基于生物信息学方法的喉鳞癌相分离相关基因调控机制分型分析与预后预测模型构建

Regulation mechanism typing and prognosis prediction model construction of laryngeal squamous cell carcinoma using liquid-liquid phase separation-related genes based on bioinformatic analysis

目的根据喉鳞癌患者转录组数据和临床参数,分析喉鳞癌相分离相关基因调控机制分型,并建立用于预测喉鳞癌患者生存时间的预后预测模型。方法从TCGA数据库下载111例喉鳞癌、12例正常组织转录组测序数据作为TCGA队列,以本实验室自有的107例喉鳞癌及其对应癌旁组织转录组测序数据作为自测队列,进行生物信息学分析。使用R语言DESeq2分析癌组织中的差异表达的相分离相关基因。使用GSEA Preranked算法对每例患者个体进行调控机制评分,使用一致性聚类对患者进行调控机制分子分型;使用STRING在线数据库构建相分离相关基因互作网络,使用cluster Profiler对相分离相关基因进行富集分析;使用Cox回归法从喉鳞癌表达差异的相分离相关基因中筛选与喉鳞癌患者生存时间相关的基因,以此建立喉鳞癌预后预测模型,采用Kaplan-Meier显示相分离相关基因对喉鳞癌患者生存时间的影响;使用LASSO算法构建相分离相关基因喉鳞癌患者预后预测模型。基于预后预测模型和喉鳞癌患者的临床参数构建诺莫图,评估预后预测模型对喉鳞癌患者生存时间的预测能力。结果从两个独立患者队列中共筛选出105个在喉鳞癌组织中呈显著差异表达的相分离相关基因。基于GSEA Preranked算法进行调控机制评分,将喉鳞癌患者分成3个具有不同调控机制的亚型。经过生存分析,13个相分离相关基因(TFRC、SLC7A5、CPEB3、PIMREG、KIF2C、SERPINH1、POU4F1、PRKAA2、SLC3A2、LGALS7B、PIWIL2、SPANXC和TUBB3)与喉鳞癌患者总生存率相关。预后预测模型公式为:0.2067×Ex_(TUBB3)+0.0530×Ex_(SPANXC)+(-0.0129)×Ex_(PIWIL2)+(-0.0897)×Ex_(LGALS7B)+0.1047×Ex_(SERPINH1)+(-0.1032)×Ex_(KIF2C)+(-0.0626)×Ex_(PIMREG)+(-0.1744)×Ex_(CPEB3)+0.1279×Ex_(TFRC)。根据模型对患者进行危险因子评分,将患者分为高、低风险组。高风险组总生存率低于低风险组(HR=5.47,95%CI为2.68~11.15,P<0.01);ROC曲线分析显示,模型风险评分预测喉鳞癌患者1、2、5年总生存率的AUC分别为0.650、0.794和0.727。高危险组无进展间隔率(HR=1.02,95%CI为1.00~1.04,P<0.05)和疾病特异性生存率(HR=1.02,95%CI为1.00~1.04,P<0.05)低于低危险组;ROC曲线分析显示,模型风险评分对喉鳞癌患者1、2、5年的无进展间隔率预测的AUC分别为0.697、0.687和0.71,对疾病特异性生存率预测的AUC分别为0.649、0.654和0.725。诺莫图决策曲线显示,预后预测模型对喉鳞癌患者生存时间具有良好的预测能力。结论喉鳞癌相分离相关基因调控机制可分为3种不同亚型;基于喉鳞癌组织中差异表达的相分离相关基因,成功构建了预后预测模型。

Objective To analyze the regulation mechanism typing of liquid-liquid phase separation related genes(LLPSGs)of laryngeal squamous cell carcinoma(LSCC)and to establish a LLPSGs-related model that can be used to predict the prognosis survival of LSCC patients based on their transcriptome data and clinical features.Methods The transcriptome data consisting of 111 LSCC and 12 normal tissues obtained from TCGA database and RNA-Seq data consisting of 107 paired LSCC and normal tissues from our laboratory were deeply analyzed using bioinformatics methods.DESeq2program in R was utilized to screen the differentially expressed LLPSGs.GSEAPreranked algorithm was employed to mark the scores of the regulated mechanism in individual levels.Regulatory mechanism molecular typing was performed using the consensus clustering.The protein-protein interaction networks among LLPSGs were constructed in STRING online database.The enrichment analysis was performed by clusterProfiler package in R.Cox regression was used to screen the genes related to the survival time of LSCC patients from the differentially expressed LLPSGs between LSCC and normal tissues,and then we built the prognosis prediction model of LSCC.Kaplan-Meier was also used to analyzed the effects of LLPSGs on the survival time of LSCC patients.LLPSGs-related prognosis prediction model for patients with LSCC was established using LASSO algorithm.The nomogram was finally constructed based on LLPSGs-related prognosis prediction model and clinical features,and was used to assess the predictive ability of prognosis prediction model on survival time.Results A total of 105 LLPSGs with significantly differential expression in LSCC tissues were screened out from two independent cohort.Based on enrichment scores of individuals given by GSEAPreranked algorithm,patients with LSCC were divided into three subtypes with different regulatory mechanisms.Thirteen LLPSGs were identified to tightly correlate with prognosis of LSCC patients,including TFRC,SLC7A5,CPEB3,PIMREG,KIF2C,SERPINH1,POU4F1,PRKAA2,SLC3A2,LGALS7B,PIWIL2,SPANXC and TUBB3 though survival analysis.The risk score for each patient could be calculated with the formula:0.2067×Ex_(TUBB3)+0.0530×Ex_(SPANXC)+(-0.0129)×Ex_(PIWIL2)+(-0.0897)×Ex_(LGALS7B)+0.1047×Ex_(SERPINH1)+(-0.1032)×Ex_(KIF2C)+(-0.0626)×Ex_(PIMREG)+(-0.1744)×Ex_(CPEB3)+0.1279×Ex_(TFRC).Based on the risk score,LSCC patients could be divided into the high risk group and low risk group.Overall survival of patients in the high risk group was lower than that in the low risk group(HR=5.47,95%CI:2.68-11.15,P<0.01).Results of ROC curves showed that AUC of risk model for 1-year,2-year,and 5-year overall survival prediction was 0.650,0.794 and 0.727.Progressionfree interval(HR=1.02,95%CI:1.00-1.04,P<0.05)and disease-specific survival(HR=1.02,95%CI:1.00-1.04,P<0.05)probability of LSCC patients in the high risk group were lower than those in the low risk group.Results of ROC curves showed that AUC of risk model for 1-,2-and 5-year progression-free interval prediction was 0.697,0.687 and 0.71,and that for disease-specific survival prediction was 0.649,0.654 and 0.725,respectively.Decision curve of nomogram showed that prognosis prediction model has good ability to predict the survival time of LSCC patients.Conclusions The regulation mechanism typing of LLPSGs of LSCC may be divided into three subtypes.Based on differentially expressed LLPSGs in the laryngeal tissues,we have successfully constructed the prognostic prediction model.