关键词: 胰腺肿瘤, 预后, RNA,长链非编码, ROC曲线

Abstract:

Objective To identify and screen out immune-related long non-coding RNAs (lncRNAs) in pancreatic cancer, and construct a prognostic risk assessment model to predict the prognostic factors of patients with pancreatic cancer. Methods RNA-Seq data and corresponding clinicopathological and follow-up information of 177 pancreatic cancer patients were downloaded from The Cancer Genome Atlas (TCGA) database. The 177 pancreatic cancer samples were randomly divided into discovery cohort (n=89) and validation cohort (n=88) using random number table method. Immune-related lncRNAs were identified by Pearson correlation coefficient analysis. Univariate and multivariate Cox analysis were used to select prognosis-related lncRNAs and construct the risk score formula based on the data of discovery cohort. The conclusion generated from discovery cohort would be verified in the validation cohort. Results A total of 788 immune-related lncRNAs were screened out using Pearson correlation coefficient calculation formula, and 5 lncRNAs (AC006237.1, AC025154.2, RASSF8-AS1, AL122010.1 and AC073896.3) were selected by univariate and multivariate Cox analysis to build the risk score formula based on the discovery cohort. Based on the above risk score formula, pancreatic cancer patients from the discovery cohort were divided into high-risk group (n=44) and low-risk group (n=45). Survival analysis indicated that the median survival time of high-risk group (1.09 years) was significantly shorter than that of low-risk group (4.11 years; χ²=26.016, P<0.001). The validation cohort was also divided into high-risk group (n=44) and low-risk group (n=44) based on the above risk score formula. Survival analysis showed that the median survival time of high-risk group (1.28 years) was also significantly shorter than that of low-risk group (1.90 years; χ²=4.422, P=0.035). Besides, univariate and multivariate analyses suggested that the prognostic risk assessment model could effectively predict the prognosis of patients with pancreatic cancer, and could be used as an independent prognostic prediction model (HR=2.618, 95%CI: 1.285-5.332, P=0.008). The predictive efficiency of this model was better than that of common clinicopathological information such as age, gender and tumor histopathological grade [1-year area under curve (AUC)=0.687, 3-year AUC=0.725 and 5-year AUC=0.782]. The expression levels of AC025154.2, AC073896.3, AL122010.1 and RASSF8-AS1 were significantly different in various clinical characteristics of pancreatic cancer (all P<0.05), and might serve as potential new targets for diagnosis and treatment in pancreatic cancer. Interferon α, mammalian target of rapamycin complex 1 (mTORC1), MYC regulatory genes and transforming growth factor-β (TGF-β) signaling pathways were significantly activated, and myogenesis and pancreas β cells signaling pathways were significantly suppressed in the high-risk group, which may explain the underlying molecular mechanisms of the prognostic risk assessment model. Conclusion The prognostic risk assessment model based on 5 immune-related lncRNAs can effectively predict the prognosis of pancreatic cancer patients. Besides, the above immune-related lncRNAs may serve as new biomarkers in the diagnosis and therapy in pancreatic cancer.

Key words: Pancreatic neoplasms, Prognosis, RNA,long noncoding, ROC curve