When the safety net becomes the tripwire

On the same Monday, the FDA stopped one gene therapy over five asymptomatic MRIs from its own follow-up protocol — and quietly reopened a review the advisory committee had voted dead. Plus the first oral drug for dermatomyositis, a tau tracer with a label at last, and US$380 million of cell-therapy automation gone.

The one that matters

On 24 August, Regenxbio disclosed a clinical hold on RGX-121, its intrathecally delivered gene therapy for Hunter syndrome. Not because a patient got worse. Because expanded MRI surveillance found small nodules or cystic masses on the spine scans of five participants dosed three to six years ago. All five are asymptomatic. Their neurocognitive data are stable or improved. The investigators call the findings non-serious, the radiologists likely benign. The programme is stopped anyway.

The same day, Capricor announced that the FDA had moved the decision date for deramiocel, its cell therapy for Duchenne muscular dystrophy, from 22 August to 22 November. Remember the context: an advisory committee voted 9–3 in July that the evidence does not support effectiveness in DMD cardiomyopathy, and on 20 August STAT reported a rejection was all but certain. Instead of the expected refusal, Capricor filed 24-month extension data plus a narrower indication — upper-limb function, the endpoint its phase 3 trial actually met — and the agency classified the package as a major amendment and reset the clock.

Read together, the two decisions describe how CBER actually works in 2026, and it is not how most development plans assume. A review is never as finished as the advisory vote suggests, and a programme is never as safe as its clinical data suggest. The long-term follow-up imaging that regulators ask for — and sponsors agree to without much negotiation, because it feels like a formality — produces findings. Findings in asymptomatic patients, years after dosing, with no agreed rule for what happens next. In the absence of that rule, the default answer is a hold.

I have sat in the meetings where these commitments get made. The imaging schedule is agreed in twenty minutes at the end of a Type B discussion, because everyone's attention is on the efficacy endpoints. Nobody asks the only question that matters ten years later: what finding triggers what action, decided by whom, on what timeline?

Negotiate the adjudication path before you agree the surveillance. If your protocol mandates long-term imaging, it should also define which findings are reportable, which are actionable, and who classifies them. An incidental finding without a pre-agreed category becomes whatever the review division decides it is that week.

Treat late-cycle amendments as a tool, not a failure. Capricor traded three months of clock for the chance to anchor its claim on the endpoint it actually met. If your extension data sharpen the population or the claim, filing them mid-review is a legitimate move — provided the deficiencies from any earlier letter are genuinely fixed, because the amendment reopens everything.

Budget for the file staying open. In both directions. The programme you think is dead may have one move left; the approval pathway you think is secured carries surveillance obligations that can stop it years later. Neither belongs in a plan as a residual risk footnote.

Also this week

Three more worth a click

FDA granted RMAT designation to Kyverna's CD19 CAR-T in non-active secondary progressive MS — on six patients from an investigator-initiated phase 1. CAR-T's autoimmune franchise is walking from rheumatology into core neurology (Kyverna, 11 Aug).
Two more immunoglobulin lots — Gamunex-C and GammaKed — were withdrawn in the US over hypersensitivity reactions, extending an eighteen-month run of Ig quality events; if IVIg is background or rescue in your protocol, write lot-switch provisions in now (IDF safety notices, 14 Aug).
Ultragenyx won accelerated approval for the first GSDIa gene therapy on a surrogate you can explain to a family: how much cornstarch the patient still needs. A functional surrogate precedent worth citing in rare-disease Type B meetings (FDA, 19 Aug).