Mouse Models of Metastasis and Dormancy

Abstract

Metastasis is the ultimate and often lethal stage of cancer. Metastasis occurs in three phases that may vary across individuals: First, dissemination from the primary tumor. Second, tumor dormancy at the metastatic site where micrometastatic cancer cells remain quiescent or, in dynamic cycles of proliferation and elimination, remaining clinically undetectable. Finally, cancer cells are able to overcome microenvironmental constraints for outgrowth, or the formation of clinically detectable macrometastases that colonize distant organs and are largely incurable. A variety of approaches have been used to model metastasis to elucidate molecular mechanisms and identify putative therapeutic targets. In particular, metastatic dormancy has been challenging to model in vivo due to the sparse numbers of cancer cells in micrometastasis nodules and the long latency times required for tumor outgrowth. Here, we review state-of-the art genetically engineered mouse, syngeneic, and patient-derived xenograft approaches for modeling metastasis and dormancy. We describe the advantages and limitations of various metastasis models, novel findings enabled by such approaches, and highlight opportunities for future improvement.

Metastasis remains the leading cause of cancer deaths. Patients with primary tumors but no metastasis, corresponding to clinical stage I–III cancer, can often be cured with surgery and/or radiation and systemic therapy, but once a cancer has established growth in a distant organ (clinical stage IV disease), such metastatic disease typically becomes resistant to therapies and is largely incurable (Siegel et al. 2020; Seer Cancer Statistics Review (CSR) 1975–2016, seer.cancer.gov/archive/csr/1975_2016). Lineage-tracing of both human and mouse cancers suggest that only a small subset of cancer cells from the primary tumor successfully initiates distant tumor regrowth (Luzzi et al. 1998; Lambert et al. 2017; Massagué and Ganesh 2021). Such cancer cells, dubbed metastasis-initiating cells (MICs), emerge through selection of preexisting tumor subclones or through dynamic phenotypic plasticity, can survive dissemination through the lymphovasculature, evade immune surveillance, extravasate at a distant site, and seed metastases. In some cases, disseminated cancer cells (DCCs) can enter into a period of dormancy, when microscopic cells persist but are clinically undetectable. Some MICs, either immediately after seeding or by overcoming barriers to exit dormancy, eventually grow out into clinically detectable macrometastatic tumors. MICs must accumulate a succession of phenotypic traits to overcome the hurdles of the invasion-metastasis cascade (Fig. 1). The ability of MICs to overcome these hurdles and adapt to a new microenvironment in the absence of newly acquired mutations highlights their phenotypic plasticity—the ability to dynamically adopt distinct cell states (Gerstberger et al. 2023).

Figure 1.

Invasion-metastasis cascade. (A) Dissemination: cancer cells at the invasion front of invasive carcinomas that have spread into the submucosa can undergo epithelial-to-mesenchymal transition (EMT), remodel the extracellular matrix, and interact with other cell types (e.g., fibroblasts, macrophages, and endothelial cells). Migrating cancer cells can intravasate as individual cells or cell clusters. (B) Dormancy: once cancer cells disseminate to distant organs, they can settle in a dormant niche where they can exit the cell cycle by suppressive signaling from the native tissue or the absence of required mitogens. Dormant lesions can achieve dynamic equilibrium with their microenvironment where hypoxic conditions and/or immune cells cull proliferating tumor cells, preventing the outgrowth of clinically detectable metastases. (C) Colonization: metastatic colonization is the final phase of the invasion-metastasis cascade. Once permissive conditions arise, through cellintrinsic reprogramming, undergoing an angiogenic switch, and/or suppression of the immune system, dormant lesions can progress into clinically detectable metastatic tumors. Notably, these metastases can reinitiate the invasion-metastasis cascade and, in turn, disseminate to and colonize additional distant organs. (CTC) Circulating tumor cell.

Mouse models of cancer have yielded great insight into the biology of primary tumor initiation, yet studying metastasis with mouse models remains complex and technically challenging, due to the inherent long time-course and genotypic and phenotypic complexity of human metastatic cancer. In patients, metastasis can present in one of two forms: as micrometastatic lesions that are radiologically invisible and may persist after the surgical removal of a primary tumor. Such metastasis may remain dormant for months or years before eventually regrowing to establish macrometastasis; alternatively, metastasis can present as disease that is already widespread and macrometastatic at the time of patient presentation to the clinic. Here, we discuss advances in mouse modeling that have revealed novel insights on dormancy and metastatic outgrowth.

THE INVASION-METASTASIS CASCADE

Metastatic diseases develop through three main phases: (1) dissemination from the primary tumor, (2) dormancy, and (3) colonization (Massagué and Ganesh 2021). Understanding these stages of metastasis is crucial to selecting the most appropriate mouse model to address the hypothesis being tested.

Dissemination

Cancer cells at the invasion front of a primary tumor detach from their neighbors and gain migratory and invasive properties that resemble embryonic epithelial-to-mesenchymal transition (EMT) (Nieto et al. 2016). Such cells can enter the venous or lymphatic capillaries within the submucosa, gaining access to routes for systemic dissemination. Cells that are present in the blood circulation, known as circulating tumor cells (CTCs), can be detected as single cells or as cell clusters. CTCs interact with other cell types in circulation and can be coated with platelets that protect from shear stress and immune cell attack (Gay and Felding-Habermann 2011). Upon arrival at the capillary bed or a distant organ, CTCs can extravasate through the interstitial space of endothelial cells and settle in the perivascular niche (Reymond et al. 2013; Carlson et al. 2019).

Dormancy

Patients who have been treated for a primary tumor and gone into remission can relapse with metastatic disease after a variable disease-free interval that can last decades in some cancer types, such as hormone receptor-positive breast cancer (Karrison et al. 1999; Weckermann et al. 2001). Dormancy may not apply to all cancer types or individuals. For example, small-cell lung cancer is characterized by rapid development of macrometastatic disease within months of primary tumor removal. Two opposing models, the “uninterrupted growth” model and “tumor dormancy” model, have been proposed to explain the long latency period between the treatment of the primary tumor and manifestation of metastatic disease (Salmon and Kyle 1995). Clinical, experimental, and mathematical models support the tumor dormancy model, which posits that residual disease, tumor cells that cannot be detected using conventional imaging, enter dormancy—a period of tumor-intrinsic or niche-imposed tumor growth arrest—prior to manifesting in some cases as overt metastatic disease (Meltzer 1990; Democheli et al. 1994; Demicheli 2001). Data from clinical trials have shown that, despite benefiting subsets of patients, adjuvant therapy fails to prevent metastatic relapse in many patients with breast, colorectal, and pancreatic cancers, underscoring the need for an improved understanding of the dormant state and its therapeutic vulnerabilities (Schwartz et al. 2002; Demicheli et al. 2005; Bonadonna et al. 2009; Sinicrope et al. 2011; Nomura et al. 2019). The term “tumor dormancy” has been used to refer to two distinct phenomena: (1) cancer cell dormancy, in which cancer cells reversibly exit the cell cycle and enter a G₀ nonproliferative state; and (2) tumor mass dormancy, in which a cluster of cancer cells is at equilibrium with its surrounding microenvironment, where the rate of cancer cell proliferation is opposed by an equal death rate, due to hypoxia or immunosurveillance.

Dormant DCCs can be detected in the bone marrow, kidney, and other organs long after the initial removal and treatment of the primary tumor (MacKie et al. 2003; Shiozawa et al. 2011; Xiao et al. 2013). In some cases, such dormant cells may never progress to form macrometastases, whereas in others the dormant period may be minimal and metastatic tumors rapidly progress. Modeling and understanding the determinants of dormancy and regrowth is therefore critical and represents a major opportunity for the prevention of macrometastasis and cancer mortality. Micrometastatic lesions may reside in this dormant state until a microenvironment permissive for colonization outgrowth arises, either due to stochastic changes in the cancer cells or host, or due to extrinsic changes related to aging, diet, or other systemic factors (Chambers 2009; Koelwyn et al. 2017; Pascual et al. 2017; Hopkins et al. 2018; Cannioto et al. 2021).

Colonization

Metastatic colonization, or the outgrowth of clinically detectable tumors, is the final phase of the invasion-metastasis cascade. Many cancer types preferentially metastasize to specific target organs (e.g., the liver in colorectal carcinomas [CRCs], the bone in hormone receptor-positive breast cancer and prostate cancer, and the lung in renal and thyroid carcinomas) (Nguyen et al. 2022). Such organotropism arises due to two distinct phenomena. One is the ability of invading cancer cells to reach a distant organ based on the anatomy of blood flow or direct proximity, and the second is finding or reengineering a microenvironment to support metastatic outgrowth, first described by Stephen Paget in 1889 as the “seed and soil” hypothesis (Paget 1889). Cancer cells disseminating from specific primary sites may preferentially lodge in the capillary beds of select target organs (e.g., the majority of the venous blood flow from the intestines goes through the portal vein into the liver, where it filters through hepatic sinusoids before eventually entering the systemic circulation via the hepatic vein) (Moore and Bridenbaugh 1951; Dionne 1965). Thus, intestinal cancer cells may have a greater likelihood of metastasizing to the liver simply due to the higher numbers of cancer cells reaching the liver first. On the other hand, studies have also shown that most cancers will disseminate dormant lesions throughout the body, suggesting that MICs may only colonize tissues that can support metastatic outgrowth. While DCCs are frequently found in the bone marrow and kidney, these sites are not common sites of metastatic colonization in most cancers. Organotropism of metastatic outgrowth is therefore likely a combination of both phenomena, where circulation delivers CTCs to specific tissues in a skewed distribution, increasing the chances of stochastic metastatic outgrowth, and MICs preferentially adapting and proliferating in specific microenvironments.

APPROACHES AND CONSIDERATIONS FOR HUMAN-IN-MOUSE MODELING OF METASTASIS

A number of mouse models have been developed for studying metastasis, including genetically engineered mouse models (GEMMs), syngeneic murine allografts, and human-derived xenograft models (Fig. 2). In the modern era, analysis of patient samples offers many opportunities to directly study tumor and microenvironmental changes in primary tumors, CTCs, and metastatic biospecimens directly from patients (e.g., using single-cell transcriptomics and multisite copy number alteration-based lineage tracing). However, mouse models, which can be perturbed using sophisticated genetic tools as well as pharmacological approaches, offer powerful opportunities to mechanistically interrogate cancer cells in the distinct tissue contexts of the primary and metastatic sites, and to probe the relative contributions of tumor and microenvironment to each step of the metastatic cascade. In particular, critical insights into dormancy, which cannot be readily studied in patients, have been gleaned from sophisticated mouse models. The choice of an ideal mouse model is heavily influenced by phase of metastasis being investigated—dissemination, dormancy, or colonization, the source of the biological material (murine- or human-derived cancer cells), the specific cancer and target organ of interest, and practical implications such as the availability of GEMMs and the ability to perform technically challenging mouse surgeries. While these models have led to many advances in metastasis research, the pros and cons of each model must be considered.

Figure 2.

Patient- and murine-derived models of metastasis. (A) Xenograft models: patient-derived cell lines, organoids, or xenografts, either from the primary tumors or metastases, can be transplanted into immunocompromised mice. With the source material being patient tumors, these cells retain the mutations, karyotypic alterations, and heterogeneity seen in human disease. (B) Genetically engineered mouse models (GEMMs): GEMMs give rise to primary tumors that can spontaneously metastasize to distant sites. GEMMs recapitulate invasion-metastasis cascade and allow for the interrogation of the native tumor microenvironment (TME), but metastatic tumors arise unpredictably with low penetrance. (C) Murine-derived cell lines, organoids, and xenografts can be derived from GEMMs or wild-type mice (whereafter mutations are introduced using gene-editing strategies) and injected into immunocompetent mice from the same strain. These tumors often carry fewer mutations than seen in patients and fail to capture the full scope of cancer cells’ abnormalities and heterogeneity.

Human xenograft models involve injecting human cancer cells, such as organoids, cell lines, or patient-derived tumor cell suspensions of patient-derived xenografts (PDXs), into mice. These injections could be orthotopic, at the same site or organ from where the cancer cells were derived, or ectopic, at a different site or organ. Human-derived xenograft models allow for the investigation of human cancer-specific mechanisms of metastasis that reflects metastatic disease in patients with high fidelity. This has the advantage of preserving patient-specific genotypic alterations, including complex karyotypic alterations and germline single-nucleotide polymorphisms that may be difficult to accurately model in mice. In principle, patient and potentially metastatic, site-specific intratumor heterogeneity and phenotypic plasticity could also be captured, although the extent to which such dynamic features of the tumor are retained in vitro or upon long-term passaging outside the patient is unknown, and likely to be variable. A major limitation of most patient-derived models, however, is that they lack intact stromal, including immune, microenvironments, and must be transplanted into immunocompromised mice, typically NOD.Cg-Prkdc^scid Il2rg^tm1Wjl/SzJ (NSG) mice that lack functional B, T, and NK cells, and have defective myeloid differentiation (Shultz et al. 2005; Coughlan et al. 2016). Thus, features ofmetastasis that depend on interactions with the host-immune system cannot be accurately modeled. Promising developments in the use of “humanized” mice, mice that have been engrafted with a human immune system (Guil-Luna et al. 2021), could offer a way to circumvent this limitation in the future.

Source of Patient Cancer Cells

Historically, metastasis has been studied by injecting cancer cell lines into subcutaneous fat, kidney capsule, or directly into the circulation of immunocompromised mice. Such cancer cell lines can be derived from diverse source materials including patient primary tumors and various metastatic sites including pleural effusions (e.g., MDA-MB-231 triple-negative breast cancer) and ascites. While these models have been used to great effect and are simple and scalable, they rely on cell lines that have been in culture for decades and are largely homogeneous. Clinical annotation of the source material is typically minimal and the extent to which such lines have drifted genetically and phenotypically from their source material is unknown. With the advent of short tandem repeat (STR) profiling for cell line authentication, some issues of misidentification of cell lines have been identified, rendering conclusions of earlier studies performed with such models suspect. More recently, efforts have been made to access early passage, fully clinically annotated source materials for introduction into mice with the goal of generating more representative models. Perhaps most widespread are PDX models, wherein tumor tissue surgically removed from a patient is minced, mixed with Matrigel, and implanted into the subcutaneous fat (Rygaard and Povlsen 1969; Johnson et al. 2001). Such models can sometimes spontaneously metastasize from the subcutaneous site to distant organs (Eyre et al. 2016; Ma et al. 2021). Cell lines derived from such xenografts can also be inoculated into the circulation or orthotopically into metastatic sites (see below). Maintaining PDX models can be costly and labor intensive, requiring passaging mostly through mice, and, given poor penetrance of metastasis and limited ability to genetically engineer cells, such models have been used sparingly for studying metastasis (Caeser et al. 2022).

More recently, patient-derived organoid (PDO) models have provided a powerful alternative to study metastasis directly from patient samples. Organoids involve ex vivo culture of cancer (or noncancer) cells in three-dimensional cultures in defined stem-cell-based media (Sato et al. 2009; Barker et al. 2010). Organoids are readily genetically engineered and retain genetic and phenotypic heterogeneity observed in vivo, as well as patient-specific drug sensitivity, and can be transplanted into orthotopic metastatic sites and circulation (Vlachogiannis et al. 2018; Ganesh et al. 2020). For some cancer types, such as colorectal cancer, where organoid technology is well established and is highly efficient for patient samples, patient-derived models have become established as the standard patient-derived models for the field. Complementing these approaches are patient-derived tumor spheroids, typically used from brain tumors and breast cancer (Xiao et al. 2018). Finally, the source of cancer cells from the same patient is worth considering. Metastasis poses an evolutionary bottleneck, with only a subset of cells in the primary tumor becoming capable of eventual macrometastatic colonization (Tentler et al. 2012; Hidalgo et al. 2014). Depending on the question being studied (e.g., mechanisms of EMT vs. mechanisms of end organ colonization), cancer cells from the primary tumor or one or more metastatic sites could be chosen for murine xenotransplantation. However, metastatic cells capable of simply proliferating in a metastatic organ (i.e., participating in metastatic outgrowth), may not be capable of de novo seeding and initiation, and vice versa. A further source of patient cells is from the circulation (CTCs), capturing cells in transit, part way through the metastatic cascade, yielding a source of cells likely to be more capable of metastasis than those in the primary tumor, but perhaps less so than those in established metastasis (Castro-Giner and Aceto 2020). In general, primary patient-derived cells are typically inefficient at metastasizing upon xenotransplantation, and extensive characterization and in vivo selection may be required to render them useful as experimental model systems.

Organotropic Selection

Cancer cells undergo attrition at each stage of the invasion-metastasis cascade, making metastasis a slow process. The slow progression makes modeling metastasis in mice particularly challenging as most models cannot simultaneously capture all stages of metastatic disease. While transplantation models that yield spontaneous metastases recapitulate dissemination, dormancy, and metastatic colonization, they require long incubation times of several months, during which mice often succumb to their primary tumor. To circumvent this limitation, investigators often rely on serial in vivo selection, which involves the inoculation of the parent cancer cells into mice and deriving secondary lines with higher metastatic capacity. Such secondary lines do not harbor new mutations but are instead a product of a serial selection of subclones that are more competent at trafficking into and colonizing specific organs (Jacob et al. 2015).

This process was elegantly utilized in the now classic work of Isaiah J. Fidler. Fidler and colleagues injected PC3, a prostate cancer cell line, into the prostate of immunosuppressed mice and were able to derive cell lines from the regional and distant metastases, which exhibited higher tumorigenic and metastatic capacity than the parent PC3 cell line (Pettaway et al. 1996). This strategy has subsequently been extensively leveraged by the group of Joan Massagué to identify organ-specific metastatic adaptations and mediators that enable colonization in brain (Bos et al. 2009), bone (Kang et al. 2003), lungs (Minn et al. 2005), leptomeninges (Boire et al. 2017), among others. The process of in vivo selection has galvanized metastasis research as it allows for the derivation of cancer cells that exhibit highly penetrant and short latency metastatic phenotypes and can be used to study organ-specific metastatic disease. However, the repeated selection of highly metastatic cells could lead to cells adopting hyperaggressive cell states that are not clinically relevant, potentially biasing studies to investigate mechanisms that are not present in patients. Further, it should be kept in mind that the parental cell lines used to derive aggressive metastatic clones are themselves typically derived from metastatic tumors in patients and should not be taken to represent features of primary tumors. A more rigorous comparison of matched primary and metastatic tumors directly from patients, with their in vivo–selected counterparts, is needed to fully understand the scope and limitations of serial in vivo selection.

Approaches for Inoculating into Circulation

There are various approaches to model and investigate specific stages of the metastatic cascade using xenograft models. The route of delivery of cancer cells influences their metastatic capacity (Fig. 3). A common approach to model metastatic seeding and colonization is to inject cancer cells directly into the circulation into the vessel supplying the metastatic organ of interest. Thus, the function of genes influencing extravasation, seeding, and outgrowth in a particular organ can be studied. Using an intrasplenic injection model, we demonstrated the role of L1 cell-adhesion molecule (L1CAM) in liver metastatic seeding and colonization (Ganesh et al. 2020). Recently, Wang et al. developed a novel strategy to model bone metastasis where cancer cells are injected into the common iliac artery in a mouse’s hind limbs (Wang et al. 2018). Using the interiliac injection model, Zhang et al. were able to demonstrate the role of the bone marrow in promoting metastasis-to-metastasis seeding, distinct from primary tumor-to-metastasis seeding (Zhang et al. 2021). To study multiorgan metastasis, intracardiac injection, introducing cancer cells into the left ventricle of the heart, enables dissemination through the aorta to all the capillary beds in the body, and allows selection of subclones particularly capable of colonizing individual organs in a physiologically relevant manner (Arguello et al. 1988).

Figure 3.

Approaches for introducing metastatic cancer cells into mice. Metastasis mouse models: cancer cells can be transplanted into the primary tumor site, injected into circulation, or directly transplanted into the metastatic site to model metastatic disease. Each model has advantages and disadvantages, so the most appropriate model for each study must be chosen with the question in mind (e.g., the stage of metastasis being modeled). (PDX) Patient-derived xenograft.

Orthotopic Transplantation

Delivering cancer cells directly into circulation allows for the focused study of metastatic dormancy and colonization but does not recapitulate the early steps of dissemination from the primary tumor. Orthotopic injection models wherein cancer cells are transplanted into the relevant organ require more sophisticated and technically challenging surgical approaches but can be leveraged to interrogate the mechanisms of CTC dissemination. Diamantopoulou et al. (2022) injected cancer cells into the fourth right mammary fat pad of female mice to demonstrate the increased rates of intravasation of CTCs during sleep. Mira et al. (2017) microinjected CTCs into mice cecal submucosa to demonstrate the role of HGF/MET signaling in promoting metastatic dissemination (Mira et al. 2017). O’Rourke et al. developed a novel spontaneous metastasis model wherein patient-derived organoids are transplanted intraluminally into the rectums of dextran sodium sulfate-treated mice and demonstrated that colorectal cancer cells disseminate to and colonize the liver and lungs (O’Rourke et al. 2017; Ganesh et al. 2019). Transplantation models are powerful tools for recapitulating the metastasis-invasion cascade. However, although many transplantation models develop primary tumors, relatively few succeed at developing detectable macrometastases within the time window of the study. This reflects two challenges: (1) The relatively slow progression of metastasis in patients, which occurs over years or decades, in contrast to the few weeks or months for which most transplanted animals are studied, and (2) xenografted mice will typically die of their primary tumors prior to the development of distant metastasis. To overcome this limitation, models have been developed to first xenotransplant cancer cells into the orthotopic primary site (e.g., mammary fat pad [Correia et al. 2021] or cecum [Cañellas-Socias et al. 2022]), then surgically remove the orthotopic primary tumor after a few weeks, with subsequent monitoring of animals for distant macrometastasis that may arise weeks to months after surgery. Such models are particularly powerful to study the biology of residual disease and dormancy but are resource intensive and technically complex to master.

Alternatively, cancer cells can be transplanted directly into the metastatic target organ to assess colonization competency and mechanisms. Ebright et al. (2020) transplanted patient-derived CTCs in the right frontal lobe of mice to uncover the role of hypoxia-inducible factor 1α (HIF-1α) in promoting brain metastatic colonization. Hepatic metastatic colonization can similarly be interrogated using an intra-hepatic injection model where patient-derived cancer cells are inoculated below the liver capsule. These models provide valuable insights into the mechanisms of metastatic colonization but cannot recapitulate the dissemination and dormancy stages of metastasis.

APPROACHES AND CONSIDERATIONS FOR MOUSE-IN-MOUSE MODELING OF METASTASIS

In contrast to human cancer cells that must be implanted into immunocompromised mice, syngeneic mouse models, with an intact immune system and the absence of MHC ligand/TCR mismatch that occurs in xenograft models, enable interrogation of immune modulation of cancer progression. On the other hand, even the most sophisticated engineered mouse models lack the complexity and heterogeneity of human tumors and evolve under less stringent selective pressures for shorter durations of time (and hence fewer replication cycles) than advanced human cancers. Further, significant differences exist between the murine and human immune systems, limiting the extent to which observations made in the mouse can be directly translated to humans, especially in the context of preclinical testing of immunotherapeutic strategies (Mestas and Hughes 2004).

Genetically Engineered Mouse Models

GEMMs rely on tissue-specific gene expression to selectively induce tumorigenesis through inducing driver mutations commonly found in various carcinomas in target organs. GEMMs capture de novo tumor evolution in its native organ and can give rise to metastases in distant organs. Such models of tumorigenesis in intact organisms with spontaneous distant metastasis can powerfully recapitulate all the steps of the invasion-metastasis cascade, yet may be unwieldy for experimentation since the timing of metastasis can be difficult to predict. Further, only a few models in a small number of cancer types exist that can capture this full spectrum of disease without early mortality due to large primary tumor burden. A further limitation is the inability to capture the heterogeneity of genomic alterations typically observed among patients with the same cancer type.

GEMMs harboring the polyoma middle T antigen (PyMT) under the mammary epithelium specific promoter, the mouse mammary tumor virus long terminal repeat (MMTV-LTR) (Fantozzi and Christofori 2006; Borowsky 2011), exhibit spontaneous metastasis and have been used in several studies. The MMTV-PyMT GEMM model yields palpable mammary tumors that spread to lymph nodes and metastasize to the lungs, but not the bone, brain, or liver, which are common sites of metastasis in breast cancer. MMTV-PyMT GEMM model relies on the PyMT protein, which has been shown to inhibit p53-dependent transcription and actuate Pi3k/Akt signaling (Oliveira et al. 1999). Recently, Ross et al. (2020) were able to identify metastasis-specific genetic alterations using the MMTV-PyMT GEM model. A major advantage of these GEMM models is that the mammary tumors can be surgically excised while allowing metastases to develop, analogous to patients relapsing with metastatic tumors after undergoing mastectomies. While breast cancer GEMMs have been used to uncover important findings related to tumor initiation and have been valuable as preclinical models, their reliance on exogenous promoters and viral proteins means they may not faithfully recapitulate human disease and limits their utility in studying advanced carcinomas.

Arriaga et al. (2020) recently developed a GEMM, NPK^EYFP, that generates adenocarcinomas through selectively editing Pten and Kras in luminal prostatic cells. While the NPK^EYFP GEMM yields tumors capable of metastasizing to the bone, it relies on mutating Kras, which is not commonly altered in prostate cancer although RAS signaling is frequently altered. In addition to breast and prostate cancer, metastatic GEMMs currently exist for cancers of the lung (DuPage et al. 2009), pancreas (Hingorani et al. 2005), and colorectum (Calon et al. 2012; Tauriello et al. 2018), although the latter are seldom used because of their inflexibility due to the need to breed multiple alleles, long generation time of metastasis, and unpredictability of metastatic timing.

Syngeneic Allograft Mouse Models

Syngeneic allografts involve transplanting murine cell lines or organoids from GEMMs or chemical carcinogenesis models or by genetically engineering normal murine cells in vitro, and then transplanting such murine cancer cells ectopically into the circulation or orthotopically into disease-relevant sites in mice with the same genetic background (i.e., of the same strain). Such transplantation-based models circumvent the problem of rapid primary tumor growth frequently noted in GEMMs, and highly penetrant injection-based metastasis models have been established for a range of cancer types. Given their scalability, reproducibility, flexibility, and technical simplicity, such models have become the mainstay of immunotherapy research (e.g., CT26, MC38 mismatch repair-deficient colorectal cancer, and B16 melanoma) (Pilon-Thomas et al. 2010; Germano et al. 2017).

Fidler and colleagues were among the first to establish syngeneic mouse models for studying metastasis when they injected murine melanoma cells into mice and derived lines from spontaneous lung metastases (Fidler 1973). Using these models, Fiddler conducted foundational work that established basic concepts such as metastatic progression, tumor heterogeneity, and organotropism (Fidler and Kripke 1977). The 4T1 mammary carcinoma cell line is a syngeneic mouse model that has been widely used to interrogate metastatic triple-negative breast cancer. Once transplanted in the mammary gland of syngeneic Balb/c mice, 4T1 tumors can metastasize to the brain, liver, lung, and bone, all common metastatic sites for breast cancer (Pulaski and Ostrand-Rosenberg 2000). An advantage of this model, and other spontaneously metastatic mammary cancer models, is that the primary tumor can be surgically removed, allowing metastases to develop in the absence of the primary tumor, which corresponds to clinical scenarios where cancer patients relapse with metastatic tumors after the primary tumor has been eliminated. Like their human-derived counterparts, murine cell lines can be phenotypically relatively homogeneous following decades of selection for two-dimensional growth.

The advent of murine cancer organoid models in a wide range of tissues has prompted their adoption for metastasis studies. Normal mouse intestinal organoids from C57BL6 mice engineered in vitro with common oncogenic mutations in APC, KRAS, TP53, and SMAD4 or PIK3CA can be transplanted into the cecum or rectum of recipient syngeneic C57BL6 animals, where they form orthotopic primary tumors that invade, and, over the period of several months, metastasize spontaneously to distant organs including the liver and lungs. Such organoids can also be inoculated into the splenic vein (to enter the portal venous circulation) upon which they seed and colonize the liver (de Sousa e Melo et al. 2017; O’Rourke et al. 2017). These liver metastases could further be used to derive organoids with increased metastatic capacity through rounds of in vivo selection.

MODELING METASTATIC DORMANCY

Of the different stages of metastasis, dormancy is notoriously challenging to study in the clinic, since by definition this implies studying cells that cannot be clinically detected. Mouse models have therefore played a critical role in elucidating mechanisms of both cancer cell–intrinsic and microenvironment-mediated dormancy.

Modeling Cancer Cell Dormancy

In normal physiology, quiescent cells serve an important role in tissue repair (Marescal and Cheeseman 2020). These cells must regulate their entrance and exit from the cell cycle through fine-tuning the balance of pro-proliferative proteins, such as cyclins and cyclin-dependent kinases (CDKs) and CDK inhibitors. In metastatic disease, DCCs can hijack these pathways to enter a nonproliferative state that allows them to reside in foreign tissue until a microenvironment conducive to outgrowth develops. To identify quiescent cells, studies most commonly rely on immunostaining of common cell-cycle-related proteins to differentiate between quiescent and proliferative cells in fixed tissue and cells. Alternative strategies rely on the use of nucleoside analogs, such as EdU or BrdU, that are incorporated in proliferating cells during DNA synthesis in mitotic cells. Cells can be stratified based on the level of nucleoside incorporation with cells that do not incorporate any analogs being classified as nonproliferative or quiescent. Using ki67, pH3, and EdU staining, Basak et al. revealed the molecular mechanisms underlying the transition of quiescent LGR5⁺ cells to hormone-producing enteroendocrine cells (Basak et al. 2017). Similarly, studies have used dye-retention assays, where fluorescent lipophilic dyes are incubated with cancer cells, incorporated into the plasma membrane, and diluted as cells undergo cytokinesis, to isolate quiescent cells (Chung et al. 2017). While these tools have been used to elicit the underlying biology regarding cancer cell quiescence and proliferation, the need for exogenous reagents is a barrier to their incorporation into metastasis mouse models. To circumvent these challenges, studies rely on endogenous fluorescent reporters to interrogate the cell cycle in vivo. Using an H2B-GFP fluorescent reporter, which relies on the accumulation of the GFP-conjugated H2B histones in nonproliferating cells, Brown et al. demonstrated that TGF-β promotes tumor-propagating cells quiescence in squamous cell carcinomas (Brown et al. 2017). More recently, fluorescent protein-based assays have been developed to isolate cells in various phases of the cell cycle, including G₀. In 2008, Sakaue-Sawano etal. (2008) developed the groundbreaking fluorescent ubiquitination-based cell-cycle indicator (FUCCI) reporter system that allows for the fluorescent labeling of cells in various stages of the cell cycle. While the FUCCI system could label cells in various stages of the cell cycle, quiescent cells could not reliably be identified. Oki et al. (2014) developed a similar reporter system that selectively labels quiescent cells, which relies on a mutant p27 protein—a CDK inhibitor. Various versions of these reporter systems have been developed to interrogate the biology of proliferating and quiescent cancer cells (Zerjatke et al. 2017; Oren et al. 2021). Using the p27k⁻ reporter, Sato and colleagues identified a specific quiescent cancer stem cell population in colorectal cancer that relies on the hemidesmosome formation and cell–matrix interactions through the expression of collagen XVII (COL17A1) to induce quiescence in a basement membrane–dependent manner (Ohta et al. 2022). These dormant cancer cells exhibit increased resistance to chemotherapy and could be implicated in cancer relapse. Using the same reporter in an immunocompetent mouse breast cancer model, quiescent cells were found to be uniquely competent in evading elimination by cytotoxic T cells, driven by the remodeling of the tumor microenvironment (TME) into a hypoxic dormant immunosuppressive niche (Baldominos et al. 2022). Together, the use of in vivo quiescence reporters in mouse models has elucidated the bidirectional interactions between quiescent cancer cells and the TME: peritumoral factors promote cancer dormancy, and quiescent cancer cells engineer a dormant niche that shields them against immune attacks and cytotoxic agents. By elucidating these interactions, novel drug targets can be identified to sensitize nonproliferative cells to antimitotic agents, preventing cancer relapse.

Tumor Mass Dormancy and the Micrometastatic Niche

In addition to entering quiescence, growing evidence supports a model in which tumor mass dormancy is a result of complex interactions between DCCs attempting to colonize foreign tissue and a suppressive metastatic niche (Zou 2005). MICs thus maintain a dynamic equilibrium with the antagonistic microenvironment where their outgrowth attempts are thwarted by an antagonistic microenvironment consisting of immune cells, such as T cells, NK cells, and tissue-resident immune cells, such as Kupffer cells in the liver and microglia in the brain. Evidence for the role of immune-induced metastatic dormancy comes from organ donors who have had undetectable metastatic disease and passed their malignancies to immunosuppressed organ recipients (Chapman et al. 2013; Amara et al. 2021). These data underscore the role of the TME in maintaining the dormancy of micrometastases.

Using MDA-MB-231 triple-negative breast cancer cells, Ghajar et al. (2013) leveraged a spontaneous metastasis model, where they transplanted cancer cells in the mammary glands of immune-compromised mice, and an intracardiac metastasis model to study the interaction between the perivascular niche and cancer cells. They established the role of endothelial-derived thrombospondin-1 in maintaining cancer cell quiescence, an interaction that is lost in the presence of newly formed neovasculature. To further dissect the role of the metastatic niche in maintaining metastatic dormancy, the Ghajar group employed two-photon intravital imaging with an intracardiac injection mouse model to visualize the interactions between the perivascular niche and cancer cells (Dai et al. 2021). They demonstrated the role of astrocyte-derived laminin-211 in maintaining cancer cell dormancy by modulating the Hippo pathway. Correia et al. (2021) injected breast cancer cells into mouse mammary glands, followed by resections of the primary tumors, to establish a complex network between NK, hepatic stellate, and cancer cells that maintain cancer dormancy through CXC-chemokine receptor 4 (CXCR4), CXCL12, and IFN-γ signaling. Fluegen et al. (2017) further showed the role of the microenvironment in mediating metastatic dormancy using transplantation, GEMMs, and intravital imaging to analyze the role of hypoxia in determining cancer cell proliferation. By combining a novel in situ induction nano-intravital device (iNANIVID) imaging with fluorescent cell labeling, they demonstrated the role of NR2F1 and HIF-1α in modulating the cell cycle and therapy resistance. Using various approaches to model metastasis in mice, these studies reveal the complex interplay between the TME and cancer cells that give rise to nonproliferative cancer cells, which can resist therapy and induce cancer relapse in a permissive environment.

Dormancy and Immune Evasion

MICs undergo significant attrition as they attempt to colonize distant organs, largely due to elimination by the immune system (Garner and de Visser 2020). Depletion of T and NK cells, as well as other immune cell types using depleting antibodies injected into mice, or by transplanting cancer cells into mice engineered to lack distinct immune subsets, leads to rapid metastatic outgrowth (Eyles et al. 2010; Romero et al. 2014; Malladi et al. 2016; Laughney et al. 2020). Such immune cell-depletion approaches have become powerful tools for modeling immune-mediated dormancy. Using an intracardiac injection model, Malladi et al. (2016) were able to show that cancer cells hijack the dormancy program to self-impose a slow-cycling state through autocrine signaling of DKK1, which allows them to evade surveilling NK cells (Malladi et al. 2016). Immune cells also apply a selective pressure to enrich dormant DCCs, which down-regulate immune recognition programs. By injecting murine-derived pancreatic ductal adenocarcinoma cells into the spleens of preimmunized mice, Pommier et al. showed that cells undergoing unresolved endoplasmic reticulum stress enter a nonproliferative immunosuppressed state (Hingorani et al. 2005; Pommier et al. 2018). Dormant DCCs have also been shown to suppress immune cell function by creating unfavorable hypoxic conditions for T-cell function (Baldominos et al. 2022).

Studies have shown that a disruption of the homeostatic equilibrium that maintains tumor dormancy promotes metastatic outgrowth. De Cock et al. (2016) leveraged lipopolysaccharide (LPS), combined with a tail-vein injection model, to induce inflammation in the lung, which, in turn, led to metastatic outgrowth. Using a similar inflammation model, Albrengues et al. (2018) showed that neutrophil extracellular traps (NETs) converted dormant cells to aggressive metastatic cells in response to LPS-mediated inflammation.

Another form of tumor mass dormancy relates to the rate-limiting need of growing tumors to induce and maintain a source of nutrients and oxygen to support their outgrowth, termed angiogenic dormancy. Neoangiogenesis, the development of new blood vessels from existing vasculature, has long been a hallmark of cancer progression (Folkman et al. 1989; Hanahan and Weinberg 2000). In their groundbreaking work, Holmgren et al. (1995) further implicated neoangiogenesis as a driver of tumor dormancy. Through subcutaneous inoculation of Lewis lung carcinoma cells, followed by surgical resection of the primary tumors, they demonstrated that lung-metastatic lesions were undergoing rapid proliferation, and, as the lesions outstripped their blood supply, cancer cells underwent high rates of cell death (Holmgren et al. 1995). These dormant lesions must undergo an angiogenic switch, transition from nonangiogenic to angiogenic (e.g., through the expression of HIF-1α), to develop into macrometastases (Hanahan and Folkman 1996; Carmeliet et al. 1998; Bergers and Benjamin 2003; Naumov et al. 2006).

MODELING METASTASIS IN VIVO REVEALS CONTEXT-SPECIFIC TUMOR–TME INTERACTIONS

Both human-in-mouse and mouse-in-mouse approaches have yielded powerful insights into the molecular mechanisms of tumor–TME cross talk during each step of the metastatic cascade, which we have recently reviewed (Massagué and Ganesh 2021; Gerstberger et al. 2023). Importantly, both the tumor cells and specific cells of the TME can be selectively perturbed using lineage-specific Cre drivers in the tumor or host or transplantation of genetically engineered cancer cells, often in a temporospatially inducible manner. The availability of these sophisticated genetic approaches has enabled deconvolution of the relative contributions of discrete cancer and stromal cell populations to tumor progression. Further, modern in vivo lineage-tracing tools have enabled quantification of clonal dynamics and evolutionary bottlenecks during metastasis at unprecedented resolution (Rogers et al. 2017; Murray et al. 2019; Karras et al. 2022; Yang et al. 2022). Such insights could not be gained through analysis of clinical samples alone.

Disseminating MICs must invade through basement membrane, intravasate through blood vessels or lymphatics, withstand the stress of loss of cell-to-cell contact and mechanical stress of circulation, and, ultimately, extravasate to the target distant organ. GEMMs and transplantation models in which cancer cells are injected into a primary tumor site have been powerfully used to investigate MIC intravasation, while direct inoculation of cancer cells into circulation has revealed mechanisms of extravasation and distant seeding. For example, hypoxia in the primary tumor has been shown to promote dissemination through promoting tumor cell intravasation. Hypoxic stress-mediated activation of HIF-1α led to increased expression of the L1CAM and CXCR4 on cancer cells, promoting intravasation (Zhang et al. 2012). Studies using mammary fatpad orthotopic transplantation models revealed the interplay between metabolic regulation and cancer cell intravasation. Down-regulation of serine synthesis through the loss of phosphoglycerate dehydrogenase (PHGDH) reduces tumor growth while promoting cell dissemination. Metabolic stress has been further linked to increased levels of CTCs and CTC clusters via the response to insulin treatment (Rossi et al. 2022). Such tissue context-dependent intratumoral metabolic heterogeneity could not be effectively studied using in vitro systems.

Inoculation of cancer cells into the circulation has revealed that cancer cells can extravasate to and seed distant organs through modulating the expression of cell-adhesion proteins and secreting signaling molecules, which promotes vascular permeability. Using an intracardiac injection model, breast cancer CTCs were revealed to promote the expression of α2,6-sialyltransferase ST6GALNAC5, whose expression is normally restricted to the brain, to enhance cancer cell adhesion to brain endothelial cells and traverse the blood-brain barrier (Bos et al. 2009). Using a tail-vein inoculation model, a novel role of CTC chemokine release was discovered, where circulating CRC cell-derived C-C motif chemokine ligand 2 (CCL2) binds to C-C motif chemokine receptor 2 (CCR2) on the endothelial cell surface. CTC-mediated endothelial cell CCR2 activation promotes JAK2-Stat5 and p38MAPK signaling in endothelial cells, increasing vascular permeability and CTC extravasation (Wolf et al. 2012).

To model late-stage metastatic disease, cancer cells are often transplanted directly into the metastatic site or inoculated into the capillary beds of the target organ. While flooding the organ with cancer cells may not accurately recapitulate clinical progression, these strategies offer high penetrance and allow for the study of overt metastatic tumors. Such approaches have revealed that metastatic colonization relies on both cancer cell–intrinsic and microenvironment-dependent mechanisms. Using intracranial xenograft models, breast cancer brain-metastatic tumors have been shown to rely on HIF-1α-mediated signaling to promote proliferation (Ebright et al. 2020). Mex3a⁺ CRC cells were shown to be chemoresistant persister cells capable of tumor regeneration after ablation through the use of GEMM and intrasplenic injection models (Álvarez-Varela et al. 2022). To examine the cross talk between cancer cells and their microenvironment, breast cancer cells were injected intracranially, and astrocyte-derived micro-RNA exosomes were shown to suppress PTEN expression in cancer cells to promote metastatic colonization (Ma et al. 2021). Macrometastases derived from intrasplenic injections of cancer cells revealed the role of peritumoral YAP/TAZ signaling in regulating metastatic colonization (Moya et al. 2019).

CONCLUDING REMARKS AND FUTURE DIRECTIONS

There is a plethora of approaches that can be used to study all stages of metastatic disease, including cancer dormancy. The model chosen for any study must account for the variables that accompany any scientific investigation: the choice of starting biological material (cell lines, spheroids, or organoids; human- or murine-derived), the mouse model (GEMM, transplantation, or injection models in immunocompetent or immunocompromised mice), and the stage of metastasis being investigated (dissemination, dormancy, or outgrowth). There is no single model that can recapitulate disease with complete fidelity, but combining multiple models, coupled with in vitro validation and interrogation of patient-derived biospecimens, has provided a paradigm for rigorous scientific discovery.

A major challenge in mouse modeling of cancer lies in the inherent limitation of cross-species interactions, where human-derived cancer cells must, almost exclusively, be studied in the partial or complete absence of a competent immune system. On the other hand, while mouse-in-mouse models retain immunocompetence, differences in human and murine cancer cells and immune cells may limit clinical translation. One future approach to at least partially overcoming this limitation may be the development of humanized mouse models: immunocompromised mouse models that have been engrafted with human immune cells (Guil-Luna et al. 2021). These models allow for the study of the interactions of metastatic tumors with the human immune system; however, these models often suffer from graft-versus-host disease, limiting the experimental window to a few weeks. Recently, aging has been implicated in metastatic progression (Fane et al. 2022). It has long been known that aged and younger tissues have distinct microenvironments; combined with the knowledge that cancer risks increase with aging, many lines of investigation have begun to study cancer in aged mice to more faithfully model human disease. Future studies will reveal the utility of such approaches in improving the fidelity of modeling metastasis. It is our expectation that the use of matched patient-derived primary and metastatic tumors, combined with advances in computational biology strategies such as single-cell RNA sequencing and spatial-resolved transcriptomics and lineage tracing strategies that map the evolution of metastatic disease, will not only reveal the properties that make metastatic cancer a uniquely challenging disease but will pave the way to improved anticancer therapeutics.