基于生物信息学及机器学习算法筛选前列腺癌的关键表达基因及相关免疫细胞浸润分析

Bioinformatics and machine learning algorithm were used to screen key expressed genes and analyze related immune cell infiltration in prostate cancer

目的通过生物信息学及机器学算法筛选及分析前列腺癌的关键表达基因,探究诊断前列腺癌的生物标志物及与前列腺癌免疫细胞浸润的相关性。方法使用生物信息学方法从基因表达谱(GEO)数据库中下载3个前列腺癌组织信使RNA(mRNA)芯片数据集:其中GSE46602和GSE69334作为训练集,GSE32571作为验证集。对数据集GSE46602及数据集GSE69223两个数据集进行合并分析后获得差异表达基因(DEGs),京都基因与基因组百科全书(KEGG)、基因本体论(GO)、疾病富集分析(DO)与基因富集分析(GSEA)用于功能富集分析。Lasso回归筛选特征基因11个,支持向量机(SVM)筛选特征基因2个,取交集为两个特征基因丝氨酸蛋白酶(HPN)、角蛋白23(KRT23),将两个基因在数据集GSE32571中进行验证,同时通过实时荧光定量聚合酶链反应在前列腺癌相关细胞系中进行验证,最后进一步分析了两个特征基因与免疫细胞浸润相关联系,两组间使用Student’s t检验评估统计学意义。结果通过对GEO数据库3个前列腺癌数据集使用R语言及机器学习等方法进行分析,总共发现35个DEGs和两个核心基因,其中20个为下调基因,15个为上调基因。通过GO、KEGG、DO及GSEA通路分析发现这些基因富集在表皮细胞分化、角质形成等功能中,以细胞外基质受体相互作用及雌激素受体通路等信号上。通过套索算法(LASSO)及SVM筛选出的特征基因与数据集GSE32571进行检验,发现HPN、KRT23是前列腺癌的2个诊断生物标志物,在前列腺腺癌细胞系中mRNA水平进行验证符合生物信息分析结果,HPN在Du145、PC3、Vcap、Lncap、C4-2、22RV1组相对表达量(1.10±0.29、0.46±0.12、3.02±0.79、1.58±0.09,0.39±0.02,0.41±0.07)指标高于RWPE1组(0.09±0.01),差异有统计学意义(t=6.000、5.030、6.400、27.980、15.600、6.870,P<0.05),KRT23在Du145、PC3、Vcap、Lncap、C4-2、22RV1组相对表达量(0.42±0.01、0.15±0.03、0.15±0.02、0.15±0.03、0.62±0.09、0.04±0.01)指标低于RWPE1组(1.01±0.19),差异有统计学意义(t=5.210、7.600、7.620、7.580、3.120、8.630,P<0.05),且HPN、KRT23与免疫细胞相关,HPN与T细胞CD8、静息肥大细胞、静息树突状细胞呈负相关,与巨噬细胞M0呈正相关;KRT23与巨噬细胞M0呈负相关,与静息树突状细胞、静息肥大细胞呈正相关。结论HPN、KRT23可以作为前列腺癌的诊断性生物标志物,且HPN、KRT23与滤泡辅助性T细胞及调节性T细胞等免疫细胞相关。

Objective Through bioinformatics and machine science algorithm to screen and analyze the key expression genes of prostate cancer,explore the biomarkers for the diagnosis of prostate cancer and the correlation with immune cell infiltration of prostate cancer.Methods Three prostate cancer tissue mRNA microarray datasets were downloaded from the gene expression profile(GEO)database by bioinformatics methods:GSE46602 and GSE69334 were used as training sets,and GSE32571 as validation sets.Differential expression genes(DEGs)were obtained by combining data sets GSE46602 and GSE69223.Kyoto Encyclopedia of Genes and Genomes(KEGG),Gene Ontology(GO),disease enrichment analysis(DO)and gene enrichment analysis(GSEA)were used for functional enrichment analysis.Lasso gene 11 regression filter characteristics,support vector machine(SVM)gene 2 filter characteristics,characteristics of intersection for two gene characteristics of intersection for two gene hepsin(HPN),keratin23(KRT23),the two genes in the data set GSE32571 verification,At the same time,real-time fluorescence quantitative polymerase chain reaction was carried out to verify the relationship between two characteristic genes and immune cell infiltration in prostate cancer-related cell lines.Results A total of 35 DEGs and two core genes were found by using R language and machine learning methods in three prostate cancer datasets of GEO database,including 20 down-regulated genes and 15 up-regulated genes.Analysis of GO,KEGG,DO and GSEA pathways revealed that these genes were enriched in epidermal cell differentiation,keratinosis and other functions,and in extracellular matrix receptor interaction and estrogen receptor pathway.The characteristic genes screened by least absolute shrinkage and selection operator(LASSO)and support vector machuines(SVM)and the data set GSE32571 were tested.It was found that HPN and KRT23 were two diagnostic biomarkers of prostate cancer,and the mRNA level in the prostatic adenocarcinoma cell line was verified in line with the results of bioinformatics analysis.The expression levels of HPN in Du145,PC3,Vcap,Lncap,C4-2 and 22RV1 groups(1.10±0.29,0.46±0.12,3.02±0.79,1.58±0.09,0.39±0.02,0.41±0.07)was higher than RWPE1 group(0.09±0.01).The difference was statistically significant(t=6.000,5.030,6.400,27.980,15.600,6.870,P<0.05).The expression of KRT23 in Du145,PC3,Vcap,Lncap,C4-2 and 22RV1 groups(0.42±0.01,0.15±0.03,0.15±0.02,0.15±0.03,0.62±0.09,0.04±0.01)was lower than that in RWPE1 group(1.01±0.19).The difference was statistically significant(t=5.210,7.600,7.620,7.580,3.120,8.630,P<0.05),HPN and KRT23 were correlated with immune cells,HPN was negatively correlated with T cell CD8,resting mast cell and resting dendritic cell,and positively correlated with macrophage M0.KRT23 was negatively correlated with macrophage M0 and positively correlated with resting dendritic cells and resting mast cells.Conclusion HPN and KRT23 can be used as diagnostic biomarkers of prostate cancer,and HPN and KRT23 are related to immune cells such as follicular helper T cells and regulatory T cells.