Molecular classification of bladder cancer: global mRNA classification versus tumor cell phenotype classification.

In this study gene expression profiles for 307 cases of advanced bladder cancers were compared to molecular phenotype at the tumor cell level. TUR-B tissue for RNA extraction was macrodissected from the close vicinity of the tissue sampled for immunohistochemistry to ensure high-quality sampling and to minimize the effects of intra-tumor heterogeneity. Despite excellent agreement between gene expression values and IHC-score at the single marker level, broad differences emerge when samples are clustered at the global mRNA versus tumor cell (IHC) levels. Classification at the different levels give different results in a systematic fashion, which implicates that analysis at both levels is required for optimal subtype-classification of bladder cancer. 307 FFPE embedded bladder cancers and 28 lymph node metastasis were analyzed using the GeneChip Human Gene 1.0 ST Arrays (Affymetrix). Total RNA was extracted using the High Pure FFPET RNA isolation kit (Roche), assessed for quality using Nanodrop ND-1000, and labeled using SensationPlus FFPE Amplification and WT labeling kit. The full dataset of 335 CEL files were normalized using RMA by labeling batches. Batches were combined and RMA normalized. The data was filtered for probes with signal intensity less than the median of negative control probes for >80% of the values. Gene symbols and EntrezID were appended using Bioconductor (hugene10stprobeset.db Version 8.3.1). Probes not mapping to a gene symbol were removed. The probes were median merged by gene symbol and log-transformed. Batch correction was performed using ComBat. 307/335 samples were used in this cohort. These samples were median centered. ConsensusClusterPlus was run on the data with 7031 genes (50% variance filter).