P08.18 A single-cell and machine learning-derived prognostic signature of CD8 + exhausted T cells for colorectal cancer liver metastasis

The tumor microenvironment (TME) is pivotal in the progression and treatment resistance of colorectal cancer liver metastasis (CRLM). The role of CD8 + exhausted T cells (CD8+ Tex) in CRLM remains incompletely understood. This study aimed to construct a single-cell atlas of CRLM to decipher the cellular interactions of CD8+ Tex and develop a prognostic model for clinical application.We integrated single-cell and bulk transcriptomic data from GEO and TCGA. Using Seurat and CellChat, we characterized the cellular composition and communication in CRLM TME. Metastasis-associated differentially expressed genes were identified by comparing CD8+ Tex from liver metastases and primary tumors. A prognostic signature was built via LASSO-Cox regression and validated in CRC cell lines using RT-qPCR.The CRLM single-cell atlas revealed marked enrichment of CD8+ Tex in liver metastases. CellChat analysis indicated active cancer-associated fibroblasts (CAFs)–CD8+ Tex crosstalk mediated by collagen signaling. A six-gene prognostic signature (FKBP5, CMC1, NDUFC1, TMSB10, SERF2, SNW1) was established and validated, showing consistent predictive power. Differential gene expression was confirmed in CRC cells. A nomogram combining risk score and clinical features was developed to facilitate prognosis assessment and personalized treatment.Our study elucidates an immunosuppressive CRLM microenvironment driven by CD8+ Tex–CAFs interactions and presents a clinically applicable CD8+ Tex–based prognostic signature. These insights enhance the understanding of immune evasion in CRLM and offer a potential tool for prognosis and targeted therapy development.