Fighting the Sixth Decade of the Cancer War with Better Cancer Models

Models of human cancer have played an essential role in cancer research due to the limited availability of human tissue samples that are representative of all stages of cancer progression. As surrogates of primary human cancer, cancer models have helped establish key principles regarding disease etiology, and future investments in cancer models will continue to provide valuable insights regarding facets of cancer pathogenesis that remain obscure. Our advances in determining the mechanistic basis of neoplasia have naturally elicited great expectations that these cancer models should facilitate the rapid development of effective therapeutic strategies for patients. However, the direct extension of preclinical activity to clinical success has been frustratingly infrequent, and this has motivated some to question the predictive relevance of such models. Here, I will reflect upon our prior notable accomplishments and current challenges with cancer models, in an effort to highlight new opportunities for researchers to consider over this sixth decade of our “War on Cancer.”

For the past century, autochthonous cancer models have played a critical role in determining the foundational principles of cancer biology. The pioneering studies of chicken sarcoma by Peyton Rous established early assays that led to the functional description of transforming retroviruses (1), and this work concluded with the molecular and cellular mechanistic work by Harold Varmus and Michael Bishop that identified membrane-localized and truncated SRC as the first human oncogene (2). The production of transgenic “oncomice” by Ralph Brinster, Richard Palmiter, and colleagues (3) confirmed the transforming potential of multiple cellular and viral oncogenes following their insertion into fertilized oocytes. This was closely followed by equally important work that applied gene disruption approaches in embryonic stem (ES) cells pioneered by Bob Weinberg, Allan Bradley, Anton Berns, Neal Copeland, and colleagues (reviewed in ref. 4) to confirm the existence of alleles that suppress tumorigenesis as inferred previously by kindred linkage studies and genome sequencing. Thus, the evidence that cancer occurs due to the acquisition of oncogenic mutations and the loss of tumor suppressor alleles—the veritable “nuts and bolts of cancer”—was initially confirmed using mouse models several decades ago. Subsequently, the focus has shifted toward producing mouse models that most accurately represent many forms of human cancer, and this has been partially accomplished through the development of increasingly sophisticated conditional and inducible genetically engineered mouse models (GEMM) by Jos Jonkers, Mariano Barbacid, and colleagues (reviewed in refs. 4, 5). One class of GEMMs was cleverly developed by Ronald DePinho, Lynda Chin, Gerard Evan, Scott Lowe, and colleagues and uses reversible gene expression to identify a role of certain cancer genes in tumor maintenance (reviewed in ref. 6). Such reversible mouse models reinforce the relevance of therapeutic development around these genes and their effector pathways. Finally, the incorporation of CRISPR/Cas9 and analogous gene editing approaches in ES cells or in somatic tissues was pioneered by Tyler Jacks and colleagues (reviewed in ref. 7) and accelerates the functional assessment of cancer-associated alleles in mice. In the coming decade, many new cancer-associated variants should be evaluated using this panoply of approaches with GEMMs.

The alternative to GEMMs has historically been human tumor tissue transplantation into immune-suppressed rodents. Despite their shortcomings, xenografts have informed our understanding of the tumor microenvironment and cancer metastasis, and have been preclinical therapeutic workhorses in academia and the private sector. Newer xenograft models that are established in situ also afford the ability to evaluate tissue invasion as a primary step in cancer initiation. Syngeneic tumor transplants of murine neoplastic cells afford the ability to engage an intact immune system, and indeed this simpler alternative to GEMMs was used by Max Krummel and James Allison to correctly identify the T-cell immune-activating potential of antibodies directed to CTLA4 (8), where mice harboring syngeneic melanomas exhibited complete responses after a lag of several months.

Another model system that has gained popularity over the past decade is the use of human tumor fragments and derivative outgrowths such as organoids. Organoid technology was advanced and popularized by Toshiro Sato and Hans Clevers to produce cultures of normal and malignant tissues (9). Organoids accelerate both basic and translational cancer research due to the ease of establishing and manipulating organoid cultures. The coculturing of organoids with additional stromal cell types enabled the identification of T-cell clones that reacted to neoantigens in the laboratory of Calvin Kuo (10), and such models are now being used to investigate immune therapy approaches for patients. A pressing translational objective is to determine whether the organoid cultures can be used as predictive models for personalized cancer treatments, similar to bacteriology drug testing, and studies are currently under way to address this.

Although the framework of our current understanding of neoplasia has been broadly assembled, many molecular and cellular aspects of cancer biology remain relatively unexplored or poorly understood. For example, human cancers oftentimes express a litany of mutant oncogene and tumor suppressor gene alleles, and very little is known about the potentially different tropic functions of these specific alleles in tumor progression. Interestingly, an analysis of an allelic series of Kras mutations in GEMMs by the laboratories of Luke Dow (11) and Kevin Haigis (12) demonstrated that the orthologous human mutations in KRAS play distinct roles in activating biochemical cascades and promoting neoplasia in the pancreas, lung, and intestines. These findings motivate similar questions for additional tumor-associated alleles, and the application of gene editing technologies both in ES cells and in vivo can accelerate these studies. Additionally, the relevance of variant germline alleles that are linked to cancer predisposition is an area that has been highlighted by comprehensive whole-genome sequencing in large populations of patients. These alleles are postulated to modify the functions of cancer pathways in neoplastic cells and alter homeostatic mechanisms in nonneoplastic cells to influence cancer progression, and these putative epistatic interactions could be systematically probed in GEMMs. Additional nongenetic mechanisms that may alter gene function in human cancers, such as noncoding RNAs, mRNA editing, differential RNA splicing, mRNA stability, and chromatin regulation, are also areas in which future GEMMs could be deployed to determine the in vivo relevance of these processes in cancer pathogenesis.

Cellular analyses of human tumors and cancer models have revealed that cancer cells recapitulate some aspects of normal tissue development, with progenitor or stem-like cell states existing in a dynamic relationship with more differentiated and proliferative neoplastic cells, as first reported by John Dick (13) and Irving Weissman (reviewed in ref. 14). These stem-like neoplastic cells oftentimes are more resilient when exposed to stressful environments, including hypoxia and nutrient deprivation, and are more refractory to therapeutic challenge—thereby contributing to therapeutic resistance. Investigations of stem-like cells have shown that the selective depletion of these cells reduces tumor progression in GEMMs substantially, and further explorations using additional methods could reveal additional dependencies in these cell states that may be amenable to therapeutic targeting. A related feature of cancer progression applies to quiescent or “dormant” metastatic cancer cells that Douglas Fearon found had evaded T cell–mediated clearance by assuming a mesenchymal fate and downregulating MHC class I (15). Cancer cell dormancy may contribute to the failure of many therapies in patients, and cancer models should be leveraged to more deeply interrogate this feature of cancer cell biology to establish the parameters that promote cancer cell dormancy as well as the escape from dormancy. For example, local inflammation that induces neutrophil activation and NETosis was recently demonstrated by Mikala Egeblad to stimulate dormant metastatic cancer cells to regain proliferative and migratory capacity in pulmonary tissues (16). Indeed, the mechanisms regulating the acquiescence of cancer cell dormancy and its escape represent important goals of cancer models because they would represent potential new therapeutic avenues for patients.

Intratumoral heterogeneity extends beyond the stem-like and dormant cell states that are epigenetic in nature, as patients with cancer additionally harbor multiple genetic clones simultaneously due to continued genomic instability and promoted by therapeutic pressure. Indeed, these distinct genetic clones likely contribute to drug resistance in patients, as shown by the seminal finding of Charles Swanton in kidney cancer (17). Early findings by Anton Berns demonstrated that the combination of two distinct cell populations derived from a lung cancer GEMM synergistically promoted metastasis following transplantation and suggested a symbiotic relationship between each (18). Furthermore, GEMMs of pancreatic cancer were used by Ben Stanger to confirm that multiple primary tumor clones contribute to individual metastatic foci, providing a system to investigate whether such mixed metastases are stochastic or symbiotic in nature (19). The ability of GEMMs and other cancer models to explore neoplastic cell heterogeneity and its coevolution with the tumor microenvironment are important goals for the decade ahead.

Despite these notable contributions of cancer models, sadly most patients with cancer still suffer and die of their disease, prompting the concern that the models deployed in biomedical research are inaccurate recapitulations of the human disease. Cancer models typically have undergone a detailed assessment of the histologic, cellular, molecular, and pathophysiologic features of the cognate human condition throughout all stages of tumor progression to benchmark the models prior to dissemination in the research community, and this has been performed for GEMMs that represent multiple human cancers. The comparative assessment of cancer models has led to the observation that many GEMMs harbor neoplastic transcriptomes that are very similar to human cancers and display many additional pathophysiologic features reminiscent of human cancer. Nonetheless, these same GEMMs usually harbor a lower density of secondary genetic alterations that are typical of their human counterparts, potentially due to the shorter time period that the tumors evolve and the lack of additional carcinogen exposures (20). Therefore, such cancer models that possess a reduced diversity of genomic alterations compared with human cancer may provide an unreasonably optimistic assessment of therapeutic responses due to their lack of intratumoral heterogeneity. Likewise, these GEMMs would harbor fewer neoantigens and therefore lack an optimal response to immune therapies. Accordingly, efforts to increase secondary genetic alterations have led to mutagenizing GEMM-derived cell lines to make them more responsive to immune therapy, as expected (21). By extension, the expression of DNA-altering enzymes and inhibition of DNA damage repair genes in future GEMMs could improve this aspect of genetic and cellular diversity and be useful for measuring therapeutic responses to systemic and immunologic therapies.

Comparative assessments between GEMMs and human cancers have also led to new findings about human malignancy. Indeed, some features of human cancers were discovered first in GEMMs and then confirmed in humans, such as the hypovascular nature of pancreatic cancer by Ken Olive and colleagues (22). In addition, the deployment of organoid coculture and GEMMs led to the initial observation of cancer fibroblast subtypes in cancer, and this was later confirmed in human cancers (23). Differences between cancers in our two species have also led to new insights. For example, the glycosylation patterns of human cancers are distinct from GEMMs due to the absence of certain glycosyltransferases in rodents, and the “humanization” of the mouse glycome revealed that aberrant glycosylation triggers inflammation and promotes pancreatic cancer, providing new pathways to explore for therapeutic interrogation in patients (24). In the decade ahead, the ongoing detailed analyses of primary and metastatic tumors by deploying emergent “omic” technologies await conclusion for the most widely used GEMMs, in an effort to complete this important comparative assessment of cancer.

The lack of predictive value between certain preclinical studies and the follow-up clinical trials may alternatively reflect that cancer models have been improperly deployed in preclinical therapeutic studies. A historical shortcoming in therapeutic studies involving cancer models is that the slowing of tumor volume expansion following treatment has been considered a positive response. Any tumor growth in humans undergoing treatment is considered progressive disease, and therefore many therapies that demonstrated only tumor growth inhibition potentially were pushed forward with an incomplete preclinical rationale. In addition, the durability of responses in animal models is oftentimes not extensively studied, which is a critical parameter in oncology patients. Most promising preclinical therapeutic strategies that subsequently fail in patients with cancer are never rigorously studied in a scientific fashion due to the lack of accessible tumor samples or infrastructure support, and this aspect of drug development also deserves serious attention. Indeed, the implementation of proof-of-concept or “phase 0” trials to ascertain tumor drug levels, target engagement, and pharmacodynamic effectiveness that one can pursue in cancer models should be resourced in selected centers to determine whether this can facilitate a more successful translational path.

Finally, these models can be used to establish new paradigms for cancer treatment. For example, the demonstration by Laurence Zitvogel and Guido Kroemer that immunogenic cell death is a central feature of clinical response to all therapeutic modalities (25) challenges our field to develop new therapeutic strategies that selectively target neoplastic cells and alleviate local immune suppression mechanisms, while simultaneously sparing the cells involved in generating an effective immune response. Cancer models should lead the way in investigating this concept in the coming decade. Also, the parallel development of new diagnostic approaches that portend early response in cancer models could facilitate the more rapid assessment of therapeutics in patients, and this is another immediate goal for our field.

Cancer models have provided tremendous value to our basic understanding of human cancer, and will continue to do so over the coming decade. The systematic incorporation of nongenetic processes known to promote human cancer, including inflammation, obesity, stress, and aging, will be important additional goals. In addition, the translational importance of cancer models is currently underappreciated, in part because insufficient attention has been given to pertinent aspects of therapeutic development, and future work should herald new methods to more meaningfully make this transition for the benefit of patients.

D.A. Tuveson reports other from Leap Therapeutics, Mestag Therapeutics, Surface Oncology, Cygnal Therapeutics, ONO, and Fibrogen outside the submitted work; and is President-Elect, AACR, and Chief Scientist, Lustgarten Foundation. No other disclosures were reported.

I apologize for the inability to cite the primary work of many colleagues due to space restrictions. This work was supported by the CSHL Cancer Center and NIH P30CA45508.