Where CGT Manufacturing Breaks Down—And Why Those Same Failures Get You Put on Hold

This post is the second in our three-part series on the trends and challenges in cell and gene therapy (CGT) manufacturing. You can read the first part in the series here.

Clinical hold notices are written in such a way you'd think the problem lives somewhere in a submission document: "Inadequate CMC (chemistry, manufacturing, and controls) information." "Insufficient process controls." "Comparability data not provided." Perhaps a gap in a table, a missing appendix, or a section that didn't satisfy a reviewer.

However, every one of those deficiencies has a root cause on the manufacturing floor—a deviation that wasn't caught until the batch was gone, a process parameter that drifted across shifts without anyone noticing, or a change that seemed minor at the time and created a comparability gap the U.S. Food and Drug Administration (FDA) couldn't reconcile six months later.

The first post in this series established that roughly 21% of CGT clinical holds are triggered by CMC deficiencies¹—the category most likely to be preventable, and the one that takes the longest to resolve. This post dives into exactly where they happen and what it takes to stop them.

When Documentation and Execution Come Apart

A CAR-T manufacturing run involves somewhere between 20 and 30 discrete steps across 10 to 14 days. Each step carries critical process parameters—temperatures, timing, volumes, and reagent lot numbers—that must be recorded accurately and completely. In a paper-based system, operators hand-write those parameters at each step, often while managing multiple tasks under time pressure. And somewhere across those steps an error occurs. A temperature is transposed. A step is marked complete before it finished. The batch keeps moving, and no one knows there's a problem until a quality reviewer reconciles the batch record days later.

By then, the product has either been infused, shipped, or is sitting in a freezer with a data integrity question that blocks release. For autologous products, that patient's cells can't be re-collected. They either wait for another manufacturing attempt—if they're still eligible—or they don't get treated.

The Documentation Gap That Becomes a Regulatory Problem

The FDA expects you to demonstrate that your process is controlled, reproducible, and capable of consistently producing product that meets quality standards. When batch records are incomplete, inconsistently maintained, or reconstructed after the fact, you've lost the ability to make that demonstration. A reviewer of an investigational new drug (IND) application finds process descriptions that don't align with batch record data, or a comparability argument that depends on Phase 1 records that don't contain enough detail to support the comparison: that's the clinical hold. Not because the FDA is being unreasonable, but because your documentation can't prove the process generating your product is under control.

The solution isn't telling operators to be more careful. It's removing transcription as a failure point. When critical parameters feed directly from equipment into the batch record, the human transcription step disappears. When the system enforces process sequence and deviations are flagged in real time rather than discovered during post-execution review, there's still an opportunity to intervene. In CGT, electronic batch records aren't a system upgrade—they're the difference between a manufacturing process that can be demonstrated to be controlled and one that can't.

Process Consistency Across Variables You Can't Control

Clean documentation proves your process ran correctly, but it doesn't prove your process ran consistently. And in CGT manufacturing, where no two batches start from identical material, those are different questions with different answers.

In autologous CAR-T manufacturing, the same process can produce 10-fold expansion in one patient and 100-fold in another.² Both batches have to meet release specifications, and proving the difference reflects known biology rather than an uncontrolled process is a documentation and monitoring challenge that most programs underestimate until they're in the middle of it. The manufacturing question that the FDA will ask is direct: how do you control for patient-to-patient variability? Answering it requires real-time phenotypic characterization of apheresis material and in-process monitoring that shows—for that patient's cells—that the run stayed within defined operating ranges.

In adeno-associated virus (AAV) manufacturing, most of what a production run yields may be empty capsids³, and batch quality can vary by orders of magnitude within the same process. The FDA's objection isn't the variability. It's sponsors who can't demonstrate that their process consistently delivers a product with defined potency. That argument requires upstream process controls that detect drift during production, not end-point testing that tells you the batch failed after the fact.

In lentiviral manufacturing, counting particles and measuring potency are not the same thing.⁴ Most programs measure particles. Switch measurement methods between Phase 1 and Phase 2 without bridging data, and you've created a comparability gap that requires data you almost certainly didn't collect.

The root cause in each case is identical: quality measured only at the end rather than monitored throughout. In-process controls are what allow you to distinguish biological variability from process drift—and they're the evidence base the FDA is reviewing when they ask you to demonstrate process understanding.

The Changes You Made and Didn't Document

The FDA's 2020 CMC guidance for gene therapy INDs is built on the expectation that sponsors will improve their processes as they move through clinical development.⁵ The problem isn't the changes. It's the gap between the change and the documentation.

A media formulation gets optimized. A purification step is adjusted to improve yield. A supplier is switched because the original one discontinued a material. Each change is reasonable and, in the moment, none of them feel like the kind of thing that needs a formal comparability package.

But the FDA doesn't see incremental improvements—they see a manufacturing process that looks different between IND amendments, without documented rationale or comparability data to explain why the product is still the same. In each case, the manufacturing team made a well-intentioned decision, but the absence of formal documentation turned that decision into a regulatory problem that couldn't be defended. Change control embedded in manufacturing execution—not bolted on as a quality approval step—prevents that gap. Changes are documented before implementation, comparability data is generated at the time of the change when the historical baseline is still accessible, and IND amendments are filed before the FDA finds the discrepancy rather than after.

The Materials That Can Derail a Batch Before It Starts

CGT starting materials are the most variable and least standardized inputs in any manufacturing category, and supplier qualification gaps are a consistent source of IND deficiencies. A batch built on inadequately documented materials is potentially unreviewable regardless of how the product tested.

The gap usually isn't that the material was bad—it's that there's no evidence it was consistently good. A vendor that's been used for years, Certificates of Analysis (CoAs) that were reviewed and looked fine, no problems observed in production. When FDA asks for supplier qualification records, acceptance criteria specifications, and lot-to-lot consistency trending, familiarity isn't documentation. Programs that treat supplier management as an administrative function rather than a manufacturing control discover this at the worst possible moment.

For autologous programs, the raw material problem runs deeper. The starting material is the patient's own cells, and traceability from that specific patient through to finished product isn't a documentation formality—it's a patient safety requirement that has no workaround.

What This Series Has Established and Where It's Going

The pattern across every section is consistent: CMC deficiencies aren't generated in submissions—they're generated on the manufacturing floor, and they reach submissions because the systems that would have caught them weren't there. Paper batch records that discover problems too late. Process monitoring that can't distinguish biology from drift. Change control that can't keep pace with development speed. Supplier management that mistakes familiarity for qualification.

Every one of these failure modes has a corresponding solution—not as a compliance layer added on top of manufacturing, but as infrastructure that makes it possible to execute under the constraints of this product class. The final post in this series will lay out what an ideal infrastructure looks like, how to build it in a way that scales from Phase 1 to commercial, and why the architecture decisions made at the beginning of a program determine whether the comparability, consistency, and traceability arguments hold up at the end.

Read part 1 in this CGT manufacturing series here.

References

1. "Clinical holds for cell and gene therapy trials: Risks, impact, and lessons learned," Carolyn A. Wills, Daniela Drago, and Rober G. Pietrusko, Molecular Therapy – Methods & Clinical Development, Oct. 20, 2023.
2. "The Challenge of Variability in Chimeric Antigen Receptor T cell Manufacturing," Andrew D. Fesnak, Regenerative Engineering and Translational Medicine, Aug. 19, 2019.
3. "Methods to reduce empty capsid content in AAV manufacturing at scale," Eureka by Patsnap, Sept. 2, 2025.
4. "Challenges and Opportunities in Lentivirus Viral Vector Manufacturing for In Vivo Applications," Eduardo Barbieri and Caryn L. Heldt, Biomedicines, Feb. 5, 2026.
5. "Chemistry, Manufacturing, and Control (CMC) Information for Human Gene Therapy Investigational New Drug Applications (INDs)," FDA Guidance for Industry, January 2020.