An Integrative Study On Molecular Classification And Molecular Networks Of Bladder Urothelial Carcinoma

The biological characteristics of the two subtypes of bladder urothelial carcinoma, non-invasive superficial tumor and invasive tumor, are closely associated with the clinical performance, the therapeutical plan. However there is a lack of effective approaches to the preoperative discrimination for the two subtypes of bladder cancer. In this study, array comparative genome hybridization (array CGH) analysis was performed first, to obtain the genome DNA copy number alteration profiles that specifically related to the non-invasive or invasive bladder cancer, respectively. From these, a set of gain or loss genes/regions could be identified to generate a mathematical model for molecular classification of the two subtypes of tumor.The tumor tissue samples derived from 45 patients of bladder cancer were analyzed by array CGH on the Agilent human genome 4×44K microarrays. The genome-wide patterns of copy number alteration were obtained after analyzing by the Agilent software CGH Analytics 4.0. In euchromosomes, the alterations were mainly occurred in 1p, 1q, 2q,4q,5q,6q,8p,8q,9p,9q, 10q,11p, 11q,13q,14q, 17p and 20q. Combining the data with that of 20 bladder cancer samples previously analyzed by Agilent human genome 1×44K microarrays, a genomic profile was identified for the 65 cases of bladder cancer (29 non-invasive and 36 invasive). Besides,29 significantly variable fragments among the stage Ta, T1, T2 and T3 tumors were detected. Using them, a recursive partitioning decision tree model was constructed with 5 fragments located in 1p36.13, 1q32.1,5q11.2,9q21.31-q21.33 and 9q33.2, respectively. The accuracy of the model was 90.8%(59/65) for discriminating non-invasive and invasive bladder cancer. The reliability of the tree model was demonstrated by cross validation with the 'leave-one-out' algorithm. The 9 genes involved in the tree model (A₁4_P103301, FAM131C, CHIT1, CDC20B, RASEF, RMI1, DAPK1, TRAF1 and C5) were then validated by real time polymerase chain reaction (RT-PCR) in both the tumor tissue samples used for the array CGH analysis (n=57) and that derived from an independent group of bladder cancers (n=49). There was a considerable correlation between the results of the array CGH and real time PCR. profile and a microRNA (miRNA) expression profile were obtained from 35 and 32 tissue samples of bladder cancer, respectively, using the Agilent microarrays. There were 772 differentially expressed genes between non-invasive and invasive bladder cancer; and all these genes were associated with extracellular matrix by Gene Ontology (GO) searching. There were 146 differentially expressed miRNAs between the two types of bladder cancer. The target genes of the 5 most significantly different miRNAs were predicted, and subsequently analyzed by the GO.In this study,30 of the 65 cases were parallelly screened by array CGH. mRNA microarrays and miRNA microarrays; and the data mining was performed with bioinformatic approaches. Grounded on the mRNA expression profile,11 gene-modules correlated with tumor invasion were identified, using weighted correlation network analysis (WGCNA). Based on the data derived from GSEA (gene set enrichment analysis) and eQTL (expression Quantitative Trait Locus), the networks for each module were built up. which integrated the information from the molecular levels of DNA, mRNA and miRNA, and the biological significance of the molecular networks was discussed.In conclusion, data obtained from the present study indicated that there were differences in the genomic profiles between the non-invasive and invasive bladder cancers. The two types of bladder urothelial carcinomas could be classified with the decision tree model with 5 fragments, which was certified in an independent sample set. This integrative study on molecular classification and molecular networks provided considerable information for better understanding bladder urothelial carcinoma.