The local microenvironment matters in preclinical basic and translational studies of cancer immunology and immunotherapy

Preclinical models are important tools for studying the beneficial and harmful effects of novel therapies, alone or in combination, and help guide their translation into human clinical studies. Because clinical trials can cost millions of dollars, and only a fraction of therapies proven effective in animal models are also effective in humans, it is of utmost importance to use relevant preclinical cancer models. For instance, it is crucial that the experimental techniques do not unnaturally influence the biology of the disease (Grzelak et al., 2022) and that the experiments are performed in a way that ensures rigor and reproducibility (Day et al., 2022). These Cancer Cell letters provide specific guidance on these points.

Equally important, we need to ensure that preclinical cancer studies are clinically relevant and thus increase probability of successful translation to the clinic. We posit that modeling tumors in their natural orthotopic microenvironment improves the relevance of preclinical animal studies of cancer progression and treatment (Figure S1A). For example, in mice, colon cancers should be grown in the colonic wall to model primary colon tumors, or in the liver or lungs to model colon cancer metastasis. In our laboratories, we have been examining the role of the local microenvironment in tumor biology and its response to various therapies and have repeatedly shown that the local microenvironment governs not only tumor biology but also therapeutic response (Fukumura et al., 1997; Ho et al., 2021; Hobbs et al., 1998; Kodack et al., 2017).

We have shown that the same human colon cancer cells, when grown under the skin of nude mice versus in the liver, developed distinct angiogenesis and microcirculation characteristics (Fukumura et al., 1997). We have also shown that mammary carcinomas and gliomas developed distinct transvascular transport properties when grown in the skin versus in the brain microenvironment of mice (Hobbs et al., 1998). These host-tissue-driven microenvironmental differences influenced tumor response to both low- and high-molecular-weight-targeted therapeutics. For example, the same luminal breast cancer cells, when grown in the mammary fat pad versus the brain microenvironment, developed distinct HER3 signaling and responded differently to PI3K inhibition, even though the drug reached effective concentration in both microenvironments (Kodack et al., 2017). This study provided a mechanistic basis for why breast cancer brain metastases are more resistant to certain therapies than primary breast cancer in patients and revealed strategies to overcome this resistance—confirmed in a clinical trial.

Moreover, in a study of mismatch-repair-proficient colorectal cancer (pMMR CRC), we showed that treatment with immune checkpoint blockers (ICBs) such as anti-PD-1 and anti-CTLA-4 resulted in dramatically different outcomes when tumors were implanted in the flank, colon, or liver (Ho et al., 2021). When pMMR CRC cells were grown subcutaneously, they responded to ICBs. However, when the same pMMR CRC cells were implanted in their natural microenvironment, for example, in the colonic wall as primary tumors, or in the liver as hepatic metastases, they did not respond to this treatment (Figure S1B). Thus, since human pMMR CRCs are refractory to immunotherapy, only the preclinical models using orthotopic tumors recapitulated the clinical observations, while the subcutaneous tumors incorrectly suggested the efficacy of ICBs against this cancer. Of note, orthotopic liver metastases showed significantly decreased infiltration of antitumor immune cells, including activated CD8⁺ T cells and dendritic cells, compared to subcutaneous tumors, which may explain the differences in treatment efficacy against tumors growing at these different sites.

These discrepant findings are not unique. For example, Horton et al. showed that lung cancer cells grown in the lung microenvironment lack CD8⁺ T cells with an effector phenotype and are resistant to treatment with ICBs, compared to the same lung cancer cells grown in the skin (Horton et al., 2021). Similarly, Jiao et al. showed that prostate cancer cells grown as bone metastases have distinct immune infiltrates and respond differently to ICBs, compared to the same cells grown in the skin (Jiao et al., 2019). Systemic immunosuppression—caused, for example, by growing a tumor in the liver—can affect a companion/secondary tumor’s immune response no matter where it is implanted (Yu et al., 2021). Therefore, both the tumor immune context and the response to immunotherapy are defined, at least in part, by the tissue in which the cancer cells are grown. Tissue variables that may affect treatment response include not only the various cellular components of the tumor microenvironment (immune cells, endothelial cells, fibroblasts) but also specific pH and oxygen levels, metabolites, osmotic pressure, mechanical forces, and microbiota. Unfortunately, most preclinical cancer studies aimed at identifying new therapeutic approaches use mouse models in which non-skin cancer cells are ectopically injected into the flank.

Given the above, what are the potential approaches moving forward? We propose that, to provide better information for human clinical trials, preclinical studies should include orthotopic tumor models. These can be animal models in which tumor cell lines are implanted at their orthotopic site or genetically engineered mouse models (GEMMs). GEMMs are attractive options because tumors in these mice naturally develop in their native tumor microenvironment and have shown value in certain experimental settings. For example, a STING agonist is ineffective against spontaneous mammary tumors arising in GEMMs, consistent with clinical data, whereas it rejects mammary tumors transplanted orthotopically into the fat pad (Guerin et al., 2019). However, orthotopic tumor grafts are particularly useful because they are much faster and easier to handle in large cohorts and may carry more mutations than those in GEMMs. Indeed, tumors in GEMMs may not capture the mutational burdens seen in human tumors. This may be limiting when testing new therapeutic options such as ICBs, since tumor mutational burden may be associated with their efficacy (Figure S1A).

Finally, in some disease contexts, even orthotopically grown tumors may not recapitulate clinical response and need further refinement. For example, the widely used glioblastoma model GL261 responded to ICBs when grown orthotopically in mice, whereas human glioblastoma is refractory to the same treatment (Reardon et al., 2016). A solution to this problem may be establishing tumor variants that are refractory to ICBs (Amoozgar et al., 2021) as more relevant cancer models for future translational studies. Regardless of the choice of preclinical models, it is important that preclinical results are validated using patient data, if available, to ensure the clinical relevance to the human disease.

We write this article not to point out weaknesses in published work but to strive to establish some guidance for future preclinical studies to develop new therapeutic approaches in academic, governmental, and industrial settings. The data obtained in the optimized preclinical models will not necessarily guarantee clinical success, but they will more likely provide clinically relevant information and reduce the number of clinical failures and their associated costs.