Mouse models of tumourigenesis have been vital in our attempts to unravel the complex, multistage processes which confer a neoplastic phenotype. Mouse models of many of the major human cancers are available and are widely used in both basic research and clinical and therapeutic trials. Mice have many advantages as a model; they are small, easy to house, and have a short gestation time. Our increased understanding of the mouse genome over the last few years has also been a powerful force, allowing precise manipulation of the mouse genome to produce ever more sophisticated models.
It is also important to recognize the limitations of the mouse in modelling human pathologies. Mice have shorter lifespans and many differences in basic cellular processes (see Rangarajan and Weinberg 1.) The spectrum of common sporadic tumours in mouse and man is also different. Mice tend to develop sarcomas and lymphomas, derived from mesenchymal tissues, whereas humans are more likely to develop carcinomas derived from epithelial tissues, such as carcinomas of the colon, breast, lung, skin, and pancreas 1, 2. Immortalization kinetics in mice and humans are also different, due to the differences in telomeres and telomerase expression 1, 3.
Certain mouse models recapitulate human disease extremely well. For example, overexpression of c-Myc in the mouse leads to similar pathologies (B-cell lymphomas), as it does in man 4, 5. However, identical genetic lesions do sometimes produce very different pathologies in the two species. A good example of this is the retinoblastoma gene product Rb. Rb transduces anti-proliferative signals and is an important tumour suppressor 6. In humans, loss of the retinoblastoma tumour suppressor gene RB leads to the development of retinoblastoma at an early age, followed by osteosarcomas and small cell lung cancer 7. In mice, however, loss of Rb very rarely causes retinoblastoma 5, 7; Rb null mice exhibit embryonic lethality, and heterozygotes develop pituitary carcinomas and thyroid tumours at high frequency 5, 7, 8. Prima facie, then, the mouse model of retinoblastoma may appear to be of little use. However, even if the tumour spectra are different, the underlying cellular processes that lead to the human disease may well be similar. For example, it was hypothesized that there may be redundancy or compensation between Rb and family members such as p107 and p1309–11. Studies by Dyer et al in chimeric models 10, 11 found that p107 was up-regulated in the developing mouse retina and that Rb+/p107−/− null mice do develop bilateral retinoblastomas at high frequency 11. This suggests a compensatory mechanism in the mouse which does not exist in the human retina. Such experiments show the complexity of signalling pathways and provide important insights into their mechanisms of action in both the mouse and the human.
Even models which do not appear to recapitulate the human disease exactly can be valuable tools for understanding the mechanisms of tumourigenesis. Recent advances in gene manipulation such as conditional gene targeting, high throughput screening strategies, and informatics are all facilitating a new generation of mouse models which promise to recapitulate human disease more faithfully and so bring us closer to the goal of curing human neoplasias.
The first generation of mouse models of cancer
The first generation of models was created by constitutively expressing cellular and viral oncogenes such as c-Myc4, or by ‘knocking out’ tumour suppressor genes such as Rb5, 7, 12, 13. These models continue to contribute greatly to our understanding of the cellular process underlying cancer development. However, they have a number of drawbacks; first, expression of an exogenous gene, or ablation of a vital tumour suppressor gene, is often incompatible with normal development, leading to embryonic lethality or severe developmental disruption or sterility in the adult 14–16. Second, whole body expression or ablation of a gene does not mimic sporadic tumour development in vivo, where abnormal cells carrying the genetic lesion are surrounded by normal tissue. Third, with addition transgenesis there is usually little control over site of integration and copy number. In these circumstances, the exogenous gene may affect genes near the insertion site, or be affected by endogenous control elements 14–16. Fourth, spatial and temporal control of the transgene is limited, although some models do achieve good tissue specificity by the use of a tissue-specific promoter. One such successful example is a model of mouse pancreatic cancer that uses the SV40 T-antigen under control of the insulin promoter to drive pancreatic transformation 17.
Recent advances in gene targeting technology have led to models in which the expression of single or multiple genes can be tightly controlled, both spatially and temporally. Using this approach, the problem of embryonic lethality can be circumvented, as has been shown for many genes, including the adenomatosis polyposis coli gene (Apc) 18. The gene of interest can then be mutated in a specific tissue at a defined point in the life cycle, allowing the effect of removal to be precisely defined. Cell-targeted alteration of gene expression also recapitulates the in vivo development of sporadic cancers 19 and allows the influence of normal surrounding tissue to be examined.
A range of control systems have now been developed which allow the precise spatial and temporal control of gene expression, and are predominantly based on a bitransgenic approach, as described in detail below. Mice carrying a tissue-specific inducible transactivator gene are crossed to mice carrying the allele of interest which has been engineered so as to be controlled by the transactivator. Offspring that carry both transgenic elements can then be treated with an appropriate inducer to express the transactivator gene in a specific tissue, which then acts on the desired allele.
Transgenic tools. Cre-Lox and FLP: site-specific recombinases
Cre (Causes recombination) recombinase is a site-specific recombinase of the integrase family, isolated from bacteriophage P1 20–22. Cre catalyses site-specific recombination between defined 34 bp ‘Lox P’ (locus χ of crossover P1) sites. If a gene is placed between two Lox P sites and exposed to Cre, then the gene will be excised or ‘floxed’ out. A similar recombinase, FLP, isolated from S cerevisiae, also catalyses recombination from similar 34 bp FRT (FLP recombination target) sites 20, 23, 24. The Cre and FLP systems can be used to create tissue-specific conditional deletions of an allele, overcoming the problem of embryonic lethality or developmental defects 20. Mice carrying the Cre recombinase under the control of an inducible or tissue-specific promoter are crossed with mice carrying the gene of interest that has been flanked by Lox P sites 20. When the mice are given the appropriate inducer, Cre is expressed in a spatially defined manner and the gene of interest is ‘floxed out’ in a specific tissue (Figure 1A). The Cre-Lox and FLP systems can also be used to activate a gene. In this instance, a STOP cassette flanked by Lox P or Frt sites is placed before the gene of interest, and expression of the Cre results in the cassette being excised, allowing expression of the gene 20 (Figure 1B).
Because Cre recombinase is exogenous to the mouse 25, 26, it was thought that expression would have no effects other than to target the specific Lox P sites. However, recent reports show that Cre appears to act, albeit with low affinity, with ‘pseudo-Lox P sites’ in the mouse genome 26 and its expression in mammalian cells can have deleterious effects on the stability of the mouse genome, including chromosome rearrangements in mouse spermatids 25. Although the implications of this are as yet unknown, this has in part stimulated the development of self-excising Cre vectors 27.
Several forms of Cre delivery have been developed; thus, Cre may be ligand-induced, such as through creation of the fusion protein CreER™ transgene, which is tamoxifen-dependent 20, 28. Cre may also be delivered packaged in adenovirus 20, 29, or may be placed under the control of a promoter which has a well-defined spatio-temporal expression pattern, such as the use of the WAP30 and BLG promotors to deliver Cre expression to the mouse mammary gland 31. Finally, Cre activity may be controlled through the use of an inducible promoter; for example, through use of the Cyp1A promoter 32. Despite the apparent versatility of these systems, the one caveat that remains is their irreversibility. Thus, if a model requires a gene to be switched on and off, then a more suitable alternative may be constructed with, for example, a tetracycline-responsive system.
The tetracycline (tet)-dependent system developed by Gossen and Bujard 33, 34 allows tight spatial and temporal control through the use of a tissue-specific transactivator and an effector gene (Figure 2). Unlike the Cre and FLP systems, the tet system allows genes to be turned on and off at will, via the administration of the inducer (tetracycline, or more usually doxycycline) The system can also be tuned to either switch on gene expression (the ‘tet-on’ or tTA system) or switch it off (the tet-off or rtTA system 35).
The tet-on (tTA) system uses the Tn10-specific tetracycline resistance operon of E coli (Figure 2) and is composed of two parts; a transactivator and an effector 35. The transactivator is composed of the DNA binding domain of the E coli tetR gene fused to the transactivation domain of the VP16 herpes simplex virion protein 16 gene, under the control of a tissue-specific promoter which drives expression of the tTA (tetR/VP16) fusion protein in the desired tissue. The second component is a construct containing the gene of interest driven by the minimal promoter of human cytomegalovirus (Pcmv), under the control of the tet operator, tetO. In the absence of doxycycline, the tTA protein binds tetO and activates the minimal promoter, driving expression of the gene. However, when doxycycline is added, it binds to the tTA protein and causes a conformational change which prevents it from binding tetO. The promoter is not activated and the gene is not expressed 35 (Figure 2A).
In the tet-on system (rtTa), the second component of the binary system is identical to the tet-off system, but the transactivator component is made up of a tissue-specific promoter and a mutant version of the tetR DNA binding domain fused to the VP16 transactivator 35. This produces a mutant rtTA protein: in the absence of doxycycline, rtTA does not bind to the tetO operator, and there is no gene expression. When doxycycline is added, the protein undergoes a conformational change and can bind the operator, leading to expression of the gene of interest. If absolute repression of a gene is required, the transactivator can be constructed with a silencer with the KRAB-AB silencing domain of the Kid1 gene inserted instead of the VP16 domain. This prevents leaky transcription caused by the basal level of affinity of tTA for tetO in the absence of doxycycline. The tet-on, tet-off, and silencer systems can be used together to control more than one gene at a time.
Other conditionally inducible systems are under development, although they are not as progressed as the tet and Cre-Lox systems. These include systems based on the insect steroid moulting hormone ecdysone 36, the progesterone analogue mifepristone 37, the Lac operator-repressor system 38, and the GAL4/UAS system 39.
Modelling six of the hallmarks of cancer
In their 2000 paper 40, Hanahan and Weinberg proposed that cancer cells can be seen as having six characteristics, or ‘hallmarks:’ self-sufficiency in growth signals; non-responsiveness to anti-proliferative signals; resistance to apoptosis; unlimited replicative capacity; angiogenesis; and the ability to metastasize and invade. Aspects of each of these different facets have been modelled in the mouse, providing valuable insights into the relevance of these mechanisms. The remainder of this review will therefore focus on a number of examples relating to each of these ‘hallmarks’. We will also briefly discuss the modelling of other factors which assist or trigger carcinogenesis, such as genomic instability or alterations of the epigenetic imprint of the cell.
Self-sufficiency in growth signals
Oncogenes and tumour suppressor genes (TSGs) are involved in regulation of the cell cycle. The core of the cell cycle machinery consists of a family of cyclin-dependent kinases (CDKs), which drive the cell through the cell cycle by phosphorylating key effector substrates 41. The CDKs are primarily positively regulated by the cyclins and negatively regulated by the members of the Ink4 family and by p21kip and p19Arf. In turn, these primary regulators are themselves regulated by many other molecules in response to signals from inside (eg DNA damage surveillance) and outside (eg growth hormones) the cell. The tight control of cell growth and division is vital to a multicellular organism and mutations in these key cycle regulators can lead to uncontrolled division—one of the hallmarks of the cancer cell.
Mouse models of cell cycle regulators are allowing us to unravel the complex web of control that dictates when and how fast cells divide. For example, PTEN (phosphotin and tensin homologue deleted on chromosome 10) is a phosphatase which antagonizes the PI3K pathway 42. PTEN is a tumour suppressor gene and is mutated in a variety of human sporadic tumours. Constitutive Pten homozygous knockouts show early embryonic lethality; Pten heterozygotes develop a range of tumours of the intestine, prostate, mammary, thyroid, endometrial, and adrenal tissues 42. Furthermore, conditional knockouts using Cre-Lox technology have revealed Pten loss as an early event in prostate tumourigenesis 43, 44. Homozygous conditional deletion of Pten in the prostate leads to hyperplasia by 4 weeks post-floxing, prostate intraepithelial neoplasia by 6 weeks, invasive adenocarcinomas by as early as 9 weeks, and metastasis as early as 12 weeks. The conditional approach also shows heterozygous Pten deletion to result in much slower disease progression, with failure to progress to invasive carcinoma. Even in such an excellent model, however, caveats remain; metastatic lesions in mice were found mainly in the lung, whereas human prostate cancer metastases tend to develop in bone. Nonetheless, the prostate Pten model could be used to test the effects of other genes on the progression of the disease, or for therapeutic testing. Indeed, co-operativity has already been demonstrated for the cell cycle regulators Ink4aArf and Pten45. The Ink4a/Arf locus encodes two TSGs: p16Ink4a, which regulates the Rb pathway, and p19Arf, which regulates p53. The compound mutant Ink4a/Arf+/−Pten+/− mouse shows shortened tumour latency and Ink4a/Arf−/−Pten+/− mice shorter latency still (although, intriguingly, progression to invasive carcinoma is not seen in this model.)
Murine models are also illustrating how incompletely we understand cell cycle control. Thus, although the cell cycle regulators Cdk4 and Cdk6 have been considered critical initiators of the cell cycle, Cdk4 has been shown to be dispensable for proliferation in many cell types 46, 47. Furthermore, Cdk6 knockout mice are viable with only minor defects 48, and no synergism has been observed in mice doubly null for Cdk2 and Cdk6. By contrast, deficiency of both Cdk4 and Cdk6 does lead to embryonic lethality, although this is due to severe anaemia, rather than a general cell division defect 46. MEFs (murine embryonic fibroblasts) from Cdk4−/−Cdk6−/− mice were capable of division in culture (albeit more slowly than wild-type cells) and underwent immortalization after repeated passaging. Mutant cells also became quiescent in response to serum withdrawal but re-entered the cycle normally with the appropriate stimuli. Cells also had lowered Rb phosphorylation—a serious challenge to the accepted theory that full Rb phosphorylation by both Cdk2 and Cdk4/6 is required for G1/S transition 48. Taken together, the data from these mice therefore challenge our understanding of the role of the D-type cyclins and suggest that alternative mechanisms are capable of initiating proliferation.
If a cell has activated a gene allowing self-sufficiency in growth factor signals, deactivating that gene may lead to tumour regression. Marinkovic et al have used a mouse model in which c-Myc is conditionally expressed under the control of a lymphoid-specific promoter using the tet-off system 49. Treatment with doxycycline led to regression of the tumours, implying that the requirement for c-Myc activation is continuous 46. A reversible model of skin neoplasia has been created by Pelengaris et al50 using a modified version of Myc (MycER™) targeted to skin using the involucrin promoter. Upon administration of the inducer 4-hydroxytamoxifen (4-OHT), Myc is turned on and induces proliferation in the normally post-mitotic cells of the suprabasal layer. This leads to hyperplastic lesions and papillomatous epidermal lesions. These lesions have not been seen to progress to malignancy, as the differentiation pathway is thought to override the influence of Myc in these cells. Deactivation of the Myc transgene leads to rapid regression of the lesions and their associated vasculature. Even transient deactivation of Myc is sufficient to exclude cells permanently from the cell cycle.
Jain et al have used a tetracycline-controlled ‘tet-off’ system to transiently inactivate Myc in mice with transplanted osteogenic sarcoma cells or transgenic tumours 51. When Myc was inactivated, the sarcomas and osteogenic tumours underwent differentiation to mature bone and showed significant regression. Cultured tumour cells in which Myc was subsequently reactivated did not resume proliferation but instead underwent apoptosis. In all of these tumour models, it will be interesting to see if the requirement for c-Myc continues even after other enabling mutations have occurred in the cell. If so, targeting c-Myc could be an extremely productive therapeutic strategy for some tumour types.
The product of the Ras oncogene is involved in a wide range of cellular processes, including progression through the cell cycle, transcription and translation, cell survival, and apoptosis, via interactions with a wide range of effectors 52. Several mouse models with defects in Ras have been developed, both conditional and conventional. K-Ras−/− mice die at E12–E14 from liver defects and anaemia 53, 54, while N-Ras−/− and H-Ras−/− mice are viable, with the former having defects in immune and T-cell function and the latter having no obvious phenotype except for decreased carcinogen-induced tumourigenesis. Conditional Ras alleles are a relatively recent development. Johnson et al have created a mouse model which carries a latent oncogenic allele of K-Ras, which can undergo spontaneous activation in vivo55, 56, 57. One hundred per cent of animals carrying this allele developed multiple lung tumours, with a high proportion also developing thymic lymphomas and skin papillomas. The lung tumours appear to be similar to human non small-cell lung cancer (NSCLC) progressing through hyperplastic and dysplastic stages before progressing to carcinoma. The spontaneous and stochastic nature of the activation events makes this mouse an excellent model of spontaneous K-Ras-induced lung cancer in humans, and also may mimic the interaction of adjacent mutant and normal tissues seen in human cancer in vivo.
K-Ras is widely mutated in human tumours, although this varies tremendously with tumour type, so cellular context may be highly important 52. A recent paper demonstrates this, using a K-Ras allele activated by the Cre-Lox-mediated removal of a STOP cassette to drive K-ras expression 58. Guerra et al also engineered a bicistronic colour marker into their system, and thus show that expression of the endogenous K-Ras allele appears to have no discernible consequences in most cell and tissue types 58. A low incidence of sarcomas and anal papillomas was seen in these mice; however, 100% developed multiple lung lesions, which derived from type I pneumocytes, although a small proportion were derived from Clara cells. This replicates the K-ras-induced lung tumours found by Johnson et al57, Jackson et al59, and Meuwissen et al60.
If Ras oncogenes are expressed under the control of a highly active promoter, then cells enter G1arrest and senescence due to oncogenic stress 61. However, the models of Johnson et al57 and Guerra et al58 show that under endogenous control, K-RasV12 does not induce senescence, but instead causes cells to acquire hyperproliferative capacity akin to that of immortalized cells. This shows how important levels of expression are in determining the exact reaction of a system.
Oncogenic Ras has also been shown to be important for tumour maintenance. Wong and Chin have created a mouse melanoma model using doxycycline-induced H-RasV12G on an Ink4a null background 62. Induction and maintenance of H-RasV12G were shown to be absolutely required for induction and maintenance of melanomas. When doxycycline was removed and H-Ras V12G was down-regulated, extensive apoptosis and regression of tumours were observed.
Lack of response to anti-proliferative signals
Rb is a major transducer of anti-proliferative signals 63, and mutations in RB1 in humans produce the rare childhood malignancy retinoblastoma 63. Efforts to model this disease in mice were hampered by the fact that Rb knockout mice do not develop retinoblastoma at high frequencies 7. The first true phenotypic mouse model of retinoblastoma was created in 2004 by Zhang, Schweers, and Dyer 63, and contains six engineered alleles: Rb1, p53 and p107 null alleles; floxed p53 and Rb alleles; and a Chx 10-Cre which targets retinal progenitors. Retinal progenitor cells lacking Rb and p107 keep proliferating past the time when they should undergo terminal cell-cycle exit, showing that p107 compensates for lack of Rb by preventing deregulated proliferation. Tumours which lacked Rb, p107, and p53 were highly aggressive and metastatic; hence p53-mediated apoptosis may also be a barrier to aggressive retinoblastoma formation in the mouse. This model will no doubt be extremely use for preclinical research and is being refined still further—a key difference between the murine and human diseases is that the number of Rb inactivation events in the human retina is much lower, leading to a small number of focal, clonal tumours. In an attempt to mimic this in the mouse, a retrovirally delivered Cre is being used to target small numbers of cells and increase the accuracy of the model 63.
Endogenous and exogenous DNA damage are considered key initiators of the apoptotic response 40, and many TSGs are involved in either the repair or sensing of DNA damage. It is now widely assumed that cells with damaged DNA will attempt repair, but if this is not possible, they may default to an apoptotic pathway. Avoidance of this pathway is a frequent feature of neoplastic cells, which may progress through the cell cycle regardless of damage, resulting in mutation or karyotypic alterations 40. One key mediator of the apoptotic response is p53, which integrates cellular stress signals and triggers apoptosis, repair or senescence.
Constitutive defects in human p53 are associated with the Li–Fraumeni syndrome, which predisposes to a wide range of tumours 64 and the first mouse models of p53 defects were null alleles which developed a variety of neoplasias 65–69. Although Li–Fraumeni patients mainly develop epithelial tumours 70, p53 null mice predominantly develop lymphomas 71. The short lifespan of p53 null mice due to early-onset lymphoma has largely precluded the study of tumours with longer latencies. It was also soon realized that most spontaneous and familial human mutations of p53 are missense mutations in the DNA binding domain (DBD) 72 and so the conventional null alleles were not a faithful model of p53 mutation in vivo. Missense p53 mutations associated with cancer are generally gain-of-function mutations, which have been ascribed dominant negative effects 73.
Several new mouse models of p53 function are therefore being developed which better mimic the real (and diverse) range of human mutations. For example, Liu et al have produced a p53 DBD missense mutation that carries a common mutation (R172H) and also a splice variant that reduces expression levels to almost wild type 74. Clear differences were found between the p53+/− and p53R172H gain of function mouse, with the latter developing fewer lymphomas and more carcinomas, with a higher degree of metastasis in the tumours observed.
Another engineered version of the p53 protein shows the sheer complexity of the effects each of the mutations in p53 may have in vivo. Lozano's group have recently created a mouse carrying a version of p53 which cannot induce apoptosis but can still induce cell cycle arrest 75. Mice homozygous for this mutation (p53515c/515c) have significantly delayed tumour onset when compared with p53 null mice. Tumours which arose in the p53515c/515c mice showed a longer latency and had more stable genomes than those arising in p53 null mice. Cells from p53 null mice had extensive aneuploidy and chromosome breaks and abnormalities, whereas cells from p53515c/515c mice were generally 2N, 4N or 8N with many fewer abnormalities. This suggests that the p53515c mutation is capable of suppressing genomic instability, and also argues strongly that abrogation of the apoptotic programme does predispose to malignancy.
It is not just specific mutations in p53 that need to be explored, but also its basic functions. It had been questioned whether loss of heterozygosity (LOH) of the remaining wild-type p53 allele was an important factor in tumourigenesis in Li–Fraumeni patients 70. In p53+/− models 76, around 50% of tumours showed LOH, but in the p53R172H model, only 9% of tumours exhibited LOH, implying that LOH is not absolutely necessary for tumour formation (although it may still represent a significant route to tumourigenesis in a subset of tumours). Another important mechanistic question answered by new p53 models is whether the transcriptional transactivation activity of p53 is essential for p53-mediated apoptosis and tumour suppression 69. A mouse model has been developed with missense mutations in the transactivation domain by Jimenez et al(p53L25Q/W26S) 77. The mutant p53 protein produced still has a functional DBD. Mice carrying this mutation have lowered p53-mediated apoptosis in thymocytes in vivo and their cultured embryonic fibroblasts become neoplastic. This strongly argues that transcriptional transactivation is indeed essential for p53's function as a tumour suppressor gene.
Although these studies therefore support differing roles for p53 in tumour suppression (which includes a role in initiating apoptosis), the precise relevance of p53-mediated apoptosis to tumour suppression remains somewhat unclear. The original observation that p53 mediates the apoptotic response to DNA damage gave rise to the simple hypothesis that p53 mediates its primary tumour suppressive activities through initiation of the apoptotic response, and that failure to delete cells harbouring DNA damage would result in increased mutation and thereby to accelerated tumourigenesis. This hypothesis has, however, been somewhat difficult to prove, with clear tissue-specific differences in the role of p53-mediated apoptosis. For example, it appears that the loss of p53-dependent apoptosis is a good predictor of mutation burden and tumour predisposition within the haematological system but not the intestine 78.
Despite these complexities, conventional models and null alleles still have a central role in p53 research. For example, combinatorial use of alleles null for p53 and p19Arf and ectopically expressed Mdm2 have been used to explore the in vivo interactions between these genes 79. Increased Arf levels are known to inhibit Mdm2-mediated proteolytic degradation of p53 80, 81, and Arf expression can lead to increased oncogenic signalling, such as overexpression of c-Myc or activation of Ras 82, 83. Various combinations of the three alleles have been used to investigate co-operativity between the three gene products 78. Thus, p19Arf deficiency accelerated tumour development in the Mdm2-overexpressing mouse, whilst the principal TSG activity of Arf appears to be p53-dependent.
As a solid tumour grows, its size exceeds its blood supply and the inner part of the tumour may become hypoxic 84, 85. To maintain its rate of growth, the tumour must establish its own blood supply to ensure a constant supply of oxygen and to remove catabolites. Angiogenesis is difficult to model in vivo; many models require tissue to be transplanted into the mouse, which causes injury, one of the normal triggers of apoptosis 84.
Although Ras and Myc are implicated in angiogenesis, the process is more complex than may have first been appreciated. Both oncogenic K-Ras and H-Ras can stimulate VEGF expression 62. In the mouse model of melanoma developed by Wong and Chin and described earlier 62, decreased expression of H-Ras resulted in vascular regression before the main tumour regression. High rates of apoptosis were seen in cells lining tumour-associated vessels, indicating that continued activation of Ras may be needed for a stable tumour vasculature. Wong and Chin also showed that loss of Ras-stimulated VEGF expression is not responsible for tumour regression and that high VEGF levels are not sufficient to maintain the tumour vasculature if H-Ras is absent. There must therefore be another mechanism by which Ras promotes angiogenesis.
Modulation of angiogenesis can also occur through deregulation of thrombospondin 1 (TSP1), an inhibitor of angiogenesis. Thus, expression of TSP1 can be inhibited by many of the oncogenes associated with malignant progression, namely oncogenic Ras, c-Myc, v-Src, c-Jun and Id185. Furthermore, expression of TSP1 has recently been shown to be under the control of TSGs such as p53 and Pten86.
Although the mechanisms which control angiogenesis are still incompletely understood, evidence is growing for the existence of an ‘angiogenic switch’, whereby the normal balance of inhibitors and promoters of angiogenesis is disrupted, allowing the tumour to form a new vasculature 84, 85.
If a cell acquires mutations that allow it to supply its own growth signals and ignore anti-proliferative and pro-apoptotic signals, then in theory, that cell should be capable of unlimited growth and division 40. In reality, somatic cells have a limited number of divisions ‘programmed’ into them, with the absolute number of divisions varying with tissue and cell type 87. Cultured cells will proliferate for a number of generations and then enter senescence. This state can be overcome by various mutations, including those in p53 and Rb, leading to further growth and division until cells enter a ‘crisis’ phase. This is characterized by karyotypic abnormalities such as chromosome fusions and a high proportion of cell death 88. In a population of such cells, clones will emerge which are immortalized and capable of replicating indefinitely. The mechanisms surrounding immortalization appear to be largely dependent on maintaining telomeres 89. Telomeres stabilize chromosomes by preventing recombination and fusion events, and also by preventing the cell from recognizing the end of the chromosome as a double-strand break 90. With every cell division, the ends of chromosomes progressively ‘fray off’, counting off the cell generations until they reach a certain non-permissive limiting length which triggers senescence or apoptosis 91–93. In order to overcome this, a cell may either acquire activating mutations in the telomerase gene, which acts to maintain telomeres and is strongly repressed in normal cells, or activate a mechanism that maintains telomeres through recombination 89.
Mice have very long telomeres (40–60 kb as opposed to 10 kb in human) and also have a wider pattern of expression of telomerase than in the human 94. For a human cell to escape replicative senescence, it must acquire mutations in the telomerase maintenance system, but this restraint apparently does not apply to mouse cells 94. Mice engineered to have shorter telomeres have a higher rate of spontaneous tumourigenesis 94. One possible reason why mice and humans have different tumour spectra was hinted at when these short-telomered mice were crossed onto a late-generational p53 null allele. The resultant mice developed a high rate of epithelial-derived carcinomas, similar to the tumour spectrum in humans 95.
Mice have also been generated which provide insights into the relationship between telomeres, premature ageing, and cancer 96. Chang et al generated mice null for both the telomerase RNA component Terc, and Wrn. Inactivation of WRN (a RecQ helicase family member) causes Werner syndrome, which presents with premature ageing, genomic instability, and an increased incidence of cancer 97, 98. Cells with WRN deficiency have an increased rate of telomere loss and undergo premature senescence, which can be rescued by forced expression of Tert 99, 100. Doubly null Wrn−/−Terc−/− mice experience premature ageing, cataracts, and hair loss, as well as increased chromosomal instability and incidence of non-epithelial cancers such as osteosarcomas and soft tissue sarcomas 96. Interestingly, the effects of the Wrn deficiency on Terc null mice increased through subsequent generations, with the first and second generations of Terc−/− mice being unaffected by Wrn status. In contrast, by the fourth to sixth generations, Terc−/−Wrn−/− mice had lower body weights and a shorter life expectancy than Terc−/−Wrn+/− mice 96. Later generations of Terc−/−Wrn−/− mice also showed age-related degeneration in mesenchymal tissues, unlike Terc−/− mice, which show phenotypes in the highly proliferative compartments of skin, blood, and intestine. The lower impact of Wrn deficiency in epithelial tissues could be due to higher telomerase activity or functional redundancy in the RecQ helicase family such as the Bloom helicase.
The nature of the checkpoints which monitor telomere length are still not understood in detail, and no doubt new mouse models such as the Wrn/Terc null mouse will be vital in unravelling the mechanisms by which cells acquire unlimited replicative capacity. Furthermore, despite the clear importance of telomeres, they are not solely responsible for inducing senescence, as evidenced by the fact that some mouse cell cultures can undergo senescence with telomeres of 50 kb or more remaining 101.
One gene recognized to induce cellular senescence is p53 64. In this light, it is intriguing that mice with a deletion of the first six exons of p53, which encodes a truncated RNA forming a carboxy-terminal p53 fragment, have been shown to have an altered lifespan 102. This mutation confers some properties of activated p53, and the mice display an early ageing phenotype including reduced lifespan, osteoporosis, weight loss, lordokyphosis (hunched spine), and muscle loss. Notably, none of the mice developed tumours, compared with around 45% of wild types. These data argue for a role for p53 in regulating senescence in vivo and also suggest that there may be a ‘pay-off’ between increased lifespan and tumour susceptibility.
Invasion and metastasis
In order to metastasize, a tumour must break down surrounding tissue such as a basement membrane to escape its tissue of origin; it must then travel within the body and re-establish itself at a new site, supplying its own growth factors 40. It must be resistant to anoikis (apoptosis induced by lack of correct positional information) and establish a new vasculature at the distant site. Metastases are the cause of an estimated 90% of cancer deaths 40 and thus the study of metastatic progression is of vital importance.
Several types of adhesion molecule are altered in many cancer cells, such as the calcium-dependent cadherins, which mediate cell–cell interaction, and the integrins, which mediate cell–extracellular matrix (ECM) interactions 103.E-cadherin mediates cell–cell adhesions in epithelial tissues; the intracellular domain interacts with β-catenin to transduce anti-proliferative signals such as those relayed via the Lef/Tcf system. Loss of E-cadherin expression is associated with de-differentiation and invasion in a variety of human cancers 104, 105—the forced expression of E-cadherin in a transgenic model of pancreatic β-cell carcinogenesis has already been shown to prevent the transition from adenoma to carcinoma, underlining the importance of correct adhesion and positional signals in invasion and metastasis 106.
Integrins consist of a non-covalently linked α and β subunit (there are 18 known α and eight known β units) which spans the membrane and transduces information about the cell's environment into the cell 107. Different combinations of α and β subunits have different affinities for ECM components. The expression profiles of some integrins are known to change during tumourigenesis, allowing the cell to alter its migration and adhesion; for example, α6β4 and α3β5 are known to be up-regulated during tumourigenesis, whereas reduced expression of α1, α6, β1 or β4 is associated with breast epithelial neoplasms 107. The α3β5 integrin binds a wide range of ECM components and is considered an attractive target for therapeutic intervention, as it is expressed at low levels in resting endothelial cells and at higher levels on vascular cells in tumours 107, 108.
Integrins are also thought to play a key role in directing tissue remodelling via their ability to activate MMP precursors. α3β5 interacts with MMP2, a key player in tissue remodelling. MMP2 knockout mice show reduced angiogenesis and tumourigenesis 109 and inhibitors of the MMP2α–α3β5 interaction are potent suppressors of angiogenesis, gliomas, and melanomas 110, 111. However, knockout mice lacking β3 or β3 and β5 subunits, generated by Reynolds et al, show enhanced tumour growth and enhanced angiogenesis within the tumours that form 108. This shows that the α3β5 and αvβ5 integrins, which were thought to be vital for angiogenesis, are in fact not essential. The apparent discrepancy between the effects of antagonists and knockouts may be due to compensation, or the antagonists being non-specific 108. However, it seems more likely that we simply do not yet have a complete enough understanding of the complex interactions between cell and stroma that are vital in driving invasion and metastasis.
Genomic instability, a helping hand on the road to cancer
Although not one of the six hallmarks identified by Hanahan and Weinberg, genomic instability is increasingly being identified as a major contributor to many cancer types. Genomic instability is seen in many cancer cells, either as an initiating event or as a later stage. It is estimated that up to 30% of genes in the genome code for proteins that regulate genomic fidelity 112 and many inherited cancer syndromes are due to mutations in these genes [eg Li–Fraumeni (p53, Chk2113; OMIM #151 623), Bloom syndrome (BLM; OMIM #210 900), Nijmegen breakage syndrome (NBS1; OMIM #251 260), ataxia telangiectasia (ATR/ATM; OMIM #208 900), and HNPCC (OMIM #114 500; the MMR system)]. Genomic instability can have many causes: hypomethylation, inefficient MMR (mismatch repair), increased mitotic recombination, chromosomal translocations or defects in genes which code for proteins that monitor and repair lesions, or monitor genomic fidelity through the cell cycle 40.
Chromosomal translocations frequently give rise to fusion proteins which may participate in the cell cycle and drive clonal expansion of the tumour. For example, translocations at human 11q23 can fuse the 5′ end of the MLL (myeloid lymphoid leukaemia) gene to a range of potential genes, giving rise to a diverse number of leukaemia aetiologies 114. Forster et al have developed a highly elegant model which accurately mimics the translocation that fuses the MLL and ENL genes to give rise to myeloid leukaemia and mixed myeloid/lymphoid leukaemia 114. Forster et al placed Lox P sites at defined breakpoints on separate chromosomes. Cre recombinase was then expressed under the control of the haematopoietic promoter Lmo2, leading to inter-chromosomal reciprocal translocations and the rapid onset of myeloid tumours with high penetrance. This accurately recapitulates the randomly occurring spontaneous translocations and clonal expansion of the human leukaemia and is the first model to generate both translocation products.
Genomic instability can also be caused by hypomethylation, as recently shown by Gaudet et al115. The Dnmt1 protein maintains methylation in somatic cells and is required for embryonic development—Dnmt1 null mice die during gestation 116, 117, so Gaudet et al created a heterozygous Dnmt1 model with one Dnmt1 null allele and one Dnmt1 hypomorphic allele (Dnmt1chip/−) These mice are estimated to have only 10% Dnmt1 activity and have substantially reduced levels of genomic methylation. The mice developed aggressive T-cell lymphomas at around 4–8 months of age. Array CGH (comparative genome hybridization) showed an increased rate of chromosome gains, most notably gain of chromosome 15 and the c-Myc gene, which it carries. This shows that methylation levels can directly contribute to tumourigenesis via genomic instability.
The role of methylation in tumourigenesis is, however, complex. Thus, loss of DNMT1 can also be shown to have a tumour protective effect within the intestine, presumably as a consequence of blocking transcriptional repression of TSGs 118. Similarly, Sansom et al119 used a mouse model (ApcMin) of the human intestinal cancer syndrome FAP (familial adenomatous polyposis coli) to show that constitutive loss of the Mbd2 gene (a member of the methyl binding domain family) in the intestine results in greatly reduced tumour burden. Clearly, the role of methylation patterns in cancer development is complex, and mouse models of altered epigenetic states will no doubt provide insights into these divergent mechanisms.
The few thousand mutant mouse strains in existence represent only about 10% of the total genes of the mouse genome 120. Although we have almost complete sequence information available for the mouse, we still do not know the biological functions of most of these genes 120. Functional annotation of the mouse genome will necessarily be slower than sequencing, given the volume of work needed to characterize mutant phenotypes. However, several concerted programmes are now underway 121–123 to systematically mutate each gene in the mouse genome using high throughput approaches such as ENU (ethylnitrosourea) mutation. γENU is a potent alkylating agent which causes point mutations at high frequency in many tissues 120. When male mice are injected with ENU, point mutations occur in the pre-meiotic spermatogonial stem cells 120. The F1 progeny can then be screened for dominant phenotypic abnormalities (behavioural or biochemical) or bred for a further two generations to expose recessive phenotypes. The ENU approach has a number of advantages: point mutations can often produce both null alleles and hypomorphic alleles, which are more informative than null alleles alone; novel functions in both known and unknown genes can be exposed; and the whole approach is ‘phenotype-driven’, removing any bias towards known or expected gene functions. A large-scale ENU mutagenesis programme (especially one screening for dominant and recessive mutations) is, however, a massive logistical undertaking, requiring co-ordination of animal resources, personnel, and informatics. Several such projects are now underway and will no doubt provide a vast number of novel mutant lines, many of which will aid in the study of tumourigenesis.
Conclusions and future directions
The current generation of mouse models is extremely sophisticated, yielding insights into the fundamental processes underlying normal cell physiology and cancer. Future mouse models will no doubt provide more precise control over gene expression at various stages of development. As we understand more about the expression profiles of genes through embryogenesis, it is also becoming possible to activate mutant alleles at precise points in the developmental programme, for example through the use of inducible Cre-Lox technology 124, 125. Inevitably, some problems remain with these technologies. For example, certain Cre-expressing lines are ‘leaky’ and drive recombination in tissues other than the target tissue. This issue will need to be carefully addressed, although it could itself be used as a useful tool. Ultimately, it is possible to envisage the development of a modular ‘toolbox’ containing a range of defined transgenes and systems which will permit controlled gene expression of many different alleles of a given gene throughout all stages of the life cycle of an organism. With novel reporter systems and imaging technologies 126, it should also be possible to follow the initiation and spread of tumours in vivo and to integrate all these data with a bio-informatic database. Indeed, the integration of such informatics will be key in allowing the full potential of the mouse (and human) genomes to be exploited. Finally, it should be stressed that the primary purpose of these systems is the modelling of human disease, and by implication the resolution of human disease. The need to develop cohesive, productive links between the basic murine studies described here and translational and clinical research cannot be overstated.