Switching Off Genes One by One to Find Tumors’ Survival Lifelines: Large Organoid Biobank Maps Cancer Dependencies

The same type of cancer may rely on entirely different survival mechanisms in different patients. Looking only at which mutations a tumor carries is often insufficient to determine where to attack. Researchers are now growing patient tumors as three-dimensional organoids that can be used repeatedly in experiments, then switching off genes one by one in an effort to directly identify the vulnerabilities that cancer cells truly cannot survive without.

A study published in *Nature* established a research biobank containing 256 patient-derived organoids across five difficult-to-treat cancers. The team conducted large-scale CRISPR screens in 162 of these models. By disabling specific genes and observing which changes impeded organoid growth or caused cancer cell death, the researchers identified thousands of “survival dependencies.” These functional experiments ask not which abnormalities a tumor possesses, but which gene activities remain indispensable to it.

The researchers then compared the screening results with the models’ genomic, molecular, and clinical data, identifying 1,733 associations between tumor vulnerabilities and specific features. These links can be used to formulate more precise target hypotheses—for example, first defining a potentially sensitive group of tumors based on a particular mutation or molecular characteristic, then testing whether blocking the corresponding dependency has therapeutic potential. However, an association does not itself mean that an applicable drug already exists, nor can it prove that patients will necessarily benefit after receiving treatment.

These models were derived from patient biopsy or surgically resected tissue and were expanded, cryopreserved, and genetically validated for use as three-dimensional culture systems that can be updated over the long term. The complete resource publicly described by the Wellcome Sanger Institute currently lists 281 clinically annotated organoids covering colorectal, esophageal, ovarian, pancreatic, gastric, and mesothelioma cancers. This number and range of cancer types are broader than the 256 models analyzed in the paper, reflecting that the biobank resource and the samples included in this study are not entirely identical.

The accompanying data are also continuing to be expanded. In March 2026, Cell Model Passports extended mutation and copy-number variation data to 256 models, with expression data covering 255. In June, it added downloadable whole-genome sequencing data for 176 paired organoid and original-tumor samples. Individual models can also be linked to raw data in the European Genome-phenome Archive, while some models are available to researchers through the Human Cancer Models Initiative system, allowing other teams to repeat analyses, select models, and test candidate targets.

Organoids remain simplified in vitro systems. Although they can preserve many of the genetic characteristics and heterogeneity of patient tumors, they may not fully reproduce immune cells, blood vessels, stroma, drug metabolism, or toxicity in the human body. Long-term culture may also favor cell populations adapted to experimental conditions. In addition, the CRISPR screens covered only some of the models in the biobank, and whether robust conclusions can be reached for less common molecular subtypes will still depend on expanding the sample set. The near-term value of this map therefore lies in narrowing the range of candidate targets. Advancing them into treatments will still require stepwise validation in independent models, animal studies, and clinical trials.