IL-15-engineered GPC3 CAR-T cells enhance tumor control in low-antigen liver cancer xenografts

Hepatocellular carcinoma (HCC) is a highly lethal tumor with limited effective treatments, especially for metastatic forms. While Chimeric Antigen Receptor (CAR)-T cell therapy has revolutionized hematologic malignancies, its efficacy in solid tumors like liver cancer is hampered by antigen heterogeneity, low target antigen density, and an immunosuppressive tumor microenvironment (TME). Cytokine engineering offers a promising avenue to boost CAR-T cell persistence and effector function, but identifying optimal cytokine payloads for specific tumor contexts remains a significant challenge due to the complexity and labor-intensive nature of experimental comparisons. This study addresses this gap by using an in silico platform to guide cytokine selection.

Researchers developed a large language model (LLM)-based CAR-T in silico platform to evaluate cytokine engineering strategies for glypican-3 (GPC3)-targeted CAR-T cells in liver cancer. The platform systematically assessed IL-2, IL-7, IL-12, IL-15, and IL-18 as potential enhancers. Guided by computational predictions, IL-15 was identified as the most effective. Subsequently, cytokine-armored GPC3 CAR-T cells were generated and validated in vitro for proliferation, persistence, and serial cytotoxicity, and in vivo in human liver cancer xenograft models. The primary endpoint was improved tumor control compared to conventional and other cytokine-engineered CAR-T cells.

Computational predictions from the LLM-guided platform identified IL-15 as the most effective cytokine enhancer, particularly against tumor cells with low GPC3 expression. > IL-15-engineered CAR-T cells exhibited superior proliferation, persistence, and serial cytotoxicity against GPC3-low liver cancer cells in vitro. In human liver cancer xenograft models, IL-15-enhanced CAR-T cells achieved improved tumor control compared with both conventional and other cytokine-engineered CAR-T cells. The study highlighted that the platform successfully recovered known CAR-T cell-relevant cytokine biology, serving as a positive benchmark for the LLM-guided in silico workflow. This validation supports the framework's utility for rational cytokine selection in CAR-T engineering.