机器学习衍生的 NK 细胞特征预测透明细胞肾细胞癌预后与治疗反应

Natural killer cells, interconnected with patient prognosis and treatment response, play a pivotal role in the tumor immune microenvironment and may serve as potential novel predictive biomarkers for renal cell carcinoma.

Clear cell renal cell carcinoma transcriptome data and the corresponding clinical data were obtained from the Cancer Genome Atlas (TCGA) database. Single-cell sequencing data were sourced from the Gene Expression Omnibus (GEO) database. A risk model was established by integrating ten different machine learning algorithms, which resulted in 101 combined models. The model with the highest average C-index was selected for further analysis, and was assessed using nomogram, time-dependent receiver operating characteristics (ROC) and Kaplan-Meier survival analysis. The differences in immune infiltration fractions, clinicopathological features, and response to various targeted therapies and immunotherapy between high- and low-risk groups were investigated. Furthermore, qRT-PCR, IHC, colony formation test, CCK8 assay and flow cytometry were conducted to explore the expression pattern and function of ARHGAP9 in our own patient samples and renal cancer cell lines.

Totally, 156 NK cell-related genes and 5189 prognosis-related genes were identified, and 36 genes of their intersection demonstrated prognostic value. A risk model with 18 genes was established by Coxboost plus plsRcox, which can accurately predict the prognosis of ccRCC patients. Significant correlations were determined between risk score and tumor malignancy and immune cell infiltration. Meanwhile, a combination of tumor mutation burden plus risk score could have higher accuracy of predicting clinical outcomes. Moreover, high-risk group patients were more likely to be responsive to targeted therapy but show no response to immunotherapy.

Intricate signaling interactions between NK cells and various cellular subgroups were depicted and the developmental trajectory of NK cells was elucidated. A NK cells-related risk model was established, which can provide reliable prognostic information and identified patients with more probability of benefiting from therapy.