Decoding CAR T cell phenotype using combinatorial signaling motif libraries and machine learning.

Area of Science:

• Immunology
• Molecular Biology
• Bioengineering

Background:

• Chimeric antigen receptor (CAR) costimulatory domains influence therapeutic T cell function.
• Understanding how signaling motifs dictate T cell fate is crucial for CAR T-cell therapy development.

Purpose of the Study:

• To construct and analyze a large library of synthetic CAR costimulatory domains.
• To identify key design rules governing CAR signaling and T cell phenotypes using machine learning.

Main Methods:

• Creation of a library of approximately 2300 CARs with synthetic costimulatory domains, composed of 13 signaling motifs.
• Utilizing neural networks to decode the combinatorial grammar of CAR signaling motifs.
• Evaluating the impact of different motif combinations on human T cell fates and functions.

Main Results:

• Diverse human T cell fates were observed, sensitive to specific motif combinations and configurations.
• Machine learning models successfully extracted critical design rules for CAR signaling.
• Non-native motif combinations, particularly those involving tumor necrosis factor receptor-associated factors (TRAFs) and phospholipase C gamma 1 (PLCγ1), enhanced T cell cytotoxicity and stemness.

Conclusions:

• Synthetic signaling motif libraries, combined with machine learning, offer an efficient approach to engineer CARs.
• This strategy can guide the development of CARs with desired phenotypes for improved cancer immunotherapy.
• The findings provide a framework for rational design of CAR costimulatory domains.