胃癌患者预后相关微RNA预测模型的构建及其应用价值探讨

doi:10.3969/j.issn.1674-8115.2021.11.006

摘要/Abstract

摘要：

目的·通过生物信息学方法建立胃癌患者预后相关微RNA（microRNA，miRNA）的预测模型并探讨其应用价值。方法·收集癌症和肿瘤基因图谱计划（TCGA）数据库中397例胃癌患者的临床病理资料，其中356例临床病理资料完整。397例患者中，男258例，女139例；中位年龄为67岁。将397例患者采用随机抽样法，按7∶3比例分为训练集278例和测试集119例。另从基因表达数据库（Gene Expression Omnibus，GEO）中下载包含20对胃癌及对应癌旁正常组织的miRNA测序数据集GSE93415，从癌组织和癌旁正常组织差异表达的miRNA中筛选出候选差异表达miRNA。基于训练集患者的信息，利用LASSO回归分析，将候选差异表达miRNA拟合成一个可以预测胃癌患者生存率的预后相关miRNA模型。对构建的预后相关miRNA模型的预测效能，分别在训练集和测试集中进行验证，以Log-Rank检验进行生存分析来验证模型的可靠性；以受试者操作特征曲线下面积（area under curve，AUC）分析该模型的预测效能。使用基因表达数据和pRRophetic包中的算法预测患者对化学治疗药物的敏感性。使用Calibration曲线验证恩诺图的准确性。使用一致性指数（consistency index，C-index）检测恩诺图和其他因素建模的一致性。使用决策曲线分析（decision curve analysis，DCA）推测候选因素对于临床决策的帮助。结果·①训练集与测试集患者的临床资料和总体生存率比较，差异均无统计学意义（均P>0.05）。②差异表达miRNA筛选结果：从GSE93415测序数据集中计算得到111个候选差异表达miRNA，其中20个在癌组织中上调，91个在癌组织中下调。对111个候选差异表达miRNA进行过滤后，得到59个候选差异表达miRNA。③构建预后相关miRNA模型：从59个候选miRNA中，筛选出5个与生存相关的miRNA，分别为let-7i-5p、let-7f-5p、miR-708-5p、miR-135b-5p、miR-100-5p，差异表达模式（癌组织对比癌旁组织）均为降低，差异表达倍数分别2.55、2.78、2.17、3.08、3.26倍。由5个与生存相关miRNA构建的预后表达方程为：风险分数=（-0.049×let-7i-5p表达量-0.033 2×let-7f-5p表达量+0.202 9×miR-708-5p表达量-0.088 9×miR-135b-5p表达量+0.016 3×miR-100-5p表达量）。④预后相关miRNA模型的验证：在训练集和测试集中，高风险组患者的总体生存率低于低风险组患者，差异均有统计学意义（均P<0.05）。在训练集中，预后相关miRNA模型的1年、3年、5年生存时间预测概率AUC值分别为0.640、0.763、0.853；在测试集中，则分别为0.631、0.735、0.750。⑤影响胃癌患者预后的相关因素分析：单因素分析结果显示年龄、肿瘤病理分期、T分期、N分期、M分期和预后相关miRNA模型评分是胃癌患者预后的相关因素（HR=1.017、1.633、1.353、1.346、2.652、15.874，95%CI为1.002~1.033、1.333~2.001、1.109~1.650、1.169~1.548、1.553~4.529、5.729~43.985，P<0.05）。多因素分析结果显示年龄、M分期和预后相关miRNA模型评分是胃癌患者预后的独立危险因素（HR=1.03、2.27、18.72，95%CI为1.01~1.05、1.09~4.70、5.96~58.77，P<0.05）。⑥预后相关miRNA模型与临床病理因素预测效能的比较：在356例临床资料完整的胃癌患者中，预后相关miRNA模型对胃癌患者5年生存时间预测的AUC值为0.818，高于年龄、性别、肿瘤病理分期、T分期、N分期、M分期的预测效能。⑦预后相关恩诺图的评估：Calibration验证曲线、C-index以及DCA分析均提示，胃癌患者从预后相关恩诺图中的获益程度高于年龄、性别和肿瘤病理分期。结论·由5个差异表达的miRNA构建了可用于预测胃癌患者生存的预后相关miRNA模型；该模型对胃癌患者的生存状态和预后具有一定的区分预测能力，可为胃癌患者的临床治疗提供参考。

关键词: 胃癌, LASSO回归模型, 微RNA, 预测模型, 受试者操作特征曲线

Abstract:

Objective·To construct a prognosis-associated microRNA (miRNA) prediction model in gastric cancer patients based on bioinformatics analysis and evaluate its application value.

Methods·The clinicopathological data of gastric cancer patients were downloaded from the Cancer Genome Atlas (TCGA). There were 258 males and 139 females with a median age of 67 years. Three hundred and fifty-six of the 397 patients had complete clinicopathological data. The 397 patients were allocated into training cohort consisting of 278 patients and validation cohort consisting of 119 patients using the random sampling method, with a ratio of 7∶3. A miRNA sequencing dataset GSE93415 containing 20 pairs of gastric cancer and corresponding adjacent normal tissue was downloaded from Gene Expression Omnibus database. The candidate miRNAs were selected from differentially expressed miRNAs in gastric cancer and adjacent tissue. A prognosis associated miRNA prediction model was constructed upon survival-associated miRNAs which were selected from candidate miRNAs through LASSO-Cox regression analysis. The performance of prognosis-associated miRNA prediction model was validated in the training cohort and validation cohort. The reliability of the model was evaluated by using Log-Rank test, and the accuracy of the model was evaluated by using the area under curve (AUC) of the receiver operating characteristic curves. Gene expression profiles and algorithms in the pRRophetic package were utilized to predict patients' sensitivity to chemotherapy drugs. Calibration curves were used to verify the accuracy of the nomogram. Consistency index (C-index) was used to check the consistency of the nomogram and other factors. Decision curve analysis (DCA) was employed to predict the contribution of candidate factors to clinical decision making.

Results·① There was no significant difference in the baseline and overall survival between the training cohort and validation cohort (P>0.05). ② There were 111 differentially expressed miRNAs calculated from GSE93415 dataset, of which 20 were up-regulated in tumor tissue while 91 were down-regulated. Fifty-nine miRNAs were selected as candidate miRNAs after filtration. ③ Among the 59 candidate miRNAs, 5 survival-associated miRNAs were selected, including let-7i-5p, let-7f-5p, miR-708-5p, miR-135b-5p and miR-100-5p. The differential expression patterns of gastric cancer to adjacent tissue were all down-regulation, with the fold change of 2.55, 2.78, 2.17, 3.08 and 3.26. Risk score = (-0.049×let-7i-5p expression level -0.033 2×let-7f-5p expression level +0.202 9×miR-708-5p expression level -0.088 9×miR-135b-5p expression level +0.016 3×miR-100-5p expression level). ④ In the training cohort and the validation cohort, the overall survival rate of patients in the high-risk group was lower, and the difference was statistically significant (P<0.05) . The AUC of the prognosis-associated miRNA model for 1-, 3- and 5-year survival prediction was 0.640, 0.763 and 0.853, and was 0.631, 0.735 and 0.750 in validation cohort. ⑤Results of univariate analysis showed that age, tumor pathological stage, T stage, N stage, M stage and prognosis-associated miRNA model score were related factors for prognosis of gastric cancer patients (HR=1.017, 1.633, 1.353, 1.346, 2.652, 15.874; 95%CI 1.002?1.033, 1.333?2.001, 1.109?1.650, 1.169?1.548, 1.553?4.529, 5.729?43.985; P<0.05). Results of multivariate analysis showed that age, M stage and prognosis-associated miRNA model score were independent risk factors for prognosis of gastric cancer patients (HR=1.03, 2.27, 18.72; 95%CI 1.01?1.05, 1.09?4.70, 5.96?58.77; P<0.05). ⑥The AUC of the prognosis-associated miRNA model for 5-year survival prediction was 0.818 in 356 gastric cancer patients with complete clinicopathological data, higher than that of age, gender, tumor pathological stage, T stage, N stage, M stage and merged clinical factors. ⑦The results of Calibration curves, C-index, and DCA all indicated that patients with gastric cancer may get more net benefits from the prognostic nomogram than age, gender and tumor pathological stage.

Conclusion·A prognosis-associated miRNA prediction model that can be used to predict the survival of gastric cancer patients is constructed based on 5 miRNAs, which has a certain predictive ability to distinguish the survival status and prognosis of gastric cancer patients and can provide reference for clinical treatment.

Key words: gastric cancer, LASSO regression model, microRNA, prediction model, receiver operating characteristic curve

中图分类号:

R735.2

岳犇, 王高明, 杨鹿笛, 崔然, 郁丰荣. 胃癌患者预后相关微RNA预测模型的构建及其应用价值探讨[J]. 上海交通大学学报(医学版), 2021, 41(11): 1436-1445.

Ben YUE, Gao-ming WANG, Lu-di YANG, Ran CUI, Feng-rong YU. Construction and application value of prognosis-associated miRNA prediction model in gastric cancer patients[J]. JOURNAL OF SHANGHAI JIAOTONG UNIVERSITY (MEDICAL SCIENCE), 2021, 41(11): 1436-1445.

图/表 7

表1 训练集与测试集患者一般资料比较

Tab 1 Baseline of patients in the training and validation cohorts

Characteristics	Training cohort	Validation cohort	Statistic	Pvalue
Number of patients/n	278	119
Gender/n
Male	179	79	χ²=0.072	0.789
Female	99	40
Age/year	65±10	64±10	t=0.736	0.463
Stage/n
Ⅰ	30	19	χ²=2.738	0.603
Ⅱ	92	34
Ⅲ	117	52
Ⅳ	27	10
Not report	12	4
Race/n
Asian	61	22	χ²=1.294	0.731
Black	7	5
White	173	76
Not report	37	16
Pharmaceutical treatment/n
Yes	139	49	χ²=3.011	0.222
No	121	63
Not report	18	7
Radiation treatment/n
Yes	62	18	χ²=2.783	0.249
No	201	95
Not report	15	6
Site/n
Body of stomach	67	24	χ²=6.418	0.268
Cardia	75	26
Fundus of stomach	27	21
Gastric antrum	99	43
Lesser curvature of stomach	1	0
Other	9	5
Status/n
Alive	169	67	χ²=0.523	0.470
Dead	109	52
Follow up/d	495 （275‒810）	398 （275‒657）	W=18 013	0.160

图1 278例训练集和119例测试集胃癌患者的生存曲线Note： The dotted line is the median.

Fig 1 Survival curves of patients with gastric cancer in the training and validation cohorts

图2 LASSO-Cox回归分析筛选miRNAsNote： A. Lasso-Cox regression analysis was performed based on 59 differentially expressed miRNAs and the training cohort. Different colored lines represent different candidate miRNAs. The horizontal axis above represents the number of variables, and the vertical axis represents the coefficient of variables. B. The 10-fold cross validation method was used to determine the modeling position. The partial likelihood deviation tended to be stable with a λ of 0.062, so this point was selected for modeling. The upper horizontal axis represents the number of variables and the vertical axis represents the partial likelihood deviation. C. Cox analysis of 5 selected miRNAs. D. Kaplan-Meier survival curves of 5 selected miRNAs. Patients with high expression of let-7F-5p, hsa-let-7i-5p and hsa-miR-135b-5p and low expression of hsa-miR-100-5p and hsa-miR-708-5p in gastric cancer samples had better overall survival.

Fig 2 miRNAs screened by Lasso-Cox regression analysis

图3 训练集预后相关miRNA模型评分、时间依赖ROC分析以及Kaplan-Meier生存分析Note： A. The distribution of risk score. B. The survival status and survival time of gastric cancer patients. C. Heatmap of the expression distribution of 5 selected miRNAs. D. Kaplan-Meier survival curves. E. Time-dependent ROC analysis.

Fig 3 Prognostic miRNA model score, time-dependent ROC analysis, and Kaplan-Meier survival analysis of the training cohort

图4 测试集预后相关miRNA模型评分、时间依赖ROC分析以及Kaplan-Meier生存分析Note： A. The distribution of risk score. B. The survival status and survival time of gastric cancer patients. C. Heatmap of the expression distribution of 5 selected miRNAs. D. Kaplan-Meier survival curves. E. Time-dependent ROC analysis.

Fig 4 Prognostic miRNA model score, time-dependent ROC analysis, and Kaplan-Meier survival analysis of the validation cohort

图5 影响胃癌患者预后的相关因素分析Note： A. Univariate Cox regression analysis of prognostic factors in patients with gastric cancer. B. Multivariate Cox regression analysis of prognostic factors in patients with gastric cancer. C. ROC analysis of the risk score and other clinical factors. D. The estimated IC50 of cisplatin between low- and high-risk groups.

Fig 5 Analysis of prognostic factors in patients with gastric cancer

图6 预后相关恩诺图的建立及评估Note： A. Prognostic nomogram for patients with gastric cancer. ①P<0.01; ②P<0.001. B. Calibration curves to validate the accuracy of the nomogram. C. C-index of the nomogram and traditional factors. D. DCA curves.

Fig 6 Construction and evaluation of the prognostic nomogram

参考文献 23

1	Smyth EC, Nilsson M, Grabsch HI, et al. Gastric cancer[J]. Lancet, 2020, 396(10251): 635-648.
2	Orditura M, Galizia G, Sforza V, et al. Treatment of gastric cancer[J]. World J Gastroenterol, 2014, 20(7): 1635-1649.
3	Shin VY, Chu KM. MiRNA as potential biomarkers and therapeutic targets for gastric cancer[J]. World J Gastroenterol, 2014, 20(30): 10432-10439.
4	Lu TX, Rothenberg ME. MicroRNA[J]. J Allergy Clin Immunol, 2018, 141(4): 1202-1207.
5	Lee YS, Dutta A. MicroRNAs in cancer[J]. Annu Rev Pathol, 2009, 4: 199-227.
6	Tutar Y. miRNA and cancer; computational and experimental approaches[J]. Curr Pharm Biotechnol, 2014, 15(5): 429.
7	Colaprico A, Silva TC, Olsen C, et al. TCGAbiolinks: an R/Bioconductor package for integrative analysis of TCGA data[J]. Nucleic Acids Res, 2016, 44(8): e71.
8	Barrett T, Wilhite SE, Ledoux P, et al. NCBI GEO: archive for functional genomics data sets--update[J]. Nucleic Acids Res, 2013, 41(Database issue): D991-995.
9	Love MI, Huber W, Anders S. Moderated estimation of fold change and dispersion for RNA-seq data with DESeq2[J]. Genome Biol, 2014, 15(12): 550.
10	Engebretsen S, Bohlin J. Statistical predictions with glmnet[J]. Clin Epigenetics, 2019, 11(1): 123.
11	Bose E, Maganti S, Bowles KH, et al. Machine learning methods for identifying critical data elements in nursing documentation[J]. Nurs Res, 2019, 68(1): 65-72.
12	Zhou ZR, Wang WW, Li Y, et al. In-depth mining of clinical data: the construction of clinical prediction model with R[J]. Ann Transl Med, 2019, 7(23): 796.
13	Kamarudin AN, Cox T, Kolamunnage-Dona R. Time-dependent ROC curve analysis in medical research: current methods and applications[J]. BMC Med Res Methodol, 2017, 17(1): 53.
14	Geeleher P, Cox N, Huang RS. pRRophetic: an R package for prediction of clinical chemotherapeutic response from tumor gene expression levels[J]. PLoS One, 2014, 9(9): e107468.
15	Ritchie ME, Phipson B, Wu D, et al. Limma powers differential expression analyses for RNA-sequencing and microarray studies[J]. Nucleic Acids Res, 2015, 43(7): e47.
16	Ke Y, Zhao W, Xiong J, et al. Downregulation of miR-16 promotes growth and motility by targeting HDGF in non-small cell lung cancer cells[J]. FEBS Lett, 2013, 587(18): 3153-3157.
17	Yang M, Tang X, Wang Z, et al. miR-125 inhibits colorectal cancer proliferation and invasion by targeting TAZ[J]. Biosci Rep, 2019,39(12): BSR20190193.
18	Hu W, Lei L, Xie X, et al. Heterogeneous nuclear ribonucleoprotein L facilitates recruitment of 53BP1 and BRCA1 at the DNA break sites induced by oxaliplatin in colorectal cancer[J]. Cell Death Dis, 2019, 10(8): 550.
19	鲁艳明, 王月, 周梦雅, 等. miR-125b靶基因鉴定及对卵巢癌细胞增殖和凋亡能力影响实验研究[J].中华肿瘤防治杂志, 2018, 25(1): 8-14.
20	Masood N, Basharat Z, Khan T, et al. Entangling relation of micro RNA-let7, miRNA-200 and miRNA-125 with various cancers[J]. Pathol Oncol Res, 2017, 23(4): 707-715.
21	Ssran MO, Meli LE, Dobru ED. MicroRNA modulation of host immune response and inflammation triggered by Helicobacter pylori[J]. Int J Mol Sci, 2021, 22(3): 1406.
22	Wang Z, Yao W, Li K, et al. Reduction of miR-21 induces SK-N-SH cell apoptosis and inhibits proliferation via PTEN/PDCD4[J]. Oncol Lett, 2017, 13(6): 4727-4733.
23	Lai KW, Koh KX, Loh M, et al. MicroRNA-130b regulates the tumour suppressor RUNX3 in gastric cancer[J]. Eur J Cancer, 2010, 46(8): 1456-1463.

[1]	马江磊, 李晓瑶, 赵世富, 杨德君. 胃癌临床分期诊断方法的应用进展[J]. 上海交通大学学报(医学版), 2021, 41(6): 821-825.
[2]	万淑君, 孔祥, 吕坤. 非编码RNA与糖尿病血管病变的关系[J]. 上海交通大学学报(医学版), 2021, 41(5): 665-670.
[3]	顾琦晟, 张米粒, 曹灿, 李继坤. 基于TCGA数据库分析胃癌可变剪接与肿瘤免疫的关系[J]. 上海交通大学学报(医学版), 2021, 41(4): 448-458.
[4]	林梁俊, 王卫娣, 王佩, 林关宁, 王振. 强迫症的表观遗传学研究进展[J]. 上海交通大学学报(医学版), 2021, 41(2): 267-272.
[5]	王昱欢, 丁奕岑, 蔡瑶雨, 康亚妮. 差异表达微RNA作为多囊卵巢综合征生物标志物的研究[J]. 上海交通大学学报(医学版), 2021, 41(11): 1429-1435.
[6]	奚黎婷, 朱锦舟, 虞晨燕, 倪柳菁, 许春芳, 吴爱荣. 急性非静脉曲张性上消化道出血患者再出血预测模型和新型评分系统的构建[J]. 上海交通大学学报(医学版), 2021, 41(11): 1491-1497.
[7]	张靖. 川楝素通过下调环状RNA circDLST对胃癌细胞BGC-823的抑制作用[J]. 上海交通大学学报(医学版), 2020, 40(09): 1202-1206.
[8]	何沁，张伟滨，沈宇辉. 乳腺癌脊柱转移患者预后预测模型的构建[J]. 上海交通大学学报(医学版), 2020, 40(09): 1243-1248.
[9]	吕恒宇，黄晨，夏翔，赵刚. 预测根治性胃癌切除术后并发症危险因素的列线图模型的建立[J]. 上海交通大学学报(医学版), 2020, 40(07): 894-900.
[10]	吴珈悦，蒋萌，林思涵，狄文. 系统性红斑狼疮妊娠丢失预测模型的建立与验证[J]. 上海交通大学学报(医学版), 2020, 40(07): 908-914.
[11]	窦敏1, 2，郑英霞2，韩丽2，赵倩1. PRMT4在胃癌发生和发展中的作用和机制研究[J]. 上海交通大学学报(医学版), 2020, 40(05): 609-618.
[12]	余志龙，戎泽印，罗再，章建明，黄陈. 基于高通量测序对人胃癌组织长链非编码RNA差异表达的分析[J]. 上海交通大学学报（医学版）, 2019, 39(9): 955-.
[13]	袁园，糜军. 谷丙转氨酶2对胃癌顺铂治疗抵抗的影响及其作用机制[J]. 上海交通大学学报（医学版）, 2019, 39(10): 1134-.
[14]	孙颖1，胡梅洁2，顾玮1，李健1，王吉1，郑雄1. lincRNA-BBOX1-2在胃癌组织中的表达及临床意义[J]. 上海交通大学学报（医学版）, 2019, 39(10): 1178-.
[15]	王振寰 1，黄永辉 1，林艳萍 1，陈华江 2，张颖 2，袁文 2. 骨组织钻削过程有限元模拟研究进展[J]. 上海交通大学学报（医学版）, 2018, 38(5): 561-.