Why preclinical cancer models still fails in the clinic?  

Why preclinical cancer models still fail in the clinic?  

February 2023 l Olivier DUCHAMP, Head of Translational Pharmacology at Oncodesign Services

The failure rate of new cancer medicines in the clinic remains extremely high – maybe 95% of compounds tested in clinical trials will never become a medicine. Yet in the discovery lab and preclinical work, all will have looked promising. At the heart of the problem is the fact early testing might be good at showing that a compound can kill cancer cells in a preclinical model system, but these models are poor at pinpointing which are likely to be similarly effective when given to humans.

Should we continue to push anti-cancer drug candidates into the clinic knowing that 95% of them will only bring side effects to patients involved in early phases of clinical trials ? Should we continue to spend so much money for so little clinical benefit ? Should we continue to select drug candidates with the same preclinical trial systems without changing anything ?

For sure at Oncodesign Services, we do not want to continue in this way. That’s why we are always pushing alternative models and technological innovations into the preclinical drug development process in the hope of reversing this virtuous trend, to limit failures in late-stage studies and to minimize potential for harm in clinical trials.

The unique preclinical cancer model does not exist.

We still have to play with panels of models, each having its own limitations, each patient having its own cancer characteristics. Lots of research papers are based on animal models, highlighting that these models still play a crucial role in translational medicine, even in a historical moment in which the need to find alternative methodologies is increasingly pressing.

For decades, we have relied on in vivo models where tumors are engrafted in mice or rats. To engraft human tumors in mice or rats, the animals must be immunodeficient. Yet we know the immune system plays an important role in cancer, and numerous drugs targeting the human immune system are already available to treat cancer. As these xenografted rodents have poor immune systems, a critical part of the model is missing.

To counter this, human immune cells can be re-introduced into mice bearing human tumors to better mimic the clinical reality and to make them a little bit more predictive of what might happen in the clinic.

Combining these “immune-humanized” mice with patient-derived xenograft (PDX) tumors taken directly from the patient rather than historical in vitro cultured cancer cells should clearly help the identifications of patients who would be more responsive to a novel treatment¹. While this maintains the tumor’s histology and genomic situation, it’s still not enough, as many tumor micro-environment components are still missing.

Can we continue to humanize the animals ?

Giving them human immune cells as well as an engrafted human tumor, this makes an even more realistic but complex model. We demonstrated that a “hot” immune-infiltrated patient tumor remain “hot” when engrafted in immune-humanized mice and inversly a “cold” immune-infiltrated patient tumor remain “cold” when engrafted in immune-humanized mice. These results demonstrated that the human tumor cells are able to educate and drive the human immune cells also in a murine host. Additionally, we demonstrated that the tumor itself or the aggressive chemotherapies are modifying the gut microbiota (with bacteria as most numerous members) of patients or tumor-bearing mice, and that these modifications clearly impact the response of the tumors to cancer immune-therapies (e.g. anti-PD1). This tumor sensitivity to immune checkpoint inhibitors should be restored by re-engrafting human beneficial bacteria in mice.

We are also able to humanize the liver of mice to better predict drug metabolism and potential drug toxicities, and put all these human compartments in one animal. However, although this would be more ‘human’, there is still another problem, as yet, we still don’t know how to humanize the cancer-associated fibroblasts (CAF), a component representing 10 to 80% of the tumor stromal compartment. So, when patient-derived tumors are engrafted in a mouse, the human fibroblasts are rejected and replaced by murine ones. The gulf between rodent and human still requires closing.

What about the development of novel drug candidates without considering the cancer metastasis process ?

In reality, most patients who die of cancer will have developed metastases. If metastasis can be prevented, cancer would be far easier to treat, and even cure. But vanishingly few of our models are effective at identifying a tumor’s metastatic potential. The key to metastasis seems to lie in the interaction of the tumor cells with its microenvironment including the immune cells, the fibroblast, the host natural organ cells, the microbiota…

In syngeneic models, where the mouse still has a complete microenvironment with the same origin of the tumor cells but the tumor is not human, it is far more likely to metastasize than in the xenogeneic models where the tumor cells and the microenvironment are from different origin.

At Oncodesign Services, we also know that rat tumor models are better predictive pharmacological models as compared to mouse models. But many drug makers are still reluctant, incriminating the size of animals that should impact the quantity of compounds and the budget needed for the preclinical development.

During the next decade, we probably will have to change or at least to reconsider the paradigm of the preclinical oncology models, in the objective of increasing our chance of clinical success and benefits for the patients. Cancer is now generally accepted to be an evolutionary and ecological process with complex interactions between tumor cells and their environment sharing many similarities with organismal evolution, and is typically scrutinized as a pathological process characterized by chromosomal aberrations and clonal expansion subject to stochastic Darwinian selection within adaptive cellular ecosystems.

Developing tumor models that better reflect human biology remains a work in progress.

It’s an important problem to solve if we are to become better at predicting the likelihood of a compound becoming an effective anticancer drug, and reduce the huge attrition rate of molecules in the clinic.

Since 1995, Oncodesign Services has been developing models from chemo-induced syngeneic to xenogeneic including genetically-modified, immune-humanized and PDX in vivo models. Our extensive knowledge covers a wide in vivo and in vitro pharmacological scope. More recently, we developed in vitro human tumor organoids, or “tumor-on-chip”, that are now becoming robust system to evaluate anti-cancer drug efficacy. In a recent publication, Wang et al.² illustrated enormous value of patient lung cancer organoids as a useful in vitro system for predicting and personalizing the treatment of patients with lung cancer. All these models are more and more robust and complex, but taking alone no one is realistically predicting the clinical benefits of a novel drug candidate.

At Oncodesign Services, we are thinking that a combination of models is always necessary to allow the selection of better drug candidates and a better picture of the clinical predictivity, as demonstrated by Hoare et al.³ presented for the first time the complementarity of 3 different types of preclinical models of pancreatic adenocarcinoma (in vivo PDX, ex vivo organoids and in vitro primary tumor cells from the same patients). That’s why, we have been investing for 30 years to build a comprehensive and holistic preclinical platform with a large panel of technologies around our cancer models, including proteomics, genomics and pharmaco-imaging.

Now, we hope that the implementation of data driven “artificial intelligence” technologies will help us to better characterized human-like models and to analyse these complex combinations of models to avoid misleading conclusions and consequently the overlook of clinical relevant features. Always thinking of the patients who are impatiently awaiting the solution that will cure their cancer. It is our common responsibility.

About the author

This blog post was written by Olivier Duchamp, head of translational pharmacology department at Oncodesign Services. Olivier was present at the creation of the company and contributed largely to its development and innovation in preclinical models and pharmaco-imaging.

Interested in developing a suitable preclinical model for your research, we would love to discuss about it. Get in touch via the contact form here.

KNOW MORE ABOUT TUMOR MODELS

References : 

1. Julien et al., Clin Cancer Res, 2012 
2. Cell report, 2023
3. Cancer, 2021