The standard murine model utilizes a transplantable tumor system in which cultured cell lines derived from murine tumors of various origins can be injected, typically into the subcutaneous region of a mouse. These tumor cells must be derived in the same genetic background of the mouse and allows for rapid growth, simple growth assessment, and peripheral blood can be collected or mice can be euthanized at various times for toxicity analysis. These models are particularly useful for rapid studies of potential immunotherapy drug combinations and allow for inclusion of appropriate treatment controls. The model, however, has numerous limitations, including the lack of appropriate tumor microenvironment, potential problems with limited host immune system interactions since tumors often grow quickly, and local injection may induce inflammation that can influence tumor growth or drug response. Further, if the tumor cells harbor foreign transgenes, these may serve as xenoantigens and inadvertently promote tumor rejection. In some cases, murine tumor-associated antigens have been identified and these can be used to monitor immune responses and determine if tolerance can be broken during combination immunotherapy. While transplantable tumors are usually established in the subcutaneous location of the flanks, it is possible to inject the cells orthotopically, or in natural locations in which the tumor arises, to replicate the normal local environment. The potential importance of orthotopic models has been confirmed in at least one study in which the therapeutic efficacy of immunotherapy was less prominent when the cells were implanted into the kidney compared to subcutaneous tumors [52].

In order to better mimic human tumors, spontaneous tumor models have been developed in which tumors arise in the histologic tissue of origin, and these offer the benefit of more accurately reflecting patterns of tumor growth and treatment response kinetics. Such models may also be more appropriate for development and detection of irAEs. A major drawback to these models is that they generally require carcinogen induction or genetic manipulation, which may limit their clinical relevance. Examples of carcinogen-induced spontaneous tumors include methylcholanthrene (MCA)-induced fibrosarcomas and 7,12-dimethylbenz[a]anthracene (DMBA)/12-O-tetradecanoylphorbol-13-acetate (TPA)-induced skin papillomas [52, 53]. While these are more physiologically relevant, they often take longer for cancers to develop and may be associated with considerable heterogeneity requiring many more animals to obtain the required number of tumors and determine therapeutic activity. Further, establishing the timing of treatment may be particularly challenging since neoplastic transformation may occur over variable time periods with these models.

Genetically-engineered mouse models (GEMMs) utilize forced oncogene expression or knockout of known tumor suppressor genes, usually in a tissue-specific and/or temporally controlled manner [53, 54]. There are now several well-established, genetically mediated spontaneous tumor models in use. These include several breast cancer models in which selected oncogenic transgenes are driven by promoters that drive transgene expression in the mouse mammary epithelium [55]. The transgenes include Her-2/neu (ErbB2), polyoma middle T antigen (PyMT), simian virus 40 (SV40) T antigen, Ha-Ras, Wnt-1, TGF-α, and c-Myc. Oncogene expression in the MMTV-Neu and MMTV-PyMT mice is driven by the mouse mammary tumor virus promoter, and these mice develop multifocal mammary tumors and can exhibit metastatic disease in the lungs and lymph nodes, which typically occur after the first pregnancy. In contrast, the SV40 transgenic mice develop invasive tumors without the need for hormonal manipulation. In some cases, tumors require two genetic defects to promote tumor development.

There have been over 60 spontaneous murine models of melanoma reported to date [56, 57]. This includes a model in which the RET oncogene is fused to the metallothionein-I (MT) promoter-enhancer in a mixed murine strain background (C57BL/6xBALB/c), which produced systemic skin melanosis and spontaneous benign melanocytic tumors [58]. By backcrossing the mice over 10 times into C57BL/6 mice, a line in which melanocytic lesions progressed to invasive melanoma after several months was developed, and tumors eventually metastasize to lymph nodes and visceral organs. Bosenberg has developed an especially useful model in which melanoma is driven by both BRAF mutation and PTEN loss. These mice, characterized as Braf^CATyr-creER^T2Pten^fl/fl, develop melanoma after exposure to 4-hydroxytamoxifen (4-HT), which induces de novo melanoma initiation [56]. A murine model of autochthonous lung tumors has been reported using adenoviral vectors encoding Cre recombinase, KRAS, and p53 in the pulmonary epithelia [59]. In general, the carcinogen-induced models are considered highly immunogenic with emergence of numerous neoantigens that can be recognized by the immune system. In contrast, models driven by germline mutations are typically not very immunogenic [60, 61].

Although it may seem counterintuitive, a number of immunodeficient murine models have been used to conduct mechanistic studies of immunotherapy. Today, there are numerous such models with selective as well as more global deficiencies in immune cells or immunologic function. These models can, thus, be classified as those with severe combined immunodeficiencies (SCID) and those with selective immunodeficiencies. The simplest immune deficient mouse was termed the nude (nu) mouse, in which thymic development is thwarted and results in deficient T cell maturation [62]. Nude mice were originally derived from mice with defects in the Forkhead box protein N1 (FOXN1) gene [63]. These mice typically live 6–12 months, accept xenografts and allow reasonable time for tumor treatment experiments. Nude females may be unable to nurse their young due to defects in mammary gland milk production and have largely been supplanted by SCID models.

A commonly used SCID model is based on knocking out the V(D)J recombination activation gene (RAG-1). RAG-1^−/− mice lack mature B and T cells, and these mice generally will not reject transplanted tumors [64]. The potential with these mice is that subpopulations of lymphocytes can be adoptively transferred, and the therapeutic impact of various treatment regimens can be assessed with and without specific lymphocyte populations. Because of this capability, these models have provided valuable insights into the mechanisms underlying antitumor immunity [65–67]. Another SCID model uses mice with mutations in the Prkdc gene, which encodes a protein that resolves DNA strand breaks during V(D)J recombination and results in the absence of functional B and T cells [68, 69]. These Prkdc^scid (also known as non-obese diabetic or NOD^scid) mice do have a normal hematopoietic microenvironment, can accept both allogeneic and xenogeneic grafts, allow adoptive transfer experiments, and rarely develop mature lymphocytes. A particularly immunodeficient SCID model is the NSG mouse (NOD^scidgamma; NOD.Cg-Prkdc ^scid Il2rg ^tm1Wjl /SzJ), a strain of inbred mice in which the Prkdc gene and IL-2 receptor gamma gene, which is critical for IL-2-mediated signaling, are knocked out [70]. NSG mice lack both innate and adaptive immunity with loss of B cell, T cell, and NK cell function as well as reducedmacrophage and antigen-presenting cell function [71]. These mice are highly permissive for xenogeneic tumor engraftment and have been instrumental in studies of tumor immunotherapy and other human diseases. A variety of murine strains in which single molecular pathways are disrupted have also been generated and can be used for selective mechanistic studies.

The selection of immunodeficient murine strains for experimental studies depends on several features. The strain background is important as this may influence the H2 haplotype, tumor cell engraftment potential, and disease susceptibility. The NOD mice, for example, are prone to diabetes and lack innate immunity. The functional consequences of the genetic defects also need to be considered in strain selection. Some mice exhibit “leakiness”, in which the mice may start to generate functional immune cells as they age. The emergence of mature B and T cells has been reported in Prkdc^scid mice when they are older, especially if they are housed in non-specific pathogen-free conditions. Leakiness may also be more common in certain genetic backgrounds, such as C57BL/6 J and BALB/cByJ mice. The lifespan of individual mouse strains is also an important consideration as some immunodeficient mice die at a young age, with some becoming susceptible to thymic lymphomas, and this may limit their potential for long-term experiments. Some strains may also have difficulty breeding, as occurs with female nude mice, and this can limit usefulness. Some strains are highly sensitive to radiation (e.g., Prkdc^scid mice) and this can limit irradiation prior to engraftment or prevent studies of combination approached that utilize radiation therapy. Some mice also require pathogen-free environments, and husbandry capabilities may limit the choice of model. Finally, the impact of genetic mutations and how these influence cell function should be considered. For example, mutations in perforin can decrease NK cell activity whereas defects in the IL-2 receptor gamma chain can completely eliminate NK cell function.

The improved engraftment of human tumor cells in some of the SCID mouse models has allowed the generation of so-called humanized mouse models. These models further utilize the transfer of hematopoietic stem cells or, more recently autologous peripheral blood, to reconstitute the normal human immune system [72]. While several models have been proposed with considerable progress in replicating human immune-tumor components, there is still debate about how closely the humanized mice mimic the human host. Investigators are exploring the number of cells transferred, the route of transfer, the timing and age of transfer, and irradiation sources to optimize immune engraftment. Whether these mice truly organize a relevant tumor microenvironment remains unclear, but there is some evidence that mild graft-versus-host disease (GVHD) can develop, suggesting these models may be helpful for evaluating immune system activation and emergence of irAEs [72]. Other strategies in development include engineering expression of various cytokines into the mice to allow for more efficient immune function. Further refinements may be necessary before these mice can be endorsed as a significant improvement over other models. The generation of humanized mice is also complicated by the need for access to human tumor tissue and hematopoietic cells, institutional review board (IRB) approval, and often a need for rapid execution of cell transfer and frequently high cost to conduct experiments.

Many of the murine models have been helpful for evaluating therapeutic activity of monotherapy and combination immunotherapy agents, but few have faithfully replicated the toxicity profiles observed in humans [73–76]. It is possible, however, that the manifestations of irAEs may be present but challenging to detect in the murine models and may depend on the length of tumor establishment, background strain of the mouse used, or subtle impact of genetic changes in some models. Despite these limitations, there has been some progress in which autoimmune side effects have been observed. This includes the appearance of vitiligo in C57BL/6 mice bearing melanoma tumors and treated with a variety of immunotherapy strategies or the induction of hypophysitis in SJL/J mice treated with multiple treatments of CTLA-4 blockade [73, 74]. A strategy to better evaluate toxicity may be to use murine models in which the particular mouse strain is more susceptible to development of autoimmune symptoms (e.g., NOD, SJL/J, etc.). Another approach may be to add additional immune regulation to the model. For example, eradication of CD4 + FoxP3+ Tregs has been reported in the DEREG mouse model where mice have been engineered to express a diphtheria toxin (DT) receptor-enhanced green fluorescent protein fusion protein driven by the FoxP3 gene locus [75]. These mice permit conditional depletion of Tregs using injections of diphtheria toxin and some experiments have shown that DEREG mice may be more susceptible to autoimmune-related side effects with immunotherapy treatment [76].

The availability of numerous murine models that allow establishment of human tumors and immune system components provides an important resource for more rapidly testing rational combinations of immunotherapy agents. The large number of models further promotes more relevant systems to assess both therapeutic response and propensity for irAEs. While all models have limitations (see Table 1), the range of models allows selection of systems that most closely resemble the particular cancer, immunologic targets, and genetic factors that most closely mirror the human host and permit more rapid development of novel combination treatment strategies for clinical trials.

There has been considerable controversy regarding limitations, both real and perceived, in the utility of preclinical tumor models as tools to inform the clinical development of new oncology agents. While some limitations are certainly clear, in other instances, concerns may be driven by mouse models being used inappropriately or unrealistically, as opposed to intrinsic flaws in the models themselves. For several reasons, preclinical models may be particularly useful for cancer immunotherapy and in the development of new combination immunotherapy regimens [77]. Optimizing the dose, schedule, and configuration of immunotherapeutic combinations may be complex, yet as discussed above, is critical to additively or synergistically engage immunoregulatory mechanisms and maximize the risk-benefit profile of a given regimen. This may necessitate the comparison of a range of distinct schedules and configurations for combinations to maximize both their pharmacodynamic activity and their antitumor efficacy with acceptable tolerability. While the need for additional clinical optimization of dose and schedule is often inevitable, the assessment of new regimens in rigorous preclinical models may help to substantially focus the scope and cost of these efforts, and also may enable the interrogation of candidate clinical biomarkers to monitor the biologic activity of these combinations. Preclinical models may also enable more thorough understanding of the interaction between tumor and the host immune system in vivo, and may be utilized to enable rational, hypothesis-driven identification of mechanism-based combinations for clinical testing. In tandem with more rigorous early clinical development of combination regimens, preclinical models may play an important role in identifying and optimizing the safety, clinical activity, and overall risk-benefit profile of immunotherapeutic combinations.

Historically, clinical development of many oncology combinations has proceeded using traditional development paradigms, where individual molecules undergo rigorous clinical testing as monotherapy, and often, combination trials have been deferred until clear monotherapy proof-of-concept has been established. As such, single agent phase I and phase II trials with each agent were typically executed, and only then, would phase I/II trials be triggered to investigate specific combinations. In many instances, failure to demonstrate substantive monotherapy activity led to cessation of further clinical development. This approach has been well-established using small molecules, but may be less well-suited for some immunotherapy agents, where clinical development efforts can be terminated prematurely based on unrealistic expectations for monotherapy activity using conventional criteria. More recently, driven by increasing recognition of the potential of immunotherapeutic combinations, as well as the reality that some agents may yield only modest clinical activity as monotherapy yet be highly active in the context of a therapeutic combination, innovative trial designs have been deployed increasingly to test these agents [103, 104]. These include run-in trial designs, zig-zag designs with or without de-escalation, and bifurcated designs among others. Run-in trial designs offer the prospect, where appropriate, for staged, sequential combination of an investigational agent with another standard-of-care drug, within the same patient after an initial monotherapy run-in window. At the study level, run-in trial designs may offer the prospect for getting an initial characterization of monotherapy safety before each patient is exposed to the combination, and conceptually, may enable more rapid identification of the MTD for a regimen depending on the extent of dose escalation. This approach may be most appropriate when there is particularly well-substantiated data suggesting that the safety profile of a given combination is anticipated to be highly favorable, and the respective agents clearly have non-overlapping safety profiles. Such studies can be executed with or without the option for de-escalation, but are likely used most appropriately when deployed with the flexibility for dose de-escalation of either agent. So-called “zig-zag” escalation trials afford the flexibility to explore various dose combinations of the respective agents by alternating the increases in dosing of each agent during dose escalation. This design may be particularly useful when the key driver of the clinical activity and/or safety of a given combination is less clear or when there is an anticipation that a given combination may have a narrower therapeutic window. Bifurcated designs also have been used recently for the investigation of some immunotherapeutic combinations. These are particularly well-suited for the testing of combinations where it is anticipated that an investigational agent will have a modest safety profile and limited potential for monotherapy clinical activity, yet there is clear rationale that this agent may synergize when administered in combination with another drug. In this approach, monotherapy dose escalation is executed through several dose levels with the novel agent. Presuming acceptable safety, the trial may then “bifurcate” down two distinct paths for subsequent escalation. On one arm, continued monotherapy escalation of the novel agent is pursued as appropriate until the MTD, maximum biologically-effective dose (MBED), or maximum administered dose (MAD) is defined. In parallel, escalation of the combination is pursued on a second arm, typically by combining a dose of the novel agent that is 1–2 dose levels below the highest monotherapy dose that already has been shown to be safe, in combination with another agent. Escalation can then proceed using either a fixed dose of the second agent or can proceed using a “zig-zag” approach guided by features of the specific combination. This approach allows for more rapid triggering of combination testing in clinical development, but may not be appropriate for some combinations. Further, a bifurcated-design trial should typically be structured so that the dose of the novel agent being tested in the combination arm does not exceed the dose that has been deemed safe in the monotherapy arm of the study.

We are entering an exciting era for combination immunotherapy that offers the prospect to build upon the powerful proof-of-principle established by the clinical experience with combination checkpoint blockade. Several key considerations may play an important role in enabling future progress with this approach. These include: a) rigorous assessment of the optimal dose, sequence, and schedule of agents in both preclinical models and the clinical setting; b) flexible approaches to decision-making in the selection of dose and schedule, and application of this decision-making across multiple indications; c) recognition that some agents may have limited monotherapy clinical activity, yet have high potential for clinical activity in the setting of an immunotherapy combination; d) careful attention to supportive care, including education of both patients and all members of the health-care team regarding the importance of early recognition and intervention for the management of irAEs; and e) the use of properly designed trials that enable efficient testing of the safety and clinical activity of combination immunotherapy regimens.