A New Generation of Mouse Models of Cancer for Translational Research

Although the spontaneous occurrence of carcinoma in mice is rare, mice have become an attractive model for studying human cancer because of the ease of manipulating their genome. Although the earliest mouse models were derived from viral insertions, nowadays, when we refer to mouse models of cancer, we are typically referring to mice that have been genetically modified to express oncogenes or to delete tumor suppressor genes, either of which can lead to cancer phenotypes. Generally speaking, these genetically modified mouse models enable the investigation of cancer phenotypes in the context of the tumor microenvironment and an intact immune system. These are critical distinctions from the widely used xenograft models in which tumor cells or tissues are typically grown in immunodeficient mice, which, as argued by Olive and Tuveson (1), Fomchenko and Holland (2), and Carver and Pandolfi (3), limit their “predictive use” for drug development.

The original genetically engineered mouse models of cancer were engineered to express dominantly acting oncogenes, such as the SV40 T antigen, in particular compartments using tissue-specific promoters (e.g., refs. 11–13). Such mouse models were enabled by the availability of tissue-specific promoters driving oncogene expression. Although they are limited by the need for such promoters as well as practical issues about variability of transgene integration site, expression levels, and cell types of expression, these transgenic models established the feasibility of studying cancer phenotypes in genetically engineered mice and continue to provide a valuable resource for preclinical studies.

Complementing these gain-of-function transgenic models are loss-of-function models in which targeted gene deletion of tumor suppressor genes results in cancer phenotypes. Indeed, since the landmark studies showing that loss of function of p53 could lead to development of cancer in mice (14, 15), many subsequent studies have shown that the individual or combinatorial loss of function of tumor suppressor (or other) genes in mice can lead to cancer. Moreover, as a consequence of a veritable explosion in the technological advances of gene targeting in recent years (16), mouse models based on targeted alterations of the genome have become increasingly more common and significantly more sophisticated. Nowadays, the conditional as well as inducible deletion of tumor suppressor genes or activation of oncogenes in tissue-specific compartments have become the norm (see discussions in refs. 1–3). This has led to the emergence of a new generation of mouse models that are highly relevant for drug development and translational studies.

Nonetheless, despite these technological advances, there remain significant differences between cancer development in humans and in genetically engineered mouse models (6). One important distinction is that, in many genetically engineered mice, the genomic alteration(s) (i.e., the loss or gain of gene expression) typically occurs in the germ line or at least in a large portion of the cells, which contrasts with most instances of human cancer in which gene alterations are typically rare and stochastic. Therefore, as discussed by Fomchenko and Holland (2), most genetically engineered mouse models are more representative of human cancer predisposition syndromes rather than random tumorigenesis. They and others have used a somatic gene transfer approach (the RCAS/tv-A system) to circumvent this potential limitation (2). Another notable exception is the spontaneous activation of Kras following somatic recombination in the whole animal (17).

Does this difference in cancer development of most genetically engineered mouse models limit their use in studying human cancer? If one's intent is to study the evolution of the disease as it occurs in humans, then it very well might. However, if one's intent is to investigate the molecular mechanisms underlying a cancer phenotype based on a particular genetic alteration or to facilitate biomarker analyses, then perhaps not. Indeed, analyses of gene expression changes in relatively “homogeneous” mouse models can help to identify biomarkers that may have otherwise been obscured in more “heterogeneous” human cancer specimens. For example, comparison of gene expression profiles from a mouse model of lung cancer model based on activation of Kras2 and human tumors enabled the identification of a Ras signature that was not identifiable by analyzing only the human tumor data (18).

Moreover, the relative homogeneity of cancer development in mouse models, relative to humans, can be of considerable benefit for drug testing. As discussed in the reviews by Degenhardt and White (4), Fomchenko and Holland (2), and Carver and Pandolfi (3), mouse models based on the perturbation of specific molecular pathways can enable a detailed interrogation of the involvement of such pathways in tumorigenesis as well as investigations of potential therapeutics that specifically target these pathways. Generally, cancer develops more uniformly and more rapidly in mouse models, which from an industry perspective can be highly advantageous for drug testing (5). Moreover, mouse models provide ready access to cancer tissue at all disease stages; this can greatly facilitate pharmacokinetic and pharmacodynamic studies (1, 5), thereby providing valuable information about whether a given agent is effective in the target tissue. Finally, the ability to monitor the consequences of drug delivery by in vivo imaging of mouse models (1, 2) can enable a quantitative assessment of the consequences of drug action for tumor development.

However, not all of the unique features of mouse models are advantageous either for interrogating human cancer or pursuing translational studies. Perhaps the most significant hindrance of mouse models is that relatively few develop metastases and even those that do tend to display metastases with different tissue specificity than seen in human cancer. This is of particular concern because most cancer lethality is a consequence of metastasis, and the availability of mouse models that facilitate investigation of therapeutics targeting metastatic disease would be highly advantageous.

Can this be overcome? It is certainly plausible that the genesis of metastases in mice and humans is so inherently different as to preclude their occurrence in most mouse models. However, a new generation of mouse models able to closely recapitulate the metastatic features of human cancers seems to be in evolution as exemplified by recently described mouse models of pancreatic cancer (19, 20).

We are approaching a new era in modeling cancer in mice. Although our models are still imperfect, they have become considerably more accurate in their ability to recapitulate human cancer and, thereby, have become increasingly more valuable and more widely used for preclinical and translational research. The critical next step will be to determine whether mouse models have significant predictive use for drug development and whether such use can be directly translated to human therapeutic applications. Considering recent successes in this arena (21), the outlook is very promising!

I thank all of my colleagues in the Mouse Models of Human Cancer Consortium, whose thoughtful discussions over the past 7 years provided the foundation for this review, and Michael Shen, Arnold Rabson, and Bob DiPaola for helpful comments on the manuscript.