CELL INTELLIGENCE · 肿瘤细胞治疗研究

Claudin 18.2 (CLDN18.2)-targeted CAR-T cell therapies have shown promising clinical efficacy in gastric cancer. However, early-phase trials have reported gastrointestinal adverse events due to on-target off-tumor recognition of CLDN18.2 in the gastric mucosa. By leveraging shared CLDN18.2 epitopes and expression in humans and mice, we establish an in vivo model that replicates the on-target off-tumor toxicity of CLDN18.2 CAR-T. Our findings confirm that this toxicity is independent of the CAR construct's design, co-stimulatory domain, and tumor model. Additionally, we demonstrate the utility of this model in testing strategies to mitigate on-target toxicity, such as Boolean-logic AND-gate approaches. Our results offer insights into the use of mouse models that recapitulate on-target off-tumor toxicities, with the caveat that although we are often concerned that models will undercall toxicities in humans, they may also overcall the incidence and severity of toxicities, prematurely discarding promising therapeutic agents from further clinical development.