The study, led by Kasumi Ota and Yasushi Okuno of Kyoto University’s Graduate School of Medicine, together with Jumpei Kondo and colleagues across multiple Japanese institutions, focused on patient-derived colorectal cancer organoids—miniature three-dimensional tumors grown in the laboratory from tissue obtained during surgery at Kyoto University Hospital and the Osaka International Cancer Institute. These organoids preserve much of the genetic and molecular identity of the original tumors, making them a powerful platform for testing how individual cancers react to individual drugs. The team grew organoids from ten patients and exposed them to five different targeted agents, capturing the molecular response at a remarkably early time point: just eight hours after drug exposure.

The experimental design combined three layers of molecular measurement. Whole-exome sequencing catalogued the mutations and copy-number alterations carried by each organoid line. Paired RNA sequencing, performed before and after drug treatment, recorded changes in gene expression. The critical third layer, however, was a computational method called organoid-line-specific regulatory-network analysis. Rather than treating genes as isolated variables, this approach, built on Bayesian network inference with B-spline nonparametric regression, models the regulatory relationships—the edges—connecting genes into coordinated circuits. Each edge carries an edge contribution value, or ECv, quantifying how strongly a regulatory relationship behaves in a given sample. When a drug perturbs the system, the change in these values, denoted ΔECv, reveals which regulatory connections have been rewired.

Before testing the network method, the researchers established a baseline with the conventional tools. Mutation profiles, copy-number alterations, expression-only clustering, and differential-expression analysis each provided only partial stratification of the drug responses. Some genomic features were directionally informative: SMAD4 alterations aligned with response to LDN-193189, a bone morphogenetic protein pathway inhibitor, and KRAS mutation status, as expected, correlated with resistance to cetuximab. But the standard analyses could not fully resolve the discordant cases. The most striking example was an organoid line designated C45, which carried a KRAS mutation—normally a predictor of cetuximab resistance—yet nonetheless proved sensitive to the drug. For such patients, conventional biomarkers offer little guidance, and this is precisely where the network approach promised new insight.

Applying their method across the ten organoid lines, the team identified heterogeneous post-treatment rewiring patterns associated with sensitivity and resistance. For LDN-193189, the analysis yielded a common sensitivity subnetwork of fifteen edges—fifteen regulatory relationships that changed consistently among the organoid lines that responded strongly to the drug. When the researchers applied a more stringent ΔECv threshold, twelve of those fifteen edges survived, indicating that the core signal was robust rather than an artifact of an arbitrary cutoff. For cetuximab, the picture was more conservative: when all organoids meeting the predefined high-sensitivity threshold were included, the stringent core consisted of just two edges, a reminder that the strength and consistency of network signals varies considerably between drugs and between biological contexts.

Crucially, the team did not stop at identifying these subnetworks. Recognizing the limitations of a ten-patient cohort, they subjected their findings to a battery of internal robustness analyses. They tested the sensitivity of the results to the ΔECv threshold, mapped network support across the full cohort of ten organoid lines spanning all response categories, and performed exact same-size subset permutation analyses to check whether the common subnetworks could have emerged by chance from arbitrary groupings of organoids. The full-cohort support analysis revealed something important: network support was heterogeneous among the moderate-response and resistant organoids. In other words, the common sensitivity subnetworks should not be interpreted as validated classifiers that cleanly separate responders from non-responders across every response category. The authors are explicit that their findings are exploratory, a framework for generating hypotheses rather than a finished diagnostic tool.

To validate the network-derived findings at the molecular level, the researchers selected genes from within the identified subnetworks and examined them by quantitative polymerase chain reaction. This follow-up provided node-level support for the network predictions, confirming that at least some of the computationally inferred regulatory changes correspond to measurable changes in gene expression. The qPCR primers and results for both the LDN-193189 and cetuximab settings were published alongside the paper, allowing other groups to replicate the measurements independently.

The technical machinery behind the analysis deserves attention because it addresses a persistent problem in network medicine: most regulatory network methods produce a single consensus network for a group of samples, which can obscure the individuality of each tumor. By computing organoid-line-specific networks and then comparing the changes in edge contribution values before and after treatment, the method captures how each patient’s tumor circuitry responds to perturbation in its own way. The mathematical details, including the Bayesian network inference procedure and the extraction of line-specific subnetworks, are provided in the supplementary materials, and the computational work was supported by the Human Genome Center at the University of Tokyo and the NIG supercomputer at ROIS National Institute of Genetics.

Why does this matter for patients? Targeted therapy decisions in colorectal cancer currently rest on a handful of biomarkers, most famously KRAS and NRAS mutation status for anti-EGFR therapy. Patients whose tumors lack these mutations may still fail to respond, and patients whose tumors carry them occasionally respond anyway—the C45 organoid being a case in point. Network rewiring analysis offers a potential way to understand these exceptions mechanistically. If the regulatory circuits that reorganize in the first hours of drug exposure can be characterized, they may point to compensatory pathways that tumors activate to survive treatment, and those pathways in turn may become targets for combination therapies. The framework also fits naturally with the growing infrastructure of patient-derived organoid drug screening, since organoids can be generated, drugged, and sequenced within clinically relevant timeframes.

The authors and the field alike caution that much work remains. Ten organoid lines and five drugs constitute a proof of concept, not a clinical validation set, and the empirical response categories used in the study reflect the inherent difficulty of classifying continuous drug-response data into discrete bins. The researchers themselves emphasize that the identified subnetworks are candidate response-associated features requiring further study in larger cohorts. Still, the study demonstrates that interpretable, drug-response-associated network features can be extracted from patient-derived colorectal cancer organoids beyond what conventional genomic and expression-only analyses provide. As organoid biobanks expand and perturbational transcriptomics becomes routine, network-level analysis of this kind could become a standard layer in the translational pipeline, helping clinicians understand not just which tumors respond to targeted drugs, but why—and, ultimately, how to convert resistance into sensitivity. The research was supported by the Japanese Cabinet Office’s PRISM program, the Fujitsu-Kyoto University Large Scale Medical AI joint research laboratory, and the Japan Agency for Medical Research and Development, and the full article is available open access under a Creative Commons license.

Subject of Research: Regulatory network analysis of targeted drug response in patient-derived colorectal cancer organoids

Article Title: Organoid-line-specific regulatory network analysis identifies targeted-drug response-associated subnetworks in patient-derived colorectal cancer organoids

Article References: Ota, K., Sakuragi, M., Nakazawa, M. A., Harada, Y., Shimizu, S., Onuma, K., Coppo, R., Kawada, K., Obama, K., Aoki, S., Miyoshi, E., Tanaka, Y., Asada, R., Kamada, M., Inoue, M., Tamada, Y., Kondo, J., & Okuno, Y. (2026). Organoid-line-specific regulatory network analysis identifies targeted-drug response-associated subnetworks in patient-derived colorectal cancer organoids. Journal of Translational Medicine. https://doi.org/10.1186/s12967-026-08914-4

Image Credits: AI Generated

DOI: 10.1186/s12967-026-08914-4

Keywords: colorectal cancer, patient-derived organoids, targeted therapy, drug resistance, gene regulatory networks, network medicine, Bayesian networks, cetuximab, EGFR, biomarkers, transcriptomics, personalized medicine