Pharmacology & Therapeutics 142 (2014) 351–361

Contents lists available at ScienceDirect

Pharmacology & Therapeutics journal homepage: www.elsevier.com/locate/pharmthera

Associate editor: B. Teicher

Fit-for purpose use of mouse models to improve predictivity of cancer therapeutics evaluation K. Wartha ⁎, F. Herting, M. Hasmann Discovery Oncology, Pharmaceutical Research and Early Development (pRED), Roche Diagnostics GmbH, Penzberg, Germany

a r t i c l e

i n f o

Available online 8 January 2014 Keywords: Tumor models Mouse Cancer therapeutics Predictivity Preclinical Drug evaluation

a b s t r a c t Preclinical animal models are useful tools to better understand tumor initiation and progression and to predict the activity of an anticancer agent in the clinic. Ideally, these models should recapitulate the biological characteristics of the tumor and of the related tumor microenvironment (e.g. vasculature, immune cells) in patients. Even if several examples of translational success have been reported it is a matter of fact that clinical trials in oncology often fail to meet their primary endpoints despite encouraging preclinical data. For this reason, there is an increasing need of improved and more predictive models. This review aims to give an overview on existing mouse models for preclinical evaluation of cancer therapeutics and their applicability. Different types of mouse models commonly used for the evaluation of cancer therapeutics are described and considerations for a “fit-for purpose” application of these models for the evaluation of different cancer therapeutics dependent on their mode of action are outlined. Furthermore, considerations for study design and data interpretation to translatability of findings into the clinics are given. Conclusion: Detailed knowledge of the molecular/biological properties of the respective model, diligent experimental setup, and awareness of its limitations are indispensable prerequisites for the successful translational use of animal models. © 2014 Elsevier Inc. All rights reserved.

Contents 1. Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 2. Overview and characteristics of available preclinical models . . . . . . . . . . . . . . . . . . 3. Preclinical model selection for the evaluation of cancer therapeutics with a specific mode of action 4. Translational implications of preclinical study design . . . . . . . . . . . . . . . . . . . . . 5. Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6. Conflict of interest statement . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . References . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

. . . . . . . .

352 351 352 352 355 352 358 352 360 353 360 353 360 353 360

Abbreviations: SCID, severe combined immunodeficiency; NOD, non-obese diabetes; MHC, major histocompatibility complex; NK cell, natural killer cell; Th1 or 2, t helper cell type 1 or 2; AOM, azoxymethane; DSS, dextran sodium sulfate; MNNG, N-methyl-N-nitro-N-nitrosoguanidine; GEMM, genetically engineered mouse model; ADCC, antibody dependent cellular cytotoxicity; EGFR, epidermal growth factor receptor; HER, human epidermal growth factor receptor; PI3K, phosphoinositide 3-kinase; MDM2, mouse double minute 2 homolog; PK, pharmacokinetic; FAP, fibroblast associated protein; HGF, hepatocyte growth factor; ALL, acute lymphatic leukemia; AML, acute myeloid leukemia; CML, chronic myeloid leukemia; CLL, chronic lymphocytic leukemia; CTLA-4, cytotoxic T lymphocyte antigen-4; PD-1, programmed cell death protein-1; PD-L1, programmed cell death ligand 1; RECIST, response evaluation criteria in solid tumors; LD, longest diameter; CR, complete response; PR, partial response; PD, progressive disease; SD, stable disease; PFS, progression-free survival; DFS, disease-free survival; OS, overall survival; pCR, pathological complete response; TGI, tumor growth inhibition; NSCLC, non-small-cell lung carcinoma; SEM, standard error of the mean; Cy5, cyanin-5. ⁎ Corresponding author at: Nonnenwald 2, D-82377 Penzberg, Germany. Tel.: +49 8856 606564; fax: +49 8856 60796564. E-mail address: [email protected] (K. Wartha). 0163-7258/$ – see front matter © 2014 Elsevier Inc. All rights reserved. http://dx.doi.org/10.1016/j.pharmthera.2014.01.001

352

K. Wartha et al. / Pharmacology & Therapeutics 142 (2014) 351–361

1. Introduction Cancer is a major health problem worldwide and one of the leading causes of death accounting for 7.6 million deaths (around 13% of all deaths) per year. Moreover, worldwide mortality rates from cancer are projected to continue rising, with an estimated 13.1 million deaths in 2030 (Ferlay et al., 2013). Therefore novel and superior approaches for early diagnosis and treatment of cancer are highly warranted and in the past decades large amounts of public and industry resources have been invested into identification of novel approaches to fight cancer. Despite the huge effort, clinical success rate from first-in-man to registration for novel cancer therapeutics is at a discouraging rate of 5% (Kola & Landis, 2004). Therefore while significant progress in therapeutic options has been made for selected cancer types such as breast cancer and Non-Hodgkin's Lymphoma (NHL), efficacious treatment options remain poor for a large number of patients with other types of cancer such as for example pancreatic and lung cancer. Consequently, the 5-year relative survival rates have improved for female breast cancer patients from 63% in the early 1960s to 90% today, and for NHL from 47% in 1975 to 70% (ASCO, 2012). In contrast, 5-year survival rates remain low at 16% and 6% for lung and pancreatic cancer patients, respectively (ASCO, 2012). As virtually all novel cancer therapeutics undergo extensive preclinical evaluation prior to entry into clinical trials the overwhelming clinical attrition rate of 95% indicates a strong discrepancy between preclinical efficacy and clinical response. This review has the aim to give an overview on existing mouse models for preclinical evaluation of cancer therapeutics and their applicability. It is clear that all mouse models have a “model” character with several intrinsic limitations. However, we believe that a “fit for purpose use” of preclinical cancer models, selecting models closely representing the tissue in focus and the drug target, combined with suitable study design and interpretation of results may be a way to improve predictivity of preclinical cancer therapeutics evaluation. We first describe different types of mouse models commonly used for the evaluation of cancer therapeutics, followed by a section outlining the application of these models for the evaluation of different cancer therapeutics dependent on their mode of action. Considerations for study design and data interpretation to translatability of findings into the clinics are given in the final section. 2. Overview and characteristics of available preclinical models

the interleukin-2 receptor of NOD/SCID mice were required to allow efficient establishment of human immune cells. Detailed knowledge of the respective immunodeficient mouse strain regarding the presence and functionality of immune cells is mandatory to select the most appropriate variant for engraftment studies or experiments where an immune response is required. The selection of the most suitable mouse strain is also important for the use of syngeneic tumor models since immunocompetent strains can differ in their predominant immune response type, C57BL/6 and BALB/c mice being prototypical Th1- and Th2-type mouse strains (Watanabe et al., 2004). 2.2. Xenograft models Xenograft models, usually established by implanting human tumor cell lines subcutaneously into the flank of immunodeficient mice or rats, have long been the standard model for preclinical evaluation of novel cancer therapeutics. In comparison to other in vivo models, xenograft models are easy to handle, produce results relatively quickly and offer high throughput, low variability and good reproducibility. A large set of well characterized xenograft models, in terms of mutations, signaling pathway activity as well as drug sensitivity/resistance, are available. As the cell lines used for the generation of xenograft models are derived from human tumors the effect of novel therapeutics on a human tumor can be relatively easily studied in an in vivo setting. However, it is arguable how closely a xenograft model represents a patient tumor situation: Most of the cell lines that are routinely used as xenografts have undergone a high number of in vitro passages and thus differ strongly from the original tumor due to long-term in vitro selection. The cell line origin also implicates a very homogeneous tumor cell population that does not reflect the heterogeneity of human tumors. Furthermore, the fast, subcutaneous growth of xenografts often leads to only limited build-up of tumor stroma and therefore tumor–stroma interactions are difficult to study in standard xenograft models. The subcutaneous location also does not reflect the organ environment of the tumor that may be necessary e.g. for the provision of tissue-specific growth factors. Last but not least, the use of immunodeficient host animals makes it difficult if not impossible to study certain drugs whose mode of action depends on immune effector functions. Nevertheless, xenograft models are useful for studying targeted drugs or chemotherapeutics that act by direct interaction with the tumor cell. Especially for mechanistic studies where a mode of action hypothesis is investigated or for combination studies that need large numbers of study groups and precise readouts xenograft models are a useful tool.

2.1. Mouse strains commonly used for preclinical tumor models 2.3. Patient-derived models The discovery of the nude mutation in 1966 on chromosome 11 of athymic mice lacking T-lymphocytes opened the door for the engraftment of human tissues, hematopoietic stem cells or peripheral blood mononuclear cells. The animals' deficient immune system composed mainly of B-cells lacking T-cell support enabled the study of human biological processes in vivo. Subsequently, various genetic modifications were described that led to a higher grade of immunodeficiency, like in SCID (severe combined immunodeficiency) or NOD (non-obese diabetes)/SCID mice. Mice carrying the recessive SCID mutation on chromosome 16 are virtually devoid of T- and B-lymphocyte function despite having a thymus, lymph nodes, splenic follicles, and normal numbers of NK-cells. NOD/SCID mice, however, have an additional mutation causing Beta-2-Microglobulin deficiency. They lack mature lymphocytes, serum immunoglobulin, MHC class 1 expression, and NK-cell activity. Beige mice are deficient in cytotoxic T-cells, NK-cells, and macrophages. Therefore, the combined immunodeficient SCID/beige mouse variant probably has the weakest immune competence that could be turned against engrafted cells or tissue. Nevertheless, for a high success rate in the creation of humanized mice, additional targeted mutations at

An alternative to classical xenograft models are patient-derived models. Instead of the transplantation of an established cell line, patient-derived tumor materials (cell suspension or tumor fragments) are transplanted onto immunodeficient animals and then passaged directly from mouse to mouse in vivo. For recent reviews on patientderived models see Lum et al. (2012) and Tentler et al. (2012). An advantage of patient-derived models is their direct origin from human tumors without any previous in vitro culture and clonal selection. Thereby the genetic background and heterogeneity of human tumor tissue is preserved in a better way and may retain several characteristics more closely reflecting the patient situation. Another plus of patient-derived models compared to classical cell line based xenograft models is the higher amount of stroma that is initially present within the tumor and may support its growth and tissue homeostasis. During the first passages the stroma is of human origin but is replaced by mouse stroma after several passages (Reyal et al., 2012). Compared to classical xenografts, patient-derived models are labor intensive (due to in vivo passaging of tumors) and also come with an increased variability and longer timelines. They are less well characterized

K. Wartha et al. / Pharmacology & Therapeutics 142 (2014) 351–361

and difficult to standardize/reproduce, as the model may change its characteristics with increasing number of in vivo passages. As for cell line derived xenograft models one major disadvantage is the use of immunodeficient hosts, not allowing the investigation of immune effects. One key application of patient-derived models currently lies in studies with translational objectives. To identify/confirm a potential clinical indication, responsiveness of a panel of patient-derived models from a certain tumor type is evaluated with the clinical candidate drug. Information on sensitivity/responsiveness in a representative panel of patient-derived models can then be used for hypothesis generation in respect to potential response prediction markers. Even further goes the recent advancement in applying banks of patient-derived models as avatars, thus using contemporaneously forming patientderived models for drug sensitivity testing (Malaney et al., 2013). While this approach is mainly used to select the most effective medication for treatment of a particular tumor before administering it to the respective patient, the added benefit of using avatars in anticancer drug development remains to be demonstrated by further investigations. A potential complication is the sometimes low take rate of patientderived material. Dependent on the tumor indication, engraftment rates vary between 23% and 75% (Siolas & Hannon, 2013). As a consequence, patient-derived models in some cases do not represent entire tumor indications but rather a subset thereof. This should be kept in mind when interpreting data generated in panels of patient-derived models representing a specific human tumor indication, especially when drawing conclusions in respect to response rates and biomarker hypothesis generation. 2.4. Orthotopic models Another approach to make tumor models more “patient-like” is the use of orthotopic models. For this type of tumor model, cell lines or patient material is transplanted orthotopically, i.e. at the respective tissue site of tumor origin. Thus for example cells of a pancreatic carcinoma cell line or a pancreatic tumor fragment is transplanted into the pancreatic tissue of the host animal. For a review on the state-ofthe-art of different orthotopic models the reader is referred to several recently published articles (Ottewell et al., 2006; Tseng et al., 2007; Chan et al., 2009; Sano & Myers, 2009). This organ specific transplantation ensures a more natural environment (growth factors, stromal cells etc.) for the transplanted tumor thereby reflecting the tumor microenvironment more closely. Furthermore the organ localization of the tumor may allow investigation of angiogenesis, tissue invasion and metastatic spread (Loi et al., 2011). The use of orthotopic models is technically more challenging than subcutaneous transplantation, and highly skilled personnel is required to ensure appropriate quality and reproducibility of orthotopic tumor models. As orthotopic tumors oftentimes are not palpable and measureable from the outside, monitoring of tumors requires surgical intervention or imaging approaches that are labor and cost intensive. 2.5. Metastasis models Various in vivo metastasis models exist that are able to mimic different steps of the complex metastasis process of tumor cells which are: invasive growth, intravasation, survival and dissemination in the blood stream, extravasation and outgrowth at a distant site. Basically, one can distinguish between the direct implantation into an organ to monitor the invasive growth, the spontaneous metastasis assay where tumor cells spread from a primary tumor site and the so-called experimental metastasis or colonization assay where tumor cells are injected into different vessel types (peripheral or portal vein; carotid artery) or into an organ (intrasplenic; intracardiac). The choice of a model mainly depends on the scientific question one intends to answer since each of the approaches covers more or less different stages of the metastatic processes and bears its specific pros and cons. For example, orthotopically

353

growing brain or pancreatic tumors (Sennino et al., 2012) are suitable to investigate the invasive growth. Recent reviews describe appropriate models for bone, liver or brain metastasis and are recommended to the readers for understanding further details (Rosol et al., 2003; de Jong et al., 2009; Daphu et al., 2013). Experimental approaches like the colonization assay or specific metastasis assays like the “foot pad” or ear model as well as the resection of already metastasized subcutaneous tumors may reflect the typical clinical situation of metastatic outgrowth to some extent. Nevertheless, the investigation of neo-adjuvant or adjuvant therapies in preclinical metastasis experiments after resecting the primary tumor remains challenging. One of the major problems is the differential growth and invasion of the tumors at the time point of resection and the resulting asynchronous appearance of spontaneous metastases following the surgical removal of the primary tumors (Francia et al., 2011). Besides classical histology for the detection of metastatic cells methods to monitor and measure metastatic burden include confocal or ultramicroscopy, Alu-PCR of lymph nodes and organs (Schneider et al., 2002), or measurement of CTCs (circulating tumor cells). Further methods to visualize metastases are described in recent reviews (Bos et al., 2010; Saxena & Christofori, 2013). 2.6. Syngeneic models In the same way as human cell lines or tumor fragments, also mouse tumor cell lines or tumor fragments can be transplanted onto host animals, either subcutaneously or orthotopically. The mouse tumors used for transplantation mostly originate from tumors occurring in laboratory mouse strains either spontaneously or induced e.g. by carcinogens, or in genetically-engineered tumor models (see following sections). A broad range of long established, well-characterized models (Bertram & Janik, 1980; Overwijk & Restifo, 2001; Pulaski & OstrandRosenberg, 2001) as well as newly emerging models with characteristics closely reflecting molecular features of human cancer exist (Cho et al., 2013; Torres et al., 2013). The advantage of these models is that they can be transplanted “syngeneically”, meaning in a mouse strain with the same genetic background. This allows the use of fully immunocompetent host mice as the transplanted syngeneic tumor is not recognized as a foreign tissue by the host's immune system. Similarly to xenograft tumors, a large set of well characterized syngeneic models is available, offering an easy to handle, fast and low variability approach to assess cancer therapeutics in a fully immunocompetent host. While this is a great setting to study immune modulatory effects of compounds in a meaningful setting with the full repertoire of immune cells present some limitations have to be kept in mind when using syngeneic tumor models: as the models are derived from mouse tumors, they may differ in their repertoire of genetic alterations when compared to human tumors. Also, depending on the drug target investigated, the target biology in the mouse may differ from the human system. Factors to keep in mind in respect to this are target homology, target expression pattern, immune cell function and expression of alternate receptors/pathways. Moreover, especially when working with therapeutic antibodies, the therapeutic compound may not be cross-reactive to the mouse homolog, in which case the use of a mouse specific surrogate compound may be required. Additionally, as for human cell line based xenografts, high passaging numbers of the syngeneic cell lines and in vitro selection may alter the properties of the tumor cell lines not closely reflecting the patient situation. A further problematic feature is the extremely fast tumor growth of many syngeneic models that prevents a proper co-establishment of stromal/ immune infiltrate in the tumor, a process that takes time to develop. In addition, due to their fast growth most syngeneic tumor models only allow a treatment period of two to three weeks. Keeping these limitations in mind, syngeneic tumor models are key models for the evaluation of therapeutics with immune involvement (see sections below) as they offer a fast and reproducible way especially

354

K. Wartha et al. / Pharmacology & Therapeutics 142 (2014) 351–361

for setups where relatively high throughput is needed, such as assessment of different treatment schedules, drug combinations and mechanistic studies. 2.7. Chemically-induced models Another, traditional way to induce tumor formation in mice is the application of carcinogens. Several well established experimental models such as the AOM/DSS colon carcinoma model, the urethane lung carcinoma model or the MNNG gastric cancer model representing different types of mostly inflammation associated cancer, have been used for decades (for reviews of different chemically-induced models see Tsukamoto et al., 2007; Rosenberg et al., 2009; Hayakawa et al., 2013; Vikis et al., 2013). One of the major advantages of chemically-induced models is the slow and relatively natural way of tumor induction. The carcinogen induces mutations in host cells and/or a chronic inflammatory stimulus that ultimately leads to tumor formation. Tumor development in this type of models is slow and allows co-development of tumor microenvironment and immune suppression in the tumor similar to a naturally occurring human tumor. Several types of human tumors, such as gastric cancer following stomach ulcer, colorectal cancer following colitis or smoking-related lung cancer arise under inflammatory conditions or are caused by carcinogens. Thus, chemically-induced models nicely represent a subset of human tumors that, due to the absence of a transplantation event and therefore slow development of tumors in fully immunocompetent host animals, allow a meaningful investigation of immunomodulatory therapeutic effects. Among the disadvantages of chemically-induced models are the large heterogeneity of tumor formation and long timelines until tumor formation occurs. Moreover, tumor formation is difficult to monitor due to the organ-location of the tumors and an important question is when to start treatment in these models to avoid a preventative study design. Similarly to syngeneic models the problems of “mouse tumors” exist: limited homology of the human vs. mouse system and lack of cross-reactivity of compounds (see section Syngeneic models). Overall, chemically-induced models are suitable for the evaluation of therapeutic agents with immune involvement or therapeutics that specifically address inflammatory tumor indications. 2.8. Genetically-engineered mouse models Genetically-engineered mouse models (GEMMs) offer a large set of possibilities for the generation and adaptation of tumor models for specific purposes. Principally one can distinguish two main categories of GEMMs with relevance for preclinical cancer models: i) human transgenic mice carrying specific human genes, and ii) transgenic mice generated to carry cloned oncogenes, or knockout mice lacking tumor suppressor genes that make them prone to develop cancer, so-called “oncomice”. 2.8.1. Human transgenic mice For studying the mode of action of certain therapeutic approaches it can be useful to generate mice that carry specific human transgenes. One possibility is to generate a transgenic mouse expressing the human target protein which allows the analysis of in vivo efficacy and pharmacokinetics for compounds (mainly therapeutic antibodies) that lack cross-reactivity to the respective mouse protein. In other cases it may be useful to generate a transgenic mouse model to allow a more close representation of the mode of action of the drug. The analysis of Fc-receptor mediated immune effector functions of ADCC-enhanced antibodies, for example, is limited in mouse models by divergence in Fcγ receptor expression between mouse and human. While in humans the main effector cells for mediation of ADCC are thought to be NK cells bearing Fcγ receptor IIIa, mouse macrophages but not NK cells express the mouse homologue of human FcγRIIIa, FcγRIV. Therefore, the

generation of mice that express the human FcγRIIIa in NK cells creates a useful tool to fully investigate the mode of action of ADCC-enhanced therapeutic antibodies as shown for the glyco-engineered EGFR antibody GA201 (Gerdes et al., 2013). As the generation of human transgenic mice is a time consuming, labor intensive endeavor, the feasibility is restricted to a few genetic modifications within one engineered mouse line. Moreover, close attention needs to be paid whether the expression levels as well as the spatial and temporal expression patterns of the human transgene in the mouse reflect the physiological situation in patients. 2.8.2. Oncomice Genetically engineered mice that either are transgenic for cloned human oncogenes or bear a knockout of tumor suppressor gene function are popular spontaneous tumor models that closely reflect the natural process of tumor development. Tumors arise slowly over the course of several months and clear progression through premalignant stages up to malignant stages with invasive properties and metastasis formation are described for different strains of oncomice. A wide variety of oncomouse strains have been developed in recent years and are constantly improved and refined using conditional systems. For recent reviews on this topic the reader is referred to Walrath et al. (2010), Westphalen & Olive (2012) and Singh et al. (2012). The clear advantage of oncomice is the slow and spontaneous tumor development process that closely reflects the onset and progression stages of human malignancies. The gradual process of tumor formation also allows a natural recruitment and modulation of stromal cells and immune cells by the tumor and thus a co-development of a functional tumor microenvironment. This, together with the fully immunocompetent mouse background used for most models, predisposes oncomice as highly relevant models to study immunomodulatory effects of cancer therapeutics as well as tumor–stromal interactions and metastatic processes. One striking example for the relevance of GEMMs is represented by the treatment effects observed in preclinical models of pancreatic ductal adenocarcinoma. Most human pancreatic adenocarcinomas are characterized by a strong stromal infiltrate that occupies the majority of the tumor mass consisting of extracellular matrix components and non-neoplastic cells including fibroblasts, vascular and immune cells. This so-called desmoplastic reaction has recently been recognized to be the likely cause of the high treatment resistance of pancreatic cancer. However most cell-line based xenograft models of human pancreatic cancer do not show this desmoplastic reaction and therefore they are poor models to study the role of stroma in pancreatic cancer (Vonlaufen et al., 2008). The KPC model of pancreatic ductal adenocarcinoma, however, where spontaneous tumor development in the pancreas occurs due to a conditional expression of the tumorassociated p53R172H mutation and K-rasG12D mutation, shows a strong desmoplastic reaction and thus more closely represents this important hallmark of human pancreatic cancer. Indeed several examples support the notion that the KPC model and other GEMMs for pancreatic ductal adenocarcinoma may be more predictive of clinical performance than other preclinical models (Feig et al., 2012). The translational publication by Beatty et al. (Beatty et al., 2011) describing the preclinical and clinical findings on the combination of an agonistic CD40 antibody with gemcitabine for the treatment of pancreatic cancer nicely demonstrated that the efficacy and mode of action, namely the activation of macrophages but not T cells, observed in patients could be demonstrated in the KPC model but not in transplanted pancreatic cancer models. Similarly pancreatic cancer GEMMs mirror the resistance of human pancreatic cancer to gemcitabine and the enhanced treatment effect observed in combination with nab-paclitaxel more closely than transplanted models (Singh et al., 2010; Frese et al., 2012). The applicability of GEMMs for preclinical studies is limited by the extremely long timelines, high costs and large inherent variation of the models, and again the possible implications of treating “mouse

K. Wartha et al. / Pharmacology & Therapeutics 142 (2014) 351–361

tumors” (limited homology of the human vs. mouse system, issues with cross-reactivity of compounds, see section Syngeneic models). 2.9. Mouse models with humanized immune system Mice with a humanized immune system, so-called “humanized” mouse models, have been established over the past decades to study the complex interaction of the human immune system or parts thereof during infectious or autoimmune diseases and neoplasia. Generally, human hematopoietic stem cells are grafted into immunodeficient mice with the aim to reconstitute the human hemato-lymphoid system in mice as closely as possible. Since 1988 various immunodeficient mouse strains have been generated to improve the human cell engraftment. Nevertheless, a preconditioning of newborn or juvenile mice by applying irradiation or chemotherapy is required to deplete mouse phagocytic cells, to ablate the mouse bone marrow, and to create a niche for the human stem cells. The source and expansion of human hematopoietic stem cells is a further critical step in the process of generating humanized mice and needs to be optimized to obtain sufficient numbers of comparably reconstituted mice for therapeutic experiments. In spite of all progress, there are still a number of challenges that have to be overcome. Issues as low grade or inadequate maintenance of reconstitution, especially for myeloid cells, are major challenges. Low levels of human cytokines may hinder adequate development and functionality of human immune cells and can be optimized by supplementation or genetic expression of human cytokines in mice. Several excellent publications in the field of “humanized mice” are recommended to the reader for more detailed information (Shultz et al., 2007; Nomura et al., 2008; Drake et al., 2012; Ito et al., 2012; Rongvaux et al., 2013). 3. Preclinical model selection for the evaluation of cancer therapeutics with a specific mode of action 3.1. Preclinical models for the evaluation of tumor cell targeting drugs For the preclinical in vivo evaluation of compounds that interfere with targets expressed on the tumor cells such as cell membrane localized growth factor receptors (e.g. EGFR, HER2, HER3) or components of intracellular signaling pathways (e.g. PI3K inhibitors, MDM2 inhibitors, BRAF inhibitors) a relatively broad set of tumor models exists. In the past, mostly conventional xenograft models (subcutaneous or orthotopic injection of human tumor cell lines) have been used for the evaluation of this type of compounds. This approach is

A

B

355

useful for understanding the basic functionality of the compound, e.g. whether an EGFR inhibitor is capable of binding its target in vivo, inhibits signal transduction, and whether inhibition results in tumor growth inhibition of an EGFR dependent tumor. An example for a successful application of preclinical xenograft data to the clinics is the recently published combination of Herceptin (trastuzumab) + Perjeta (pertuzumab), two HER2 targeting therapeutic antibodies with complementary modes of action (Scheuer et al., 2009). Preclinical evaluation of Herceptin + Perjeta in different HER2 positive xenograft models showed strong synergistic effects of the two compounds (Fig. 1). These preclinical data triggered clinical trials testing the dual HER2-targeting combination and finally resulted in FDA approval of the Herceptin + Perjeta combination in HER2 overexpressing metastatic breast cancer in 2012 (Baselga & Swain, 2010; Baselga et al., 2012). Another example is the HER2 targeted antibody-drug conjugate T-DM1 where preclinical evaluation revealed strong efficacy in HER2 overexpressing but not HER2 normal xenograft models (Lewis Phillips et al., 2008). This preclinical finding was confirmed in a clinical study where T-DM1 demonstrated benefit for patients with HER2 overexpressing (HER2 IHC score 3+ or gene amplified) but not in patients with HER2 low tumors (Burris et al., 2011). However there are some aspects to keep in mind when selecting models and evaluating data in this type of setting. 1) Cross-reactivity of compound: especially therapeutic antibodies often bind only to the human target but are not cross-reactive to the respective mouse protein. In this case the only site where the drug can bind in a standard xenograft model is the human receptor expressed on the tumor cells and no binding in normal (mouse) tissue can occur. This needs to be kept in mind when drawing conclusions in respect to tumor targeting, side effects, PK properties and dose selection. This limitation could be overcome by using mouse strains that are humanized for the target (e.g. human EGFR transgenic mice for the evaluation of EGFR antibodies) or by the use of mouse cross-reactive compounds. A nice example illustrating this aspect is displayed in Fig. 2 showing the tumor targeting properties of different CD44 antibodies. If the human specific but not mouse cross-reactive therapeutic CD44 antibody is injected into tumor-bearing mice the antibody accumulates at the tumor only (Fig. 2A + B). However if a mouse cross-reactive surrogate is employed, binding of the antibody to the spleen, liver and hematopoietic bone marrow can be observed (Fig. 2C). This tissue distribution of the antibody is comparable to the tissue distributions observed in cynomolgus monkeys and humans (Herting, 2012; Vugts et al., 2013). 2) Influence of tumor stroma: The tumor stroma can provide growth factors and cytokines for the tumor cells that can mediate resistance to targeted therapeutics. There is a large body of evidence that stromal factors,

C

Adapted by permission from the American Association for Cancer Research Scheuer et al., Cancer Res. 69: 9330-9336, 2009 Fig. 1. Combination of pertuzumab with trastuzumab has synergistic antitumor activity even after progression on trastuzumab. Antitumor activity of trastuzumab and/or pertuzumab in NSCLC (Calu-3) and breast cancer (KPL-4) xenograft tumor models. (A) Mice bearing Calu-3 and (B) KPL-4 xenograft tumors were treated for the duration of the study with vehicle, pertuzumab, trastuzumab and trastuzumab + pertuzumab. Combination of the two agents resulted in enhanced antitumor activity compared with either agent alone. (C) Mice with KPL-4 xenograft tumors were treated for the duration of the study with vehicle, trastuzumab and trastuzumab followed by pertuzumab starting at day 35. Second line treatment with pertuzumab was sufficient to inhibit tumor growth and reduced tumor mass for further 45 days. Data are indicated as mean tumor volume in mm3 ± SEM. Adapted by permission from the American Association for Cancer Research.

356

K. Wartha et al. / Pharmacology & Therapeutics 142 (2014) 351–361

A

C

B

Fig. 2. Preclinical in vivo imaging of CD44 antibodies. (A) Accumulation of Cy5 labeled RG7356 to human tumor cells is shown by optical imaging in a CD44 expressing xenograft model. (B) Specific binding to the tumor cell membranes was demonstrated by histological investigations. (C) Biodistribution in mice was evaluated with a mouse-surrogate antibody. Measurements 24 h after a single administration of the Cy5 labeled antibody at 50 μg per animal yielded an accumulation in the spleen, bone marrow, tumor and to a less extent in the liver which presents a similar distribution pattern like in humans.

e.g. produced by fibroblasts, mediate resistance to therapies (Paraiso & Smalley, 2013). This needs to be taken into account when interpreting preclinical data. While the presence of stroma is low in most cell line based xenograft models (Ostermann et al., 2008) it has been shown that patient-derived fragment based models frequently have a higher amount of stroma content more comparable to human tumors. As an example, Fig. 3 shows IHC stainings of fibroblast associated protein (FAP), a marker for cancer associated fibroblasts demonstrating a

similar content of FAP + stromal cells in patient-derived fragment based models and human tumors. This finding may indicate that patient-derived fragment based models may be more suitable to study tumor stromal interactions than most cell line based models. However, one also should keep in mind that in xenograft models the tumor cells are human but the surrounding tumor stroma is of mouse origin. It is known that certain factors produced by the tumor stroma, such as e.g. HGF, are not cross-reactive between mouse and human receptors.

Fig. 3. FAP Immunohistochemistry staining of patient-derived mouse models and human tumors. Immunohistochemistry staining of FAP (fibroblast associated protein) was conducted in tissue sections of patient-derived mouse models of colorectal cancer (A) or non-small cell lung cancer (B) and human primary tumors of the same indications (colorectal cancer (C) and non-small cell lung cancer (D)). FAP in human tumors was stained on Paraffin embedded tissue sections using Vitatex D8 antibody. FAP in mouse tumors was stained on fresh frozen tissue sections using an inhouse developed antibody. 4-fold magnifications are shown.

K. Wartha et al. / Pharmacology & Therapeutics 142 (2014) 351–361

Thus potential resistance mechanisms, such as for example HGF/cMET driven tumor growth as a resistance mechanism for EGFR inhibition, may be overlooked in standard xenograft models (Zucali et al., 2008). 3) Selection of a representative model: in order to find the most suitable in vivo model for the evaluation of a novel drug and to limit the number of in vivo models that need to be run, a pre-selection of the model based on the properties of the underlying cell line (e.g. expression of the target and sensitivity to the drug in an in vitro assay) is useful and frequently done. This is suitable for a first in vivo proof of mechanism; however one should keep the pre-selection criteria in mind when interpreting data in respect to predictivity for the clinical setting. One may ask the question: how representative is the selected cell line for the clinical indication the drug shall be used for? How often does this mutation/overexpression/ dependency on a single signaling receptor occur in patient tumors? One way to address this in a more representative way for the clinical setting is the use of patient-derived models. Here, in a more translational approach, a panel of models representing a clinical cancer indication can be selected and the drug sensitivity can be assessed in a more unbiased approach. This can be cost and resource intensive, however it can be useful to better understand the performance of the drug in a more patient-near setting and, if the panel of models is large enough, may even help identifying markers of resistance/sensitivity. How patientderived models may be valuable for response prediction in the clinical setting is indicated by a pilot clinical study published by Hidalgo et al. In this innovative study tumors resected from 14 patients with refractory advanced solid cancers were propagated in immunodeficient mice and treated with 63 drugs. Treatment schedules were selected based on preclinical response and a remarkable correlation between drug activity in the model and clinical outcome was observed (Hidalgo et al., 2011). 3.2. Preclinical models for the evaluation of anti-angiogenic compounds The use of in vivo assays for studying angiogenesis and the interference with anti-angiogenic compounds is broadly applied in drug research. Numerous in vivo assays are available that focus on the formation of new vessels induced by different materials carrying angiogenic cytokines or by implantation of tumor cells or fragments like the chicken chorioallantoic membrane, the cornea pocket, the dorsal air sac or different chamber assays (dorsal skinfold, cranial window) (Staton et al., 2007). These mechanistical angiogenesis models allow investigating vessel growth and their microhemodynamic perfusion on a longitudinal basis by using different imaging methods and thus can be applied to assess specific mode of action questions. On the other hand one should keep in mind that in these models vessels do not grow within the typical tumor microenvironment which may limit their applicability to a significant extent. Alternatively, anti-angiogenic compounds can be studied in subcutaneous or even better orthotopic tumor models where the presence of tumor microenvironment allows studying the effect of stromal factors on tumor vessel formation. Thereby, orthotopic may be favored over subcutaneous implantation as the natural organ environment has been shown to provide a more representative tumor environment and may allow a better buildup of tumor vessel infiltration (Fidler, 2001; Keyes et al., 2003). One also has to keep in mind that growth factors, endothelial cells and pericytes are at least in part derived from the murine host. This is relevant for the application of antibodies which have to be mouse cross-reactive if they shall hit the murine targets. A new era was opened by the development of GEMMs in the field of tumor angiogenesis during the last two decades. The Rip-Tag model for example represents a pancreatic islet ß-cell tumor that exhibits an angiogenic switch through the different stages of tumor formation (Hanahan et al., 1996) and is therefore a powerful model to investigate the mode of action of anti-angiogenic drugs. Although the preclinical evaluation of a multitude of anti-angiogenic drugs in the past years frequently showed strong single-agent activity in

357

numerous xenograft models and GEMMs, this only translated into single-agent clinical efficacy for some indications like renal cell, hepatocellular and ovarian carcinoma (Ebos & Kerbel, 2011). Based on this experience caution has to be taken with the interpretation of preclinical data of anti-angiogenic compounds (Singh & Ferrara, 2012). However, a better understanding of tumor angiogenesis including its key drivers in mouse models and patients with various cancer diseases will be required to enable more consistent translation of preclinical results to the clinical situation. Hence it is essential to better understand the molecular characteristics of neovascularization in mouse models and identify corresponding human tumor types, as well as to characterize patient tumors in detail and establish mouse models that reflect the angiogenic drivers in particular cancer tissues. In an attempt to overcome the disadvantage of missing vessels of human origin in xenografts new mouse models have recently been explored. Coinjection of human embryonic stem cell-derived teratoma cells together with human tumor cells leads to the generation of vessels expressing human tumor vascular markers in the engraftments (Burgos-Ojeda et al., 2013). 3.3. Preclinical models for compounds targeting hematologic malignancies Hematologic malignancies represent about 15% of global oncology patients with particularly high unmet medical need. Leukemia is very heterogeneous with respect to subtypes and in vivo models need to reflect the heterogeneity of hematologic malignancies. As human material from leukemia patients can engraft on mice, classical xenografts are the preferred model. In that approach, primary patient material is transplanted into immunodeficient mice and is thereby recapitulating the original patients' leukemia. The major advantages of such xenografts are: i) the starting material is usually from advanced cancers and relatively homogeneous due to its clonal origin, ii) the tumors represent the original tumors in patients, including their genetic complexity, iii) multiple drugs can be tested in different settings against the same set of cells or against different tumors from different patients (Politi & Pao, 2011). Most of these models are generated by using the severely immunodeficient strains of mice that have been developed at various institutions. A comparison of different immune-compromised strains showed that the new generation NSG strain had superior engraftment kinetics with human cord-blood (McDermott et al., 2010). This strain also bears the advantage not to develop lymphoma over time, in contrast to other NOD/SCID mouse strains. For acute leukemia of both the lymphoid lineage (T-ALL, B-ALL) and myeloid lineage (AML), the generation of patient-derived models works quite uncomplicated. Cells from around 70% of patients will engraft and models can be established. Engraftment can be monitored in the peripheral blood and will be visible in most cases after 4–12 weeks, depending on the cell number and individual patient characteristics. Generally, in the bone marrow the highest human vs. mouse chimerism will be observed, often reaching more than 80% at time of euthanasia (after 1–8 months). Leukemic cells are also found at high numbers in the spleen and sometimes in the liver and brain, and they can even be harvested from these organs. For chronic leukemia the generation of patient-derived models is more complicated. In the case of chronic myeloid leukemia (CML) CD34+ progenitor cells need to be enriched prior to intra-venous injection, which is not only enriching for the leukemic stem cells, but also the normal stem cells. For that reason, normal co-engraftments can occur. Regarding chronic lymphocytic leukemia (CLL) data on how to establish preclinical models are still very heterogeneous and no common protocol could be established yet. The major problem is, that CLL cells engraft only transiently and at low levels. Thus, for CLL the use of cell lines (MEC-1, JVM-3; injected intra-venous, expansion in bone marrow,

358

K. Wartha et al. / Pharmacology & Therapeutics 142 (2014) 351–361

blood, lymph nodes and peritoneum) must be accepted until a way is found to create more robust models.

3.4. Preclinical models for the evaluation of immunomodulatory drugs For the preclinical evaluation of drugs that aim to activate or modulate the anti-tumor immune response a key requirement is the presence of a fully functional immune system. Therefore human xenograft models and patient-derived models are in most cases not appropriate to study this type of drugs. Among the models described above, syngeneic models, chemically-induced models and GEMMs all can be run in fully immunocompetent mice and are therefore potentially suitable models for the evaluation of immunomodulatory drugs. Clearly the selection of the model strongly depends on the questions that need to be addressed by the experiment and the mode of action of the drug that shall be evaluated. Due to ease of handling most preclinical testing of immunomodulatory drugs has so far been done in syngeneic transplant models. For example therapeutic blockade of immune inhibitory checkpoints with antibodies targeting CTLA-4 or PD-1/PD-L1 were evaluated in multiple syngeneic models. These models indicated that CTLA-4 monotherapy efficacy may be expected in immunogenic tumors whereas less immunogenic tumors require combination with other therapeutics, e.g. vaccines. This could be confirmed in clinical trials leading to FDA approval of the CTLA-4 antibody ipilimumab in advanced melanoma in 2011 (Grosso & Jure-Kunkel, 2013). Moreover preclinical data predict that combined inhibition of CTLA-4 and PD-1 leads to improved efficacy. This hypothesis is currently investigated in multiple clinical trials and first data from a recent Phase I study look promising (Callahan & Wolchok, 2013; Intlekofer & Thompson, 2013). As described above these models have the advantage to be well characterized, fast growing, reproducible and fully immunocompetent. However dependent on the drug and mode of action to be studied there are some aspects to be considered before relying solely on syngeneic models. i) Syngeneic models are cell line based and due to in vitro selection and transplantation of large amounts of tumor cells grow fast and largely independent of stromal infiltrate. This fast growth may limit the formation of an immune suppressive tumor microenvironment. The use of chemically-induced mouse models or GEMMs may overcome this issue. In these models the tumors form slowly, thus allowing the development of an immune suppressive microenvironment as found in many human tumors. The evaluation of the CD40 antibody in the KPC model of pancreatic cancer is a good example for the relevance of GEMMs for the evaluation of immunomodulatory drugs (see section on Genetically Engineered Mouse Models, (Beatty et al., 2011)). ii) Differences in mouse and human immune systems are another factor that needs to be taken into account when selecting models and interpreting data. Mouse and human immune cells can differ in their immune cell repertoires and the arsenal of receptors expressed on the different cell types (for a review on differences between human and mouse immune system see Mestas & Hughes, 2004). Especially myeloid cell populations such as for example macrophage subtypes (M1 and M2) and myeloid derived suppression cells (MDSC) frequently do not have exact matching human homologues and are characterized by different sets of markers in human versus mouse (Murdoch et al., 2008). Therefore in general, it is of great importance to closely characterize the immune infiltrate of the model to select an appropriate model for the evaluation of the drug at hand and understand the potential limitations of the model due to differences in the immune cell repertoire. Tumor models with a humanized immune system might be one way to overcome these issues in the future (see also section on Mouse models with humanized immune system).

4. Translational implications of preclinical study design In the clinical trial situation, Response Evaluation Criteria in Solid Tumors (RECIST) have been agreed upon by several international organizations and clinical study groups (Therasse et al., 2000; Eisenhauer et al., 2009). These criteria have been widely accepted by regulatory agencies and are nowadays used in most clinical trials. The basis of response evaluation in patients is the calculation of the sum of longest diameters (LD) of predefined tumor target lesions. Briefly, complete response (CR) is defined as disappearance of all target lesions, partial response (PR) as a decrease in the sum of LD by at least 30%, progressive disease (PD) by at least a 20% increase in the sum of LD, and stable disease (SD) when neither sufficient shrinkage to qualify for PR nor sufficient increase to qualify for PD occurs. RECIST criteria also form the basis for another commonly used endpoint in clinical studies, progression-free survival (PFS). PFS refers to the time elapsed from treatment initiation to the occurrence of progressive disease according to RECIST. In analogy to PFS, calculation of disease-free survival (DFS) time is used in adjuvant studies (see below), describing the duration of absent cancer phenotype after complete remission until eventual disease recurrence. The most indisputable read-out and therefore most widely accepted by regulatory agencies, however, is the median overall survival (OS) time, which reflects the period from the start of cancer therapy until the point in time when half of the patients have passed away. Another frequently used endpoint in clinical trials, for debulking the disease or evaluating efficacy of a novel treatment applied before radical surgery, is the rate of pathological complete response (pCR), defined as the complete disappearance of all cancer disease including at the site of the primary tumor. The apparent different properties of mouse models compared to patient tumors imply difficulties in comparing commonly used readouts for the evaluation of treatment effects. In preclinical models, the profoundly different growth characteristics of tumors require different criteria for the evaluation of drug treatment effects. As slowing down the rapid growth of xenograft tumors often indicates therapeutic efficacy of a compound that is also active in a clinical trial setting, the concept of calculating tumor growth inhibition (TGI) has been established on an empirical basis. Tumor volumes are usually determined by two perpendicular measurements of a palpable nodule using a caliper, and the relative effect on tumor growth from treatment inception to study termination is calculated. A result of at least 50% TGI is commonly considered to indicate relevant antitumor activity that may translate into clinical benefit. Another aspect to be taken into account for the set-up of mouse tumor models is the fact that the therapeutic use of anticancer drugs in patients is established for different treatment strategies depending on the type of tumor, stage of disease, and line of treatment. The following clinical settings in which anticancer drugs are commonly used may be considered for preclinical study design. 4.1. Metastatic disease Frequently, cancer is diagnosed in humans only when metastatic spread to one or several organs or tissues distant to the site of primary tumor emergence has already occurred. Together with the relatively prolonged etiology this advanced stage may imply a higher phenotypic/histopathologic and genetic diversity of the disease compared to cell line based xenograft models. This situation may be better reflected in GEMMs or carcinogen induced mouse models (see above) which develop several tumors per animal embodying some histopathological and molecular biological variations. Patient tumor heterogeneity may be further enhanced by the administration of several lines of treatment including mutagenic chemotherapy drugs, e.g. alkylating agents and topoisomerase inhibitors. Drug resistance models can be established for preclinical investigation of alternate therapeutic approaches, but as multiple factors may render

K. Wartha et al. / Pharmacology & Therapeutics 142 (2014) 351–361

patient tumors insensitive to treatment the respective mechanisms involved need to be well-defined. Despite the presence of multiple genetic lesions and the progressive nature of metastatic cancer, tumor xenograft models usually have a significantly higher proliferation rate compared to patient tumors. This implies a bias in favor of preclinical models when the antitumor efficacy of cytotoxic chemotherapy agents which preferentially act on dividing cells is tested. The discrepant proliferation rates of patient versus xenograft tumors should therefore be considered for the choice of preclinical models for the evaluation of compounds with cell cycle specific activity. Another aspect in the metastatic disease situation of patients is the relatively high tumor burden which is not appropriately reflected in common xenograft models. High tumor burden entails various challenges for drug activity, like high interstitial pressure limiting access for compounds, genetic diversity, and development of a more complex tumor microenvironment. Taken together, these disparities explain why compounds highly active in preclinical models often produced disappointing clinical results. However, it is well-known that advanced metastatic disease with high tumor burden is generally difficult to treat. Therefore, the traditional way of clinical development of novel anticancer agents should be challenged: instead of initial trials in advanced metastatic cancer, testing of new compounds should be moved to earlier stages of disease. Such strategies will be even more important for novel targeted agents, because accumulating genetic variations along disease progression are more likely to enable escape mechanisms to single target assault approaches. On the other hand, preclinical models supposed to represent advanced stages of human cancer, like carcinogen-induced models, need to be characterized in molecular detail in order to conclude on which tumor properties they actually represent. The use of patient-derived tumors may ensure more heterogeneity, but it needs to be considered whether the respective set-up

359

of the model (e.g. ectopic, orthotopic) has influence on its growth properties, development of a functional tumor microenvironment, and the relevance of the drug target investigated. 4.2. Adjuvant treatment After the introduction of chemotherapy for the treatment of patients with tumors, oncologists found out that additional treatment with chemotherapy (and/or radiotherapy) after complete removal of all detectable disease, usually by radical surgery, could delay or prevent eventual relapse. The prophylactic treatment, now called adjuvant therapy, obviously destroys occult tumor cells that pose a certain statistical risk for disease recurrence. Independent of their mode of action, basically all compounds used in the adjuvant setting have also shown activity on established patient tumors (metastases), including different cytotoxic chemotherapy principles, as well as endocrine and biological therapies (Senkus et al., 2013). Therefore, no particular set-up of mouse models seems to be required for preclinical testing of compounds useful for adjuvant treatment. Nevertheless, it should be noted that some models used to investigate inhibition of metastasis formation, like the footpad model mentioned in Section 2.4, may actually represent a situation with several physiological similarities to the adjuvant therapy setting. 4.3. Neoadjuvant therapy The administration of therapeutic agents is often used to reduce the size and extent of the cancer before surgery. In addition such treatment may act on micro-metastases, thus reducing the risk of disease relapse. Today, neoadjuvant studies are more and more used to compare the potency of different drugs and combination therapies, particularly with targeted agents, because this setting allows most relevant

Model profiling

Mouse with Human Immune System

In vivo imaging Signaling / oncogenic drivers

Ex vivo signaling analysis

Immune infiltrate analysis

GEMMs

Tumorstromal interactions

Human Transgenic Models

Carcinogenesis Models

Metastasis Models

Angiogenesi s Models

Patient-derived Models

Orthotopic Models

Ex vivo vessel analysis

Syngeneic Models

Xenograft Models

Database on model characteristics Fig. 4. Schematic drawing of collection of available preclinical tumor models and suggested ‘fit-for-purpose’ application. The pyramid depicts the different classes of currently available mouse tumor models that are used for preclinical evaluation of anti-tumor therapeutics. While classical xenograft models still build the basis for preclinical testing, more sophisticated models (such as GEMMs, patient-derived xenograft tumors) may be used to gain a better understanding of the drugs' mode of action. For a fit-for purpose application of models a thorough characterization of the tumor models (e.g. by characterization of the immune infiltrate, vessel content and signaling analysis) is required to understand the effects observed in preclinical experiments and thereby improves translatability of the findings.

360

K. Wartha et al. / Pharmacology & Therapeutics 142 (2014) 351–361

biomarker investigations by taking tumor biopsies before, during, and after treatment. The clinical endpoint in neoadjuvant trials usually is the rate of pathological complete response (pCR). Considering the therapeutic set-up of commonly used xenograft models, where treatment is initiated as soon as the growth of palpable tumors can be reliably monitored, they seem to be most closely related to the clinical neoadjuvant setting. However, the rate of complete tumor regression did not turn out to be a practical read-out for the evaluation of drug potency, because complete remission is rarely seen in xenograft experiments. And a direct comparison to the clinical situation would not be justified, because, in addition to the different growth characteristics of xenografts, the heterogeneity of tumors in a patient population is not represented in a model based on a clonal cell line. 5. Conclusions In the present review we have tried to summarize the current view on preclinical models for the evaluation of anti-cancer drugs and advocate a fit-for-purpose application of models (Fig. 4). While a large array of models exists, it is clear that no single model can capture all the different aspects of tumor growth characteristics and treatment approaches. Therefore, one needs to be aware of the limitations of each respective model when drawing conclusions on efficacy and clinical applicability of a novel therapeutic compound. Key points to consider for model selection and evaluation of preclinical data are: i) Detailed molecular and pathological characterization of the model (e.g. oncogenic drivers, immune infiltrate and tumor vasculature, mouse strain) is critical to understand the properties and limitations of the model and thus allow better interpretation of data, ii) Study design and tumor model should be adapted to the translational questions asked, and iii) A set of models that capture different aspects of the mode of action of the drug may be combined to give an as comprehensive picture of the drug's properties as possible. If these points are considered and data are observed in the context of a good understanding of the model and their limitations it may help to improve trial design and translatability of findings. However, in addition to a thoughtful and reasonable fit-for purpose application of currently existing tumor models work on improved models such as GEMMs and humanized models is key to further improve preclinical models and thereby drug evaluation and patient benefit. 6. Conflict of interest statement All authors are employed by Roche Diagnostics GmbH. Acknowledgments We thank Maike Schmitz for her contribution to “Preclinical models for compounds targeting hematologic malignancies”. We also thank Suzana Vega Harring and Hadassah Sade for providing immunohistochemistry analysis and images. References ASCO (2012). American society of clinical oncology. Progress against cancer (http://www. cancer.net/sites/cancer.net/files/vignette/Progress_Against_Cancer_Timeline.pdf [On-line]) Baselga, J., Cortes, J., Kim, S. B., Im, S. A., Hegg, R., Im, Y. H., et al. (2012). Pertuzumab plus trastuzumab plus docetaxel for metastatic breast cancer. N Engl J Med 366, 109–119. Baselga, J., & Swain, S. M. (2010). CLEOPATRA: A phase III evaluation of pertuzumab and trastuzumab for HER2-positive metastatic breast cancer. Clin Breast Cancer 10, 489–491. Beatty, G. L., Chiorean, E. G., Fishman, M. P., Saboury, B., Teitelbaum, U. R., Sun, W., et al. (2011). CD40 agonists alter tumor stroma and show efficacy against pancreatic carcinoma in mice and humans. Science 331, 1612–1616. Bertram, J. S., & Janik, P. (1980). Establishment of a cloned line of Lewis lung carcinoma cells adapted to cell culture. Cancer Lett 11, 63–73. Bos, P. D., Nguyen, D. X., & Massague, J. (2010). Modeling metastasis in the mouse. Curr Opin Pharmacol 10, 571–577.

Burgos-Ojeda, D., McLean, K., Bai, S., Pulaski, H., Gong, Y., Silva, I., et al. (2013). A novel model for evaluating therapies targeting human tumor vasculature and human cancer stem-like cells. Cancer Res 73, 3555–3565. Burris, H. A., III, Rugo, H. S., Vukelja, S. J., Vogel, C. L., Borson, R. A., Limentani, S., et al. (2011). Phase II study of the antibody drug conjugate trastuzumab-DM1 for the treatment of human epidermal growth factor receptor 2 (HER2)-positive breast cancer after prior HER2-directed therapy. J Clin Oncol 29, 398–405. Callahan, M. K., & Wolchok, J.D. (2013). At the bedside: CTLA-4- and PD-1-blocking antibodies in cancer immunotherapy. J Leukoc Biol 94, 41–53. Chan, E., Patel, A., Heston, W., & Larchian, W. (2009). Mouse orthotopic models for bladder cancer research. BJU Int 104, 1286–1291. Cho, S., Sun, Y., Soisson, A. P., Dodson, M. K., Peterson, C. M., Jarboe, E. A., et al. (2013). Characterization and evaluation of pre-clinical suitability of a syngeneic orthotopic mouse ovarian cancer model. Anticancer Res 33, 1317–1324. Daphu, I., Sundstrom, T., Horn, S., Huszthy, P. C., Niclou, S. P., Sakariassen, P. O., et al. (2013). In vivo animal models for studying brain metastasis: Value and limitations. Clin Exp Metastasis 30, 695–710. de Jong, G. M., Aarts, F., Hendriks, T., Boerman, O. C., & Bleichrodt, R. P. (2009). Animal models for liver metastases of colorectal cancer: Research review of preclinical studies in rodents. J Surg Res 154, 167–176. De, R. M., Massi, E., Poeta, M. L., Carotti, S., Morini, S., Cecchetelli, L., et al. (2011). The AOM/DSS murine model for the study of colon carcinogenesis: From pathways to diagnosis and therapy studies. J Carcinog 10, 9. Drake, A.C., Chen, Q., & Chen, J. (2012). Engineering humanized mice for improved hematopoietic reconstitution. Cell Mol Immunol 9, 215–224. Ebos, J. M., & Kerbel, R. S. (2011). Antiangiogenic therapy: Impact on invasion, disease progression, and metastasis. Nat Rev Clin Oncol 8, 210–221. Eisenhauer, E. A., Therasse, P., Bogaerts, J., Schwartz, L. H., Sargent, D., Ford, R., et al. (2009). New response evaluation criteria in solid tumours: Revised RECIST guideline (version 1.1). Eur J Cancer 45, 228–247. Feig, C., Gopinathan, A., Neesse, A., Chan, D. S., Cook, N., & Tuveson, D. A. (2012). The pancreas cancer microenvironment. Clin Cancer Res 18, 4266–4276. Ferlay, J., Steliarova-Foucher, E., Lortet-Tieulent, J., Rosso, S., Coebergh, J. W., Comber, H., et al. (2013). Cancer incidence and mortality patterns in Europe: Estimates for 40 countries in 2012. Eur J Cancer 49, 1374–1403. Fidler, I. J. (2001). Angiogenic heterogeneity: Regulation of neoplastic angiogenesis by the organ microenvironment. J Natl Cancer Inst 93, 1040–1041. Francia, G., Cruz-Munoz, W., Man, S., Xu, P., & Kerbel, R. S. (2011). Mouse models of advanced spontaneous metastasis for experimental therapeutics. Nat Rev Cancer 11, 135–141. Frese, K. K., Neesse, A., Cook, N., Bapiro, T. E., Lolkema, M. P., Jodrell, D. I., et al. (2012). nab-Paclitaxel potentiates gemcitabine activity by reducing cytidine deaminase levels in a mouse model of pancreatic cancer. Cancer Discov 2, 260–269. Gerdes, C. A., Nicolini, V. G., Herter, S., Van, P. E., Lang, S., Roemmele, M., et al. (2013). GA201 (RG7160): A novel, humanized, glycoengineered anti-EGFR antibody with enhanced ADCC and superior in vivo efficacy compared with cetuximab. Clin Cancer Res 19, 1126–1138. Grosso, J. F., & Jure-Kunkel, M. N. (2013). CTLA-4 blockade in tumor models: An overview of preclinical and translational research. Cancer Immun 13, -5. Hanahan, D., Christofori, G., Naik, P., & Arbeit, J. (1996). Transgenic mouse models of tumour angiogenesis: The angiogenic switch, its molecular controls, and prospects for preclinical therapeutic models. Eur J Cancer 32A, 2386–2393. Hayakawa, Y., Fox, J. G., Gonda, T., Worthley, D. L., Muthupalani, S., & Wang, T. C. (2013). Mouse models of gastric cancer. Cancers 5, 92–130. Herting, F. (2012). Antibodies to target cancer. Preclinical assays in cancer therapy. Amsterdam: NKI-AVL. Hidalgo, M., Bruckheimer, E., Rajeshkumar, N. V., Garrido-Laguna, I., De, O. E., Rubio-Viqueira, B., et al. (2011). A pilot clinical study of treatment guided by personalized tumorgrafts in patients with advanced cancer. Mol Cancer Ther 10, 1311–1316. Intlekofer, A.M., & Thompson, C. B. (2013). At the bench: Preclinical rationale for CTLA-4 and PD-1 blockade as cancer immunotherapy. J Leukoc Biol 94, 25–39. Ito, R., Takahashi, T., Katano, I., & Ito, M. (2012). Current advances in humanized mouse models. Cell Mol Immunol 9, 208–214. Keyes, K. A., Mann, L., Teicher, B., & Alvarez, E. (2003). Site-dependent angiogenic cytokine production in human tumor xenografts. Cytokine 21, 98–104. Kola, I., & Landis, J. (2004). Can the pharmaceutical industry reduce attrition rates? Nat Rev Drug Discov 3, 711–715. Lewis Phillips, G. D., Li, G., Dugger, D. L., Crocker, L. M., Parsons, K. L., Mai, E., et al. (2008). Targeting HER2-positive breast cancer with trastuzumab-DM1, an antibodycytotoxic drug conjugate. Cancer Res 68, 9280–9290. Loi, M., Di, P. D., Becherini, P., Zorzoli, A., Perri, P., Carosio, R., et al. (2011). The use of the orthotopic model to validate antivascular therapies for cancer. Int J Dev Biol 55, 547–555. Lum, D. H., Matsen, C., Welm, A. L., & Welm, B. E. (2012). Overview of human primary tumorgraft models: Comparisons with traditional oncology preclinical models and the clinical relevance and utility of primary tumorgrafts in basic and translational oncology research. Curr Protoc Pharmacol. http://dx.doi.org/10.1002/0471141755 (Chapter 14, Unit). Malaney, P., Nicosia, S. V., & Dave, V. (2013). One mouse, one patient paradigm: New avatars of personalized cancer therapy. Cancer Lett. http://dx.doi.org/10.1016/j.canlet.2013. 10.010. McDermott, S. P., Eppert, K., Lechman, E. R., Doedens, M., & Dick, J. E. (2010). Comparison of human cord blood engraftment between immunocompromised mouse strains. Blood 116, 193–200. Mestas, J., & Hughes, C. C. (2004). Of mice and not men: Differences between mouse and human immunology. J Immunol 172, 2731–2738. Murdoch, C., Muthana, M., Coffelt, S. B., & Lewis, C. E. (2008). The role of myeloid cells in the promotion of tumour angiogenesis. Nat Rev Cancer 8, 618–631.

K. Wartha et al. / Pharmacology & Therapeutics 142 (2014) 351–361 Nomura, T., Habu, S., & Watanabe, T. (2008). Humanized mice. Berlin, Heidelberg: Springer-Verlag Berlin Heidelberg. Ostermann, E., Garin-Chesa, P., Heider, K. H., Kalat, M., Lamche, H., Puri, C., et al. (2008). Effective immunoconjugate therapy in cancer models targeting a serine protease of tumor fibroblasts. Clin Cancer Res 14, 4584–4592. Ottewell, P. D., Coleman, R. E., & Holen, I. (2006). From genetic abnormality to metastases: murine models of breast cancer and their use in the development of anticancer therapies. Breast Cancer Res Treat 96, 101–113. Overwijk, W. W., & Restifo, N.P. (2001). B16 as a mouse model for human melanoma. Curr Protoc Immunol. http://dx.doi.org/10.1002/0471142735 (Chapter 20, Unit). Paraiso, K. H., & Smalley, K. S. (2013). Fibroblast-mediated drug resistance in cancer. Biochem Pharmacol 85, 1033–1041. Politi, K., & Pao, W. (2011). How genetically engineered mouse tumor models provide insights into human cancers. J Clin Oncol 29, 2273–2281. Pulaski, B.A., & Ostrand-Rosenberg, S. (2001). Mouse 4T1 breast tumor model. Curr Protoc Immunol. http://dx.doi.org/10.1002/0471142735 (Chapter 20, Unit). Reyal, F., Guyader, C., Decraene, C., Lucchesi, C., Auger, N., Assayag, F., et al. (2012). Molecular profiling of patient-derived breast cancer xenografts. Breast Cancer Res 14, R11. Rongvaux, A., Takizawa, H., Strowig, T., Willinger, T., Eynon, E. E., Flavell, R. A., et al. (2013). Human hemato-lymphoid system mice: Current use and future potential for medicine. Annu Rev Immunol 31, 635–674. Rosenberg, D. W., Giardina, C., & Tanaka, T. (2009). Mouse models for the study of colon carcinogenesis. Carcinogenesis 30, 183–196. Rosol, T. J., Tannehill-Gregg, S. H., LeRoy, B. E., Mandl, S., & Contag, C. H. (2003). Animal models of bone metastasis. Cancer 97, 748–757. Sano, D., & Myers, J. N. (2009). Xenograft models of head and neck cancers. Head Neck Oncol 1, -32. Saxena, M., & Christofori, G. (2013). Rebuilding cancer metastasis in the mouse. Mol Oncol 7, 283–296. Scheuer, W., Friess, T., Burtscher, H., Bossenmaier, B., Endl, J., & Hasmann, M. (2009). Strongly enhanced antitumor activity of trastuzumab and pertuzumab combination treatment on HER2-positive human xenograft tumor models. Cancer Res 69, 9330–9336. Schneider, T., Osl, F., Friess, T., Stockinger, H., & Scheuer, W. V. (2002). Quantification of human Alu sequences by real-time PCR—an improved method to measure therapeutic efficacy of anti-metastatic drugs in human xenotransplants. Clin Exp Metastasis 19, 571–582. Senkus, E., Kyriakides, S., Penault-Llorca, F., Poortmans, P., Thompson, A., Zackrisson, S., et al. (2013). Primary breast cancer: ESMO Clinical Practice Guidelines for diagnosis, treatment and follow-up. Ann Oncol 24(Suppl. 6), vi7–vi23. Sennino, B., Ishiguro-Oonuma, T., Wei, Y., Naylor, R. M., Williamson, C. W., Bhagwandin, V., et al. (2012). Suppression of tumor invasion and metastasis by concurrent inhibition of c-Met and VEGF signaling in pancreatic neuroendocrine tumors. Cancer Discov. 2, 270–287. Shultz, L. D., Ishikawa, F., & Greiner, D. L. (2007). Humanized mice in translational biomedical research. Nat Rev Immunol 7, 118–130.

361

Singh, M., & Ferrara, N. (2012). Modeling and predicting clinical efficacy for drugs targeting the tumor milieu. Nat Biotechnol 30, 648–657. Singh, M., Lima, A., Molina, R., Hamilton, P., Clermont, A.C., Devasthali, V., et al. (2010). Assessing therapeutic responses in Kras mutant cancers using genetically engineered mouse models. Nat Biotechnol 28, 585–593. Singh, M., Murriel, C. L., & Johnson, L. (2012). Genetically engineered mouse models: Closing the gap between preclinical data and trial outcomes. Cancer Res 72, 2695–2700. Siolas, D., & Hannon, G. J. (2013). Patient-derived tumor xenografts: Transforming clinical samples into mouse models. Cancer Res 73, 5315–5319. Staton, C. A., Lewis, C., & Bicknell, R. (2007). Angiogenesis assays: A critical appraisal of current techniques. (Wiley.com) Tentler, J. J., Tan, A.C., Weekes, C. D., Jimeno, A., Leong, S., Pitts, T. M., et al. (2012). Patient-derived tumour xenografts as models for oncology drug development. Nat Rev Clin Oncol 9, 338–350. Therasse, P., Arbuck, S. G., Eisenhauer, E. A., Wanders, J., Kaplan, R. S., Rubinstein, L., et al. (2000). New guidelines to evaluate the response to treatment in solid tumors. European Organization for Research and Treatment of Cancer, National Cancer Institute of the United States, National Cancer Institute of Canada. J Natl Cancer Inst 92, 205–216. Torres, M. P., Rachagani, S., Souchek, J. J., Mallya, K., Johansson, S. L., & Batra, S. K. (2013). Novel pancreatic cancer cell lines derived from genetically engineered mouse models of spontaneous pancreatic adenocarcinoma: Applications in diagnosis and therapy. PLoS One 8, e80580. Tseng, W., Leong, X., & Engleman, E. (2007). Orthotopic mouse model of colorectal cancer. J Vis Exp, -484. Tsukamoto, T., Mizoshita, T., & Tatematsu, M. (2007). Animal models of stomach carcinogenesis. Toxicol Pathol 35, 636–648. Vikis, H. G., Rymaszewski, A. L., & Tichelaar, J. W. (2013). Mouse models of chemically-induced lung carcinogenesis. Front Biosci (Elite Ed) 5, 939–946. Vonlaufen, A., Joshi, S., Qu, C., Phillips, P. A., Xu, Z., Parker, N. R., et al. (2008). Pancreatic stellate cells: Partners in crime with pancreatic cancer cells. Cancer Res 68, 2085–2093. Vugts, D. J., Heuveling, D. A., Stigter-van Walsum, M., Weigand, S., Bergstrom, M., van Dongen, G. A., et al. (2013). Preclinical evaluation of 89Zr-labeled anti-CD44 monoclonal antibody RG7356 in mice and cynomolgus monkeys: Prelude to Phase 1 clinical studies. mAbs 6, 0–1. http://dx.doi.org/10.4161/mabs.27415. Walrath, J. C., Hawes, J. J., Van, D. T., & Reilly, K. M. (2010). Genetically engineered mouse models in cancer research. Adv Cancer Res 106, 113–164. Watanabe, H., Numata, K., Ito, T., Takagi, K., & Matsukawa, A. (2004). Innate immune response in Th1- and Th2-dominant mouse strains. Shock 22, 460–466. Westphalen, C. B., & Olive, K. P. (2012). Genetically engineered mouse models of pancreatic cancer. Cancer J 18, 502–510. Zucali, P. A., Ruiz, M. G., Giovannetti, E., Destro, A., Varella-Garcia, M., Floor, K., et al. (2008). Role of cMET expression in non-small-cell lung cancer patients treated with EGFR tyrosine kinase inhibitors. Ann Oncol 19, 1605–1612.