基于放化疗敏感性相关基因的原发胶质母细胞瘤预后预测模型的构建和验证

Construction and verification of a prognostic prediction model for primary glioblastomas based on genes related to radiotherapy and chemotherapy sensitivity

目的筛选原发胶质母细胞瘤放化疗敏感性相关基因,基于相关基因构建原发胶质母细胞瘤患者的预后预测模型并验证。方法回顾性分析中国脑胶质瘤基因组图谱计划(CGGA)2019数据库(102例,试验组)和CGGA数据库(54例,验证组)中术后接受规范放化疗的原发胶质母细胞瘤患者的临床资料及转录组测序数据。在试验组中筛选长生存期亚组(≥18个月,49例)与短生存期亚组(≤9个月,22例)患者的差异基因,进一步将患者的年龄、肿瘤异柠檬酸脱氢酶(IDH)突变状态、染色体1p/19q共缺失状态、O6-甲基鸟嘌呤DNA甲基转移酶(MGMT)启动子区甲基化状态以及差异基因均纳入多因素Cox回归分析,筛选其中为独立预后因素的目标差异基因,取其风险比的自然对数作为各基因相应的系数,计算预后风险评分。采用单因素和多因素Cox回归分析法评估预后风险评分是否为独立预后因素;采用Kaplan-Meier法绘制生存曲线,log-rank检验比较高风险组(预后风险评分>0分)与低风险组(预后风险评分≤0分)患者生存期的差异。通过Pearson相关性分析筛选与风险评分呈正相关的基因,并对其进行功能富集分析。结果试验组中,长生存期亚组与短生存期亚组患者的性别、年龄、肿瘤切除程度、IDH突变状态、染色体1p/19q缺失状态以及MGMT启动子甲基化状态的差异均无统计学意义(均P>0.05)。长生存期亚组与短生存期亚组比较,9个基因表达量显著升高(均P<0.05),28个基因表达量显著降低(均P<0.05)。采用多因素Cox回归分析法筛选出4个目标差异基因,分别为AKR1C1(HR=0.910,95%CI:0.850~0.974,P=0.006)、CPZ(HR=0.947,95%CI:0.898~0.999,P=0.044)、HIST1H3H(HR=1.299,95%CI:1.025~1.647,P=0.031)以及TBX5(HR=1.104,95%CI:1.034~1.179,P=0.003),其对应的预测模型中的系数分别为-0.0947、-0.0547、0.2624、0.0994。试验组和验证组中,预后风险评分(高风险)均为预后的独立危险因素(试验组中,HR=2.407,95%CI:1.470~3.939,P<0.001;验证组中,HR=2.054,95%CI:1.101~3.830,P=0.024)。试验组和验证组中,高风险亚组(分别为51、22例)患者的总生存期均比低风险亚组(分别为51例、32例)短,差异均有统计学意义(均P<0.05)。功能富集分析结果显示,在试验组与验证组中,与预后风险评分正相关的基因更多地富集在细胞增殖、基因表观遗传学调控、DNA修复及核因子κB信号通路的激活等生物学功能上。结论基于放化疗敏感性相关基因AKR1C1、CPZ、HIST1H3H以及TBX5构建的预后预测模型或可用于预测原发胶质母细胞瘤患者的预后。

Objective To screen the genes related to radiotherapy and chemotherapy sensitivity of primary glioblastomas and to construct a prognostic prediction model for primary glioblastomas based on those genes.Methods A retrospective analysis was conducted on the clinical data and transcriptome sequencing data of 156 patients with primary glioblastomas who underwent standardized radiotherapy and chemotherapy post operation which were obtained from the Chinese Glioma Genome Atlas(CGGA)Project 2019 database(n=102,experimental group)and CGGA database(n=54,validation group).In the experimental group,the differentially expressed genes between tumors of long-term survivors(n=49,overall survival≥18 months)and short-term survivors(n=22,overall survival≤9 months)were screened out.Subsequently,candidate genes with prognostic values independent of age,tumor isocitrate dehydrogenase(IDH)mutation status,chromosome 1p/19q co-deletion status,and O6-methylguanine DNA methyltransferase(MGMT)promoter methylation status were further screened by multivariate Cox regression analysis.A prognostic prediction model was constructed based on the natural logarithm of the hazard ratio of candidate genes.Whether the high-risk group was an independent prognostic factor was tested by univariate and multivariate Cox regression analyses.Kaplan-Meier method was used to draw the survival curve of the high-risk group and low-risk group.The difference in survival curve between high-risk group(>0 point)and low-risk group(≤0 point)was tested by log-rank test.Pearson correlation analysis was used to screen genes that were positively related to risk scores,and functional enrichment analysis was conducted for them.Results There were no statistically differences in the gender,age,degree of tumor resection,IDH mutation status,chromosome 1p/19q deletion status,or MGMT promoter methylation status between the long-term survivors and short-term survivors(all P>0.05).Compared with the short-term survivors,the expressions of 9 genes were significantly increased(all P<0.05)and the expressions of 28 genes were significantly decreased(all P<0.05)in the tumors of long-term survivors.Four candidate genes were selected out by multivariate Cox regression analysis method,which were AKR1C1(HR=0.910,95%CI:0.850-0.974,P=0.006),CPZ(HR=0.947,95%CI:0.898-0.999,P=0.044),HIST1H3H(HR=1.299,95%CI:1.025-1.647,P=0.031)and TBX5(HR=1.104,95%CI:1.034-1.179,P=0.003).The coefficients of the corresponding genes were-0.0947,-0.0547,0.2624 and 0.0994.In the experimental group and validation groups,the prognostic risk score(high risk)was an independent risk prognostic factor(HR=2.407,95%CI:1.470-3.939,P<0.001 in experimental group,HR=2.054,95%CI:1.101-3.830,P=0.024 in validation group).In the experimental group and validation group,the high-risk group(n=51 and 22 respectively)had shorter overall survival than low-risk group(n=51 and 32 respectively),and the differences were statistically significant(both P<0.05).In the experimental and validation groups,genes positively related to the prognostic risk score were enriched in biological functions such as cell proliferation,gene epigenetic regulation,DNA repair,and activation of the nuclear factor κB signaling pathway by functional enrichment analysis.Conclusion The prognostic prediction model of primary glioblastomas based on genes related to radiotherapy and chemotherapy sensitivity(AKR1C1,CPZ,HIST1H3H and TBX5)could be used to predict the prognosis of primary glioblastoma patients.