The expansion addresses a central limitation of cancer genomics. Sequencing and other molecular methods can catalogue alterations in tumors, yet the clinically important question is often which of those alterations cancer cells actually require to survive. Project Achilles at the Broad Institute and Project DRIVE at Novartis addressed this through large-scale RNA interference screens, while the Cancer Cell Line Encyclopedia linked dependencies in hundreds of cell lines to genomic features. Computational filtering reduced off-target effects, and CRISPR subsequently offered more specific gene inactivation. Together, these resources created DepMap, which connects molecular features with essential genes across thousands of conventional cell lines.

Conventional models nevertheless represent only part of human cancer. Some tumors, including many prostate cancers, adapt poorly to two-dimensional culture. Established cell lines may also lack the treatment history of the patients from whom they originated and retain little of the cellular diversity present within a tumor. The three new studies therefore assembled complementary collections of patient-derived organoids and other next-generation models. In addition to enlarging the number of systems available for perturbation, they asked whether the new models retain tumor identity, can support rigorous genome-wide screens, and expose dependencies missed by conventional cell lines.

The first assessment compared models with their parental tumors and produced encouraging but incomplete results. Across assay-specific subsets, 92% to 98% of Human Cancer Models Initiative models met genomic or transcriptional measures of concordance. Sanger researchers reported strong genomic concordance across 171 organoid–tumor pairs. Broad’s Celligner analysis found that 69% of next-generation models matched their expected lineage, compared with 35% of conventional cell lines. Fewer than 2% of the newer models occupied an undifferentiated transcriptional cluster, whereas more than 25% of conventional cell lines did. These findings indicate that patient-derived models can preserve cancer-associated molecular features substantially better than many long-established lines.

That fidelity has clear practical limits. Patient-derived models are difficult to establish, and their genomes and cellular identities can change in culture. Sanger generated 256 organoids from 907 submitted samples, with low tumor cellularity and cultures that could not be renewed accounting for many failures. HCMI models retained 81% of tumor driver events, but 10% developed whole-genome doubling that was absent from the parental tumor. Sanger also observed whole-genome doubling and chromothripsis more frequently in organoids. In a more specific pattern, extrachromosomal EGFR copy number was higher in the parental tumor in 80% of glioblastoma pairs examined by HCMI.

Cellular state proved even less stable than the presence of a cancer-driving mutation. At single-cell resolution, only three of seven profiled glioblastoma models retained the malignant cell states seen in their parental tumors, while three shifted toward a predominantly mesenchymal state. Broad found that breast cancer organoids sometimes lacked enough ESR1 and ERBB2 expression to reproduce the corresponding dependencies, and none of 12 prostate cancer models retained a recognizable prostate lineage identity by Celligner. The results show why retaining a driver mutation is not enough: a model can preserve a genomic lesion while losing the state that makes a tumor dependent on it, or it can acquire a dependency without faithfully retaining the original cellular program.

Despite these complications, the new models proved suitable for genome-scale experiments. Broad’s organoid screens separated essential from non-essential genes as effectively as conventional screens, while Sanger obtained quality-controlled data from 85% of models in which screens were attempted. These results establish that patient-derived systems can support large perturbations across multiple genes. They do not remove every interpretive problem, however. Dependencies in steroid and isoprenoid biosynthesis may be influenced by culture conditions rather than the original tumor environment. Bulk measurements can also conceal dependencies restricted to particular cellular subpopulations, an issue especially relevant in tumors that contain several distinct states.

The clearest value of the expanded maps lies in their ability to resolve dependencies that broad genomic labels cannot explain. In esophageal and gastric organoids, KRAS amplification predicted dependency on stearoyl-CoA desaturase most strongly, and the finding was validated with two inhibitors. Colorectal organoids carrying KRAS G12 variants remained dependent on EGFR and PTPN11, whereas a KRAS Q61H model resisted EGFR inhibition. Microsatellite instability was likewise insufficient to predict WRN dependency: four organoids lacked that dependency and did not carry the expanded TA repeats associated with it. These are model observations, not clinical efficacy findings, but they demonstrate that the same cancer label can conceal different underlying requirements.

Cellular composition added another dimension to the maps. Low ENO2 expression predicted ENO1 dependency more strongly in patient-derived models than in conventional cell lines. Glial glioblastoma models showed CDK6 dependency associated with CDKN2A loss, while organoids expressing a PDAC-classical and MUC-high program depended more heavily on WNT pathway components. Single-cell profiling of 22 advanced prostate cancer organoids identified seven lineage states that frequently coexisted within individual models, helping explain why bulk gene-depletion measurements can vary. This analysis also implicated the aryl hydrocarbon receptor in plasticity. Dependencies could shift with treatment as well: an esophageal organoid obtained at relapse had lost MYC amplification, showed reduced KRAS and DNMT1 dependencies, gained PSMB5 dependency and exhibited corresponding changes in drug sensitivity.

The central implication is that dependency is not determined by a tumor’s broad genomic category alone. It reflects the particular alteration present, the cellular state expressing that alteration and the treatment history embedded in the model. Patient-derived organoids help expose those combinations, offering a more detailed map of survival requirements than conventional cell lines. The evidence remains preclinical, and the studies do not establish that the new dependencies will produce better drugs or improve patient outcomes. Conventional DepMap screens have nevertheless already motivated clinical development programs for targets such as PRMT5 in MTAP-deleted tumors and WRN in microsatellite-unstable cancers. Whether the added biological resolution of the expanded map can be converted into useful therapies will depend on further validation, and on building models that faithfully preserve the lineage states of common hormonally driven cancers such as breast and prostate cancer.

Subject of Research: Cancer dependency mapping in patient-derived organoid models and organoid-specific genetic vulnerabilities

Article Title: Expanding DepMap: is the juice worth the squeeze?

Article References: Zaidi, S., & Sawyers, C. L. (2026). Expanding DepMap: is the juice worth the squeeze?. Cell Research. https://doi.org/10.1038/s41422-026-01311-y

Image Credits: AI Generated

DOI: 10.1038/s41422-026-01311-y

Keywords: DepMap, cancer dependency maps, patient-derived models, organoids, cancer genomics, genetic vulnerabilities, KRAS, SCD, cellular plasticity, drug discovery, Expanding, juice