Pancreatic cancer kills nine out of ten patients within five years. A Chinese team publishes in Cell Stem Cell a biobank of pancreatic organoids and links it to a phase II clinical trial testing a combined therapeutic approach.
The essentials
- The biobank comprises 260 organoid lines. Analysis identified 2,794 multi-omics characteristics associated with drug sensitivity (Cell Stem Cell, 2025).
- In the associated phase II trial (37 patients), 26 showed a decrease greater than 20% in tumor markers.
- The trial is single-arm: without a comparison group, the response rate is insufficient to establish the superiority of the approach over standard care.
- The main bottleneck is now regulatory and methodological, no longer biological: how to transform a discovery platform into robust clinical decision-making, and how to ensure benefits are not confined to large centers.
Pancreatic cancer resists where others retreat
Most cancers have seen their prognosis improve markedly over the past twenty years. Pancreatic cancer’s prognosis remains among the most unfavorable, with a five-year survival rate around 12% according to data from the American National Cancer Institute, though it has improved over two decades. Two reasons dominate: the disease is detected late, often after the tumor has already invaded neighboring tissues, and it resists standard chemotherapies with discouraging frequency.
Resistance stems in part from the heterogeneity of pancreatic tumors. Two patients carrying the same KRAS mutation, present in over 90% of cases, can respond radically differently to the same treatment. Genomic profiles alone offer limited predictability of response. Sequencing tumor DNA tells us what genetic alterations the cell carries; it does not tell us how it will react when exposed to a drug.
This is where the organoid comes in. Taken from the patient’s tumor, cultured in the laboratory for a few weeks, it constitutes a living three-dimensional model that retains the biological characteristics of the original tumor. Treatments can be tested on it ex vivo and potential sensitivities identified, but their use to guide initial treatment still needs confirmation in interventional trials.
260 lines, 2,794 markers: what the biobank makes possible
The team established 260 organoid lines from pancreatic cancers in a collaboration led notably by Dong Gao’s team and Changhai Hospital in Shanghai. The study identified 2,794 multi-omics characteristics associated with drug sensitivity, what researchers call multi-omics analysis. The ambition is to map the links between these biological profiles and responses to a battery of molecules tested on the organoids.
This mapping work produces two types of results. The first is fundamental: it reveals associations between multi-omics profiles and clinical classifications. The analysis identifies links between certain biological characteristics and responses to MAPK pathway inhibitors or platinum compounds. The biobank thus becomes a repository of molecular vulnerabilities.
The second result is clinical. Based on these profiles, the team designed a phase II trial (reference NCT06241352 on the ClinicalTrials.gov registry) testing atorvastatin combined with standard chemotherapy in patients with advanced pancreatic cancer. Of 37 patients enrolled, 26 showed a decrease greater than 20% in tumor markers.
This figure deserves careful reading. A single-arm trial, without a group of patients treated according to standard protocol for comparison, does not allow us to conclude that the organoid-guided approach outperforms the conventional approach. It demonstrates that it is feasible and that the observed rates are encouraging enough to justify the next step: a randomized trial with a comparator arm.
Statins as a signal, not a solution
The choice of statins deserves explanation. These drugs, prescribed for decades against cholesterol, have long intrigued cancer researchers. Several observational studies suggested that prolonged statin use could be associated with reduced risk of certain cancers or improved prognosis. The proposed mechanisms involve the mevalonate pathway, implicated in cell proliferation, and pro-apoptotic effects—that is, a tendency to push cancer cells toward self-destruction.
The problem, until now, was patient selection. Administering statins to all patients with pancreatic cancer produces no measurable effect in a heterogeneous population. The organoid provides a filter: statins showed activity against chemoresistant pancreatic organoids in the study’s experiments. The Chinese trial evaluates an atorvastatin-chemotherapy combination inspired by analyses of chemoresistant organoids, without demonstrating individualized patient selection by organoid.
The advance lies in demonstrating that a living biological model can generate hypotheses or predictions of sensitivity for treatments whose utility for patient selection still needs prospective validation. Sequencing alone only imperfectly predicts therapeutic sensitivity, and adding functional profiling by organoids could improve stratification. This logic can apply to drugs whose response depends on the tumor profile, provided proper validation specific to each drug and clinical context.
The missing link between biopsy and clinical decision
Precision medicine promised much over the past decade. It has kept part of its promises—immunotherapy guided by PD-L1 expression, BRCA inhibitors in ovarian and breast cancers, therapies targeting EGFR in lung cancer. But pancreatic cancer has largely remained outside these successes. Robust predictive biomarkers are rare there, and KRAS mutations, ubiquitous, long remained untargetable.
Recent advances in primary care show that well-designed infrastructure can profoundly alter health trajectories. In precision oncology, technology must integrate into repeatable clinical decision-making. The Shanghai biobank attempts to build this chain: it has served to link multi-omics profiles and therapeutic sensitivities within a research program. This sequence takes several weeks according to published protocols and requires laboratory logistics that most hospitals do not yet control.
The biobank’s data suggest this approach is biologically valid. The challenge now lies in its integration into existing care pathways, with timelines compatible with oncologic urgency and at a cost a public health system can absorb.
A single-arm trial opens a door, not a highway
The results deserve their context. In standard chemotherapy trials for metastatic pancreatic cancer, objective response rates hover around 20 to 30% for reference protocols like FOLFIRINOX or gemcitabine-nab-paclitaxel. A non-randomized cohort of 37 patients does not directly compare to populations in landmark trials.
This precision does not invalidate the trial. It defines its scope. A phase II single-arm trial serves to demonstrate feasibility and calibrate the expected effect for sizing the next step. It can provide a signal of feasibility or activity, but is insufficient to establish the clinical utility of patient selection. Regulators—whether the American FDA, the European Medicines Agency, or its Chinese equivalent, the Center for Drug Evaluation—could request adequate controlled data for regulatory approval.
The EMA evaluates data for drug authorization, while treatment recommendations fall to competent health authorities and clinical organizations. The NMPA evaluates data required for drug registration.
This regulatory hurdle is legitimate. It protects patients from premature generalization of promising but insufficiently validated techniques. But it also outlines a concrete methodological challenge: designing a randomized trial where the control arm receives standard treatment without organoid guidance, and the experimental arm is selected by the biobank. Such a trial will need to recruit several hundred patients across multiple centers over several years.
Who builds tomorrow’s biobank, and for whom
The implications of this research extend beyond the Shanghai trial. If pancreatic organoids become a standard therapeutic decision tool by 2040—a conditional scenario, not a certain projection—two trajectories are possible.
In the first, a few large academic centers accumulate samples, build proprietary databases, patent protocols for culture and characterization. General hospitals refer their patients to these centers without participating in evidence production or developing their own expertise. Patients living far from these hubs or treated in less well-resourced health systems have no access to organoid-guided selection. The precision tool reproduces the geography of healthcare access inequality, a phenomenon that research on primary care concentration already documents in other domains.
In the second trajectory, interoperable biobanks, funded by public consortiums, equipped with common culture and characterization protocols, governed by shared rules on biological data, make organoids a common good of clinical research. Adaptive multisite trials allow medium-sized hospitals to contribute patients and share results. The time between biopsy and sensitivity result decreases as protocols standardize.
Three signals would allow tracking which trajectory prevails. First: the success rate for organoid culture in non-specialized centers, and the average delay between biopsy and profile return. If this delay remains above three weeks outside large centers, the technology will remain de facto reserved for the best-resourced. Second: the number of comparative multisite trials launched in coming years; each multisite trial is a signal that governance is distributed. Third: the geographic distribution of patients enrolled in these trials, which will indicate whether evidence production remains concentrated in China, the United States, and Western Europe, or expands.
Pathways to progress exist. Funding hospital networks of mutualized biobanks, modeled on what the European Union built for genomic biobanks under the “1+ Million Genomes” initiative, would provide shared infrastructure. Requiring a robust clinical comparator before any integration into national recommendations protects against premature generalization while maintaining pressure for phase III trials to launch quickly.
The obstacle is not demonstrating that the organoid improves predictability compared to genome alone—an approach for which biological data are beginning to accumulate. The obstacle is organizing evidence production so that an oncologist in Dakar, Lyon, or Hanoi might one day prescribe treatment guided by the same living model as an oncologist in Shanghai or Boston. This challenge resembles less a scientific problem than a governance problem for global medical infrastructure, a problem biology has already begun to produce answers for.



