Modeling cancer patient populations in mice

Abstract

Genetic differences among individuals contribute to differential susceptibility to cancer and, undoubtedly, to variable efficacy and toxicity of pharmacological-based therapeutics. Many of the specific molecular processes involved in human tumorigenesis have been elucidated and accurately modeled in mice. However, the current models used for drug testing do not accurately predict how new treatments will fare in clinical trials. More sophisticated models that treat cancer as a complex disease present within heterogenous patient populations will provide better predictive power to identify patients that may benefit from specific therapies or that may develop potential drug-induced toxicities.

INTRODUCTION

It is increasingly clear that pharmaceutical-based cancer therapies provide variable and often unpredictable patterns of efficacy among patients. Additionally, unpredicted or off-target toxicities within patient populations are a growing concern that has been highlighted in the lay press over the last several years. While animal models, especially mice, have been used extensively by the pharmaceutical industry for mechanistic studies and safety testing, the commonly used approaches and mouse models are not appropriate to address many of the complex genetic and environmental contributions underlying varying therapeutic effects and/or toxicities in patient populations. Herein we compare current models and suggest new approaches to more accurately model heterogenous patient populations.

IN VITRO MODEL SYSTEMS

Cell-Based Platforms

Cell-based screens are an efficient and cost effective platform to identify potential therapeutic targets, to investigate mechanisms of action, and to assess cellular toxicity in early stages of drug discovery. While these screening approaches may be effective in identifying potential therapeutic lead compounds, their predictive value for the clinic is limited. Readily available cancer cell lines have been maintained in culture for many generations of passage and usually have acquired many unknown genetic alterations. Uncharacterized mutations may affect the validity of conclusions and long-term reproducibility. It is also not uncommon for cell lines to be cross contaminated with other cell lines after many years of repeated culturing. The German Cell Bank (DSMZ) has estimated that 29% of all human-tumor cell line submissions contain cross contamination [1]. While certain drugs may have positive results in cell line screens many other physiological factors may affect the overall performance of lead compounds including the surrounding microenvironment of cancers in vivo and the underlying genetic heterogeneity of the patient populations that may affect efficacy, metabolism, as well as susceptibility to toxicity. Strategies using panels of cell lines have been developed to overcome some of these limitations, such as the National Cancer Institute’s panel of 60 cancer cell lines (http://dtp.nci.nih.gov) that have been used to test thousands of chemicals. Another strategy is to combine in vitro cell-based platforms with in vivo growth assays like xenograft models.

Xenografts: An In Vivo Petri Dish

The current gold standard for testing the efficacy of cancer pharmaceutical therapeutics in vivo is the xenograft mouse model. In this model human tumors are subcutaneously implanted into immunodeficient mice and drugs are administered either locally or systemically. Following treatment, tumors are assessed for changes in size, vasculature, metastasis to distant sites, among other measures. Although xenografts may be efficient in detecting therapy-induced changes to specific cancer cell growth characteristics, the predictive power of this model has been modest at best. Previous studies have shown that only 3.8% of patients in Phase I cancer drug trials show a significant clinical response [2]. Most, if not all, drugs that failed in humans worked on subcutaneous xenograft tumors in mice. It is apparent that there is a large discrepancy in the relationship between artificially-implanted xenograft models and in vivo arising cancers.

While these models may be effective for determining whether pharmaceutical drugs hit their appropriate targets and cause changes in specific cancer cell characteristics, they have significant shortcomings when compared to spontaneously occurring tumors in humans. One of the main issues is the lack of an intact immune system in xenograft recipients, which is a well-established component of cancers that can have both growth promoting and growth inhibiting properties. Another issue is that these implanted tumors are grown in a subcutaneous environment lacking a host stromal compartment that evolved with the cancer. Differences in the tumor microenvironment compared with the normal tissue of origin may have a profound impact on growth and survival of cancer cells as this artificial environment has different cellular compositions and vasculature processes. Orthotopic implantation, injecting the human tumors into the corresponding mouse organ tissue, may resolve some of these issues but this route of implantation is not practical due to extensive procedure time and the inability to accurately mimic the location and growth environment of the human tumors from which the cells were derived.

Both of the immunodeficient mice commonly used for xenografts, severe combined immunodeficient mice (SCID) and Nude mice, are usually maintained on a single or mixed genetic background which eliminates the ability to detect host specific genetic modifiers that may contribute to varying pharmacological and toxicological effects in humans. Finally, xenograft models platforms rarely incorporate important environmental factors, such as diet, in assessing the efficacy and toxicity of pharmaceuticals. With the current knowledge on the role of diet in cancer, drug-food interactions, and differential effects of diet across mouse strains and humans, the human diet, which is substantially different than normal mouse chow, likely contributes to reduced predictive power of mouse models as described below.

The simplistic xenograft model system is generally regarded as not being predictive for therapeutic efficacy and has lead to significant criticisms of the value of mouse models for drug discovery and testing [3], despite the fact that xenografts are not mouse models of human cancer but are primarily a complex in vivo cell culture system. These models may even be responsible for prematurely canceling development of candidate drugs that may have efficacy in humans, but data is not available for drug leads not pursued based upon failure in xenograft studies. With the vast majority of cancer pharmaceutical companies using these models, despite the low success of subsequently developed drugs in human clinical trials, it is apparent that more accurate mouse cancer models must be utilized to specifically model the complexity of tumorigenesis as it occurs in humans.

IN VIVO MODEL SYSTEMS

Advanced Molecular Mouse Models of Cancer

Transgenic, knock-out, and knock-in mouse models with spontaneously arising tumors have been developed that demonstrate many of the molecular and histopathological changes associated with human cancer progression from initiation through metastasis (see the National Cancer Institute’s Mouse Models of Human Cancer Consortium (MMHCC) web site: http://emice.nci.nih.gov). Although these models have been used extensively in basic research investigations into the mechanisms of tumorigenesis, their use in pharmaceutical preclinical testing has been limited. With the shortcomings of the xenograft models in cancer drug development becoming more widely acknowledged, it is appropriate to entertain the use of more complex but realistic mouse models of human cancer for preclinical testing. Unlike xenograft models, engineered mouse models retain an intact immune system and a host-derived tumor microenvironment that accurately models that present in human cancers.

The first transgenic mouse models of human cancer were generated in the 1980s when MYC overexpression in the mouse mammary epithelium led to development of adenocarcinoma [4] and mutant KRAS expression led to the oncogenic transformation of pancreatic cells [5]. These models demonstrated that expression of human oncogenes in the mouse could successfully initiate tumor development. Subsequently it was shown that knock-outs of tumor suppressors such as Trp53 or Rb would induce a spectrum of tumors including lymphomas, sarcomas and pituitary adenomas [6-8]. Unfortunately, the concept of engineering mice to develop cancer and their use for preclinical trials was patented. DuPont licensed the patented ‘OncoMouse’ in 1988, preventing the production and exchange of these models without a license. Many have argued that the cost and restrictions imposed by these licenses have hindered the adoption of more accurate mouse models by the pharmaceutical industry for preclinical therapeutic development and delayed development of more intelligently tested drugs, which has increased the cost of drug development despite the lower cost of the xenograft models compared to more clinically-accurate engineered and spontaneously arising mouse cancer models [9,10].

Since the first generation mouse models of human cancer were developed, numerous and ever more accurate models have become available for most cancers including brain, breast, colon, lung, and prostate cancers (Table 1). Models have been designed and built to display varying stages of tumorigenesis ranging from initiation through metastatic disease making them useful for evaluating therapies against particular stages of cancer progression. Cancer is a disease with multiple mutations and epigenetic alterations usually involving activation of oncogenes (Myc, Kras, etc.) and loss of tumor suppressor genes (Trp53, Rb, etc.). While single gene transgenic and knock-out models are sufficient to induce tumorigenic events in the mice, more complex models that recapitulate the multiple genetic events occurring in human cancers have been developed. Illustrative of the remarkable similarity between the latest generation of mouse models and human cancer is the development of a mouse pancreatic cancer model that progresses from initiation to invasive, metastatic disease that was engineered by introducing an activated Kras gene with pancreas-specific expression along with loss of the tumor suppressor Cdkn2a [11]. These two mutations occur in almost all cases of human pancreatic adenocarcinomas, yet mice expressing Kras only display signs of early stage pancreatic tumor growth while mice lacking Cdkn2a have no neoplastic lesions in the pancreas. This mouse model should have specific utility in preclinical testing for pancreatic cancer drug development since it recapitulates both the molecular and histopathological changes observed in human pancreatic cancer. Greater use of genetically engineered immunocompetent mice to model the order and spectrum of mutations observed in humans can only improve the predictive value of mouse models for the discovery of cancer therapeutics and will lead to lower drug development costs because of more informed evaluations in realistic models before the expense of clinical trials in incurred.

Table 1. Mouse models of human cancer.

Cancer Site	Mouse Model	Reference
Brain medulloblastoma astrocytoma glioblastoma	Ptc+/-;p53, GFAP-Cre;Rbloxp/loxp GFAP-v-src, GFAP-HRas NPcis	[30] [30] [31]
Breast low-grade mammary intraepithelial neoplasia high-grade mammary intraepithelial neoplasia papillary carcinoma human ductal carcinoma in situ (DCIS)	MMTV-LTR/int3, MT/HGF C(3)1/SV40 tag, WAP/TGFα MMTV-LTR/cyclin D1, MMTV-PyV-mt MMTV-c-erb-B2	[32] [32] [32] [32]
Colon adenoma adenocarcinoma mucinous carcinoma	Apcmin/+, ApcΔ716, Apc1638N/+ Mlh1-/-;Apc1638N/+, Msh6-/-;Apc1638N/+, Msh3-/-;Apc1638N/+ Tgfb-/-;Rag2-/-	[33] [33] [33]
Lung bronchiolar neuroendocrine cell hyperplasia adenocarcinoma non-small-cell lung carcinoma	Rb+/-;p53-/- Tg(Sp-C-LTag), Tg(CC10-Tag) K-rasLSL;p53, CC10-hASH1	[34] [34] [34,35]
Prostate low-grade prostatic intraepithelial neoplasia (LGPIN) high-grade intraepithelial neoplasia (HGPIN) invasive carcinoma metastasis	PB-Ras, Nkx3.1-/- ARR2PB-Skp2 LPB-SV40 Tag (Lady model) Pb-Cre4;PTENloxp/loxp, LPB-Tag;ARR2PB-hepsin, TRAMP model	[36] [36] [36] [36]

Genetic Modifiers

Pharmacology studies have traditionally used mice from inbred or undefined genetic backgrounds for drug discovery and toxicity testing. However, studies based solely on results obtained using one inbred strain or model do not support clinically-important insights or the full appreciation of the response spectrum likely to be encountered in the clinic. This insight can only be achieved by investigating the genetic context of efficacy and toxicity. Different inbred mouse strains have varying susceptibility to spontaneous or chemically-induced tumor formation [12,13], and specific modifier loci have been identified in mice that lead to susceptibility to specific cancers [14,15] and varying toxicological responses [16]. For example, the Pas1 locus has been identified to predispose mice to lung tumor development [17]. Depending on genetic background, the activity of the same allele may even switch from resistance to susceptibility to cancer [18].

These differences demonstrate the importance of genetic background in cancer development and use of this information may allow better understanding of how humans differ in their response to pharmaceutical therapies and/or susceptibility to drug-induced toxicities. A ‘one-size-fits all’ mouse model approach will not be sufficient to provide the best preclinical predictive power.

Genetic background can dramatically modify the effect of engineered mutations on tumor growth and survival and by extension, on the efficacy and toxicity of preclinical therapeutic trials. Patients developing cancers with specific genetic mutations often have different survival rates and respond to therapy with varying degrees of success, presumably as a result of unidentified genetic modifiers. Although requiring substantial changes from the status quo, incorporating knowledge of genetic modifiers into preclinical tests will improve extrapolation of results to patients and identification of patient populations best matched to particular therapies, leading to more accurate and cost-effective translation to the clinic.

Humans are not inbred nor genetically homogeneous so it is over-simplistic to assume that a single inbred mouse model will be sufficiently representative of the genetic heterogeneity present in patient populations. By identifying and understanding modifier loci in mice that alter therapeutic response or susceptibility to toxicity, polymorphisms in orthologous genetic loci in patients enrolled in clinical trials can be considered to increase the probability of successful translation to the clinic. Inevitably, future cancer therapies will be personalized to the patient’s spectrum of tumor mutations and their constitutional genetic makeup.

Diet and Nutrition

In addition to genetic predisposition, environmental factors like diet can have a major impact on cancer risk and response to therapy. It is becoming increasingly apparent that knowledge about gene-environment interactions will be important for improving cancer prevention and treatment. According to the American Cancer Society an estimated one-third of all cancer deaths in the U.S. are related to lifestyle, such as diet and physical activity (http://www.cancer.org). The interaction between environment and genetic susceptibility and their influence on cancer deaths also varies among cancer type; environment has the greatest impact on cervical, lung and oesphageal cancers while leukemia and colorectal cancers are more heavily influenced by genetic susceptibility in combination with environmental factors [19]. The role of diet in particular has been examined extensively in animal studies and human epidemiological and clinical trials.

The ‘Western-style’ diet consumed in North America and Europe that has high levels of fat and low levels of vitamin D and calcium has been modeled in mouse experiments. Maintenance of mice on this diet induces hyperproliferation in pancreatic, prostate, mammary and intestinal epithelial cells of wild-type C57BL/6 mice [20,21]. Further modification of the diet to include decreased levels of folate and other nutrients essential for DNA methylation induces intestinal adenomas and carcinomas in wild-type mice [22]. Combining a Western diet with inactivation of Cdkn1b, coding for a key cell cycle regulater. results in an additive effect on intestinal tumorigenesis compared to Western diet or Cdkn1b mutations alone [23]. Analogous effects on therapeutic response or susceptibility to drug-induced toxicity are predicted to occur despite the dearth of studies in this area.

Another factor not usually considered is the more than fifty percent of adults taking at least one type of dietary supplement. Recent studies into the molecular mechanisms underlying the effects of specific nutrients are being revealed and found to intersect with many drug-target pathways known to be involved in cancer [24-26]. In addition, individuals vary in their ability to metabolize specific nutrients, and this difference has been correlated with cancer risk [27-29].

With the development and use of more targeted therapies, and with the increasing knowledge of how diet impacts the same pathways, consideration should be given to how diet and nutrients interact with cancer therapies. Results from clinical trials have revealed that efficacy and toxicity of chemotherapeutics is individualized with few preclinical studies being predictive for these effects. Better designed preclinical studies utilizing more than one mouse model, genetic background, and base diet have the power to account for these variable circumstances, further improving the translational potential of mouse models by improving the predictive power and lowering the total cost of drug development.

IMPROVING MODEL TRANSLATION TO HUMANS

Mouse models have been used extensively to determine the underlying causes of many human diseases long before the genetic similarities between humans and mice were appreciated. Their small size, relative affordability, short gestational times, and similarity to human organ systems have made the mouse a convenient model to study human disease. With the completion of the mouse and human genome sequence, it is now known that 40% of the human and mouse genomes can be directly aligned and greater than 80% of human genes have a direct orthologous gene in the mouse genome [26]. This data suggests that the mouse can serve not only as a physiological model for preclinical drug testing but can also be utilized as a model for studying the underlying genetic variation among individuals that modulates therapeutic response.

It is evident that traditional methods of pharmaceutical research are not adequate for painting the full picture of how a population of patients will respond to new therapeutics (Table 2). The use of homogenous ‘in vitro’ mouse models has proven to be economically inefficient when the total cost of drug development from target identification to clinical trial is considered. This is evident in the fact that a large number of drugs reaching clinical trials never achieve approval by the Food and Drug Administration (FDA). As the identification of potential new pharmaceuticals increases, it is essential that parallel improvements be made in the use of preclinical models that accurately recapitulate the natural development and progression of human cancers. Similarly, models that incorporate host genetic variability and environmental conditions observed in clinical trials will improve the efficiencies and reduce the costs associated with drug development.

Table 2. Comparison of models for preclinical testing.

	In vitro	In vivo
		xenograft	engineered	population
Pros	cost and time efficiency, high throughput	cost and time efficiency (relative to other in vivo systems)	mirrors human histopathological progression, intact immune system, host tumor microenvironment retained
				incorporates genetic modifiers
Cons	contamination, genetic selection, lack of physiological relevance	lack immune system, tumor microenvironment different from host, inability to detect genetic modifiers	patents, higher direct cost, long timeline, genotyping, lower throughput
Best Use of Model	high throughput screening for initial target identification	validation of drug access to target	pharmacological and toxicological preclinical testing
Access to Model	ATCC, Cambrex/Clonetics	MMHCC, Jackson Laboratories, Harlan, Taconic Charles River Laboratories

CONCLUSIONS

The sequencing of the human and model organism genomes is beginning to have a beneficial impact on the progress of drug discovery. An individual’s genetic background and environment has the potential to significantly alter their response to therapy or susceptibility to toxicity. As the genome era continues to transform our understanding of the mechanisms behind human disease and response, the field of cancer therapeutics must incorporate new insights based upon models that accurately recapitulate the genetics and environmental circumstances of human cancers if the success of drug development is to improve.