Over the past decade, there has been a growing drive across the biomedical research sector to reduce the use of animals, particularly mice, and to develop in vitro models that better replicate the disease observed in patients.

This shift is certainly evident in oncology drug discovery – a field where the later stages of preclinical development heavily rely on the use of animal models to evaluate the safety and efficacy of new treatments.

The shift in policy and regulatory guidance from animal-based to in vitro human-based approaches – both in the UK and the US – has been motivated in large part by the low success rate of cancer drugs implemented in the clinic. Despite extensive testing both in vitro and in animal models and promising pre-clinical data, over 90% of drug candidates fail in clinical trials without ever reaching patients.

The biggest contributors to these failures are the fundamental biological differences between human patients and model organisms that don’t fully capture the complexity of human diseases. This limits our ability to predict both toxicity and clinical efficacy, or indeed the lack of it. As a result, it remains very challenging to develop safe therapeutics based on current in vivo studies.

An accurate reflection of human biology

Increasingly, regulators, funders and researchers recognise that improving translational success requires models that more accurately reflect human biology. Human-derived models are promising and offer the opportunity to complement and, in some cases, replace animal studies while generating more clinically relevant data.

Until recently however, the availability of alternative non-animal models (NAM) was limited mainly to human cancer cell lines derived from a small set of tumours. Since 1951 when the first human cell line – HeLa cells – was established, these models have been extensively used in cancer research. However, their suitability as cancer models is severely limited. They capture only part of the phenotypic complexity and heterogeneity of human tumours, have limited clinical annotation and due to their continuous passaging have diverged significantly from the original tumours from which they were derived.

It was in this context of scarce availability of cancer models that the Human Cancer Model Initiative (HCMI) was established in 2016.

The HCMI formed as an international collaboration between the National Cancer Institute (NCI), part of the US National Institutes of Health, Cancer Research UK, the Wellcome Sanger Institute and the foundation Hubrecht Organoid Technology (now part of Merck) in the Netherlands. The goal of the consortium was to address the deficiencies in available human cancer models and provide researchers with access to a comprehensive resource of clinically and molecularly annotated patient-derived in vitro models. These models span diverse cancers and are made available to the research community through American Type Culture Collection (ATCC).

The consortium included several research centres and hospital networks across four different countries (USA, UK, Italy and the Netherlands). Patients were consented for the study at surgery and tissue samples – including blood, tumour tissue and, where possible, matched normal tissue – were collected and transferred to Cancer Model Development Centres (CMDCs).

A valuable resource

The results of this decade-long effort have been significant. Between 2016 and 2021, over 2,500 patients were enrolled into the HCMI resulting in the generation of 665 patient-derived cancer models representing 25 distinct tumour types. These include organoids, neurospheres and cell lines derived from both primary tumours and metastatic lesions, creating one of the largest and most diverse collections of next-generation cancer models available to the research community.

What makes this compendium of models extremely valuable is their extensive molecular and phenotypic characterisation and clinical diversity. Each model is associated with comprehensive genomic, transcriptomic, epigenomic and clinical information linking directly patients’ data with the molecular characteristics of the tumours found in patients. Crucially, analyses conducted by the consortium demonstrated that these models retain many of the key biological features of the original tumours even after growing in laboratory conditions. In fact, it was shown that there was over 95% concordance observed for genetic alterations and epigenetic features and over 90% similarity in RNA expression to the corresponding tumour.