Manufacturing Intelligence: Organoids, Organ-on-a-Chip, and the Trust Gap

New approach methodologies (NAMs) may be shortening the drug development pipeline more than AI is right now, but organoids, organ-on-a-chip systems, and in silico models still have to earn the kind of trust that animal models have held for decades.

In Part 1 of the 3-part episode 7 of PharmTech's Manufacturing Intelligence, co-hosts Christopher Cole and Richard Jaenisch examine where NAMs are compressing timelines and what FDA's latest moves mean for sponsors and CMC teams. Jaenisch notes that AI is being placed throughout development and its speed gains will depend on how it is deployed. NAMs, by contrast, are positioned to shorten the pipeline more directly, and he points out that roughly half involve AI in some form, though in silico models are not the chatbot-style generative AI most people picture. He highlights a recent in silico model for drug-induced liver injury that moved through FDA's ISTAND program, the pathway he expects many NAMs to follow, calling liver injury and toxicity one of the core concerns for any advancing therapy and an area where NAMs are particularly well suited.

Animal studies, Jaenisch explains, remain lengthy when measured holistically, even with vivariums running efficient, well-established systems. NAMs could trim a few months from that process today and possibly years in the future, though the savings depend on the specific challenge and how far a developer is from reaching its drug target. For CMC and manufacturing teams, the near-term value lies in human-relevant data that animal models, with their non-human biology and unexpected results, cannot provide. "Largely speaking, I don't see a huge time savings on the production component yet," he observes.

The discussion then turns to FDA's direct final rule replacing animal-test language with nonclinical tests, alongside a database of roughly 25 real-world NAM use cases that includes organoid and organ-on-a-chip data. Jaenisch says many in the industry do not yet fully trust these models compared with animal models, but investment trends suggest companies are putting money behind them. He draws on his own experience as a hepatitis C survivor: chimpanzees were the only animal model that worked with the virus, and once they fell out of favor in the 2010s, investment shied away because sponsors wanted a sure path to approval. Early development of sofosbuvir relied on modified zebrafish and as many as 10 different models, each contributing a fragment of toxicity data. Organoids, built from human cells as mock human organs, cannot replicate the whole body's environment but offer organ-specific insight that makes liver toxicity testing more feasible.

Multi-organ chips introduce their own questions. Jaenisch asks whether a drug moves through the tubing linking a liver to a kidney the same way it moves through human vessels of different material and composition, issues that will need vetting by both the experts using these systems and those reviewing them. A leaner FDA adds another variable: with only a handful of specialists in the field, He questions whether the agency can upskill quickly enough, and whether market investment will expand only to be throttled by regulatory capacity or spark a surge of new applications.

With the rule issued on September 22, Jaenisch expects it will take a year or two before the industry can see where confidence in NAMs is truly being established. "I think it's going to take a minute for that trust to be earned, genuinely," he notes. For CMC leaders, the takeaway is to build familiarity with NAM data now, before faster, human-relevant safety packages reshape development timelines.