Ninety-five percent of cancer drugs that clear animal testing fail when they reach human patients. That number has been quoted so often it has lost its shock value. But every data point behind it represents a failed trial, a discarded pipeline candidate, and a patient who ran out of time. The standard model of drug discovery has a precision problem, and a regulatory shift now underway is starting to crack it open.
In October 2025, the FDA approved an investigational new drug application for a combination therapy involving BAL0891 and the checkpoint inhibitor tislelizumab. Qureator announced the approval on October 6, 2025. The approval itself was unremarkable. What was remarkable was the evidence behind it: for the first time, the FDA cleared an oncology IND based entirely on efficacy data from human vascularized organoids — lab-grown clusters of patient-derived tumor tissue — with no animal proof-of-concept testing at all. The study was run by Qureator, a San Diego-based organ-on-a-chip company, and the data were presented at AACR 2025 before the agency signed off.
The approval is a data point, not a revolution. But it arrives at a moment when the infrastructure supporting animal-based drug discovery is being deliberately dismantled. In April 2025, the FDA announced it would begin phasing out requirements for animal testing in the development of monoclonal antibody therapies and other drugs — a policy shift that sent reverberations through the preclinical services industry overnight. The timing matters. Organoids have been in research labs for two decades, since Hans Clevers at the Hubrecht Institute first grew intestinal crypts from stem cells in 2009 and coined the term. They have never lacked scientific credibility. What they lacked was a regulatory opening.
Qureator's platform — called vTIME, for vascularized tumor immune microenvironment model — was developed by a team led by Dr. Sanghee Yoo, Qureator's Head of Biology. The platform attempts to do something animal models cannot: replicate the specific architecture of a patient's tumor, including its blood vessel structure and immune environment, in a dish. The company grows 3D tumor organoids from patient tissue and tests drug combinations against them, using AI to analyze the results. When it found a synergistic effect between BAL0891 and an immune checkpoint inhibitor, it had human data — not mouse data — to show the FDA.
Roche's Institute of Human Biology has offered a closely watched example of what a compressed timeline looks like. According to a review the institute published in Nature Reviews Drug Discovery, one antibody went from initial concept to first-in-human trial in two and a half years using organoid testing alone — no animal models, no cell lines — and is now in Phase 3 clinical trials. That molecule exists because someone bet that human data could substitute for animal data, and the bet paid off. It remains, as the Roche team acknowledges, an exception.
The gap between proof-of-concept and standard practice is where the real story lives. Organoids are not yet woven into the operational fabric of most pharmaceutical companies. Animal models have been the substrate of preclinical drug development for so long that swapping them out requires more than scientific validation — it requires restructuring supply chains, retraining teams, and recalibrating the relationship between a company and its regulators. Clevers, now ad interim director of Roche's Institute of Human Biology, has a word for the industry's hesitation: koudwatervrees, Dutch for "cold water fear" — the hesitation to step into something unfamiliar even when the water is warm. "You cannot simply discard those without providing the same level of evidence," he said at a GEN conference in March. He is also skeptical of the FDA's stated five-year timeline to phase out animal efficacy testing. "I think that's overly optimistic," he told EL PAÍS in November 2025. "The strength of organoids, but also their weakness, is that they are very simple."
That simplicity is the core technical challenge. A tumor organoid in a dish is not a tumor in a body. It lacks the systemic context of a circulatory system, a microbiome, a nervous system signaling back to the tumor site. Vascularized organoids — ones that include blood vessel-like structures — are a partial answer, but they are still missing things scientists have not fully enumerated. The question is not whether organoids can replace animal models outright. The more pressing question is whether they can supplement them well enough to catch the failures that animal models systematically miss.
Alif Saleh, CEO of 28bio — a company that grows human brain organoids at scale for neurodegenerative disease and cancer neurotoxicity screening — frames the problem in economic terms. "The risk of failure is built in all the way to Wall Street," he said. "If everybody accepts that risk, there's very little incentive to change." His company is not waiting for a regulatory mandate. It is building organoid platforms as a commercial service and hoping that a visible commercial success, not a policy announcement, is what finally changes minds.
The October IND approval is the closest thing to that visible success the field has produced so far. It did not take a catastrophe or a regulatory showdown — it took a company willing to generate the data, an agency willing to accept it, and a drug developer willing to submit the filing. The next test is whether that trick can be replicated at scale, with different tumor types, different drug classes, and a skeptical industry watching to see whether the exception becomes a precedent.
