Off-the-Shelf Base-Edited CAR33 T Cells Miss Primary Endpoints in First-in-Human AML Trial

Three Enrolled Out of Five Screened

Five patients under 16 with relapsed or refractory AML were screened under ISRCTN14430213 and three were enrolled, and a fourth patient, an adult, received the product through compassionate access. All were treated at one center in an open-label, nonrandomized phase 1 design. Participants received fludarabine, cyclophosphamide and alemtuzumab as lymphodepletion, then a single infusion of between 1.2 million and 1.8 million BE-CAR33 cells per kilogram of body weight, given ahead of allogeneic stem cell transplantation.

Treatment-emergent adverse events included cytokine release syndrome at grade 2 or below, grade 3 neurotoxicity, grade 4 cytopenias and transient rashes. Two patients showed reduced minimal residual disease and moved to transplant. The authors state plainly that primary endpoints were not met.

Autologous T Cells Are Hard to Get in Pretreated AML

CAR T cell therapy in AML runs into two problems the authors name at the top of the paper, antigen heterogeneity and expression of the target on healthy compartments. Obtaining autologous T cells from heavily pretreated AML patients is frequently difficult, which is the practical argument for a banked donor product. The team, led from the UCL Great Ormond Street Institute of Child Health and Great Ormond Street Hospital for Children NHS Trust with collaborators at Queen Mary University of London, King's College Hospital and Hannover Medical School, used base editing to knock out TRAC, CD52 and CD7 at once, described in the paper as preventing graft-versus-host disease and evading immunotherapy effects.

The Laboratory Trail Behind Each Patient

Three assays ran in parallel on every treated patient. Serial flow cytometry, chimerism quantification and vector copy number analysis followed BE-CAR33 T cells until they were eliminated during transplant, which means the transplant itself cleared the engineered product rather than the product persisting as a long-term surveillance population. Differential gene expression carried the editing signature and shifted over time, moving from a manufacturing-related profile toward postexpansion effector and exhaustion profiles.

Georgiadis and Colleagues Stop at Feasibility

Four treated patients, three enrolled on the trial and one dosed through compassionate access, cannot support an efficacy claim, and the authors do not make one. Their stated conclusion is that the trial "demonstrated the feasibility of an 'off-the-shelf' base-edited CAR T cell approach and informs future multiantigen strategies against AML." A Science Translational Medicine subscription stands between the reader and the manuscript. Available from outside it were the published abstract, the author list and the full affiliation record. Response durations, individual patient trajectories and the manufacturing release data sit in the parts nobody outside can see.

Why This Matters to the APO|APE Reader

The same London group reported a very different picture in T-cell ALL. Their universal BE-CAR7 program, published in the New England Journal of Medicine on January 8, 2026, treated 11 patients under ISRCTN15323014, brought 9 of them to deep remission and transplant, and left 7 of 11 in ongoing remission at 3 to 36 months, following a 2023 report on the first three children. Set against that record, a CD33-directed version enrolling three children and missing its endpoints says the transferable part of base editing is the manufacturing platform, and the target biology in AML remains the harder problem. Any center weighing an edited donor product will also be weighing the immunophenotyping and chimerism workload described here, which ran per patient and did not stop at infusion.