Abstract: Objective This study aimed to investigate impact of immune microenvironment feature on liver metastasis in patients with colorectal cancer (CRC) and to construct a predictive model based on immune biomarkers to assess the risk of liver metastasis. Methods 216 patients with CRC, including liver metastasis in 92 cases and no liver metastasis in 124 cases, were encountered in our hospital beween January 2019 and December 2022, and all underwent surgery for resection of tumors. Pathological examination determined lymphovascular invasion (LVI), perineural invasion (PNI) and tumor-stroma ratio (TSR). Immunohistochemistry was performed for Ki67, p53, CD8, PD-L1, FOXP3, galectin-9, VISTA, CD39, CCR8 and CXCL13 expression, and semiquantitative H-scores were calculated. The effector score (CD8+CXCL13), suppressor score (FOXP3+PD-L1), and dysregulation score were subsequently constructed. Multivariate Logistic regression analysis was performed to identify independent factors, and an XGBoost model was developed based on selected variables to assess predictive performance, and decision curve analysis (DCA) was established to determine its clinical application. Results The Ki67 index was 62.0 (51.0-71.0)%, much higher than 39.0 (31.8-54.0)% in non-liver metastasis group (P<0.05); positive rates of LVI and PNI in the liver metastasis group were 65.2% and 44.6%, both significantly higher than 1.6% and 2.4% (P<0.05) in the non-liver metastasis group; the CD8 H-score was much lower than that in the non-liver metastasis group [130.5 (87.5-183.0) vs. 203.5 (144.0-255.5), P<0.05], while the H-scores of PD-L1, FOXP3, Galectin-9, VISTA, CD39 and CCR8 were higher than those in the non-liver metastasis group (all P<0.05); the effector score was 272.0 (208.0-330.0), much lower than that in the non-liver metastasis group [393.5 (298.2-460.5), P<0.05]; the suppressor score was 387.5 (315.0-461.0), much higher than that in the non-liver metastasis group [260.5 (198.5-321.2), P<0.05]; multivariate Logistic regression analysis showed that CXCL13 (OR = 0.04, 95% CI: 0.01-0.10, P<0.05) and dysregulation score (OR=0.29, 95% CI: 0.09-0.84, P<0.05) were protective factors, while CCR8 (OR=2.15, 95% CI: 1.27-3.92, P<0.05) and VISTA (OR=1.85, 95% CI: 1.03-3.46, P<0.05) were the risk factors; the XGBoost model predicted liver metastasis with an AUC of 0.828 (P<0.05), and within the threshold range of 1% to 43%, the net benefit was superior to the "treat all" and "treat none" strategies. Conclusion Patients with CRC and liver metastasis exhibit an immunosuppressive microenvironment characterized by decreased effector factors and increased suppressor factors, which might promote metastasis. The XGBoost model we constructed based on immune biomarkers demonstrate a good predictive performance and clinical application, which provide a reference for early prediction and individualized management of patients with CRC.

Key words: Liver metastasis, Colorectal cancer, Tumor microenvironment, PD-L1, FOXP3, XGBoost, Prediction