Overexpressed tumor-associated antigens [for example, epidermal growth factor receptor (EGFR) and human epidermal growth factor receptor 2 (HER2)] are attractive targets for therapeutic T cells, but toxic "off-tumor" cross-reaction with normal tissues that express low levels of target antigen can occur with chimeric antigen receptor (CAR)-T cells. Inspired by natural ultrasensitive response circuits, we engineered a two-step positive-feedback circuit that allows human cytotoxic T cells to discriminate targets on the basis of a sigmoidal antigen-density threshold. In this circuit, a low-affinity synthetic Notch receptor for HER2 controls the expression of a high-affinity CAR for HER2. Increasing HER2 density thus has cooperative effects on T cells-it increases both CAR expression and activation-leading to a sigmoidal response. T cells with this circuit show sharp discrimination between target cells expressing normal amounts of HER2 and cancer cells expressing 100 times as much HER2, both in vitro and in vivo.

Chimeric antigen receptor (CAR) costimulatory domains derived from native immune receptors steer the phenotypic output of therapeutic T cells. We constructed a library of CARs containing ~2300 synthetic costimulatory domains, built from combinations of 13 signaling motifs. These CARs promoted diverse human T cell fates, which were sensitive to motif combinations and configurations. Neural networks trained to decode the combinatorial grammar of CAR signaling motifs allowed extraction of key design rules. For example, non-native combinations of motifs that bind tumor necrosis factor receptor-associated factors (TRAFs) and phospholipase C gamma 1 (PLCγ1) enhanced cytotoxicity and stemness associated with effective tumor killing. Thus, libraries built from minimal building blocks of signaling, combined with machine learning, can efficiently guide engineering of receptors with desired phenotypes.