic inhibition of COX2, through administration of non-steroidal anti-inflammatory agents, also reduces polyp formation in both Apc heterozygous mice and people (Giardiello et al., 1993; Jacoby et al., 1996). These studies are illustrative of the power of mouse genetics to uncover relevant drug targets for human cancer, and the Min/Mom-1 interaction serves as a paradigm for identifying modifier genes in tumorigenesis. Functional genomics Much of our knowledge of genes altered in tumors, whether from human or from mouse, has derived from studies examining the role of one or at most a few genes. Functional genomics tools now permit experimentalists to cast a much wider net by simultaneously monitoring large proportions of the genome for aberrant expression or sequence alterations. The Cancer Genome Anatomy Project (http://cgap.nci.nih.gov) implemented by the National Cancer Institute in 1997 is aimed at cataloging all cancer-related changes in both human and mouse tumors using EST and SAGE cDNA sequencing methods (Strausberg et al., 2000). The sequence data can be mined using a web browser to extract sequences present in tumor but not normal tissue libraries, a process termed digital differential display (Scheurle et al., 2000). Several web sites that specifically support mouse bioinformatics are shown in Table 1. In addition to direct sequence analysis, a collection of high-throughput methods have been developed and applied to cancer gene discovery (Table 2). Once again yeast genetics has set the paradigm for most of this methodology, owing to the reduced complexity of the yeast genome. Array construction for mouse experiments has also lagged behind human array development, but many arrays are now available or under development. Transcriptional profiling has been Oncogene
Inducible tumor models L Jackson-Grusby
5510 Table 1
Web resources for transgenic mouse tumor models
Tumor biology resources Cancer Genome Anatomy Project Mouse Tumor Biology Database: Carcinogenic Potency Database
http://cgap.nci.nih.gov/ http://tumor.informatics.jax.org/ http://potency.berkeley.edu/
Mouse genetics resources Mouse Genome Informatics Ensembl mouse genome server Mouse Genetic and Physical Maps: WICGR
http://www.informatics.jax.org/ http://www.ensembl.org/Mus musculus/ http://www-genome.wi.mit.edu/cgi-bin/mouse/index
Table 2 High throughput assays for detecting genetic and epigenetic abnormalities in tumors Assay
References
Comparative Genomic Hybridization Loss of Heterozygosity DNA Methylation Profiling Transcriptional Profiling Genome-wide location analysis (ChIP)
Hogdson et al., 2001; Pollack et al., 1999 Lindblad-Toh et al., 2000a,b Costello et al., 2000; Eads et al., 2000 Clark et al., 2000; Jackson-Grusby et al., 2001 Iyer et al., 2001; Ren et al., 2000
Cited references provide examples using mouse genomics tools where available
successfully used to monitor expression changes in a mouse conditional knockout for the DNA methyltransferase Dnmt1 (Jackson-Grusby et al., 2001), and has identified the RhoC gene as a mediator of metastasis in a mouse melanoma model (Clark et al., 2000). Comparative genomic hybridization has been implemented using the microarray platform (Pollack et al., 1999), and applied to a mouse islet cell tumor model to identify loci that were either amplified or exhibited LOH in tumors (Hodgson et al., 2001). SNPbased LOH screens can also employ oligonucleotide arrays (Lindblad-Toh et al., 2000a), and abnormal patterns of DNA methylation can be identified using CpG island microarrays (Costello et al., 2000) or by high-throughput PCR assays (Eads et al., 2000). Localization of DNA binding proteins by chromatin immunoprecipitation assays performed on arrays of all yeast genes and intergenic sequences has been coupled to transcriptional profiling to identify expression changes mediated directly by factor binding (Iyer et al., 2001; Ren et al., 2000). Indeed, the collection of tools for characterizing the genetic and epigenetic alterations in tumors is growing. The staggering amount of data that emerge from these studies include genes that are markers of disease progression and those that are intimately involved in tumor progression. The challenge is to identify those genes that offer greatest prognostic, epidemiologic or therapeutic value. Mutational analysis of these genes in the mouse, and examination of their role in tumor models provides one powerful tool toward that end. Lessons from mouse models on the role of DNA methylation in cancer Mouse models provide a crucial test for validating the significance of gene-specific mutations in the initiation Oncogene
and progression of tumors. Aberrant patterns of DNA methylation, such as epi-mutations caused by promoter CpG island hypermethylation, are by contrast impossible to mimic by genetic engineering. The value of mouse models in DNA methylation research has to date been the discovery of tumor-specific effects on incidence or progression that are caused by manipulating DNA methyltransferase levels and concomitantly altering DNA methylation levels in vivo. Mouse tumor models have also provided a platform for discovering genes silenced by epi-mutation through aberrant DNA methylation in tumors. Analysis of DNA methylation state in human tumor samples has demonstrated three significant alterations in DNA methylation profiles: global hypomethylation, gene-specific promoter CpG island hypermethylation, and mutational loss by cytosine deamination (Baylin and Herman, 2000; Jones and Laird, 1999; Laird and Jaenisch, 1996). The critical questions from these observations are what are the causes of the aberrant DNA methylation state, and what consequences do they confer upon the incipient tumor cell? While the mechanisms causing aberrant DNA methylation in tumors are yet to be clearly defined, several compelling models for the effects of these changes on tumor progression have been experimentally substantiated. Global hypomethylation Genome-wide hypomethylation is observed in early tumors from a range of tissue-types, implicating its role in tumor progression (Feinberg and Vogelstein, 1983; Goelz et al., 1985). As a large fraction of genomic methylation is contained within CpG rich repetitive transposon-like sequences, Bestor and colleagues have proposed that DNA methylation may serve to protect the genome from propagation of these parasitic sequences (Yoder et al., 1997). An additional conse-
Inducible tumor models L Jackson-Grusby
5511
quence of hypomethylation of these repetitive sequences is illegitimate recombination, which could explain the selection for DNA hypomethylation during tumor progression. In order to test the effect of genome-wide hypomethylation on chromosomal mutational mechanisms, Chen et al. (1998) measured mutation rates caused by gene deletion and mitotic recombination in mouse embryonic stem cells with varying levels of the maintenance DNA methyltransferase Dnmt1. A five- to 10-fold elevated mutation rate was observed in Dnmt1 mutant ES cells consistent with a role for DNA methylation in suppressing intra- and inter-chromosomal recombination. Evidence to support this model in somatic cells and in a mouse tumor model in particular are at present lacking, but offer an exciting avenue for further work. The disruptions in metabolism or gene regulation that lead to hypomethylation of repetitive sequences in human tumors are not understood, however recent evidence implicates an interaction between the maintenance and de novo methyltransferases in this process. Using ES cells deficient for either Dnmt1 (the maintenance methyltransferase) or double mutants for Dnmt3a and Dnmt3b (the de novo methyltransferases) Liang et al. (2002) have demonstrated that certain repetitive sequences are dependent on both activities for sustained methylation. Consistent with this, Rhee et al. (2002) have shown that human tumor cell lines are also dependent on a cooperative interaction between Dnmt1 and Dnmt3b for maintaining both repetitive DNA methylation and aberrant methylation of promoter CpG islands. Unraveling the details of how this genetic interaction occurs, be it through a common protein co-factor or through direct interaction, will undoubtedly provide essential clues to uncovering the deficits within the tumor cell that lead to the loss of methylation at these recombinogenic sequences. Tumor-suppressor gene hypermethylation Promoter CpG island hypermethylation is observed among a diverse set of tumor suppressor genes in both spontaneous and familial cancers (Esteller et al., 2001). Through genetic manipulation of the DNA methyltransferase genes aimed at either amplifying or ablating expression, mouse models provide an opportunity to explore the significance of promoter hypermethylation in tumor initiation and progression. Mutant mice for the key genes for establishing (Dnmt3a, Dnmt3b; Okano et al., 1999) and maintaining (Dnmt1; Lei et al., 1996; Li et al., 1992) DNA methylation state are in hand. However, each of these mutations are lethal, which has complicated their use in tumor models. The first breakthrough in this area was made by Laird et al. (1995) who demonstrated that Dnmt1 heterozygotes showed reduced polyp incidence in Apc Min mice. Tumor incidence was further abrogated by pharmacologic administration of the methyltransferase inhibitor 5-aza-CdR. Using a viable hypomorphic allele of Dnmt1, Eads et al. (2002) have recently shown that
polyp incidence in these mice can be entirely blocked by DNA hypomethylation. The significance of DNA methylation in polyp formation has therefore been convincingly established, yet the underlying mechanism through which hypomethylation is protective has not been shown. It is likely that the effect is epigenetic and is mediated by tumor suppressor gene hypermethylation. This model would therefore be an excellent test case for exploring the effects of loss of Dnmt3a or Dnmt3b function. Additional recent reports of mouse mutants will increase the repertoire of questions one can ask in exploring effects of DNA hypomethylation on tumor models. The availability of a Dnmt1 conditional knockout allele (Jackson-Grusby et al., 2001), opens up the possibility of examining the effects of DNA hypomethylation in defined tumor cell types, or in cells that support tumor development using the Cre-lox technology described above. When combined with the inducible systems, this will allow eliminating DNA methylation at defined times in tumor progression to determine the impact at varying stages of progression. Furthermore, a mouse knockout for the SWI/SNF helicase homologue lsh has been demonstrated to show reduced DNA hypomethylation, similar to the phenotype described in a homologous gene DDM in plants (Dennis et al., 2001). Test of the effects of the loss of lsh, the de novo methyltransferases, and additional chromatin components that mediate DNA methylation levels will be essential for a full appreciation of the role this set of genes plays in tumor progression. Emerging technologies Genetically engineered mice will clearly continue to provide important clues to the basic cell biology important for tumor initiation and progression. The desire to construct complex mouse models that incorporate multiple tissue specific knockout and transgenic alleles necessitates developing strategies to expedite their production and analysis. As an example, to produce a double conditional knockout regulated by a single unlinked transgene using conventional breeding yields only three doubly conditional mutants in 64 total offspring (equal to roughly eight litters from an F1 intercross). While the frequency of mutants can be improved by homozygosing the conditional alleles, each mouse generation adds 3 months to the project timeline. Fortunately new developments in both transgenic and embryonic stem cell technologies hold promise for expediting the process of making compound mutant mice. Lois et al. (2002) have recently shown that lentiviral vectors efficiently infect mouse zygotes and importantly retain expression during development and after passage through the germline. Lentiviral transgenesis does not have the same strain limitations as pronuclear injection, and alleles could therefore be directly introduced into mutant strains. This approach may be well suited for bringing several different tissueOncogene
Inducible tumor models L Jackson-Grusby
5512
specific recombinase alleles into a given homozygous conditional knockout background. The production of mice derived entirely from ES cells using tetraploid embryo complementation (Nagy et al., 1990) is another approach that may be useful for creating compound mutant mice. Eggan et al. (2002) have exploited this technology to create both male and female mice from a targeted ES cell line by screening for XO subclones from an otherwise XY cell line. Both males and females produced using this approach are viable and fertile, however greatest efficiencies are achieved using ES cell lines derived from F1 hybrid mice (Eggan et al., 2001). Hybrid mice created by this approach could be used for genome-wide SNP-based loss of heterozygosity screens (Lindblad-Toh et al., 2000a), yet to be broadly adopted this method needs to be extended to work with inbred ES cell lines. In principle there is no reason to expect that inbred ES cell lines will not eventually work as the differences between inbred and hybrid ES lines are better understood. One can envision that desired compound mutant ES cell lines could be established and used to generate many ES cell-derived mice by this method. The addition of a transgene using this approach is reduced to a simple gene targeting event in the mutant ES cell line, which ultimately saves months of breeding time and expense. Finally, recent reports using in vivo imaging methodologies suggest that the technological hurdle of monitoring gene expression and tumor growth in living mice has been surpassed. Sweeney et al. (1999) have utilized bioluminescence detection of tumor cells engineered to express luciferase in order to follow their
growth in mice and to monitor their response to chemotherapeutics. Using the same imaging strategy, Hasan et al. (2001) have created a transgenic mouse strain that expresses both luciferase and Cre recombinase under the regulation of tetracycline responsive promoter such that bioluminescence serves as a surrogate marker for Cre expression. Consequently, the kinetics of transgene induction can be determined in individual animals, which allows for testing of various drug induction regimens using limited numbers of mice. Ideally methods to image tumors noninvasively should be independent of a particular transgenic expression signal to have broadest utility. Fisher et al. (2001) have accomplished this using magnetic resonance imaging to monitor tumor growth and regression of ras-dependent lung tumors. The greatest advantage for analysing mouse tumor models using non-invasive imaging strategies may be the ability to collect multiple datapoints from single animals. As the latest breeds of oncomice are being engineered with complex sets of alleles, more effort will be needed to broaden the availability of these imaging technologies to squeeze the maximum amount of information from these valuable mouse resources. The power of the mouse geneticist’s toolbox is indeed coming to fruition.
Acknowledgments I would like to thank K Eggan for helpful discussions, and R Jaenisch for support. I apologize to those authors whose papers I did not cite owing to space limitations.
References Adams JM, Harris AW, Pinkert CA, Corcoran LM, Alexander WS, Cory S, Palmiter RD and Brinster RL. (1985). Nature, 318, 533 – 538. Baylin SB and Herman JG. (2000). Trends Genet., 16, 168 – 174. Berns A, Mikkers H, Krimpenfort P, Allen J, Scheijen B and Jonkers J. (1999). Cancer Res., 59, 1773s – 1777s. Brinster RL, Chen HY, Messing A, van Dyke T, Levine AJ and Palmiter RD. (1984). Cell, 37, 367 – 379. Buchholz F, Angrand PO and Stewart AF. (1998). Nat. Biotechnol., 16, 657 – 662. Buchholz F, Refaeli Y, Trumpp A and Bishop JM. (2000). EMBO Rep. J., 1, 133 – 139. Chen RZ, Pettersson U, Beard C, Jackson-Grusby L and Jaenisch R. (1998). Nature, 395, 89 – 93. Chin L, Tam A, Pomerantz J, Wong M, Holash J, Bardeesy N, Shen Q, O’Hagan R, Pantginis J and Zhou H et al. (1999). Nature, 400, 468 – 472. Clark EA, Golub TR, Lander ES and Hynes RO. (2000). Nature, 406, 532 – 535. Collins EC, Pannell R, Simpson EM, Forster A and Rabbitts TH. (2000). EMBO Rep. J., 1, 127 – 132. Cormier RT, Hong KH, Halberg RB, Hawkins TL, Richardson P, Mulherkar R, Dove WF and Lander ES. (1997). Nat. Genet., 17, 88 – 91.
Oncogene
Costello JF, Fruhwald MC, Smiraglia DJ, Rush LJ, Robertson GP, Gao X, Wright FA, Feramisco JD, Peltomaki P and Lang JC et al. (2000). Nat. Genet., 24, 132 – 138. Coussens LM, Tinkle CL, Hanahan D and Werb Z. (2000). Cell, 103, 481 – 490. Cronin CA, Gluba W and Scrable H. (2001). Genes Dev., 15, 1506 – 1517. Dennis K, Fan T, Geiman T, Yan Q and Muegge K. (2001). Genes Dev., 15, 2940 – 2944. Diaz V, Servert P, Prieto I, Gonzalez MA, Martinez AC, Alonso JC and Bernad A. (2001). J. Biol. Chem., 276, 16257 – 16264. Donehower LA, Harvey M, Slagle BL, McArthur MJ, Montgomery Jr CA, Butel JS and Bradley A. (1992). Nature, 356, 215 – 221. Eads CA, Lord RV, Kurumboor SK, Wickramasinghe K, Skinner ML, Long TI, Peters JH, DeMeester TR, Danenberg KD and Danenberg PV et al. (2000). Cancer Res., 60, 5021 – 5026. Eads CA, Nickel AE and Laird PW. (2002). Cancer Res., 62, 1296 – 1299. Eggan K, Akutsu H, Loring J, Jackson-Grusby L, Klemm M, Rideout III WM, Yanagimachi R and Jaenisch R. (2001). Proc. Natl. Acad. Sci. USA, 98, 6209 – 6214.
Inducible tumor models L Jackson-Grusby
5513
Eggan K, Rode A, Jentsch I, Samuel C, Hennek T, Tintrup H, Zevnik B, Erwin J, Loring J and Jackson-Grusby L et al. (2002). Nat. Biotechnol., 20, 455 – 459. Esteller M, Fraga MF, Guo M, Garcia-Foncillas J, Hedenfalk I, Godwin AK, Trojan J, Vaurs-Barriere C, Bignon YJ and Ramus S et al. (2001). Hum. Mol. Genet., 10, 3001 – 3007. Feinberg AP and Vogelstein B. (1983). Nature, 301, 89 – 92. Felsher DW and Bishop JM. (1999). Mol. Cell, 4, 199 – 207. Fisher GH, Wellen SL, Klimstra D, Lenczowski JM, Tichelaar JW, Lizak MJ, Whitsett JA, Koretsky A and Varmus HE. (2001). Genes Dev., 15, 3249 – 3262. Furth PA, St Onge L, Boger H, Gruss P, Gossen M, Kistner A, Bujard H and Hennighausen L. (1994). Proc. Natl. Acad. Sci. USA, 91, 9302 – 9306. Gallahan D and Callahan R. (1997). Oncogene, 14, 1883 – 1890. Giardiello FM, Hamilton SR, Krush AJ, Piantadosi S, Hylind LM, Celano P, Booker SV, Robinson CR and Offerhaus GJ. (1993). N. Engl. J. Med., 328, 1313 – 1316. Giovannini M, Robanus-Maandag E, van der Valk M, Niwa-Kawakita M, Abramowski V, Goutebroze L, Woodruff JM, Berns A and Thomas G. (2000). Genes Dev., 14, 1617 – 1630. Goelz SE, Vogelstein B, Hamilton SR and Feinberg AP. (1985). Science, 228, 187 – 190. Gu H, Marth JD, Orban PC, Mossmann H and Rajewsky K. (1994). Science, 265, 103 – 106. Hanahan D. (1985). Nature, 315, 115 – 122. Hanahan D and Weinberg RA. (2000). Cell, 100, 57 – 70. Hasan MT, Schonig K, Berger S, Graewe W and Bujard H. (2001). Genesis, 29, 116 – 122. Ho SN, Biggar SR, Spencer DM, Schreiber SL and Crabtree GR. (1996). Nature, 382, 822 – 826. Hodgson G, Hager JH, Volik S, Hariono S, Wernick M, Moore D, Nowak N, Albertson DG, Pinkel D and Collins C et al. (2001). Nat. Genet., 29, 459 – 464. Holland EC, Hively WP, DePinho RA and Varmus HE. (1998). Genes Dev., 12, 3675 – 3685. Indra AK, Warot X, Brocard J, Bornert JM, Xiao JH, Chambon P and Metzger D. (1999). Nucleic Acids Res., 27, 4324 – 4327. Iyer VR, Horak CE, Scafe CS, Botstein D, Snyder M and Brown PO. (2001). Nature, 409, 533 – 538. Jacks T, Fazeli A, Schmitt EM, Bronson RT, Goodell MA and Weinberg RA. (1992). Nature, 359, 295 – 300. Jackson EL, Willis N, Mercer K, Bronson RT, Crowley D, Montoya R, Jacks T and Tuveson DA. (2001). Genes Dev., 15, 3243 – 3248. Jackson-Grusby L, Beard C, Possemato R, Tudor M, Fambrough D, Csankovszki G, Dausman J, Lee P, Wilson C, Lander E and Jaenisch R. (2001). Nat. Genet., 27, 31 – 39. Jacoby RF, Marshall DJ, Newton MA, Novakovic K, Tutsch K, Cole CE, Lubet RA, Kelloff GJ, Verma A, Moser AR and Dove WF. (1996). Cancer Res., 56, 710 – 714. Johnson L, Mercer K, Greenbaum D, Bronson RT, Crowley D, Tuveson DA and Jacks T. (2001). Nature, 410, 1111 – 1116. Jones PA and Laird PW. (1999). Nat. Genet., 21, 163 – 167. Laird PW, Jackson-Grusby L, Fazeli A, Dickinson SL, Jung WE, Li E, Weinberg RA and Jaenisch R. (1995). Cell, 81, 197 – 205. Laird PW and Jaenisch R. (1996). Annu. Rev. Genet., 30, 441 – 464.
Lakso M, Sauer B, Mosinger Jr B, Lee EJ, Manning RW, Yu SH, Mulder KL and Westphal H. (1992). Proc. Natl. Acad. Sci. USA, 89, 6232 – 6236. Lei H, Oh SP, Okano M, Juttermann R, Goss KA, Jaenisch R and Li E. (1996). Development, 122, 3195 – 3205. Lewandoski M. (2001). Nat. Rev. Genet., 2, 743 – 755. Li E, Bestor TH and Jaenisch R. (1992). Cell, 69, 915 – 926. Liang G, Chan MF, Tomigahara Y, Tsai YC, Gonzales FA, Li E, Laird PW and Jones PA. (2002). Mol. Cell. Biol., 22, 480 – 491. Lindblad-Toh K, Tanenbaum DM, Daly MJ, Winchester E, Lui WO, Villapakkam A, Stanton SE, Larsson C, Hudson TJ and Johnson BE et al. (2000a). Nat. Biotechnol., 18, 1001 – 1005. Lindblad-Toh K, Winchester E, Daly MJ, Wang DG, Hirschhorn JN, Laviolette JP, Ardlie K, Reich DE, Robinson E and Sklar P et al. (2000b). Nat. Genet., 24, 381 – 386. Littlewood TD, Hancock DC, Danielian PS, Parker MG and Evan GI. (1995). Nucleic Acids Res., 23, 1686 – 1690. Logie C and Stewart AF. (1995). Proc. Natl. Acad. Sci. USA, 92, 5940 – 5944. Lois C, Hong EJ, Pease S, Brown EJ and Baltimore D. (2002). Science, 295, 868 – 872. MacPhee M, Chepenik KP, Liddell RA, Nelson KK, Siracusa LD and Buchberg AM. (1995). Cell, 81, 957 – 966. Mattioni T, Louvion JF and Picard D. (1994). Methods Cell. Biol., 43, (Pt A): 335 – 352. McClatchey AI, Saotome I, Mercer K, Crowley D, Gusella JF, Bronson RT and Jacks T. (1998). Genes Dev., 12, 1121 – 1133. McCreath KJ, Howcroft J, Campbell KH, Colman A, Schnieke AE and Kind AJ. (2000). Nature, 405, 1066 – 1069. Moser AR, Pitot HC and Dove WF. (1990). Science, 247, 322 – 324. Nagy A, Gocza E, Diaz EM, Prideaux VR, Ivanyi E, Markkula M and Rossant J. (1990). Development, 110, 815 – 821. No D, Yao TP and Evans RM. (1996). Proc. Natl. Acad. Sci. USA, 93, 3346 – 3351. O’Gorman S, Fox DT and Wahl GM. (1991). Science, 251, 1351 – 1355. Okano M, Bell DW, Haber DA and Li E. (1999). Cell, 99, 247 – 257. Pelengaris S, Littlewood T, Khan M, Elia G and Evan G. (1999). Mol. Cell., 3, 565 – 577. Pollack JR, Perou CM, Alizadeh AA, Eisen MB, Pergamenschikov A, Williams CF, Jeffrey SS, Botstein D and Brown PO. (1999). Nat. Genet., 23, 41 – 46. Rabbitts TH. (1994). Nature, 372, 143 – 149. Ramirez-Solis R, Liu P and Bradley A. (1995). Nature, 378, 720 – 724. Reichmann E, Schwarz H, Deiner EM, Leitner I, Eilers M, Berger J, Busslinger M and Beug H. (1992). Cell, 71, 1103 – 1116. Ren B, Robert F, Wyrick JJ, Aparicio O, Jennings EG, Simon I, Zeitlinger J, Schreiber J, Hannett N and Kanin E et al. (2000). Science, 290, 2306 – 2309. Rhee I, Bachman KE, Park BH, Jair KW, Yen RW, Schuebel KE, Cui H, Feinberg AP, Lengauer C and Kinzler KW et al. (2002). Nature, 416, 552 – 556. Ringrose L, Angrand PO and Stewart AF. (1997). Eur. J. Biochem., 248, 903 – 912.
Oncogene
Inducible tumor models L Jackson-Grusby
5514
Robanus-Maandag E, Dekker M, van der Valk M, Carrozza ML, Jeanny JC, Dannenberg JH, Berns A and te Riele H. (1998). Genes Dev., 12, 1599 – 1609. Sauer B and Henderson N. (1988). Proc. Natl. Acad. Sci. USA, 85, 5166 – 5170. Scheurle D, DeYoung MP, Binninger DM, Page H, Jahanzeb M and Narayanan R. (2000). Cancer Res., 60, 4037 – 4043. Shackleford GM, MacArthur CA, Kwan HC and Varmus HE. (1993). Proc. Natl. Acad. Sci. USA, 90, 740 – 744. Shibata H, Toyama K, Shioya H, Ito M, Hirota M, Hasegawa S, Matsumoto H, Takano H, Akiyama T and Toyoshima K et al. (1997). Science, 278, 120 – 123. Sinn E, Muller W, Pattengale P, Tepler I, Wallace R and Leder P. (1987). Cell, 49, 465 – 475. Smith AJ, De Sousa MA, Kwabi-Addo B, Heppell-Parton A, Impey H and Rabbitts P. (1995). Nat. Genet., 9, 376 – 385. Soriano P. (1999). Nat. Genet., 21, 70 – 71. Stewart TA, Pattengale PK and Leder P. (1984). Cell, 38, 627 – 637. Strausberg RL, Buetow KH, Emmert-Buck MR and Klausner RD. (2000). Trends Genet., 16, 103 – 106.
Oncogene
Struhl K. (1983). Nature, 305, 391 – 397. Su LK, Kinzler KW, Vogelstein B, Preisinger AC, Moser AR, Luongo C, Gould KA and Dove WF. (1992). Science, 256, 668 – 670. Sweeney TJ, Mailander V, Tucker AA, Olomu AB, Zhang W, Cao Y, Negrin RS and Contag CH. (1999). Proc. Natl. Acad. Sci. USA, 96, 12044 – 12049. Thomas KR and Capecchi MR. (1987). Cell, 51, 503 – 512. Urlinger S, Baron U, Thellmann M, Hasan MT, Bujard H and Hillen W. (2000). Proc. Natl. Acad. Sci. USA, 97, 7963 – 7968. Wang XJ, Liefer KM, Tsai S, O’Malley BW and Roop DR. (1999). Proc. Natl. Acad. Sci. USA, 96, 8483 – 8488. Wutz A, Rasmussen TP and Jaenisch R. (2002). Nat. Genet., 30, 167 – 174. Xu X, Wagner KU, Larson D, Weaver Z, Li C, Ried T, Hennighausen L, Wynshaw-Boris A and Deng CX. (1999). Nat. Genet., 22, 37 – 43. Yoder JA, Walsh CP and Bestor TH. (1997). Trends Genet., 13, 335 – 340.