Due to the high recurrence, the HCC prognosis remains poor. Yet, the biomarkers for predicting the recurrence of high-risk patients are currently lacking. We aimed to develop a signature to predict the recurrence of HCC based on NKG2D ligands.The multivariate Cox proportional hazards regression was used to select recurrence-related variables of NKG2D ligands in HCC patients from The Cancer Genome Atlas (TCGA). HCC patients from the OEP000321 dataset and Guilin cohort were used to validate the predictive signature. The mRNA expression of NKG2D ligands was measured by QRT-PCR. Immunohistochemistry analysis of HCC tissue microarray samples was used to identify the expression of NKG2D ligands.In this study, NKG2D ligands expression in the mRNA and protein level was both abnormally expressed in HCC and associated with recurrence-free survival (RFS). Then, the recurrence-related variables of NKG2D ligands in HCC were selected by the multivariate Cox proportional hazards regression. Among the eight NKG2D ligands, MICA (HR = 1.347; 95% CI = 1.012-1.793; p = 0.041), ULBP3 (HR = 0.453; 95% CI = 0.231-0.889; p = 0.021) and ULBP5 (HR = 3.617; 95% CI = 1.819-7.194; p 0.001) were significantly correlated with RFS in the TCGA-LIHC cohort. Then, the signature was constructed by the three NKG2D ligands. The predictive effectiveness of this signature was also validated in the OEP000321 dataset and Guilin cohort. Further, HCC patients were classified into low-risk and high-risk subgroups by the predictive score. Compared with the low-risk group, the high-risk group had poor RFS in both training and validation cohorts. Importantly, compared with the low-risk patients with the G1-G2 stage, the levels of infiltrated NK-activated cells and NKG2D expression were both lower in the high-risk patients.The signature based on MICA, ULBP3, and ULBP5 could predict HCC recurrence.
