Owkin’s lab is at the forefront of building 3D organoids

Before a candidate drug is approved for trial in patients, its effects on humans must first be understood in the laboratory. So we create model systems to mimic the human body. But creating accurate model systems is one of the main challenges of biomedical science. 

Historically, preclinical researchers have approximated human biology using in vitro models and mouse models. In vitro models are cells and tissue, removed from the body, and grown flat on a plate (‘2D models’). Mouse models are mice bred to recapitulate human disease in living mice. Studies conducted with such models are an essential part of the preclinical studies required for Investigational New Drug (IND) approval of novel drugs.

But these models are flawed. For example, in vitro cell lines lack the tissue architecture, molecular diversity, and biological context of real human physiology. Mouse models have improved over the last decades, with the emergence of engineered mouse models and patient-derived xenografts - but these models suffer from a strong selection mechanism that makes them different from the tumor of origin.¹

So testing drugs in these models can give misleading results. Each fails to capture full tumor heterogeneity. This can partly explain why 90% of new oncology drugs fail when they reach clinical trials - these models lack translability into humans.

The pharmaceutical and biotech industries desperately want to increase the probability of success of clinical trials. Now, in the last ten years, a new kind of model has been developed that brings the laboratory closer to the patient's physiology - and could increase the chance of new drugs succeeding at trial.

The emergence of patient-derived organoids

Organoids are 3-dimensional biological structures grown in vitro from a patient’s tumor cells or stem cells. As they grow, they can recapitulate the shape of the organ from which their tissue was taken - hence ‘organoid’. These powerful and faithful models retain key features of human physiology that previous models have not, including:

• The cellular and molecular heterogeneity of the primary tissues.²
• The supracellular structures and functions of the primary tissues³(collections of cells & larger structures).
• The genetic background (e.g. mutational signatures).⁴
• High correlation with drug response observed in clinical trials.⁵

Organoids are revolutionizing research and pharmaceutical companies have taken note. Industry has started acquiring organoid companies (e.g. Merck acquired HUB Organoids at the beginning of 2025⁶) and developing internal organoid capabilities (e.g. Roche⁷ and Novartis⁸). Owkin has set up its own organoid lab, growing organoids models of several disease indications. These organoids have mutational profiles and specific indication subtypes with high unmet need; they are derived from patient samples, delivered through our worldwide network of research partners.

This turn towards organoids is mirrored in the regulatory stance. The FDA has approved models including organoids for drug testing when preclinical animal models do not have sufficient relevance.⁹ Moreover, the FDA recently updated its guidelines, lifting the need to use mouse models in preclinical studies for antibody-based drugs. They now encourage researchers to perform work on patient-derived models and to use AI to better predict drug efficacy.¹⁰

Several clinical trials are ongoing that incorporate organoid models, or that evaluate a drug for which preclinical studies used mostly organoid.^{11 12} The outcome of these clinical trials will set a new gold standard for the use of organoids to expedite preclinical work. The shifting dynamics of the market suggest that organoids will become a central part of biological modelling.

A work in progress

But the technology is not without drawbacks. From a technical perspective, their culturing protocol strongly differs from that of 2D cell lines - they require live tissues obtained from surgeries to be sent to a lab. At Owkin we’ve found that partner relationships are crucial to obtain these tissues - not only do you need to be connected to partners that are producing the live tissue required for the model, but the goodwill and understanding to work through knotty legal agreements and complex logistics. For example, our deep existing relationships with a broad range of institutions has enabled us to expand our network to four centers in less than a year. This allows us to quickly receive regular live samples.

The skillsets required to be able to generate and culture organoids in a long-term fashion are also rare. As a growing field, organoid specialists are in demand. Culture reagents and material used for organoid generation and culture are also expensive - so there needs to be investment and an appetite for the long term benefits that the technology will bring.

One major drawback of organoids remains the fact that culturing them requires a partial dissociation of the tumor of origin. Hence these models will still fail at recapitulating the spatial architecture of tumors at a large scale (like blood-vessels, tumour-stroma edge etc). This is where organoid data can be combined with other patient data that retains these larger architectures. Owkin’s MOSAIC is a good example of this - the largest spatial omics dataset in cancer, this can provide the broader tumor environment picture, in parallel with the causal and drug response data from specific organoid modelling.

The organoids of the future

Organoids are providing researchers with rich biological data, closer to physiological reality. This will help close the translational gap and boost the probability of success for drugs entering clinical trials. 

But such rich, complex, even multimodal data are also perfect as substrate to train AI models for use in drug discovery and development. At Owkin, we believe that a deep understanding of biology is necessary to build this kind of AI. 

To train AI we need large amounts of data. Several datasets describing human cancers are available online, but these can be seen as snapshots of a disease at a certain timepoint. What is needed are huge databases for 3D in vitro models - similar to those built by Recursion in 2D - that will give AI the data needed to uncover useful insights. The organoid data Owkin collects will allow us to map cancer ecosystems over time as well as investigate genetic and chemical perturbations to understand causality in disease. All this will dramatically improve our AI models’ power and soon aid in the design of organoid experiments themselves.

As AI advances, we move towards the possibility of digital twins - in silico models of individual patients - that will allow us to test treatments on patients artificially to predict their real-life response. Organoids are a form of in vitro twin - allowing us to test drugs on model systems that closely recapitulate the biology of individual patients, capturing a patient’s unique physiology.

Deeper understanding of physiologically relevant biology through organoids, improved understanding of patient data (from modalities such as spatial omics), and next generation agentic artificial intelligence are all changing what it is possible to study. Together, this will bring our understanding of disease closer to the reality of what is happening in humans, close the translational gap, and increase the probability of success of clinical trials to bring more successful drugs to patients.

‍

[1] Ben-David U, Patient-derived xenografts undergo mouse-specific tumor evolution (2017) Nature Genetics.

[2] Sachs N, A Living Biobank of Breast Cancer Organoids Captures Disease Heterogeneity (2018) Cell.

[3] Sato T, Single Lgr5 stem cells build crypt-villus structures in vitro without a mesenchymal niche (2009) Nature.

[4] Lee SH, Tumor Evolution and Drug Response in Patient-Derived Organoid Models of Bladder Cancer (2018) Cell.

[5] Vlachogiannis G, Patient-derived organoids model treatment response of metastatic gastrointestinal cancers (2018) Science.

[6] Press Release: Merck Acquires HUB Organoids Holding B.V., Expands Next-Gen Biology Portfolio

[7] Institute of Human Biology website: https://institutehumanbiology.com/

[8] Novartis website: https://www.novartis.com/research-and-development/research-disease-areas/dax-exploratory-disease-research-novartis

[9] https://www.science.org/content/article/fda-no-longer-needs-require-animal-tests-human-drug-trials

[10] Press release: FDA Announces Plan to Phase Out Animal Testing Requirement for Monoclonal Antibodies and Other Drugs.

[11] Herpers B, Functional patient-derived organoid screenings identify MCLA-158 as a therapeutic EGFR × LGR5 bispecific antibody with efficacy in epithelial tumors (2022) Nature Cancer.

[12] Clinical trial number NCT06496178 evaluating petosemtamab versus standard of care in recurring head and neck cancers.