Their Gene Therapy Worked. The FDA Said No Anyway.

On 11 July 2025, Ultragenyx received a Complete Response Letter for UX111, an AAV gene therapy for Sanfilippo syndrome type A. The reviewers had no quarrel with the science. They called the clinical data robust and the biomarker data supportive. What stopped the approval was additional Chemistry, Manufacturing and Controls (CMC) information, plus observations from a recently completed inspection of the manufacturing facility. Ultragenyx characterised those observations as facility and process related, not tied to product quality.

So the drug worked, and the company that made it still could not prove, to the agency's satisfaction, that it could make the same thing the same way, lot after lot, at scale. A therapy for a fatal paediatric neurodegenerative disease did not reach the children who needed it on the timeline its data earned.

Ultragenyx is not alone. This has become the defining failure mode of cell and gene therapy in 2025 and 2026, and there is now a number attached to it.

Three numbers that should reorder your development plan

For cell and gene therapy (CGT), the question the FDA is really asking has changed. The old question was whether the therapy worked. The question now is whether you can make it the same way twice. Three figures frame the stakes.

The $4.25M Lenmeldy price, the gene therapy for metachromatic leukodystrophy, is the economic weight riding on a process you have to validate. When the cure costs more than a house, the consistency of the manufacturing stops being a back-office line item and becomes the asset itself. And the 3 programmes, Ultragenyx, Capricor, and Rocket, are three names you know, three strong stories, three programmes stalled at the same wall inside roughly a year.

The CMC Cliff is where strong programmes go to wait

Call it the CMC Cliff. A programme builds momentum on clinical efficacy, clears the science, then runs off the edge of its own manufacturing maturity. Strong data gets you to that edge. It does not carry you over.

Look at the three programmes together, because the differences matter as much as the pattern.

Read the table carefully, because the contrast carries the lesson. Capricor's Deramiocel is not a clean CMC story. That letter questioned the substantial evidence of effectiveness and asked for more clinical data alongside the CMC clarifications, so efficacy still mattered there. Ultragenyx and Rocket are the purer illustrations of the cliff. In both, the science was settled and manufacturing was the holdup. Rocket's Kresladi drew a CMC-only letter in June 2024, was resubmitted, and now carries a 28 March 2026 PDUFA date. A manufacturing question cost that programme roughly two years.

Two years is no rounding error for a therapy priced in the millions, aimed at a small population, racing competitors and patent clocks. And the delay does not wait for the BLA. Trade-press estimates put roughly 40% of CGT investigational new drug (IND) submissions as delayed for CMC-related reasons even at the investigational stage. That figure comes from industry coverage rather than a primary FDA count, so treat it as directional. The point holds either way: the friction shows up at the first filing and compounds all the way to the approval decision.

Set that against the commercial appetite. Analysts at GlobalData project the cell and gene therapy market to grow at a 31.3% compound annual rate through 2030. That is a forecast, not a fact on the ground, but it captures the mismatch cleanly. Capital and demand are sprinting, and approvals are stalling on process readiness. The science exists. What programmes cannot reliably produce is a manufacturing case the agency will accept.

11 January 2026: the FDA concedes the point

On 11 January 2026, the FDA's Center for Biologics Evaluation and Research (CBER) published a communication titled "Flexible Requirements for Cell and Gene Therapies to Advance Innovation." The agency described CGTs as inherently complex biologic products, often individualised for patients, that may require sophisticated manufacturing under real time pressure. It said it would apply flexibilities already permitted under its regulations while keeping quality standards rigorous. The press release crossed the wires the next day.

Here is the regulator that has been writing these letters now stating, in writing, that manufacturing maturity is the rate limiter for this class of medicine, and adjusting how it oversees that maturity. Three concrete flexibilities stand out.

• Deferred cGMP for early trials: Before an investigational product is manufactured for phase 2 or phase 3, the sponsor is not expected to fully comply with 21 CFR part 211 (current good manufacturing practice for finished pharmaceuticals). The agency also signalled it will not demand stringent comparability data for minor manufacturing changes during development.
• Flexible commercial specifications: CBER will consider flexibility on product release specifications for BLAs, especially for small patient populations, and may revise specifications as post-approval manufacturing experience accumulates.
• No mandatory three PPQ lots: The classic expectation of three process performance qualification (PPQ) lots is gone as a hard rule. The FDA will evaluate whether the proposed number of lots is justified, allows concurrent release of qualification lots, and centres the review on process understanding rather than a lot count.

You could read all this as the cliff disappearing. It is not.

Flexibility is not absolution

This is the part that should shape how you staff and sequence your programme. Legal analysis of the announcement (Covington & Burling) puts it bluntly: the communication largely reflects existing policy and practice. The underlying legal requirements under the Federal Food, Drug, and Cosmetic Act and the Public Health Service Act have not changed. Flexibility clarifies the path while leaving the obligation to demonstrate a defensible control strategy fully in place.

The real question is therefore technical. What, specifically, has to be locked into that control strategy for an AAV or cell therapy product?

The science that has to hold

For an AAV gene therapy, the empty-to-full capsid ratio is a recognised critical quality attribute, and a hard one. Empty and partially filled capsids carry no therapeutic genome yet still count towards capsid load. They dilute the effective dose, complicate your PK/PD modelling, and raise the immunogenicity and safety profile of the product. The gold-standard method for measuring the ratio, analytical ultracentrifugation, is slow, expensive, and poorly suited to high-throughput screening, which is one reason these control strategies are so hard to pin down early. Vector genome titer, replication-competent AAV, residual host-cell DNA and protein, aggregation, and potency all sit beside it on the specification sheet.

Viral safety can no longer be deferred. ICH Q5A(R2), the viral safety guideline, was finalised in November 2023, adopted by the FDA in January 2024 and by the EMA in June 2024. The revision expressly extends viral safety expectations to genetically engineered viral vectors and endorses NGS (next generation sequencing) and PCR for adventitious-agent detection on a risk basis. If you are making an AAV product, this is the live framework your package has to satisfy, not a legacy document you can wave at.

For cell therapy, the difficulty migrates to the cold chain and the batch. Cryopreservation media and excipient selection, post-thaw viability and potency, tonicity and buffer for autologous products such as CD19 CAR-T: each one is a control-strategy decision with a safety and efficacy tail. Underneath all of it sits the small, individualised batch. When every lot is one patient's cells, conventional statistical process control, built for thousands of identical units, simply does not apply. You cannot average your way to confidence across an n of one.

Potency assays for these modalities are notoriously immature, and the analytical maturity bar set by ICH Q2(R2) and Q14 is where a large share of CMC gaps actually live. Layer on comparability under ICH Q5E, which bites the moment you change your process between phase 2/3 and commercial scale (exactly the change the FDA's relaxed comparability stance touches), and the shape of the work comes clear. The control strategy is a connected argument running across half a dozen ICH guidelines, and the agency can pull on any link in it.

This work decides whether your strong data ever reaches a patient, and most programmes start it far too late.

How DeepC helps build the CMC package the cliff still demands

DeepC is an AI co-scientist for pharmaceutical formulation scientists. Its biologics studies are built for exactly this problem: assembling a defensible, ICH-anchored CMC package for advanced modalities, grounded in real regulatory precedent and a mapped IP landscape, with every claim linked back to a source. Three studies map directly onto the CMC Cliff, and they work in sequence: precedent, then IP, then the design candidates.

One boundary up front, because the honest framing is the credible one. DeepC produces design candidates with their specifications, stability profile, and CMC red flags. It does not run your assays, manufacture your lots, or file your BLA, and it will not get your therapy approved on its own. What it does is build, fast and with citations, the package your team still has to execute, and surface the gaps before the agency does.

Biologics research review: anchor the strategy in what has actually cleared

Before a single specification gets proposed, the biologics research review grounds the work in precedent. The question it answers is the one that de-risks everything downstream: what has actually cleared the FDA for this modality, and what specifications and viral-safety package did those precedent products defend? You design your control strategy against the bar real approvals have already met, with a documented reference point instead of guesswork.

Biologics FTO analysis: map the capsid IP before you build around it

A capsid choice is two decisions at once. It is a CMC decision (how the serotype behaves on empty-full ratio, how manufacturable it is) and a freedom-to-operate decision (who already claims that serotype or engineered variant). Locking a control strategy around a capsid you cannot freely use is an expensive way to find an IP problem. The freedom-to-operate analysis maps that question, returning design-around strategies each with its own risk rating, before capital and process development commit to it.

Biologics formulation and CMC study: design candidates with severity-tagged red flags

This study is the primary one for the cliff. Ask it for buffer and tonicity considerations for an autologous CD19 CAR-T cryopreserved product, and you get structured, cited design candidates rather than a literature dump. Two parts of that output land on the cliff with no translation needed. The CMC red flags, each tagged with its severity, are the literal artefact you need to avoid an Ultragenyx-style CMC and facility letter: it surfaces the gaps in your package while you can still close them, well ahead of the agency finding them for you. And against the FDA's January 2026 flexibilities, the study's Q6B-anchored specification rationale is exactly the input the small-population specification-flexibility path requires, while its stability and specification framing remains what the agency still demands even where the three-PPQ-lot mandate has lifted. The study designs the drug product; it does not design your capsid or run your analytical ultracentrifugation.