Comprehensive microRNA profiling coupled with function-based feature selection reveals a biomarker panel for predicting CAR-T cell exhaustion.

Abstract

Chimeric antigen receptor (CAR)-T cell exhaustion limits durable therapeutic efficacy, particularly under persistent antigen stimulation. While transcriptomic and epigenomic analyses have advanced the understanding of CAR-T cell exhaustion, the contribution of microRNAs (miRNAs) remains poorly characterized. Here, we present the first comprehensive landscape of miRNA expression in exhausted CAR-T cells generated using an in vitro repeated antigen stimulation model. Bulk miRNA sequencing identified 39 differentially expressed miRNAs between exhausted and control CAR-T cells. Subsequent reverse transcription-quantitative polymerase reaction validation reduced the candidate list to 18 miRNAs, which retained enrichment in pathways associated with cellular proliferation. To select an optimal biomarker panel for predicting exhaustion, we further applied functional analysis-based feature selection to minimize pathway redundancy, resulting in a six-miRNA panel. Machine learning models using these miRNAs achieved superior predictive performance (area under the curve = 0.958) compared with larger panels. Our findings identify miRNAs as key molecular hallmarks of CAR-T cell exhaustion and establish a rational framework for biomarker panel selection, with potential applications in CAR-T cell quality control, therapeutic response prediction, and manufacturing optimization.