How Much of a Tumor’s Original Identity Remains When It Is Brought Into the Lab? NCI Releases 665 Patient-Derived Cancer Models

Cancer drugs that work in a culture dish may not work in the human body. One fundamental challenge is whether experimental models can faithfully reproduce patients’ tumors. The National Cancer Institute’s (NCI) genomic data-sharing platform recently released a large collection of patient-derived models, including 665 organoids, spheroids, and cell lines established by the Human Cancer Models Initiative (HCMI) across 25 cancer types. The resource allows researchers to investigate cancer-causing mechanisms and drug resistance using materials that more closely resemble actual tumors.

The international project’s overall donor cohort comprises 2,780 people. Within the model repository, 522 models are accompanied by more comprehensive clinical data, 153 are derived from rare cancers, and another 71 come from participants of non-European ancestry. These inclusion criteria are intended to address gaps in traditional cancer cell lines, which are weighted toward common cancer types and certain populations. However, actual representation across cancer types and populations remains uneven, and the repository’s scale should not be equated directly with comprehensive coverage of cancer diversity.

The research team generated whole-genome, exome, transcriptome, and DNA methylation data for the models and their source tumors. NCI’s updated analysis covered 421 matched sample sets and found genetic and epigenetic concordance of 97.8% and 95%, respectively. An earlier conference abstract estimated, based on 417 models, that approximately 96% closely reproduced the overall molecular profiles of the original tumors. The scope and metrics of the two analyses are not entirely the same, but both point to the same conclusion: most models preserve important tumor characteristics.

However, the models are not static replicas of patients’ tumors. Single-nucleus or single-cell RNA sequencing showed that the cellular states of some models may change because of the culture medium, the loss of tumor stroma, or the selection of certain cell populations in vitro. In other words, high concordance in genetic mutations does not mean that cellular composition, signaling activity, and the microenvironment remain the same. When comparing drug responses, researchers must account for culture conditions in their interpretation.

Another use of this resource is tracing the marks left by treatment pressure. The repository includes samples collected both before and after treatment, along with treatment response and clinical follow-up information from before and after model generation. Some models retain amplification of oncogenes on extrachromosomal DNA and also display post-treatment mutational signatures. They can be used to study how tumors evolve under drug selection pressure, develop drug resistance, and to screen treatment hypotheses for subsequent validation.

The database provides multimodal molecular data, clinical attributes, and integrated analysis tools, while some sensitive data must be obtained through controlled-access procedures. Its value lies in providing a preclinical research foundation for reproducible comparisons, rather than directly predicting treatment outcomes for individual patients. Any candidate drugs or biomarkers identified through the models must still be validated in independent samples, animal studies, and clinical trials.