Researchers at Massachusetts General Hospital in Boston, working with colleagues at the Broad Institute of MIT and Harvard and California-based Kite Pharma, report that inherited germline variants in three genes — STXBP2, ADAMTSL3, and PTPN22 — predict severe toxicity and CAR T cell expansion in patients treated with the CD-19-targeted axicabtagene ciloleucel (axi-cel), according to a study published July 28, 2026 in Science Immunology. The findings suggest that a patient's own germline DNA, rather than tumor characteristics alone, encodes meaningful variation in CAR T cell therapy outcomes. The study represents the first replicated germline whole-genome sequencing analysis across two randomized CAR-T trials.
Beyond autologous CAR-T therapy, the findings may have even greater implications for allogeneic cell therapies. Because a single healthy donor can provide source T cells for dozens to hundreds of off-the-shelf CAR-T products, inherited germline variants that influence T-cell expansion or inflammatory signaling could affect the performance of an entire manufacturing campaign. If validated, germline genomic screening could become an additional criterion for donor selection alongside existing manufacturing and safety requirements.
Response rates to CAR T cell therapy remain highly variable across patients with aggressive lymphoma, and the biological basis for that variability is incompletely understood. Tumor burden, disease characteristics, and manufacturing variables have all been examined as predictors, but germline factors — the inherited genetic background of the patient and the T cells drawn from them — have received comparatively little systematic attention. Because autologous CAR T cells are manufactured from a patient's own lymphocytes, they carry that individual's full complement of germline variants, making the product biologically distinct from patient to patient in ways that go beyond manufacturing process controls.
The study, funded solely by Kite Pharma, applied whole-genome sequencing to samples from participants in two large randomized trials of axi-cel — ZUMA-1 (NCT02348216) and ZUMA-7 (NCT03391466) — in patients with aggressive large B cell lymphoma.
The analysis focused on 191 patients of European ancestry drawn from both trials — 78 from ZUMA-1 and 113 from ZUMA-7 — comparing those who developed severe cytokine release syndrome (CRS) or immune effector cell-associated neurotoxicity syndrome (ICANS) against controls who did not. Variant enrichment was assessed against large population references including gnomAD and the UK Biobank. Functional validation used CRISPR-Cas9 knockout and variant overexpression in primary human CAR T cells from healthy donors, with degranulation assays, intracellular cytokine staining, macrophage coculture, and real-time cytotoxicity assays.
STXBP2 and the Degranulation–Cytokine Storm Axis
The most functionally characterized finding centers on STXBP2, which encodes syntaxin binding protein 2, a regulator of SNARE complex assembly and cytotoxic granule fusion with the plasma membrane. Rare pathogenic germline variants in STXBP2 were significantly enriched in patients who developed severe toxicity in both ZUMA-1 and ZUMA-7, and were depleted in population-level controls — supporting pathogenicity rather than common variation. All six STXBP2 variant carriers in ZUMA-1 reached the combined toxicity endpoint, with a hazard ratio of 3.72 for toxicity and a median onset of day 5 versus day 7 in non-carriers, a stronger predictive signal than tumor burden. Functional validation using CRISPR-edited primary CAR T cells confirmed that the identified variants altered degranulation, cytokine production and macrophage activation.
ADAMTSL3 as a Potential Toxicity Modifier
Germline variants in ADAMTSL3, a negative regulator of TGF-β signaling, showed the opposite pattern: carriers were enriched in the non-toxic control cohort across both ZUMA-1 and ZUMA-7, making it the only gene nominally significant in the same direction in both trials. Carriers also showed lower peak IL-15, a cytokine associated with ICANS. The proposed mechanism — that putative loss-of-function variants in ADAMTSL3 reduce inhibition of TGF-β, thereby dampening the inflammatory cascade — positions this gene as a potential endogenous brake on CAR T cell-mediated toxicity.